Datasets:

sugam11

/

covid-qa

like 0

Estimating the Unreported Number of Novel Coronavirus (2019-nCoV) Cases in China in the First Half of January 2020: A Data-Driven Modelling Analysis of the Early Outbreak https://doi.org/10.3390/jcm9020388 SHA: bf20dda99538a594eafc258553634fd9195104cb Authors: Zhao, Shi; Musa, Salihu S.; Lin, Qianying; Ran, Jinjun; Yang, Guangpu; Wang, Weiming; Lou, Yijun; Yang, Lin; Gao, Daozhou; He, Daihai; Wang, Maggie H. Date: 2020 DOI: 10.3390/jcm9020388 License: cc-by Abstract: Background: In December 2019, an outbreak of respiratory illness caused by a novel coronavirus (2019-nCoV) emerged in Wuhan, China and has swiftly spread to other parts of China and a number of foreign countries. The 2019-nCoV cases might have been under-reported roughly from 1 to 15 January 2020, and thus we estimated the number of unreported cases and the basic reproduction number, R0, of 2019-nCoV. Methods: We modelled the epidemic curve of 2019-nCoV cases, in mainland China from 1 December 2019 to 24 January 2020 through the exponential growth. The number of unreported cases was determined by the maximum likelihood estimation. We used the serial intervals (SI) of infection caused by two other well-known coronaviruses (CoV), Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS) CoVs, as approximations of the unknown SI for 2019-nCoV to estimate R0. Results: We confirmed that the initial growth phase followed an exponential growth pattern. The under-reporting was likely to have resulted in 469 (95% CI: 403−540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18−25) in comparison to the situation from 1 to 17 January 2020 on average. We estimated the R0 of 2019-nCoV at 2.56 (95% CI: 2.49−2.63). Conclusion: The under-reporting was likely to have occurred during the first half of January 2020 and should be considered in future investigation. Text: A novel coronavirus (2019-nCoV) infected pneumonia infection, which is deadly [1] , was first identified in Wuhan, China in December 2019 [2] . The virus causes a range of symptoms including fever, cough, and shortness of breath [3] . The cumulative number of reported cases slowly increased to cumulative 41 cases by 1 January 2020, and rapidly increased after 16 January 2020. As of 26 January 2020, the still ongoing outbreak had resulted in 2066 (618 of them are in Wuhan) confirmed cases and 56 (45 of them were in Wuhan) deaths in mainland China [4] , and sporadic cases exported from Wuhan were reported in Thailand, Japan, Republic of Korea, Hong Kong, Taiwan, Australia, and the United States, please see the World Health Organization (WHO) news release via https://www.who.int/csr/don/en/ from 14 to 21 January 2020. Using the number of cases exported from Wuhan to other countries, a research group at Imperial College London estimated that there had been 4000 (95%CI: 1000-9700) cases in Wuhan with symptoms onset by 18 January 2020, and the basic reproduction number (R 0 ) was estimated at 2.6 (95%CI: 1.5-3.5) [5] . Leung et al. drew a similar conclusion and estimated the number of cases exported from Wuhan to other major cities in China [6] , and the potentials of travel related risks of disease spreading was also indicated by [7] . Due to an unknown reason, the cumulative number of cases remained at 41 from 1 to 15 January 2020 according to the official report, i.e., no new case was reported during these 15 days, which appears inconsistent with the following rapid growth of the epidemic curve since 16 January 2020. We suspect that the 2019-nCoV cases were under-reported roughly from 1 to 15 January 2020. In this study, we estimated the number of unreported cases and the basic reproduction number, R 0 , of 2019-nCoV in Wuhan from 1 to 15 January 2020 based on the limited data in the early outbreak. The time series data of 2019-nCoV cases in mainland China were initially released by the Wuhan Municipal Health Commission from 10 to 20 January 2020 [8] , and later by the National Health Commission of China after 21 January 2020 [9] . The case time series data in December 2019 were obtained from a published study [3] . All cases were laboratory confirmed following the case definition by the national health commission of China [10] . We chose the data up to 24 January 2020 instead of to the present study completion date. Given the lag between timings of case confirmation and news release of new cases, the data of the most recent few days were most likely to be tentative, and thus they were excluded from the analysis to be consistent. We suspected that there was a number of cases, denoted by ξ, under-reported from 1 to 15 January 2020. The cumulative total number of cases, denoted by C i , of the i-th day since 1 December 2019 is the summation of the cumulative reported, c i , and cumulative unreported cases, Ξ i . We have C i = c i + Ξ i , where c i is observed from the data, and Ξ i is 0 for i before 1 January and ξ for i after 15 January 2020. Following previous studies [11, 12] , we modelled the epidemic curve, i.e., the C i series, as an exponential growing Poisson process. Since the data from 1 to 15 January 2020 appeared constant due to unclear reason(s), we removed these data from the fitting of exponential growth. The ξ and the intrinsic growth rate (γ) of the exponential growth were to be estimated based on the log-likelihood, denoted by , from the Poisson priors. The 95% confidence interval (95% CI) of ξ was estimated by the profile likelihood estimation framework with cutoff threshold determined by a Chi-square quantile [13] , χ 2 pr = 0.95, df = 1 . With γ estimated, the basic reproduction number could be obtained by R 0 = 1/M(−γ) with 100% susceptibility for 2019-nCoV presumed at this early stage. Here, the function M(·) was the Laplace transform, i.e., the moment generating function, of the probability distribution for the serial interval (SI) of the disease [11, 14] , denoted by h(k) and k is the mean SI. Since the transmission chain of 2019-nCoV remained unclear, we adopted the SI information from Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS), which share the similar pathogen as 2019-nCoV [15] [16] [17] . We modelled h(k) as Gamma distributions with mean of 8.0 days and standard deviation (SD) of 3.6 days by averaging the SI mean and SD of SARS, mean of 7.6 days and SD of 3.4 days [18] , and MERS, mean of 8.4 days and SD of 3.8 days [19] . We were also interested in inferring the patterns of the daily number of cases, denoted by ε i for the i-th day, and thus it is obviously that C i = C i−1 + ε i . A simulation framework was developed for the iterative Poisson process such that E[ denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403-540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R 0 was estimated at 2.56 (95% CI: 2.49-2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R 0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (C i ) remarkably well, see Figure 1c iterative Poisson process such that denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403−540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R0 was estimated at 2.56 (95% CI: 2.49−2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (Ci) remarkably well, see Figure 1c , referring to McFadden's pseudo-R-squared of 0.99. show the exponential growth fitting results of the cumulative number of cases (Ci) and the daily number of cases (εi) respectively. In panels (c) and (d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an panels (a,b) , the green shading area represents the 95% CI (on the horizontal axis), and the vertical green line represents the maximum likelihood estimate (MLE) of the number of unreported cases. With the MLE of R 0 at 2.56, panels (c,d) show the exponential growth fitting results of the cumulative number of cases (C i ) and the daily number of cases (ε i ) respectively. In panels (c,d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R 0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an insight into the transmission potential of 2019-nCoV at the early outbreak. We note that slightly varying the mean and SD of SI would not affect our main conclusions. The R 0 of 2019-nCoV was estimated at 2.56 (95% CI: 2.49-2.63), and it is generally in line with those of SARS, i.e., 2-5 [19, 24, 25] , and MERS, i.e., 2.7-3.9 [26] . For the simulated daily number of cases (ε i ), see Figure 1d , we found that ε i matched the observed daily number after 17 January 2020, but was significantly larger than the observations from 1 to 17 January 2020. This finding implied that under-reporting was likely to have occurred in the first half of January 2020. We estimated that the reporting rate after 17 January 2020 increased 21-fold (95% CI: [18] [19] [20] [21] [22] [23] [24] [25] compared to the situation from 1 to 17 January 2020 on average. One of the possible reasons was that the official diagnostic protocol was released by WHO on 17 January 2020 [27] , and the diagnosis and reporting efforts of 2019-nCoV infections probably increased. Thereafter, the daily number of newly reported cases started increasing rapidly after 17 January 2020, see Figure 1d . We conducted additional sensitivity analysis by varying the starting date of the under-reporting time window, e.g., 1 January 2020 in the main results, from 2 December 2019 to 3 January 2020, and we report our estimates largely hold. The exact value of the reporting rate was difficult to determine due to lack of serological surveillance data. The reporting rate can be determined if serological surveillance data are available for a population; we would know who was infected (seropositive) and who was not (seronegative), with high confidence. The reporting rate is the ratio of reported cases over the number of seropositive individuals. It was statistically evident that increasing in reporting was likely, and thus it should be considered in the future investigation of this outbreak. Previous preprint suggested cumulative cases of 1723 (95% CI: 427-4471) as of 12 January 2020, and 4000 (95% CI: 1000-9700) as of 18 January 2020 based on the aggregated international export cases [5] . Our analysis yielded cumulative cases of 280 (95% CI: 128-613) as of 12 January 2020, and 609 (95% CI: 278-1333) as of 18 January 2020 based on the exponential growing mechanistic in the early outbreak. Although our estimate case number appeared to have a lower mean than those estimated by Imai et al. [5] , they are not statistically different. This study applied a different screening effort to detect the 2019-nCoV cases from that in Imai et al. [5] . Imai et al. assumed the average screening effort at overseas airports that covered travelers arriving from Wuhan. Whereas we assumed a constant screening effort applied in Wuhan at the same point of time, and then a number of cases (i.e., ξ) should have been reported yet failed to be reported in the first half of January 2020 due to all sorts of reasons. It is not surprising that different assumptions yielded different results, and this difference in screening effort also partly explained why the detected cases out of China mainly presented mild symptoms. Thus, it was reasonable that our estimates appeared lower than those estimated by Imai et al. [5] . It must be emphasized that such a gap in the knowledge would be resolved by serological survey study (for a large population to approximate the actual positive rate) or an explicit estimation of the actual reporting rate. Under-reporting was likely to have occurred and resulted in 469 (95% CI: 403-540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18-25) compared with the situation from 1 to 17 January 2020 on average, and it should be considered in future investigation. We estimated the R 0 at 2019-nCoV to be 2.56 (95% CI: 2.49-2.63). Author Contributions: All authors conceived the study, carried out the analysis, discussed the results, drafted the first manuscript. All authors have read and agreed to the published version of the manuscript.

What was the result of the Imperial College estimation?

Estimating the Unreported Number of Novel Coronavirus (2019-nCoV) Cases in China in the First Half of January 2020: A Data-Driven Modelling Analysis of the Early Outbreak https://doi.org/10.3390/jcm9020388 SHA: bf20dda99538a594eafc258553634fd9195104cb Authors: Zhao, Shi; Musa, Salihu S.; Lin, Qianying; Ran, Jinjun; Yang, Guangpu; Wang, Weiming; Lou, Yijun; Yang, Lin; Gao, Daozhou; He, Daihai; Wang, Maggie H. Date: 2020 DOI: 10.3390/jcm9020388 License: cc-by Abstract: Background: In December 2019, an outbreak of respiratory illness caused by a novel coronavirus (2019-nCoV) emerged in Wuhan, China and has swiftly spread to other parts of China and a number of foreign countries. The 2019-nCoV cases might have been under-reported roughly from 1 to 15 January 2020, and thus we estimated the number of unreported cases and the basic reproduction number, R0, of 2019-nCoV. Methods: We modelled the epidemic curve of 2019-nCoV cases, in mainland China from 1 December 2019 to 24 January 2020 through the exponential growth. The number of unreported cases was determined by the maximum likelihood estimation. We used the serial intervals (SI) of infection caused by two other well-known coronaviruses (CoV), Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS) CoVs, as approximations of the unknown SI for 2019-nCoV to estimate R0. Results: We confirmed that the initial growth phase followed an exponential growth pattern. The under-reporting was likely to have resulted in 469 (95% CI: 403−540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18−25) in comparison to the situation from 1 to 17 January 2020 on average. We estimated the R0 of 2019-nCoV at 2.56 (95% CI: 2.49−2.63). Conclusion: The under-reporting was likely to have occurred during the first half of January 2020 and should be considered in future investigation. Text: A novel coronavirus (2019-nCoV) infected pneumonia infection, which is deadly [1] , was first identified in Wuhan, China in December 2019 [2] . The virus causes a range of symptoms including fever, cough, and shortness of breath [3] . The cumulative number of reported cases slowly increased to cumulative 41 cases by 1 January 2020, and rapidly increased after 16 January 2020. As of 26 January 2020, the still ongoing outbreak had resulted in 2066 (618 of them are in Wuhan) confirmed cases and 56 (45 of them were in Wuhan) deaths in mainland China [4] , and sporadic cases exported from Wuhan were reported in Thailand, Japan, Republic of Korea, Hong Kong, Taiwan, Australia, and the United States, please see the World Health Organization (WHO) news release via https://www.who.int/csr/don/en/ from 14 to 21 January 2020. Using the number of cases exported from Wuhan to other countries, a research group at Imperial College London estimated that there had been 4000 (95%CI: 1000-9700) cases in Wuhan with symptoms onset by 18 January 2020, and the basic reproduction number (R 0 ) was estimated at 2.6 (95%CI: 1.5-3.5) [5] . Leung et al. drew a similar conclusion and estimated the number of cases exported from Wuhan to other major cities in China [6] , and the potentials of travel related risks of disease spreading was also indicated by [7] . Due to an unknown reason, the cumulative number of cases remained at 41 from 1 to 15 January 2020 according to the official report, i.e., no new case was reported during these 15 days, which appears inconsistent with the following rapid growth of the epidemic curve since 16 January 2020. We suspect that the 2019-nCoV cases were under-reported roughly from 1 to 15 January 2020. In this study, we estimated the number of unreported cases and the basic reproduction number, R 0 , of 2019-nCoV in Wuhan from 1 to 15 January 2020 based on the limited data in the early outbreak. The time series data of 2019-nCoV cases in mainland China were initially released by the Wuhan Municipal Health Commission from 10 to 20 January 2020 [8] , and later by the National Health Commission of China after 21 January 2020 [9] . The case time series data in December 2019 were obtained from a published study [3] . All cases were laboratory confirmed following the case definition by the national health commission of China [10] . We chose the data up to 24 January 2020 instead of to the present study completion date. Given the lag between timings of case confirmation and news release of new cases, the data of the most recent few days were most likely to be tentative, and thus they were excluded from the analysis to be consistent. We suspected that there was a number of cases, denoted by ξ, under-reported from 1 to 15 January 2020. The cumulative total number of cases, denoted by C i , of the i-th day since 1 December 2019 is the summation of the cumulative reported, c i , and cumulative unreported cases, Ξ i . We have C i = c i + Ξ i , where c i is observed from the data, and Ξ i is 0 for i before 1 January and ξ for i after 15 January 2020. Following previous studies [11, 12] , we modelled the epidemic curve, i.e., the C i series, as an exponential growing Poisson process. Since the data from 1 to 15 January 2020 appeared constant due to unclear reason(s), we removed these data from the fitting of exponential growth. The ξ and the intrinsic growth rate (γ) of the exponential growth were to be estimated based on the log-likelihood, denoted by , from the Poisson priors. The 95% confidence interval (95% CI) of ξ was estimated by the profile likelihood estimation framework with cutoff threshold determined by a Chi-square quantile [13] , χ 2 pr = 0.95, df = 1 . With γ estimated, the basic reproduction number could be obtained by R 0 = 1/M(−γ) with 100% susceptibility for 2019-nCoV presumed at this early stage. Here, the function M(·) was the Laplace transform, i.e., the moment generating function, of the probability distribution for the serial interval (SI) of the disease [11, 14] , denoted by h(k) and k is the mean SI. Since the transmission chain of 2019-nCoV remained unclear, we adopted the SI information from Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS), which share the similar pathogen as 2019-nCoV [15] [16] [17] . We modelled h(k) as Gamma distributions with mean of 8.0 days and standard deviation (SD) of 3.6 days by averaging the SI mean and SD of SARS, mean of 7.6 days and SD of 3.4 days [18] , and MERS, mean of 8.4 days and SD of 3.8 days [19] . We were also interested in inferring the patterns of the daily number of cases, denoted by ε i for the i-th day, and thus it is obviously that C i = C i−1 + ε i . A simulation framework was developed for the iterative Poisson process such that E[ denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403-540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R 0 was estimated at 2.56 (95% CI: 2.49-2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R 0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (C i ) remarkably well, see Figure 1c iterative Poisson process such that denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403−540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R0 was estimated at 2.56 (95% CI: 2.49−2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (Ci) remarkably well, see Figure 1c , referring to McFadden's pseudo-R-squared of 0.99. show the exponential growth fitting results of the cumulative number of cases (Ci) and the daily number of cases (εi) respectively. In panels (c) and (d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an panels (a,b) , the green shading area represents the 95% CI (on the horizontal axis), and the vertical green line represents the maximum likelihood estimate (MLE) of the number of unreported cases. With the MLE of R 0 at 2.56, panels (c,d) show the exponential growth fitting results of the cumulative number of cases (C i ) and the daily number of cases (ε i ) respectively. In panels (c,d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R 0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an insight into the transmission potential of 2019-nCoV at the early outbreak. We note that slightly varying the mean and SD of SI would not affect our main conclusions. The R 0 of 2019-nCoV was estimated at 2.56 (95% CI: 2.49-2.63), and it is generally in line with those of SARS, i.e., 2-5 [19, 24, 25] , and MERS, i.e., 2.7-3.9 [26] . For the simulated daily number of cases (ε i ), see Figure 1d , we found that ε i matched the observed daily number after 17 January 2020, but was significantly larger than the observations from 1 to 17 January 2020. This finding implied that under-reporting was likely to have occurred in the first half of January 2020. We estimated that the reporting rate after 17 January 2020 increased 21-fold (95% CI: [18] [19] [20] [21] [22] [23] [24] [25] compared to the situation from 1 to 17 January 2020 on average. One of the possible reasons was that the official diagnostic protocol was released by WHO on 17 January 2020 [27] , and the diagnosis and reporting efforts of 2019-nCoV infections probably increased. Thereafter, the daily number of newly reported cases started increasing rapidly after 17 January 2020, see Figure 1d . We conducted additional sensitivity analysis by varying the starting date of the under-reporting time window, e.g., 1 January 2020 in the main results, from 2 December 2019 to 3 January 2020, and we report our estimates largely hold. The exact value of the reporting rate was difficult to determine due to lack of serological surveillance data. The reporting rate can be determined if serological surveillance data are available for a population; we would know who was infected (seropositive) and who was not (seronegative), with high confidence. The reporting rate is the ratio of reported cases over the number of seropositive individuals. It was statistically evident that increasing in reporting was likely, and thus it should be considered in the future investigation of this outbreak. Previous preprint suggested cumulative cases of 1723 (95% CI: 427-4471) as of 12 January 2020, and 4000 (95% CI: 1000-9700) as of 18 January 2020 based on the aggregated international export cases [5] . Our analysis yielded cumulative cases of 280 (95% CI: 128-613) as of 12 January 2020, and 609 (95% CI: 278-1333) as of 18 January 2020 based on the exponential growing mechanistic in the early outbreak. Although our estimate case number appeared to have a lower mean than those estimated by Imai et al. [5] , they are not statistically different. This study applied a different screening effort to detect the 2019-nCoV cases from that in Imai et al. [5] . Imai et al. assumed the average screening effort at overseas airports that covered travelers arriving from Wuhan. Whereas we assumed a constant screening effort applied in Wuhan at the same point of time, and then a number of cases (i.e., ξ) should have been reported yet failed to be reported in the first half of January 2020 due to all sorts of reasons. It is not surprising that different assumptions yielded different results, and this difference in screening effort also partly explained why the detected cases out of China mainly presented mild symptoms. Thus, it was reasonable that our estimates appeared lower than those estimated by Imai et al. [5] . It must be emphasized that such a gap in the knowledge would be resolved by serological survey study (for a large population to approximate the actual positive rate) or an explicit estimation of the actual reporting rate. Under-reporting was likely to have occurred and resulted in 469 (95% CI: 403-540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18-25) compared with the situation from 1 to 17 January 2020 on average, and it should be considered in future investigation. We estimated the R 0 at 2019-nCoV to be 2.56 (95% CI: 2.49-2.63). Author Contributions: All authors conceived the study, carried out the analysis, discussed the results, drafted the first manuscript. All authors have read and agreed to the published version of the manuscript.

Who release the time series data from 10th to 20th January 2020?

Estimating the Unreported Number of Novel Coronavirus (2019-nCoV) Cases in China in the First Half of January 2020: A Data-Driven Modelling Analysis of the Early Outbreak https://doi.org/10.3390/jcm9020388 SHA: bf20dda99538a594eafc258553634fd9195104cb Authors: Zhao, Shi; Musa, Salihu S.; Lin, Qianying; Ran, Jinjun; Yang, Guangpu; Wang, Weiming; Lou, Yijun; Yang, Lin; Gao, Daozhou; He, Daihai; Wang, Maggie H. Date: 2020 DOI: 10.3390/jcm9020388 License: cc-by Abstract: Background: In December 2019, an outbreak of respiratory illness caused by a novel coronavirus (2019-nCoV) emerged in Wuhan, China and has swiftly spread to other parts of China and a number of foreign countries. The 2019-nCoV cases might have been under-reported roughly from 1 to 15 January 2020, and thus we estimated the number of unreported cases and the basic reproduction number, R0, of 2019-nCoV. Methods: We modelled the epidemic curve of 2019-nCoV cases, in mainland China from 1 December 2019 to 24 January 2020 through the exponential growth. The number of unreported cases was determined by the maximum likelihood estimation. We used the serial intervals (SI) of infection caused by two other well-known coronaviruses (CoV), Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS) CoVs, as approximations of the unknown SI for 2019-nCoV to estimate R0. Results: We confirmed that the initial growth phase followed an exponential growth pattern. The under-reporting was likely to have resulted in 469 (95% CI: 403−540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18−25) in comparison to the situation from 1 to 17 January 2020 on average. We estimated the R0 of 2019-nCoV at 2.56 (95% CI: 2.49−2.63). Conclusion: The under-reporting was likely to have occurred during the first half of January 2020 and should be considered in future investigation. Text: A novel coronavirus (2019-nCoV) infected pneumonia infection, which is deadly [1] , was first identified in Wuhan, China in December 2019 [2] . The virus causes a range of symptoms including fever, cough, and shortness of breath [3] . The cumulative number of reported cases slowly increased to cumulative 41 cases by 1 January 2020, and rapidly increased after 16 January 2020. As of 26 January 2020, the still ongoing outbreak had resulted in 2066 (618 of them are in Wuhan) confirmed cases and 56 (45 of them were in Wuhan) deaths in mainland China [4] , and sporadic cases exported from Wuhan were reported in Thailand, Japan, Republic of Korea, Hong Kong, Taiwan, Australia, and the United States, please see the World Health Organization (WHO) news release via https://www.who.int/csr/don/en/ from 14 to 21 January 2020. Using the number of cases exported from Wuhan to other countries, a research group at Imperial College London estimated that there had been 4000 (95%CI: 1000-9700) cases in Wuhan with symptoms onset by 18 January 2020, and the basic reproduction number (R 0 ) was estimated at 2.6 (95%CI: 1.5-3.5) [5] . Leung et al. drew a similar conclusion and estimated the number of cases exported from Wuhan to other major cities in China [6] , and the potentials of travel related risks of disease spreading was also indicated by [7] . Due to an unknown reason, the cumulative number of cases remained at 41 from 1 to 15 January 2020 according to the official report, i.e., no new case was reported during these 15 days, which appears inconsistent with the following rapid growth of the epidemic curve since 16 January 2020. We suspect that the 2019-nCoV cases were under-reported roughly from 1 to 15 January 2020. In this study, we estimated the number of unreported cases and the basic reproduction number, R 0 , of 2019-nCoV in Wuhan from 1 to 15 January 2020 based on the limited data in the early outbreak. The time series data of 2019-nCoV cases in mainland China were initially released by the Wuhan Municipal Health Commission from 10 to 20 January 2020 [8] , and later by the National Health Commission of China after 21 January 2020 [9] . The case time series data in December 2019 were obtained from a published study [3] . All cases were laboratory confirmed following the case definition by the national health commission of China [10] . We chose the data up to 24 January 2020 instead of to the present study completion date. Given the lag between timings of case confirmation and news release of new cases, the data of the most recent few days were most likely to be tentative, and thus they were excluded from the analysis to be consistent. We suspected that there was a number of cases, denoted by ξ, under-reported from 1 to 15 January 2020. The cumulative total number of cases, denoted by C i , of the i-th day since 1 December 2019 is the summation of the cumulative reported, c i , and cumulative unreported cases, Ξ i . We have C i = c i + Ξ i , where c i is observed from the data, and Ξ i is 0 for i before 1 January and ξ for i after 15 January 2020. Following previous studies [11, 12] , we modelled the epidemic curve, i.e., the C i series, as an exponential growing Poisson process. Since the data from 1 to 15 January 2020 appeared constant due to unclear reason(s), we removed these data from the fitting of exponential growth. The ξ and the intrinsic growth rate (γ) of the exponential growth were to be estimated based on the log-likelihood, denoted by , from the Poisson priors. The 95% confidence interval (95% CI) of ξ was estimated by the profile likelihood estimation framework with cutoff threshold determined by a Chi-square quantile [13] , χ 2 pr = 0.95, df = 1 . With γ estimated, the basic reproduction number could be obtained by R 0 = 1/M(−γ) with 100% susceptibility for 2019-nCoV presumed at this early stage. Here, the function M(·) was the Laplace transform, i.e., the moment generating function, of the probability distribution for the serial interval (SI) of the disease [11, 14] , denoted by h(k) and k is the mean SI. Since the transmission chain of 2019-nCoV remained unclear, we adopted the SI information from Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS), which share the similar pathogen as 2019-nCoV [15] [16] [17] . We modelled h(k) as Gamma distributions with mean of 8.0 days and standard deviation (SD) of 3.6 days by averaging the SI mean and SD of SARS, mean of 7.6 days and SD of 3.4 days [18] , and MERS, mean of 8.4 days and SD of 3.8 days [19] . We were also interested in inferring the patterns of the daily number of cases, denoted by ε i for the i-th day, and thus it is obviously that C i = C i−1 + ε i . A simulation framework was developed for the iterative Poisson process such that E[ denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403-540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R 0 was estimated at 2.56 (95% CI: 2.49-2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R 0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (C i ) remarkably well, see Figure 1c iterative Poisson process such that denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403−540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R0 was estimated at 2.56 (95% CI: 2.49−2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (Ci) remarkably well, see Figure 1c , referring to McFadden's pseudo-R-squared of 0.99. show the exponential growth fitting results of the cumulative number of cases (Ci) and the daily number of cases (εi) respectively. In panels (c) and (d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an panels (a,b) , the green shading area represents the 95% CI (on the horizontal axis), and the vertical green line represents the maximum likelihood estimate (MLE) of the number of unreported cases. With the MLE of R 0 at 2.56, panels (c,d) show the exponential growth fitting results of the cumulative number of cases (C i ) and the daily number of cases (ε i ) respectively. In panels (c,d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R 0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an insight into the transmission potential of 2019-nCoV at the early outbreak. We note that slightly varying the mean and SD of SI would not affect our main conclusions. The R 0 of 2019-nCoV was estimated at 2.56 (95% CI: 2.49-2.63), and it is generally in line with those of SARS, i.e., 2-5 [19, 24, 25] , and MERS, i.e., 2.7-3.9 [26] . For the simulated daily number of cases (ε i ), see Figure 1d , we found that ε i matched the observed daily number after 17 January 2020, but was significantly larger than the observations from 1 to 17 January 2020. This finding implied that under-reporting was likely to have occurred in the first half of January 2020. We estimated that the reporting rate after 17 January 2020 increased 21-fold (95% CI: [18] [19] [20] [21] [22] [23] [24] [25] compared to the situation from 1 to 17 January 2020 on average. One of the possible reasons was that the official diagnostic protocol was released by WHO on 17 January 2020 [27] , and the diagnosis and reporting efforts of 2019-nCoV infections probably increased. Thereafter, the daily number of newly reported cases started increasing rapidly after 17 January 2020, see Figure 1d . We conducted additional sensitivity analysis by varying the starting date of the under-reporting time window, e.g., 1 January 2020 in the main results, from 2 December 2019 to 3 January 2020, and we report our estimates largely hold. The exact value of the reporting rate was difficult to determine due to lack of serological surveillance data. The reporting rate can be determined if serological surveillance data are available for a population; we would know who was infected (seropositive) and who was not (seronegative), with high confidence. The reporting rate is the ratio of reported cases over the number of seropositive individuals. It was statistically evident that increasing in reporting was likely, and thus it should be considered in the future investigation of this outbreak. Previous preprint suggested cumulative cases of 1723 (95% CI: 427-4471) as of 12 January 2020, and 4000 (95% CI: 1000-9700) as of 18 January 2020 based on the aggregated international export cases [5] . Our analysis yielded cumulative cases of 280 (95% CI: 128-613) as of 12 January 2020, and 609 (95% CI: 278-1333) as of 18 January 2020 based on the exponential growing mechanistic in the early outbreak. Although our estimate case number appeared to have a lower mean than those estimated by Imai et al. [5] , they are not statistically different. This study applied a different screening effort to detect the 2019-nCoV cases from that in Imai et al. [5] . Imai et al. assumed the average screening effort at overseas airports that covered travelers arriving from Wuhan. Whereas we assumed a constant screening effort applied in Wuhan at the same point of time, and then a number of cases (i.e., ξ) should have been reported yet failed to be reported in the first half of January 2020 due to all sorts of reasons. It is not surprising that different assumptions yielded different results, and this difference in screening effort also partly explained why the detected cases out of China mainly presented mild symptoms. Thus, it was reasonable that our estimates appeared lower than those estimated by Imai et al. [5] . It must be emphasized that such a gap in the knowledge would be resolved by serological survey study (for a large population to approximate the actual positive rate) or an explicit estimation of the actual reporting rate. Under-reporting was likely to have occurred and resulted in 469 (95% CI: 403-540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18-25) compared with the situation from 1 to 17 January 2020 on average, and it should be considered in future investigation. We estimated the R 0 at 2019-nCoV to be 2.56 (95% CI: 2.49-2.63). Author Contributions: All authors conceived the study, carried out the analysis, discussed the results, drafted the first manuscript. All authors have read and agreed to the published version of the manuscript.

Who released the time series data from after 21st January 2020?

Estimating the Unreported Number of Novel Coronavirus (2019-nCoV) Cases in China in the First Half of January 2020: A Data-Driven Modelling Analysis of the Early Outbreak https://doi.org/10.3390/jcm9020388 SHA: bf20dda99538a594eafc258553634fd9195104cb Authors: Zhao, Shi; Musa, Salihu S.; Lin, Qianying; Ran, Jinjun; Yang, Guangpu; Wang, Weiming; Lou, Yijun; Yang, Lin; Gao, Daozhou; He, Daihai; Wang, Maggie H. Date: 2020 DOI: 10.3390/jcm9020388 License: cc-by Abstract: Background: In December 2019, an outbreak of respiratory illness caused by a novel coronavirus (2019-nCoV) emerged in Wuhan, China and has swiftly spread to other parts of China and a number of foreign countries. The 2019-nCoV cases might have been under-reported roughly from 1 to 15 January 2020, and thus we estimated the number of unreported cases and the basic reproduction number, R0, of 2019-nCoV. Methods: We modelled the epidemic curve of 2019-nCoV cases, in mainland China from 1 December 2019 to 24 January 2020 through the exponential growth. The number of unreported cases was determined by the maximum likelihood estimation. We used the serial intervals (SI) of infection caused by two other well-known coronaviruses (CoV), Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS) CoVs, as approximations of the unknown SI for 2019-nCoV to estimate R0. Results: We confirmed that the initial growth phase followed an exponential growth pattern. The under-reporting was likely to have resulted in 469 (95% CI: 403−540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18−25) in comparison to the situation from 1 to 17 January 2020 on average. We estimated the R0 of 2019-nCoV at 2.56 (95% CI: 2.49−2.63). Conclusion: The under-reporting was likely to have occurred during the first half of January 2020 and should be considered in future investigation. Text: A novel coronavirus (2019-nCoV) infected pneumonia infection, which is deadly [1] , was first identified in Wuhan, China in December 2019 [2] . The virus causes a range of symptoms including fever, cough, and shortness of breath [3] . The cumulative number of reported cases slowly increased to cumulative 41 cases by 1 January 2020, and rapidly increased after 16 January 2020. As of 26 January 2020, the still ongoing outbreak had resulted in 2066 (618 of them are in Wuhan) confirmed cases and 56 (45 of them were in Wuhan) deaths in mainland China [4] , and sporadic cases exported from Wuhan were reported in Thailand, Japan, Republic of Korea, Hong Kong, Taiwan, Australia, and the United States, please see the World Health Organization (WHO) news release via https://www.who.int/csr/don/en/ from 14 to 21 January 2020. Using the number of cases exported from Wuhan to other countries, a research group at Imperial College London estimated that there had been 4000 (95%CI: 1000-9700) cases in Wuhan with symptoms onset by 18 January 2020, and the basic reproduction number (R 0 ) was estimated at 2.6 (95%CI: 1.5-3.5) [5] . Leung et al. drew a similar conclusion and estimated the number of cases exported from Wuhan to other major cities in China [6] , and the potentials of travel related risks of disease spreading was also indicated by [7] . Due to an unknown reason, the cumulative number of cases remained at 41 from 1 to 15 January 2020 according to the official report, i.e., no new case was reported during these 15 days, which appears inconsistent with the following rapid growth of the epidemic curve since 16 January 2020. We suspect that the 2019-nCoV cases were under-reported roughly from 1 to 15 January 2020. In this study, we estimated the number of unreported cases and the basic reproduction number, R 0 , of 2019-nCoV in Wuhan from 1 to 15 January 2020 based on the limited data in the early outbreak. The time series data of 2019-nCoV cases in mainland China were initially released by the Wuhan Municipal Health Commission from 10 to 20 January 2020 [8] , and later by the National Health Commission of China after 21 January 2020 [9] . The case time series data in December 2019 were obtained from a published study [3] . All cases were laboratory confirmed following the case definition by the national health commission of China [10] . We chose the data up to 24 January 2020 instead of to the present study completion date. Given the lag between timings of case confirmation and news release of new cases, the data of the most recent few days were most likely to be tentative, and thus they were excluded from the analysis to be consistent. We suspected that there was a number of cases, denoted by ξ, under-reported from 1 to 15 January 2020. The cumulative total number of cases, denoted by C i , of the i-th day since 1 December 2019 is the summation of the cumulative reported, c i , and cumulative unreported cases, Ξ i . We have C i = c i + Ξ i , where c i is observed from the data, and Ξ i is 0 for i before 1 January and ξ for i after 15 January 2020. Following previous studies [11, 12] , we modelled the epidemic curve, i.e., the C i series, as an exponential growing Poisson process. Since the data from 1 to 15 January 2020 appeared constant due to unclear reason(s), we removed these data from the fitting of exponential growth. The ξ and the intrinsic growth rate (γ) of the exponential growth were to be estimated based on the log-likelihood, denoted by , from the Poisson priors. The 95% confidence interval (95% CI) of ξ was estimated by the profile likelihood estimation framework with cutoff threshold determined by a Chi-square quantile [13] , χ 2 pr = 0.95, df = 1 . With γ estimated, the basic reproduction number could be obtained by R 0 = 1/M(−γ) with 100% susceptibility for 2019-nCoV presumed at this early stage. Here, the function M(·) was the Laplace transform, i.e., the moment generating function, of the probability distribution for the serial interval (SI) of the disease [11, 14] , denoted by h(k) and k is the mean SI. Since the transmission chain of 2019-nCoV remained unclear, we adopted the SI information from Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS), which share the similar pathogen as 2019-nCoV [15] [16] [17] . We modelled h(k) as Gamma distributions with mean of 8.0 days and standard deviation (SD) of 3.6 days by averaging the SI mean and SD of SARS, mean of 7.6 days and SD of 3.4 days [18] , and MERS, mean of 8.4 days and SD of 3.8 days [19] . We were also interested in inferring the patterns of the daily number of cases, denoted by ε i for the i-th day, and thus it is obviously that C i = C i−1 + ε i . A simulation framework was developed for the iterative Poisson process such that E[ denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403-540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R 0 was estimated at 2.56 (95% CI: 2.49-2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R 0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (C i ) remarkably well, see Figure 1c iterative Poisson process such that denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403−540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R0 was estimated at 2.56 (95% CI: 2.49−2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (Ci) remarkably well, see Figure 1c , referring to McFadden's pseudo-R-squared of 0.99. show the exponential growth fitting results of the cumulative number of cases (Ci) and the daily number of cases (εi) respectively. In panels (c) and (d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an panels (a,b) , the green shading area represents the 95% CI (on the horizontal axis), and the vertical green line represents the maximum likelihood estimate (MLE) of the number of unreported cases. With the MLE of R 0 at 2.56, panels (c,d) show the exponential growth fitting results of the cumulative number of cases (C i ) and the daily number of cases (ε i ) respectively. In panels (c,d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R 0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an insight into the transmission potential of 2019-nCoV at the early outbreak. We note that slightly varying the mean and SD of SI would not affect our main conclusions. The R 0 of 2019-nCoV was estimated at 2.56 (95% CI: 2.49-2.63), and it is generally in line with those of SARS, i.e., 2-5 [19, 24, 25] , and MERS, i.e., 2.7-3.9 [26] . For the simulated daily number of cases (ε i ), see Figure 1d , we found that ε i matched the observed daily number after 17 January 2020, but was significantly larger than the observations from 1 to 17 January 2020. This finding implied that under-reporting was likely to have occurred in the first half of January 2020. We estimated that the reporting rate after 17 January 2020 increased 21-fold (95% CI: [18] [19] [20] [21] [22] [23] [24] [25] compared to the situation from 1 to 17 January 2020 on average. One of the possible reasons was that the official diagnostic protocol was released by WHO on 17 January 2020 [27] , and the diagnosis and reporting efforts of 2019-nCoV infections probably increased. Thereafter, the daily number of newly reported cases started increasing rapidly after 17 January 2020, see Figure 1d . We conducted additional sensitivity analysis by varying the starting date of the under-reporting time window, e.g., 1 January 2020 in the main results, from 2 December 2019 to 3 January 2020, and we report our estimates largely hold. The exact value of the reporting rate was difficult to determine due to lack of serological surveillance data. The reporting rate can be determined if serological surveillance data are available for a population; we would know who was infected (seropositive) and who was not (seronegative), with high confidence. The reporting rate is the ratio of reported cases over the number of seropositive individuals. It was statistically evident that increasing in reporting was likely, and thus it should be considered in the future investigation of this outbreak. Previous preprint suggested cumulative cases of 1723 (95% CI: 427-4471) as of 12 January 2020, and 4000 (95% CI: 1000-9700) as of 18 January 2020 based on the aggregated international export cases [5] . Our analysis yielded cumulative cases of 280 (95% CI: 128-613) as of 12 January 2020, and 609 (95% CI: 278-1333) as of 18 January 2020 based on the exponential growing mechanistic in the early outbreak. Although our estimate case number appeared to have a lower mean than those estimated by Imai et al. [5] , they are not statistically different. This study applied a different screening effort to detect the 2019-nCoV cases from that in Imai et al. [5] . Imai et al. assumed the average screening effort at overseas airports that covered travelers arriving from Wuhan. Whereas we assumed a constant screening effort applied in Wuhan at the same point of time, and then a number of cases (i.e., ξ) should have been reported yet failed to be reported in the first half of January 2020 due to all sorts of reasons. It is not surprising that different assumptions yielded different results, and this difference in screening effort also partly explained why the detected cases out of China mainly presented mild symptoms. Thus, it was reasonable that our estimates appeared lower than those estimated by Imai et al. [5] . It must be emphasized that such a gap in the knowledge would be resolved by serological survey study (for a large population to approximate the actual positive rate) or an explicit estimation of the actual reporting rate. Under-reporting was likely to have occurred and resulted in 469 (95% CI: 403-540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18-25) compared with the situation from 1 to 17 January 2020 on average, and it should be considered in future investigation. We estimated the R 0 at 2019-nCoV to be 2.56 (95% CI: 2.49-2.63). Author Contributions: All authors conceived the study, carried out the analysis, discussed the results, drafted the first manuscript. All authors have read and agreed to the published version of the manuscript.

How was the epidemic curve modelled?

Estimating the Unreported Number of Novel Coronavirus (2019-nCoV) Cases in China in the First Half of January 2020: A Data-Driven Modelling Analysis of the Early Outbreak https://doi.org/10.3390/jcm9020388 SHA: bf20dda99538a594eafc258553634fd9195104cb Authors: Zhao, Shi; Musa, Salihu S.; Lin, Qianying; Ran, Jinjun; Yang, Guangpu; Wang, Weiming; Lou, Yijun; Yang, Lin; Gao, Daozhou; He, Daihai; Wang, Maggie H. Date: 2020 DOI: 10.3390/jcm9020388 License: cc-by Abstract: Background: In December 2019, an outbreak of respiratory illness caused by a novel coronavirus (2019-nCoV) emerged in Wuhan, China and has swiftly spread to other parts of China and a number of foreign countries. The 2019-nCoV cases might have been under-reported roughly from 1 to 15 January 2020, and thus we estimated the number of unreported cases and the basic reproduction number, R0, of 2019-nCoV. Methods: We modelled the epidemic curve of 2019-nCoV cases, in mainland China from 1 December 2019 to 24 January 2020 through the exponential growth. The number of unreported cases was determined by the maximum likelihood estimation. We used the serial intervals (SI) of infection caused by two other well-known coronaviruses (CoV), Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS) CoVs, as approximations of the unknown SI for 2019-nCoV to estimate R0. Results: We confirmed that the initial growth phase followed an exponential growth pattern. The under-reporting was likely to have resulted in 469 (95% CI: 403−540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18−25) in comparison to the situation from 1 to 17 January 2020 on average. We estimated the R0 of 2019-nCoV at 2.56 (95% CI: 2.49−2.63). Conclusion: The under-reporting was likely to have occurred during the first half of January 2020 and should be considered in future investigation. Text: A novel coronavirus (2019-nCoV) infected pneumonia infection, which is deadly [1] , was first identified in Wuhan, China in December 2019 [2] . The virus causes a range of symptoms including fever, cough, and shortness of breath [3] . The cumulative number of reported cases slowly increased to cumulative 41 cases by 1 January 2020, and rapidly increased after 16 January 2020. As of 26 January 2020, the still ongoing outbreak had resulted in 2066 (618 of them are in Wuhan) confirmed cases and 56 (45 of them were in Wuhan) deaths in mainland China [4] , and sporadic cases exported from Wuhan were reported in Thailand, Japan, Republic of Korea, Hong Kong, Taiwan, Australia, and the United States, please see the World Health Organization (WHO) news release via https://www.who.int/csr/don/en/ from 14 to 21 January 2020. Using the number of cases exported from Wuhan to other countries, a research group at Imperial College London estimated that there had been 4000 (95%CI: 1000-9700) cases in Wuhan with symptoms onset by 18 January 2020, and the basic reproduction number (R 0 ) was estimated at 2.6 (95%CI: 1.5-3.5) [5] . Leung et al. drew a similar conclusion and estimated the number of cases exported from Wuhan to other major cities in China [6] , and the potentials of travel related risks of disease spreading was also indicated by [7] . Due to an unknown reason, the cumulative number of cases remained at 41 from 1 to 15 January 2020 according to the official report, i.e., no new case was reported during these 15 days, which appears inconsistent with the following rapid growth of the epidemic curve since 16 January 2020. We suspect that the 2019-nCoV cases were under-reported roughly from 1 to 15 January 2020. In this study, we estimated the number of unreported cases and the basic reproduction number, R 0 , of 2019-nCoV in Wuhan from 1 to 15 January 2020 based on the limited data in the early outbreak. The time series data of 2019-nCoV cases in mainland China were initially released by the Wuhan Municipal Health Commission from 10 to 20 January 2020 [8] , and later by the National Health Commission of China after 21 January 2020 [9] . The case time series data in December 2019 were obtained from a published study [3] . All cases were laboratory confirmed following the case definition by the national health commission of China [10] . We chose the data up to 24 January 2020 instead of to the present study completion date. Given the lag between timings of case confirmation and news release of new cases, the data of the most recent few days were most likely to be tentative, and thus they were excluded from the analysis to be consistent. We suspected that there was a number of cases, denoted by ξ, under-reported from 1 to 15 January 2020. The cumulative total number of cases, denoted by C i , of the i-th day since 1 December 2019 is the summation of the cumulative reported, c i , and cumulative unreported cases, Ξ i . We have C i = c i + Ξ i , where c i is observed from the data, and Ξ i is 0 for i before 1 January and ξ for i after 15 January 2020. Following previous studies [11, 12] , we modelled the epidemic curve, i.e., the C i series, as an exponential growing Poisson process. Since the data from 1 to 15 January 2020 appeared constant due to unclear reason(s), we removed these data from the fitting of exponential growth. The ξ and the intrinsic growth rate (γ) of the exponential growth were to be estimated based on the log-likelihood, denoted by , from the Poisson priors. The 95% confidence interval (95% CI) of ξ was estimated by the profile likelihood estimation framework with cutoff threshold determined by a Chi-square quantile [13] , χ 2 pr = 0.95, df = 1 . With γ estimated, the basic reproduction number could be obtained by R 0 = 1/M(−γ) with 100% susceptibility for 2019-nCoV presumed at this early stage. Here, the function M(·) was the Laplace transform, i.e., the moment generating function, of the probability distribution for the serial interval (SI) of the disease [11, 14] , denoted by h(k) and k is the mean SI. Since the transmission chain of 2019-nCoV remained unclear, we adopted the SI information from Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS), which share the similar pathogen as 2019-nCoV [15] [16] [17] . We modelled h(k) as Gamma distributions with mean of 8.0 days and standard deviation (SD) of 3.6 days by averaging the SI mean and SD of SARS, mean of 7.6 days and SD of 3.4 days [18] , and MERS, mean of 8.4 days and SD of 3.8 days [19] . We were also interested in inferring the patterns of the daily number of cases, denoted by ε i for the i-th day, and thus it is obviously that C i = C i−1 + ε i . A simulation framework was developed for the iterative Poisson process such that E[ denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403-540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R 0 was estimated at 2.56 (95% CI: 2.49-2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R 0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (C i ) remarkably well, see Figure 1c iterative Poisson process such that denoted the expectation. The simulation was implemented starting from 1 January 2020 with a cumulative number of cases seed of 40, the same as reported on 31 December 2019. We conducted 1000 samples and calculated the median and 95% CI. The number of 2019-nCoV unreported cases was estimated at 469 (95% CI: 403−540), see Figure 1a , which was significantly larger than 0. This finding implied the occurrence of under-reporting between 1 and 15 January 2020. After accounting for the effect of under-reporting, the R0 was estimated at 2.56 (95% CI: 2.49−2.63), see Figure 1b , which is consistent with many existing online preprints with range from 2 to 4 [5, [20] [21] [22] . With the R0 of 2.56 and ξ of 469, the exponential growing framework fitted the cumulative total number of cases (Ci) remarkably well, see Figure 1c , referring to McFadden's pseudo-R-squared of 0.99. show the exponential growth fitting results of the cumulative number of cases (Ci) and the daily number of cases (εi) respectively. In panels (c) and (d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an panels (a,b) , the green shading area represents the 95% CI (on the horizontal axis), and the vertical green line represents the maximum likelihood estimate (MLE) of the number of unreported cases. With the MLE of R 0 at 2.56, panels (c,d) show the exponential growth fitting results of the cumulative number of cases (C i ) and the daily number of cases (ε i ) respectively. In panels (c,d), the gold squares are the reported cases, the blue bold curve represents the median of the fitting results, the dashed blue curves are the 95% CI of the fitting results, and the purple shading area represents the time window from 1 to 15 January 2020. In panel (c), the blue dots are the cumulative total, i.e., reported and unreported, number of cases. In panel (d), the grey curves are the 1000 simulation samples. Our estimation of R 0 rely on the SI of 2019-nCoV, which remains unknown as of 26 January 2020. In this work, we employed the SIs of SARS and MERS as approximations to that of 2019-nCoV. The determination of SI requires the knowledge of the chain of disease transmission that needs a sufficient number of patient samples and periods of time for follow-up [23] , and thus this is unlikely to be achieved shortly. However, using SIs of SARS and MERS as approximation could provide an insight into the transmission potential of 2019-nCoV at the early outbreak. We note that slightly varying the mean and SD of SI would not affect our main conclusions. The R 0 of 2019-nCoV was estimated at 2.56 (95% CI: 2.49-2.63), and it is generally in line with those of SARS, i.e., 2-5 [19, 24, 25] , and MERS, i.e., 2.7-3.9 [26] . For the simulated daily number of cases (ε i ), see Figure 1d , we found that ε i matched the observed daily number after 17 January 2020, but was significantly larger than the observations from 1 to 17 January 2020. This finding implied that under-reporting was likely to have occurred in the first half of January 2020. We estimated that the reporting rate after 17 January 2020 increased 21-fold (95% CI: [18] [19] [20] [21] [22] [23] [24] [25] compared to the situation from 1 to 17 January 2020 on average. One of the possible reasons was that the official diagnostic protocol was released by WHO on 17 January 2020 [27] , and the diagnosis and reporting efforts of 2019-nCoV infections probably increased. Thereafter, the daily number of newly reported cases started increasing rapidly after 17 January 2020, see Figure 1d . We conducted additional sensitivity analysis by varying the starting date of the under-reporting time window, e.g., 1 January 2020 in the main results, from 2 December 2019 to 3 January 2020, and we report our estimates largely hold. The exact value of the reporting rate was difficult to determine due to lack of serological surveillance data. The reporting rate can be determined if serological surveillance data are available for a population; we would know who was infected (seropositive) and who was not (seronegative), with high confidence. The reporting rate is the ratio of reported cases over the number of seropositive individuals. It was statistically evident that increasing in reporting was likely, and thus it should be considered in the future investigation of this outbreak. Previous preprint suggested cumulative cases of 1723 (95% CI: 427-4471) as of 12 January 2020, and 4000 (95% CI: 1000-9700) as of 18 January 2020 based on the aggregated international export cases [5] . Our analysis yielded cumulative cases of 280 (95% CI: 128-613) as of 12 January 2020, and 609 (95% CI: 278-1333) as of 18 January 2020 based on the exponential growing mechanistic in the early outbreak. Although our estimate case number appeared to have a lower mean than those estimated by Imai et al. [5] , they are not statistically different. This study applied a different screening effort to detect the 2019-nCoV cases from that in Imai et al. [5] . Imai et al. assumed the average screening effort at overseas airports that covered travelers arriving from Wuhan. Whereas we assumed a constant screening effort applied in Wuhan at the same point of time, and then a number of cases (i.e., ξ) should have been reported yet failed to be reported in the first half of January 2020 due to all sorts of reasons. It is not surprising that different assumptions yielded different results, and this difference in screening effort also partly explained why the detected cases out of China mainly presented mild symptoms. Thus, it was reasonable that our estimates appeared lower than those estimated by Imai et al. [5] . It must be emphasized that such a gap in the knowledge would be resolved by serological survey study (for a large population to approximate the actual positive rate) or an explicit estimation of the actual reporting rate. Under-reporting was likely to have occurred and resulted in 469 (95% CI: 403-540) unreported cases from 1 to 15 January 2020. The reporting rate after 17 January 2020 was likely to have increased 21-fold (95% CI: 18-25) compared with the situation from 1 to 17 January 2020 on average, and it should be considered in future investigation. We estimated the R 0 at 2019-nCoV to be 2.56 (95% CI: 2.49-2.63). Author Contributions: All authors conceived the study, carried out the analysis, discussed the results, drafted the first manuscript. All authors have read and agreed to the published version of the manuscript.

How was the epidemic curve modeled?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What were the most common viruses sampled from nasal swabs in Ilorin, Nigeria

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What was the most common virus detected in community members in this sample?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

How bad is the burden of disease in developing countries?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

Where do the majority of all infectious disease outbreaks happen?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What are some risk factors for countries to experience a high prevalence of Acute Respiratory Infections?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What symptoms are associated with acute respiratory infections?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What was the most common virus detected in community samples in Ilorin, Nigeria?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What was the prevalence of Coronavirus OC43 in community samples in Ilorin, Nigeria?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What was the prevalence of Coronavirus OC 229 E/NL63 in clinical subjects in Ilorin, Nigeria?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

What was the difference between community and clinic cases of acute respiratory infections?

Etiology of respiratory tract infections in the community and clinic in Ilorin, Nigeria https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5719735/ SHA: f2e835d2cde5f42054dbd0c20d4060721135c518 Authors: Kolawole, Olatunji; Oguntoye, Michael; Dam, Tina; Chunara, Rumi Date: 2017-12-07 DOI: 10.1186/s13104-017-3063-1 License: cc-by Abstract: OBJECTIVE: Recognizing increasing interest in community disease surveillance globally, the goal of this study was to investigate whether respiratory viruses circulating in the community may be represented through clinical (hospital) surveillance in Nigeria. RESULTS: Children were selected via convenience sampling from communities and a tertiary care center (n = 91) during spring 2017 in Ilorin, Nigeria. Nasal swabs were collected and tested using polymerase chain reaction. The majority (79.1%) of subjects were under 6 years old, of whom 46 were infected (63.9%). A total of 33 of the 91 subjects had one or more respiratory tract virus; there were 10 cases of triple infection and 5 of quadruple. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses in the clinical sample; present in 93.8% (15/16) of clinical subjects, and 6.7% (5/75) of community subjects (significant difference, p < 0.001). Coronavirus OC43 was the most common virus detected in community members (13.3%, 10/75). A different strain, Coronavirus OC 229 E/NL63 was detected among subjects from the clinic (2/16) and not detected in the community. This pilot study provides evidence that data from the community can potentially represent different information than that sourced clinically, suggesting the need for community surveillance to enhance public health efforts and scientific understanding of respiratory infections. Text: Acute Respiratory Infections (ARIs) (the cause of both upper respiratory tract infections (URIs) and lower respiratory tract infections (LRIs)) are a major cause of death among children under 5 years old particularly in developing countries where the burden of disease is 2-5 times higher than in developed countries [1] . While these viruses usually cause mild cold-like symptoms and can be self-limiting, in recent years novel coronaviruses such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS) have evolved and infected humans, causing severe illness, epidemics and pandemics [2] . Currently, the majority of all infectious disease outbreaks as recorded by the World Health Organization (WHO) occur in the continent of Africa where there is high transmission risk [3, 4] . Further, in developing areas (both rural and urban), there are increasing risk factors such as human-animal interfaces (due to residential-proximity to livestock). These changing epidemiological patterns have resulted in calls for improved ARI surveillance, especially in places of high transmission risk [5] . Nigeria is one such place with high prevalence of many of the risk factors implicated in ARI among children including; age, sex, overcrowding, nutritional status, socio-economic status, and where study of ARIs is currently limited [6] . These broad risk factors alongside limited resources have indicated the need for community-based initiatives for surveillance and interventions [6, 7] . For ARI surveillance in particular, infections in the community are those that do not get reported clinically. Clinical data generally represents the most severe cases, and those from locations with access to healthcare institutions. In Nigeria, hospitals are visited only when symptoms are very severe. Thus, it is hypothesized that viral information from clinical sampling is insufficient to either capture disease incidence in general populations or its predictability from symptoms [8] . Efforts worldwide including in East and Southern Africa have been focused on developing community-based participatory disease surveillance methods [9] [10] [11] [12] [13] . Community-based approaches have been shown useful for learning more about emerging respiratory infections such as assessing under-reporting [14] , types of viruses prevalent in communities [10] , and prediction of epidemics [15] . Concurrently, advancements in molecular identification methods have enabled studies regarding the emergence and epidemiology of ARI viruses in many locations (e.g. novel polyomaviruses in Australia [16, 17] , human coronavirus Erasmus Medical Center (HCoV-EMC) in the Middle East and United Kingdom [18, 19] , SARS in Canada and China [20] [21] [22] ), yet research regarding the molecular epidemiology of ARI viruses in Nigeria is limited. Diagnostic methods available and other constraints have limited studies there to serological surveys of only a few of these viruses and only in clinical populations [23, 24] . Thus, the utility of community-based surveillance may be appropriate in contexts such as in Nigeria, and the purpose of this pilot study was to investigate if clinical cases may describe the entire picture of ARI among children in Nigeria. We performed a cross-sectional study in three community centers and one clinical in Ilorin, Nigeria. Ilorin is in Kwara state and is the 6th largest city in Nigeria by population [25] . Three Local Government Areas (Ilorin East, Ilorin South and Ilorin West LGAs) were the community sites and Children's Specialist Hospital, Ilorin the clinical site. Convenience sampling was used for the purposes of this pilot study, and samples were obtained from March 28 to April 5 2017. Inclusion criteria were: children less than 14 years old who had visible symptoms of ARI within the communities or those confirmed at the hospital with ARI. Exclusion criteria were: children who were 14 and above, not showing signs of ARI and subjects whose parents did not give consent. Twenty-five children with symptoms were selected each from the three community locations while 16 symptomatic children were sampled from the hospital. The total sample size (n = 91) was arrived at based on materials and processing cost constraints, as well as to provide enough samples to enable descriptive understanding of viral circulation patterns estimated from other community-based studies [10] . Disease Surveillance and Notification Officers, who are employed by the State Ministry of Health and familiar with the communities in this study, performed specimen and data collection. Symptoms considered were derived in accordance with other ARI surveillance efforts: sore throat, fever, couch, running nose, vomiting, body ache, leg pain, nausea, chills, shortness of breath [10, 26] . Gender and age, type of residential area (rural/urban), education level, proximity of residence to livestock, proximity to an untarred road and number of people who sleep in same room, were all recorded. The general difference between the two settings was that those from the hospital had severe illnesses, while those from the community were generally "healthy" but exhibiting ARI symptoms (i.e. mild illness). Nasal swabs were collected from the subjects and stored in DNA/RNA shield (Zymo Research, Irvine, California). Collected samples were spinned and the swab removed. Residues containing the nasal samples were stored at -20 °C prior to molecular analysis. Viral RNA was isolated using ZR Viral RNA ™ Kit (Zymo Research, Irvine, California) per manufacturer instructions (http://www.zymoresearch.com/downloads/dl/file/ id/147/r1034i.pdf ). Real-time PCR (polymerase chain reaction), commonly used in ARI studies [10, 19, 27] , was then carried out using RV15 One Step ACE Detection Kit, catalogue numbers RV0716K01008007 and RV0717B01008001 (Seegene, Seoul, South Korea) for detection of 15 human viruses: parainfluenza virus 1, 2, 3 and 4 (PIV1-4), respiratory syncytial virus (RSV) A and B, influenza A and B (FLUA, FLUB), rhinovirus type A-C, adenovirus (ADV), coronavirus (OC 229 E/NL63, OC43), enterovirus (HEV), metapneumovirus (hMPV) and bocavirus (BoV). Reagents were validated in the experimental location using an inbuilt validation protocol to confirm issues of false negative and false positive results were not of concern. Amplification reaction was carried out as described by the manufacturer: reverse transcription 50 °C-30′, initial activation 94°-15′, 45 cycles: denaturation 94°-30″, annealing 60°-1′ 30″, extension 72°-1, final extension 72°-10′, hold 4°. Visualization was performed using electrophoresis on a 2% agarose gel in TBE 1X with EtBr, in presence of RV15 OneStep A/B/C Markers; molecular weight marker. Specimen processing was not blinded as there was no risk of experimental bias. Standardized procedures were used for community and clinic sampling. All statistical analyses were performed using R version 3.2.4. Univariate statistics [mean and 95% confidence interval (CI)] are described. Bivariate statistics (difference in proportions) were assessed using a two-proportion z-test. A p value < 0.001 was considered significant. No observations used in this study had any missing data for analyses in this study. Basic participant demographics are summarized in PCR results showed that ten different viruses (influenza A, coronavirus OC 229 E/NL63, RSVA, RSV B, parainfluenza 1-4) were detected. Figure 1 shows how these infections were distributed across virus types as well as in the community versus clinic samples. In sum, a total of 33 of the 91 subjects surveyed had one or more respiratory tract virus (36.3%, 95% CI 26.6-47.0%, Fig. 1 ). Furthermore, 10 of those cases were triple infections and 5 were quadruple infections (illustrated by color of bars in Fig. 1 ). Figure 2 indicates how frequently each pair of viruses were found in the same participant; co-infections were most common among enterovirus and parainfluenza virus 4 (Fig. 2) . We also compared and contrasted the clinical and community results. Parainfluenza virus 4, respiratory syncytial virus B and enterovirus were the most common viruses found in the clinical sample. These three infections resulted in 41 viruses detected in 15 subjects clinically, and eight infections detected in five people in the community. Together they infected 94% (15/16, 95% CI 67.7-99.7%) of clinical subjects, and 7% (5/75, 95% CI 2.5-15.5%) in the community (significant difference, p < 0.001). The most common virus detected in community samples was Coronavirus OC43; this virus was detected in 13.3% (95% CI 6.9-23.6%) people in the community and not in any of the clinical samples. However a different strain, coronavirus OC 229 E/NL63 was detected in 12.5% of the clinical subjects (2/16, 95% CI 2.2-39.6%) and not detected in the community. Double, triple and quadruple infections were another common feature of note. We identified ten different respiratory tract viruses among the subjects as shown in Fig. 1 . Samples collected from the Children's specialist hospital showed 100% prevalence rate of infection with one or more viruses. This might not be surprising, as the basic difference between the community and clinic samples was an increased severity of illness in the clinical sample. This may also explain the high level of co-infection found among the clinical subjects. The most prevalent virus in the clinical sample (coronavirus OC43) was not detected in the community sample. Further, there was a significant difference between prevalence of the most common viruses in the clinical sample (parainfluenza virus 4, respiratory syncytial virus B and enterovirus) and their prevalence in the community. Finally, some of the viruses detected in this study have not been detected and implicated with ARIs in Nigeria. There is no report, to the best of our knowledge, implicating coronavirus in ARIs in Nigeria, and it was detected in 12 subjects in this study. Although cases of double and triple infections were observed in a study in Nigeria in 2011 [28] , as far as we are aware, reports of quadruple infections are rare and have not been reported in Nigeria previously. Due to the unique nature of the data generated in this study and novelty of work in the setting, it is not possible to exactly compare results to other studies. For example, though we found a similar study regarding ARIs in clinical subjects in Burkina Faso [27] , due to the small sample size from this study it would not be feasible to infer or compare prevalence rates. Studies of ARI etiology have mostly been generally focused in areas of the world that are more developed [29] , and it is important to note that the availability of molecular diagnostic methods as employed in this study substantially improve the ability to detect viruses which hitherto have not been detected in Nigeria. Further, findings from this work also add to the growing body of research that shows value of community-data in infectious disease surveillance [8] . As most of the work to-date has been in higher resource areas of the world this study adds perspective from an area where healthcare resources are lower. In conclusion, results of this study provide evidence for active community surveillance to enhance public health surveillance and scientific understanding of ARIs. This is not only because a minority of children with severe infection are admitted to the hospital in areas such this in Nigeria, but also findings from this pilot study which indicate that viral circulation in the community may not get detected clinically [29] . This pilot study indicates that in areas of Nigeria, etiology of ARIs ascertained from clinical samples may not represent all of the ARIs circulating in the community. The main limitation of the study is the sample size. In particular, the sample is not equally representative across all ages. However, the sample size was big enough to ascertain significant differences in community and clinic sourced viruses, and provides a qualitative understanding of viral etiology in samples from the community and clinic. Moreover, the sample was largely concentrated on subjects under 6 years, who are amongst the groups at highest risk of ARIs. Despite the small sample size, samples here indicate that circulation patterns in the community may differ from those in the clinic. In addition, this study resulted in unique findings Given that resources are limited for research and practice, we hope these pilot results may motivate further systematic investigations into how community-generated data can best be used in ARI surveillance. Results of this study can inform a larger study, representative across demographic and locations to systematically assess the etiology of infection and differences in clinical and community cohorts. A larger study will also enable accounting for potential confounders such as environmental risk factors. Finally, while it may be intuitive, findings from this pilot study shed light on the scope of differences in ARI patterns including different types and strains of circulating viruses. Also, because PCR was used for viral detection, the study was limited to detection of viruses in the primer sets. Given that these are the most up-to-date and common viruses, this approach was deemed sufficient for this initial investigation. The study was conceived by RC and OK. RC and OK, MO and TD were involved in the design of the study, which was conducted by MO and TD. RC and OK analyzed the data. RC and OK wrote and revised the manuscript. All authors read and approved the final manuscript.

How can countries enhance public health surveillance?

Bioinformatics analysis of rabbit haemorrhagic disease virus genome https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3377956/ SHA: eff26d8739498efca2d32fe2e66cdbebf0569c50 Authors: Tian, Xiao-ting; Li, Bao-yu; Zhang, Liang; Jiao, Wen-qiang; Liu, Ji-xing Date: 2011-11-01 DOI: 10.1186/1743-422x-8-494 License: cc-by Abstract: BACKGROUND: Rabbit haemorrhagic disease virus (RHDV), as the pathogeny of Rabbit haemorrhagic disease, can cause a highly infectious and often fatal disease only affecting wild and domestic rabbits. Recent researches revealed that it, as one number of the Caliciviridae, has some specialties in its genome, its reproduction and so on. RESULTS: In this report, we firstly analyzed its genome and two open reading frameworks (ORFs) from this aspect of codon usage bias. Our researches indicated that mutation pressure rather than natural is the most important determinant in RHDV with high codon bias, and the codon usage bias is nearly contrary between ORF1 and ORF2, which is maybe one of factors regulating the expression of VP60 (encoding by ORF1) and VP10 (encoding by ORF2). Furthermore, negative selective constraints on the RHDV whole genome implied that VP10 played an important role in RHDV lifecycle. CONCLUSIONS: We conjectured that VP10 might be beneficial for the replication, release or both of virus by inducing infected cell apoptosis initiate by RHDV. According to the results of the principal component analysis for ORF2 of RSCU, we firstly separated 30 RHDV into two genotypes, and the ENC values indicated ORF1 and ORF2 were independent among the evolution of RHDV. Text: Synonymous codons are not used randomly [1] . The variation of codon usage among ORFs in different organisms is accounted by mutational pressure and translational selection as two main factors [2, 3] . Levels and causes of codon usage bias are available to understand viral evolution and the interplay between viruses and the immune response [4] . Thus, many organisms such as bacteria, yeast, Drosophila, and mammals, have been studied in great detail up on codon usage bias and nucleotide composition [5] . However, same researches in viruses, especially in animal viruses, have been less studied. It has been observed that codon usage bias in human RNA viruses is related to mutational pressure, G +C content, the segmented nature of the genome and the route of transmission of the virus [6] . For some vertebrate DNA viruses, genome-wide mutational pressure is regarded as the main determinant of codon usage rather than natural selection for specific coding triplets [4] . Analysis of the bovine papillomavirus type 1 (BPV1) late genes has revealed a relationship between codon usage and tRNA availability [7] . In the mammalian papillomaviruses, it has been proposed that differences from the average codon usage frequencies in the host genome strongly influence both viral replication and gene expression [8] . Codon usage may play a key role in regulating latent versus productive infection in Epstein-Barr virus [9] . Recently, it was reported that codon usage is an important driving force in the evolution of astroviruses and small DNA viruses [10, 11] . Clearly, studies of synonymous codon usage in viruses can reveal much about the molecular evolution of viruses or individual genes. Such information would be relevant in understanding the regulation of viral gene expression. Up to now, little codon usage analysis has been performed on Rabbit haemorrhagic disease virus (RHDV), which is the pathogen causing Rabbit haemorrhagic disease (RHD), also known as rabbit calicivirus disease (RCD) or viral haemorrhagic disease (VHD), a highly infectious and often fatal disease that affects wild and domestic rabbits. Although the virus infects only rabbits, RHD continues to cause serious problems in different parts of the world. RHDV is a single positive stranded RNA virus without envelope, which contains two open reading frames (ORFs) separately encoding a predicted polyprotein and a minor structural protein named VP10 [12] . After the hydrolysis of self-coding 3C-like cysteinase, the polyprotein was finally hydrolyzed into 8 cleavage products including 7 nonstructural proteins and 1 structural protein named as VP60 [13, 14] . Studies on the phylogenetic relationship of RHDVs showed only one serotype had been isolated, and no genotyping for RHDV was reported. It reported that the VP10 was translated with an efficiency of 20% of the preceding ORF1 [15] . In order to better understand the characteristics of the RHDV genome and to reveal more information about the viral genome, we have analyzed the codon usage and dinucleotide composition. In this report, we sought to address the following issues concerning codon usage in RHDV: (i) the extent and causes of codon bias in RHDV; (ii) A possible genotyping of RHDV; (iii) Codon usage bias as a factor reducing the expression of VP10 and (iiii) the evolution of the ORFs. The 30 available complete RNA sequences of RHDV were obtained from GenBank randomly in January 2011. The serial number (SN), collection dates, isolated areas and GenBank accession numbers are listed in Table 1 . To investigate the characteristics of synonymous codon usage without the influence of amino acid composition, RSCU values of each codon in a ORF of RHDV were calculated according to previous reports (2 Sharp, Tuohy et al. 1986 ) as the followed formula: Where g ij is the observed number of the ith codon for jth amino acid which has n i type of synonymous codons. The codons with RSCU value higher than 1.0 have positive codon usage bias, while codons with value lower than 1.0 has relative negative codon usage bias. As RSCU values of some codons are nearly equal to 1.0, it means that these codons are chosen equally and randomly. The index GC3s means the fraction of the nucleotides G+C at the synonymous third codon position, excluding Met, Trp, and the termination codons. The ENC, as the best estimator of absolute synonymous codon usage bias [16] , was calculated for the quantification of the codon usage bias of each ORF [17] . The predicted values of ENC were calculated as ENC = 2 + s + 29 where s represents the given (G+C) 3 % value. The values of ENC can also be obtained by EMBOSS CHIPS program [18] . Analyses were conducted with the Nei-Gojobori model [19] , involving 30 nucleotide sequences. All positions containing gaps and missing data were eliminated. The values of dn, ds and ω (dn/ds) were calculated in MEGA4.0 [20] . Multivariate statistical analysis can be used to explore the relationships between variables and samples. In this study, correspondence analysis was used to investigate the major trend in codon usage variation among ORFs. In this study, the complete coding region of each ORF was represented as a 59 dimensional vector, and each dimension corresponds to the RSCU value of one sense codon (excluding Met, Trp, and the termination codons) [21] . Correlation analysis was used to identify the relationship between nucleotide composition and synonymous codon usage pattern [22] . This analysis was implemented based on the Spearman's rank correlation analysis way. All statistical processes were carried out by with statistical software SPSS 17.0 for windows. The values of nucleotide contents in complete coding region of all 30 RHDV genomes were analyzed and listed in Table 2 and Table 3 . Evidently, (C+G)% content of the ORF1 fluctuated from 50.889 to 51.557 with a mean value of 51.14557, and (C+G)% content of the ORF2 were ranged from 35.593 to 40.113 with a mean value of 37.6624, which were indicating that nucleotides A and U were the major elements of ORF2 against ORF1. Comparing the values of A 3 %, U 3 %, C 3 % and G 3 %, it is clear that C 3 % was distinctly high and A 3 % was the lowest of all in ORF1 of RHDV, while U 3 % was distinctly high and C 3 % was the lowest of all in ORF2 of Table 2 Identified nucleotide contents in complete coding region (length > 250 bps) in the ORF1 of RHDV (30 isolates) genome Table 4 . Most preferentially used codons in ORF1 were C-ended or G-ended codons except Ala, Pro and Ser, however, A-ended or G-ended codons were preferred as the content of ORF2. In addition, the dn, ds and ω(dN/dS) values of ORF1 were separately 0.014, 0.338 and 0.041, and the values of ORF2 were 0.034, 0.103 and 0.034, respectively. The ω values of two ORFs in RHDV genome are generally low, indicating that the RHDV whole genome is subject to relatively strong selective constraints. COA was used to investigate the major trend in codon usage variation between two ORFs of all 30 RHDV selected for this study. After COA for RHDV Genome, one major trend in the first axis (f' 1 ) which accounted for 42.967% of the total variation, and another major trend in the second axis (f' 2 ) which accounted for 3.632% of the total variation. The coordinate of the complete coding region of each ORF was plotted in Figure 1 defining by the first and second principal axes. It is clear that coordinate of each ORF is relatively isolated. Interestingly, we found that relatively isolated spots from ORF2 tend to cluster into two groups: the ordinate value of one group (marked as Group 1) is To estimate whether the evolution of RHDV genome on codon usage was regulated by mutation pressure or natural selection, the A%, U%, C%, G% and (C+G)% were compared with A 3 %, U 3 %, C 3 %, G 3 % and (C 3 +G 3 )%, respectively (Table 5 ). There is a complex correlation among nucleotide compositions. In detail, A 3 %, U 3 %, C 3 % and G 3 % have a significant negative correlation with G%, C%, U% and A% and positive correlation with A%, U%, C% and G%, respectively. It suggests that nucleotide constraint may influence synonymous codon usage patterns. However, A 3 % has non-correlation with U% and C%, and U 3 % has noncorrelation with A% and G%, respectively, which haven't indicated any peculiarity about synonymous codon usage. Furthermore, C 3 % and G 3 % have non-correlation with A%, G% and U%, C%, respectively, indicating these data don't reflect the true feature of synonymous codon usage as well. Therefore, linear regression analysis was implemented to analyze the correlation between synonymous codon usage bias and nucleotide compositions. Details of correlation analysis between the first two principle axes (f' 1 and f' 2 ) of each RHDV genome in COA and nucleotide contents were listed in Table 6 . In surprise, only f2 values are closely related to base nucleotide A and G content on the third codon position only, suggesting that nucleotide A and G is a factor influencing the synonymous codon usage pattern of RHDV genome. However, f' 1 value has non-correlation with base nucleotide contents on the third codon position; it is observably suggest that codon usage patterns in RHDV were probably influenced by other factors, such as the second structure of viral genome and limits of host. In spite of that, compositional constraint is a factor shaping the pattern of synonymous codon usage in RHDV genome. Figure 1 A plot of value of the first and second axis of RHDV genome in COA. The first axis (f' 1 ) accounts for 42.967% of the total variation, and the second axis (f' 2 ) accounts for 3.632% of the total variation. Table 5 Summary of correlation analysis between the A, U, C, G contents and A 3 , U 3 , C 3 , G 3 contents in all selected samples There have been more and more features that are unique to RHDV within the family Caliciviridae, including its single host tropism, its genome and its VP10 as a structural protein with unknown function. After we analyzed synonymous codon usage in RHDV (Table 2) , we obtained several conclusions and conjectures as followed. 4.1 Mutational bias as a main factor leading to synonymous codon usage variation ENC-plot, as a general strategy, was utilized to investigate patterns of synonymous codon usage. The ENC-plots of ORFs constrained only by a C 3 +G 3 composition will lie on or just below the curve of the predicted values [18] . ENC values of RHDV genomes were plotted against its corresponding (C 3 +G 3 ) %. All of the spots lie below the curve of the predicted values, as shown in Figure 2 , suggesting that the codon usage bias in all these 30 RHDV genomes is principally influenced by the mutational bias. As we know, the efficiency of gene expression is influenced by regulator sequences or elements and codon usage bias. It reported that the RNA sequence of the 3terminal 84 nucleotides of ORF1were found to be crucial for VP10 expression instead of the encoded peptide. VP10 coding by ORF2 has been reported as a low expressive structural protein against VP60 coding by ORF1 [5] . And its efficiency of translation is only 20% of VP60. According to results showed by Table 4 , it revealed the differences in codon usage patterns of two ORFs, which is a possible factor reducing the expression of VP10. Although VP10 encoded by ORF2, as a minor structural protein with unknown functions, has been described by LIU as a nonessential protein for virus infectivity, the ω Figure 2 Effective number of codons used in each ORF plotted against the GC3s. The continuous curve plots the relationship between GC3s and ENC in the absence of selection. All of spots lie below the expected curve. value of ORF2 suggests VP10 plays an important role in the certain stage of whole RHDV lifecycle. After combining with low expression and ω value of VP10, we conjectured that VP10 might be beneficial for the replication, release or both of virus by inducing infected cell apoptosis initiate by RHDV. This mechanism has been confirmed in various positive-chain RNA viruses, including coxsackievirus, dengue virus, equine arterivirus, footand-mouth disease virus, hepatitis C virus, poliovirus, rhinovirus, and severe acute respiratory syndrome [23] [24] [25] [26] [27] [28] [29] , although the details remain elusive. As preceding description, ENC reflects the evolution of codon usage variation and nucleotide composition to some degree. After the correlation analysis of ENC values between ORF1 and ORF2 (Table 7) , the related coefficient of ENC values of two ORFs is 0.230, and p value is 0.222 more than 0.05. These data revealed that no correlation existed in ENC values of two ORFs, indicating that codon usage patterns and evolution of two ORFs are separated each other. Further, this information maybe helps us well understand why RSCU and ENC between two ORFs are quite different. Interestingly, we found that relatively isolated spots from ORF2 tend to cluster into two groups: the ordinate value of one group (marked as Group 1) is positive value and the other one (marked as Group 2) is negative value. And all of those strains isolated before 2000 belonged to Group 2, including Italy-90, RHDV-V351, RHDV-FRG, BS89, RHDV-SD and M67473.1. Although RHDV has been reported as only one type, this may be a reference on dividing into two genotypes. In this report, we firstly analyzed its genome and two open reading frameworks (ORFs) from this aspect of codon usage bias. Our researches indicated that mutation pressure rather than natural is the most important determinant in RHDV with high codon bias, and the codon usage bias is nearly contrary between ORF1 and ORF2, which is maybe one of factors regulating the expression of VP60 (encoding by ORF1) and VP10 (encoding by ORF2). Furthermore, negative selective constraints on the RHDV whole genome implied that VP10 played an important role in RHDV lifecycle. We conjectured that VP10 might be beneficial for the replication, release or both of virus by inducing infected cell apoptosis initiate by RHDV. According to the results of the principal component analysis for ORF2 of RSCU, we firstly separated 30 RHDV into two genotypes, and the ENC values indicated ORF1 and ORF2 were independent among the evolution of RHDV. All the results will guide the next researches on the RHDV as a reference.

What factor may influence viral replication and gene expression?

Bioinformatics analysis of rabbit haemorrhagic disease virus genome https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3377956/ SHA: eff26d8739498efca2d32fe2e66cdbebf0569c50 Authors: Tian, Xiao-ting; Li, Bao-yu; Zhang, Liang; Jiao, Wen-qiang; Liu, Ji-xing Date: 2011-11-01 DOI: 10.1186/1743-422x-8-494 License: cc-by Abstract: BACKGROUND: Rabbit haemorrhagic disease virus (RHDV), as the pathogeny of Rabbit haemorrhagic disease, can cause a highly infectious and often fatal disease only affecting wild and domestic rabbits. Recent researches revealed that it, as one number of the Caliciviridae, has some specialties in its genome, its reproduction and so on. RESULTS: In this report, we firstly analyzed its genome and two open reading frameworks (ORFs) from this aspect of codon usage bias. Our researches indicated that mutation pressure rather than natural is the most important determinant in RHDV with high codon bias, and the codon usage bias is nearly contrary between ORF1 and ORF2, which is maybe one of factors regulating the expression of VP60 (encoding by ORF1) and VP10 (encoding by ORF2). Furthermore, negative selective constraints on the RHDV whole genome implied that VP10 played an important role in RHDV lifecycle. CONCLUSIONS: We conjectured that VP10 might be beneficial for the replication, release or both of virus by inducing infected cell apoptosis initiate by RHDV. According to the results of the principal component analysis for ORF2 of RSCU, we firstly separated 30 RHDV into two genotypes, and the ENC values indicated ORF1 and ORF2 were independent among the evolution of RHDV. Text: Synonymous codons are not used randomly [1] . The variation of codon usage among ORFs in different organisms is accounted by mutational pressure and translational selection as two main factors [2, 3] . Levels and causes of codon usage bias are available to understand viral evolution and the interplay between viruses and the immune response [4] . Thus, many organisms such as bacteria, yeast, Drosophila, and mammals, have been studied in great detail up on codon usage bias and nucleotide composition [5] . However, same researches in viruses, especially in animal viruses, have been less studied. It has been observed that codon usage bias in human RNA viruses is related to mutational pressure, G +C content, the segmented nature of the genome and the route of transmission of the virus [6] . For some vertebrate DNA viruses, genome-wide mutational pressure is regarded as the main determinant of codon usage rather than natural selection for specific coding triplets [4] . Analysis of the bovine papillomavirus type 1 (BPV1) late genes has revealed a relationship between codon usage and tRNA availability [7] . In the mammalian papillomaviruses, it has been proposed that differences from the average codon usage frequencies in the host genome strongly influence both viral replication and gene expression [8] . Codon usage may play a key role in regulating latent versus productive infection in Epstein-Barr virus [9] . Recently, it was reported that codon usage is an important driving force in the evolution of astroviruses and small DNA viruses [10, 11] . Clearly, studies of synonymous codon usage in viruses can reveal much about the molecular evolution of viruses or individual genes. Such information would be relevant in understanding the regulation of viral gene expression. Up to now, little codon usage analysis has been performed on Rabbit haemorrhagic disease virus (RHDV), which is the pathogen causing Rabbit haemorrhagic disease (RHD), also known as rabbit calicivirus disease (RCD) or viral haemorrhagic disease (VHD), a highly infectious and often fatal disease that affects wild and domestic rabbits. Although the virus infects only rabbits, RHD continues to cause serious problems in different parts of the world. RHDV is a single positive stranded RNA virus without envelope, which contains two open reading frames (ORFs) separately encoding a predicted polyprotein and a minor structural protein named VP10 [12] . After the hydrolysis of self-coding 3C-like cysteinase, the polyprotein was finally hydrolyzed into 8 cleavage products including 7 nonstructural proteins and 1 structural protein named as VP60 [13, 14] . Studies on the phylogenetic relationship of RHDVs showed only one serotype had been isolated, and no genotyping for RHDV was reported. It reported that the VP10 was translated with an efficiency of 20% of the preceding ORF1 [15] . In order to better understand the characteristics of the RHDV genome and to reveal more information about the viral genome, we have analyzed the codon usage and dinucleotide composition. In this report, we sought to address the following issues concerning codon usage in RHDV: (i) the extent and causes of codon bias in RHDV; (ii) A possible genotyping of RHDV; (iii) Codon usage bias as a factor reducing the expression of VP10 and (iiii) the evolution of the ORFs. The 30 available complete RNA sequences of RHDV were obtained from GenBank randomly in January 2011. The serial number (SN), collection dates, isolated areas and GenBank accession numbers are listed in Table 1 . To investigate the characteristics of synonymous codon usage without the influence of amino acid composition, RSCU values of each codon in a ORF of RHDV were calculated according to previous reports (2 Sharp, Tuohy et al. 1986 ) as the followed formula: Where g ij is the observed number of the ith codon for jth amino acid which has n i type of synonymous codons. The codons with RSCU value higher than 1.0 have positive codon usage bias, while codons with value lower than 1.0 has relative negative codon usage bias. As RSCU values of some codons are nearly equal to 1.0, it means that these codons are chosen equally and randomly. The index GC3s means the fraction of the nucleotides G+C at the synonymous third codon position, excluding Met, Trp, and the termination codons. The ENC, as the best estimator of absolute synonymous codon usage bias [16] , was calculated for the quantification of the codon usage bias of each ORF [17] . The predicted values of ENC were calculated as ENC = 2 + s + 29 where s represents the given (G+C) 3 % value. The values of ENC can also be obtained by EMBOSS CHIPS program [18] . Analyses were conducted with the Nei-Gojobori model [19] , involving 30 nucleotide sequences. All positions containing gaps and missing data were eliminated. The values of dn, ds and ω (dn/ds) were calculated in MEGA4.0 [20] . Multivariate statistical analysis can be used to explore the relationships between variables and samples. In this study, correspondence analysis was used to investigate the major trend in codon usage variation among ORFs. In this study, the complete coding region of each ORF was represented as a 59 dimensional vector, and each dimension corresponds to the RSCU value of one sense codon (excluding Met, Trp, and the termination codons) [21] . Correlation analysis was used to identify the relationship between nucleotide composition and synonymous codon usage pattern [22] . This analysis was implemented based on the Spearman's rank correlation analysis way. All statistical processes were carried out by with statistical software SPSS 17.0 for windows. The values of nucleotide contents in complete coding region of all 30 RHDV genomes were analyzed and listed in Table 2 and Table 3 . Evidently, (C+G)% content of the ORF1 fluctuated from 50.889 to 51.557 with a mean value of 51.14557, and (C+G)% content of the ORF2 were ranged from 35.593 to 40.113 with a mean value of 37.6624, which were indicating that nucleotides A and U were the major elements of ORF2 against ORF1. Comparing the values of A 3 %, U 3 %, C 3 % and G 3 %, it is clear that C 3 % was distinctly high and A 3 % was the lowest of all in ORF1 of RHDV, while U 3 % was distinctly high and C 3 % was the lowest of all in ORF2 of Table 2 Identified nucleotide contents in complete coding region (length > 250 bps) in the ORF1 of RHDV (30 isolates) genome Table 4 . Most preferentially used codons in ORF1 were C-ended or G-ended codons except Ala, Pro and Ser, however, A-ended or G-ended codons were preferred as the content of ORF2. In addition, the dn, ds and ω(dN/dS) values of ORF1 were separately 0.014, 0.338 and 0.041, and the values of ORF2 were 0.034, 0.103 and 0.034, respectively. The ω values of two ORFs in RHDV genome are generally low, indicating that the RHDV whole genome is subject to relatively strong selective constraints. COA was used to investigate the major trend in codon usage variation between two ORFs of all 30 RHDV selected for this study. After COA for RHDV Genome, one major trend in the first axis (f' 1 ) which accounted for 42.967% of the total variation, and another major trend in the second axis (f' 2 ) which accounted for 3.632% of the total variation. The coordinate of the complete coding region of each ORF was plotted in Figure 1 defining by the first and second principal axes. It is clear that coordinate of each ORF is relatively isolated. Interestingly, we found that relatively isolated spots from ORF2 tend to cluster into two groups: the ordinate value of one group (marked as Group 1) is To estimate whether the evolution of RHDV genome on codon usage was regulated by mutation pressure or natural selection, the A%, U%, C%, G% and (C+G)% were compared with A 3 %, U 3 %, C 3 %, G 3 % and (C 3 +G 3 )%, respectively (Table 5 ). There is a complex correlation among nucleotide compositions. In detail, A 3 %, U 3 %, C 3 % and G 3 % have a significant negative correlation with G%, C%, U% and A% and positive correlation with A%, U%, C% and G%, respectively. It suggests that nucleotide constraint may influence synonymous codon usage patterns. However, A 3 % has non-correlation with U% and C%, and U 3 % has noncorrelation with A% and G%, respectively, which haven't indicated any peculiarity about synonymous codon usage. Furthermore, C 3 % and G 3 % have non-correlation with A%, G% and U%, C%, respectively, indicating these data don't reflect the true feature of synonymous codon usage as well. Therefore, linear regression analysis was implemented to analyze the correlation between synonymous codon usage bias and nucleotide compositions. Details of correlation analysis between the first two principle axes (f' 1 and f' 2 ) of each RHDV genome in COA and nucleotide contents were listed in Table 6 . In surprise, only f2 values are closely related to base nucleotide A and G content on the third codon position only, suggesting that nucleotide A and G is a factor influencing the synonymous codon usage pattern of RHDV genome. However, f' 1 value has non-correlation with base nucleotide contents on the third codon position; it is observably suggest that codon usage patterns in RHDV were probably influenced by other factors, such as the second structure of viral genome and limits of host. In spite of that, compositional constraint is a factor shaping the pattern of synonymous codon usage in RHDV genome. Figure 1 A plot of value of the first and second axis of RHDV genome in COA. The first axis (f' 1 ) accounts for 42.967% of the total variation, and the second axis (f' 2 ) accounts for 3.632% of the total variation. Table 5 Summary of correlation analysis between the A, U, C, G contents and A 3 , U 3 , C 3 , G 3 contents in all selected samples There have been more and more features that are unique to RHDV within the family Caliciviridae, including its single host tropism, its genome and its VP10 as a structural protein with unknown function. After we analyzed synonymous codon usage in RHDV (Table 2) , we obtained several conclusions and conjectures as followed. 4.1 Mutational bias as a main factor leading to synonymous codon usage variation ENC-plot, as a general strategy, was utilized to investigate patterns of synonymous codon usage. The ENC-plots of ORFs constrained only by a C 3 +G 3 composition will lie on or just below the curve of the predicted values [18] . ENC values of RHDV genomes were plotted against its corresponding (C 3 +G 3 ) %. All of the spots lie below the curve of the predicted values, as shown in Figure 2 , suggesting that the codon usage bias in all these 30 RHDV genomes is principally influenced by the mutational bias. As we know, the efficiency of gene expression is influenced by regulator sequences or elements and codon usage bias. It reported that the RNA sequence of the 3terminal 84 nucleotides of ORF1were found to be crucial for VP10 expression instead of the encoded peptide. VP10 coding by ORF2 has been reported as a low expressive structural protein against VP60 coding by ORF1 [5] . And its efficiency of translation is only 20% of VP60. According to results showed by Table 4 , it revealed the differences in codon usage patterns of two ORFs, which is a possible factor reducing the expression of VP10. Although VP10 encoded by ORF2, as a minor structural protein with unknown functions, has been described by LIU as a nonessential protein for virus infectivity, the ω Figure 2 Effective number of codons used in each ORF plotted against the GC3s. The continuous curve plots the relationship between GC3s and ENC in the absence of selection. All of spots lie below the expected curve. value of ORF2 suggests VP10 plays an important role in the certain stage of whole RHDV lifecycle. After combining with low expression and ω value of VP10, we conjectured that VP10 might be beneficial for the replication, release or both of virus by inducing infected cell apoptosis initiate by RHDV. This mechanism has been confirmed in various positive-chain RNA viruses, including coxsackievirus, dengue virus, equine arterivirus, footand-mouth disease virus, hepatitis C virus, poliovirus, rhinovirus, and severe acute respiratory syndrome [23] [24] [25] [26] [27] [28] [29] , although the details remain elusive. As preceding description, ENC reflects the evolution of codon usage variation and nucleotide composition to some degree. After the correlation analysis of ENC values between ORF1 and ORF2 (Table 7) , the related coefficient of ENC values of two ORFs is 0.230, and p value is 0.222 more than 0.05. These data revealed that no correlation existed in ENC values of two ORFs, indicating that codon usage patterns and evolution of two ORFs are separated each other. Further, this information maybe helps us well understand why RSCU and ENC between two ORFs are quite different. Interestingly, we found that relatively isolated spots from ORF2 tend to cluster into two groups: the ordinate value of one group (marked as Group 1) is positive value and the other one (marked as Group 2) is negative value. And all of those strains isolated before 2000 belonged to Group 2, including Italy-90, RHDV-V351, RHDV-FRG, BS89, RHDV-SD and M67473.1. Although RHDV has been reported as only one type, this may be a reference on dividing into two genotypes. In this report, we firstly analyzed its genome and two open reading frameworks (ORFs) from this aspect of codon usage bias. Our researches indicated that mutation pressure rather than natural is the most important determinant in RHDV with high codon bias, and the codon usage bias is nearly contrary between ORF1 and ORF2, which is maybe one of factors regulating the expression of VP60 (encoding by ORF1) and VP10 (encoding by ORF2). Furthermore, negative selective constraints on the RHDV whole genome implied that VP10 played an important role in RHDV lifecycle. We conjectured that VP10 might be beneficial for the replication, release or both of virus by inducing infected cell apoptosis initiate by RHDV. According to the results of the principal component analysis for ORF2 of RSCU, we firstly separated 30 RHDV into two genotypes, and the ENC values indicated ORF1 and ORF2 were independent among the evolution of RHDV. All the results will guide the next researches on the RHDV as a reference.

What accounts for the variation of codon usage among open reading frameworks?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

How many deaths each year are caused by gastroenteritis?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What percentage of sporadic diarrhea are caused by norovirus?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What is the most common cause of viral gastroenteritis in children?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

Which types of adenovirus are associated with diarrhea?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

When was the first tissue culture system developed?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What are the most common DNA-based techniques for detecting viruses?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What is Universal primer-PCR used for in viral studies?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What is Koch's first postulate?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What is Koch's second postulate?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

What is Koch's third postulate?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

If all 3 of Koch's postulates are met, what does this indicate?

Viruses Causing Gastroenteritis: The Known, The New and Those Beyond https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4776197/ SHA: f7b30ee89775bc82607cc6bc87feb5934b47625f Authors: Oude Munnink, Bas B.; van der Hoek, Lia Date: 2016-02-19 DOI: 10.3390/v8020042 License: cc-by Abstract: The list of recently discovered gastrointestinal viruses is expanding rapidly. Whether these agents are actually involved in a disease such as diarrhea is the essential question, yet difficult to answer. In this review a summary of all viruses found in diarrhea is presented, together with the current knowledge about their connection to disease. Text: The gastrointestinal tract is a vulnerable organ for infections as there is constant contact with the outside, mainly via the oral route. Inflammation of the stomach and the intestines (gastroenteritis) can cause nausea, vomiting and diarrhea. Gastroenteritis is responsible for two to three million deaths each year, making it one of the most common causes of mortality [1] . Mainly children in developing countries, but also immuno-compromised individuals in developed countries, suffer from diarrhea. While bacterial and parasitic gastrointestinal infections are declining as a result of proper disposal of sewage and safe drinking water, viral gastroenteritis is not declining in developing countries [2] . In the developed world, viruses are already the most common pathogens causing diarrhea [3] . Although viruses infecting humans had already been described since 1901 [4] and viruses were suspected to play a role in diarrhea, it lasted until 1972, when the first virus causing gastroenteritis (norovirus) was identified in an outbreak of diarrhea in Norwalk (California, United States) [5] . Shortly after the discovery of norovirus several other viruses causing gastroenteritis were discovered: rotavirus in epithelial cells of children with gastroenteritis [6] , astrovirus in infantile diarrhea cases [7] , enteric adenoviruses in the feces of children with acute diarrhea [8] , and sapovirus during an outbreak of gastroenteritis in an orphanage in Sapporo, Japan [9] . All these viruses spread via the fecal-oral route through person-to-person transmission and are described in more detail below. Noroviruses are part of the family Caliciviridae and outbreaks of norovirus gastroenteritis have been reported in cruise ships, health care settings, schools, and in the military, but norovirus is also responsible for around 60% of all sporadic diarrhea cases (diarrhea cases where an enteropathogen could be found), reviewed in the literature [10, 11] . The pathogenesis of norovirus infection has been tested in vivo. Filtrated norovirus was given to healthy volunteers after which most of them developed diarrhea [12] . Culturing of the virus, however, has been a problem since its discovery, yet one study has recently described the cultivation of norovirus in B cells, and has revealed that co-factors, such as histo-blood antigen expressing enteric bacteria, are probably needed before enteric viruses can be cultured in vitro [13] . Sapoviruses are also members of the Caliciviridae. There are five human genogroups of sapovirus described [14] which account for 2.2%-12.7% of all gastroenteritis cases around the globe [14, 15] . Sapovirus outbreaks occur throughout the year and can be foodborne [16] . For sapoviruses it has been described that the virus was not found before onset of an outbreak, and that it was found in 95% of the patients during an outbreak, while it declined to 50% after an outbreak, indicating that the virus introduces disease in a naturally infected host [17] . Rotavirus infection is the most common cause of viral gastroenteritis among children; however, parents of infected children also often become ill and as a result rotavirus is the second most common cause of gastroenteritis in adults [18] . Studies in human volunteers have shown that infection with rotavirus causes diarrhea, results in shedding of the virus and a rise in antibody anti-virus titer after infection [19] . Additionally, astroviruses infections are common, accounting for about 10% of all sporadic diarrhea cases [20] . Astrovirus has been isolated from diseased people, filtrated and administered to healthy individuals after which in some of the volunteers diarrheal disease was observed and astrovirus was shed in their stools [21] . The virus can replicate in human embryonic kidney cells and was detected by electron microscopy (EM) [21] . Adenoviruses are responsible for around 1.5%-5.4% of the diarrhea cases in children under the age of 2 years, reviewed in the literature [22] . Of the 57 identified adenovirus types [23] , only adenoviruses type 40 and 41 are associated with diarrhea [24] . Next to these two types, adenovirus type 52 can also cause gastroenteritis [25] , although it has been argued whether type 52 is actually a separate type since there is not sufficient distance to adenovirus type 41 [26] . Adenoviruses can generally be propagated in cell lines; however, enteric adenovirus 40/41 are difficult to culture, reviewed in the literature [27] . In the 1980s and 1990s some viral agents were identified for which the direct association with disease is less clear. Aichi viruses are members of the Picornaviridae identified in fecal samples of patients with gastroenteritis [28] . Aichi virus infection has been shown to elicit an immune response [29] . Since their discovery, two case-control studies were performed, but, although both studies only found Aichi virus in stools of diarrheic patients, the prevalence of Aichi virus (0.5% and 1.8%) was too low to find a significant association with diarrhea [30, 31] . In immuno-compromised hosts the virus is found in higher quantities and is not associated with diarrhea [32] . Toroviruses, part of the Coronaviridae, were first identified in 1984 in stools of children and adults with gastroenteritis [33] . Torovirus infection is associated with diarrhea [34] and is more frequently observed in immuno-compromised patients and in nosocomial infected individuals [34] . Retrospective analysis of nosocomial viral gastroenteritis in a pediatric hospital revealed that in 67% of the cases torovirus could be detected [35] . However, only a limited number of studies report the detection of torovirus and therefore the true pathogenesis and prevalence of this virus remains elusive. Picobirnaviruses belong to the Picobirnaviridae and were first detected in the feces of children with gastroenteritis [36] . Since the initial discovery, the virus has been detected in fecal samples of several animal species, and it has been shown that the viruses are genetically highly diverse without a clear species clustering, reviewed in the literature [37] . This high sequence diversity has also been observed within particular outbreaks of gastroenteritis [38, 39] , limiting the likelihood that picobirnaviruses are actually causing outbreaks, as no distinct single source of infection can be identified. In 1907 the first tissue culture system was developed which was regarded as the golden standard for virus detection for a long time, reviewed in the literature [40] . In the 1930's serology and electron microscopy were introduced which boosted the discovery of new viruses. During these years, these methods developed fruitfully but viruses infecting the gastrointestinal tract were especially difficult to culture. Throughout the last several decades, several DNA-based techniques have been developed for virus discovery that boosted the identification of novel viruses in stool samples. The four most used methods are: 1. Universal primer-PCR [41] ; 2. Random priming-based PCR [42] ; 3. Virus Discovery cDNA, Amplified Fragment Length Polymorphism (VIDISCA) [43] ; and 4. Sequence-Independent Single Primer Amplification (SISPA) [44] . Universal primer-PCR is a virus discovery technique that uses universal primers designed on conserved parts of a specific viral family, which can be used to detect novel variants of this viral family. Random priming-based PCR is a technique that randomly amplifies all nucleic acids present in samples, after which the resulting PCR products can be cloned and sequenced. SISPA and VIDISCA are virus discovery techniques that are based on digestion with restriction enzymes, after which adaptors can be ligated. These methods have been successful in the discovery of novel viruses, but there are some limitations. Universal primers are useful for discovering novel viruses of a chosen family, but the primers, based on our present knowledge of the viral family, may not fit on all unknown variants. Random priming PCR, SISPA and VIDISCA are sequence independent amplification techniques. The disadvantage of random priming PCR, SISPA and VIDISCA is that the virus needs to be present at a high concentration, while the host background DNA and/or RNA should be minimal and preferably not complex. In recent years, sequence independent amplification techniques improved considerably by coupling these techniques to next-generation sequencing platforms and as a result several novel viruses have been described in gastroenteritis cases, such as cosavirus [45] , Saffold virus [46] , klassevirus/salivirus [47, 48] , polyomavirus [49] , bufavirus [50] , tusavirus [51] , and recovirus [52] . Although these viruses are found in individuals with diarrhea, for most of them the degree of circulation (prevalence) and the ability to cause morbid conditions or disease (pathogenesis) remains to be determined, as described below (also see Table 1 ). Only found in low prevalence; **: Only limited data is available about this virus; ***: Antibodies against astrovirus HMO-C were observed whereas no antibodies against astrovirus HMO-A were found (HMO = human-mink-ovine-like astrovirus); -No published data available;ˆPicobirnavirus, tusavirus and recovirus were identified in the gastrointestinal tract after next-generation sequencing, but no information regarding antibody response or association with diarrhea is available. In the last decade, two novel clades of astroviruses have been discovered in stool samples from patients with diarrhea that are genetically far distinct from the classical astroviruses. The first clade consists of the VA-1, VA-2, VA-3, VA-4, and VA-5 astroviruses, which are genetically related to feline and porcine astroviruses, while the second clade consists of the MLB1, MLB2 and MLB3 astroviruses and form a separate cluster [55, 57, [74] [75] [76] [77] [78] . For these novel clades the pathogenesis remains to be determined since the viruses have been identified in patients with and without diarrhea, and in some studies the viruses were associated with diarrhea whilst in others no association could be found [55] [56] [57] . In addition an antibody response was observed against some but not all novel astrovirus types [54, 58] . Recently, astrovirus MLB2 has also been detected in blood plasma of a febrile child [79] and astrovirus VA1 in a frontal cortex biopsy specimen from a patient with encephalitis [80] , suggesting that astrovirus infection may not be limited to the gastrointestinal tract. In 2008, Saffold virus was detected in a stool sample from a pediatric patient with fever of unknown origin [46] . Although Saffold virus type 3 was cultured on a human epithelial cervical carcinoma (HeLa) cell line, cytopathic effects were observed and neutralizing antibodies have been found in serum samples [59] , subsequent case-control studies showed that the virus was not significantly associated with diarrhea [53, 60, 61] . Additionally, in 2008 cosavirus was identified in a patient with diarrhea [45] . However, a case-control study showed that this virus was also detected in a substantial amount of individuals without diarrhea and is not associated with diarrhea [32, 62, 63] . Klassevirus/salivirus was identified in 2009 in two fecal samples from infants with gastrointestinal disorders [47, 48] . In two studies the detection of this virus was associated with diarrhea [48, 53] , while in another study no association with disease was found [65] . Serological evidence of human klassevirus infection was obtained, suggesting that the virus infects human cells [64] . With the use of next-generation sequencing techniques, three novel polyomaviruses were also identified in human fecal samples. MW polyomavirus was identified in the stool of a healthy child from Malawi in 2012 [49] , and in the same year MX polyomavirus was found in stool samples of patients with and without diarrhea from Mexico, United States and Chili [68] . One year later, STL polyomavirus was found in the stool of a healthy child from Malawi [71] . An antibody response against MX polyomavirus [66] and MW polyomavirus [69] was observed, although MW polyomavirus [67] and STL polyomavirus [70] were not significantly associated with diarrhea in two independent case-control studies. Bufavirus is a member of the Parvoviridae and was first described in 2012 [50] . Two case-controls in Thailand and in Turkey showed that the virus was only found in patients with diarrhea and not in controls [72, 73] ; however, because of the low prevalence (respectively 0.3% in Thailand and 1.4% in Turkey), no significant association with disease was found. Tusavirus, another recently described member of the Parvoviridae, was identified in the feces of a child from Tunisia with unexplained diarrhea [51] , and thus far this is the only study describing this virus. Recovirus is a novel member of the Caliciviridae and was found in diarrhea samples from Bangladesh [52] . Similar to tusavirus, this is the only study describing this virus thus far. The identification of the above-mentioned novel viruses certainly increased our knowledge about viruses that can be found in the gastrointestinal tract of humans, yet it is unknown how many of these novel viruses are actually enteropathogens. Human stool contains a wide variety of viruses which can be derived from different hosts: Besides genuine human viruses, plant dietary viruses [32, 81] and animal dietary viruses [82] can also be found in human stool, as well as bacteriophages and viruses infecting protozoa [32] . Even viruses derived from other parts of the body can be found in fecal samples, such as the John Cunningham Polyoma virus originating from the kidney ending up in feces via urine [83] , and rhinoviruses [84] , bocaviruses [85] and coronaviruses [86] originating from the respiratory tract and probably swallowed. Furthermore, viruses infecting blood cells such as human immunodeficiency virus (HIV)-1 can also be detected in fecal samples [87] . Therefore, once a novel virus has been identified in human stool samples it is does not indicate that this virus is replicating in human intestinal cells. Koch recognized as early as 1891 that associating the presence of a certain agent with a certain disease is complex, and he therefore postulated guidelines that should be followed before an agent can be classified as a pathogen [88] . His postulates can be summarized in three points: (1) The microbe occurs in every case of the disease in question and under circumstances which can account for the pathological changes and clinical course of the disease; (2) the microbe occurs in no other disease as a fortuitous and nonpathogenic parasite; and (3), after being fully isolated from the body and repeatedly grown in pure culture, the microbe can induce the disease anew. If a microbe has fulfilled these three postulates it can be stated that "the occurrence of the microbe in the disease can no longer be accidental, but in this case no other relation between it and the disease except that the microbe is the cause of the disease can be considered". For enteric viruses, however, these postulates are not applicable. Firstly, the enteric viruses are not easily cultured [89] [90] [91] , and, secondly, prolonged sheading of viral agents and asymptomatic infection have been described [92] , reviewed in the literature [93] . Although attempts have been made to adjust the Koch's postulates specifically for viruses and the current methodologies deployed [94] [95] [96] , fulfilling these postulates is still not feasible on most occasions due to the lack of an efficient cell culture system, difficulties in antigen synthesis and high levels of viral genetic diversity within viral groups, reviewed in the literature [97] . Several approaches have been made to develop a methodology that adds more significance to the discovery of a novel virus. One approach is based on the enrichment of immunogenic viruses before next-generation sequencing by making use of autologous antibody capture prior to sequencing. This method was tested and validated on several fecal samples containing adenovirus, sapovirus and norovirus, and has shown to enrich immunogenic viruses, while plant viruses and bacteriophages were not enriched after antibody capture [98] . Another method to enrich for relevant viruses prior to next-generation sequencing is the so-called virome capture sequencing platform for vertebrate viruses (VirCapSeq-VERT) which uses~2 million probes which cover the genomes of all members of the viral taxa known to infect vertebrates [99] . However, both methods have limitations: For the antibody capture method, viruses need to be present in high viral loads, and convalescent blood, serum or plasma needs to be available. A disadvantage of the VirCapSeq-VERT technique is that completely novel viruses, e.g., viruses from a novel virus family, will not be identified. The most straightforward method to demonstrate association with disease is using case-control studies. In order to perform such studies, matched stool samples have to be collected in case and control groups from the same geographical locations in the same period of the year. Additionally, whereas in recent years case-control studies have been performed using conventional real-time PCRs (RT-PCR), in the future, sequence independent next-generation sequencing techniques can be used for such case-control studies. Since it allows detection of virtually all nucleic acids, next-generation sequencing has several advantages compared to specific RT-PCRs. Next-generation sequencing prevents the necessity to perform numerous RT-PCRs to screen for all viruses suspected to be associated with disease, and novel variants of currently known viral families or novel virus species can be detected which can be particularly beneficial if only few reference genomes are available. The major benefit of such a database is that in the immediate future the most important question can be answered if a novel virus is identified in diarrhea cases: Is the virus likely to cause disease? In conclusion, the long list of viruses identified in the gastrointestinal tract is most probably not final yet. It is to be expected that several novel viruses will be described in the near future, since detection of these agents using the current next-generation sequence technologies is no longer a difficulty. Therefore, adding relevance to the discovery of novel viruses should be the main goal for future studies.

Is Koch's postulate applicable to enteric viruses?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are examples of delivery vectors for commercial anti-Salmonella vaccines?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What can be a factor in using common vectors for the delivery of vaccines?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

Is a pre-existing immune response to commonly used delivery vector an advantage or a disadvantage?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What bacterial delivery vectors have been tested in animal hosts?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

Which bacteial delivery vectors have gained favor for vaccines?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

Why are E Coli and lactic acid are safe choices as delivery vectors for vaccines?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is Listeria?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the advantage of Listeria as a delivery vector for vaccines?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are examples of viral vectors for delivering vaccines?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

Which viral vaccine delivery vector was first licensed?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are examples of attenuated poxvirus vaccine delivery vectors?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the connection between chicken and Salmonella?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

Why are some poxvirus ideally suited as vaccine delivery vectors?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the advantage of adenovirus as vaccine delivery vector?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are important criteria for selecting vaccine delivery vectors?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are important criteria for selecting vaccine delivery vectors?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are important criteria for selecting vaccine delivery vectors?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What happens when a recipient of a vaccine has immune response to the delivery vector?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the effect of host immune response to the delivery vector on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the effect of host immune response to the delivery vector on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is an example of the effect of immunity to the delivery vector on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the effect of host immune response to viral delivery vectors in the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the effect of host immune response to the viral delivery vector on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What is the effect of host immune response to the delivery vector on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are methods to avoid the effect vector immunity on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

What are methods to avoid the effect of vector immune response on the efficacy of vaccination?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

How does cell-mediated immunity to viral delivery vector, reduce the immune response to vaccine?

Pre-existing immunity against vaccine vectors – friend or foe? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3542731/ SHA: f5bdf18567bb3760e1ce05008135f0270badbd5c Authors: Saxena, Manvendra; Van, Thi Thu Hao; Baird, Fiona J.; Coloe, Peter J.; Smooker, Peter M. Date: 2013-01-27 DOI: 10.1099/mic.0.049601-0 License: cc-by Abstract: Over the last century, the successful attenuation of multiple bacterial and viral pathogens has led to an effective, robust and safe form of vaccination. Recently, these vaccines have been evaluated as delivery vectors for heterologous antigens, as a means of simultaneous vaccination against two pathogens. The general consensus from published studies is that these vaccine vectors have the potential to be both safe and efficacious. However, some of the commonly employed vectors, for example Salmonella and adenovirus, often have pre-existing immune responses in the host and this has the potential to modify the subsequent immune response to a vectored antigen. This review examines the literature on this topic, and concludes that for bacterial vectors there can in fact, in some cases, be an enhancement in immunogenicity, typically humoral, while for viral vectors pre-existing immunity is a hindrance for subsequent induction of cell-mediated responses. Text: In the fields of medicine and veterinary medicine, there are numerous live, attenuated bacterial and viral vaccines in use today worldwide. The safety and efficacy of such vaccines is well established and allows further development as vector systems to deliver antigen originating from other pathogens. Various attenuated bacteria, including Escherichia coli, Vibrio cholerae, lactic acid bacteria (LAB), specifically Lactococcus lactis, Mycobacterium, Listeria, Shigella and Salmonella, have been tested for the targeted delivery of heterologous antigens of bacterial, viral and parasitic origin into a variety of animal hosts (Bahey-El-Din et al., 2010; Innocentin et al., 2009; Johnson et al., 2011; Tobias et al., 2008 Tobias et al., , 2010 Tobias & Svennerholm, 2012) . Bacteria such as E. coli and lactic acid bacteria have recently gained favour, as E. coli is a commensal and lactic acid bacteria are present in most fermented food items and are therefore naturally present in the host. They are also a much safer option than traditional attenuated vaccines in children and immunecompromised people. As this review discusses the effects of pre-existing immune responses to attenuated vaccines, further discussion of LAB and E. coli as potential vectors will not be undertaken; however, the reader is directed to several interesting reviews (Bermú dez-Humarán et al., 2011; Wells & Mercenier, 2008) . Intracellular bacteria from the genera Mycobacterium (Guleria et al., 1996) , Listeria (Gentschev et al., 2001) , Shigella (Levine et al., 1997) and Salmonella (Dougan et al., 1987) are considered to be suitable candidates for the delivery of vaccine antigens due to their capability to induce robust T cell immune responses (Alderton et al., 1991; Lo et al., 1999; Mastroeni et al., 2001; Mittrücker & Kaufmann, 2000; Nauciel, 1990) . Salmonella is one genus that has been well examined as a vector, building on the extensive research available on the micro-organism's physiology and pathogenesis (Basso et al., 2000; Killeen & DiRita, 2000; Sirard et al., 1999; Ward et al., 1999) . There exist several commercial vaccines that are used as anti-Salmonella vaccines in humans and animals (e.g. Ty21a for typhoid fever in humans, several Salmonella serovars against salmonellosis in chickens and other animals). The general strategy for vectoring heterologous antigen is depicted in Fig. 1 . The first clinical trial of a recombinant, which was conducted over 20 years ago using an attenuated Salmonella as a delivery vector, led to the widespread testing of this bacterium as a mucosal delivery system for antigens from non-Salmonella pathogens (Dougan et al., 1987) . These studies have demonstrated the utility of live bacteria to deliver expressed antigens and DNA vaccines to the host immune system (Atkins et al., 2006; Husseiny & Hensel, 2008; Jiang et al., 2004; Kirby et al., 2004) . Since then several other intracellular bacterial vectors have been successfully tested for their capability to deliver a variety of antigens from various pathogens, as well as vaccination against cancer. One genus which has been widely tested as vector is Listeria. Listeria species are Gram-positive intracellular food-borne pathogens. The advantages of Listeria are that it can invade a variety of cells, including antigen presenting cells (APCs). After invading the host cell, Listeria resides inside the phagosome; however, it can escape the phagosome with the help of listeriolysin O (LLO; Hly) and reside in the cytoplasm of the cells, thereby efficiently presenting antigen to both CD8 and CD4 T cells (Cossart & Mengaud, 1989; Kaufmann, 1993; Pamer et al., 1997) . Several studies have demonstrated the effectiveness and ease of using Listeria monocytogenes to deliver heterologous vaccine antigens and DNA vaccines Jensen et al., 1997; Johnson et al., 2011; Peters et al., 2003; Shen et al., 1995; Yin et al., 2011) . Similarly, various viral vectors have been successfully tested for their capability to deliver heterologous vaccine antigens, and this generally results in the induction of strong CTL immune responses. In the veterinary field, there are numerous viral vector vaccines that are currently licensed for use in livestock and domesticated animals. These recombinant vaccines are based on both DNA viruses (such as fowlpox virus-based vaccines which target avian influenza virus and fowlpox virus, or vaccinia virusbased vectors against the rabies virus in wildlife) and RNA viruses [such as Newcastle disease virus-based vaccines to be used in poultry or yellow fever virus (YFV)-based vaccines to be used in horses against West Nile virus] (Draper & Heeney, 2010) . Based on the safety record in the veterinary field, many viruses have been studied for human use as a vector in vaccine development (Beukema et al., 2006; Esteban, 2009; Schirrmacher & Fournier, 2009; Stoyanov et al., 2010; Weli & Tryland, 2011) . Amongst them, YFV (YF-17D strain) was the first to be licensed for use in humans, where the cDNAs encoding the envelope proteins of YFV were replaced with the corresponding genes of an attenuated Japanese encephalitis virus strain, SA14-14-2 (Appaiahgari & Vrati, 2010; Rollier et al., 2011) . Poxviruses are also studied extensively as candidate vectors for human use, among which attenuated derivatives of vaccinia virus [such as modified vaccinia virus Ankara (MVA) and New York attenuated vaccinia virus NYVAC strains] are the most promising vectors (Esteban, 2009; Gó mez et al., 2008; Rimmelzwaan & Sutter, 2009 ). They are ideal candidate vectors due to their large DNA-packing capacity and their thermal and genetic stability (Minke et al., 2004) . The NYVAC vector has been shown to induce CD4 + T cell-dominant responses, and MVA induces both CD4 + and CD8 + T cell responses (Mooij et al., 2008) . The adenovirus (Ad) vector is another of the most widely evaluated vectors to date to express heterologous antigens, due to ease of production, safety profile, genetic stability, the ease of DNA genome manipulation, and the ability to stimulate both innate and adaptive immune responses and induce both T and B cell responses (Alexander et al., 2012; Fitzgerald et al., 2003; Gabitzsch & Jones, 2011; Lasaro & Ertl, 2009; Vemula & Mittal, 2010; Weyer et al., 2009) . They have been extensively examined as a delivery vector in several preclinical and clinical studies for infectious diseases such as anthrax, hepatitis B, human immunodeficiency virus (HIV)-1, influenza, measles, severe acute respiratory syndrome (SARS), malaria and tuberculosis M. Saxena and others (Chengalvala et al., 1994; Gao et al., 2006; Hashimoto et al., 2005; Hsu et al., 1992; Limbach & Richie, 2009; Radosevic et al., 2007; Shiver et al., 2002) . However, before vectored vaccines can be used in the human population they need to satisfy several important criteria. Safety is a major concern, as even a low level of toxicity is unacceptable (of course the minor discomfort that accompanies many vaccinations is normal). Secondly, a vaccine should be inexpensive, so that it can be administered to a large population at minimal cost, and this is particularly important in resource-poor countries (Killeen & DiRita, 2000) . Similar constraints apply to veterinary vaccines, with cost often an even more important consideration. Finally, long-lasting cellular and (where appropriate) humoral immune responses to the vectored antigen must be induced following administration of these vaccines, preferably with a single dose (Atkins et al., 2006) . As some of the vectors in use will have been seen by the host immune system prior to vaccination, whether the presence of pre-existing immune responses is detrimental for the further development of a vector-based vaccine scheme, or can augment responses to the vectored antigen, needs to be considered in detail. This is the subject of this review. In discussing the possible effects on pre-existing immunity, the natural immunity to the vector needs to be considered. Therefore, considering a vector such as Salmonella, if a host has previously been infected there will exist robust B and T memory responses, and as such, when a vaccination is delivered, an anamnestic response to the Salmonella antigens will be induced (while the response to the vectored antigen will be a primary response). This will theoretically reduce the exposure of the heterologous antigen to the immune system, as the vector is rapidly cleared. Surprisingly, as will be seen in some of the examples given below, this can have results that differ depending on the magnitude of the response to the vectored antigen. Similarly, for virally vectored antigens, the existence of pre-existing immunity to the vector (particularly neutralizing antibody) will restrict delivery of the virus into cells, thereby effectively reducing the dose of the vectored antigen. Again, this might be expected to result in a reduction in the antigenicity of the vectored antigen. In the case of bacterial vectors, the effect of pre-existing immune responses has only been tested using Salmonella serovars and Listeria spp. Concern that prior immunological experience of the host with either the homologous Salmonella vector strain or a related strain might compromise its ability to deliver heterologous vaccine antigen was first raised in 1987 (Dougan et al., 1987) . Bao and Clements subsequently reported experimental evidence of the consequences of prior exposure of animals to the vector strain (Bao & Clements, 1991) . This work showed that both serum and mucosal antibody responses against the foreign antigen were in fact upregulated in animals with prior exposure to the vector strain. Whittle & Verma (1997) reported similar findings. Mice immunized via the intra-peritoneal route with a Salmonella dublin aroA mutant expressing heterologous antigen after being exposed to the same vector showed a higher immune response to the vectored antigen in comparison to mice without any immunological memory against the vector. Subsequently, several studies have been conducted to examine the effect of pre-existing immunity in the host against Salmonella. These results are summarized in Table 1 . The various reports are contradictory in their findings and seem to paint a rather confusing picture. Some studies concluded that pre-existing immunity against the Salmonella vector leads to stronger immune responses against the delivered antigen (Bao & Clements, 1991; Jespersgaard et al., 2001; Kohler et al., 2000a, b; Metzger et al., 2004; Saxena et al., 2009; Sevil Domènech et al., 2008; Whittle & Verma, 1997) , with others considering pre-existing immunity to be a limiting factor in the long-term use of Salmonella as an efficient vector for antigen delivery (Attridge et al., 1997; Gahan et al., 2008; Roberts et al., 1999; Sevil Domènech et al., 2007; Vindurampulle & Attridge, 2003a, b) . A slight majority of the studies listed in Table 1 (10 versus eight) indicate the upregulation of immune responses after animals have been exposed to either homologous or related strains before the delivery of heterologous antigen using a Salmonella vector. A study by Metzger and co-workers on human volunteers using Salmonella Typhi as a vector suggested that there was no change in the T cell immune response against the heterologous antigen in human volunteers who were exposed to empty vector in comparison with volunteers who were immunologically naive of the vector strain (Metzger et al., 2004) . In these subjects, humoral responses were moderately elevated in preexposed individuals. Similarly, Saxena et al. (2009) indicated higher humoral and T cell responses in mice pre-exposed to homologous or heterologous Salmonella strains. The interleukin 4 (IL4) response was significantly higher when the animal host was exposed to the homologous strain, whereas pre-exposure to a related species did not have such an impact on IL4 responses. Conversely interferon (IFN)-c responses were higher, irrespective of the strain to which mice were pre-exposed. This study also indicated that the presence of homologous or heterologous opsonizing antibodies leads to a higher uptake of Salmonella by macrophages in vitro, which may explain the higher immune responses in exposed mice. As may be expected, uptake was higher when homologous sera were used as the opsonin rather than heterologous sera. This is depicted in Fig. 2 . Conversely, there are reports that indicate that pre-existing immunity against the bacterial vector downregulates immune responses against the delivered heterologous antigen using similar or related vectors. Attridge and coworkers reported that the presence of immunity against the bacterial vector prior to the delivery of vectored antigenic Microbiology 159 protein can downregulate immune responses in mice against the delivered antigen (Attridge et al., 1997) . Similar results were reported by Roberts et al. (1999) and Vindurampulle & Attridge (2003a, b) . However, the latter authors found that the hypo-responsiveness could be largely eliminated by exposing animals to the foreign antigen prior to vectorpriming (Vindurampulle & Attridge, 2003b) . Unfortunately, this would appear to be impractical for an immunization regimen! A study presented by Gahan et al. (2008) immunized mice with S. Typhimurium expressing C fragment of tetanus toxin antigen from an expression plasmid or as a DNA vaccine. Vaccinated mice developed humoral responses to LPS and tetC (for the plasmid-bearing vaccines). Animals from all groups (including a previously unvaccinated group) were immunized on day 182 with Salmonella expressing tetC. At this time, the anti-LPS and tetC titres were beginning to wane. Fourteen days after the second immunization, the colonization of various mouse organs was assessed. The ability to colonize was found to be significantly reduced in groups that had been previously vaccinated with Salmonella. In view of this finding, it was perhaps not surprising that at day 210 the LPS titres were not significantly different between groups receiving one or two vaccinations. More interestingly, mice that had been primed with Salmonella alone, and then boosted with Salmonella expressing tetC, induced much lower anti-tetC responses than mice that had not been primed. This argues strongly that prior immunological immunity to the vector can seriously dampen subsequent antigen-specific humoral responses. Whether the same is true for cellular responses was not evaluated. Other studies have evaluated cellular responses. A study by Sevil Domènech and colleagues reported that pre-existing anti-vector immunity seriously compromises CD8 + responses in mice when exposed to a similar strain used as vector (Sevil Domènech et al., 2007) . In contrast, another study by the same authors reported that animals exposed to related vectors induce much higher CD8 + responses when compared with animals which do not have any pre-existing Salmonella immunity (Sevil Domènech et al., 2008) . The difference between these two studies was that in the first, the prime and boost were with identical serovars, while in the second study, different serovars were used. This may point to a way of avoiding downregulation of CD8 responses by pre-existing immunity. This is important, as one of the advantages of using Salmonella (an intracellular pathogen) is that strong cellular immune responses can be induced. It must be noted that in the case of Salmonella vaccines, effects other than strictly immunological responses (particularly adaptive responses) should be considered. In the context of innate immunity, it was shown that administration of non-virulent Salmonella to gnobiotic pigs eliminated disease following challenge with a virulent strain (Foster et al., 2003) . Interestingly, protection was not by competitive exclusion, as the virulent strain was in high numbers in the gut but did not distribute systemically. The protection was proposed to be mediated by the infiltration of a large number of polymorphonuclear leukocytes into the gut, and although perhaps impractical as a general prophylactic (as the time between vaccination and infection is short), this may be an option for short-term or perhaps therapeutic vaccination (as reviewed by Foster et al., 2012) . Chickens (Gallus gallus) are a natural animal reservoir for Salmonella, which makes them an important source of Salmonella-associated gastroenteritis in humans. The ability to use oral Salmonella vaccines to immunize against heterologous pathogens would be of enormous benefit to Uptake of STM-1 by J774 macrophages, relative to the highest uptake percentage. X, Opsonized with naive sera; m, opsonized with serum from mice exposed to Salmonella enteriditis; &, opsonized with serum from mice exposed to STM-1. Pre-existing immunity against vaccine vectors the poultry industry in both broiler and layer flocks. Both vertical and horizontal transmission is associated with Salmonella in chickens (Liljebjelke et al., 2005) . Vertical transmission via in ovo transmission is particularly important, because if there is prior exposure to the vaccine strain, subsequent vaccination using an oral Salmonella vector could be severely compromised. A considerable number of studies on cross-protective immunity and competitive exclusion have been undertaken in chickens. Protective cross-reactive immunity against Salmonella strains has been demonstrated against both homologous and heterologous challenges (Beal et al., 2006) , although cross-serogroup protection was not strong. Furthermore, a recent study reported that pretreatment of newly hatched chickens with different Salmonella strains could produce a complete invasioninhibition effect on any subsequent exposure to both homologous and heterologous strains (Methner et al., 2010) . Pre-exposure with a highly invasive form of Salmonella Enteritidis caused a large influx of heterophils to the caecal mucosa in 1-day-old chicks, and subsequent heterologous caecal colonization was inhibited for a period of 48 h (Methner et al., 2010) . The implications of this kind of colonization-inhibition study on the immunological status of the affected chickens are yet to be fully elucidated. It should be noted that the studies listed in Tables 1 and 2 are controlled laboratory studies, with the possibility of a competitive exclusion component to immunity not discussed. Similarly studies of L. monocytogenes and the effects of preexisting immune responses indicate conflicting results. A study by Bouwer et al. (1999) indicates that pre-existing immune responses against the Listeria vector do not diminish immune responses against the delivered heterologous antigen, and a similar study by Starks et al. (2004) also concluded that prior exposure of mice to the empty Listeria vector did not influence anti-cancer immune responses when a similar mutant was used as a carrier of a melanoma cancer antigen. Similar findings were reported by Whitney et al. (2011) in rhesus macaques in which L. monocytyogens was used as a carrier of gag-HIV antigen. Conversely, studies by Stevens et al. (2005) in which L. monocytogens was used to deliver feline immunodeficiency virus (FIV) gag protein and as a carrier of DNA vaccines to vaccinate cats against FIV envelope protein indicated lower immune responses against the delivered antigen in cats exposed to empty Listeria vector in comparison with naive animals (Stevens et al., 2005) . Similar findings have been reported by Tvinnereim et al. (2002) and Leong et al. (2009) . However, taken together, these studies conclude that prior exposure of host animals to empty vector does not abrogate immune responses to the vectored antigen, but only reduces them somewhat. Only the study by Vijh et al. (1999) indicated that exposure to the empty vector may completely abrogate immune responses against the delivered antigens (Vijh et al., 1999) . However, these studies also indicate that downregulation of antigenspecific immune responses is highly dependent on dose and time. Leong et al. (2009) also demonstrated that the negative impact of vector-specific immune responses can also be countered by repeated immunization with the same vaccine and dose; this in effect leads to higher priming of naive T cells against the delivered antigen. Of course, such repeated vaccination may not be practicable in real-world situations. Despite the many advantages which viral vectoring can offer, pre-existing immunity is a major obstacle of many viralvectored vaccines, such as Ad serotype 5 or herpes simplex virus type 1 (HSV-1), where the rate of seroprevalence to these viruses is very high [40-45 % and 70 % (or more) of the US population, respectively] (Hocknell et al., 2002; Pichla-Gollon et al., 2009) . Vector-specific antibodies may impede the induction of immune responses to the vaccine-encoded antigens, as they may reduce the dose and time of exposure of the target cells to the vaccinated antigens (Pichla-Gollon et al., 2009; Pine et al., 2011) . In a large-scale clinical trial (STEP) of an Ad serotype 5 (AdHu5)-based HIV-1 vaccine, the vaccines showed a lack of efficacy and tended to increase the risk of HIV-1 infection in vaccine recipients who had pre-existing neutralizing antibodies to AdHu5 (Buchbinder et al., 2008) . For an HSV-1-based vector vaccine, it has been demonstrated that pre-existing anti-HSV-1 immunity reduced, but did not abolish, humoral and cellular immune responses against the vaccine-encoded antigen (Hocknell et al., 2002; Lauterbach et al., 2005) . However, Brockman and Knipe found that the induction of durable antibody responses and cellular proliferative responses to HSVencoded antigen were not affected by prior HSV immunity (Brockman & Knipe, 2002) . Similarly, pre-existing immunity to poliovirus has little effect on vaccine efficacy in a poliovirus-vectored vaccine (Mandl et al., 2001) . Different effects of pre-existing immunity on the efficacy of recombinant viral vaccine vectors are summarized in Table 2 . There are several approaches to avoiding pre-existing vector immunity, such as the use of vectors derived from nonhuman sources, using human viruses of rare serotypes (Kahl et al., 2010; Lasaro & Ertl, 2009) , heterologous prime-boost approaches (Liu et al., 2008) , homologous reimmunization (Steffensen et al., 2012) and removing key neutralizing epitopes on the surface of viral capsid proteins (Gabitzsch & Jones, 2011; Roberts et al., 2006) . The inhibitory effect of pre-existing immunity can also be avoided by masking the Ad vector inside dendritic cells (DCs) (Steffensen et al., 2012) . In addition, mucosal vaccination or administration of higher vaccine doses can overcome pre-existing immunity problems (Alexander et al., 2012; Belyakov et al., 1999; Priddy et al., 2008; Xiang et al., 2003) . As we search for new vaccine approaches for the array of pathogens for which none is yet available, revisiting proven vaccines and developing these further has gained M. Saxena and others momentum. Hence, attenuated bacteria and viruses which have a long history of efficacy and safety are being brought into use. While very attractive, a common theme in these experimental approaches has been the limitations that preexisting immunity to the vector may pose. However, as this examination of the relevant literature shows, there is a rather confusing picture, with some studies in fact indicating that pre-existing immunity may be a friend, rather than foe. Few studies using viral vectors have reported on the influence of pre-existing immunity on humoral responses. Generally speaking, for bacterial-delivered antigens, the humoral responses were influenced by pre-existing immunity, with slightly more studies finding augmentation rather than diminution. Why is there variation? This may be due to several factors, including the type of Salmonella used and its invasiveness. Dunstan and colleagues tested the ability of six isogenic Salmonella serovar Typhimurium strains harbouring different mutations for their ability to induce immune responses against the C fragment of tetanus toxin and concluded that the strain which had the least ability to colonize Peyer's patches induced the lowest immune responses (Dunstan et al., 1998) . Similarly, the boosting time and nature of the antigen used might be important. Attridge and colleagues indicated the importance of boosting time. In one experiment, boosting mice at 10 weeks led to complete inhibition of antibody responses against the delivered heterologous antigen; however, when the mice were boosted at 4 weeks, the downregulation of antibody responses was not so prominent (Attridge et al., 1997) . A similar study conducted by Kohlers and colleagues shows that boosting at 7 weeks after pre-exposing animals to empty vector leads to lower antigen-specific IgG and secretory IgA responses; however, boosting at 14 weeks leads to higher IgG and secretory IgA responses (Kohler et al., 2000b) . This is in conflict with the above result, although it should be mentioned that they used different Salmonella species. Vindurampulle and Attridge also examined the impact of the Salmonella strain and the nature of the antigens used. In their study, they used S. Dublin and Salmonella Stanley aroA mutants to deliver E. coli K88 and LT-B antigens, and concluded that the effect of pre-existing immunity depends on both the strain used and the type of antigen delivered (Vindurampulle & Attridge, 2003b) . All these studies on the effect of pre-existing immunity discuss the impact on humoral responses. Sevil Domenech and colleagues reported that pre-exposing animals to the homologous Salmonella vector leads to a significant reduction in CD8 + responses; however, exposure of animals to a heterologous strain leads to significantly higher CD8 + responses (Sevil Domènech et al., 2007 , 2008 . Saxena and colleagues also reported that antigenspecific T cell responses were either similar or significantly higher, with no downregulation in T cell responses observed after pre-exposing mice to either homologous or heterologous strains (Saxena et al., 2009) . For viral vectors, the impact of cell-mediated immunity was more pronounced, and as depicted in Table 2 , almost always resulted in a reduction in the subsequent immune response. Presumably this is because viruses will induce neutralizing antibody on the first dose, and in subsequent doses this antibody will limit the number of transduced cells, therefore limiting the responses. This is particularly a problem with a common viral vector such as Ad, where a large proportion of the population will have immunological memory against common serotypes (Lasaro & Ertl, 2009) . As these authors conclude, it will be possible to utilize such vectors only by developing vaccines from alternative serotypes. It may be that a vector such as Pre-existing immunity against vaccine vectors attenuated influenza virus, with the ability to easily develop reassortants, will be useful in this context. In addition, immunological memory in the form of opsonizing antibody certainly plays an important role in the early uptake of Salmonella by macrophages and DC. This may be beneficial, as the live bacterial vector used for delivery purposes harbours mutations in genes encoding proteins responsible for their survival in the animal host. This not only encumbers their ability to cause disease, making them safe live vectors, but also limits the number of replications. The presence of opsonizing antibodies should mean a higher level of bacterial uptake, leading to higher presentation to the immune system and therefore a better immune response. We have previously shown that this is indeed the case (Saxena et al., 2009 ) (depicted in Fig. 2 ). It would be of great benefit to address these issues not only in mice but also in other organisms such as chickens, which are the most likely host to be targeted for the use of live Salmonella vectors, specifically where the vaccines are developed for use in livestock and poultry. To summarize, bacterial vectors such as Salmonella and viral vectors such as Ad show great promise as delivery vehicles for heterologous antigens; however, prior exposure to the vector must be considered. By judicious selection of the strain/serotype it will be possible to avoid the negative effects and it may indeed be possible to positively influence the response, particularly for humoral immunity.

How can vectors for which host has immunity, be used differently to increase the efficacy of vaccination?

A focus reduction neutralization assay for hepatitis C virus neutralizing antibodies https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1852297/ SHA: ee8dca216514deeed4c9415bc2ad8a78dc3d9670 Authors: Fournier, Carole; Duverlie, Gilles; François, Catherine; Schnuriger, Aurelie; Dedeurwaerder, Sarah; Brochot, Etienne; Capron, Dominique; Wychowski, Czeslaw; Thibault, Vincent; Castelain, Sandrine Date: 2007-03-30 DOI: 10.1186/1743-422x-4-35 License: cc-by Abstract: BACKGROUND/AIM: The role of humoral immunity in hepatitis C virus (HCV) infection is poorly understood. Nevertheless, there is increasing interest in characterizing the neutralizing antibodies in the serum of HCV-infected patients. Focus reduction assays have been widely used to evaluate neutralizing antibody responses against a range of non-cytopathic viruses. Based on the recent development of a HCV cell culture system using the genotype 2 JFH-1-strain, we developed a focus reduction assay for HCV-neutralizing antibodies. METHODS: The focus reduction assay was based on a standard microneutralization assay in which immunostained foci on tissue culture plates are counted. The neutralizing anti-HCV antibodies titers of purified serum immunoglobulin samples from seventy-seven individuals were determined using a 50% focus reduction neutralization assay. Each titer was determined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. IgG antibodies were first purified from each serum in order to avoid the facilitating effect of HDL on HCV entry. RESULTS: The assay's cut-off using an ELISA and RNA HCV-negative samples was found to be 1.25 log, corresponding to a dilution of 1:18. The assay was compared with a commercial HCV ELISA and exhibited specificity and sensitivity values of 100% and 96.5%, respectively, and good reproducibility (with intra-assay and inter-assay coefficients of variation of 6.7% and 12.6%, respectively). The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples. The neutralizing antibodies titers were 2.13 log (1:134) for homologous samples from HCV genotype 2 infected patients harboring the same genotype as JFH-1 and 1.93 log (1:85) for heterologous samples from patients infected by genotypes other than type 2. These results confirm the presence of broadly cross-neutralizing antibodies already reported using the HCV pseudoparticles system. CONCLUSION: This study presents a simple, specific and reproducible cell culture-based assay for determination of HCV-neutralizing antibodies in human sera. The assay should be an important tool for gauging the relationship between the neutralizing antibodies response and viral load kinetics in acutely or chronically infected patients and for investigating the possible eradication or prevention of HCV infection by neutralizing antibodies. Text: Hepatitis C virus (HCV, a member of the Flaviviridae family) is an enveloped, positive-stranded RNA virus that preferentially replicates in hepatocytes. At least 170 million people worldwide are persistently infected with hepatitis C virus. Chronic HCV infection is associated with a significant risk of progression to cirrhosis and hepatocellular carcinoma [1] . Antiviral therapy with pegylated alpha-interferon and ribavirin (the current best therapeutic regimen) is only successful in about 50% of all treated patients. Better knowledge of the viral and host factors that determine HCV clearance or persistence during the acute stage of infection is needed in order to improve antiviral therapy and to develop efficient vaccines. Studies focusing on innate and cellular immune responses have shown that a sufficiently large HCV inoculum is able to evade, subvert or circumvent the host's defences. At present, the chimpanzee is the only reliable experimental animal model in which the initial post-HCV infection events and the efficacy of vaccine candidates can be evaluated [2] . It has been shown that HCV-specific T-cell immunity is important in the control of HCV infection [3, 4] . Several studies have indicated a role for humoral immunity in the acute stage of HCV infection but this aspect remains poorly characterized. The E1 and E2 glycoproteins are thought to be the viral attachment proteins and thus the main targets for HCV-neutralizing antibodies; identification of protective epitopes conserved across different strains of HCV is therefore a major challenge in vaccine design. A number of antibodies capable of blocking E2 binding to cells or cell receptors have been described, [5] [6] [7] [8] some of which neutralize HCV entry in animal or cellular models [9, 10] . Cell entry has been shown to involve several surface molecules (notably including the tetraspanin CD81 and the SR-BI receptor [11, 12] ), although further studies are needed to better understand how viral entry occurs and how it might be neutralized. Detection of neutralizing antibodies in human blood had been problematical until an efficient and reliable cell culture system for HCV became available. Hence, the development of an in vitro neutralization assay for HCV could be extremely valuable for characterizing the humoral immune response to HCV and for evaluating the potential of passive and active immunization against hepatitis C. Recent studies using an in vitro neutralization assay system (based on infectious retroviral pseudoparticles (HCVpp) bearing HCV envelope glycoproteins) have confirmed that HCV-infected patient sera can indeed neutralize infection [13, 14] . However, it has also been shown that the neutralizing activity of antibodies from HCV-infected patients is attenuated by a factor present in human serum, identified as the highdensity lipoprotein (HDL) fraction [11, 13, 15] . HDL facilitation of HCVpp entry is a post-binding event [16] , sug-gesting that HDLs favour internalization of virions and thus the latter's escape from neutralizing antibodies. Recently, an HCV cell culture model (HCVcc) has been developed [17] [18] [19] , allowing the production of virus particles that can be efficiently propagated in cell culture. Some preliminary neutralization assays have been carried out by these authors. In this study, we describe how we set up a standardized focus reduction neutralization assay based on HCVcc. Focus reduction assays have been widely used to evaluate the neutralizing antibody responses to viruses that can form foci in infected cells. Following the recent development of the HCVcc model, the principle of the focus reduction assay has been applied to HCV-neutralizing antibodies detection. The JFH-1 HCV 2a viral strain was grown on a Huh-7 human hepatoma cell line. After three days of infection and cell permeabilization, detection of the HCV foci was carried out using an inactivated HCVpositive patient serum primary antibody and a peroxidase-coupled, Fc-specific anti-human IgG-antibody. The reaction was revealed with DAB peroxidase substrate. The viral foci were thus stained brown, making them easy to count (see Fig. 1a ). It has been recently shown that the neutralizing activity of HCV antibodies is attenuated by a serum factor associated with the HDL fraction. Hence, HDLs were able to facilitate HCVpp and HCVcc entry via a mechanism which depended on the expression of the scavenger receptor BI (SR-BI) and its selective lipid-uptake function [11, 15, 16, 20] . In view of the role of HDL in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibody titer (see Fig. 1b ). The specificity of the HCV neutralization assay was assessed by testing 20 anti-HCV-ELISA-negative samples, including five positive for hepatitis B virus surface antibodies (anti-HBs) and five positive for heterophile antibodies. All samples tested negative with two commercial anti-HCV antibody detection assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany; Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom) and HCV-RNA-negative with a qualitative, commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France). These anti-HCV-negative samples were compared with 11 samples from patients chronically infected with HCV genotype 2. The neutralization titers of anti-HCV-negative serum samples are shown in Fig. 2 ., with a mean value of 1.083 ± 0.083 (corresponding to a dilution of 1:12). The assay's cut-off (determined as the mean value for negative samples plus two standard deviations) corresponded to a dilution of 1:18. The assay exhibited specificity and sensibility values of 100% and 96.5%, respectively. The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples (data not shown). Conversely, the chronically HCV genotype 2-positive samples displayed strong reactions, with a mean value of 2.128 ± 0.365 (corresponding to a dilution of 1:134) (p < 0.001). Inter-assay variability was determined by testing one HCV genotype 2 sample in 10 consecutive experiments (n = 10), whereas intra-assay variability was evaluated by testing the same sample 10 times (n = 10) in the same experiment, whilst running the dilution series. The intra-assay and inter-assay coefficients of variation (CV) of the log neutralization titers were 6.7% and 12.6%, respectively. Fifty-seven HCV-positive antibodies samples were evaluated using the HCV focus reduction neutralization assay. The genotypes were distributed as follows; for types 1a, 1b, 2, 3, 4 and 5, we studied 11, 11, 11, 12, 10 and 2 samples, respectively. The mean values of the different genotypes is shown in Fig. 3 . and Table 1 . The mean log neutralization titers for genotypes 1a, 2 and 3 are very similar (2.046 ± 0.671 for genotype 1a, 2.128 ± 0.365 for genotype 2 and 2.148 ± 0.478 for genotype 3). The mean average values are lower for genotype 1b (1.747 ± 0.462) and genotype 4 (1.786 ± 0.236). Strikingly, very high heterologous titers were observed for five patients -three infected with HCV genotype 1a and two infected with HCV genotype 3 (see Fig. 3a ). There were too few genotype 5 samples to compare with the other genotypes but the corresponding results nevertheless indicate that the neutralization assay is suitable for this genotype. The two The distribution of the log neutralization titers across all the HCV ELISA and RNA-positive samples as a function of the HCV genotype is shown in Fig. 3b . More than 60% of the neutralizing antibodies titers fell in the range from 1.7 to 2.69 log titers, corresponding to dilutions of 1:50 and 1:500, respectively. Overall, 3.5% of the samples displayed a titer greater than log 3.0 (1:1000) and, conversely, 3.5% displayed a titer below the cut-off value, i.e. log 1.25 (1:10). Thus, of 57 HCV-infected patients, only two did not test positive for neutralizing antibodies in this assay (the titers were 0.960 and 0.932, respectively). The role of neutralizing antibodies during acute and chronic viral infection remains an important question and has generated controversial results. Initially, the presence of neutralizing antibodies was shown to control the HCV load and to contribute to viral eradication in patients capable of clearing the infection [13] . In other studies, the appearance of neutralizing antibodies was delayed and restricted to IgG1 antibodies in patients who develop a chronic infection [2, 21] . The chimpanzee model has been critical for the study of HCV transmission and host immune responses; however, neutralizing antibodies were not detected in some animals that resolved their infection -suggesting a minimal role in viral clearance, as also observed in human studies [14, 15] . Experimentally infected chimpanzees and naturally infected humans can be re-infected with homologous and heterologous HCV strains, suggesting that the humoral immunity that develops after spontaneous resolution of acute hepatitis C is not sterilizing [22] [23] [24] . During chronic infection in humans, the presence and/or production of neutralizing antibodies do not suffice for curing the infection but could regulate the spread of the virus. Thus, it can be postulated that during chronic infection, viral mutants can continuously escape the renewed production of neutralizing antibodies. Retroviral pseudoparticles have been used to develop a very interesting tool for measuring neutralizing antibodies in vitro [14] . The assay has demonstrated the presence of HCV-neutralizing antibodies in human sera with relatively high titers (>1:320) and broadly neutralizing activity against different HCV genotypes. However, this model does not represent genuine HCV virions; in particular, the budding of retroviral particles is thought to be very different and may involve a variety of cellular pathways. Characterization of infectious retroviral pseudotype particles bearing HCV glycoproteins have been shown to be very heterogeneous, and so it is possible that these pseudoparticles may not be as relevant as the native HCV virions [25] . The recent development of a cell culture model for HCV enables the production of native HCV virions that can be efficiently propagated in cell culture [17] [18] [19] . This cell culture system has allowed us to develop a neutralization assay for evaluating the level and the proportion of HCVneutralizing antibodies in chronically infected HCV patients. We analysed a number of parameters (such as practicability, reproducibility and specificity) and tested the effect of a range of variables (viral inoculum size, incubation time, fixation and permeabilization methods, blocking and revelation reagents) on these parameters (data not shown). Overall, the neutralization assay described in this study performs similarly to standardized neutralization assays for many other viruses [26] [27] [28] . The assay relies on the ability of the specific JFH-1 genotype 2 viral strain to replicate and multiply on a Huh-7 human hepatoma cell line in a cell culture model, enabling the rapid detection of viral foci after 72 hours of infection. Moreover, no secondary foci were detectable at this time point. Fixation with paraformaldehyde and permeabilization with Triton X-100 were chosen in order to preserve antigenicity and prevent the cell monolayer from detaching during washes. Development with DAB peroxide substrate made it easy to count specifically coloured viral foci. The viral inoculum size is an important parameter; it has to be low enough to enable good assay sensitivity but high enough to produce a statistically significant number of foci, i.e. allowing the reduction in the number of foci (and thus the effect of neutralization) to be monitored. Thus, 100 FFUs were used as the inoculum in this neutralization assay. In order to test different human samples, we had to take into account the ability of HDL to facilitate HCVcc entry via a mechanism which depends on expression of the scavenger receptor BI [11, 15, 16, 20] . Given HDL's role in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibodies titer; this frees the assay of the risk of non-specific neutralization activity of the serum via the effects of HDL, the complement system and/or serum amyloid A protein (SAA) [29] . The HCV neutralization assay exhibited good reproducibility, for both duplicate assays and independent tests. As expected, the intra-assay coefficient of variation (CV) was lower than the interassay CV. The test also showed good specificity, since there was no interaction with anti-HIV, anti-HBV or heterophile antibodies. Very low titers were found with HCV ELISA and RNA-negative samples, and the assay's cut-off was determined as the mean titer for negative samples plus two standard deviations (1.25 log, corresponding to a dilution of 1:18). Given that only the JFH-1 strain of HCV genotype 2a was available for the assay, we evaluated the neutralization titer of sera from patients chronically infected with other HCV genotypes, i.e. 1, 2, 3, 4 and 5. Most of these sera were detected as positive by the neutralization assay, except for two sera from HCV genotype 1-infected patients. These two samples presented a high specific antibody ratio according to the ELISA but only very low inhibition by neutralization assay (far below the cut-off, in fact). We conclude that either the samples lacked neutralizing antibodies or that any such antibodies that were present did not cross-neutralize with HCV genotype 2a. The sensitivity was 100% -not only for genotype 2 (the genotype of the strain used for the assay) but also for other HCV genotypes (except genotype 1). HCV genotype 5 antibodies were also measured but there were too few samples for accurate testing. Moreover, the positive sera (96.5%) had comparable and significantly high titers (1.99 ± 0.63), whatever the genotype. This finding suggests that most neutralizing antibodies are cross-reactive. Another possibility is that most of the patients had been previously infected by a genotype 2 strain. However, this is unlikely because few genotype 2 strains are circulating in France [30] . As expected for a neutralization test, the assay presented in the present study appeared to be very specific (independently of the genotype) and usable in most circumstances. For most viral infections, neutralization assays such as that described in this study are used as reference assays. Thus, we are confident that as other HCVcc genotypes become available, these assays will replace the pseudoparticle assay in the near future because they are probably more relevant. Our assay is somewhat time-consuming and could be simplified by using one dilution to count the foci; however, this type of "short cut" would make it difficult to extrapolate to the dilution neutralizing 50% of the inoculum. Another approach would consist in using recombinant HCV capable of expressing reporter genes (such as luciferase) in order to use a single dilution and obtain a quantitative result [31] . However, further neutralization studies using other genotypes are needed in order to complete our observations and to char- A simple, specific and reproducible cell culture-based neutralization assay was developed for the determination of neutralizing anti-HCV antibodies in human sera. This test should be an important tool for gauging the relationship between the neutralizing response and viral load kinetics in acutely and chronically infected patients. The Huh-7 human hepatoma cells [32] were grown in Dulbecco's minimum essential medium (Invitrogen) supplemented with 10% fetal bovine serum. All cell cultures were maintained in 5% CO 2 at 37°C. The plasmid pJFH-1 containing the full-length cDNA of the JFH-1 isolate (which belongs to subtype 2a (GenBank accession no. AB047639)), was a gift from Dr Wakita (Department of Microbiology, Tokyo Metropolitan Institute for Neuroscience, Tokyo, Japan) and has been described previously [17] . To generate genomic HCV RNA, the plasmid pJFH-1 was linearized at the 3' end of the HCV cDNA and used as a template for in vitro transcription, as described previously [33] . Viral stocks were obtained by harvesting cell culture supernatants and freezing them at -80°C. Virus titration was performed on Huh-7 cells with 6-well microtiter plates (Corning, NY) 72 hours after incubation, by immunostaining the cells with antibodies from a HCV-positive patient serum that had previously been inactivated at 56°C (see the section on the virus neutralization assay). The viral titer was determined in triplicate from the mean number of foci and expressed as focus forming units/mL (FFU/mL). Seventy-seven human serum samples were tested. Collection of the sera was approved by the local Ethics Committee and informed consent had been obtained from the donors. Fifty-seven of these samples were obtained from chronically infected HCV patients. The presence of HCV antibodies was determined and confirmed using two third-generation HCV EIA assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany and Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom). HCV RNA was determined with a qualitative commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France) and HCV genotyping was performed by direct sequencing, as described elsewhere [34] . The genotypes were distributed as follows: 11, 11, 11, 12, 10 and 2 samples of types 1a, 1b, 2, 3, 4 and 5, respectively. A set of 20 anti-HCV-negative serum samples was used to evaluate the assay's specif-icity, including five serum samples with positive hepatitis B virus surface antibody (anti-HBs) status and five sera from Epstein-Barr virus-infected patients that had tested positive for heterophile antibodies. All serum samples had been stored at -80°C upon collection and had not been thawed until the time of assay. Serum immunoglobulins G (IgG) fraction was purified using protein G-Sepharose (GE Healthcare, Orsay, France The HCV focus reduction neutralization assay was performed in 96-well microtiter plates. Serial dilutions of purified IgG (10 μg) ranging from 1:10 to 1:1,280 were established. Each dilution was tested twice. 25 μL of each sample was mixed with 25 μL of virus (100 FFU) in 96well microtiter plates and incubated for 1 hour at 37°C, 5% CO 2 . A volume of 100 μL of Huh-7 cell suspension (10,000 cells/well) in culture medium was added and incubated for 5 hours at 37°C, 5% CO2. After 5 hours, the supernatants were removed and 100 μL of culture medium were added to the monolayers. After 72 hours, the cells were fixed with paraformaldehyde and permeabilized with 0.5% Triton X-100. Primary antibody (a HCVpositive patient serum inactivated at 56°C) was diluted to 1:500 prior to use and then incubated for 1 h at room temperature. A peroxidase-coupled, Fc-specific anti-human IgG antibody (Sigma, Saint Quentin Fallavier, France) diluted to 1:200 was dispensed onto the cell monolayer and incubated for 30 min at room temperature. The reaction was developed with DAB peroxidase substrate (Sigma, Saint Quentin Fallavier, France) and stopped after 10 min of incubation with distilled water. The number of HCV foci in each dilution was determined. Controls were included in each assay (non-neutralized virus, purified IgG from each patient at a 1:10 dilution). The dilution that neutralized 50% of the virus was calculated by curvilinear regression analysis using XLSTAT 2006 software (Addinsoft SARL, Paris, France) [35] . Each titer was deter-mined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. Titers were expressed as logarithmic values and means ± standard deviation were calculated. Student's t-test was used to compare data between groups. p values below 0.05 were considered to be significant.

What is the Hepatitis C virus?

A focus reduction neutralization assay for hepatitis C virus neutralizing antibodies https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1852297/ SHA: ee8dca216514deeed4c9415bc2ad8a78dc3d9670 Authors: Fournier, Carole; Duverlie, Gilles; François, Catherine; Schnuriger, Aurelie; Dedeurwaerder, Sarah; Brochot, Etienne; Capron, Dominique; Wychowski, Czeslaw; Thibault, Vincent; Castelain, Sandrine Date: 2007-03-30 DOI: 10.1186/1743-422x-4-35 License: cc-by Abstract: BACKGROUND/AIM: The role of humoral immunity in hepatitis C virus (HCV) infection is poorly understood. Nevertheless, there is increasing interest in characterizing the neutralizing antibodies in the serum of HCV-infected patients. Focus reduction assays have been widely used to evaluate neutralizing antibody responses against a range of non-cytopathic viruses. Based on the recent development of a HCV cell culture system using the genotype 2 JFH-1-strain, we developed a focus reduction assay for HCV-neutralizing antibodies. METHODS: The focus reduction assay was based on a standard microneutralization assay in which immunostained foci on tissue culture plates are counted. The neutralizing anti-HCV antibodies titers of purified serum immunoglobulin samples from seventy-seven individuals were determined using a 50% focus reduction neutralization assay. Each titer was determined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. IgG antibodies were first purified from each serum in order to avoid the facilitating effect of HDL on HCV entry. RESULTS: The assay's cut-off using an ELISA and RNA HCV-negative samples was found to be 1.25 log, corresponding to a dilution of 1:18. The assay was compared with a commercial HCV ELISA and exhibited specificity and sensitivity values of 100% and 96.5%, respectively, and good reproducibility (with intra-assay and inter-assay coefficients of variation of 6.7% and 12.6%, respectively). The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples. The neutralizing antibodies titers were 2.13 log (1:134) for homologous samples from HCV genotype 2 infected patients harboring the same genotype as JFH-1 and 1.93 log (1:85) for heterologous samples from patients infected by genotypes other than type 2. These results confirm the presence of broadly cross-neutralizing antibodies already reported using the HCV pseudoparticles system. CONCLUSION: This study presents a simple, specific and reproducible cell culture-based assay for determination of HCV-neutralizing antibodies in human sera. The assay should be an important tool for gauging the relationship between the neutralizing antibodies response and viral load kinetics in acutely or chronically infected patients and for investigating the possible eradication or prevention of HCV infection by neutralizing antibodies. Text: Hepatitis C virus (HCV, a member of the Flaviviridae family) is an enveloped, positive-stranded RNA virus that preferentially replicates in hepatocytes. At least 170 million people worldwide are persistently infected with hepatitis C virus. Chronic HCV infection is associated with a significant risk of progression to cirrhosis and hepatocellular carcinoma [1] . Antiviral therapy with pegylated alpha-interferon and ribavirin (the current best therapeutic regimen) is only successful in about 50% of all treated patients. Better knowledge of the viral and host factors that determine HCV clearance or persistence during the acute stage of infection is needed in order to improve antiviral therapy and to develop efficient vaccines. Studies focusing on innate and cellular immune responses have shown that a sufficiently large HCV inoculum is able to evade, subvert or circumvent the host's defences. At present, the chimpanzee is the only reliable experimental animal model in which the initial post-HCV infection events and the efficacy of vaccine candidates can be evaluated [2] . It has been shown that HCV-specific T-cell immunity is important in the control of HCV infection [3, 4] . Several studies have indicated a role for humoral immunity in the acute stage of HCV infection but this aspect remains poorly characterized. The E1 and E2 glycoproteins are thought to be the viral attachment proteins and thus the main targets for HCV-neutralizing antibodies; identification of protective epitopes conserved across different strains of HCV is therefore a major challenge in vaccine design. A number of antibodies capable of blocking E2 binding to cells or cell receptors have been described, [5] [6] [7] [8] some of which neutralize HCV entry in animal or cellular models [9, 10] . Cell entry has been shown to involve several surface molecules (notably including the tetraspanin CD81 and the SR-BI receptor [11, 12] ), although further studies are needed to better understand how viral entry occurs and how it might be neutralized. Detection of neutralizing antibodies in human blood had been problematical until an efficient and reliable cell culture system for HCV became available. Hence, the development of an in vitro neutralization assay for HCV could be extremely valuable for characterizing the humoral immune response to HCV and for evaluating the potential of passive and active immunization against hepatitis C. Recent studies using an in vitro neutralization assay system (based on infectious retroviral pseudoparticles (HCVpp) bearing HCV envelope glycoproteins) have confirmed that HCV-infected patient sera can indeed neutralize infection [13, 14] . However, it has also been shown that the neutralizing activity of antibodies from HCV-infected patients is attenuated by a factor present in human serum, identified as the highdensity lipoprotein (HDL) fraction [11, 13, 15] . HDL facilitation of HCVpp entry is a post-binding event [16] , sug-gesting that HDLs favour internalization of virions and thus the latter's escape from neutralizing antibodies. Recently, an HCV cell culture model (HCVcc) has been developed [17] [18] [19] , allowing the production of virus particles that can be efficiently propagated in cell culture. Some preliminary neutralization assays have been carried out by these authors. In this study, we describe how we set up a standardized focus reduction neutralization assay based on HCVcc. Focus reduction assays have been widely used to evaluate the neutralizing antibody responses to viruses that can form foci in infected cells. Following the recent development of the HCVcc model, the principle of the focus reduction assay has been applied to HCV-neutralizing antibodies detection. The JFH-1 HCV 2a viral strain was grown on a Huh-7 human hepatoma cell line. After three days of infection and cell permeabilization, detection of the HCV foci was carried out using an inactivated HCVpositive patient serum primary antibody and a peroxidase-coupled, Fc-specific anti-human IgG-antibody. The reaction was revealed with DAB peroxidase substrate. The viral foci were thus stained brown, making them easy to count (see Fig. 1a ). It has been recently shown that the neutralizing activity of HCV antibodies is attenuated by a serum factor associated with the HDL fraction. Hence, HDLs were able to facilitate HCVpp and HCVcc entry via a mechanism which depended on the expression of the scavenger receptor BI (SR-BI) and its selective lipid-uptake function [11, 15, 16, 20] . In view of the role of HDL in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibody titer (see Fig. 1b ). The specificity of the HCV neutralization assay was assessed by testing 20 anti-HCV-ELISA-negative samples, including five positive for hepatitis B virus surface antibodies (anti-HBs) and five positive for heterophile antibodies. All samples tested negative with two commercial anti-HCV antibody detection assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany; Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom) and HCV-RNA-negative with a qualitative, commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France). These anti-HCV-negative samples were compared with 11 samples from patients chronically infected with HCV genotype 2. The neutralization titers of anti-HCV-negative serum samples are shown in Fig. 2 ., with a mean value of 1.083 ± 0.083 (corresponding to a dilution of 1:12). The assay's cut-off (determined as the mean value for negative samples plus two standard deviations) corresponded to a dilution of 1:18. The assay exhibited specificity and sensibility values of 100% and 96.5%, respectively. The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples (data not shown). Conversely, the chronically HCV genotype 2-positive samples displayed strong reactions, with a mean value of 2.128 ± 0.365 (corresponding to a dilution of 1:134) (p < 0.001). Inter-assay variability was determined by testing one HCV genotype 2 sample in 10 consecutive experiments (n = 10), whereas intra-assay variability was evaluated by testing the same sample 10 times (n = 10) in the same experiment, whilst running the dilution series. The intra-assay and inter-assay coefficients of variation (CV) of the log neutralization titers were 6.7% and 12.6%, respectively. Fifty-seven HCV-positive antibodies samples were evaluated using the HCV focus reduction neutralization assay. The genotypes were distributed as follows; for types 1a, 1b, 2, 3, 4 and 5, we studied 11, 11, 11, 12, 10 and 2 samples, respectively. The mean values of the different genotypes is shown in Fig. 3 . and Table 1 . The mean log neutralization titers for genotypes 1a, 2 and 3 are very similar (2.046 ± 0.671 for genotype 1a, 2.128 ± 0.365 for genotype 2 and 2.148 ± 0.478 for genotype 3). The mean average values are lower for genotype 1b (1.747 ± 0.462) and genotype 4 (1.786 ± 0.236). Strikingly, very high heterologous titers were observed for five patients -three infected with HCV genotype 1a and two infected with HCV genotype 3 (see Fig. 3a ). There were too few genotype 5 samples to compare with the other genotypes but the corresponding results nevertheless indicate that the neutralization assay is suitable for this genotype. The two The distribution of the log neutralization titers across all the HCV ELISA and RNA-positive samples as a function of the HCV genotype is shown in Fig. 3b . More than 60% of the neutralizing antibodies titers fell in the range from 1.7 to 2.69 log titers, corresponding to dilutions of 1:50 and 1:500, respectively. Overall, 3.5% of the samples displayed a titer greater than log 3.0 (1:1000) and, conversely, 3.5% displayed a titer below the cut-off value, i.e. log 1.25 (1:10). Thus, of 57 HCV-infected patients, only two did not test positive for neutralizing antibodies in this assay (the titers were 0.960 and 0.932, respectively). The role of neutralizing antibodies during acute and chronic viral infection remains an important question and has generated controversial results. Initially, the presence of neutralizing antibodies was shown to control the HCV load and to contribute to viral eradication in patients capable of clearing the infection [13] . In other studies, the appearance of neutralizing antibodies was delayed and restricted to IgG1 antibodies in patients who develop a chronic infection [2, 21] . The chimpanzee model has been critical for the study of HCV transmission and host immune responses; however, neutralizing antibodies were not detected in some animals that resolved their infection -suggesting a minimal role in viral clearance, as also observed in human studies [14, 15] . Experimentally infected chimpanzees and naturally infected humans can be re-infected with homologous and heterologous HCV strains, suggesting that the humoral immunity that develops after spontaneous resolution of acute hepatitis C is not sterilizing [22] [23] [24] . During chronic infection in humans, the presence and/or production of neutralizing antibodies do not suffice for curing the infection but could regulate the spread of the virus. Thus, it can be postulated that during chronic infection, viral mutants can continuously escape the renewed production of neutralizing antibodies. Retroviral pseudoparticles have been used to develop a very interesting tool for measuring neutralizing antibodies in vitro [14] . The assay has demonstrated the presence of HCV-neutralizing antibodies in human sera with relatively high titers (>1:320) and broadly neutralizing activity against different HCV genotypes. However, this model does not represent genuine HCV virions; in particular, the budding of retroviral particles is thought to be very different and may involve a variety of cellular pathways. Characterization of infectious retroviral pseudotype particles bearing HCV glycoproteins have been shown to be very heterogeneous, and so it is possible that these pseudoparticles may not be as relevant as the native HCV virions [25] . The recent development of a cell culture model for HCV enables the production of native HCV virions that can be efficiently propagated in cell culture [17] [18] [19] . This cell culture system has allowed us to develop a neutralization assay for evaluating the level and the proportion of HCVneutralizing antibodies in chronically infected HCV patients. We analysed a number of parameters (such as practicability, reproducibility and specificity) and tested the effect of a range of variables (viral inoculum size, incubation time, fixation and permeabilization methods, blocking and revelation reagents) on these parameters (data not shown). Overall, the neutralization assay described in this study performs similarly to standardized neutralization assays for many other viruses [26] [27] [28] . The assay relies on the ability of the specific JFH-1 genotype 2 viral strain to replicate and multiply on a Huh-7 human hepatoma cell line in a cell culture model, enabling the rapid detection of viral foci after 72 hours of infection. Moreover, no secondary foci were detectable at this time point. Fixation with paraformaldehyde and permeabilization with Triton X-100 were chosen in order to preserve antigenicity and prevent the cell monolayer from detaching during washes. Development with DAB peroxide substrate made it easy to count specifically coloured viral foci. The viral inoculum size is an important parameter; it has to be low enough to enable good assay sensitivity but high enough to produce a statistically significant number of foci, i.e. allowing the reduction in the number of foci (and thus the effect of neutralization) to be monitored. Thus, 100 FFUs were used as the inoculum in this neutralization assay. In order to test different human samples, we had to take into account the ability of HDL to facilitate HCVcc entry via a mechanism which depends on expression of the scavenger receptor BI [11, 15, 16, 20] . Given HDL's role in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibodies titer; this frees the assay of the risk of non-specific neutralization activity of the serum via the effects of HDL, the complement system and/or serum amyloid A protein (SAA) [29] . The HCV neutralization assay exhibited good reproducibility, for both duplicate assays and independent tests. As expected, the intra-assay coefficient of variation (CV) was lower than the interassay CV. The test also showed good specificity, since there was no interaction with anti-HIV, anti-HBV or heterophile antibodies. Very low titers were found with HCV ELISA and RNA-negative samples, and the assay's cut-off was determined as the mean titer for negative samples plus two standard deviations (1.25 log, corresponding to a dilution of 1:18). Given that only the JFH-1 strain of HCV genotype 2a was available for the assay, we evaluated the neutralization titer of sera from patients chronically infected with other HCV genotypes, i.e. 1, 2, 3, 4 and 5. Most of these sera were detected as positive by the neutralization assay, except for two sera from HCV genotype 1-infected patients. These two samples presented a high specific antibody ratio according to the ELISA but only very low inhibition by neutralization assay (far below the cut-off, in fact). We conclude that either the samples lacked neutralizing antibodies or that any such antibodies that were present did not cross-neutralize with HCV genotype 2a. The sensitivity was 100% -not only for genotype 2 (the genotype of the strain used for the assay) but also for other HCV genotypes (except genotype 1). HCV genotype 5 antibodies were also measured but there were too few samples for accurate testing. Moreover, the positive sera (96.5%) had comparable and significantly high titers (1.99 ± 0.63), whatever the genotype. This finding suggests that most neutralizing antibodies are cross-reactive. Another possibility is that most of the patients had been previously infected by a genotype 2 strain. However, this is unlikely because few genotype 2 strains are circulating in France [30] . As expected for a neutralization test, the assay presented in the present study appeared to be very specific (independently of the genotype) and usable in most circumstances. For most viral infections, neutralization assays such as that described in this study are used as reference assays. Thus, we are confident that as other HCVcc genotypes become available, these assays will replace the pseudoparticle assay in the near future because they are probably more relevant. Our assay is somewhat time-consuming and could be simplified by using one dilution to count the foci; however, this type of "short cut" would make it difficult to extrapolate to the dilution neutralizing 50% of the inoculum. Another approach would consist in using recombinant HCV capable of expressing reporter genes (such as luciferase) in order to use a single dilution and obtain a quantitative result [31] . However, further neutralization studies using other genotypes are needed in order to complete our observations and to char- A simple, specific and reproducible cell culture-based neutralization assay was developed for the determination of neutralizing anti-HCV antibodies in human sera. This test should be an important tool for gauging the relationship between the neutralizing response and viral load kinetics in acutely and chronically infected patients. The Huh-7 human hepatoma cells [32] were grown in Dulbecco's minimum essential medium (Invitrogen) supplemented with 10% fetal bovine serum. All cell cultures were maintained in 5% CO 2 at 37°C. The plasmid pJFH-1 containing the full-length cDNA of the JFH-1 isolate (which belongs to subtype 2a (GenBank accession no. AB047639)), was a gift from Dr Wakita (Department of Microbiology, Tokyo Metropolitan Institute for Neuroscience, Tokyo, Japan) and has been described previously [17] . To generate genomic HCV RNA, the plasmid pJFH-1 was linearized at the 3' end of the HCV cDNA and used as a template for in vitro transcription, as described previously [33] . Viral stocks were obtained by harvesting cell culture supernatants and freezing them at -80°C. Virus titration was performed on Huh-7 cells with 6-well microtiter plates (Corning, NY) 72 hours after incubation, by immunostaining the cells with antibodies from a HCV-positive patient serum that had previously been inactivated at 56°C (see the section on the virus neutralization assay). The viral titer was determined in triplicate from the mean number of foci and expressed as focus forming units/mL (FFU/mL). Seventy-seven human serum samples were tested. Collection of the sera was approved by the local Ethics Committee and informed consent had been obtained from the donors. Fifty-seven of these samples were obtained from chronically infected HCV patients. The presence of HCV antibodies was determined and confirmed using two third-generation HCV EIA assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany and Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom). HCV RNA was determined with a qualitative commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France) and HCV genotyping was performed by direct sequencing, as described elsewhere [34] . The genotypes were distributed as follows: 11, 11, 11, 12, 10 and 2 samples of types 1a, 1b, 2, 3, 4 and 5, respectively. A set of 20 anti-HCV-negative serum samples was used to evaluate the assay's specif-icity, including five serum samples with positive hepatitis B virus surface antibody (anti-HBs) status and five sera from Epstein-Barr virus-infected patients that had tested positive for heterophile antibodies. All serum samples had been stored at -80°C upon collection and had not been thawed until the time of assay. Serum immunoglobulins G (IgG) fraction was purified using protein G-Sepharose (GE Healthcare, Orsay, France The HCV focus reduction neutralization assay was performed in 96-well microtiter plates. Serial dilutions of purified IgG (10 μg) ranging from 1:10 to 1:1,280 were established. Each dilution was tested twice. 25 μL of each sample was mixed with 25 μL of virus (100 FFU) in 96well microtiter plates and incubated for 1 hour at 37°C, 5% CO 2 . A volume of 100 μL of Huh-7 cell suspension (10,000 cells/well) in culture medium was added and incubated for 5 hours at 37°C, 5% CO2. After 5 hours, the supernatants were removed and 100 μL of culture medium were added to the monolayers. After 72 hours, the cells were fixed with paraformaldehyde and permeabilized with 0.5% Triton X-100. Primary antibody (a HCVpositive patient serum inactivated at 56°C) was diluted to 1:500 prior to use and then incubated for 1 h at room temperature. A peroxidase-coupled, Fc-specific anti-human IgG antibody (Sigma, Saint Quentin Fallavier, France) diluted to 1:200 was dispensed onto the cell monolayer and incubated for 30 min at room temperature. The reaction was developed with DAB peroxidase substrate (Sigma, Saint Quentin Fallavier, France) and stopped after 10 min of incubation with distilled water. The number of HCV foci in each dilution was determined. Controls were included in each assay (non-neutralized virus, purified IgG from each patient at a 1:10 dilution). The dilution that neutralized 50% of the virus was calculated by curvilinear regression analysis using XLSTAT 2006 software (Addinsoft SARL, Paris, France) [35] . Each titer was deter-mined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. Titers were expressed as logarithmic values and means ± standard deviation were calculated. Student's t-test was used to compare data between groups. p values below 0.05 were considered to be significant.

How many people have persistent hepatitis C virus?

A focus reduction neutralization assay for hepatitis C virus neutralizing antibodies https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1852297/ SHA: ee8dca216514deeed4c9415bc2ad8a78dc3d9670 Authors: Fournier, Carole; Duverlie, Gilles; François, Catherine; Schnuriger, Aurelie; Dedeurwaerder, Sarah; Brochot, Etienne; Capron, Dominique; Wychowski, Czeslaw; Thibault, Vincent; Castelain, Sandrine Date: 2007-03-30 DOI: 10.1186/1743-422x-4-35 License: cc-by Abstract: BACKGROUND/AIM: The role of humoral immunity in hepatitis C virus (HCV) infection is poorly understood. Nevertheless, there is increasing interest in characterizing the neutralizing antibodies in the serum of HCV-infected patients. Focus reduction assays have been widely used to evaluate neutralizing antibody responses against a range of non-cytopathic viruses. Based on the recent development of a HCV cell culture system using the genotype 2 JFH-1-strain, we developed a focus reduction assay for HCV-neutralizing antibodies. METHODS: The focus reduction assay was based on a standard microneutralization assay in which immunostained foci on tissue culture plates are counted. The neutralizing anti-HCV antibodies titers of purified serum immunoglobulin samples from seventy-seven individuals were determined using a 50% focus reduction neutralization assay. Each titer was determined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. IgG antibodies were first purified from each serum in order to avoid the facilitating effect of HDL on HCV entry. RESULTS: The assay's cut-off using an ELISA and RNA HCV-negative samples was found to be 1.25 log, corresponding to a dilution of 1:18. The assay was compared with a commercial HCV ELISA and exhibited specificity and sensitivity values of 100% and 96.5%, respectively, and good reproducibility (with intra-assay and inter-assay coefficients of variation of 6.7% and 12.6%, respectively). The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples. The neutralizing antibodies titers were 2.13 log (1:134) for homologous samples from HCV genotype 2 infected patients harboring the same genotype as JFH-1 and 1.93 log (1:85) for heterologous samples from patients infected by genotypes other than type 2. These results confirm the presence of broadly cross-neutralizing antibodies already reported using the HCV pseudoparticles system. CONCLUSION: This study presents a simple, specific and reproducible cell culture-based assay for determination of HCV-neutralizing antibodies in human sera. The assay should be an important tool for gauging the relationship between the neutralizing antibodies response and viral load kinetics in acutely or chronically infected patients and for investigating the possible eradication or prevention of HCV infection by neutralizing antibodies. Text: Hepatitis C virus (HCV, a member of the Flaviviridae family) is an enveloped, positive-stranded RNA virus that preferentially replicates in hepatocytes. At least 170 million people worldwide are persistently infected with hepatitis C virus. Chronic HCV infection is associated with a significant risk of progression to cirrhosis and hepatocellular carcinoma [1] . Antiviral therapy with pegylated alpha-interferon and ribavirin (the current best therapeutic regimen) is only successful in about 50% of all treated patients. Better knowledge of the viral and host factors that determine HCV clearance or persistence during the acute stage of infection is needed in order to improve antiviral therapy and to develop efficient vaccines. Studies focusing on innate and cellular immune responses have shown that a sufficiently large HCV inoculum is able to evade, subvert or circumvent the host's defences. At present, the chimpanzee is the only reliable experimental animal model in which the initial post-HCV infection events and the efficacy of vaccine candidates can be evaluated [2] . It has been shown that HCV-specific T-cell immunity is important in the control of HCV infection [3, 4] . Several studies have indicated a role for humoral immunity in the acute stage of HCV infection but this aspect remains poorly characterized. The E1 and E2 glycoproteins are thought to be the viral attachment proteins and thus the main targets for HCV-neutralizing antibodies; identification of protective epitopes conserved across different strains of HCV is therefore a major challenge in vaccine design. A number of antibodies capable of blocking E2 binding to cells or cell receptors have been described, [5] [6] [7] [8] some of which neutralize HCV entry in animal or cellular models [9, 10] . Cell entry has been shown to involve several surface molecules (notably including the tetraspanin CD81 and the SR-BI receptor [11, 12] ), although further studies are needed to better understand how viral entry occurs and how it might be neutralized. Detection of neutralizing antibodies in human blood had been problematical until an efficient and reliable cell culture system for HCV became available. Hence, the development of an in vitro neutralization assay for HCV could be extremely valuable for characterizing the humoral immune response to HCV and for evaluating the potential of passive and active immunization against hepatitis C. Recent studies using an in vitro neutralization assay system (based on infectious retroviral pseudoparticles (HCVpp) bearing HCV envelope glycoproteins) have confirmed that HCV-infected patient sera can indeed neutralize infection [13, 14] . However, it has also been shown that the neutralizing activity of antibodies from HCV-infected patients is attenuated by a factor present in human serum, identified as the highdensity lipoprotein (HDL) fraction [11, 13, 15] . HDL facilitation of HCVpp entry is a post-binding event [16] , sug-gesting that HDLs favour internalization of virions and thus the latter's escape from neutralizing antibodies. Recently, an HCV cell culture model (HCVcc) has been developed [17] [18] [19] , allowing the production of virus particles that can be efficiently propagated in cell culture. Some preliminary neutralization assays have been carried out by these authors. In this study, we describe how we set up a standardized focus reduction neutralization assay based on HCVcc. Focus reduction assays have been widely used to evaluate the neutralizing antibody responses to viruses that can form foci in infected cells. Following the recent development of the HCVcc model, the principle of the focus reduction assay has been applied to HCV-neutralizing antibodies detection. The JFH-1 HCV 2a viral strain was grown on a Huh-7 human hepatoma cell line. After three days of infection and cell permeabilization, detection of the HCV foci was carried out using an inactivated HCVpositive patient serum primary antibody and a peroxidase-coupled, Fc-specific anti-human IgG-antibody. The reaction was revealed with DAB peroxidase substrate. The viral foci were thus stained brown, making them easy to count (see Fig. 1a ). It has been recently shown that the neutralizing activity of HCV antibodies is attenuated by a serum factor associated with the HDL fraction. Hence, HDLs were able to facilitate HCVpp and HCVcc entry via a mechanism which depended on the expression of the scavenger receptor BI (SR-BI) and its selective lipid-uptake function [11, 15, 16, 20] . In view of the role of HDL in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibody titer (see Fig. 1b ). The specificity of the HCV neutralization assay was assessed by testing 20 anti-HCV-ELISA-negative samples, including five positive for hepatitis B virus surface antibodies (anti-HBs) and five positive for heterophile antibodies. All samples tested negative with two commercial anti-HCV antibody detection assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany; Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom) and HCV-RNA-negative with a qualitative, commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France). These anti-HCV-negative samples were compared with 11 samples from patients chronically infected with HCV genotype 2. The neutralization titers of anti-HCV-negative serum samples are shown in Fig. 2 ., with a mean value of 1.083 ± 0.083 (corresponding to a dilution of 1:12). The assay's cut-off (determined as the mean value for negative samples plus two standard deviations) corresponded to a dilution of 1:18. The assay exhibited specificity and sensibility values of 100% and 96.5%, respectively. The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples (data not shown). Conversely, the chronically HCV genotype 2-positive samples displayed strong reactions, with a mean value of 2.128 ± 0.365 (corresponding to a dilution of 1:134) (p < 0.001). Inter-assay variability was determined by testing one HCV genotype 2 sample in 10 consecutive experiments (n = 10), whereas intra-assay variability was evaluated by testing the same sample 10 times (n = 10) in the same experiment, whilst running the dilution series. The intra-assay and inter-assay coefficients of variation (CV) of the log neutralization titers were 6.7% and 12.6%, respectively. Fifty-seven HCV-positive antibodies samples were evaluated using the HCV focus reduction neutralization assay. The genotypes were distributed as follows; for types 1a, 1b, 2, 3, 4 and 5, we studied 11, 11, 11, 12, 10 and 2 samples, respectively. The mean values of the different genotypes is shown in Fig. 3 . and Table 1 . The mean log neutralization titers for genotypes 1a, 2 and 3 are very similar (2.046 ± 0.671 for genotype 1a, 2.128 ± 0.365 for genotype 2 and 2.148 ± 0.478 for genotype 3). The mean average values are lower for genotype 1b (1.747 ± 0.462) and genotype 4 (1.786 ± 0.236). Strikingly, very high heterologous titers were observed for five patients -three infected with HCV genotype 1a and two infected with HCV genotype 3 (see Fig. 3a ). There were too few genotype 5 samples to compare with the other genotypes but the corresponding results nevertheless indicate that the neutralization assay is suitable for this genotype. The two The distribution of the log neutralization titers across all the HCV ELISA and RNA-positive samples as a function of the HCV genotype is shown in Fig. 3b . More than 60% of the neutralizing antibodies titers fell in the range from 1.7 to 2.69 log titers, corresponding to dilutions of 1:50 and 1:500, respectively. Overall, 3.5% of the samples displayed a titer greater than log 3.0 (1:1000) and, conversely, 3.5% displayed a titer below the cut-off value, i.e. log 1.25 (1:10). Thus, of 57 HCV-infected patients, only two did not test positive for neutralizing antibodies in this assay (the titers were 0.960 and 0.932, respectively). The role of neutralizing antibodies during acute and chronic viral infection remains an important question and has generated controversial results. Initially, the presence of neutralizing antibodies was shown to control the HCV load and to contribute to viral eradication in patients capable of clearing the infection [13] . In other studies, the appearance of neutralizing antibodies was delayed and restricted to IgG1 antibodies in patients who develop a chronic infection [2, 21] . The chimpanzee model has been critical for the study of HCV transmission and host immune responses; however, neutralizing antibodies were not detected in some animals that resolved their infection -suggesting a minimal role in viral clearance, as also observed in human studies [14, 15] . Experimentally infected chimpanzees and naturally infected humans can be re-infected with homologous and heterologous HCV strains, suggesting that the humoral immunity that develops after spontaneous resolution of acute hepatitis C is not sterilizing [22] [23] [24] . During chronic infection in humans, the presence and/or production of neutralizing antibodies do not suffice for curing the infection but could regulate the spread of the virus. Thus, it can be postulated that during chronic infection, viral mutants can continuously escape the renewed production of neutralizing antibodies. Retroviral pseudoparticles have been used to develop a very interesting tool for measuring neutralizing antibodies in vitro [14] . The assay has demonstrated the presence of HCV-neutralizing antibodies in human sera with relatively high titers (>1:320) and broadly neutralizing activity against different HCV genotypes. However, this model does not represent genuine HCV virions; in particular, the budding of retroviral particles is thought to be very different and may involve a variety of cellular pathways. Characterization of infectious retroviral pseudotype particles bearing HCV glycoproteins have been shown to be very heterogeneous, and so it is possible that these pseudoparticles may not be as relevant as the native HCV virions [25] . The recent development of a cell culture model for HCV enables the production of native HCV virions that can be efficiently propagated in cell culture [17] [18] [19] . This cell culture system has allowed us to develop a neutralization assay for evaluating the level and the proportion of HCVneutralizing antibodies in chronically infected HCV patients. We analysed a number of parameters (such as practicability, reproducibility and specificity) and tested the effect of a range of variables (viral inoculum size, incubation time, fixation and permeabilization methods, blocking and revelation reagents) on these parameters (data not shown). Overall, the neutralization assay described in this study performs similarly to standardized neutralization assays for many other viruses [26] [27] [28] . The assay relies on the ability of the specific JFH-1 genotype 2 viral strain to replicate and multiply on a Huh-7 human hepatoma cell line in a cell culture model, enabling the rapid detection of viral foci after 72 hours of infection. Moreover, no secondary foci were detectable at this time point. Fixation with paraformaldehyde and permeabilization with Triton X-100 were chosen in order to preserve antigenicity and prevent the cell monolayer from detaching during washes. Development with DAB peroxide substrate made it easy to count specifically coloured viral foci. The viral inoculum size is an important parameter; it has to be low enough to enable good assay sensitivity but high enough to produce a statistically significant number of foci, i.e. allowing the reduction in the number of foci (and thus the effect of neutralization) to be monitored. Thus, 100 FFUs were used as the inoculum in this neutralization assay. In order to test different human samples, we had to take into account the ability of HDL to facilitate HCVcc entry via a mechanism which depends on expression of the scavenger receptor BI [11, 15, 16, 20] . Given HDL's role in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibodies titer; this frees the assay of the risk of non-specific neutralization activity of the serum via the effects of HDL, the complement system and/or serum amyloid A protein (SAA) [29] . The HCV neutralization assay exhibited good reproducibility, for both duplicate assays and independent tests. As expected, the intra-assay coefficient of variation (CV) was lower than the interassay CV. The test also showed good specificity, since there was no interaction with anti-HIV, anti-HBV or heterophile antibodies. Very low titers were found with HCV ELISA and RNA-negative samples, and the assay's cut-off was determined as the mean titer for negative samples plus two standard deviations (1.25 log, corresponding to a dilution of 1:18). Given that only the JFH-1 strain of HCV genotype 2a was available for the assay, we evaluated the neutralization titer of sera from patients chronically infected with other HCV genotypes, i.e. 1, 2, 3, 4 and 5. Most of these sera were detected as positive by the neutralization assay, except for two sera from HCV genotype 1-infected patients. These two samples presented a high specific antibody ratio according to the ELISA but only very low inhibition by neutralization assay (far below the cut-off, in fact). We conclude that either the samples lacked neutralizing antibodies or that any such antibodies that were present did not cross-neutralize with HCV genotype 2a. The sensitivity was 100% -not only for genotype 2 (the genotype of the strain used for the assay) but also for other HCV genotypes (except genotype 1). HCV genotype 5 antibodies were also measured but there were too few samples for accurate testing. Moreover, the positive sera (96.5%) had comparable and significantly high titers (1.99 ± 0.63), whatever the genotype. This finding suggests that most neutralizing antibodies are cross-reactive. Another possibility is that most of the patients had been previously infected by a genotype 2 strain. However, this is unlikely because few genotype 2 strains are circulating in France [30] . As expected for a neutralization test, the assay presented in the present study appeared to be very specific (independently of the genotype) and usable in most circumstances. For most viral infections, neutralization assays such as that described in this study are used as reference assays. Thus, we are confident that as other HCVcc genotypes become available, these assays will replace the pseudoparticle assay in the near future because they are probably more relevant. Our assay is somewhat time-consuming and could be simplified by using one dilution to count the foci; however, this type of "short cut" would make it difficult to extrapolate to the dilution neutralizing 50% of the inoculum. Another approach would consist in using recombinant HCV capable of expressing reporter genes (such as luciferase) in order to use a single dilution and obtain a quantitative result [31] . However, further neutralization studies using other genotypes are needed in order to complete our observations and to char- A simple, specific and reproducible cell culture-based neutralization assay was developed for the determination of neutralizing anti-HCV antibodies in human sera. This test should be an important tool for gauging the relationship between the neutralizing response and viral load kinetics in acutely and chronically infected patients. The Huh-7 human hepatoma cells [32] were grown in Dulbecco's minimum essential medium (Invitrogen) supplemented with 10% fetal bovine serum. All cell cultures were maintained in 5% CO 2 at 37°C. The plasmid pJFH-1 containing the full-length cDNA of the JFH-1 isolate (which belongs to subtype 2a (GenBank accession no. AB047639)), was a gift from Dr Wakita (Department of Microbiology, Tokyo Metropolitan Institute for Neuroscience, Tokyo, Japan) and has been described previously [17] . To generate genomic HCV RNA, the plasmid pJFH-1 was linearized at the 3' end of the HCV cDNA and used as a template for in vitro transcription, as described previously [33] . Viral stocks were obtained by harvesting cell culture supernatants and freezing them at -80°C. Virus titration was performed on Huh-7 cells with 6-well microtiter plates (Corning, NY) 72 hours after incubation, by immunostaining the cells with antibodies from a HCV-positive patient serum that had previously been inactivated at 56°C (see the section on the virus neutralization assay). The viral titer was determined in triplicate from the mean number of foci and expressed as focus forming units/mL (FFU/mL). Seventy-seven human serum samples were tested. Collection of the sera was approved by the local Ethics Committee and informed consent had been obtained from the donors. Fifty-seven of these samples were obtained from chronically infected HCV patients. The presence of HCV antibodies was determined and confirmed using two third-generation HCV EIA assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany and Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom). HCV RNA was determined with a qualitative commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France) and HCV genotyping was performed by direct sequencing, as described elsewhere [34] . The genotypes were distributed as follows: 11, 11, 11, 12, 10 and 2 samples of types 1a, 1b, 2, 3, 4 and 5, respectively. A set of 20 anti-HCV-negative serum samples was used to evaluate the assay's specif-icity, including five serum samples with positive hepatitis B virus surface antibody (anti-HBs) status and five sera from Epstein-Barr virus-infected patients that had tested positive for heterophile antibodies. All serum samples had been stored at -80°C upon collection and had not been thawed until the time of assay. Serum immunoglobulins G (IgG) fraction was purified using protein G-Sepharose (GE Healthcare, Orsay, France The HCV focus reduction neutralization assay was performed in 96-well microtiter plates. Serial dilutions of purified IgG (10 μg) ranging from 1:10 to 1:1,280 were established. Each dilution was tested twice. 25 μL of each sample was mixed with 25 μL of virus (100 FFU) in 96well microtiter plates and incubated for 1 hour at 37°C, 5% CO 2 . A volume of 100 μL of Huh-7 cell suspension (10,000 cells/well) in culture medium was added and incubated for 5 hours at 37°C, 5% CO2. After 5 hours, the supernatants were removed and 100 μL of culture medium were added to the monolayers. After 72 hours, the cells were fixed with paraformaldehyde and permeabilized with 0.5% Triton X-100. Primary antibody (a HCVpositive patient serum inactivated at 56°C) was diluted to 1:500 prior to use and then incubated for 1 h at room temperature. A peroxidase-coupled, Fc-specific anti-human IgG antibody (Sigma, Saint Quentin Fallavier, France) diluted to 1:200 was dispensed onto the cell monolayer and incubated for 30 min at room temperature. The reaction was developed with DAB peroxidase substrate (Sigma, Saint Quentin Fallavier, France) and stopped after 10 min of incubation with distilled water. The number of HCV foci in each dilution was determined. Controls were included in each assay (non-neutralized virus, purified IgG from each patient at a 1:10 dilution). The dilution that neutralized 50% of the virus was calculated by curvilinear regression analysis using XLSTAT 2006 software (Addinsoft SARL, Paris, France) [35] . Each titer was deter-mined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. Titers were expressed as logarithmic values and means ± standard deviation were calculated. Student's t-test was used to compare data between groups. p values below 0.05 were considered to be significant.

What is the long-term risk of chronic hepatitis C infection?

A focus reduction neutralization assay for hepatitis C virus neutralizing antibodies https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1852297/ SHA: ee8dca216514deeed4c9415bc2ad8a78dc3d9670 Authors: Fournier, Carole; Duverlie, Gilles; François, Catherine; Schnuriger, Aurelie; Dedeurwaerder, Sarah; Brochot, Etienne; Capron, Dominique; Wychowski, Czeslaw; Thibault, Vincent; Castelain, Sandrine Date: 2007-03-30 DOI: 10.1186/1743-422x-4-35 License: cc-by Abstract: BACKGROUND/AIM: The role of humoral immunity in hepatitis C virus (HCV) infection is poorly understood. Nevertheless, there is increasing interest in characterizing the neutralizing antibodies in the serum of HCV-infected patients. Focus reduction assays have been widely used to evaluate neutralizing antibody responses against a range of non-cytopathic viruses. Based on the recent development of a HCV cell culture system using the genotype 2 JFH-1-strain, we developed a focus reduction assay for HCV-neutralizing antibodies. METHODS: The focus reduction assay was based on a standard microneutralization assay in which immunostained foci on tissue culture plates are counted. The neutralizing anti-HCV antibodies titers of purified serum immunoglobulin samples from seventy-seven individuals were determined using a 50% focus reduction neutralization assay. Each titer was determined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. IgG antibodies were first purified from each serum in order to avoid the facilitating effect of HDL on HCV entry. RESULTS: The assay's cut-off using an ELISA and RNA HCV-negative samples was found to be 1.25 log, corresponding to a dilution of 1:18. The assay was compared with a commercial HCV ELISA and exhibited specificity and sensitivity values of 100% and 96.5%, respectively, and good reproducibility (with intra-assay and inter-assay coefficients of variation of 6.7% and 12.6%, respectively). The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples. The neutralizing antibodies titers were 2.13 log (1:134) for homologous samples from HCV genotype 2 infected patients harboring the same genotype as JFH-1 and 1.93 log (1:85) for heterologous samples from patients infected by genotypes other than type 2. These results confirm the presence of broadly cross-neutralizing antibodies already reported using the HCV pseudoparticles system. CONCLUSION: This study presents a simple, specific and reproducible cell culture-based assay for determination of HCV-neutralizing antibodies in human sera. The assay should be an important tool for gauging the relationship between the neutralizing antibodies response and viral load kinetics in acutely or chronically infected patients and for investigating the possible eradication or prevention of HCV infection by neutralizing antibodies. Text: Hepatitis C virus (HCV, a member of the Flaviviridae family) is an enveloped, positive-stranded RNA virus that preferentially replicates in hepatocytes. At least 170 million people worldwide are persistently infected with hepatitis C virus. Chronic HCV infection is associated with a significant risk of progression to cirrhosis and hepatocellular carcinoma [1] . Antiviral therapy with pegylated alpha-interferon and ribavirin (the current best therapeutic regimen) is only successful in about 50% of all treated patients. Better knowledge of the viral and host factors that determine HCV clearance or persistence during the acute stage of infection is needed in order to improve antiviral therapy and to develop efficient vaccines. Studies focusing on innate and cellular immune responses have shown that a sufficiently large HCV inoculum is able to evade, subvert or circumvent the host's defences. At present, the chimpanzee is the only reliable experimental animal model in which the initial post-HCV infection events and the efficacy of vaccine candidates can be evaluated [2] . It has been shown that HCV-specific T-cell immunity is important in the control of HCV infection [3, 4] . Several studies have indicated a role for humoral immunity in the acute stage of HCV infection but this aspect remains poorly characterized. The E1 and E2 glycoproteins are thought to be the viral attachment proteins and thus the main targets for HCV-neutralizing antibodies; identification of protective epitopes conserved across different strains of HCV is therefore a major challenge in vaccine design. A number of antibodies capable of blocking E2 binding to cells or cell receptors have been described, [5] [6] [7] [8] some of which neutralize HCV entry in animal or cellular models [9, 10] . Cell entry has been shown to involve several surface molecules (notably including the tetraspanin CD81 and the SR-BI receptor [11, 12] ), although further studies are needed to better understand how viral entry occurs and how it might be neutralized. Detection of neutralizing antibodies in human blood had been problematical until an efficient and reliable cell culture system for HCV became available. Hence, the development of an in vitro neutralization assay for HCV could be extremely valuable for characterizing the humoral immune response to HCV and for evaluating the potential of passive and active immunization against hepatitis C. Recent studies using an in vitro neutralization assay system (based on infectious retroviral pseudoparticles (HCVpp) bearing HCV envelope glycoproteins) have confirmed that HCV-infected patient sera can indeed neutralize infection [13, 14] . However, it has also been shown that the neutralizing activity of antibodies from HCV-infected patients is attenuated by a factor present in human serum, identified as the highdensity lipoprotein (HDL) fraction [11, 13, 15] . HDL facilitation of HCVpp entry is a post-binding event [16] , sug-gesting that HDLs favour internalization of virions and thus the latter's escape from neutralizing antibodies. Recently, an HCV cell culture model (HCVcc) has been developed [17] [18] [19] , allowing the production of virus particles that can be efficiently propagated in cell culture. Some preliminary neutralization assays have been carried out by these authors. In this study, we describe how we set up a standardized focus reduction neutralization assay based on HCVcc. Focus reduction assays have been widely used to evaluate the neutralizing antibody responses to viruses that can form foci in infected cells. Following the recent development of the HCVcc model, the principle of the focus reduction assay has been applied to HCV-neutralizing antibodies detection. The JFH-1 HCV 2a viral strain was grown on a Huh-7 human hepatoma cell line. After three days of infection and cell permeabilization, detection of the HCV foci was carried out using an inactivated HCVpositive patient serum primary antibody and a peroxidase-coupled, Fc-specific anti-human IgG-antibody. The reaction was revealed with DAB peroxidase substrate. The viral foci were thus stained brown, making them easy to count (see Fig. 1a ). It has been recently shown that the neutralizing activity of HCV antibodies is attenuated by a serum factor associated with the HDL fraction. Hence, HDLs were able to facilitate HCVpp and HCVcc entry via a mechanism which depended on the expression of the scavenger receptor BI (SR-BI) and its selective lipid-uptake function [11, 15, 16, 20] . In view of the role of HDL in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibody titer (see Fig. 1b ). The specificity of the HCV neutralization assay was assessed by testing 20 anti-HCV-ELISA-negative samples, including five positive for hepatitis B virus surface antibodies (anti-HBs) and five positive for heterophile antibodies. All samples tested negative with two commercial anti-HCV antibody detection assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany; Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom) and HCV-RNA-negative with a qualitative, commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France). These anti-HCV-negative samples were compared with 11 samples from patients chronically infected with HCV genotype 2. The neutralization titers of anti-HCV-negative serum samples are shown in Fig. 2 ., with a mean value of 1.083 ± 0.083 (corresponding to a dilution of 1:12). The assay's cut-off (determined as the mean value for negative samples plus two standard deviations) corresponded to a dilution of 1:18. The assay exhibited specificity and sensibility values of 100% and 96.5%, respectively. The assay did not show any cross-reactivity with anti-HIV, anti-HBs or heterophile antibody-positive samples (data not shown). Conversely, the chronically HCV genotype 2-positive samples displayed strong reactions, with a mean value of 2.128 ± 0.365 (corresponding to a dilution of 1:134) (p < 0.001). Inter-assay variability was determined by testing one HCV genotype 2 sample in 10 consecutive experiments (n = 10), whereas intra-assay variability was evaluated by testing the same sample 10 times (n = 10) in the same experiment, whilst running the dilution series. The intra-assay and inter-assay coefficients of variation (CV) of the log neutralization titers were 6.7% and 12.6%, respectively. Fifty-seven HCV-positive antibodies samples were evaluated using the HCV focus reduction neutralization assay. The genotypes were distributed as follows; for types 1a, 1b, 2, 3, 4 and 5, we studied 11, 11, 11, 12, 10 and 2 samples, respectively. The mean values of the different genotypes is shown in Fig. 3 . and Table 1 . The mean log neutralization titers for genotypes 1a, 2 and 3 are very similar (2.046 ± 0.671 for genotype 1a, 2.128 ± 0.365 for genotype 2 and 2.148 ± 0.478 for genotype 3). The mean average values are lower for genotype 1b (1.747 ± 0.462) and genotype 4 (1.786 ± 0.236). Strikingly, very high heterologous titers were observed for five patients -three infected with HCV genotype 1a and two infected with HCV genotype 3 (see Fig. 3a ). There were too few genotype 5 samples to compare with the other genotypes but the corresponding results nevertheless indicate that the neutralization assay is suitable for this genotype. The two The distribution of the log neutralization titers across all the HCV ELISA and RNA-positive samples as a function of the HCV genotype is shown in Fig. 3b . More than 60% of the neutralizing antibodies titers fell in the range from 1.7 to 2.69 log titers, corresponding to dilutions of 1:50 and 1:500, respectively. Overall, 3.5% of the samples displayed a titer greater than log 3.0 (1:1000) and, conversely, 3.5% displayed a titer below the cut-off value, i.e. log 1.25 (1:10). Thus, of 57 HCV-infected patients, only two did not test positive for neutralizing antibodies in this assay (the titers were 0.960 and 0.932, respectively). The role of neutralizing antibodies during acute and chronic viral infection remains an important question and has generated controversial results. Initially, the presence of neutralizing antibodies was shown to control the HCV load and to contribute to viral eradication in patients capable of clearing the infection [13] . In other studies, the appearance of neutralizing antibodies was delayed and restricted to IgG1 antibodies in patients who develop a chronic infection [2, 21] . The chimpanzee model has been critical for the study of HCV transmission and host immune responses; however, neutralizing antibodies were not detected in some animals that resolved their infection -suggesting a minimal role in viral clearance, as also observed in human studies [14, 15] . Experimentally infected chimpanzees and naturally infected humans can be re-infected with homologous and heterologous HCV strains, suggesting that the humoral immunity that develops after spontaneous resolution of acute hepatitis C is not sterilizing [22] [23] [24] . During chronic infection in humans, the presence and/or production of neutralizing antibodies do not suffice for curing the infection but could regulate the spread of the virus. Thus, it can be postulated that during chronic infection, viral mutants can continuously escape the renewed production of neutralizing antibodies. Retroviral pseudoparticles have been used to develop a very interesting tool for measuring neutralizing antibodies in vitro [14] . The assay has demonstrated the presence of HCV-neutralizing antibodies in human sera with relatively high titers (>1:320) and broadly neutralizing activity against different HCV genotypes. However, this model does not represent genuine HCV virions; in particular, the budding of retroviral particles is thought to be very different and may involve a variety of cellular pathways. Characterization of infectious retroviral pseudotype particles bearing HCV glycoproteins have been shown to be very heterogeneous, and so it is possible that these pseudoparticles may not be as relevant as the native HCV virions [25] . The recent development of a cell culture model for HCV enables the production of native HCV virions that can be efficiently propagated in cell culture [17] [18] [19] . This cell culture system has allowed us to develop a neutralization assay for evaluating the level and the proportion of HCVneutralizing antibodies in chronically infected HCV patients. We analysed a number of parameters (such as practicability, reproducibility and specificity) and tested the effect of a range of variables (viral inoculum size, incubation time, fixation and permeabilization methods, blocking and revelation reagents) on these parameters (data not shown). Overall, the neutralization assay described in this study performs similarly to standardized neutralization assays for many other viruses [26] [27] [28] . The assay relies on the ability of the specific JFH-1 genotype 2 viral strain to replicate and multiply on a Huh-7 human hepatoma cell line in a cell culture model, enabling the rapid detection of viral foci after 72 hours of infection. Moreover, no secondary foci were detectable at this time point. Fixation with paraformaldehyde and permeabilization with Triton X-100 were chosen in order to preserve antigenicity and prevent the cell monolayer from detaching during washes. Development with DAB peroxide substrate made it easy to count specifically coloured viral foci. The viral inoculum size is an important parameter; it has to be low enough to enable good assay sensitivity but high enough to produce a statistically significant number of foci, i.e. allowing the reduction in the number of foci (and thus the effect of neutralization) to be monitored. Thus, 100 FFUs were used as the inoculum in this neutralization assay. In order to test different human samples, we had to take into account the ability of HDL to facilitate HCVcc entry via a mechanism which depends on expression of the scavenger receptor BI [11, 15, 16, 20] . Given HDL's role in HCV entry, immunoglobulins were purified from each serum sample prior to determination of the neutralizing antibodies titer; this frees the assay of the risk of non-specific neutralization activity of the serum via the effects of HDL, the complement system and/or serum amyloid A protein (SAA) [29] . The HCV neutralization assay exhibited good reproducibility, for both duplicate assays and independent tests. As expected, the intra-assay coefficient of variation (CV) was lower than the interassay CV. The test also showed good specificity, since there was no interaction with anti-HIV, anti-HBV or heterophile antibodies. Very low titers were found with HCV ELISA and RNA-negative samples, and the assay's cut-off was determined as the mean titer for negative samples plus two standard deviations (1.25 log, corresponding to a dilution of 1:18). Given that only the JFH-1 strain of HCV genotype 2a was available for the assay, we evaluated the neutralization titer of sera from patients chronically infected with other HCV genotypes, i.e. 1, 2, 3, 4 and 5. Most of these sera were detected as positive by the neutralization assay, except for two sera from HCV genotype 1-infected patients. These two samples presented a high specific antibody ratio according to the ELISA but only very low inhibition by neutralization assay (far below the cut-off, in fact). We conclude that either the samples lacked neutralizing antibodies or that any such antibodies that were present did not cross-neutralize with HCV genotype 2a. The sensitivity was 100% -not only for genotype 2 (the genotype of the strain used for the assay) but also for other HCV genotypes (except genotype 1). HCV genotype 5 antibodies were also measured but there were too few samples for accurate testing. Moreover, the positive sera (96.5%) had comparable and significantly high titers (1.99 ± 0.63), whatever the genotype. This finding suggests that most neutralizing antibodies are cross-reactive. Another possibility is that most of the patients had been previously infected by a genotype 2 strain. However, this is unlikely because few genotype 2 strains are circulating in France [30] . As expected for a neutralization test, the assay presented in the present study appeared to be very specific (independently of the genotype) and usable in most circumstances. For most viral infections, neutralization assays such as that described in this study are used as reference assays. Thus, we are confident that as other HCVcc genotypes become available, these assays will replace the pseudoparticle assay in the near future because they are probably more relevant. Our assay is somewhat time-consuming and could be simplified by using one dilution to count the foci; however, this type of "short cut" would make it difficult to extrapolate to the dilution neutralizing 50% of the inoculum. Another approach would consist in using recombinant HCV capable of expressing reporter genes (such as luciferase) in order to use a single dilution and obtain a quantitative result [31] . However, further neutralization studies using other genotypes are needed in order to complete our observations and to char- A simple, specific and reproducible cell culture-based neutralization assay was developed for the determination of neutralizing anti-HCV antibodies in human sera. This test should be an important tool for gauging the relationship between the neutralizing response and viral load kinetics in acutely and chronically infected patients. The Huh-7 human hepatoma cells [32] were grown in Dulbecco's minimum essential medium (Invitrogen) supplemented with 10% fetal bovine serum. All cell cultures were maintained in 5% CO 2 at 37°C. The plasmid pJFH-1 containing the full-length cDNA of the JFH-1 isolate (which belongs to subtype 2a (GenBank accession no. AB047639)), was a gift from Dr Wakita (Department of Microbiology, Tokyo Metropolitan Institute for Neuroscience, Tokyo, Japan) and has been described previously [17] . To generate genomic HCV RNA, the plasmid pJFH-1 was linearized at the 3' end of the HCV cDNA and used as a template for in vitro transcription, as described previously [33] . Viral stocks were obtained by harvesting cell culture supernatants and freezing them at -80°C. Virus titration was performed on Huh-7 cells with 6-well microtiter plates (Corning, NY) 72 hours after incubation, by immunostaining the cells with antibodies from a HCV-positive patient serum that had previously been inactivated at 56°C (see the section on the virus neutralization assay). The viral titer was determined in triplicate from the mean number of foci and expressed as focus forming units/mL (FFU/mL). Seventy-seven human serum samples were tested. Collection of the sera was approved by the local Ethics Committee and informed consent had been obtained from the donors. Fifty-seven of these samples were obtained from chronically infected HCV patients. The presence of HCV antibodies was determined and confirmed using two third-generation HCV EIA assays (Axsym ® HCV Version 3.0, Abbott, Wiesbaden, Germany and Vitros ® Anti-HCV reagent pack, Ortho-Clinical Diagnostic, High Wycombe, United Kingdom). HCV RNA was determined with a qualitative commercial assay (Cobas Amplicor HCV test Version 2.0, Roche Diagnostics, Meylan, France) and HCV genotyping was performed by direct sequencing, as described elsewhere [34] . The genotypes were distributed as follows: 11, 11, 11, 12, 10 and 2 samples of types 1a, 1b, 2, 3, 4 and 5, respectively. A set of 20 anti-HCV-negative serum samples was used to evaluate the assay's specif-icity, including five serum samples with positive hepatitis B virus surface antibody (anti-HBs) status and five sera from Epstein-Barr virus-infected patients that had tested positive for heterophile antibodies. All serum samples had been stored at -80°C upon collection and had not been thawed until the time of assay. Serum immunoglobulins G (IgG) fraction was purified using protein G-Sepharose (GE Healthcare, Orsay, France The HCV focus reduction neutralization assay was performed in 96-well microtiter plates. Serial dilutions of purified IgG (10 μg) ranging from 1:10 to 1:1,280 were established. Each dilution was tested twice. 25 μL of each sample was mixed with 25 μL of virus (100 FFU) in 96well microtiter plates and incubated for 1 hour at 37°C, 5% CO 2 . A volume of 100 μL of Huh-7 cell suspension (10,000 cells/well) in culture medium was added and incubated for 5 hours at 37°C, 5% CO2. After 5 hours, the supernatants were removed and 100 μL of culture medium were added to the monolayers. After 72 hours, the cells were fixed with paraformaldehyde and permeabilized with 0.5% Triton X-100. Primary antibody (a HCVpositive patient serum inactivated at 56°C) was diluted to 1:500 prior to use and then incubated for 1 h at room temperature. A peroxidase-coupled, Fc-specific anti-human IgG antibody (Sigma, Saint Quentin Fallavier, France) diluted to 1:200 was dispensed onto the cell monolayer and incubated for 30 min at room temperature. The reaction was developed with DAB peroxidase substrate (Sigma, Saint Quentin Fallavier, France) and stopped after 10 min of incubation with distilled water. The number of HCV foci in each dilution was determined. Controls were included in each assay (non-neutralized virus, purified IgG from each patient at a 1:10 dilution). The dilution that neutralized 50% of the virus was calculated by curvilinear regression analysis using XLSTAT 2006 software (Addinsoft SARL, Paris, France) [35] . Each titer was deter-mined as the log value of the reciprocal antibody dilution that reduced the number of viral foci by 50%. Titers were expressed as logarithmic values and means ± standard deviation were calculated. Student's t-test was used to compare data between groups. p values below 0.05 were considered to be significant.

What antiviral treatments are used for hepatitis C infection?

Can’t RIDD off viruses https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4061530/ SHA: ef58c6e2790539f30df14acc260ae2af4b5f3d1f Authors: Bhattacharyya, Sankar Date: 2014-06-18 DOI: 10.3389/fmicb.2014.00292 License: cc-by Abstract: The mammalian genome has evolved to encode a battery of mechanisms, to mitigate a progression in the life cycle of an invasive viral pathogen. Although apparently disadvantaged by their dependence on the host biosynthetic processes, an immensely faster rate of evolution provides viruses with an edge in this conflict. In this review, I have discussed the potential anti-virus activity of inositol-requiring enzyme 1 (IRE1), a well characterized effector of the cellular homeostatic response to an overloading of the endoplasmic reticulum (ER) protein-folding capacity. IRE1, an ER-membrane-resident ribonuclease (RNase), upon activation catalyses regulated cleavage of select protein-coding and non-coding host RNAs, using an RNase domain which is homologous to that of the known anti-viral effector RNaseL. The latter operates as part of the Oligoadenylate synthetase OAS/RNaseL system of anti-viral defense mechanism. Protein-coding RNA substrates are differentially treated by the IRE1 RNase to either augment, through cytoplasmic splicing of an intron in the Xbp1 transcript, or suppress gene expression. This referred suppression of gene expression is mediated through degradative cleavage of a select cohort of cellular RNA transcripts, initiating the regulated IRE1-dependent decay (RIDD) pathway. The review first discusses the anti-viral mechanism of the OAS/RNaseL system and evasion tactics employed by different viruses. This is followed by a review of the RIDD pathway and its potential effect on the stability of viral RNAs. I conclude with a comparison of the enzymatic activity of the two RNases followed by deliberations on the physiological consequences of their activation. Text: Establishment of infection by a virus, even in permissive host cells, is beset with a plethora of challenges from innate-antiviral and cell-death pathways. Therefore, the host response to a virus infection might prove to be inhibitory for the viral life cycle in a direct or an indirect manner. The direct mechanism involves expression of multiple anti-viral genes that have evolved to recognize, react, and thereby rid the infected host of the viral nucleic acid (Zhou et al., 1997; Thompson et al., 2011) . On the other hand the pathways, e.g., those that culminate in initiating an apoptotic death for the host cell, indirectly serve to limit the spread of virus (Roulston et al., 1999) . A major difference between these two mechanisms is that while the former response is transmissible to neighboring uninfected cells through interferon (IFN) signaling, the latter is observed mostly in cis. Recent reports, however, have demonstrated transmission of an apoptotic signal between cells that are in contact through gap junctions, although such a signaling from an virus infected host cell to an uninfected one is not known yet (Cusato et al., 2003; Udawatte and Ripps, 2005; Kameritsch et al., 2013) . Successful viral pathogens, through a process of active selection, have evolved to replicate and simultaneously evade or block either of these host responses. The viral nucleic acids which could be the genome (positive-sense singlestranded RNA virus) or RNA derived from transcription of the genome [negative-stranded single-sense RNA or double-stranded RNA (dsRNA) or DNA virus], offer critical targets for both detection and eradication. The viral nucleic acid targeting armaments in the host arsenal include those that recognize the associated molecular patterns like toll-like receptors (TLRs), DDX58 (or RIG-1), IFIH1 (or MDA5), IFIT proteins [IFN-stimulated genes (ISG)56 and ISF54], etc. (Aoshi et al., 2011; Bowzard et al., 2011; Jensen and Thomsen, 2012) . This is followed by IFN signaling and expression or activation of factors that target the inducer for degradation or modification like OAS/ribonuclease L (RNaseL) system, APOBEC3, MCPIP1, the ZC3HAV1/exosome system and RNAi pathways (Gao et al., 2002; Sheehy et al., 2002; Guo et al., 2007; Daffis et al., 2010; Sidahmed and Wilkie, 2010; Schmidt et al., 2012; Cho et al., 2013a; Lin et al., 2013) . In this review we focus on two proteins containing homologous RNase domains, RNaseL with a known direct antiviral function and Inositolrequiring enzyme 1 (IRE1 or ERN1) which has an RNaseL-like RNase domain with a known role in homeostatic response to unfolded proteins in the endoplasmic reticulum (ER) and a potential to function as an antiviral (Figure 1 ; Tirasophon et al., 2000) . In mammalian cells the tell-tale signs of RNA virus infection, like the presence of cytosolic RNA having 5 -ppp or extensive (>30 bp) dsRNA segments are detected by dedicated pathogen associated molecular pattern receptors (PAMPs) or pattern recognition receptors (PRRs) in the host cell, like RIG-1, MDA5, and the IFIT family of proteins (Aoshi et al., 2011; Bowzard et al., 2011; Vabret and Blander, 2013) . The transduction of a signal of this recognition results in the expression of IFN genes the products www.frontiersin.org FIGURE 1 | Schematic representation of the ribonuclease activity of IRE1 and RNaseL showing cross-talk between the paths catalysed by the enzymes. The figure shows activation of RNase activity following dimerization triggered by either accumulation of unfolded proteins in the ER-lumen or synthesis of 2-5A by the enzyme OAS, respectively, for IRE1 and RNaseL. The cleavage of Xbp1u by IRE1 releases an intron thus generating Xbp1s. The IRE1 targets in RIDD pathway or all RNaseL substrates are shown to undergo degradative cleavage. The cleavage products generated through degradation of the respective substrate is shown to potentially interact with RIG-I thereby leading to Interferon secretion and trans-activation of Oas genes through Interferon signaling. Abbreviations: RIG-I = retinoic acid inducible gene-I, Ifnb = interferon beta gene loci, IFN = interferons, ISG = interferon-sensitive genes, 2-5A = 2 -5 oligoadenylates. of which upon secretion outside the cell bind to cognate receptors, initiating further downstream signaling (Figure 1 ; Randall and Goodbourn, 2008) . The genes that are regulated as a result of IFN signaling are termed as IFN-stimulated or IFN-regulated genes (ISGs or IRGs; Sen and Sarkar, 2007; Schoggins and Rice, 2011) . Oligoadenylate synthetase or OAS genes are canonical ISGs that convert ATP into 2 -5 linked oligoadenylates (2-5A) by an unique enzymatic mechanism (Figure 1 ; Hartmann et al., 2003) . Further, they are RNA-binding proteins that function like PRRs, in a way that the 2-5A synthesizing activity needs to be induced through an interaction with dsRNA (Minks et al., 1979; Hartmann et al., 2003) . In a host cell infected by an RNA virus, such dsRNA is present in the form of replication-intermediates (RI), which are synthesized by the virus-encoded RNA-dependent RNA polymerases (RdRp) and subsequently used by the same enzyme to synthesize more genomic RNA, through asymmetric transcription (Weber et al., 2006) . However, the replications complexes (RCs) harboring these RI molecules are found secluded inside host-membrane derived vesicles, at least in positive-strand RNA viruses, a group which contains many human pathogens (Uchil and Satchidanandam, 2003; Denison, 2008) . Reports from different groups suggest OAS proteins to be distributed both in the cytoplasm as well as in membrane-associated fractions, perhaps indicating an evolution of the host anti-viral methodologies towards detection of the membrane-associated viral dsRNAs (Marie et al., 1990; Lin et al., 2009) . DNA viruses on the other hand, produce dsRNA by annealing of RNA derived from transcription of both strands in the same viral genomic loci, which are probably detected by the cytoplasmic pool of OAS proteins (Jacobs and Langland, 1996; Weber et al., 2006) . Post-activation the OAS enzymes synthesize 2-5A molecules in a non-processive reaction producing oligomers which, although potentially ranging in size from dimeric to multimeric, are functionally active only in a trimeric or tetrameric form (Dong et al., 1994; Sarkar et al., 1999; Silverman, 2007) . These small ligands, which bear phosphate groups (1-3) at the 5 end and hydroxyl groups at the 2 and 3 positions, serve as co-factor which can specifically interact with and thereby allosterically activate, existing RNaseL molecules (Knight et al., 1980; Zhou et al., 1997 Zhou et al., , 2005 Sarkar et al., 1999) . As part of a physiological control system these 2-5A oligomers are quite unstable in that they are highly susceptible to degradation by cellular 5 -phosphatases and PDE12 (2 -phosphodiesterase; Silverman et al., 1981; Johnston and Hearl, 1987; Kubota et al., 2004; Schmidt et al., 2012) . Viral strategies to evade or overcome this host defense mechanism ranges from preventing IFN signaling which would hinder the induction of OAS expression or thwarting activation of expressed OAS proteins by either shielding the viral dsRNA from interacting with it or modulating the host pathway to synthesize inactive 2-5A derivatives (Cayley et al., 1984; Hersh et al., 1984; Rice et al., 1985; Maitra et al., 1994; Beattie et al., 1995; Rivas et al., 1998; Child et al., 2004; Min and Krug, 2006; Sanchez and Mohr, 2007; Sorgeloos et al., 2013) . Shielding of viral RNA from interacting with OAS is possible through enclosure of dsRNA replication intermediates in membrane enclosed compartments as observed in many flaviviruses (Ahlquist, 2006; Miller and Krijnse-Locker, 2008; Miorin et al., 2013) . RNaseL is a 741 amino acid protein containing three predominantly structured region, an N-terminal ankyrin repeat domain (ARD), a middle catalytically inactive pseudo-kinase (PK) and a C-terminal RNase domain (Figure 2A ; Hassel et al., 1993; Zhou et al., 1993) . The activity of the RNase domain is negatively regulated by the ARD, which is relieved upon binding of 2-5A molecules to ankyrin repeats 2 and 4 followed by a conformational alteration (Figure 1 ; Hassel et al., 1993; Tanaka et al., 2004; Nakanishi et al., 2005) . In support of this contention, deletion of the ARD has been demonstrated to produce constitutively active RNaseL, although with dramatically lower RNase activity (Dong and Silverman, 1997) . However, recent reports suggest that while 2-5A links the ankyrin repeats from adjacent molecules leading to formation of dimer and higher order structures, at sufficiently high in vitro concentrations, RNaseL could oligomerize even in the absence of 2-5A . Nonetheless, in vivo the RNaseL nuclease activity still seems to be under the sole regulation of 2-5A (Al-Saif and Khabar, 2012) . In order to exploit this dependence, multiple viruses like mouse hepatitis virus (MHV) and rotavirus group A (RVA) have evolved to encode phosphodiesterases capable of hydrolysing the 2 -5 linkages in 2-5A and thereby attenuate the RNaseL cleavage activity (Zhao et al., 2012; Zhang et al., 2013) . In addition to 5 -phosphatases and 2 -phosphodiesterases to reduce ClustalW alignment of primary sequence from a segment of the PK domain indicating amino acid residues which are important for interacting with nucleotide cofactors. The conserved lysine residues, critical for this interaction (K599 for IRE1 and K392 in RNaseL) are underlined. (C) Alignment of the KEN domains in RNaseL and IRE1. The amino acids highlighted and numbered in IRE1 are critical for the IRE1 RNase activity (Tirasophon et al., 2000) . the endogenous 2-5A levels, mammalian genomes encode posttranscriptional and post-translation inhibitors of RNaseL activity in the form of microRNA-29 and the protein ABCE1 (RNaseL inhibitor or RLI), respectively (Bisbal et al., 1995; Lee et al., 2013) . Direct inhibition of RNaseL function is also observed upon infection by Picornaviruses through, either inducing the expression of ABCE1 or exercising a unique inhibitory property of a segment of the viral RNA (Martinand et al., 1998 (Martinand et al., , 1999 Townsend et al., 2008; Sorgeloos et al., 2013) . Once activated by 2-5A, RNaseL can degrade single-stranded RNA irrespective of its origin (virus or host) although there seems to exist a bias towards cleavage of viral RNA (Wreschner et al., 1981a; Silverman et al., 1983; Li et al., 1998) . RNA sequences that are predominantly cleaved by RNaseL are U-rich with the cleavage points being typically at the 3 end of UA or UG or UU di-nucleotides, leaving a 5 -OH and a 3 -monophosphate in the cleavage product (Floyd-Smith et al., 1981; Wreschner et al., 1981b) . A recent report shows a more general consensus of 5 -UNN-3 with the cleavage point between the second and the third nucleotide (Han et al., 2014) . Cellular targets of RNaseL include both ribosomal RNA (rRNA) and mRNAs, the latter predominantly representing genes involved in protein biosynthesis (Wreschner et al., 1981a; Al-Ahmadi et al., 2009; Andersen et al., 2009) . Additionally, RNaseL activity can also degrade specific ISG mRNA transcripts and thereby attenuate the effect of IFN signaling (Li et al., 2000) . Probably an evolution towards insulating gene expression from RNaseL activity is observed in the coding region of mammalian genes where the UU/UA dinucleotide frequency is rarer (Bisbal et al., 2000; Khabar et al., 2003; Al-Saif and Khabar, 2012) . Perhaps not surprisingly, with a much faster rate of evolution, similar observations have been made with respect to evasion of RNaseL mediated degradation by viral RNAs too (Han and Barton, 2002; Washenberger et al., 2007) . Moreover, nucleoside modifications in host mRNAs, rarely observed in viral RNAs, have also been shown to confer protection from RNaseL (Anderson et al., 2011) . In addition to directly targeting viral RNA, the reduction in functional ribosomes and ribosomal protein mRNA affects viral protein synthesis and replication in an indirect manner. Probably, as a reflection of these effects on cellular RNAs, RNaseL is implicated as one of the factors determining the anti-proliferative effect of IFN activity . The anti-viral activity of RNaseL extends beyond direct cleavage of viral RNA, through stimulation of RIG-I by the cleavage product (Malathi et al., , 2007 (Malathi et al., , 2010 . A global effect of RNaseL is observed in the form of autophagy induced through c-jun N-terminal kinase (JNK) signaling and apoptosis, probably as a consequence of rRNA cleavage (Li et al., 2004; Chakrabarti et al., 2012; Siddiqui and Malathi, 2012) . RNaseL has also been demonstrated to play a role in apoptotic cell death initiated by pharmacological agents extending the physiological role of this pathway beyond the boundary of being only an anti-viral mechanism (Castelli et al., 1997 (Castelli et al., , 1998 . The ER serves as a conduit for maturation of cellular proteins which are either secreted or destined to be associated with a membrane for its function. An exclusive microenvironment (high Calcium ion and unique ratio of reduced to oxidized glutathione) along with a battery of ER-lumen resident enzymes (foldases, chaperones, and lectins) catalyse/mediate the necessary folding, disulfide-bond formation, and glycosylation reactions (Schroder and Kaufman, 2005) . A perturbation of the folding capacity, due to either physiological disturbances or virus infection, can lead to an accumulation of unfolded proteins in the ER lumen, which signals an unfolded protein response (UPR). UPR encompasses a networked transcriptional and translational gene-expression program, initiated by three ER-membrane resident sensors namely IRE1 or ERN1, PKR-like ER Kinase (PERK or EIF2AK3) and activating transcription factor 6 (ATF6; Hetz, 2012) . IRE1 is a type I single-pass trans-membrane protein in which, similar to what is observed with RNaseL, the N-terminal resident in the ER lumen serves as sensor and the cytosolic C-terminal as the effector (Figure 1 ; Chen and Brandizzi, 2013) . The IRE1 coding gene is present in genomes ranging from yeast to mammals and in the latter is ubiquitously expressed in all tissues (Tirasophon et al., 1998) . Signal transduction by stimulated IRE1 initiates multiple gene regulatory pathways with either pro-survival or pro-apoptotic consequences (Kaufman, 1999) . During homeostasis or unstressed conditions the sensor molecules are monomeric, a state maintained co-operatively by the " absence" of unfolded proteins and the "presence" of HSPA5 (GRP78 or Bip, an ERresident chaperone) molecules bound to a membrane-proximal disordered segment of the protein in the ER-lumen-resident Nterminus (Credle et al., 2005) . Accumulated unfolded proteins in the lumen triggers coupling of this domain from adjacent sensor molecules through a combination of (a) titration of the bound HSPA5 chaperone molecules and (b) direct tethering by malfolded protein molecules (Shamu and Walter, 1996; Credle et al., 2005; Aragon et al., 2009; Korennykh et al., 2009) . Abutting of the luminal domains juxtapose the cytosolic C-terminal segments, leading to an aggregation of the IRE1 molecules into distinct ER-membrane foci (Kimata et al., 2007; Li et al., 2010) . The C-terminal segment has a serine/threonine kinase domain and a RNase domain homologous to that of RNaseL (Figure 1 ; Tirasophon et al., 1998 Tirasophon et al., , 2000 . A trans-autophosphorylation by the kinase domain allosterically activates the RNase domain (Tirasophon et al., 2000; Lee et al., 2008; Korennykh et al., 2009) . In fact, exogenous over-expression of IRE1 in mammalian cells lead to activation suggesting that, under homeostatic conditions, the non-juxtaposition of cytosolic domains maintains an inactive IRE1 (Tirasophon et al., 1998) . Once activated, IRE1 performs cleavage of a variety of RNA substrates mediated by its RNase domain, in addition to phosphorylating and thereby activating JNK (Cox and Walter, 1996; Urano et al., 2000) . Depending on the RNA substrate, the cleavage catalyzed by IRE1 RNase produces differential consequence. Although scission of the Xbp1 mRNA transcript at two internal positions is followed by splicing of the internal segment through ligation of the terminal cleavage products, that in all other known IRE1 target RNA is followed by degradation (Figure 1 ; Sidrauski and Walter, 1997; Calfon et al., 2002) . The latter mode of negative regulation of gene expression is termed as the regulated IRE1-dependent decay (RIDD) pathway (Hollien and Weissman, 2006; Oikawa et al., 2007; Iqbal et al., 2008; Lipson et al., 2008) . Gene transcripts regulated by RIDD pathway includes that from IRE1 (i.e., selftranscripts), probably in a negative feedback loop mechanism (Tirasophon et al., 2000) . In addition to protein coding RNA, RIDD pathway down-regulates the level of a host of microRNA precursors (pre-miRNAs) and can potentially cleave in the anticodon loop of tRNA Phe (Korennykh et al., 2011; Upton et al., 2012) . The IRE1 RNase domain cleaves the Xbp1u (u for unspliced) mRNA transcript at two precise internal positions within the open reading frame (ORF) generating three segments, the terminal two of which are ligated by a tRNA ligase in yeast and by an unknown ligase in mammalian cells, to produce the Xbp1s (s for spliced) mRNA transcript (Figure 1 ; Yoshida et al., 2001) . The Xbp1s thus generated has a longer ORF, which is created by a frame-shift in the coding sequence downstream of the splice site (Cox and Walter, 1996; Calfon et al., 2002) . A similar dual endonucleolytic cleavage is also observed to initiate the XRN1 and Ski2-3-8 dependent degradation of transcripts in the RIDD degradation pathway (Hollien and Weissman, 2006) . The RIDD target transcript genes are predominantly those that encode membrane-associated or secretory proteins and which are not necessary for ER proteinfolding reactions (Hollien and Weissman, 2006) . The cleavage of Xbp1 and the RIDD-target transcripts constitute homeostatic or pro-survival response by IRE1 since XBP1S trans-activates genes encoding multiple chaperones (to fold unfolded proteins) and the ERAD pathway genes (to degrade terminally misfolded proteins) whereas RIDD reduces flux of polypeptides entering the ER lumen (Lee et al., 2003; Hollien and Weissman, 2006) . On the other hand, cleavage of pre-miRNA transcripts which are processed in the cell to generate CASPASE-2 mRNA (Casp2) controlling miRNAs, constitutes the pro-apoptotic function of IRE1 (Upton et al., 2012) . Another pro-apoptotic signal from IRE1 emanates from signaling through phosphorylation of JNK1 (Urano et al., 2000) . Although in the initial phase RIDD activity does not cleave mRNAs encoding essential ER proteins, at later stages of chronic UPR such transcripts are rendered susceptible to degradation promoting apoptosis induction (Han et al., 2009; Bhattacharyya et al., 2014) . Infection of mammalian cells by a multitude of viruses induce an UPR which is sometimes characterized by suppression of signaling by one or more of the three sensor(s; Su et al., 2002; Tardif et al., 2002; He, 2006; Yu et al., 2006 Yu et al., , 2013 Medigeshi et al., 2007; Zhang et al., 2010; Merquiol et al., 2011) . Among these at least two viruses from diverse families, HCMV (a DNA virus) and hepatitis C virus (a hepacivirus), interfere with IRE1 signaling by different mechanism (Tardif et al., 2004; Stahl et al., 2013 ). An observed inhibition of any cellular function by a virus infection could suggest a potential anti-virus function for it, which the virus has evolved to evade through blocking some critical step(s). In both the cases mentioned above, stability of the viral proteins seems to be affected by ERAD-mediated degradation, although other potential anti-viral effect of IRE1 activation are not clear yet (Isler et al., 2005; Saeed et al., 2011) . Interestingly, host mRNA fragments produced following IRE1 activation during bacterial infection, has been shown to activate RIG-I signaling (Figure 1 ; Cho et al., 2013b) . Theoretically, other functions of IRE1 can also have anti-viral effect necessitating its inhibition for uninhibited viral replication. It is, however, still not clear whether IRE1 is able to cleave any viral RNA (or mRNA) in a manner similar to that of other RIDD targets (Figure 1) . The possibilities of such a direct anti-viral function are encouraged by the fact that all these viruses encode at least one protein which, as part of its maturation process, requires glycosylation and disulfide-bond formation. Such a necessity would entail translation of the mRNA encoding such a protein, which in case of positive-sense single-stranded RNA viruses would mean the genome, in association with the ER-membrane (Figure 1 ; Lerner et al., 2003) . Additionally for many RNA viruses, replication complexes are housed in ER-derived vesicular structures (Denison, 2008; den Boon et al., 2010) . Considering the proximity of IRE1 and these virus-derived RNAs it is tempting to speculate that probably at some point of time in the viral life cycle one or more virus-associated RNA would be susceptible to cleavage by IRE1. However, studies with at least two viruses have shown that instead of increasing viral titre, inhibiting the RNase activity of activated IRE1 has an opposite effect (Hassan et al., 2012; Bhattacharyya et al., 2014) . This implies potential benefits of IRE1 activation through one or more of the following, (a) expression of chaperones or other pro-viral molecules downstream of XBP1Supregulation or JNK-activation, (b) cleavage of potential anti-viral gene mRNA transcripts by RIDD activity. However, the mode of protection for the viral RNA from RIDD activity is still not clear. It is possible that the viral proteins create a subdomain within the ER membrane, which through some mechanism excludes IRE1 from diffusing near the genomic RNA, thereby protecting the replication complexes (Denison, 2008) . It is therefore probably not surprising that single-stranded plus-sense RNA viruses encode a polyprotein, which produces replication complexes in cis, promoting formation of such subdomains (Egger et al., 2000) . The fact that IRE1 forms bulky oligomers of higher order probably aggravates such an exclusion of the activated sensor molecules from vicinity of the viral replication complexes. The UPR signaling eventually attenuate during chronic ER-stress and since that is what a virus-induced UPR mimics, probably the viral RNA needs protection only during the initial phase of UPR activation (Lin et al., 2007) . Since the choice of RIDD target seems to be grossly driven towards mRNAs that encode ER-transitory but are not ER-essential proteins, it is also possible that one or more viral protein have evolved to mimic a host protein the transcript of which is RIDD-resistant (Hollien and Weissman, 2006) . Most of the RIDD target mRNA are observed to be ER-membrane associated, the proximity to IRE1 facilitating association and cleavage (Figure 1 ; Hollien and Weissman, 2006) . Although ER-association for an mRNA is possible without the mediation of ribosomes, Gaddam and co-workers reported that continued association with polysomes for a membrane-bound mRNA can confer protection from IRE1 cleavage (Cui et al., 2012; Gaddam et al., 2013) . This would suggest important implications for the observed refractory nature of Japanese encephalitis virus (JEV) and influenza virus RNA to RIDD cleavage (Hassan et al., 2012; Bhattacharyya et al., 2014) . In contrast to Influenza virus, flaviviruses (which include JEV) do not suppress host protein synthesis implying the absence of a global inhibition on translation as would be expected during UPR (Clyde et al., 2006; Edgil et al., 2006) . Therefore, a continued translation of viral RNA in spite of UPR activation can in principle confer protection from the pattern of RNA cleavage observed in the RIDD pathway. IRE1 and RNaseL, in addition to biochemical similarities in protein kinase domain and structural similarities in their RNase domain, share the functional consequences of their activation in initiating cellular apoptosis through JNK signaling (Table 1 and Figure 2 ; Liu and Lin, 2005; Dhanasekaran and Reddy, 2008) . Though initial discoveries were made in the context of homeostatic and anti-viral role for the former and latter, differences between the pathways are narrowed by further advances in research. In the same vein, while inhibition of IRE1 signaling in virus infected cells indicates a potential anti-viral role, www.frontiersin.org (Tirasophon et al., 1998; Dong and Silverman, 1999; Papa et al., 2003; Lin et al., 2007) Nature of RNase substrates Both 28S rRNA and mRNAs IRE1β can cleave both 28S rRNA and mRNA while IRE1α substrates include only mRNAs (Iwawaki et al., 2001) Dissimilarities Cleavage substrates Beside 28S rRNA, predominantly cleaves mRNAs encoding ribosomal proteins (Andersen et al., 2009) Xbp1u and other mRNAs in addition to microRNA precursors which are targeted as part of the RIDD pathway Selection of cleavage site Cleaved between 2nd and 3rd nucleotide positions of UN/N sites (Han et al., 2014) RNA sequence with the consensus of 5 -CUGCAG-3 in association with a stem-loop (SL) structure essential for recognition of Xbp1u and other mRNAs (Oikawa et al., 2010) association of RNaseL mutations with generation of prostate cancer extends the ambit of influence of this anti-viral effector to more non-infectious physiological disorders (Silverman, 2003) . Biochemically, the similarity in their RNase domains does not extend to the choice of either substrates or cleavage point, which are downstream of UU or UA in RNaseL and downstream of G (predominantly) for IRE1 ( Figure 2C ; Yoshida et al., 2001; Hollien and Weissman, 2006; Upton et al., 2012) . Further, while RNaseL cleaves pre-dominantly in single-stranded region, IRE1 seems to cleave equally well in single-and double-stranded region (Upton et al., 2012) . However, a recent report suggested a consensus cleavage site with the sequence UN/N, in RNaseL targets and in those mRNAs that are cleaved by IRE1 as part of the RIDD pathway (Han et al., 2014) . Access to potential cleavage substrate for RNaseL is conjectured to be facilitated through its association with polyribosomes, while no such association is known for IRE1 (Salehzada et al., 1991) . Possibilities exist that IRE1 would have preferential distribution in the rough ER which, upon activation, would give it ready access to mRNAs for initiating the RIDD pathway. In the context of a virus infection, the pathway leading from both these proteins have the potential to lead to cell death. Notwithstanding the fact that this might be an efficient way of virus clearance, it also portends pathological outcomes for the infected organism. Future research would probably lead to design of drugs targeting these proteins based on the structural homology of their effector domains, regulating the pathological denouement of their activation without compromising their anti-viral or potential anti-viral functions.

What is discussed in this publication?

Can’t RIDD off viruses https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4061530/ SHA: ef58c6e2790539f30df14acc260ae2af4b5f3d1f Authors: Bhattacharyya, Sankar Date: 2014-06-18 DOI: 10.3389/fmicb.2014.00292 License: cc-by Abstract: The mammalian genome has evolved to encode a battery of mechanisms, to mitigate a progression in the life cycle of an invasive viral pathogen. Although apparently disadvantaged by their dependence on the host biosynthetic processes, an immensely faster rate of evolution provides viruses with an edge in this conflict. In this review, I have discussed the potential anti-virus activity of inositol-requiring enzyme 1 (IRE1), a well characterized effector of the cellular homeostatic response to an overloading of the endoplasmic reticulum (ER) protein-folding capacity. IRE1, an ER-membrane-resident ribonuclease (RNase), upon activation catalyses regulated cleavage of select protein-coding and non-coding host RNAs, using an RNase domain which is homologous to that of the known anti-viral effector RNaseL. The latter operates as part of the Oligoadenylate synthetase OAS/RNaseL system of anti-viral defense mechanism. Protein-coding RNA substrates are differentially treated by the IRE1 RNase to either augment, through cytoplasmic splicing of an intron in the Xbp1 transcript, or suppress gene expression. This referred suppression of gene expression is mediated through degradative cleavage of a select cohort of cellular RNA transcripts, initiating the regulated IRE1-dependent decay (RIDD) pathway. The review first discusses the anti-viral mechanism of the OAS/RNaseL system and evasion tactics employed by different viruses. This is followed by a review of the RIDD pathway and its potential effect on the stability of viral RNAs. I conclude with a comparison of the enzymatic activity of the two RNases followed by deliberations on the physiological consequences of their activation. Text: Establishment of infection by a virus, even in permissive host cells, is beset with a plethora of challenges from innate-antiviral and cell-death pathways. Therefore, the host response to a virus infection might prove to be inhibitory for the viral life cycle in a direct or an indirect manner. The direct mechanism involves expression of multiple anti-viral genes that have evolved to recognize, react, and thereby rid the infected host of the viral nucleic acid (Zhou et al., 1997; Thompson et al., 2011) . On the other hand the pathways, e.g., those that culminate in initiating an apoptotic death for the host cell, indirectly serve to limit the spread of virus (Roulston et al., 1999) . A major difference between these two mechanisms is that while the former response is transmissible to neighboring uninfected cells through interferon (IFN) signaling, the latter is observed mostly in cis. Recent reports, however, have demonstrated transmission of an apoptotic signal between cells that are in contact through gap junctions, although such a signaling from an virus infected host cell to an uninfected one is not known yet (Cusato et al., 2003; Udawatte and Ripps, 2005; Kameritsch et al., 2013) . Successful viral pathogens, through a process of active selection, have evolved to replicate and simultaneously evade or block either of these host responses. The viral nucleic acids which could be the genome (positive-sense singlestranded RNA virus) or RNA derived from transcription of the genome [negative-stranded single-sense RNA or double-stranded RNA (dsRNA) or DNA virus], offer critical targets for both detection and eradication. The viral nucleic acid targeting armaments in the host arsenal include those that recognize the associated molecular patterns like toll-like receptors (TLRs), DDX58 (or RIG-1), IFIH1 (or MDA5), IFIT proteins [IFN-stimulated genes (ISG)56 and ISF54], etc. (Aoshi et al., 2011; Bowzard et al., 2011; Jensen and Thomsen, 2012) . This is followed by IFN signaling and expression or activation of factors that target the inducer for degradation or modification like OAS/ribonuclease L (RNaseL) system, APOBEC3, MCPIP1, the ZC3HAV1/exosome system and RNAi pathways (Gao et al., 2002; Sheehy et al., 2002; Guo et al., 2007; Daffis et al., 2010; Sidahmed and Wilkie, 2010; Schmidt et al., 2012; Cho et al., 2013a; Lin et al., 2013) . In this review we focus on two proteins containing homologous RNase domains, RNaseL with a known direct antiviral function and Inositolrequiring enzyme 1 (IRE1 or ERN1) which has an RNaseL-like RNase domain with a known role in homeostatic response to unfolded proteins in the endoplasmic reticulum (ER) and a potential to function as an antiviral (Figure 1 ; Tirasophon et al., 2000) . In mammalian cells the tell-tale signs of RNA virus infection, like the presence of cytosolic RNA having 5 -ppp or extensive (>30 bp) dsRNA segments are detected by dedicated pathogen associated molecular pattern receptors (PAMPs) or pattern recognition receptors (PRRs) in the host cell, like RIG-1, MDA5, and the IFIT family of proteins (Aoshi et al., 2011; Bowzard et al., 2011; Vabret and Blander, 2013) . The transduction of a signal of this recognition results in the expression of IFN genes the products www.frontiersin.org FIGURE 1 | Schematic representation of the ribonuclease activity of IRE1 and RNaseL showing cross-talk between the paths catalysed by the enzymes. The figure shows activation of RNase activity following dimerization triggered by either accumulation of unfolded proteins in the ER-lumen or synthesis of 2-5A by the enzyme OAS, respectively, for IRE1 and RNaseL. The cleavage of Xbp1u by IRE1 releases an intron thus generating Xbp1s. The IRE1 targets in RIDD pathway or all RNaseL substrates are shown to undergo degradative cleavage. The cleavage products generated through degradation of the respective substrate is shown to potentially interact with RIG-I thereby leading to Interferon secretion and trans-activation of Oas genes through Interferon signaling. Abbreviations: RIG-I = retinoic acid inducible gene-I, Ifnb = interferon beta gene loci, IFN = interferons, ISG = interferon-sensitive genes, 2-5A = 2 -5 oligoadenylates. of which upon secretion outside the cell bind to cognate receptors, initiating further downstream signaling (Figure 1 ; Randall and Goodbourn, 2008) . The genes that are regulated as a result of IFN signaling are termed as IFN-stimulated or IFN-regulated genes (ISGs or IRGs; Sen and Sarkar, 2007; Schoggins and Rice, 2011) . Oligoadenylate synthetase or OAS genes are canonical ISGs that convert ATP into 2 -5 linked oligoadenylates (2-5A) by an unique enzymatic mechanism (Figure 1 ; Hartmann et al., 2003) . Further, they are RNA-binding proteins that function like PRRs, in a way that the 2-5A synthesizing activity needs to be induced through an interaction with dsRNA (Minks et al., 1979; Hartmann et al., 2003) . In a host cell infected by an RNA virus, such dsRNA is present in the form of replication-intermediates (RI), which are synthesized by the virus-encoded RNA-dependent RNA polymerases (RdRp) and subsequently used by the same enzyme to synthesize more genomic RNA, through asymmetric transcription (Weber et al., 2006) . However, the replications complexes (RCs) harboring these RI molecules are found secluded inside host-membrane derived vesicles, at least in positive-strand RNA viruses, a group which contains many human pathogens (Uchil and Satchidanandam, 2003; Denison, 2008) . Reports from different groups suggest OAS proteins to be distributed both in the cytoplasm as well as in membrane-associated fractions, perhaps indicating an evolution of the host anti-viral methodologies towards detection of the membrane-associated viral dsRNAs (Marie et al., 1990; Lin et al., 2009) . DNA viruses on the other hand, produce dsRNA by annealing of RNA derived from transcription of both strands in the same viral genomic loci, which are probably detected by the cytoplasmic pool of OAS proteins (Jacobs and Langland, 1996; Weber et al., 2006) . Post-activation the OAS enzymes synthesize 2-5A molecules in a non-processive reaction producing oligomers which, although potentially ranging in size from dimeric to multimeric, are functionally active only in a trimeric or tetrameric form (Dong et al., 1994; Sarkar et al., 1999; Silverman, 2007) . These small ligands, which bear phosphate groups (1-3) at the 5 end and hydroxyl groups at the 2 and 3 positions, serve as co-factor which can specifically interact with and thereby allosterically activate, existing RNaseL molecules (Knight et al., 1980; Zhou et al., 1997 Zhou et al., , 2005 Sarkar et al., 1999) . As part of a physiological control system these 2-5A oligomers are quite unstable in that they are highly susceptible to degradation by cellular 5 -phosphatases and PDE12 (2 -phosphodiesterase; Silverman et al., 1981; Johnston and Hearl, 1987; Kubota et al., 2004; Schmidt et al., 2012) . Viral strategies to evade or overcome this host defense mechanism ranges from preventing IFN signaling which would hinder the induction of OAS expression or thwarting activation of expressed OAS proteins by either shielding the viral dsRNA from interacting with it or modulating the host pathway to synthesize inactive 2-5A derivatives (Cayley et al., 1984; Hersh et al., 1984; Rice et al., 1985; Maitra et al., 1994; Beattie et al., 1995; Rivas et al., 1998; Child et al., 2004; Min and Krug, 2006; Sanchez and Mohr, 2007; Sorgeloos et al., 2013) . Shielding of viral RNA from interacting with OAS is possible through enclosure of dsRNA replication intermediates in membrane enclosed compartments as observed in many flaviviruses (Ahlquist, 2006; Miller and Krijnse-Locker, 2008; Miorin et al., 2013) . RNaseL is a 741 amino acid protein containing three predominantly structured region, an N-terminal ankyrin repeat domain (ARD), a middle catalytically inactive pseudo-kinase (PK) and a C-terminal RNase domain (Figure 2A ; Hassel et al., 1993; Zhou et al., 1993) . The activity of the RNase domain is negatively regulated by the ARD, which is relieved upon binding of 2-5A molecules to ankyrin repeats 2 and 4 followed by a conformational alteration (Figure 1 ; Hassel et al., 1993; Tanaka et al., 2004; Nakanishi et al., 2005) . In support of this contention, deletion of the ARD has been demonstrated to produce constitutively active RNaseL, although with dramatically lower RNase activity (Dong and Silverman, 1997) . However, recent reports suggest that while 2-5A links the ankyrin repeats from adjacent molecules leading to formation of dimer and higher order structures, at sufficiently high in vitro concentrations, RNaseL could oligomerize even in the absence of 2-5A . Nonetheless, in vivo the RNaseL nuclease activity still seems to be under the sole regulation of 2-5A (Al-Saif and Khabar, 2012) . In order to exploit this dependence, multiple viruses like mouse hepatitis virus (MHV) and rotavirus group A (RVA) have evolved to encode phosphodiesterases capable of hydrolysing the 2 -5 linkages in 2-5A and thereby attenuate the RNaseL cleavage activity (Zhao et al., 2012; Zhang et al., 2013) . In addition to 5 -phosphatases and 2 -phosphodiesterases to reduce ClustalW alignment of primary sequence from a segment of the PK domain indicating amino acid residues which are important for interacting with nucleotide cofactors. The conserved lysine residues, critical for this interaction (K599 for IRE1 and K392 in RNaseL) are underlined. (C) Alignment of the KEN domains in RNaseL and IRE1. The amino acids highlighted and numbered in IRE1 are critical for the IRE1 RNase activity (Tirasophon et al., 2000) . the endogenous 2-5A levels, mammalian genomes encode posttranscriptional and post-translation inhibitors of RNaseL activity in the form of microRNA-29 and the protein ABCE1 (RNaseL inhibitor or RLI), respectively (Bisbal et al., 1995; Lee et al., 2013) . Direct inhibition of RNaseL function is also observed upon infection by Picornaviruses through, either inducing the expression of ABCE1 or exercising a unique inhibitory property of a segment of the viral RNA (Martinand et al., 1998 (Martinand et al., , 1999 Townsend et al., 2008; Sorgeloos et al., 2013) . Once activated by 2-5A, RNaseL can degrade single-stranded RNA irrespective of its origin (virus or host) although there seems to exist a bias towards cleavage of viral RNA (Wreschner et al., 1981a; Silverman et al., 1983; Li et al., 1998) . RNA sequences that are predominantly cleaved by RNaseL are U-rich with the cleavage points being typically at the 3 end of UA or UG or UU di-nucleotides, leaving a 5 -OH and a 3 -monophosphate in the cleavage product (Floyd-Smith et al., 1981; Wreschner et al., 1981b) . A recent report shows a more general consensus of 5 -UNN-3 with the cleavage point between the second and the third nucleotide (Han et al., 2014) . Cellular targets of RNaseL include both ribosomal RNA (rRNA) and mRNAs, the latter predominantly representing genes involved in protein biosynthesis (Wreschner et al., 1981a; Al-Ahmadi et al., 2009; Andersen et al., 2009) . Additionally, RNaseL activity can also degrade specific ISG mRNA transcripts and thereby attenuate the effect of IFN signaling (Li et al., 2000) . Probably an evolution towards insulating gene expression from RNaseL activity is observed in the coding region of mammalian genes where the UU/UA dinucleotide frequency is rarer (Bisbal et al., 2000; Khabar et al., 2003; Al-Saif and Khabar, 2012) . Perhaps not surprisingly, with a much faster rate of evolution, similar observations have been made with respect to evasion of RNaseL mediated degradation by viral RNAs too (Han and Barton, 2002; Washenberger et al., 2007) . Moreover, nucleoside modifications in host mRNAs, rarely observed in viral RNAs, have also been shown to confer protection from RNaseL (Anderson et al., 2011) . In addition to directly targeting viral RNA, the reduction in functional ribosomes and ribosomal protein mRNA affects viral protein synthesis and replication in an indirect manner. Probably, as a reflection of these effects on cellular RNAs, RNaseL is implicated as one of the factors determining the anti-proliferative effect of IFN activity . The anti-viral activity of RNaseL extends beyond direct cleavage of viral RNA, through stimulation of RIG-I by the cleavage product (Malathi et al., , 2007 (Malathi et al., , 2010 . A global effect of RNaseL is observed in the form of autophagy induced through c-jun N-terminal kinase (JNK) signaling and apoptosis, probably as a consequence of rRNA cleavage (Li et al., 2004; Chakrabarti et al., 2012; Siddiqui and Malathi, 2012) . RNaseL has also been demonstrated to play a role in apoptotic cell death initiated by pharmacological agents extending the physiological role of this pathway beyond the boundary of being only an anti-viral mechanism (Castelli et al., 1997 (Castelli et al., , 1998 . The ER serves as a conduit for maturation of cellular proteins which are either secreted or destined to be associated with a membrane for its function. An exclusive microenvironment (high Calcium ion and unique ratio of reduced to oxidized glutathione) along with a battery of ER-lumen resident enzymes (foldases, chaperones, and lectins) catalyse/mediate the necessary folding, disulfide-bond formation, and glycosylation reactions (Schroder and Kaufman, 2005) . A perturbation of the folding capacity, due to either physiological disturbances or virus infection, can lead to an accumulation of unfolded proteins in the ER lumen, which signals an unfolded protein response (UPR). UPR encompasses a networked transcriptional and translational gene-expression program, initiated by three ER-membrane resident sensors namely IRE1 or ERN1, PKR-like ER Kinase (PERK or EIF2AK3) and activating transcription factor 6 (ATF6; Hetz, 2012) . IRE1 is a type I single-pass trans-membrane protein in which, similar to what is observed with RNaseL, the N-terminal resident in the ER lumen serves as sensor and the cytosolic C-terminal as the effector (Figure 1 ; Chen and Brandizzi, 2013) . The IRE1 coding gene is present in genomes ranging from yeast to mammals and in the latter is ubiquitously expressed in all tissues (Tirasophon et al., 1998) . Signal transduction by stimulated IRE1 initiates multiple gene regulatory pathways with either pro-survival or pro-apoptotic consequences (Kaufman, 1999) . During homeostasis or unstressed conditions the sensor molecules are monomeric, a state maintained co-operatively by the " absence" of unfolded proteins and the "presence" of HSPA5 (GRP78 or Bip, an ERresident chaperone) molecules bound to a membrane-proximal disordered segment of the protein in the ER-lumen-resident Nterminus (Credle et al., 2005) . Accumulated unfolded proteins in the lumen triggers coupling of this domain from adjacent sensor molecules through a combination of (a) titration of the bound HSPA5 chaperone molecules and (b) direct tethering by malfolded protein molecules (Shamu and Walter, 1996; Credle et al., 2005; Aragon et al., 2009; Korennykh et al., 2009) . Abutting of the luminal domains juxtapose the cytosolic C-terminal segments, leading to an aggregation of the IRE1 molecules into distinct ER-membrane foci (Kimata et al., 2007; Li et al., 2010) . The C-terminal segment has a serine/threonine kinase domain and a RNase domain homologous to that of RNaseL (Figure 1 ; Tirasophon et al., 1998 Tirasophon et al., , 2000 . A trans-autophosphorylation by the kinase domain allosterically activates the RNase domain (Tirasophon et al., 2000; Lee et al., 2008; Korennykh et al., 2009) . In fact, exogenous over-expression of IRE1 in mammalian cells lead to activation suggesting that, under homeostatic conditions, the non-juxtaposition of cytosolic domains maintains an inactive IRE1 (Tirasophon et al., 1998) . Once activated, IRE1 performs cleavage of a variety of RNA substrates mediated by its RNase domain, in addition to phosphorylating and thereby activating JNK (Cox and Walter, 1996; Urano et al., 2000) . Depending on the RNA substrate, the cleavage catalyzed by IRE1 RNase produces differential consequence. Although scission of the Xbp1 mRNA transcript at two internal positions is followed by splicing of the internal segment through ligation of the terminal cleavage products, that in all other known IRE1 target RNA is followed by degradation (Figure 1 ; Sidrauski and Walter, 1997; Calfon et al., 2002) . The latter mode of negative regulation of gene expression is termed as the regulated IRE1-dependent decay (RIDD) pathway (Hollien and Weissman, 2006; Oikawa et al., 2007; Iqbal et al., 2008; Lipson et al., 2008) . Gene transcripts regulated by RIDD pathway includes that from IRE1 (i.e., selftranscripts), probably in a negative feedback loop mechanism (Tirasophon et al., 2000) . In addition to protein coding RNA, RIDD pathway down-regulates the level of a host of microRNA precursors (pre-miRNAs) and can potentially cleave in the anticodon loop of tRNA Phe (Korennykh et al., 2011; Upton et al., 2012) . The IRE1 RNase domain cleaves the Xbp1u (u for unspliced) mRNA transcript at two precise internal positions within the open reading frame (ORF) generating three segments, the terminal two of which are ligated by a tRNA ligase in yeast and by an unknown ligase in mammalian cells, to produce the Xbp1s (s for spliced) mRNA transcript (Figure 1 ; Yoshida et al., 2001) . The Xbp1s thus generated has a longer ORF, which is created by a frame-shift in the coding sequence downstream of the splice site (Cox and Walter, 1996; Calfon et al., 2002) . A similar dual endonucleolytic cleavage is also observed to initiate the XRN1 and Ski2-3-8 dependent degradation of transcripts in the RIDD degradation pathway (Hollien and Weissman, 2006) . The RIDD target transcript genes are predominantly those that encode membrane-associated or secretory proteins and which are not necessary for ER proteinfolding reactions (Hollien and Weissman, 2006) . The cleavage of Xbp1 and the RIDD-target transcripts constitute homeostatic or pro-survival response by IRE1 since XBP1S trans-activates genes encoding multiple chaperones (to fold unfolded proteins) and the ERAD pathway genes (to degrade terminally misfolded proteins) whereas RIDD reduces flux of polypeptides entering the ER lumen (Lee et al., 2003; Hollien and Weissman, 2006) . On the other hand, cleavage of pre-miRNA transcripts which are processed in the cell to generate CASPASE-2 mRNA (Casp2) controlling miRNAs, constitutes the pro-apoptotic function of IRE1 (Upton et al., 2012) . Another pro-apoptotic signal from IRE1 emanates from signaling through phosphorylation of JNK1 (Urano et al., 2000) . Although in the initial phase RIDD activity does not cleave mRNAs encoding essential ER proteins, at later stages of chronic UPR such transcripts are rendered susceptible to degradation promoting apoptosis induction (Han et al., 2009; Bhattacharyya et al., 2014) . Infection of mammalian cells by a multitude of viruses induce an UPR which is sometimes characterized by suppression of signaling by one or more of the three sensor(s; Su et al., 2002; Tardif et al., 2002; He, 2006; Yu et al., 2006 Yu et al., , 2013 Medigeshi et al., 2007; Zhang et al., 2010; Merquiol et al., 2011) . Among these at least two viruses from diverse families, HCMV (a DNA virus) and hepatitis C virus (a hepacivirus), interfere with IRE1 signaling by different mechanism (Tardif et al., 2004; Stahl et al., 2013 ). An observed inhibition of any cellular function by a virus infection could suggest a potential anti-virus function for it, which the virus has evolved to evade through blocking some critical step(s). In both the cases mentioned above, stability of the viral proteins seems to be affected by ERAD-mediated degradation, although other potential anti-viral effect of IRE1 activation are not clear yet (Isler et al., 2005; Saeed et al., 2011) . Interestingly, host mRNA fragments produced following IRE1 activation during bacterial infection, has been shown to activate RIG-I signaling (Figure 1 ; Cho et al., 2013b) . Theoretically, other functions of IRE1 can also have anti-viral effect necessitating its inhibition for uninhibited viral replication. It is, however, still not clear whether IRE1 is able to cleave any viral RNA (or mRNA) in a manner similar to that of other RIDD targets (Figure 1) . The possibilities of such a direct anti-viral function are encouraged by the fact that all these viruses encode at least one protein which, as part of its maturation process, requires glycosylation and disulfide-bond formation. Such a necessity would entail translation of the mRNA encoding such a protein, which in case of positive-sense single-stranded RNA viruses would mean the genome, in association with the ER-membrane (Figure 1 ; Lerner et al., 2003) . Additionally for many RNA viruses, replication complexes are housed in ER-derived vesicular structures (Denison, 2008; den Boon et al., 2010) . Considering the proximity of IRE1 and these virus-derived RNAs it is tempting to speculate that probably at some point of time in the viral life cycle one or more virus-associated RNA would be susceptible to cleavage by IRE1. However, studies with at least two viruses have shown that instead of increasing viral titre, inhibiting the RNase activity of activated IRE1 has an opposite effect (Hassan et al., 2012; Bhattacharyya et al., 2014) . This implies potential benefits of IRE1 activation through one or more of the following, (a) expression of chaperones or other pro-viral molecules downstream of XBP1Supregulation or JNK-activation, (b) cleavage of potential anti-viral gene mRNA transcripts by RIDD activity. However, the mode of protection for the viral RNA from RIDD activity is still not clear. It is possible that the viral proteins create a subdomain within the ER membrane, which through some mechanism excludes IRE1 from diffusing near the genomic RNA, thereby protecting the replication complexes (Denison, 2008) . It is therefore probably not surprising that single-stranded plus-sense RNA viruses encode a polyprotein, which produces replication complexes in cis, promoting formation of such subdomains (Egger et al., 2000) . The fact that IRE1 forms bulky oligomers of higher order probably aggravates such an exclusion of the activated sensor molecules from vicinity of the viral replication complexes. The UPR signaling eventually attenuate during chronic ER-stress and since that is what a virus-induced UPR mimics, probably the viral RNA needs protection only during the initial phase of UPR activation (Lin et al., 2007) . Since the choice of RIDD target seems to be grossly driven towards mRNAs that encode ER-transitory but are not ER-essential proteins, it is also possible that one or more viral protein have evolved to mimic a host protein the transcript of which is RIDD-resistant (Hollien and Weissman, 2006) . Most of the RIDD target mRNA are observed to be ER-membrane associated, the proximity to IRE1 facilitating association and cleavage (Figure 1 ; Hollien and Weissman, 2006) . Although ER-association for an mRNA is possible without the mediation of ribosomes, Gaddam and co-workers reported that continued association with polysomes for a membrane-bound mRNA can confer protection from IRE1 cleavage (Cui et al., 2012; Gaddam et al., 2013) . This would suggest important implications for the observed refractory nature of Japanese encephalitis virus (JEV) and influenza virus RNA to RIDD cleavage (Hassan et al., 2012; Bhattacharyya et al., 2014) . In contrast to Influenza virus, flaviviruses (which include JEV) do not suppress host protein synthesis implying the absence of a global inhibition on translation as would be expected during UPR (Clyde et al., 2006; Edgil et al., 2006) . Therefore, a continued translation of viral RNA in spite of UPR activation can in principle confer protection from the pattern of RNA cleavage observed in the RIDD pathway. IRE1 and RNaseL, in addition to biochemical similarities in protein kinase domain and structural similarities in their RNase domain, share the functional consequences of their activation in initiating cellular apoptosis through JNK signaling (Table 1 and Figure 2 ; Liu and Lin, 2005; Dhanasekaran and Reddy, 2008) . Though initial discoveries were made in the context of homeostatic and anti-viral role for the former and latter, differences between the pathways are narrowed by further advances in research. In the same vein, while inhibition of IRE1 signaling in virus infected cells indicates a potential anti-viral role, www.frontiersin.org (Tirasophon et al., 1998; Dong and Silverman, 1999; Papa et al., 2003; Lin et al., 2007) Nature of RNase substrates Both 28S rRNA and mRNAs IRE1β can cleave both 28S rRNA and mRNA while IRE1α substrates include only mRNAs (Iwawaki et al., 2001) Dissimilarities Cleavage substrates Beside 28S rRNA, predominantly cleaves mRNAs encoding ribosomal proteins (Andersen et al., 2009) Xbp1u and other mRNAs in addition to microRNA precursors which are targeted as part of the RIDD pathway Selection of cleavage site Cleaved between 2nd and 3rd nucleotide positions of UN/N sites (Han et al., 2014) RNA sequence with the consensus of 5 -CUGCAG-3 in association with a stem-loop (SL) structure essential for recognition of Xbp1u and other mRNAs (Oikawa et al., 2010) association of RNaseL mutations with generation of prostate cancer extends the ambit of influence of this anti-viral effector to more non-infectious physiological disorders (Silverman, 2003) . Biochemically, the similarity in their RNase domains does not extend to the choice of either substrates or cleavage point, which are downstream of UU or UA in RNaseL and downstream of G (predominantly) for IRE1 ( Figure 2C ; Yoshida et al., 2001; Hollien and Weissman, 2006; Upton et al., 2012) . Further, while RNaseL cleaves pre-dominantly in single-stranded region, IRE1 seems to cleave equally well in single-and double-stranded region (Upton et al., 2012) . However, a recent report suggested a consensus cleavage site with the sequence UN/N, in RNaseL targets and in those mRNAs that are cleaved by IRE1 as part of the RIDD pathway (Han et al., 2014) . Access to potential cleavage substrate for RNaseL is conjectured to be facilitated through its association with polyribosomes, while no such association is known for IRE1 (Salehzada et al., 1991) . Possibilities exist that IRE1 would have preferential distribution in the rough ER which, upon activation, would give it ready access to mRNAs for initiating the RIDD pathway. In the context of a virus infection, the pathway leading from both these proteins have the potential to lead to cell death. Notwithstanding the fact that this might be an efficient way of virus clearance, it also portends pathological outcomes for the infected organism. Future research would probably lead to design of drugs targeting these proteins based on the structural homology of their effector domains, regulating the pathological denouement of their activation without compromising their anti-viral or potential anti-viral functions.

What is discussed in this publication?

Can’t RIDD off viruses https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4061530/ SHA: ef58c6e2790539f30df14acc260ae2af4b5f3d1f Authors: Bhattacharyya, Sankar Date: 2014-06-18 DOI: 10.3389/fmicb.2014.00292 License: cc-by Abstract: The mammalian genome has evolved to encode a battery of mechanisms, to mitigate a progression in the life cycle of an invasive viral pathogen. Although apparently disadvantaged by their dependence on the host biosynthetic processes, an immensely faster rate of evolution provides viruses with an edge in this conflict. In this review, I have discussed the potential anti-virus activity of inositol-requiring enzyme 1 (IRE1), a well characterized effector of the cellular homeostatic response to an overloading of the endoplasmic reticulum (ER) protein-folding capacity. IRE1, an ER-membrane-resident ribonuclease (RNase), upon activation catalyses regulated cleavage of select protein-coding and non-coding host RNAs, using an RNase domain which is homologous to that of the known anti-viral effector RNaseL. The latter operates as part of the Oligoadenylate synthetase OAS/RNaseL system of anti-viral defense mechanism. Protein-coding RNA substrates are differentially treated by the IRE1 RNase to either augment, through cytoplasmic splicing of an intron in the Xbp1 transcript, or suppress gene expression. This referred suppression of gene expression is mediated through degradative cleavage of a select cohort of cellular RNA transcripts, initiating the regulated IRE1-dependent decay (RIDD) pathway. The review first discusses the anti-viral mechanism of the OAS/RNaseL system and evasion tactics employed by different viruses. This is followed by a review of the RIDD pathway and its potential effect on the stability of viral RNAs. I conclude with a comparison of the enzymatic activity of the two RNases followed by deliberations on the physiological consequences of their activation. Text: Establishment of infection by a virus, even in permissive host cells, is beset with a plethora of challenges from innate-antiviral and cell-death pathways. Therefore, the host response to a virus infection might prove to be inhibitory for the viral life cycle in a direct or an indirect manner. The direct mechanism involves expression of multiple anti-viral genes that have evolved to recognize, react, and thereby rid the infected host of the viral nucleic acid (Zhou et al., 1997; Thompson et al., 2011) . On the other hand the pathways, e.g., those that culminate in initiating an apoptotic death for the host cell, indirectly serve to limit the spread of virus (Roulston et al., 1999) . A major difference between these two mechanisms is that while the former response is transmissible to neighboring uninfected cells through interferon (IFN) signaling, the latter is observed mostly in cis. Recent reports, however, have demonstrated transmission of an apoptotic signal between cells that are in contact through gap junctions, although such a signaling from an virus infected host cell to an uninfected one is not known yet (Cusato et al., 2003; Udawatte and Ripps, 2005; Kameritsch et al., 2013) . Successful viral pathogens, through a process of active selection, have evolved to replicate and simultaneously evade or block either of these host responses. The viral nucleic acids which could be the genome (positive-sense singlestranded RNA virus) or RNA derived from transcription of the genome [negative-stranded single-sense RNA or double-stranded RNA (dsRNA) or DNA virus], offer critical targets for both detection and eradication. The viral nucleic acid targeting armaments in the host arsenal include those that recognize the associated molecular patterns like toll-like receptors (TLRs), DDX58 (or RIG-1), IFIH1 (or MDA5), IFIT proteins [IFN-stimulated genes (ISG)56 and ISF54], etc. (Aoshi et al., 2011; Bowzard et al., 2011; Jensen and Thomsen, 2012) . This is followed by IFN signaling and expression or activation of factors that target the inducer for degradation or modification like OAS/ribonuclease L (RNaseL) system, APOBEC3, MCPIP1, the ZC3HAV1/exosome system and RNAi pathways (Gao et al., 2002; Sheehy et al., 2002; Guo et al., 2007; Daffis et al., 2010; Sidahmed and Wilkie, 2010; Schmidt et al., 2012; Cho et al., 2013a; Lin et al., 2013) . In this review we focus on two proteins containing homologous RNase domains, RNaseL with a known direct antiviral function and Inositolrequiring enzyme 1 (IRE1 or ERN1) which has an RNaseL-like RNase domain with a known role in homeostatic response to unfolded proteins in the endoplasmic reticulum (ER) and a potential to function as an antiviral (Figure 1 ; Tirasophon et al., 2000) . In mammalian cells the tell-tale signs of RNA virus infection, like the presence of cytosolic RNA having 5 -ppp or extensive (>30 bp) dsRNA segments are detected by dedicated pathogen associated molecular pattern receptors (PAMPs) or pattern recognition receptors (PRRs) in the host cell, like RIG-1, MDA5, and the IFIT family of proteins (Aoshi et al., 2011; Bowzard et al., 2011; Vabret and Blander, 2013) . The transduction of a signal of this recognition results in the expression of IFN genes the products www.frontiersin.org FIGURE 1 | Schematic representation of the ribonuclease activity of IRE1 and RNaseL showing cross-talk between the paths catalysed by the enzymes. The figure shows activation of RNase activity following dimerization triggered by either accumulation of unfolded proteins in the ER-lumen or synthesis of 2-5A by the enzyme OAS, respectively, for IRE1 and RNaseL. The cleavage of Xbp1u by IRE1 releases an intron thus generating Xbp1s. The IRE1 targets in RIDD pathway or all RNaseL substrates are shown to undergo degradative cleavage. The cleavage products generated through degradation of the respective substrate is shown to potentially interact with RIG-I thereby leading to Interferon secretion and trans-activation of Oas genes through Interferon signaling. Abbreviations: RIG-I = retinoic acid inducible gene-I, Ifnb = interferon beta gene loci, IFN = interferons, ISG = interferon-sensitive genes, 2-5A = 2 -5 oligoadenylates. of which upon secretion outside the cell bind to cognate receptors, initiating further downstream signaling (Figure 1 ; Randall and Goodbourn, 2008) . The genes that are regulated as a result of IFN signaling are termed as IFN-stimulated or IFN-regulated genes (ISGs or IRGs; Sen and Sarkar, 2007; Schoggins and Rice, 2011) . Oligoadenylate synthetase or OAS genes are canonical ISGs that convert ATP into 2 -5 linked oligoadenylates (2-5A) by an unique enzymatic mechanism (Figure 1 ; Hartmann et al., 2003) . Further, they are RNA-binding proteins that function like PRRs, in a way that the 2-5A synthesizing activity needs to be induced through an interaction with dsRNA (Minks et al., 1979; Hartmann et al., 2003) . In a host cell infected by an RNA virus, such dsRNA is present in the form of replication-intermediates (RI), which are synthesized by the virus-encoded RNA-dependent RNA polymerases (RdRp) and subsequently used by the same enzyme to synthesize more genomic RNA, through asymmetric transcription (Weber et al., 2006) . However, the replications complexes (RCs) harboring these RI molecules are found secluded inside host-membrane derived vesicles, at least in positive-strand RNA viruses, a group which contains many human pathogens (Uchil and Satchidanandam, 2003; Denison, 2008) . Reports from different groups suggest OAS proteins to be distributed both in the cytoplasm as well as in membrane-associated fractions, perhaps indicating an evolution of the host anti-viral methodologies towards detection of the membrane-associated viral dsRNAs (Marie et al., 1990; Lin et al., 2009) . DNA viruses on the other hand, produce dsRNA by annealing of RNA derived from transcription of both strands in the same viral genomic loci, which are probably detected by the cytoplasmic pool of OAS proteins (Jacobs and Langland, 1996; Weber et al., 2006) . Post-activation the OAS enzymes synthesize 2-5A molecules in a non-processive reaction producing oligomers which, although potentially ranging in size from dimeric to multimeric, are functionally active only in a trimeric or tetrameric form (Dong et al., 1994; Sarkar et al., 1999; Silverman, 2007) . These small ligands, which bear phosphate groups (1-3) at the 5 end and hydroxyl groups at the 2 and 3 positions, serve as co-factor which can specifically interact with and thereby allosterically activate, existing RNaseL molecules (Knight et al., 1980; Zhou et al., 1997 Zhou et al., , 2005 Sarkar et al., 1999) . As part of a physiological control system these 2-5A oligomers are quite unstable in that they are highly susceptible to degradation by cellular 5 -phosphatases and PDE12 (2 -phosphodiesterase; Silverman et al., 1981; Johnston and Hearl, 1987; Kubota et al., 2004; Schmidt et al., 2012) . Viral strategies to evade or overcome this host defense mechanism ranges from preventing IFN signaling which would hinder the induction of OAS expression or thwarting activation of expressed OAS proteins by either shielding the viral dsRNA from interacting with it or modulating the host pathway to synthesize inactive 2-5A derivatives (Cayley et al., 1984; Hersh et al., 1984; Rice et al., 1985; Maitra et al., 1994; Beattie et al., 1995; Rivas et al., 1998; Child et al., 2004; Min and Krug, 2006; Sanchez and Mohr, 2007; Sorgeloos et al., 2013) . Shielding of viral RNA from interacting with OAS is possible through enclosure of dsRNA replication intermediates in membrane enclosed compartments as observed in many flaviviruses (Ahlquist, 2006; Miller and Krijnse-Locker, 2008; Miorin et al., 2013) . RNaseL is a 741 amino acid protein containing three predominantly structured region, an N-terminal ankyrin repeat domain (ARD), a middle catalytically inactive pseudo-kinase (PK) and a C-terminal RNase domain (Figure 2A ; Hassel et al., 1993; Zhou et al., 1993) . The activity of the RNase domain is negatively regulated by the ARD, which is relieved upon binding of 2-5A molecules to ankyrin repeats 2 and 4 followed by a conformational alteration (Figure 1 ; Hassel et al., 1993; Tanaka et al., 2004; Nakanishi et al., 2005) . In support of this contention, deletion of the ARD has been demonstrated to produce constitutively active RNaseL, although with dramatically lower RNase activity (Dong and Silverman, 1997) . However, recent reports suggest that while 2-5A links the ankyrin repeats from adjacent molecules leading to formation of dimer and higher order structures, at sufficiently high in vitro concentrations, RNaseL could oligomerize even in the absence of 2-5A . Nonetheless, in vivo the RNaseL nuclease activity still seems to be under the sole regulation of 2-5A (Al-Saif and Khabar, 2012) . In order to exploit this dependence, multiple viruses like mouse hepatitis virus (MHV) and rotavirus group A (RVA) have evolved to encode phosphodiesterases capable of hydrolysing the 2 -5 linkages in 2-5A and thereby attenuate the RNaseL cleavage activity (Zhao et al., 2012; Zhang et al., 2013) . In addition to 5 -phosphatases and 2 -phosphodiesterases to reduce ClustalW alignment of primary sequence from a segment of the PK domain indicating amino acid residues which are important for interacting with nucleotide cofactors. The conserved lysine residues, critical for this interaction (K599 for IRE1 and K392 in RNaseL) are underlined. (C) Alignment of the KEN domains in RNaseL and IRE1. The amino acids highlighted and numbered in IRE1 are critical for the IRE1 RNase activity (Tirasophon et al., 2000) . the endogenous 2-5A levels, mammalian genomes encode posttranscriptional and post-translation inhibitors of RNaseL activity in the form of microRNA-29 and the protein ABCE1 (RNaseL inhibitor or RLI), respectively (Bisbal et al., 1995; Lee et al., 2013) . Direct inhibition of RNaseL function is also observed upon infection by Picornaviruses through, either inducing the expression of ABCE1 or exercising a unique inhibitory property of a segment of the viral RNA (Martinand et al., 1998 (Martinand et al., , 1999 Townsend et al., 2008; Sorgeloos et al., 2013) . Once activated by 2-5A, RNaseL can degrade single-stranded RNA irrespective of its origin (virus or host) although there seems to exist a bias towards cleavage of viral RNA (Wreschner et al., 1981a; Silverman et al., 1983; Li et al., 1998) . RNA sequences that are predominantly cleaved by RNaseL are U-rich with the cleavage points being typically at the 3 end of UA or UG or UU di-nucleotides, leaving a 5 -OH and a 3 -monophosphate in the cleavage product (Floyd-Smith et al., 1981; Wreschner et al., 1981b) . A recent report shows a more general consensus of 5 -UNN-3 with the cleavage point between the second and the third nucleotide (Han et al., 2014) . Cellular targets of RNaseL include both ribosomal RNA (rRNA) and mRNAs, the latter predominantly representing genes involved in protein biosynthesis (Wreschner et al., 1981a; Al-Ahmadi et al., 2009; Andersen et al., 2009) . Additionally, RNaseL activity can also degrade specific ISG mRNA transcripts and thereby attenuate the effect of IFN signaling (Li et al., 2000) . Probably an evolution towards insulating gene expression from RNaseL activity is observed in the coding region of mammalian genes where the UU/UA dinucleotide frequency is rarer (Bisbal et al., 2000; Khabar et al., 2003; Al-Saif and Khabar, 2012) . Perhaps not surprisingly, with a much faster rate of evolution, similar observations have been made with respect to evasion of RNaseL mediated degradation by viral RNAs too (Han and Barton, 2002; Washenberger et al., 2007) . Moreover, nucleoside modifications in host mRNAs, rarely observed in viral RNAs, have also been shown to confer protection from RNaseL (Anderson et al., 2011) . In addition to directly targeting viral RNA, the reduction in functional ribosomes and ribosomal protein mRNA affects viral protein synthesis and replication in an indirect manner. Probably, as a reflection of these effects on cellular RNAs, RNaseL is implicated as one of the factors determining the anti-proliferative effect of IFN activity . The anti-viral activity of RNaseL extends beyond direct cleavage of viral RNA, through stimulation of RIG-I by the cleavage product (Malathi et al., , 2007 (Malathi et al., , 2010 . A global effect of RNaseL is observed in the form of autophagy induced through c-jun N-terminal kinase (JNK) signaling and apoptosis, probably as a consequence of rRNA cleavage (Li et al., 2004; Chakrabarti et al., 2012; Siddiqui and Malathi, 2012) . RNaseL has also been demonstrated to play a role in apoptotic cell death initiated by pharmacological agents extending the physiological role of this pathway beyond the boundary of being only an anti-viral mechanism (Castelli et al., 1997 (Castelli et al., , 1998 . The ER serves as a conduit for maturation of cellular proteins which are either secreted or destined to be associated with a membrane for its function. An exclusive microenvironment (high Calcium ion and unique ratio of reduced to oxidized glutathione) along with a battery of ER-lumen resident enzymes (foldases, chaperones, and lectins) catalyse/mediate the necessary folding, disulfide-bond formation, and glycosylation reactions (Schroder and Kaufman, 2005) . A perturbation of the folding capacity, due to either physiological disturbances or virus infection, can lead to an accumulation of unfolded proteins in the ER lumen, which signals an unfolded protein response (UPR). UPR encompasses a networked transcriptional and translational gene-expression program, initiated by three ER-membrane resident sensors namely IRE1 or ERN1, PKR-like ER Kinase (PERK or EIF2AK3) and activating transcription factor 6 (ATF6; Hetz, 2012) . IRE1 is a type I single-pass trans-membrane protein in which, similar to what is observed with RNaseL, the N-terminal resident in the ER lumen serves as sensor and the cytosolic C-terminal as the effector (Figure 1 ; Chen and Brandizzi, 2013) . The IRE1 coding gene is present in genomes ranging from yeast to mammals and in the latter is ubiquitously expressed in all tissues (Tirasophon et al., 1998) . Signal transduction by stimulated IRE1 initiates multiple gene regulatory pathways with either pro-survival or pro-apoptotic consequences (Kaufman, 1999) . During homeostasis or unstressed conditions the sensor molecules are monomeric, a state maintained co-operatively by the " absence" of unfolded proteins and the "presence" of HSPA5 (GRP78 or Bip, an ERresident chaperone) molecules bound to a membrane-proximal disordered segment of the protein in the ER-lumen-resident Nterminus (Credle et al., 2005) . Accumulated unfolded proteins in the lumen triggers coupling of this domain from adjacent sensor molecules through a combination of (a) titration of the bound HSPA5 chaperone molecules and (b) direct tethering by malfolded protein molecules (Shamu and Walter, 1996; Credle et al., 2005; Aragon et al., 2009; Korennykh et al., 2009) . Abutting of the luminal domains juxtapose the cytosolic C-terminal segments, leading to an aggregation of the IRE1 molecules into distinct ER-membrane foci (Kimata et al., 2007; Li et al., 2010) . The C-terminal segment has a serine/threonine kinase domain and a RNase domain homologous to that of RNaseL (Figure 1 ; Tirasophon et al., 1998 Tirasophon et al., , 2000 . A trans-autophosphorylation by the kinase domain allosterically activates the RNase domain (Tirasophon et al., 2000; Lee et al., 2008; Korennykh et al., 2009) . In fact, exogenous over-expression of IRE1 in mammalian cells lead to activation suggesting that, under homeostatic conditions, the non-juxtaposition of cytosolic domains maintains an inactive IRE1 (Tirasophon et al., 1998) . Once activated, IRE1 performs cleavage of a variety of RNA substrates mediated by its RNase domain, in addition to phosphorylating and thereby activating JNK (Cox and Walter, 1996; Urano et al., 2000) . Depending on the RNA substrate, the cleavage catalyzed by IRE1 RNase produces differential consequence. Although scission of the Xbp1 mRNA transcript at two internal positions is followed by splicing of the internal segment through ligation of the terminal cleavage products, that in all other known IRE1 target RNA is followed by degradation (Figure 1 ; Sidrauski and Walter, 1997; Calfon et al., 2002) . The latter mode of negative regulation of gene expression is termed as the regulated IRE1-dependent decay (RIDD) pathway (Hollien and Weissman, 2006; Oikawa et al., 2007; Iqbal et al., 2008; Lipson et al., 2008) . Gene transcripts regulated by RIDD pathway includes that from IRE1 (i.e., selftranscripts), probably in a negative feedback loop mechanism (Tirasophon et al., 2000) . In addition to protein coding RNA, RIDD pathway down-regulates the level of a host of microRNA precursors (pre-miRNAs) and can potentially cleave in the anticodon loop of tRNA Phe (Korennykh et al., 2011; Upton et al., 2012) . The IRE1 RNase domain cleaves the Xbp1u (u for unspliced) mRNA transcript at two precise internal positions within the open reading frame (ORF) generating three segments, the terminal two of which are ligated by a tRNA ligase in yeast and by an unknown ligase in mammalian cells, to produce the Xbp1s (s for spliced) mRNA transcript (Figure 1 ; Yoshida et al., 2001) . The Xbp1s thus generated has a longer ORF, which is created by a frame-shift in the coding sequence downstream of the splice site (Cox and Walter, 1996; Calfon et al., 2002) . A similar dual endonucleolytic cleavage is also observed to initiate the XRN1 and Ski2-3-8 dependent degradation of transcripts in the RIDD degradation pathway (Hollien and Weissman, 2006) . The RIDD target transcript genes are predominantly those that encode membrane-associated or secretory proteins and which are not necessary for ER proteinfolding reactions (Hollien and Weissman, 2006) . The cleavage of Xbp1 and the RIDD-target transcripts constitute homeostatic or pro-survival response by IRE1 since XBP1S trans-activates genes encoding multiple chaperones (to fold unfolded proteins) and the ERAD pathway genes (to degrade terminally misfolded proteins) whereas RIDD reduces flux of polypeptides entering the ER lumen (Lee et al., 2003; Hollien and Weissman, 2006) . On the other hand, cleavage of pre-miRNA transcripts which are processed in the cell to generate CASPASE-2 mRNA (Casp2) controlling miRNAs, constitutes the pro-apoptotic function of IRE1 (Upton et al., 2012) . Another pro-apoptotic signal from IRE1 emanates from signaling through phosphorylation of JNK1 (Urano et al., 2000) . Although in the initial phase RIDD activity does not cleave mRNAs encoding essential ER proteins, at later stages of chronic UPR such transcripts are rendered susceptible to degradation promoting apoptosis induction (Han et al., 2009; Bhattacharyya et al., 2014) . Infection of mammalian cells by a multitude of viruses induce an UPR which is sometimes characterized by suppression of signaling by one or more of the three sensor(s; Su et al., 2002; Tardif et al., 2002; He, 2006; Yu et al., 2006 Yu et al., , 2013 Medigeshi et al., 2007; Zhang et al., 2010; Merquiol et al., 2011) . Among these at least two viruses from diverse families, HCMV (a DNA virus) and hepatitis C virus (a hepacivirus), interfere with IRE1 signaling by different mechanism (Tardif et al., 2004; Stahl et al., 2013 ). An observed inhibition of any cellular function by a virus infection could suggest a potential anti-virus function for it, which the virus has evolved to evade through blocking some critical step(s). In both the cases mentioned above, stability of the viral proteins seems to be affected by ERAD-mediated degradation, although other potential anti-viral effect of IRE1 activation are not clear yet (Isler et al., 2005; Saeed et al., 2011) . Interestingly, host mRNA fragments produced following IRE1 activation during bacterial infection, has been shown to activate RIG-I signaling (Figure 1 ; Cho et al., 2013b) . Theoretically, other functions of IRE1 can also have anti-viral effect necessitating its inhibition for uninhibited viral replication. It is, however, still not clear whether IRE1 is able to cleave any viral RNA (or mRNA) in a manner similar to that of other RIDD targets (Figure 1) . The possibilities of such a direct anti-viral function are encouraged by the fact that all these viruses encode at least one protein which, as part of its maturation process, requires glycosylation and disulfide-bond formation. Such a necessity would entail translation of the mRNA encoding such a protein, which in case of positive-sense single-stranded RNA viruses would mean the genome, in association with the ER-membrane (Figure 1 ; Lerner et al., 2003) . Additionally for many RNA viruses, replication complexes are housed in ER-derived vesicular structures (Denison, 2008; den Boon et al., 2010) . Considering the proximity of IRE1 and these virus-derived RNAs it is tempting to speculate that probably at some point of time in the viral life cycle one or more virus-associated RNA would be susceptible to cleavage by IRE1. However, studies with at least two viruses have shown that instead of increasing viral titre, inhibiting the RNase activity of activated IRE1 has an opposite effect (Hassan et al., 2012; Bhattacharyya et al., 2014) . This implies potential benefits of IRE1 activation through one or more of the following, (a) expression of chaperones or other pro-viral molecules downstream of XBP1Supregulation or JNK-activation, (b) cleavage of potential anti-viral gene mRNA transcripts by RIDD activity. However, the mode of protection for the viral RNA from RIDD activity is still not clear. It is possible that the viral proteins create a subdomain within the ER membrane, which through some mechanism excludes IRE1 from diffusing near the genomic RNA, thereby protecting the replication complexes (Denison, 2008) . It is therefore probably not surprising that single-stranded plus-sense RNA viruses encode a polyprotein, which produces replication complexes in cis, promoting formation of such subdomains (Egger et al., 2000) . The fact that IRE1 forms bulky oligomers of higher order probably aggravates such an exclusion of the activated sensor molecules from vicinity of the viral replication complexes. The UPR signaling eventually attenuate during chronic ER-stress and since that is what a virus-induced UPR mimics, probably the viral RNA needs protection only during the initial phase of UPR activation (Lin et al., 2007) . Since the choice of RIDD target seems to be grossly driven towards mRNAs that encode ER-transitory but are not ER-essential proteins, it is also possible that one or more viral protein have evolved to mimic a host protein the transcript of which is RIDD-resistant (Hollien and Weissman, 2006) . Most of the RIDD target mRNA are observed to be ER-membrane associated, the proximity to IRE1 facilitating association and cleavage (Figure 1 ; Hollien and Weissman, 2006) . Although ER-association for an mRNA is possible without the mediation of ribosomes, Gaddam and co-workers reported that continued association with polysomes for a membrane-bound mRNA can confer protection from IRE1 cleavage (Cui et al., 2012; Gaddam et al., 2013) . This would suggest important implications for the observed refractory nature of Japanese encephalitis virus (JEV) and influenza virus RNA to RIDD cleavage (Hassan et al., 2012; Bhattacharyya et al., 2014) . In contrast to Influenza virus, flaviviruses (which include JEV) do not suppress host protein synthesis implying the absence of a global inhibition on translation as would be expected during UPR (Clyde et al., 2006; Edgil et al., 2006) . Therefore, a continued translation of viral RNA in spite of UPR activation can in principle confer protection from the pattern of RNA cleavage observed in the RIDD pathway. IRE1 and RNaseL, in addition to biochemical similarities in protein kinase domain and structural similarities in their RNase domain, share the functional consequences of their activation in initiating cellular apoptosis through JNK signaling (Table 1 and Figure 2 ; Liu and Lin, 2005; Dhanasekaran and Reddy, 2008) . Though initial discoveries were made in the context of homeostatic and anti-viral role for the former and latter, differences between the pathways are narrowed by further advances in research. In the same vein, while inhibition of IRE1 signaling in virus infected cells indicates a potential anti-viral role, www.frontiersin.org (Tirasophon et al., 1998; Dong and Silverman, 1999; Papa et al., 2003; Lin et al., 2007) Nature of RNase substrates Both 28S rRNA and mRNAs IRE1β can cleave both 28S rRNA and mRNA while IRE1α substrates include only mRNAs (Iwawaki et al., 2001) Dissimilarities Cleavage substrates Beside 28S rRNA, predominantly cleaves mRNAs encoding ribosomal proteins (Andersen et al., 2009) Xbp1u and other mRNAs in addition to microRNA precursors which are targeted as part of the RIDD pathway Selection of cleavage site Cleaved between 2nd and 3rd nucleotide positions of UN/N sites (Han et al., 2014) RNA sequence with the consensus of 5 -CUGCAG-3 in association with a stem-loop (SL) structure essential for recognition of Xbp1u and other mRNAs (Oikawa et al., 2010) association of RNaseL mutations with generation of prostate cancer extends the ambit of influence of this anti-viral effector to more non-infectious physiological disorders (Silverman, 2003) . Biochemically, the similarity in their RNase domains does not extend to the choice of either substrates or cleavage point, which are downstream of UU or UA in RNaseL and downstream of G (predominantly) for IRE1 ( Figure 2C ; Yoshida et al., 2001; Hollien and Weissman, 2006; Upton et al., 2012) . Further, while RNaseL cleaves pre-dominantly in single-stranded region, IRE1 seems to cleave equally well in single-and double-stranded region (Upton et al., 2012) . However, a recent report suggested a consensus cleavage site with the sequence UN/N, in RNaseL targets and in those mRNAs that are cleaved by IRE1 as part of the RIDD pathway (Han et al., 2014) . Access to potential cleavage substrate for RNaseL is conjectured to be facilitated through its association with polyribosomes, while no such association is known for IRE1 (Salehzada et al., 1991) . Possibilities exist that IRE1 would have preferential distribution in the rough ER which, upon activation, would give it ready access to mRNAs for initiating the RIDD pathway. In the context of a virus infection, the pathway leading from both these proteins have the potential to lead to cell death. Notwithstanding the fact that this might be an efficient way of virus clearance, it also portends pathological outcomes for the infected organism. Future research would probably lead to design of drugs targeting these proteins based on the structural homology of their effector domains, regulating the pathological denouement of their activation without compromising their anti-viral or potential anti-viral functions.

What is reported in this publication?

Can’t RIDD off viruses https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4061530/ SHA: ef58c6e2790539f30df14acc260ae2af4b5f3d1f Authors: Bhattacharyya, Sankar Date: 2014-06-18 DOI: 10.3389/fmicb.2014.00292 License: cc-by Abstract: The mammalian genome has evolved to encode a battery of mechanisms, to mitigate a progression in the life cycle of an invasive viral pathogen. Although apparently disadvantaged by their dependence on the host biosynthetic processes, an immensely faster rate of evolution provides viruses with an edge in this conflict. In this review, I have discussed the potential anti-virus activity of inositol-requiring enzyme 1 (IRE1), a well characterized effector of the cellular homeostatic response to an overloading of the endoplasmic reticulum (ER) protein-folding capacity. IRE1, an ER-membrane-resident ribonuclease (RNase), upon activation catalyses regulated cleavage of select protein-coding and non-coding host RNAs, using an RNase domain which is homologous to that of the known anti-viral effector RNaseL. The latter operates as part of the Oligoadenylate synthetase OAS/RNaseL system of anti-viral defense mechanism. Protein-coding RNA substrates are differentially treated by the IRE1 RNase to either augment, through cytoplasmic splicing of an intron in the Xbp1 transcript, or suppress gene expression. This referred suppression of gene expression is mediated through degradative cleavage of a select cohort of cellular RNA transcripts, initiating the regulated IRE1-dependent decay (RIDD) pathway. The review first discusses the anti-viral mechanism of the OAS/RNaseL system and evasion tactics employed by different viruses. This is followed by a review of the RIDD pathway and its potential effect on the stability of viral RNAs. I conclude with a comparison of the enzymatic activity of the two RNases followed by deliberations on the physiological consequences of their activation. Text: Establishment of infection by a virus, even in permissive host cells, is beset with a plethora of challenges from innate-antiviral and cell-death pathways. Therefore, the host response to a virus infection might prove to be inhibitory for the viral life cycle in a direct or an indirect manner. The direct mechanism involves expression of multiple anti-viral genes that have evolved to recognize, react, and thereby rid the infected host of the viral nucleic acid (Zhou et al., 1997; Thompson et al., 2011) . On the other hand the pathways, e.g., those that culminate in initiating an apoptotic death for the host cell, indirectly serve to limit the spread of virus (Roulston et al., 1999) . A major difference between these two mechanisms is that while the former response is transmissible to neighboring uninfected cells through interferon (IFN) signaling, the latter is observed mostly in cis. Recent reports, however, have demonstrated transmission of an apoptotic signal between cells that are in contact through gap junctions, although such a signaling from an virus infected host cell to an uninfected one is not known yet (Cusato et al., 2003; Udawatte and Ripps, 2005; Kameritsch et al., 2013) . Successful viral pathogens, through a process of active selection, have evolved to replicate and simultaneously evade or block either of these host responses. The viral nucleic acids which could be the genome (positive-sense singlestranded RNA virus) or RNA derived from transcription of the genome [negative-stranded single-sense RNA or double-stranded RNA (dsRNA) or DNA virus], offer critical targets for both detection and eradication. The viral nucleic acid targeting armaments in the host arsenal include those that recognize the associated molecular patterns like toll-like receptors (TLRs), DDX58 (or RIG-1), IFIH1 (or MDA5), IFIT proteins [IFN-stimulated genes (ISG)56 and ISF54], etc. (Aoshi et al., 2011; Bowzard et al., 2011; Jensen and Thomsen, 2012) . This is followed by IFN signaling and expression or activation of factors that target the inducer for degradation or modification like OAS/ribonuclease L (RNaseL) system, APOBEC3, MCPIP1, the ZC3HAV1/exosome system and RNAi pathways (Gao et al., 2002; Sheehy et al., 2002; Guo et al., 2007; Daffis et al., 2010; Sidahmed and Wilkie, 2010; Schmidt et al., 2012; Cho et al., 2013a; Lin et al., 2013) . In this review we focus on two proteins containing homologous RNase domains, RNaseL with a known direct antiviral function and Inositolrequiring enzyme 1 (IRE1 or ERN1) which has an RNaseL-like RNase domain with a known role in homeostatic response to unfolded proteins in the endoplasmic reticulum (ER) and a potential to function as an antiviral (Figure 1 ; Tirasophon et al., 2000) . In mammalian cells the tell-tale signs of RNA virus infection, like the presence of cytosolic RNA having 5 -ppp or extensive (>30 bp) dsRNA segments are detected by dedicated pathogen associated molecular pattern receptors (PAMPs) or pattern recognition receptors (PRRs) in the host cell, like RIG-1, MDA5, and the IFIT family of proteins (Aoshi et al., 2011; Bowzard et al., 2011; Vabret and Blander, 2013) . The transduction of a signal of this recognition results in the expression of IFN genes the products www.frontiersin.org FIGURE 1 | Schematic representation of the ribonuclease activity of IRE1 and RNaseL showing cross-talk between the paths catalysed by the enzymes. The figure shows activation of RNase activity following dimerization triggered by either accumulation of unfolded proteins in the ER-lumen or synthesis of 2-5A by the enzyme OAS, respectively, for IRE1 and RNaseL. The cleavage of Xbp1u by IRE1 releases an intron thus generating Xbp1s. The IRE1 targets in RIDD pathway or all RNaseL substrates are shown to undergo degradative cleavage. The cleavage products generated through degradation of the respective substrate is shown to potentially interact with RIG-I thereby leading to Interferon secretion and trans-activation of Oas genes through Interferon signaling. Abbreviations: RIG-I = retinoic acid inducible gene-I, Ifnb = interferon beta gene loci, IFN = interferons, ISG = interferon-sensitive genes, 2-5A = 2 -5 oligoadenylates. of which upon secretion outside the cell bind to cognate receptors, initiating further downstream signaling (Figure 1 ; Randall and Goodbourn, 2008) . The genes that are regulated as a result of IFN signaling are termed as IFN-stimulated or IFN-regulated genes (ISGs or IRGs; Sen and Sarkar, 2007; Schoggins and Rice, 2011) . Oligoadenylate synthetase or OAS genes are canonical ISGs that convert ATP into 2 -5 linked oligoadenylates (2-5A) by an unique enzymatic mechanism (Figure 1 ; Hartmann et al., 2003) . Further, they are RNA-binding proteins that function like PRRs, in a way that the 2-5A synthesizing activity needs to be induced through an interaction with dsRNA (Minks et al., 1979; Hartmann et al., 2003) . In a host cell infected by an RNA virus, such dsRNA is present in the form of replication-intermediates (RI), which are synthesized by the virus-encoded RNA-dependent RNA polymerases (RdRp) and subsequently used by the same enzyme to synthesize more genomic RNA, through asymmetric transcription (Weber et al., 2006) . However, the replications complexes (RCs) harboring these RI molecules are found secluded inside host-membrane derived vesicles, at least in positive-strand RNA viruses, a group which contains many human pathogens (Uchil and Satchidanandam, 2003; Denison, 2008) . Reports from different groups suggest OAS proteins to be distributed both in the cytoplasm as well as in membrane-associated fractions, perhaps indicating an evolution of the host anti-viral methodologies towards detection of the membrane-associated viral dsRNAs (Marie et al., 1990; Lin et al., 2009) . DNA viruses on the other hand, produce dsRNA by annealing of RNA derived from transcription of both strands in the same viral genomic loci, which are probably detected by the cytoplasmic pool of OAS proteins (Jacobs and Langland, 1996; Weber et al., 2006) . Post-activation the OAS enzymes synthesize 2-5A molecules in a non-processive reaction producing oligomers which, although potentially ranging in size from dimeric to multimeric, are functionally active only in a trimeric or tetrameric form (Dong et al., 1994; Sarkar et al., 1999; Silverman, 2007) . These small ligands, which bear phosphate groups (1-3) at the 5 end and hydroxyl groups at the 2 and 3 positions, serve as co-factor which can specifically interact with and thereby allosterically activate, existing RNaseL molecules (Knight et al., 1980; Zhou et al., 1997 Zhou et al., , 2005 Sarkar et al., 1999) . As part of a physiological control system these 2-5A oligomers are quite unstable in that they are highly susceptible to degradation by cellular 5 -phosphatases and PDE12 (2 -phosphodiesterase; Silverman et al., 1981; Johnston and Hearl, 1987; Kubota et al., 2004; Schmidt et al., 2012) . Viral strategies to evade or overcome this host defense mechanism ranges from preventing IFN signaling which would hinder the induction of OAS expression or thwarting activation of expressed OAS proteins by either shielding the viral dsRNA from interacting with it or modulating the host pathway to synthesize inactive 2-5A derivatives (Cayley et al., 1984; Hersh et al., 1984; Rice et al., 1985; Maitra et al., 1994; Beattie et al., 1995; Rivas et al., 1998; Child et al., 2004; Min and Krug, 2006; Sanchez and Mohr, 2007; Sorgeloos et al., 2013) . Shielding of viral RNA from interacting with OAS is possible through enclosure of dsRNA replication intermediates in membrane enclosed compartments as observed in many flaviviruses (Ahlquist, 2006; Miller and Krijnse-Locker, 2008; Miorin et al., 2013) . RNaseL is a 741 amino acid protein containing three predominantly structured region, an N-terminal ankyrin repeat domain (ARD), a middle catalytically inactive pseudo-kinase (PK) and a C-terminal RNase domain (Figure 2A ; Hassel et al., 1993; Zhou et al., 1993) . The activity of the RNase domain is negatively regulated by the ARD, which is relieved upon binding of 2-5A molecules to ankyrin repeats 2 and 4 followed by a conformational alteration (Figure 1 ; Hassel et al., 1993; Tanaka et al., 2004; Nakanishi et al., 2005) . In support of this contention, deletion of the ARD has been demonstrated to produce constitutively active RNaseL, although with dramatically lower RNase activity (Dong and Silverman, 1997) . However, recent reports suggest that while 2-5A links the ankyrin repeats from adjacent molecules leading to formation of dimer and higher order structures, at sufficiently high in vitro concentrations, RNaseL could oligomerize even in the absence of 2-5A . Nonetheless, in vivo the RNaseL nuclease activity still seems to be under the sole regulation of 2-5A (Al-Saif and Khabar, 2012) . In order to exploit this dependence, multiple viruses like mouse hepatitis virus (MHV) and rotavirus group A (RVA) have evolved to encode phosphodiesterases capable of hydrolysing the 2 -5 linkages in 2-5A and thereby attenuate the RNaseL cleavage activity (Zhao et al., 2012; Zhang et al., 2013) . In addition to 5 -phosphatases and 2 -phosphodiesterases to reduce ClustalW alignment of primary sequence from a segment of the PK domain indicating amino acid residues which are important for interacting with nucleotide cofactors. The conserved lysine residues, critical for this interaction (K599 for IRE1 and K392 in RNaseL) are underlined. (C) Alignment of the KEN domains in RNaseL and IRE1. The amino acids highlighted and numbered in IRE1 are critical for the IRE1 RNase activity (Tirasophon et al., 2000) . the endogenous 2-5A levels, mammalian genomes encode posttranscriptional and post-translation inhibitors of RNaseL activity in the form of microRNA-29 and the protein ABCE1 (RNaseL inhibitor or RLI), respectively (Bisbal et al., 1995; Lee et al., 2013) . Direct inhibition of RNaseL function is also observed upon infection by Picornaviruses through, either inducing the expression of ABCE1 or exercising a unique inhibitory property of a segment of the viral RNA (Martinand et al., 1998 (Martinand et al., , 1999 Townsend et al., 2008; Sorgeloos et al., 2013) . Once activated by 2-5A, RNaseL can degrade single-stranded RNA irrespective of its origin (virus or host) although there seems to exist a bias towards cleavage of viral RNA (Wreschner et al., 1981a; Silverman et al., 1983; Li et al., 1998) . RNA sequences that are predominantly cleaved by RNaseL are U-rich with the cleavage points being typically at the 3 end of UA or UG or UU di-nucleotides, leaving a 5 -OH and a 3 -monophosphate in the cleavage product (Floyd-Smith et al., 1981; Wreschner et al., 1981b) . A recent report shows a more general consensus of 5 -UNN-3 with the cleavage point between the second and the third nucleotide (Han et al., 2014) . Cellular targets of RNaseL include both ribosomal RNA (rRNA) and mRNAs, the latter predominantly representing genes involved in protein biosynthesis (Wreschner et al., 1981a; Al-Ahmadi et al., 2009; Andersen et al., 2009) . Additionally, RNaseL activity can also degrade specific ISG mRNA transcripts and thereby attenuate the effect of IFN signaling (Li et al., 2000) . Probably an evolution towards insulating gene expression from RNaseL activity is observed in the coding region of mammalian genes where the UU/UA dinucleotide frequency is rarer (Bisbal et al., 2000; Khabar et al., 2003; Al-Saif and Khabar, 2012) . Perhaps not surprisingly, with a much faster rate of evolution, similar observations have been made with respect to evasion of RNaseL mediated degradation by viral RNAs too (Han and Barton, 2002; Washenberger et al., 2007) . Moreover, nucleoside modifications in host mRNAs, rarely observed in viral RNAs, have also been shown to confer protection from RNaseL (Anderson et al., 2011) . In addition to directly targeting viral RNA, the reduction in functional ribosomes and ribosomal protein mRNA affects viral protein synthesis and replication in an indirect manner. Probably, as a reflection of these effects on cellular RNAs, RNaseL is implicated as one of the factors determining the anti-proliferative effect of IFN activity . The anti-viral activity of RNaseL extends beyond direct cleavage of viral RNA, through stimulation of RIG-I by the cleavage product (Malathi et al., , 2007 (Malathi et al., , 2010 . A global effect of RNaseL is observed in the form of autophagy induced through c-jun N-terminal kinase (JNK) signaling and apoptosis, probably as a consequence of rRNA cleavage (Li et al., 2004; Chakrabarti et al., 2012; Siddiqui and Malathi, 2012) . RNaseL has also been demonstrated to play a role in apoptotic cell death initiated by pharmacological agents extending the physiological role of this pathway beyond the boundary of being only an anti-viral mechanism (Castelli et al., 1997 (Castelli et al., , 1998 . The ER serves as a conduit for maturation of cellular proteins which are either secreted or destined to be associated with a membrane for its function. An exclusive microenvironment (high Calcium ion and unique ratio of reduced to oxidized glutathione) along with a battery of ER-lumen resident enzymes (foldases, chaperones, and lectins) catalyse/mediate the necessary folding, disulfide-bond formation, and glycosylation reactions (Schroder and Kaufman, 2005) . A perturbation of the folding capacity, due to either physiological disturbances or virus infection, can lead to an accumulation of unfolded proteins in the ER lumen, which signals an unfolded protein response (UPR). UPR encompasses a networked transcriptional and translational gene-expression program, initiated by three ER-membrane resident sensors namely IRE1 or ERN1, PKR-like ER Kinase (PERK or EIF2AK3) and activating transcription factor 6 (ATF6; Hetz, 2012) . IRE1 is a type I single-pass trans-membrane protein in which, similar to what is observed with RNaseL, the N-terminal resident in the ER lumen serves as sensor and the cytosolic C-terminal as the effector (Figure 1 ; Chen and Brandizzi, 2013) . The IRE1 coding gene is present in genomes ranging from yeast to mammals and in the latter is ubiquitously expressed in all tissues (Tirasophon et al., 1998) . Signal transduction by stimulated IRE1 initiates multiple gene regulatory pathways with either pro-survival or pro-apoptotic consequences (Kaufman, 1999) . During homeostasis or unstressed conditions the sensor molecules are monomeric, a state maintained co-operatively by the " absence" of unfolded proteins and the "presence" of HSPA5 (GRP78 or Bip, an ERresident chaperone) molecules bound to a membrane-proximal disordered segment of the protein in the ER-lumen-resident Nterminus (Credle et al., 2005) . Accumulated unfolded proteins in the lumen triggers coupling of this domain from adjacent sensor molecules through a combination of (a) titration of the bound HSPA5 chaperone molecules and (b) direct tethering by malfolded protein molecules (Shamu and Walter, 1996; Credle et al., 2005; Aragon et al., 2009; Korennykh et al., 2009) . Abutting of the luminal domains juxtapose the cytosolic C-terminal segments, leading to an aggregation of the IRE1 molecules into distinct ER-membrane foci (Kimata et al., 2007; Li et al., 2010) . The C-terminal segment has a serine/threonine kinase domain and a RNase domain homologous to that of RNaseL (Figure 1 ; Tirasophon et al., 1998 Tirasophon et al., , 2000 . A trans-autophosphorylation by the kinase domain allosterically activates the RNase domain (Tirasophon et al., 2000; Lee et al., 2008; Korennykh et al., 2009) . In fact, exogenous over-expression of IRE1 in mammalian cells lead to activation suggesting that, under homeostatic conditions, the non-juxtaposition of cytosolic domains maintains an inactive IRE1 (Tirasophon et al., 1998) . Once activated, IRE1 performs cleavage of a variety of RNA substrates mediated by its RNase domain, in addition to phosphorylating and thereby activating JNK (Cox and Walter, 1996; Urano et al., 2000) . Depending on the RNA substrate, the cleavage catalyzed by IRE1 RNase produces differential consequence. Although scission of the Xbp1 mRNA transcript at two internal positions is followed by splicing of the internal segment through ligation of the terminal cleavage products, that in all other known IRE1 target RNA is followed by degradation (Figure 1 ; Sidrauski and Walter, 1997; Calfon et al., 2002) . The latter mode of negative regulation of gene expression is termed as the regulated IRE1-dependent decay (RIDD) pathway (Hollien and Weissman, 2006; Oikawa et al., 2007; Iqbal et al., 2008; Lipson et al., 2008) . Gene transcripts regulated by RIDD pathway includes that from IRE1 (i.e., selftranscripts), probably in a negative feedback loop mechanism (Tirasophon et al., 2000) . In addition to protein coding RNA, RIDD pathway down-regulates the level of a host of microRNA precursors (pre-miRNAs) and can potentially cleave in the anticodon loop of tRNA Phe (Korennykh et al., 2011; Upton et al., 2012) . The IRE1 RNase domain cleaves the Xbp1u (u for unspliced) mRNA transcript at two precise internal positions within the open reading frame (ORF) generating three segments, the terminal two of which are ligated by a tRNA ligase in yeast and by an unknown ligase in mammalian cells, to produce the Xbp1s (s for spliced) mRNA transcript (Figure 1 ; Yoshida et al., 2001) . The Xbp1s thus generated has a longer ORF, which is created by a frame-shift in the coding sequence downstream of the splice site (Cox and Walter, 1996; Calfon et al., 2002) . A similar dual endonucleolytic cleavage is also observed to initiate the XRN1 and Ski2-3-8 dependent degradation of transcripts in the RIDD degradation pathway (Hollien and Weissman, 2006) . The RIDD target transcript genes are predominantly those that encode membrane-associated or secretory proteins and which are not necessary for ER proteinfolding reactions (Hollien and Weissman, 2006) . The cleavage of Xbp1 and the RIDD-target transcripts constitute homeostatic or pro-survival response by IRE1 since XBP1S trans-activates genes encoding multiple chaperones (to fold unfolded proteins) and the ERAD pathway genes (to degrade terminally misfolded proteins) whereas RIDD reduces flux of polypeptides entering the ER lumen (Lee et al., 2003; Hollien and Weissman, 2006) . On the other hand, cleavage of pre-miRNA transcripts which are processed in the cell to generate CASPASE-2 mRNA (Casp2) controlling miRNAs, constitutes the pro-apoptotic function of IRE1 (Upton et al., 2012) . Another pro-apoptotic signal from IRE1 emanates from signaling through phosphorylation of JNK1 (Urano et al., 2000) . Although in the initial phase RIDD activity does not cleave mRNAs encoding essential ER proteins, at later stages of chronic UPR such transcripts are rendered susceptible to degradation promoting apoptosis induction (Han et al., 2009; Bhattacharyya et al., 2014) . Infection of mammalian cells by a multitude of viruses induce an UPR which is sometimes characterized by suppression of signaling by one or more of the three sensor(s; Su et al., 2002; Tardif et al., 2002; He, 2006; Yu et al., 2006 Yu et al., , 2013 Medigeshi et al., 2007; Zhang et al., 2010; Merquiol et al., 2011) . Among these at least two viruses from diverse families, HCMV (a DNA virus) and hepatitis C virus (a hepacivirus), interfere with IRE1 signaling by different mechanism (Tardif et al., 2004; Stahl et al., 2013 ). An observed inhibition of any cellular function by a virus infection could suggest a potential anti-virus function for it, which the virus has evolved to evade through blocking some critical step(s). In both the cases mentioned above, stability of the viral proteins seems to be affected by ERAD-mediated degradation, although other potential anti-viral effect of IRE1 activation are not clear yet (Isler et al., 2005; Saeed et al., 2011) . Interestingly, host mRNA fragments produced following IRE1 activation during bacterial infection, has been shown to activate RIG-I signaling (Figure 1 ; Cho et al., 2013b) . Theoretically, other functions of IRE1 can also have anti-viral effect necessitating its inhibition for uninhibited viral replication. It is, however, still not clear whether IRE1 is able to cleave any viral RNA (or mRNA) in a manner similar to that of other RIDD targets (Figure 1) . The possibilities of such a direct anti-viral function are encouraged by the fact that all these viruses encode at least one protein which, as part of its maturation process, requires glycosylation and disulfide-bond formation. Such a necessity would entail translation of the mRNA encoding such a protein, which in case of positive-sense single-stranded RNA viruses would mean the genome, in association with the ER-membrane (Figure 1 ; Lerner et al., 2003) . Additionally for many RNA viruses, replication complexes are housed in ER-derived vesicular structures (Denison, 2008; den Boon et al., 2010) . Considering the proximity of IRE1 and these virus-derived RNAs it is tempting to speculate that probably at some point of time in the viral life cycle one or more virus-associated RNA would be susceptible to cleavage by IRE1. However, studies with at least two viruses have shown that instead of increasing viral titre, inhibiting the RNase activity of activated IRE1 has an opposite effect (Hassan et al., 2012; Bhattacharyya et al., 2014) . This implies potential benefits of IRE1 activation through one or more of the following, (a) expression of chaperones or other pro-viral molecules downstream of XBP1Supregulation or JNK-activation, (b) cleavage of potential anti-viral gene mRNA transcripts by RIDD activity. However, the mode of protection for the viral RNA from RIDD activity is still not clear. It is possible that the viral proteins create a subdomain within the ER membrane, which through some mechanism excludes IRE1 from diffusing near the genomic RNA, thereby protecting the replication complexes (Denison, 2008) . It is therefore probably not surprising that single-stranded plus-sense RNA viruses encode a polyprotein, which produces replication complexes in cis, promoting formation of such subdomains (Egger et al., 2000) . The fact that IRE1 forms bulky oligomers of higher order probably aggravates such an exclusion of the activated sensor molecules from vicinity of the viral replication complexes. The UPR signaling eventually attenuate during chronic ER-stress and since that is what a virus-induced UPR mimics, probably the viral RNA needs protection only during the initial phase of UPR activation (Lin et al., 2007) . Since the choice of RIDD target seems to be grossly driven towards mRNAs that encode ER-transitory but are not ER-essential proteins, it is also possible that one or more viral protein have evolved to mimic a host protein the transcript of which is RIDD-resistant (Hollien and Weissman, 2006) . Most of the RIDD target mRNA are observed to be ER-membrane associated, the proximity to IRE1 facilitating association and cleavage (Figure 1 ; Hollien and Weissman, 2006) . Although ER-association for an mRNA is possible without the mediation of ribosomes, Gaddam and co-workers reported that continued association with polysomes for a membrane-bound mRNA can confer protection from IRE1 cleavage (Cui et al., 2012; Gaddam et al., 2013) . This would suggest important implications for the observed refractory nature of Japanese encephalitis virus (JEV) and influenza virus RNA to RIDD cleavage (Hassan et al., 2012; Bhattacharyya et al., 2014) . In contrast to Influenza virus, flaviviruses (which include JEV) do not suppress host protein synthesis implying the absence of a global inhibition on translation as would be expected during UPR (Clyde et al., 2006; Edgil et al., 2006) . Therefore, a continued translation of viral RNA in spite of UPR activation can in principle confer protection from the pattern of RNA cleavage observed in the RIDD pathway. IRE1 and RNaseL, in addition to biochemical similarities in protein kinase domain and structural similarities in their RNase domain, share the functional consequences of their activation in initiating cellular apoptosis through JNK signaling (Table 1 and Figure 2 ; Liu and Lin, 2005; Dhanasekaran and Reddy, 2008) . Though initial discoveries were made in the context of homeostatic and anti-viral role for the former and latter, differences between the pathways are narrowed by further advances in research. In the same vein, while inhibition of IRE1 signaling in virus infected cells indicates a potential anti-viral role, www.frontiersin.org (Tirasophon et al., 1998; Dong and Silverman, 1999; Papa et al., 2003; Lin et al., 2007) Nature of RNase substrates Both 28S rRNA and mRNAs IRE1β can cleave both 28S rRNA and mRNA while IRE1α substrates include only mRNAs (Iwawaki et al., 2001) Dissimilarities Cleavage substrates Beside 28S rRNA, predominantly cleaves mRNAs encoding ribosomal proteins (Andersen et al., 2009) Xbp1u and other mRNAs in addition to microRNA precursors which are targeted as part of the RIDD pathway Selection of cleavage site Cleaved between 2nd and 3rd nucleotide positions of UN/N sites (Han et al., 2014) RNA sequence with the consensus of 5 -CUGCAG-3 in association with a stem-loop (SL) structure essential for recognition of Xbp1u and other mRNAs (Oikawa et al., 2010) association of RNaseL mutations with generation of prostate cancer extends the ambit of influence of this anti-viral effector to more non-infectious physiological disorders (Silverman, 2003) . Biochemically, the similarity in their RNase domains does not extend to the choice of either substrates or cleavage point, which are downstream of UU or UA in RNaseL and downstream of G (predominantly) for IRE1 ( Figure 2C ; Yoshida et al., 2001; Hollien and Weissman, 2006; Upton et al., 2012) . Further, while RNaseL cleaves pre-dominantly in single-stranded region, IRE1 seems to cleave equally well in single-and double-stranded region (Upton et al., 2012) . However, a recent report suggested a consensus cleavage site with the sequence UN/N, in RNaseL targets and in those mRNAs that are cleaved by IRE1 as part of the RIDD pathway (Han et al., 2014) . Access to potential cleavage substrate for RNaseL is conjectured to be facilitated through its association with polyribosomes, while no such association is known for IRE1 (Salehzada et al., 1991) . Possibilities exist that IRE1 would have preferential distribution in the rough ER which, upon activation, would give it ready access to mRNAs for initiating the RIDD pathway. In the context of a virus infection, the pathway leading from both these proteins have the potential to lead to cell death. Notwithstanding the fact that this might be an efficient way of virus clearance, it also portends pathological outcomes for the infected organism. Future research would probably lead to design of drugs targeting these proteins based on the structural homology of their effector domains, regulating the pathological denouement of their activation without compromising their anti-viral or potential anti-viral functions.

What is the conclusion regarding IRE1 and RNaseL proteins?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

Where was the coronavirus discovered?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the number of confirmed cases reached on 8 February 2020?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is presented in this study?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the proposed model?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

When is SSA generally employed?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the main idea behind the proposed model?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the proposed model called?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

How is FPASSA-ANFIS model evaluated?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What did the comparison of the FPASSA-ANFIS model with several existing models, show?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

How was the proposed model tested?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What were the outcomes of the test?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What are the large family of viruses, called coronaviruses?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

Among whom are the coronaviruses distributed?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

In what the Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What does ANFIS offer?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

In which applications has it been applied?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is SI?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is PSO?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is SCA?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is MVO?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

For what SCA algorithm was applied to improve the ANFIS model ?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What may be a problem with individual SI algorithm?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is OWA?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is proposed in the current study?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the FPA optimization algorithm inspired by?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the SSA optimization algorithm inspired by?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What is the proposed FPASSA method?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

How does the proposed FPASSA start?

Optimization Method for Forecasting Confirmed Cases of COVID-19 in China https://doi.org/10.3390/jcm9030674 SHA: 1d7f8850c5244fdc9b387038e7eeae9bcbbde6d2 Authors: Al-Qaness, Mohammed A. A.; Ewees, Ahmed A.; Fan, Hong; Abd El Aziz, Mohamed Date: 2020 DOI: 10.3390/jcm9030674 License: cc-by Abstract: In December 2019, a novel coronavirus, called COVID-19, was discovered in Wuhan, China, and has spread to different cities in China as well as to 24 other countries. The number of confirmed cases is increasing daily and reached 34,598 on 8 February 2020. In the current study, we present a new forecasting model to estimate and forecast the number of confirmed cases of COVID-19 in the upcoming ten days based on the previously confirmed cases recorded in China. The proposed model is an improved adaptive neuro-fuzzy inference system (ANFIS) using an enhanced flower pollination algorithm (FPA) by using the salp swarm algorithm (SSA). In general, SSA is employed to improve FPA to avoid its drawbacks (i.e., getting trapped at the local optima). The main idea of the proposed model, called FPASSA-ANFIS, is to improve the performance of ANFIS by determining the parameters of ANFIS using FPASSA. The FPASSA-ANFIS model is evaluated using the World Health Organization (WHO) official data of the outbreak of the COVID-19 to forecast the confirmed cases of the upcoming ten days. More so, the FPASSA-ANFIS model is compared to several existing models, and it showed better performance in terms of Mean Absolute Percentage Error (MAPE), Root Mean Squared Relative Error (RMSRE), Root Mean Squared Relative Error (RMSRE), coefficient of determination ( R 2 ), and computing time. Furthermore, we tested the proposed model using two different datasets of weekly influenza confirmed cases in two countries, namely the USA and China. The outcomes also showed good performances. Text: A large family of viruses, called coronaviruses, are severe pathogens for human beings, which infect respiratory, hepatic, gastrointestinal, and neurologic diseases. They are distributed among humans, birds, livestock, mice, bats, and other wild animals [1] [2] [3] . The outbreaks of two previous coronaviruses, SARS-CoV and MERS-CoV in 2003 and 2012, respectively, have approved the transmission from animal to animal, and human to human [4] . In December 2019, the World Health Organization (WHO) received notifications from China for many cases of respiratory illness that were linked to some people who had visited a seafood market in Wuhan [5] . Currently, Wuhan city suffers from the spreading of a novel coronavirus, called COVID-19 (previously, it was called 2019-nCoV). In [6] , the authors concluded that COVID-19 likely originated in bats, because it is more similar to two bat-derived coronavirus strains. However, the source of the COVID-19 is not confirmed yet, and it communities, Hong Kong and Toronto, were 1.2 and 1.32, respectively. Ong et al. [20] proposed a monitoring and forecasting model for influenza A (H1N1-2009). Furthermore, Nah et al. [21] proposed a probability-based model to predict the spread of the MERS. The Adaptive Neuro-Fuzzy Inference System (ANFIS) [22] is widely applied in time series prediction and forecasting problems, and it showed good performance in many existing applications. It offers flexibility in determining nonlinearity in the time series data, as well as combining the properties of both artificial neural networks (ANN) and fuzzy logic systems. It has been applied in various forecasting applications, for example, in [23] , a stock price forecasting model was proposed using ANFIS and empirical mode decomposition. Chen et al. [24] proposed a TAIEX time series forecasting model based on a hybrid of ANFIS and ordered weighted averaging (OWA). In [25] , another time series forecasting method was presented for electricity prices based on ANFIS. Svalina et al. [26] proposed an ANFIS based forecasting model for close price indices for a stock market for five days. Ekici and Aksoy [27] presented an ANFIS based building energy consumption forecasting model. More so, ANFIS is also applied to forecast electricity loads [28] . Kumar et al. [29] proposed an ANFIS based model to forecast return products. Ho and Tsai [30] applied ANFIS to forecast product development performance. However, estimating ANFIS parameters is a challenge that needs to be improved. Therefore, in previous studies, some individual swarm intelligence (SI) methods have been applied to the ANFIS parameters to enhance time series forecasting because these parameters have a significant effect on the performance of ANFIS. The SI methods include the particle swarm optimization (PSO) [31, 32] , social-spider optimization [33] , sine-cosine algorithm (SCA) [34] , and multi-verse optimizer (MVO) [35] . For example, in [34] SCA algorithm was applied to improve the ANFIS model to forecast oil consumption in three countries, namely, Canada, Germany, and Japan. In the same context, in [35] , The MVO algorithm was used to enhance the ANFIS model to forecast oil consumption in two countries. In addition, in [36] the PSO was used with ANFIS to predict biochar yield. However, individual SI algorithms may stock at local optima. Therefore, one solution is to apply hybrid SI algorithms to avoid this problem. In [37] , a hybrid of two SI algorithms, namely GA and SSA, was presented to improve the ANFIS model. The proposed new model called GA-SSA-ANFIS was applied to forecast crude oil prices for long-term time series data. However, the previously mentioned methods suffer from some limitations that can affect the performance of the forecasting output such as slow convergence and the ability to balance between exploration and exploitation phases can influence the quality of the final output. This motivated us to propose an alternative forecasting method dependent on the hybridization concept. This concept avoids the limitations of traditional SI techniques by combining the strengths of different techniques, and this produces new SI techniques that are better than traditional ones. In the current study, we propose an improved ANFIS model based on a modified flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). The FPA is an optimization algorithm proposed by Yang [38] , which was inspired by the flow pollination process of the flowering plants. The FPA was employed in various optimization applications, for example to estimate solar PV parameter [39, 40] , solving sudoku puzzles [41] , feature selection [42] , antenna design [43] , and other applications [44] [45] [46] [47] . Moreover, SSA is also an optimization algorithm proposed by Mirjalili et al. [48] inspired by the behavior of salp chains. In recent years, the SSA was utilized to solve different optimization problems, such as feature selection [49, 50] , data classification [51] , image segmentation [52] , and others [53, 54] . The proposed method called FPASSA is a hybrid of FPA and SSA, in which the SSA is applied as a local search method for FPA. The proposed FPASSA starts by receiving the historical COVID-19 dataset. Then a set of solutions is generated where each of them represents the value for the parameters of the ANFIS model. Then the quality of each solution is calculated using the fitness value, and the solution that has the best fitness value is chosen to represent the best solution. Then the probability of each solution is computed. Then the current solution will be updated, either using global or local strategy in FPA. However, in the case of local strategy, the operators of SSA or FPA will be used according to the probability of the fitness value for each solution. The process of updating the solutions is repeated until reaching the stop condition, and the best parameter configurations are used to forecast the number of confirmed cases of COVID-19. The main contribution points of the current study are as follows: 1. We propose an efficient forecasting model to forecast the confirmed cases of the COVID-19 in China for the upcoming ten days based on previously confirmed cases. An improved ANFIS model is proposed using a modified FPA algorithm, using SSA. We compare the proposed model with the original ANFIS and existing modified ANFIS models, such as PSO, GA, ABC, and FPA. The rest of this study is organized as follows. The preliminaries of ANFIS, FPA, and SSA are described in Section 2. Section 3 presents the proposed FPASSA, and Section 4 presents the experimental setup and results. We conclude this study in Section 5. The principles of the ANFIS are given in this section. The ANFIS model links the fuzzy logic and neural networks [22] . It generates a mapping between the input and output by applying IF-THEN rules (it is also called Takagi-Sugeno inference model). Figure 1 illustrates the ANFIS model where, y and x define the inputs to Layer 1 whereas, O 1i is its output of node i that is computed as follows: where µ denotes the generalized Gaussian membership functions. A i and B i define the membership values of µ. α i and ρ i denote the premise parameters set. The output of Layer 2 (it is also known as the firing strength of a rule) is calculated as follows: Meanwhile, the output of Layer 3 (it is also known as the normalized firing strength) is calculated as follows: The output of Layer 4 (it is also known as an adaptive node) is calculated as follows: where r i , q i , and p i define the consequent parameters of the node i. Layer 5 contains only one node; its output is computed as: Flower Pollination Algorithm is an optimization method proposed by Yang [38] . It simulates the transfer of flowers' pollen by pollinators in nature. This algorithm utilizes the two types of pollination (i.e., self-pollination and cross-pollination). In self-pollination, the pollination occurs with no pollinators, whereas, in cross-pollination, the pollens are moved between different plants. In more detail, the self-pollination can be represented as a local pollination while the cross-pollination can be called global pollination. The global pollination or cross-pollination can be mathematically formed as follows: where x t i defines the pollen i at iteration t. L denotes the pollination's strength or the step size. F * is the target position or best solution. In some cases, insects can fly with different distance steps for a long space; therefore, Levy fly distribution is applied to simulate this movement. where λ = 1.5. Γ(λ) denotes the gamma function. This distribution is available for large steps s > 0. The self-pollination or local pollination can be mathematically formed as follows: where x t i and x k i represent pollens from different flower in the same plant. in the range [0,1] The process of pollination can be done using cross-pollination or self-pollination. Therefore, the random variable p, in the range [0, 1], is used to determine this process. SSA is an optimization technique introduced by [48] . It simulates the Salps' behavior in nature. This behavior is called salp chain. The mathematical model of SSA begins by splinting its population into a leader group and followers group. The leader is the front salp, whereas, the followers are the other salps. The search space is determined in n-dimensions with n variables. Equation (10) works to update the salps' positions. where x 1 j denotes the leader's position in j-th dimension. F j is the target position. ub j and lb j represent the max and min bounds, respectively. c 2 and c 3 denote random numbers in [0, 1]. c 1 is an important parameter; it balances between the exploration and exploitation phases. It is computed as follows: where the current loop number is t and the max loop' number is t max . Then, the followers' position is updated as follows: where x i j defines the i-th position of the follower in j-th dimension. i > 1. This section explains the proposed FPASSA-ANFIS method. It is a time series method for forecasting the confirmed cases of the COVID-19, as given in Figure 2 . The FPASSA-ANFIS utilizes the improved FPA to train the ANFIS model by optimizing its parameters. The FPASSA-ANFIS contains five layers as the classic ANFIS model. Layer 1 contains the input variables (the historical COVID-19 confirmed cases). Whereas Layer 5 produces the forecasted values. In the learning phase, the FPASSA is used to select the best weights between Layer 4 and Layer 5. The FPASSA-ANFIS starts by formatting the input data in a time series form. In our case, the autocorrelation function (ACF) was considered. ACF is one of the methods applied to find patterns in the data; it presents information about the correlation between points separated by various time lags. Therefore, in this paper, the variables with ACF greater than 0.2 are considered i.e., 5-lags. Besides, the training data contains 75% of the dataset, whereas the testing data contains 25% of them. The number of clusters is defined by the fuzzy c-mean (FCM) method to construct the ANFIS model. The parameters of the ANFIS model are prepared by the FPASSA algorithm. In the training phase, the calculation error (as in Equation (13)) between the real data and the predicted data is used to evaluate the parameters' quality. where T is the real data, and P is the predicted data. N s is the sample length. The smaller values of the objective function indicate good ANFIS's parameter. On the other hand, the updating phase of the followers' positions in the SSA algorithm is applied to improve the global pollination phase in the FPA algorithm. In this improvement, there is a random variable (r) used to switch between both phases. If r > 0.5, then the operators of the SSA is used; otherwise, the operators of the FPA are used. In general, The FPASSA starts by constructing the population (X); afterward, the objective function is calculated for each solution. The solution with the lowest error value is saved to the next iteration. This sequence is repeated until meeting the stop condition, which in this paper, is the maximum number of iterations. Then the best solution is passed to train the parameters of the ANFIS model. After finishing the training phase, the testing phase is started with the best solution to compute the final output. The performance of the proposed method is evaluated by comparing the real data with the predicted data using the performance measures. Finally, the FPASSA produces a foretasted value for confirmed cases of COVID-19 in China in the next day. The steps of the proposed FPASSA are presented in Algorithm 1. Input: Historical COVID-19 dataset, size of population N, total number of iterations t max . Divide the data into training and testing sets. Using Fuzzy c-mean method to determine the number of membership functions. Constructing the ANFIS network. Set the initial value for N solutions (X). Return the best solution that represents the best configuration for ANFIS. Apply the testing set to the best ANFIS model. Forecasting the COVID-19 for the next ten days. This section presents the description of the used dataset, the performance measures, the parameter setting for all methods, the experiment results, and discussions. The main dataset of this study is COVID-19 dataset. It was collected from the WHO website (https: //www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports/). It contains the daily confirmed cases in China from 21 January 2020 to 18 February 2020, as shown in Table 1 . We used 75% from the dataset to train the model while the rest is used to test it. Moreover, we evaluated the performance of the proposed method using two datasets of weekly influenza confirmed cases. The first one is called DS1; it was collected from the Centers for Disease Control and Prevention (CDC) (https://www.cdc.gov/flu/weekly/). It starts from week number 40 in 2015 and continues until week number 6 in 2020. Whereas, the second one is called DS2. It was collected from the WHO website (https://www.who.int/influenza). It contains the data of weekly influenza confirmed cases in China from week number 1 in 2016 to week number 8 in 2020. The quality of the proposed method is evaluated using a set of performance metrics as follows: • Root Mean Square Error (RMSE): where Yp and Y are the predicted and original values, respectively. • Mean Absolute Error (MAE): • Mean Absolute Percentage Error (MAPE): • Root Mean Squared Relative Error (RMSRE): N s represents the sample size of the data. • Coefficient of Determination (R 2 ): where Y represents the average of Y. The lowest value of RMSE, MAE, MAPE, and RMSRE refers to the best method. The higher value of R 2 indicates better correlation for the method. This paper aims to assess the ability of the FPASSA to forecast the COVID-19 by comparing its performance with other methods, namely the ANFIS and the trained ANFIS models using PSO, GA, ABC, FPA, and FPASSA. The parameters' setting for these models is listed in Table 2 . The common parameters, such as population size, are set to 25 and 100 iterations are applied. Besides, each algorithm is performed for 30 independent runs to fair comparisons. The selected parameters are chosen because they produced good behavior in previous experiments, such as [34, 35, 55, 56] . Table 2 . Parameters' setting. Parameters Setting Max. epochs = 100, Error goal = 0, Initial step = 0.01, Decrease rate = 0.9, Increase rate = 1. In this section, the performance of the proposed FPASSA to predict the DS1 and DS2 is discussed. It can be concluded from Table 3 that the performance of FPASSA outperformed the compared methods in all measures, whereas the FPA is ranked second. The results of DS2 indicate that the FPASSA is ranked first in terms of RMSE, MAPE, R 2 , and the CPU time. Whereas, the PSO is ranked second, followed by the FPA, GA, then ABC. These results denote that the proposed method can optimize the parameters of the ANFIS model effectively and produce good results in terms of the performance measures. Comparison results between the proposed FPASSA and other models to forecast COVID-19 are given in Table 4 . It can be concluded that the FPASSA outperforms other models. For example, by analyzing the results of RMSE, MAE, MAPE, RMSRE, and CPU time(s) it can be observed that the FPASSA achieves the smallest value among the comparison algorithms, and this indicates the high quality of the FPASSA. Meanwhile, the FPA allocates the second rank, which provides better results than the rest of the methods. Moreover, the value of R 2 refers to the high correlation between the prediction obtained by the proposed FPASSA method and the original COVID-19, which has nearly 0.97. This can also be noticed from Figure 3 , which depicts the training of the algorithms using the historical data of the COVID-19 as well as their forecasting values for ten days. Table 5 depicts the forecasting value for the confirmed cases of the COVID-19 in China from 19/2/2020 to 28/2/2020. From these results, it can be noticed that the outbreak will reach its highest level on the day 28/2/2020. The average percentage of the increase over the forecasted period is 10%, the highest percentage is 12% on 28/2/2020, and the lowest percentage is 8.7% on 19/2/2020. From the previous results, it can be concluded that the proposed FPASSA-ANFIS has a high ability to forecast the COVID-19 dataset. These results avoid the limitations of traditional ANFIS because of the combination with the modified FPA method. Moreover, the operators of SSA are combined with the local strategy of FPA to enhance their exploitation ability. However, the time computational of the proposed FPASSA method still requires more improvements. This paper proposed a modified version for the flower pollination algorithm (FPA) using the salp swarm algorithm (SSA). This modified version, called FPASSA, is applied to improve the performance of the ANFIS through determining the optimal value for its parameters. The developed FPASSA-ANFIS model is applied as a forecasting technique for a novel coronavirus, called COVID-19, that was discovered in Wuhan, China at the end of last year and January of the current year. The proposed FPASSA-ANFIS model has a high ability to predict the number of confirmed cases within ten days. Besides, FPASSA-ANFIS outperforms other forecasting models in terms of RMSE, MAE, MAPE, RMSRE, and R 2 . Furthermore, two datasets of weekly influenza confirmed cases in the USA and China were used to evaluate the proposed method, and the evaluation outcomes showed its good performance. According to the promising results obtained by the proposed FPASSA-ANFIS, it can be applied in different forecasting applications.

What do the authors propose?

MERS coronavirus: diagnostics, epidemiology and transmission https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4687373/ SHA: f6fcf1a99cbd073c5821d1c4ffa3f2c6daf8ae29 Authors: Mackay, Ian M.; Arden, Katherine E. Date: 2015-12-22 DOI: 10.1186/s12985-015-0439-5 License: cc-by Abstract: The first known cases of Middle East respiratory syndrome (MERS), associated with infection by a novel coronavirus (CoV), occurred in 2012 in Jordan but were reported retrospectively. The case first to be publicly reported was from Jeddah, in the Kingdom of Saudi Arabia (KSA). Since then, MERS-CoV sequences have been found in a bat and in many dromedary camels (DC). MERS-CoV is enzootic in DC across the Arabian Peninsula and in parts of Africa, causing mild upper respiratory tract illness in its camel reservoir and sporadic, but relatively rare human infections. Precisely how virus transmits to humans remains unknown but close and lengthy exposure appears to be a requirement. The KSA is the focal point of MERS, with the majority of human cases. In humans, MERS is mostly known as a lower respiratory tract (LRT) disease involving fever, cough, breathing difficulties and pneumonia that may progress to acute respiratory distress syndrome, multiorgan failure and death in 20 % to 40 % of those infected. However, MERS-CoV has also been detected in mild and influenza-like illnesses and in those with no signs or symptoms. Older males most obviously suffer severe disease and MERS patients often have comorbidities. Compared to severe acute respiratory syndrome (SARS), another sometimes- fatal zoonotic coronavirus disease that has since disappeared, MERS progresses more rapidly to respiratory failure and acute kidney injury (it also has an affinity for growth in kidney cells under laboratory conditions), is more frequently reported in patients with underlying disease and is more often fatal. Most human cases of MERS have been linked to lapses in infection prevention and control (IPC) in healthcare settings, with approximately 20 % of all virus detections reported among healthcare workers (HCWs) and higher exposures in those with occupations that bring them into close contact with camels. Sero-surveys have found widespread evidence of past infection in adult camels and limited past exposure among humans. Sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics have been available almost from the start of the emergence of MERS. While the basic virology of MERS-CoV has advanced over the past three years, understanding of the interplay between camel, environment, and human remains limited. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1186/s12985-015-0439-5) contains supplementary material, which is available to authorized users. Text: An email from Dr Ali Mohamed Zaki, an Egyptian virologist working at the Dr Soliman Fakeeh Hospital in Jeddah in the Kingdom of Saudi Arabia (KSA) announced the first culture of a new coronavirus to the world. The email was published on the website of the professional emerging diseases (ProMED) network on 20 th September 2012 [1] (Fig. 1) and described the first reported case, a 60 year old man from Bisha in the KSA. This information led to the rapid discovery of a second case of the virus, this time in an ill patient in the United Kingdom, who had been transferred from Qatar for care [2] . The new virus was initially called novel coronavirus (nCoV) and subsequentlty entitled the Middle East respiratoy syndrome coronavirus (MERS-CoV). As of 2 nd of September 2015, there have been 1,493 detections of viral RNA or virus-specific antibodies across 26 countries (Additional file 1: Figure S1 ) confirmed by the World Health Organization (WHO), with over a third of the positive people dying (at least 527, 35 %) [3] . Since that first report, a slow discovery process over the following two to three years revealed a virus that had infected over 90 % of adult dromedary camels (DC; Camelus dromedarius) in the KSA [4] , also DCs across the Arabian Peninsula and parts of Africa that are a source of DC imports for the KSA [5] . To date, MERS-CoV has not been detected in DCs tested in zoos or herds from other parts of the world [6] [7] [8] [9] . Occasionally, virus is transmitted from infected DCs to exposed humans. Subsequent transmission to other humans requires relatively close and prolonged exposure [10] . The first viral isolate was patented and concerns were raised that this would restrict access to both the virus and to viral diagnostics [11, 12] . However, sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics were quickly described and virus was made freely available subject to routine biosafety considerations [13] . Subsequent epidemiology and research has identified the cell receptor as exopeptidase dipeptidyl peptidase 4 (DPP4; also called CD26); that MERS-CoV has a broad tropism, replicating better in some cells lines and eliciting a more proinflammatory response than SARS-CoV; is widespread in DCs; has the potential to infect other animals and that MERS kills its human host more often than SARS did (20-40 % versus 9 % for SARS [14] ) [15] [16] [17] [18] [19] . In humans, overt disease was given the name Middle East respiratory syndrome, with the acronym MERS. From intermittent animal-to-human spill-over events, the MERS-CoV spreads sporadically among people, causing more severe disease among older adults, especially males, with pre-existing diseases. The spread of MERS-CoV among humans has often been associated with outbreaks in hospitals, with around 20 % of all cases to date involving healthcare workers (HCWs). Although DCs appear to suffer the equivalent of a 'common cold' from MERS-CoV infection, in humans, the virus can be a more serious and opportunistic pathogen associated with the death of up to 40 % of reported cases. It has yet to be established whether infections thought to have been acquired from an animal source produce a more severe outcome than those spread between humans [20] . Studies have established that the mean incubation period for MERS is five to six days, ranging from two to 16 days, with 13 to 14 days between when illness begins in one person and subsequently spreads to another [21] [22] [23] [24] . Among those with progressive illness, the median time to death is 11 to 13 days, ranging from five to 27 days [23, 24] . Fever and gastrointestinal symptoms may form a prodrome, after which symptoms decline, only to be followed by a more severe systemic and respiratory syndrome [25, 26] . The first WHO case definition [27] defined probable cases of MERS based on the presence of febrile illness, cough and requirement for hospitalization with suspicion of lower respiratory tract (LRT) involvement. It also included roles for contact with a probable or confirmed case or for travel or residence within the Arabian Peninsula. If strictly adhered to, only the severe syndrome would be subject to laboratory testing, which was the paradigm early on [21] . From July 2013, the revised WHO case definition included the importance of seeking out and understanding the role of asymptomatic cases and from June 2014, the WHO definition more clearly stated that a confirmed case included any person whose sample was RT-PCR positive for MERS-CoV, or who produced a seroconversion, irrespective of clinical signs and symptoms. [28] [29] [30] Apart from the WHO and the KSA Ministry of Health reports, asymptomatic or subclinical cases of MERS-CoV infection were documented in the scientific literature although not always as often as occurred early on [31, 32] . The KSA definition of a case became more strict on 13 th May 2014, relying on the presence of both clinical features and laboratory confirmation [33] . Testing of asymptomatic people was recommended against from December 2014 [34] , reinforced by a case definition released by the KSA Ministry of Health in June 2015 [35] . The KSA has been the source of 79 % of human cases. Severe MERS is notable for its impact among older men with comorbid diseases including diabetes mellitus, cirrhosis and various lung, renal and cardiac conditions [36] [37] [38] . Interestingly in June 2015, an outbreak in South Korea followed a similar distribution [39, 40] . Among laboratory confirmed cases, fever, cough and upper respiratory tract (URT) signs and symptoms usually occur first, followed within a week by progressive LRT distress and lymphopaenia [37] . Patients often present to a hospital with pneumonia, or worse, and secondary bacterial infections have been reported [37, 41] . Disease can progress to acute respiratory distress syndrome and multiorgan system failure [37] . MERS has reportedly killed approximately 35 % of all reported cases, 42 % of cases in the KSA, yet only 19 % of cases in South Korea, where mortality ranged from 7 % among younger age groups to 40 % among those aged 60 years and above [42] ; all may be inflated values with asymptomatic or mild infections sometimes not sought or not reported [34] . General supportive care is key to managing severe cases [43] . Children under the age of 14 years are rarely reported to be positive for MERS-CoV, comprising only 1.1 % (n = 16) of total reported cases. Between 1 st September 2012 and 2 nd December 2013, a study described the then tally of paediatric cases in the KSA, which stood at 11 (two to 16 years of age; median 13 years); nine were asymptomatic (72 %) and one infant died [44] . In Amman, Jordan, 1,005 samples from hospitalized children under the age of two years with fever and/or respiratory signs and symptoms were tested but none were positive for MERS-CoV RNA, despite being collected at a similar time to the first known outbreak of MERS-CoV in the neighbouring town of Al-Zarqa [45] . A second trimester stillbirth occurred in a pregnant woman during an acute respiratory illness and while not RT-rtPCR positive, the mother did subsequently develop antibodies to MERS-CoV, suggestive of recent infection [46] . Her exposure history to a MERS-CoV RT-rtPCR positive relative and an antibody-reactive husband, her incubation period and her symptom history met the WHO criteria for being a probable MERS-CoV case [46] . Diagnostic methods were published within days of the ProMED email announcing the first MERS case [47] , including several now gold standard in-house RT-rtPCR assays (Fig. 2 ) as well as virus culture in Vero and LLC-MK2 cells [18, 47, 48] . A colorectal adenocarcinoma (Caco-2) epithelial cell line has since been recommended for isolation of infections MERS-CoV [49] . We previously [18] .). Open reading frames are indicated as yellow rectangles bracketed by terminal untranslated regions (UTR; grey rectangles). FS-frame-shift. Predicted regions encompassing recombination break-points are indicated by orange pills. Created using Geneious v8.1 [211] and annotated using Adobe Illustrator. Beneath this is a schematic depicting the location of RT-PCR primers (blue arrows indicate direction) and oligoprobes (green rectangles) used in the earliest RT-rtPCR screening assays and conventional, semi-nested (three primers) RT-PCR confirmatory sequencing assays [47, 48] . Publication order is noted by first [27 th September 2012; red] and second [6 th December 2012; orange] coloured rectangles; both from Corman et al. [47, 48] Those assays recommended by the WHO are highlighted underneath by yellow dots [53] . The NSeq reverse primer has consistently contained one sequence mismatch with some MERS-CoV variants. An altered version of that from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] reviewed the broad tropism of MERS-CoV [5] . However, as is well described, cell culture is a slow, specialised and insensitive method [50] while PCR-based techniques are the preferred method for MERS-CoV detection. The first open reading frames (ORF 1a and 1b; Fig. 2 ) have become a key diagnostic and taxonomic target for CoV species identification. With less than 80 % identity between the amino acid sequence of MERS ORF 1ab and betacoronavirus relatives, Tylonycteris bat HKU4 and Pipistrellus bat HKU5, it can be concluded that it is a novel and distinct virus. MERS-CoV is predicted to encode ten open reading frames with 5' and 3' untranslated regions [51] . The structural proteins include the spike (S), envelope (E), membrane (M) and nucleocapsid (N) [52] . The products of ORF1a and ORF1b are predicted to encode nonstructural proteins. The majority of specimen testing to date has employed validated RT-rtPCR assays shown to be sensitive and specific [47, 48, 53] . The RealStar® kit uses these WHOrecommended assays [54] . The target sequences of these screening assays have not changed among genomes examined until at least mid-2015 (IMM observation). Other RT-rtPCR assays have been developed and validated for use as laboratory-based diagnostic tools [55] [56] [57] . Additionally, loop-mediated [58, 59] or recombinase polymerase [60] isothermal assays have been designed for field deployment. The detection of MERS-CoV antigen has not been common to date but the combination of short turnaround time from test to result, high throughput and identification of viral proteins makes this an attractive option. Detection of viral proteins rather than viral RNA indicates the likely presence of infectious virus. The first rapid immunochromatographic tool described could detect recombinant MERS-CoV nucleocapsid protein from DC nasal swabs with 94 % sensitivity and 100 % specificity compared to RT-rtPCR [61] . A different approach used a monoclonal antibody-based capture ELISA targeting the MERS-CoV nucleocapsid protein with a sensitivity of 10 3 TCID 50 and 100 % specificity [62] . Demonstration of a seroconversion to a MERS-CoV infection meets the current WHO definition of a case so optimized and thoroughly validated sero-assays employed alongside good clinical histories are useful to both identify prior MERS-CoV infection and help support transmission studies. Because serology testing is, by its nature, retrospective, it is usual to detect a viral footprint, in the form of antibodies, in the absence of any signs or symptoms of disease and often in the absence of any viral RNA [63] . Strategic, widespread sero-surveys of humans using samples collected after 2012 are infrequent. Much of the Arabian Peninsula and all of the Horn of Africa lack baseline data describing the proportion of the community who may have been infected by a MERS-CoV. However, sero-surveys have had widespread use in elucidating the role of DCs as a transmission source for MERS-CoV. Because of the identity shared between DC and human MERS-CoV (see Molecular epidemiology: using genomes to understand outbreaks), serological assays for DC sero-surveys should be transferrable to human screening with minimal re-configuration. Also, no diagnostically relevant variation in neutralization activity have been found from among a range of circulating tested MERS-CoV isolates and sera, so whole virus or specific protein-based sero-assays should perform equivalently in detecting serological responses to the single MERS-CoV serotype [49] . The development of robust serological assays requires reliable panels of wellcharacterized animal or human sera, including those positive for antibodies specific to MERS-CoV, as well as to likely sources of cross-reaction [64] . Obtaining these materials was problematic and slowed the development and commercialization of antibody detection assays for human testing [64] . A number of commercial ELISA kits, immunofluorescent assays (IFA) kits, recombinant proteins and monoclonal antibodies have been released [31, [65] [66] [67] [68] . Initially, conventional IFAs were used for human sero-surveys. These relied on MERS-CoV-infected cell culture as an antigen source, detecting the presence of human anti-MERS-CoV IgG, IgM or neutralizing antibodies in human samples [18, 48, 69] . No sign of MERS-CoV antibodies was found among 2,400 sera from patients visiting Hospital in Jeddah, from 2010 through 2012, prior to the description of MERS-CoV [18] . Nor did IFA methods detect any sign of prior MERS-CoV infection among a small sample of 130 healthy blood donors from another Hospital in Jeddah (collected between Jan and Dec 2012) [70] . Of 226 slaughterhouse workers, only eight (3.5 %) were positive by IFA, and those sera could not be confirmed by virus neutralization (NT) test. The study indicated that HCoV-HKU1 was a likely source of crossreactive antigen in the whole virus IFA [70] . Whole virus MERS-CoV IFA also suffered from some cross-reactivity with convalescent SARS patient sera and this could not be resolved by an NT test which was also cross-reactive [71] . IFA using recombinant proteins instead of whole-virus IFA, has been shown to be a more specific tool [31] . Since asymptomatic zoonoses have been posited [72] , an absence of antibodies to MERS-CoV among some humans who have regular and close contact with camels may reflect the rarity of actively infected animals at butcheries, a limited transmission risk associated with slaughtering DCs [70] , a pre-existing cross-protective immune status or some other factor(s) resulting in a low risk of disease and concurrent seroconversion developing after exposure in this group. IFA using recombinant proteins instead. Some sero-assays have bypassed the risks of working with infectious virus by creating transfected cells expressing recombinant portions of the MERS-CoV nucleocapsid and spike proteins [48, 73] , or using a recombinant lentivirus expressing MERS-CoV spike protein and luciferase [74, 75] . A pseudo particle neutralization (ppNT) assay has seen widespread used in animal studies and was at least as sensitive as the traditional microneutralization (MNT) test. [10, 74, [76] [77] [78] ] Studies using small sample numbers and ppNT found no evidence of MERS-CoV neutralizing antibody in sera from 158 children with LRT infections between May 2010 and May 2011, 110 sera from 19 to 52 year old male blood donors and 300 selfidentified animal workers from the Jazan Region of the KSA during 2012 [79, 80] . Similarly, a study of four herdsmen in contact with an infected DC herd in Al-Ahsa, eight people who had intermittent contact with the herd, 30 veterinary surgeons and support staff who were not exposed to the herd, three unprotected abattoir workers in Al-Ahsa and 146 controls who were not exposed to DCs in any professional role, found none with serological evidence of past MERS-CoV infection using the ppNT assay [10] . A delay in the neutralizing antibody response to MERS-CoV infection was associated with increased disease severity in South Korea cases with most responses detectable by week three of illness while others, even though disease was severe, did not respond for four or more weeks [81] . The implications for our ability to detect any response in mild or asymptomatic cases was not explored but may be a signifcant factor in understanding exposure in the wider community. A Jordanian outbreak of acute LRT disease in a hospital in 2012 was retrospectively found to be associated with MERS-CoV infection, initially using RT-rtPCR, but subsequently, and on a larger scale, through positivity by ELISA and IFA or MNT test. [46, 82, 83] This outbreak predated the first case of MERS in the KSA. The ELISA used a recombinant nucleocapsid protein from the group 2 betacoronavirus bat-CoV HKU5 to identify antibodies against the equivalent crossreactive MERS-CoV protein [71] . It was validated using 545 sera collected from people with prior HCoV-OC43, HCoV-229E, SARS-CoV, HCoV-NL63, HRV, HMPV or influenza A(H1N1) infections but was reportedly less specific than the recombinant IFA discussed above. It was still considered an applicable tool for screening large sample numbers [82] . A protein microarray expressing the S1 protein subunit has also been validated and widely used for DC testing [5, 84] . Detection of MERS-CoV infection using ELISA or S1 subunit protein microarray [84] is usually followed by confirmatory IFA and/ or a plaque-reduction neutralization (PRNT) [69, 70, 85] or MNT test. [74, 85, 86] This confirmatory process aims toensure the antibodies detected are able to specifically neutralize the intended virus and are not more broadly reactive to other coronaviruses found in DCs (bovine CoV, BCoV) or humans (HCoV-OC43, HCoV-229E, HCoV-NL63, HCoV-HKU1, SARS-CoV). In the largest study of human sera, a tiered diagnostic process assigned both recombinant IFA and recombinant ELISA positive sera to 'stage 1' seropositivity. A stage 2 seropositive result additionally required a suitably titred PRNT result [87] . The study found 15 sera collected in 2012 to 2013 from 10,009 (0.2 %) people in 13 KSA provinces contained MERS-CoV antibodies, but significantly higher proportions in occurred in camel shepherds (two of 87; 2.3 %) and slaughterhouse workers (five of 140; 3.6 %) [87] . Contemporary surveys are needed. MERS-CoV does not appear to be easily transmitted from DCs to humans, or perhaps it is [72] , but generally does not trigger a detectable immune response if only mild disease or asymptomatic infection results. Serology assays are in need of further validation in this area so care is required when moving newly developed diagnostic serology algorithms from a research setting to one that informs public health decisions. This was reinforced when a false positive US case, purported to have been infected after a handshake and two face-to-face meetings, did not withstand further confirmatory analysis using a more specific, NT assay and was subsequently retracted [88, 89] . The WHO recommends sampling from the LRT for MERS-CoV RT-rtPCR testing, especially when sample collection is delayed by a week or more after onset of symptoms. [53] LRT samples are also best for attempting isolation of infectious virus, although the success of culture is reduced when disease persists [49] . Recommended sample types include bronchoalveolar lavage (BAL), tracheal/tracheobronchial aspirate, pleural fluid and sputum [53, 90] . Fresh samples yield better diagnostic results than refrigerated material [69] and if delays in testing of ≥72 h are likely, samples (except for blood) should be frozen at −70°C [90] . If available, lung biopsy or autopsy tissues can also be tested [53] . The URT is a less invasive and more convenient sampling site however, and an oropharyngeal and throat swab or a nasopharyngeal aspirate/wash are recommended when URT sampling is to be conducted [90] . Paired sera, collected two to three weeks apart are preferable for serological testing while a single sample is suggested to be sufficient if collected two weeks after onset of disease or a single serum collected during the first 10-12 days if conducting RT-rtPCR [53, 90] . Human urine and stool have been found to contain MERS-CoV RNA 12 to 26 days after symptom onset [25, 69, 91] and are listed as samples that should be considered [53, 90] . In two cases that arrived in the Netherlands, urine was RT-rtPCR negative but faeces was weakly positive and sera were RT-rtPCR positive for five days or more [25] . The finding of MERS-CoV viral RNA in serum provides an avenue for retrospective PCR-based studies if respiratory samples are unavailable [83] . RNAaemia may also correlate with disease severity; signs of virus were cleared from the serum of a recovered patient, yet lingered until the death of another [92] . Clinically suspected MERS cases may return negative results by RT-rtPCR. Data have shown one or more negative URT samples may be contradicted by further URT sampling or the use of LRT samples, which is preferred [2, 43, 93] . Higher viral loads occur in the LRT compared to the URT. [22, 69, 88, 94] This fits with the observation that the majority of disease symptoms are reported to manifest as systemic and LRT disease [21] . However, on occasion, even LRT specimens from MERS cases may initially be negative, only to later become positive by RT-PCR [95] . This may be due to poor sampling when a cough is absent or non-productive or because the viral load is low [95] . Despite this both the largest human MERS-CoV studies [32, [96] [97] [98] and smaller ones [22, 25, 99] , use samples from the URT. It is then noteworthy that one study reported an association between higher loads in the URT and worse clinical outcome including intensive care and death [94] . At writing, no human data exist to define whether the virus replicates solely or preferentially in the LRT or URT, or replicates in other human tissues in vivo although MERS-CoV RNA has been detected from both the URT and LRT in a macaque monkey model [100] .The distribution of DPP4 in the human upper airways is also not well described. Individual human case studies report long periods of viral shedding, sometimes intermittently and not necessarily linked to the presence of disease symptoms. [25, 69, 99, 101] In one instance, a HCW shed viral RNA for 42 days in the absence of disease [99] . It is an area of high priority to better understand whether such cases are able to infect others. Over three quarters of MERS cases shed viral RNA in their LRT specimens (tracheal aspirates and sputum) for at least 30 days, while only 30 % of contacts were still shedding RNA in their URT specimens [91, 102] . In the only study to examine the effect of sample type on molecular analysis, 64 nasopharyngeal aspirates (NPA; an URT sample), 30 tracheal aspirates, 13 sputa and three BAL were examined. The tracheal aspirates and BAL returned the highest viral load values followed by NPA and sputum. Unsurprisingly, higher viral loads generally paralleled whole genome sequencing and culture success and, in NPA testing, were significantly correlated with severe disease and death [49, 94, 103] . This study demonstrated the importance of LRT sampling for whole genome sequencing. When tested, samples positive for MERS-CoV are often negative for other pathogens [2, 25, 93, 104] . However, many studies make no mention of additional testing for endemic human respiratory viruses [21, 23, 73, 105] . When viruses are sought, they have included human herpesvirus (HHV), rhinoviruses (HRV), enteroviruses (EV), respiratory syncytial virus (RSV), parainfluenzavirus types 1, 2 and 3 (PIVs),influenzaviruses (IFVs), endemic HCoVs, adenoviruses (AdVs) metapneumovirus (MPV) and influenza A\H1N1 virus; co-detections with MERS-CoV have been found on occasion [2, 22, 37, 69, 97] . Bacterial testing is sometimes included (for example, for Legionella and Pneumococcus) but the impact of bacterial co-presence is also unclear [22, [104] [105] [106] . Further testing of the LRT sample from the first MERS case used IFA to screen for some viruses (negative for IFV, PIVs, RSV and AdVs) and RT-PCR for others (negative for AdV, EVs, MPV and HHVs) [18] . RT-PCR also detected MERS-CoV. The WHO strongly recommends testing for other respiratory pathogens [53] but with this recommendation often discounted, there are limited data to address the occurrence and impact of co-infections or alternative viral diagnoses among both MERS cases and their contacts. Little is known of other causes of MERS-like pneumonia in the KSA or of the general burden of disease due to the known classical respiratory viruses. Testing of adult pilgrims performing the Hajj in 2012 to 2014 has not detected any MERS-CoV. In 2012, nasal swabs from 154 pilgrims collected prior to leaving for or departing from the KSA were tested [47] . In 2013, testing was significantly scaled up with 5,235 nasopharyngeal swabs from 3,210 incoming pilgrims and 2,025 swabs from outgoing pilgrims tested [98] . It should be noted that most pilgrims arrived from MERS-free countries. A further 114 swabs were taken from pilgrims with influenza-like illness [96, 107] . In earlier Hajj gatherings, it was found that influenza viruses circulated widely, whilst other viruses, often rhinoviruses, circulated more selectively, interpreted as indicating their importation along with foreign pilgrims. [107] [108] [109] Over time, increased influenza vaccination has been credited for a fall in the prevalence of influenza like illnesses among Hajj pilgrims. [110] A LRT sample is often not collected for these studies [98, 107, 109] , so false negative findings are a possibility although little is known about the initial site of MERS-CoV infection and replication; it may have been assumed it was the LRT because disease was first noticed there but the URT may be the site of the earliest replication. In Jeddah between March and July 2014 (hereafter called the Jeddah-2014 outbreak; Fig. 3 ), there was a rapid increase in MERS cases, accompanied by intense screening; approximately 5,000 samples from in and around the region were tested in a month yielding around 140 MERS-CoV detections (~3 % prevalence) [111] . Among 5,065 individuals sampled and tested across the KSA between October 2012 and September 2013,108 (2.1 %) detections were made in a hospital-centric population which included hospitalized cases (n = 2,908; 57.4 %), their families (n = 462; 9.1 %) and associated HCWs (n = 1,695; 33.5 %) [32] . Among the detections, 19 (17.8 %) were HCWs and 10 (9.3 %) were family contacts [32] . The 2-3 % prevalence of active MERS-CoV infections is not dissimilar to the hospital-based prevalence of other human CoVs. [112] However, the proportion of deaths among those infected with MERS-CoV is much higher than that known for the HCoVs NL63, HKU1, 229E or OC43 in other countries, and even above that for SARS-CoV; it is not a virus that could reasonably be described as a "storm in a teacup". It is the low transmission rate that has prevented worldwide spread, despite many "opportunities". Very early in the MERS outbreak, some animals were highly regarded as either the reservoir or intermediate host(s) of MERS-CoV with three of the first five cases having contact with DCs [73, 113, 114] . Today, animal MERS-CoV infections must be reported to the world organization for animal health as an emerging disease [115] . A summary of the first MERS cases reported by the WHO defined animal contact with humans as being direct and within 10 days prior to symptom onset [20] . This definition made no specific allowance for acquisition from DCs through a droplet-based route, which is very likely route for acquisition of a virus that initially and predominantly causes respiratory disease [23] . Camels are known to produce high levels of MERS-CoV RNA in their URT and lungs [116] . Providing support for a droplet transmission route and perhaps indicating the presence of RNA in smaller, drier droplet nuclei, MERS-CoV RNA was identified in a high volume air sample collected from a barn housing an infected DC [117] . The precise source from which humans acquire MERS-CoV remains poorly studied but it seems likely that animal and human behavioural factors may play roles (Fig. 3) [118] . These factors may prove important for human cases who do not describe any DC contact [119] nor any contact with a confirmed case. Whether the WHO definition of animal contact is sufficient to identify exposure to this respiratory virus remains unclear. Wording focuses on consumption of DC products but does not specifically ascribe risk to a droplet route for acquisition of MERS-CoV from DC [120] . Some MERS patients are listed in WHO disease notices as being in proximity to DCs or farms, but the individuals have not described coming into contact with the animals. No alternative path for acquiring infection is reported in many of these instances. What constitutes a definition of "contact" during these interviews has been defined for one study [72] . Despite this lack of clarity, the WHO consider that evidence linking MERS-CoV transmission between DCs to humans is irrefutable (Fig. 4) [120] . The possibility that bats were an animal host of MERS-CoV was initially widely discussed because of the existing diversity of coronaviruses known to reside among them [121] [122] [123] [124] . Conclusive evidence supporting bats as a source for human infections by MERS-CoV has yet to be found, but bats do appear to host ancestral representatives [53, 125] . However, these are not variants of the same virus nor always within the same phylogenetic lineage as MERS-CoV; they are each a genetically distinct virus. Bat-to-human infection by MERS-CoV is a purely speculative event. The only piece of MERS-CoV-specific evidence pointing to bats originates from amplification of a 190 nt fragment of the RNAdependent RNA polymerase gene of the MERS-CoV genome, identified in a faecal pellet from an insectivorous Emballonuridae bat, Taphozous perforatus found in Bisha, the KSA [121] . While very short, the sequence of the fragment defined it as a diagnostic discovery. Subsequently a link to DCs was reported [85] and that link has matured into a verified association [38, 126] (Fig. 4) . (See figure on previous page.) Fig. 3 Monthly detections of MERS-CoV (blue bars) and of cases who died (red bars) with some dates of interest marked for 2012 to 4 th September 2015. An approximation of when DC calving season [128] and when recently born DCs are weaned is indicated. Spring (green) and summer (orange) in the Arabian Peninsula are also shaded. Note the left-hand y-axis scale for 2014 and 2015 which is greater than for 2012/13. Sources of these public data include the WHO, Ministries of Health and FluTrackers [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] . Modified and reprinted from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] DCs, which make up 95 % of all camels, have a central presence in the Arabian Peninsula where human-DC contact ranges from little to close [119] . Contact may be commonplace and could occur in variety of ways (Fig. 4a) . There are several large well-attended festivals, races, sales and parades which feature DCs and DCs are also kept and bred close to populated areas in the KSA [127, 128] . DC milk and meat are widely consumed and the older DC is an animal of ritual significance after the Hajj pilgrimage [129] . However, MERS-CoV infection frequency is reportedly much lower than is the widespread and frequent habit of eating, drinking and preparing DC products. Daily ingestion of fresh unpasteurized DC milk is common among the desert Bedouin and many others in the KSA. DC urine is also consumed or used for supposed health benefits. Despite camel butchery being a local occupation, neither butchers nor other at-risk groups are identifiable among MERS cases; this may simply be a reporting issue rather than an unexplainable absence of MERS. A small case-control study published in 2015 identified direct DC contact, and not ingestion of products, to be associated with onset of MERS [38] . The first sero-survey of livestock living in the Middle East region was conducted during 2012-2013 [85] . DCs were sampled from a mostly Canary Island-born herd and from Omani DCs (originally imported from the Horn of Africa) [85] . A neutralising antibody assay found only 10 % of strongly seropositive Canary Island [5] . b Camel-to-human infections appear to be infrequent, while human-to-human spread of infection is regularly facilitated by poor IPC in healthcare settings where transmission is amplified, accounting for the bulk of cases. There are human MERS cases that do not fall into either category of source and it is unclear if these acquired infection through some entirely separate route, or from cases that escaped diagnosis. c Hypothetical ways in which subclinical (when infection may not meet a previously defined clinical threshold of signs and/or symptoms) or asymptomatic (no obvious signs or measured, noticed or recalled symptoms of illness) MERS-CoV infection may be implicated in transmission DC sera could neutralise MERS-CoV while all Omani DC sera had high levels of specific MERS-CoV neutralizing antibody [85] . This indicated that DCs had in the past been infected by MERS-CoV, or a very similar virus. Since this study, a host of peer-reviewed reports have looked at both DCs and other animals, and the possibility that they may host MERS-CoV infection. Seropositive DCs have been found throughout the Arabian Peninsula including Oman, the KSA, Qatar, Jordan, the United Arab Emirates (UAE), Kuwait as well as Sudan, Somalia, Egypt, Tunisia, Nigeria, Kenya and Ethiopia in Africa and the Canary Islands [85, [130] [131] [132] [133] [134] . Other animals tested include sheep, cows, pigs, horses, donkeys, mules, birds, water buffalo, goats, Bactrian camels, llamas and guanaco (south American camelids) but none had detectable neutralising antibody against MERS-CoV [4, 74, 78, 85, 86, 135, 136] . No virology or serology studies of human samples from areas in Africa where there are camels with a history of MERS-CoV have been reported to date. However,an absence of unexplained pneumonia that may be attributable to MERS-CoV infection may not signal the absence of virus among humans in each country but simply reflect a lack of expensive epidemiology studies conducted by resource-poor countries. It is thus unclear whether MERS-CoV, or an antigenically related CoV, is an unrecognized pathogen in these regions, perhaps circulating for even longer than it has been known in the Arabian Peninsula [133] . MERS-CoV RNA has also been detected in DC samples, and recovery of infectious virus has also been achieved from DC samples [4, 77, 117, 132, [137] [138] [139] [140] [141] . From some of these, full or majority length genomes of MERS-CoV have been sequenced [77, 137, 138] . DC versions of MERS-CoV were found to be as similar to each other, as were variants detected from different humans over time and across distance. Antibody screening assays have also detected crossreactive antibodies in sera. These were identified as such by screening sera against similar viruses, for example BCoV or HCoV-OC43 (as an antigenic facsimile for BCoV). It is possible that other MERS-CoV-like viruses also reside within DCs, but this does not detract from the definitive finding of MERS-CoV genetic sequences in both DCs and humans [117, 142, 143] . Screening studies have shown that juvenile DCs are more often positive for virus or viral RNA while older DCs are more likely to be seropositive and RNA or virus negative [76, 77, 144] . In adult DCs, MERS-CoV RNA has been detected among animals with pre-existing antibody, suggesting re-infection is possible [77, 144] . Viral loads among positive DCs can be very high [4, 76, 77, 139, 144] and DCs have been found positive both when ill with URT respiratory signs [77, 117, 142, 145] or when apparently healthy [137] . These findings indicate DCs host natural MERS-CoV infections. Furthermore, stored DC sera have revealed signs of MERS-CoV in DCs which date back over three decades (the earliest collected in 1983) [4, 133, 135] . Older sera have not been tested and so precisely how long DCs have been afflicted by MERS-CoV, whether the virus is enzootic among them, introduced to them decades or centuries ago from bats in Africa or the Arabian Peninsula, or they are the subject of regular but short-lived viral incursions from an as yet unknown host, cannot be answered. Researchers sought to determine a direction for infection; were DCs transmitting virus to humans or were humans infecting DCs? At a Qatari site, a farm owner and his employee became ill in mid-October 2013 and tested positive for MERS-CoV RNA in a sputum and throat swab sample, respectively. RT-rtPCRs found MERS-CoV RNA in 11 of 14 positive DC nasal swabs at the farm; six (43 %) positive by two or more assays [138] . The results indicated a recent outbreak had occurred in this herd; the first indication of MERS-CoV RNA found within DCs with a temporal association to human infections. Three positive DC samples were confirmed by sequencing a 358 nt portion of the spike gene; these sequences were identical to each other, again with close homology to other human and DC MERS-CoV sequences [138] . The DCs and human contacts yielded ORF1a and ORF4b sequences differing by only a single nucleotide each, clustering closely with the Hafr-Al-Batin_1_2013 variant [138] . Subsequent case studies found evidence of a concurrent human and DC infection and the direction of that infection was inferred to be from the ill DCs and to their human owners [117, 142, 146] . Partial genome sequences indicated that a human and a MERS-CoV RT-rtPCR positive DC had been infected by a variant of the same virus, harbouring the same distinct pattern of nucleotide polymorphisms. [142] All nine DC in the owner's herd, serially sampled, reacted in a recombinant S1 antigen ELISA, with the two animals that had been RT-rtPCR positive showing a small, verifiable rise in antibody titre [142] . A rise in titre theoretically begins 10 to 21 days after DC infection [142] . The authors suggested that the rise in titre in DC sera which occurred alongside a declining RNA load, while the patient was actively ill and hospitalized, indicated that the DCs were infected first followed by the owner [117, 142] . BCoV antibodies were also present, and rising in one of the two RT-rtPCR positive animals but no animal's antibodies could neutralise BCoV infection [142] . Camel calving season occurs in the winter months (between late October and late February; Fig. 3 ) and this may be a time when there is increased risk to humans of spill-over due to new infections among naïve DC populations [128] . What role maternal camel antibody might play in delaying infection of calves remains unknown [128, 142] . Juvenile DCs appear to host active infection more often than adult DCs and thus the sacrificial slaughter of DCs, which must be five years of age or older (termed a thane), may not be accompanied by significant risk of exposure to infection. In contrast to earlier results, slaughterhouse workers who kill both younger and older DCs, may be an occupational group with significantly higher incidence of seropositivity to MERS-CoV when animals have active MERS-CoV infections [129, 139, [147] [148] [149] . Expanded virological investigations of African DCs may lead to more seropositive animals and geographic areas in which humans may be at risk. It is possible that there are areas where humans already harbour MERS-CoV infections that have not been identified because of an absence of laboratory surveillance. Virological investigations of bats may lead to findings of ancestral viruses and viral 'missing links' and identifying any other animal sources of zoonotic spread is important to inform options for reducing human exposures [56, 76] . Infectious MERS-CoV added to DC, goat or cow milk and stored at 4°C could be recovered at least 72 h later and, if stored at 22°C, recovery was possible for up to 48 h [150] . MERS-CoV titre decreased somewhat when recovered from milk at 22°C but pasteurization completely ablated MERS-CoV infectivity [150] . In a subsequent study, MERS-CoV RNA was identified in the milk, nasal secretion and faeces of DCs from Qatar [151] . A single study has examined the ability of MERS-CoV to survive in the environment [150] . Plastic or steel surfaces were inoculated with 10 6 TCID 50 of MERS-CoV at different temperature and relative humidity (RH) and virus recovery was attempted in cell culture. At high ambient temperature (30°C) and low RH (30 %) MERS-CoV remained viable for 24 h [150] . By comparison, a well known and efficently transmitted respiratory virus, influenza A virus, could not be recovered in culture beyond four hours under any conditions [150] . Aerosol experiments found MERS-CoV viability only decreased 7 % at low RH at 20°C. In comparison, influenza A virus decreased by 95 % [150] . MERS-CoV survival is inferior to that previously demonstrated for SARS-CoV [152] . For context, pathogenic bacteria can remain viable and airborne for 45 min in a coughed aerosol and can spread 4 m. MERS-CoV's ability to remain viable over long time periods gives it the capacity to thoroughly contaminate a room's surfaces when occupied by an infected and symptomatic patient [153] . Whether MERS-CoV can remain adrift and infectious for extended periods (truly airborne) remains unknown. Such findings expand our understanding of the possibilities for droplets to transmit respiratory viruses in many settings, including hospital waiting rooms, emergency departments, treatment rooms, open intensive care facilities and private patient rooms. The nature and quality of air exchange, circulation and filtration are important variables in risk measurement and reduction as is the use of negative pressure rooms to contain known cases. Droplet spread between humans is considered the mechanism of human-to-human transmission and the need for droplet precautions was emphasized after the Al-Ahsa hospital, the KSA and the South Korean outbreaks [21, 23, 154, 155] . By extrapolation, aerosol-generating events involving DCs (urination, defecation, and preparation and consumption of DC products) should be factored into risk measurement and reduction efforts and messaged using appropriate context. The provision of evidence supporting the best formulation of personal protective equipment to be worn by HCWs who receive, manage or conduct procedures on infectious cases remains a priority. MERS-CoV was found and characterized because of its apparent association with severe, and therefore more obvious, illness in humans; we were the canaries in the coal mine. Sero-assays and prospective cohort studies have yet to determine the extent to which milder or asymptomatic cases contribute to MERS-CoV transmission chains. However, transmission of MERS-CoV is defined as sporadic (not sustained), intra-familial, often healthcare associated, inefficient and requiring close and prolonged contact [22, 31, 63, 93, 97, 102, 156] In a household study, 14 of 280 (5 %) contacts of 26 MERS-CoV positive index patients were RNA or antibody positive; the rate of general transmission, even in outbreaks is around 3 % [31] . It seems that the majority of human cases of MERS-CoV, even when numbers appear to increase suddenly, do not readily transmit to more than one other human so to date, the localized epidemic of MERS-CoV has not been self-sustaining [157] [158] [159] [160] [161] . That is to say, the basic reproduction number (R 0 ) -the average number of infections caused by one infected individual in a fully susceptible populationhas been close to one throughout various clusters and outbreaks. If R 0 was greater than 1, a sustained increase in case numbers would be expected. Some R o calculations may be affected by incomplete case contact tracing, limited community testing and how a case is defined. That MERS has had a constant presence in the Arabian Peninsula since 2012 is due to ongoing, sporadic spill-over events from DCs amplified by poorly controlled hospital outbreaks. The first known MERS human-to-human transmission event was one characterized by acute LRT disease in a healthcare setting in Jordan. In stark contrast, a sero-survey of HCW who were sometimes in close and prolonged contact with the first, fatal MERS-CoV case in 2012 [162] , found none of the HCW had seroconverted four months later, despite an absence of eye protection and variable compliance with required PPE standards [162] . Early on in the MERS story, samples for testing were mostly collected from patients with severe illness and not those with milder acute respiratory tract infections. Contacts of confirmed MERS cases were often observed for clinical illness, but not tested. These omissions may have confounded our understanding of MERS-CoV transmission and biased early data towards higher numbers of seriously ill and hospitalized patients, inflating the apparent proportion of fatal cases. Case-control studies were not a focus. As testing paradigms changed and contacts were increasingly tested, more asymptomatic and mild infections were recognized [163] . A rise in the cases termed asymptomatic (which enlarge the denominator for calculations of the proportion of fatal cases, defined in [164] ) resulted in a drop in the proportion of fatal cases during the Jeddah-2014 outbreak. Historically, such rises are consistent with changing definitions and laboratory responses and clinical management of a newly discovered virus infection that was first noted only among the severely ill. Upon follow-up, over three-quarters of such MERS-CoV RNA positive people did recall having one or more symptoms at the time, despite being reported as asymptomatic [165] raising some question over the reliability of other reported data. The proportion of fatal MERS cases within the KSA compared to outside the KSA, as well as the age, and sex distribution change in different ways when comparing MERS outbreaks. Approximately 43 % of MERS cases (549 of 1277) in the KSA were fatal betwen 2012 and December 2015 while 21 % (72 of 330) died among those occurring outside of the KSA. The total number of male cases always outnumber females and the proportion of male deaths is always greater than the proportion of females who die. However the proportion of male deaths from total males with MERS is a similar figure to that for females. In the KSA, there is a greater proportion of younger males among cases and deaths than were observed from the 2015 South Korean or the Jeddah-2014 outbreaks (Additional file 2: Figure S2 ). Why these aspects have differed may be due to differences in the time to presentation and diagnosis, the nature and quality of supportive care, the way a person became infected (habits, exposure to a human or zoonotic source, viral load, route of infection) or the extent to which different populations are burdened by underlying diseases [40] . As a group, HCWs comprised 16 % of MERS cases in the KSA and South Korea. It is apparent that the weekly proportion of infected HCWs increases alongside each steep rise in overall detections (Fig. 5) . In May 2013, the WHO published guidelines for IPC during care of probable or confirmed cases of MERS-CoV infection in a healthcare setting [166] . This is explainable because to date, each case rise has been intimately associated with healthcare-facility related outbreaks [118] . These rises in MERS-CoV detections can decrease the average age during each event because HCWs are usually younger than inpatients with MERS. Healthcare facilities have been a regular target for suggested improvements aimed at improving infection prevention and control (IPC) procedures [115, 118] . Most of the analysis of MERS-CoV genetics has been performed using high throughput or "deep" sequencing methods for complete genome deduction [167] [168] [169] . MERS-CoV was the first subject of such widespread use of deep sequencing to study an emerging viral outbreak with global reach. The technique can produce genomic [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] length coverage in a single experiment with highly repetitious measurement of each nucleotide position [52, 140] . Despite assays having been published early on, subgenomic sequencing, once the mainstay of viral outbreak studies, has less often been published during MERS-CoV characterization [48] . As more genomes from both humans and DCs have been characterized, two clades have become apparent; A and B (Fig. 6) . Clade A contains only human-derived MERS-CoV genomes from Jordan, while Clade B comprises the majority of human and camel genomes deduced thus far [168] . Two studies during 2015, one looking at Jeddah-2014 MERS-CoV variants and another looking at a variant exported from South Korea to China, have now identified signs of genetic recombination among MERS-CoV variants. While human and camel whole genome sequences have retained >99 % identity with each other, members of genetically distinct lineages can and do swap genetic material when suitable conditions and coinfections co-occur [170] [171] [172] . Shared identity implies that the major source for human acquisition is the DC, rather than another animal, although more testing of other animal species is needed to confirm that conclusion. Over a month, a DC virus sequenced on different occasions did not change at all indicating a degree of genomic stability in its host, supporting that DCs are the natural, rather than intermediate, host for the MERS-CoV we know today [77] . To date, recombination has been localised to breakpoints near the boundary between ORF1a and ORF1b regions, within the spike gene [170] and in the ORF1b region (Fig. 2) [172] . It is not unexpected that recombination should occur since it is well known among other CoVs [124] and because the majority of MERS-CoV whole genomes collected from samples spanning three years (2012-2015) and from humans, camels and different countries have shown close genetic identity to each other, with just enough subtle variation to support outbreak investigations so long as whole genome sequencing is applied [52, 77, 135, 138, 168, [173] [174] [175] . Changes in genome sequence may herald alterations to virus transmissibility, replication, persistence, lethality or response to future drugs. If we have prior knowledge of the impact of genetic changes because of thorough characterization studies, we can closely Fig. 6 The genetic relationship between MERS-CoV nucleotide sequences (downloaded from GenBank using the listed accession numbers and from virological.org [212] ). This neighbour joining tree was created in MEGA v6 using an alignment of human and DCderived MERS-CoV sequences (Geneious v8.1 [211] ). Clades are indicated next to dark (Clade A) or pale (Clade B) blue vertical bars. Camel icons denote genomes from DCs. Healthcare or community outbreaks are boxed and labelled using previously described schemes [212, 213] monitor the genomic regions and better understand any changes in transmission or disease patterns as they occur. Genetic mutations noted during the largest of human outbreaks, Jeddah-2014, did not impart any major replicative or immunomodulatory changes when compared to earlier viral variants in vitro [156, 176] . However, we understand very little of the phenotypic outcomes that result from subtle genetic change in MERS-CoV genomes. To date no clinical relevance or obvious in vivo changes to viral replication, shedding or transmission has been reported or attributed to mutations or to new recombinant viruses [156] . But vigilance and larger, more contemporary and in vivo studies are needed. Genome sequence located to a distinct clade were identified from an Egyptian DC that was probably imported from Sudan. This does not fit into either of the current clades [125, 168, 177] . A virus sequenced from a Neoromicia capensis bat was more closely related to MERS-CoV than other large bat-derived sequences had been to that point, but the genome of a variant of a MERS-CoV has yet to be discovered and deduced from any bat [125] . Analyses of MERS-CoV genomes have shown that most single nucleotide differences among variants were located in the last third of the genome (Fig. 2) , which encodes the spike protein and accessory proteins [168] . At least nine MERS-CoV genomes contained amino acid substitutions in the receptor binding domain (RBD) of the spike protein and codons 158 (N-terminal region), 460 (RBD), 1020 (in heptad repeat 1), 1202 and 1208 bear investigation as markers of adaptive change [140, 169] . The spike protein had not changed in the recombinant MERS-CoV genome identified in China in 2015 but was reported to have varied at a higher rate than that for complete MERS-CoV genomes, among South Korean variants [172, 178] . This highlights that subgenomic regions may not always contain enough genetic diversity to prove useful for differentiating viral variants. Despite this, one assay amplifying a 615 nucleotide fragment of the spike S2 domain gene for Sanger sequencing agreed with the results generated by the sequencing of a some full genomes and was useful to define additional sequence groupings [177] . Genomic sequence can also be used to define the geographic boundaries of a cluster or outbreak and monitor its progress, based on the similarity of the variants found among infected humans and animals when occurring together, or between different sites and times (Fig. 6 ) [169] . This approach was employed when defining the geographically constrained MERS hospital outbreak in Al-Ahsa, which occurred between 1 st April and 23 rd May 2013, as well as clusters in Buraidah and a community outbreak in Hafr Al-Batin, the KSA. Genomic sequencing identified that approximately 12 MERS-CoV detections from a community outbreak in Hafr Al-Batin between June and August 2013 may have been triggered by an index case becoming infected through DC contact [175] . Sequencing MERS-CoV genomes from the 2013 Al-Ahsa hospital outbreak indicated that multiple viral variants contributed to the cases but that most were similar enough to each other to be consistent with human-tohuman transmission. Molecular epidemiology has revealed otherwise hidden links in transmission chains encompassing a period of up to five months [179] . However, most outbreaks have not continued for longer than two to three months and so opportunities for the virus to adapt further to humans through co-infection and sustained serial passage have been rare [169] . In Riyadh-2014, genetic evidence supported the likelihood of multiple external introductions of virus, implicating a range of healthcare facilities in an event that otherwise looked contiguous [23, 168, 179] . Riyadh is a nexus for camel and human travel and has had more MERS cases than any other region of the KSA to date but also harbours a wide range of MERS-CoV variants [128, 167, 179] . However the South Korean outbreak originated from a single infected person, resulting in three to four generations of cases [180, 181] . Studies of this apparently recombinant viral variant did not find an increased evolutionary rate and no sign of virus adaptation thus the outbreak seems to have been driven by circumstance rather than circumstance together with mutation [181] . For many MERS cases detected outside the Arabian Peninsula, extensive contact tracing has been performed and the results described in detail. Contact tracing is essential to contain the emergence and transmission of a new virus and today it is supported by molecular epidemiology. Although it is an expensive and time consuming process, contact tracing can identify potential new infections and through active or passive monitoring, react more rapidly if disease does develop. Results of contact tracing to date have found that onward transmission among humans is an infrequent event. For example, there were 83 contacts, both symptomatic and asymptomatic, of a case treated in Germany who travelled from the UAE but no sign of virus or antibody were found in any of them [73] . The very first MERS case had made contact with 56 HCWs and 48 others, but none developed any indication of infection [162] . In a study of 123 contacts of a case treated in France, only seven matched the definition for a possible case and were tested; one who had shared a 20 m 2 hospital room while in a bed 1.5 m away from the index case for a prolonged period was positive [26] . None of the contacts of the first two MERS cases imported into the USA in 2014 contained any MERS-CoV footprint [182] and none of the 131 contacts of two travellers returning to the Netherlands developed MERS-CoV antibodies or tested RNA positive [25, 183] . Analyses of public data reveal many likely instances of nosocomial acquisition of infection in the Arabian Peninsula and these data may be accompanied by some details noting contact with a known case or facility. One example identified the likely role of a patient with a subclinical infection, present in a hospital during their admission for other reasons, as the likeliest index case triggering a family cluster [93] . Contact tracing was a significant factor in the termination of a 2015 outbreak involving multiple South Korean hospitals [184] . Such studies demonstrate the necessity of finding and understanding a role for mild and asymptomatic cases, together with restricting close contact or prolonged exposure of infected people to others, especially older family members and friends with underlying disease (Fig. 4c) . The hospital-associated outbreak in Jeddah in 2014 was the largest and most rapid accumulation of MERS-CoV detections to date. The greatest number of MERS-CoV detections of any month on record occurred in Jeddah in April. The outbreak was mostly (>60 % of cases) associated with human-to-human spread within hospital environments and resulted from a lack of, or breakdown in, infection prevention and control [37, 185, 186] . A rise in fatalities followed the rapid increase in case numbers. In 2015 two large outbreaks occurred. South Korea was the site of the first large scale outbreak outside the Arabian Peninsula and produced the first cases in both South Korea and China, occurring between May and July 2015. This was closely followed by a distinct outbreak in Ar Riyad province in the KSA which appeared to come under control in early November. After staying in Bahrain for two weeks, a 68 year old male (68 M) travelled home to South Korea via Qatar, arriving free of symptoms on the 4 th May 2015 [187] . He developed fever, myalgia and a cough nearly a week later (11 th ). He visited a clinic as an outpatient between the 12 th and 15 th of May and was admitted to Hospital A on the 15 th [188] . He was discharged from Hospital A on the 17 th then visited and was admitted to the emergency department of Hospital B on the 18 th . During this second stay, a sputum sample was taken and tested positive for MERS-CoV on the 20 th [187, 188] , triggering transfer to the designated isolation treatment facility. Over a period of 10 days, the index case was seen at three different hospitals, demonstrating a key feature of "hospital shopping" that shaped the South Korean outbreak. Approximately 34 people were infected during this time [187] . In total 186 cases were generated in this outbreak, all linked through a single transmission chain to 68 M; 37 cases died [189] . In South Korea, the national health insurance system provides for relatively low cost medical care, defraying some costs by making family members responsible for a portion of the ministration of the sick, resulting in them sometimes staying for long periods in the rooms that often have more than four beds in them [24] . Other factors thought to have enabled this outbreak included unfamiliarity of local clinicians with MERS, ease with which the public can visit and be treated by tertiary hospitals, the custom of visiting sick friends and relatives in hospitals, the hierarchical nature of Korean society, crowded emergency rooms, poor IPC measures, a lack of negative pressure isolation rooms and poor inter-hospital communication of patient disease histories [24, [190] [191] [192] . All of the reported transmission occurred across three or four generations and apart from one unknown source, were all hospital-acquired [24, 120, 181, [193] [194] [195] . Few clinical details about these cases have been reported to date and detail on transmission and contact tracing is minimal. The hospitals involved were initially not identified, governmental guidance and actions produced confusing messages and there was very limited communication at all early on which resulted in unnecessary concern, distrust and a distinct economic impact [191, [196] [197] [198] . Early in the outbreak, a infected traveller, the son of an identified case in South Korea, passed through Hong Kong on his way to China where he was located, isolated and cared for in China [91, 199, 200] . No contacts became ill. The outbreak was brought under control in late July/ early August [201] after improved IPC measures were employed, strong contact tracing monitoring and quarantine, expanded laboratory testing, hospitals were better secured, specialized personnel were dispatched to manage cases and international cooperation increased [202, 203] . A review of public data showed that, as for MERS in the KSA, older age and the presence of underlying disease were significantly associated with a fatal outcome in South Korea. [40] Even though R 0 is <1, super-spreading events facilitated by circumstances created in healthcare settings and characterized by cluster sizes over 150, such as this one, are not unexpected from MERS-CoV infection [204] . The dynamic of an outbreak depends on the R 0 and an individual's viral shedding patterns, contact type and frequency, hospital procedures and population structure and density [204] . In the region of Ar Riyad, including the capital city of Riyadh, a hospital based cluster began, within a single hospital, from late June 2015 [205] . By mid-September there had been approximately170 cases reported but the outbreak appeared to been brought under control in November. It became apparent early on that MERS-CoV spread relatively ineffectively from human-to-human. Despite ongoing and possibly seasonal introduction of virus to the human population via infected DCs and perhaps other animals yet to be identified, the vast majority of MERS-CoV transmission has occurred from infected to uninfected humans in close and prolonged contact through circumstances created by poor infection control in health care settings. This opportunistic virus has had its greatest impact on those with underlying diseases and such vulnerable people, sometimes suffering multiple comorbidities, have been most often associated with hospitals, creating a perfect storm of exposure, transmission and mortality. It remains unclear if this group are uniquely affected by MERS-CoV or if other respiratory virus infections, including those from HCoVs, produce a similarly serious impact. In South Korea, a single imported case created an outbreak of 185 cases and 36 deaths that had a disproportionate impact on economic performance, community behaviour and trust in government and the health care system. Household human-to human transmission occurs but is also limited. Educational programs will be essential tools for combatting the spread of MERS-CoV both within urban and regional communities and for the health care setting. Vigilance remains important for containment since MERS-CoV is a virus with a genetic makeup that has been observed for only three years and is not stable. Among all humans reported to be infected, nearly 40 % have died. Continued laboratory testing, sequencing, analysis, timely data sharing and clear communication are essential for such vigilance to be effective. Global alignment of case definitions would further aid accurate calculation of a case fatality ratio by including subclinical case numbers. Whole genome sequencing has been used extensively to study MERS-CoV travel and variation and although it remains a tool for experts, it appears to be the best tool for the job. MERS and SARS have some clinical similarities but they also diverge significantly [206] . Defining characteristics include the higher PFC among MERS cases (above 50 % in 2013 and currently at 30-40 %; well above the 9 % of SARS) and the higher association between fatal MERS and older males with underlying comorbidities. For the viruses, MERS-CoV has a broader tropism, grows more rapidly in vitro, more rapidly induces cytopathogenic change, triggers distinct transcriptional responses, makes use of a different receptor, induces a more proinflammatory state and has a delayed innate antiviral response compared to SARS-CoV. There appears to be a 2-3 % prevalence of MERS-CoV in the KSA with a 5 % chance of secondary transmission within the household. There is an increased risk of infection through certain occupations at certain times and a much greater chance for spread to other humans during circumstances created by humans, which drives more effective transmission than any R 0 would predict on face value. Nonetheless, despite multiple mass gatherings that have afforded the virus many millions of opportunities to spread, there have remarkably been no reported outbreaks of MERS or MERS-CoV during or immediately after these events. There is no evidence that MERS-CoV is a virus of pandemic concern. Nonetheless, hospital settings continue to describe MERS cases and outbreaks in the Arabian Peninsula. As long as we facilitate the spread of MERS-CoV among our most vulnerable populations, the world must remain on alert for cases which may be exported more frequently when a host country with infected camel reservoirs is experiencing human clusters or outbreaks. The MERS-CoV appears to be an enzootic virus infecting the DC URT with evidence of recent genetic recombination. It may once have had its origins among bats, but evidence is lacking and the relevance of that to today's ongoing epidemic is academic. Thanks to quick action, the sensitive and rapid molecular diagnostic tools required to achieve rapid and sensitive detection goal have been in place and made widely available since the virus was reported in 2012. RT-PCR testing of LRT samples remains the gold standard for MERS-CoV confirmation. Serological tools continue to emerge but they are in need of further validation using samples from mild and asymptomatic infections and a densely sampled cohort study to follow contacts of new cases may address this need. Similarly, the important question of whether those who do shed MERS-CoV RNA for extended periods are infectious while appearing well, continues to go unanswered. It is even unclear just how many 'asymptomatic' infections have been described and reported correctly which in turn raises questions about the reliability of other clinical data collection to date. While the basic virology of MERS-CoV has advanced over the course of the past three years, understanding what is happening in, and the interplay between, camel, environment and human is still in its infancy. Additional file 1: Figure S1 . The

When did the first known cases of Middle East respiratory syndrome (MERS) occur?

MERS coronavirus: diagnostics, epidemiology and transmission https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4687373/ SHA: f6fcf1a99cbd073c5821d1c4ffa3f2c6daf8ae29 Authors: Mackay, Ian M.; Arden, Katherine E. Date: 2015-12-22 DOI: 10.1186/s12985-015-0439-5 License: cc-by Abstract: The first known cases of Middle East respiratory syndrome (MERS), associated with infection by a novel coronavirus (CoV), occurred in 2012 in Jordan but were reported retrospectively. The case first to be publicly reported was from Jeddah, in the Kingdom of Saudi Arabia (KSA). Since then, MERS-CoV sequences have been found in a bat and in many dromedary camels (DC). MERS-CoV is enzootic in DC across the Arabian Peninsula and in parts of Africa, causing mild upper respiratory tract illness in its camel reservoir and sporadic, but relatively rare human infections. Precisely how virus transmits to humans remains unknown but close and lengthy exposure appears to be a requirement. The KSA is the focal point of MERS, with the majority of human cases. In humans, MERS is mostly known as a lower respiratory tract (LRT) disease involving fever, cough, breathing difficulties and pneumonia that may progress to acute respiratory distress syndrome, multiorgan failure and death in 20 % to 40 % of those infected. However, MERS-CoV has also been detected in mild and influenza-like illnesses and in those with no signs or symptoms. Older males most obviously suffer severe disease and MERS patients often have comorbidities. Compared to severe acute respiratory syndrome (SARS), another sometimes- fatal zoonotic coronavirus disease that has since disappeared, MERS progresses more rapidly to respiratory failure and acute kidney injury (it also has an affinity for growth in kidney cells under laboratory conditions), is more frequently reported in patients with underlying disease and is more often fatal. Most human cases of MERS have been linked to lapses in infection prevention and control (IPC) in healthcare settings, with approximately 20 % of all virus detections reported among healthcare workers (HCWs) and higher exposures in those with occupations that bring them into close contact with camels. Sero-surveys have found widespread evidence of past infection in adult camels and limited past exposure among humans. Sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics have been available almost from the start of the emergence of MERS. While the basic virology of MERS-CoV has advanced over the past three years, understanding of the interplay between camel, environment, and human remains limited. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1186/s12985-015-0439-5) contains supplementary material, which is available to authorized users. Text: An email from Dr Ali Mohamed Zaki, an Egyptian virologist working at the Dr Soliman Fakeeh Hospital in Jeddah in the Kingdom of Saudi Arabia (KSA) announced the first culture of a new coronavirus to the world. The email was published on the website of the professional emerging diseases (ProMED) network on 20 th September 2012 [1] (Fig. 1) and described the first reported case, a 60 year old man from Bisha in the KSA. This information led to the rapid discovery of a second case of the virus, this time in an ill patient in the United Kingdom, who had been transferred from Qatar for care [2] . The new virus was initially called novel coronavirus (nCoV) and subsequentlty entitled the Middle East respiratoy syndrome coronavirus (MERS-CoV). As of 2 nd of September 2015, there have been 1,493 detections of viral RNA or virus-specific antibodies across 26 countries (Additional file 1: Figure S1 ) confirmed by the World Health Organization (WHO), with over a third of the positive people dying (at least 527, 35 %) [3] . Since that first report, a slow discovery process over the following two to three years revealed a virus that had infected over 90 % of adult dromedary camels (DC; Camelus dromedarius) in the KSA [4] , also DCs across the Arabian Peninsula and parts of Africa that are a source of DC imports for the KSA [5] . To date, MERS-CoV has not been detected in DCs tested in zoos or herds from other parts of the world [6] [7] [8] [9] . Occasionally, virus is transmitted from infected DCs to exposed humans. Subsequent transmission to other humans requires relatively close and prolonged exposure [10] . The first viral isolate was patented and concerns were raised that this would restrict access to both the virus and to viral diagnostics [11, 12] . However, sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics were quickly described and virus was made freely available subject to routine biosafety considerations [13] . Subsequent epidemiology and research has identified the cell receptor as exopeptidase dipeptidyl peptidase 4 (DPP4; also called CD26); that MERS-CoV has a broad tropism, replicating better in some cells lines and eliciting a more proinflammatory response than SARS-CoV; is widespread in DCs; has the potential to infect other animals and that MERS kills its human host more often than SARS did (20-40 % versus 9 % for SARS [14] ) [15] [16] [17] [18] [19] . In humans, overt disease was given the name Middle East respiratory syndrome, with the acronym MERS. From intermittent animal-to-human spill-over events, the MERS-CoV spreads sporadically among people, causing more severe disease among older adults, especially males, with pre-existing diseases. The spread of MERS-CoV among humans has often been associated with outbreaks in hospitals, with around 20 % of all cases to date involving healthcare workers (HCWs). Although DCs appear to suffer the equivalent of a 'common cold' from MERS-CoV infection, in humans, the virus can be a more serious and opportunistic pathogen associated with the death of up to 40 % of reported cases. It has yet to be established whether infections thought to have been acquired from an animal source produce a more severe outcome than those spread between humans [20] . Studies have established that the mean incubation period for MERS is five to six days, ranging from two to 16 days, with 13 to 14 days between when illness begins in one person and subsequently spreads to another [21] [22] [23] [24] . Among those with progressive illness, the median time to death is 11 to 13 days, ranging from five to 27 days [23, 24] . Fever and gastrointestinal symptoms may form a prodrome, after which symptoms decline, only to be followed by a more severe systemic and respiratory syndrome [25, 26] . The first WHO case definition [27] defined probable cases of MERS based on the presence of febrile illness, cough and requirement for hospitalization with suspicion of lower respiratory tract (LRT) involvement. It also included roles for contact with a probable or confirmed case or for travel or residence within the Arabian Peninsula. If strictly adhered to, only the severe syndrome would be subject to laboratory testing, which was the paradigm early on [21] . From July 2013, the revised WHO case definition included the importance of seeking out and understanding the role of asymptomatic cases and from June 2014, the WHO definition more clearly stated that a confirmed case included any person whose sample was RT-PCR positive for MERS-CoV, or who produced a seroconversion, irrespective of clinical signs and symptoms. [28] [29] [30] Apart from the WHO and the KSA Ministry of Health reports, asymptomatic or subclinical cases of MERS-CoV infection were documented in the scientific literature although not always as often as occurred early on [31, 32] . The KSA definition of a case became more strict on 13 th May 2014, relying on the presence of both clinical features and laboratory confirmation [33] . Testing of asymptomatic people was recommended against from December 2014 [34] , reinforced by a case definition released by the KSA Ministry of Health in June 2015 [35] . The KSA has been the source of 79 % of human cases. Severe MERS is notable for its impact among older men with comorbid diseases including diabetes mellitus, cirrhosis and various lung, renal and cardiac conditions [36] [37] [38] . Interestingly in June 2015, an outbreak in South Korea followed a similar distribution [39, 40] . Among laboratory confirmed cases, fever, cough and upper respiratory tract (URT) signs and symptoms usually occur first, followed within a week by progressive LRT distress and lymphopaenia [37] . Patients often present to a hospital with pneumonia, or worse, and secondary bacterial infections have been reported [37, 41] . Disease can progress to acute respiratory distress syndrome and multiorgan system failure [37] . MERS has reportedly killed approximately 35 % of all reported cases, 42 % of cases in the KSA, yet only 19 % of cases in South Korea, where mortality ranged from 7 % among younger age groups to 40 % among those aged 60 years and above [42] ; all may be inflated values with asymptomatic or mild infections sometimes not sought or not reported [34] . General supportive care is key to managing severe cases [43] . Children under the age of 14 years are rarely reported to be positive for MERS-CoV, comprising only 1.1 % (n = 16) of total reported cases. Between 1 st September 2012 and 2 nd December 2013, a study described the then tally of paediatric cases in the KSA, which stood at 11 (two to 16 years of age; median 13 years); nine were asymptomatic (72 %) and one infant died [44] . In Amman, Jordan, 1,005 samples from hospitalized children under the age of two years with fever and/or respiratory signs and symptoms were tested but none were positive for MERS-CoV RNA, despite being collected at a similar time to the first known outbreak of MERS-CoV in the neighbouring town of Al-Zarqa [45] . A second trimester stillbirth occurred in a pregnant woman during an acute respiratory illness and while not RT-rtPCR positive, the mother did subsequently develop antibodies to MERS-CoV, suggestive of recent infection [46] . Her exposure history to a MERS-CoV RT-rtPCR positive relative and an antibody-reactive husband, her incubation period and her symptom history met the WHO criteria for being a probable MERS-CoV case [46] . Diagnostic methods were published within days of the ProMED email announcing the first MERS case [47] , including several now gold standard in-house RT-rtPCR assays (Fig. 2 ) as well as virus culture in Vero and LLC-MK2 cells [18, 47, 48] . A colorectal adenocarcinoma (Caco-2) epithelial cell line has since been recommended for isolation of infections MERS-CoV [49] . We previously [18] .). Open reading frames are indicated as yellow rectangles bracketed by terminal untranslated regions (UTR; grey rectangles). FS-frame-shift. Predicted regions encompassing recombination break-points are indicated by orange pills. Created using Geneious v8.1 [211] and annotated using Adobe Illustrator. Beneath this is a schematic depicting the location of RT-PCR primers (blue arrows indicate direction) and oligoprobes (green rectangles) used in the earliest RT-rtPCR screening assays and conventional, semi-nested (three primers) RT-PCR confirmatory sequencing assays [47, 48] . Publication order is noted by first [27 th September 2012; red] and second [6 th December 2012; orange] coloured rectangles; both from Corman et al. [47, 48] Those assays recommended by the WHO are highlighted underneath by yellow dots [53] . The NSeq reverse primer has consistently contained one sequence mismatch with some MERS-CoV variants. An altered version of that from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] reviewed the broad tropism of MERS-CoV [5] . However, as is well described, cell culture is a slow, specialised and insensitive method [50] while PCR-based techniques are the preferred method for MERS-CoV detection. The first open reading frames (ORF 1a and 1b; Fig. 2 ) have become a key diagnostic and taxonomic target for CoV species identification. With less than 80 % identity between the amino acid sequence of MERS ORF 1ab and betacoronavirus relatives, Tylonycteris bat HKU4 and Pipistrellus bat HKU5, it can be concluded that it is a novel and distinct virus. MERS-CoV is predicted to encode ten open reading frames with 5' and 3' untranslated regions [51] . The structural proteins include the spike (S), envelope (E), membrane (M) and nucleocapsid (N) [52] . The products of ORF1a and ORF1b are predicted to encode nonstructural proteins. The majority of specimen testing to date has employed validated RT-rtPCR assays shown to be sensitive and specific [47, 48, 53] . The RealStar® kit uses these WHOrecommended assays [54] . The target sequences of these screening assays have not changed among genomes examined until at least mid-2015 (IMM observation). Other RT-rtPCR assays have been developed and validated for use as laboratory-based diagnostic tools [55] [56] [57] . Additionally, loop-mediated [58, 59] or recombinase polymerase [60] isothermal assays have been designed for field deployment. The detection of MERS-CoV antigen has not been common to date but the combination of short turnaround time from test to result, high throughput and identification of viral proteins makes this an attractive option. Detection of viral proteins rather than viral RNA indicates the likely presence of infectious virus. The first rapid immunochromatographic tool described could detect recombinant MERS-CoV nucleocapsid protein from DC nasal swabs with 94 % sensitivity and 100 % specificity compared to RT-rtPCR [61] . A different approach used a monoclonal antibody-based capture ELISA targeting the MERS-CoV nucleocapsid protein with a sensitivity of 10 3 TCID 50 and 100 % specificity [62] . Demonstration of a seroconversion to a MERS-CoV infection meets the current WHO definition of a case so optimized and thoroughly validated sero-assays employed alongside good clinical histories are useful to both identify prior MERS-CoV infection and help support transmission studies. Because serology testing is, by its nature, retrospective, it is usual to detect a viral footprint, in the form of antibodies, in the absence of any signs or symptoms of disease and often in the absence of any viral RNA [63] . Strategic, widespread sero-surveys of humans using samples collected after 2012 are infrequent. Much of the Arabian Peninsula and all of the Horn of Africa lack baseline data describing the proportion of the community who may have been infected by a MERS-CoV. However, sero-surveys have had widespread use in elucidating the role of DCs as a transmission source for MERS-CoV. Because of the identity shared between DC and human MERS-CoV (see Molecular epidemiology: using genomes to understand outbreaks), serological assays for DC sero-surveys should be transferrable to human screening with minimal re-configuration. Also, no diagnostically relevant variation in neutralization activity have been found from among a range of circulating tested MERS-CoV isolates and sera, so whole virus or specific protein-based sero-assays should perform equivalently in detecting serological responses to the single MERS-CoV serotype [49] . The development of robust serological assays requires reliable panels of wellcharacterized animal or human sera, including those positive for antibodies specific to MERS-CoV, as well as to likely sources of cross-reaction [64] . Obtaining these materials was problematic and slowed the development and commercialization of antibody detection assays for human testing [64] . A number of commercial ELISA kits, immunofluorescent assays (IFA) kits, recombinant proteins and monoclonal antibodies have been released [31, [65] [66] [67] [68] . Initially, conventional IFAs were used for human sero-surveys. These relied on MERS-CoV-infected cell culture as an antigen source, detecting the presence of human anti-MERS-CoV IgG, IgM or neutralizing antibodies in human samples [18, 48, 69] . No sign of MERS-CoV antibodies was found among 2,400 sera from patients visiting Hospital in Jeddah, from 2010 through 2012, prior to the description of MERS-CoV [18] . Nor did IFA methods detect any sign of prior MERS-CoV infection among a small sample of 130 healthy blood donors from another Hospital in Jeddah (collected between Jan and Dec 2012) [70] . Of 226 slaughterhouse workers, only eight (3.5 %) were positive by IFA, and those sera could not be confirmed by virus neutralization (NT) test. The study indicated that HCoV-HKU1 was a likely source of crossreactive antigen in the whole virus IFA [70] . Whole virus MERS-CoV IFA also suffered from some cross-reactivity with convalescent SARS patient sera and this could not be resolved by an NT test which was also cross-reactive [71] . IFA using recombinant proteins instead of whole-virus IFA, has been shown to be a more specific tool [31] . Since asymptomatic zoonoses have been posited [72] , an absence of antibodies to MERS-CoV among some humans who have regular and close contact with camels may reflect the rarity of actively infected animals at butcheries, a limited transmission risk associated with slaughtering DCs [70] , a pre-existing cross-protective immune status or some other factor(s) resulting in a low risk of disease and concurrent seroconversion developing after exposure in this group. IFA using recombinant proteins instead. Some sero-assays have bypassed the risks of working with infectious virus by creating transfected cells expressing recombinant portions of the MERS-CoV nucleocapsid and spike proteins [48, 73] , or using a recombinant lentivirus expressing MERS-CoV spike protein and luciferase [74, 75] . A pseudo particle neutralization (ppNT) assay has seen widespread used in animal studies and was at least as sensitive as the traditional microneutralization (MNT) test. [10, 74, [76] [77] [78] ] Studies using small sample numbers and ppNT found no evidence of MERS-CoV neutralizing antibody in sera from 158 children with LRT infections between May 2010 and May 2011, 110 sera from 19 to 52 year old male blood donors and 300 selfidentified animal workers from the Jazan Region of the KSA during 2012 [79, 80] . Similarly, a study of four herdsmen in contact with an infected DC herd in Al-Ahsa, eight people who had intermittent contact with the herd, 30 veterinary surgeons and support staff who were not exposed to the herd, three unprotected abattoir workers in Al-Ahsa and 146 controls who were not exposed to DCs in any professional role, found none with serological evidence of past MERS-CoV infection using the ppNT assay [10] . A delay in the neutralizing antibody response to MERS-CoV infection was associated with increased disease severity in South Korea cases with most responses detectable by week three of illness while others, even though disease was severe, did not respond for four or more weeks [81] . The implications for our ability to detect any response in mild or asymptomatic cases was not explored but may be a signifcant factor in understanding exposure in the wider community. A Jordanian outbreak of acute LRT disease in a hospital in 2012 was retrospectively found to be associated with MERS-CoV infection, initially using RT-rtPCR, but subsequently, and on a larger scale, through positivity by ELISA and IFA or MNT test. [46, 82, 83] This outbreak predated the first case of MERS in the KSA. The ELISA used a recombinant nucleocapsid protein from the group 2 betacoronavirus bat-CoV HKU5 to identify antibodies against the equivalent crossreactive MERS-CoV protein [71] . It was validated using 545 sera collected from people with prior HCoV-OC43, HCoV-229E, SARS-CoV, HCoV-NL63, HRV, HMPV or influenza A(H1N1) infections but was reportedly less specific than the recombinant IFA discussed above. It was still considered an applicable tool for screening large sample numbers [82] . A protein microarray expressing the S1 protein subunit has also been validated and widely used for DC testing [5, 84] . Detection of MERS-CoV infection using ELISA or S1 subunit protein microarray [84] is usually followed by confirmatory IFA and/ or a plaque-reduction neutralization (PRNT) [69, 70, 85] or MNT test. [74, 85, 86] This confirmatory process aims toensure the antibodies detected are able to specifically neutralize the intended virus and are not more broadly reactive to other coronaviruses found in DCs (bovine CoV, BCoV) or humans (HCoV-OC43, HCoV-229E, HCoV-NL63, HCoV-HKU1, SARS-CoV). In the largest study of human sera, a tiered diagnostic process assigned both recombinant IFA and recombinant ELISA positive sera to 'stage 1' seropositivity. A stage 2 seropositive result additionally required a suitably titred PRNT result [87] . The study found 15 sera collected in 2012 to 2013 from 10,009 (0.2 %) people in 13 KSA provinces contained MERS-CoV antibodies, but significantly higher proportions in occurred in camel shepherds (two of 87; 2.3 %) and slaughterhouse workers (five of 140; 3.6 %) [87] . Contemporary surveys are needed. MERS-CoV does not appear to be easily transmitted from DCs to humans, or perhaps it is [72] , but generally does not trigger a detectable immune response if only mild disease or asymptomatic infection results. Serology assays are in need of further validation in this area so care is required when moving newly developed diagnostic serology algorithms from a research setting to one that informs public health decisions. This was reinforced when a false positive US case, purported to have been infected after a handshake and two face-to-face meetings, did not withstand further confirmatory analysis using a more specific, NT assay and was subsequently retracted [88, 89] . The WHO recommends sampling from the LRT for MERS-CoV RT-rtPCR testing, especially when sample collection is delayed by a week or more after onset of symptoms. [53] LRT samples are also best for attempting isolation of infectious virus, although the success of culture is reduced when disease persists [49] . Recommended sample types include bronchoalveolar lavage (BAL), tracheal/tracheobronchial aspirate, pleural fluid and sputum [53, 90] . Fresh samples yield better diagnostic results than refrigerated material [69] and if delays in testing of ≥72 h are likely, samples (except for blood) should be frozen at −70°C [90] . If available, lung biopsy or autopsy tissues can also be tested [53] . The URT is a less invasive and more convenient sampling site however, and an oropharyngeal and throat swab or a nasopharyngeal aspirate/wash are recommended when URT sampling is to be conducted [90] . Paired sera, collected two to three weeks apart are preferable for serological testing while a single sample is suggested to be sufficient if collected two weeks after onset of disease or a single serum collected during the first 10-12 days if conducting RT-rtPCR [53, 90] . Human urine and stool have been found to contain MERS-CoV RNA 12 to 26 days after symptom onset [25, 69, 91] and are listed as samples that should be considered [53, 90] . In two cases that arrived in the Netherlands, urine was RT-rtPCR negative but faeces was weakly positive and sera were RT-rtPCR positive for five days or more [25] . The finding of MERS-CoV viral RNA in serum provides an avenue for retrospective PCR-based studies if respiratory samples are unavailable [83] . RNAaemia may also correlate with disease severity; signs of virus were cleared from the serum of a recovered patient, yet lingered until the death of another [92] . Clinically suspected MERS cases may return negative results by RT-rtPCR. Data have shown one or more negative URT samples may be contradicted by further URT sampling or the use of LRT samples, which is preferred [2, 43, 93] . Higher viral loads occur in the LRT compared to the URT. [22, 69, 88, 94] This fits with the observation that the majority of disease symptoms are reported to manifest as systemic and LRT disease [21] . However, on occasion, even LRT specimens from MERS cases may initially be negative, only to later become positive by RT-PCR [95] . This may be due to poor sampling when a cough is absent or non-productive or because the viral load is low [95] . Despite this both the largest human MERS-CoV studies [32, [96] [97] [98] and smaller ones [22, 25, 99] , use samples from the URT. It is then noteworthy that one study reported an association between higher loads in the URT and worse clinical outcome including intensive care and death [94] . At writing, no human data exist to define whether the virus replicates solely or preferentially in the LRT or URT, or replicates in other human tissues in vivo although MERS-CoV RNA has been detected from both the URT and LRT in a macaque monkey model [100] .The distribution of DPP4 in the human upper airways is also not well described. Individual human case studies report long periods of viral shedding, sometimes intermittently and not necessarily linked to the presence of disease symptoms. [25, 69, 99, 101] In one instance, a HCW shed viral RNA for 42 days in the absence of disease [99] . It is an area of high priority to better understand whether such cases are able to infect others. Over three quarters of MERS cases shed viral RNA in their LRT specimens (tracheal aspirates and sputum) for at least 30 days, while only 30 % of contacts were still shedding RNA in their URT specimens [91, 102] . In the only study to examine the effect of sample type on molecular analysis, 64 nasopharyngeal aspirates (NPA; an URT sample), 30 tracheal aspirates, 13 sputa and three BAL were examined. The tracheal aspirates and BAL returned the highest viral load values followed by NPA and sputum. Unsurprisingly, higher viral loads generally paralleled whole genome sequencing and culture success and, in NPA testing, were significantly correlated with severe disease and death [49, 94, 103] . This study demonstrated the importance of LRT sampling for whole genome sequencing. When tested, samples positive for MERS-CoV are often negative for other pathogens [2, 25, 93, 104] . However, many studies make no mention of additional testing for endemic human respiratory viruses [21, 23, 73, 105] . When viruses are sought, they have included human herpesvirus (HHV), rhinoviruses (HRV), enteroviruses (EV), respiratory syncytial virus (RSV), parainfluenzavirus types 1, 2 and 3 (PIVs),influenzaviruses (IFVs), endemic HCoVs, adenoviruses (AdVs) metapneumovirus (MPV) and influenza A\H1N1 virus; co-detections with MERS-CoV have been found on occasion [2, 22, 37, 69, 97] . Bacterial testing is sometimes included (for example, for Legionella and Pneumococcus) but the impact of bacterial co-presence is also unclear [22, [104] [105] [106] . Further testing of the LRT sample from the first MERS case used IFA to screen for some viruses (negative for IFV, PIVs, RSV and AdVs) and RT-PCR for others (negative for AdV, EVs, MPV and HHVs) [18] . RT-PCR also detected MERS-CoV. The WHO strongly recommends testing for other respiratory pathogens [53] but with this recommendation often discounted, there are limited data to address the occurrence and impact of co-infections or alternative viral diagnoses among both MERS cases and their contacts. Little is known of other causes of MERS-like pneumonia in the KSA or of the general burden of disease due to the known classical respiratory viruses. Testing of adult pilgrims performing the Hajj in 2012 to 2014 has not detected any MERS-CoV. In 2012, nasal swabs from 154 pilgrims collected prior to leaving for or departing from the KSA were tested [47] . In 2013, testing was significantly scaled up with 5,235 nasopharyngeal swabs from 3,210 incoming pilgrims and 2,025 swabs from outgoing pilgrims tested [98] . It should be noted that most pilgrims arrived from MERS-free countries. A further 114 swabs were taken from pilgrims with influenza-like illness [96, 107] . In earlier Hajj gatherings, it was found that influenza viruses circulated widely, whilst other viruses, often rhinoviruses, circulated more selectively, interpreted as indicating their importation along with foreign pilgrims. [107] [108] [109] Over time, increased influenza vaccination has been credited for a fall in the prevalence of influenza like illnesses among Hajj pilgrims. [110] A LRT sample is often not collected for these studies [98, 107, 109] , so false negative findings are a possibility although little is known about the initial site of MERS-CoV infection and replication; it may have been assumed it was the LRT because disease was first noticed there but the URT may be the site of the earliest replication. In Jeddah between March and July 2014 (hereafter called the Jeddah-2014 outbreak; Fig. 3 ), there was a rapid increase in MERS cases, accompanied by intense screening; approximately 5,000 samples from in and around the region were tested in a month yielding around 140 MERS-CoV detections (~3 % prevalence) [111] . Among 5,065 individuals sampled and tested across the KSA between October 2012 and September 2013,108 (2.1 %) detections were made in a hospital-centric population which included hospitalized cases (n = 2,908; 57.4 %), their families (n = 462; 9.1 %) and associated HCWs (n = 1,695; 33.5 %) [32] . Among the detections, 19 (17.8 %) were HCWs and 10 (9.3 %) were family contacts [32] . The 2-3 % prevalence of active MERS-CoV infections is not dissimilar to the hospital-based prevalence of other human CoVs. [112] However, the proportion of deaths among those infected with MERS-CoV is much higher than that known for the HCoVs NL63, HKU1, 229E or OC43 in other countries, and even above that for SARS-CoV; it is not a virus that could reasonably be described as a "storm in a teacup". It is the low transmission rate that has prevented worldwide spread, despite many "opportunities". Very early in the MERS outbreak, some animals were highly regarded as either the reservoir or intermediate host(s) of MERS-CoV with three of the first five cases having contact with DCs [73, 113, 114] . Today, animal MERS-CoV infections must be reported to the world organization for animal health as an emerging disease [115] . A summary of the first MERS cases reported by the WHO defined animal contact with humans as being direct and within 10 days prior to symptom onset [20] . This definition made no specific allowance for acquisition from DCs through a droplet-based route, which is very likely route for acquisition of a virus that initially and predominantly causes respiratory disease [23] . Camels are known to produce high levels of MERS-CoV RNA in their URT and lungs [116] . Providing support for a droplet transmission route and perhaps indicating the presence of RNA in smaller, drier droplet nuclei, MERS-CoV RNA was identified in a high volume air sample collected from a barn housing an infected DC [117] . The precise source from which humans acquire MERS-CoV remains poorly studied but it seems likely that animal and human behavioural factors may play roles (Fig. 3) [118] . These factors may prove important for human cases who do not describe any DC contact [119] nor any contact with a confirmed case. Whether the WHO definition of animal contact is sufficient to identify exposure to this respiratory virus remains unclear. Wording focuses on consumption of DC products but does not specifically ascribe risk to a droplet route for acquisition of MERS-CoV from DC [120] . Some MERS patients are listed in WHO disease notices as being in proximity to DCs or farms, but the individuals have not described coming into contact with the animals. No alternative path for acquiring infection is reported in many of these instances. What constitutes a definition of "contact" during these interviews has been defined for one study [72] . Despite this lack of clarity, the WHO consider that evidence linking MERS-CoV transmission between DCs to humans is irrefutable (Fig. 4) [120] . The possibility that bats were an animal host of MERS-CoV was initially widely discussed because of the existing diversity of coronaviruses known to reside among them [121] [122] [123] [124] . Conclusive evidence supporting bats as a source for human infections by MERS-CoV has yet to be found, but bats do appear to host ancestral representatives [53, 125] . However, these are not variants of the same virus nor always within the same phylogenetic lineage as MERS-CoV; they are each a genetically distinct virus. Bat-to-human infection by MERS-CoV is a purely speculative event. The only piece of MERS-CoV-specific evidence pointing to bats originates from amplification of a 190 nt fragment of the RNAdependent RNA polymerase gene of the MERS-CoV genome, identified in a faecal pellet from an insectivorous Emballonuridae bat, Taphozous perforatus found in Bisha, the KSA [121] . While very short, the sequence of the fragment defined it as a diagnostic discovery. Subsequently a link to DCs was reported [85] and that link has matured into a verified association [38, 126] (Fig. 4) . (See figure on previous page.) Fig. 3 Monthly detections of MERS-CoV (blue bars) and of cases who died (red bars) with some dates of interest marked for 2012 to 4 th September 2015. An approximation of when DC calving season [128] and when recently born DCs are weaned is indicated. Spring (green) and summer (orange) in the Arabian Peninsula are also shaded. Note the left-hand y-axis scale for 2014 and 2015 which is greater than for 2012/13. Sources of these public data include the WHO, Ministries of Health and FluTrackers [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] . Modified and reprinted from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] DCs, which make up 95 % of all camels, have a central presence in the Arabian Peninsula where human-DC contact ranges from little to close [119] . Contact may be commonplace and could occur in variety of ways (Fig. 4a) . There are several large well-attended festivals, races, sales and parades which feature DCs and DCs are also kept and bred close to populated areas in the KSA [127, 128] . DC milk and meat are widely consumed and the older DC is an animal of ritual significance after the Hajj pilgrimage [129] . However, MERS-CoV infection frequency is reportedly much lower than is the widespread and frequent habit of eating, drinking and preparing DC products. Daily ingestion of fresh unpasteurized DC milk is common among the desert Bedouin and many others in the KSA. DC urine is also consumed or used for supposed health benefits. Despite camel butchery being a local occupation, neither butchers nor other at-risk groups are identifiable among MERS cases; this may simply be a reporting issue rather than an unexplainable absence of MERS. A small case-control study published in 2015 identified direct DC contact, and not ingestion of products, to be associated with onset of MERS [38] . The first sero-survey of livestock living in the Middle East region was conducted during 2012-2013 [85] . DCs were sampled from a mostly Canary Island-born herd and from Omani DCs (originally imported from the Horn of Africa) [85] . A neutralising antibody assay found only 10 % of strongly seropositive Canary Island [5] . b Camel-to-human infections appear to be infrequent, while human-to-human spread of infection is regularly facilitated by poor IPC in healthcare settings where transmission is amplified, accounting for the bulk of cases. There are human MERS cases that do not fall into either category of source and it is unclear if these acquired infection through some entirely separate route, or from cases that escaped diagnosis. c Hypothetical ways in which subclinical (when infection may not meet a previously defined clinical threshold of signs and/or symptoms) or asymptomatic (no obvious signs or measured, noticed or recalled symptoms of illness) MERS-CoV infection may be implicated in transmission DC sera could neutralise MERS-CoV while all Omani DC sera had high levels of specific MERS-CoV neutralizing antibody [85] . This indicated that DCs had in the past been infected by MERS-CoV, or a very similar virus. Since this study, a host of peer-reviewed reports have looked at both DCs and other animals, and the possibility that they may host MERS-CoV infection. Seropositive DCs have been found throughout the Arabian Peninsula including Oman, the KSA, Qatar, Jordan, the United Arab Emirates (UAE), Kuwait as well as Sudan, Somalia, Egypt, Tunisia, Nigeria, Kenya and Ethiopia in Africa and the Canary Islands [85, [130] [131] [132] [133] [134] . Other animals tested include sheep, cows, pigs, horses, donkeys, mules, birds, water buffalo, goats, Bactrian camels, llamas and guanaco (south American camelids) but none had detectable neutralising antibody against MERS-CoV [4, 74, 78, 85, 86, 135, 136] . No virology or serology studies of human samples from areas in Africa where there are camels with a history of MERS-CoV have been reported to date. However,an absence of unexplained pneumonia that may be attributable to MERS-CoV infection may not signal the absence of virus among humans in each country but simply reflect a lack of expensive epidemiology studies conducted by resource-poor countries. It is thus unclear whether MERS-CoV, or an antigenically related CoV, is an unrecognized pathogen in these regions, perhaps circulating for even longer than it has been known in the Arabian Peninsula [133] . MERS-CoV RNA has also been detected in DC samples, and recovery of infectious virus has also been achieved from DC samples [4, 77, 117, 132, [137] [138] [139] [140] [141] . From some of these, full or majority length genomes of MERS-CoV have been sequenced [77, 137, 138] . DC versions of MERS-CoV were found to be as similar to each other, as were variants detected from different humans over time and across distance. Antibody screening assays have also detected crossreactive antibodies in sera. These were identified as such by screening sera against similar viruses, for example BCoV or HCoV-OC43 (as an antigenic facsimile for BCoV). It is possible that other MERS-CoV-like viruses also reside within DCs, but this does not detract from the definitive finding of MERS-CoV genetic sequences in both DCs and humans [117, 142, 143] . Screening studies have shown that juvenile DCs are more often positive for virus or viral RNA while older DCs are more likely to be seropositive and RNA or virus negative [76, 77, 144] . In adult DCs, MERS-CoV RNA has been detected among animals with pre-existing antibody, suggesting re-infection is possible [77, 144] . Viral loads among positive DCs can be very high [4, 76, 77, 139, 144] and DCs have been found positive both when ill with URT respiratory signs [77, 117, 142, 145] or when apparently healthy [137] . These findings indicate DCs host natural MERS-CoV infections. Furthermore, stored DC sera have revealed signs of MERS-CoV in DCs which date back over three decades (the earliest collected in 1983) [4, 133, 135] . Older sera have not been tested and so precisely how long DCs have been afflicted by MERS-CoV, whether the virus is enzootic among them, introduced to them decades or centuries ago from bats in Africa or the Arabian Peninsula, or they are the subject of regular but short-lived viral incursions from an as yet unknown host, cannot be answered. Researchers sought to determine a direction for infection; were DCs transmitting virus to humans or were humans infecting DCs? At a Qatari site, a farm owner and his employee became ill in mid-October 2013 and tested positive for MERS-CoV RNA in a sputum and throat swab sample, respectively. RT-rtPCRs found MERS-CoV RNA in 11 of 14 positive DC nasal swabs at the farm; six (43 %) positive by two or more assays [138] . The results indicated a recent outbreak had occurred in this herd; the first indication of MERS-CoV RNA found within DCs with a temporal association to human infections. Three positive DC samples were confirmed by sequencing a 358 nt portion of the spike gene; these sequences were identical to each other, again with close homology to other human and DC MERS-CoV sequences [138] . The DCs and human contacts yielded ORF1a and ORF4b sequences differing by only a single nucleotide each, clustering closely with the Hafr-Al-Batin_1_2013 variant [138] . Subsequent case studies found evidence of a concurrent human and DC infection and the direction of that infection was inferred to be from the ill DCs and to their human owners [117, 142, 146] . Partial genome sequences indicated that a human and a MERS-CoV RT-rtPCR positive DC had been infected by a variant of the same virus, harbouring the same distinct pattern of nucleotide polymorphisms. [142] All nine DC in the owner's herd, serially sampled, reacted in a recombinant S1 antigen ELISA, with the two animals that had been RT-rtPCR positive showing a small, verifiable rise in antibody titre [142] . A rise in titre theoretically begins 10 to 21 days after DC infection [142] . The authors suggested that the rise in titre in DC sera which occurred alongside a declining RNA load, while the patient was actively ill and hospitalized, indicated that the DCs were infected first followed by the owner [117, 142] . BCoV antibodies were also present, and rising in one of the two RT-rtPCR positive animals but no animal's antibodies could neutralise BCoV infection [142] . Camel calving season occurs in the winter months (between late October and late February; Fig. 3 ) and this may be a time when there is increased risk to humans of spill-over due to new infections among naïve DC populations [128] . What role maternal camel antibody might play in delaying infection of calves remains unknown [128, 142] . Juvenile DCs appear to host active infection more often than adult DCs and thus the sacrificial slaughter of DCs, which must be five years of age or older (termed a thane), may not be accompanied by significant risk of exposure to infection. In contrast to earlier results, slaughterhouse workers who kill both younger and older DCs, may be an occupational group with significantly higher incidence of seropositivity to MERS-CoV when animals have active MERS-CoV infections [129, 139, [147] [148] [149] . Expanded virological investigations of African DCs may lead to more seropositive animals and geographic areas in which humans may be at risk. It is possible that there are areas where humans already harbour MERS-CoV infections that have not been identified because of an absence of laboratory surveillance. Virological investigations of bats may lead to findings of ancestral viruses and viral 'missing links' and identifying any other animal sources of zoonotic spread is important to inform options for reducing human exposures [56, 76] . Infectious MERS-CoV added to DC, goat or cow milk and stored at 4°C could be recovered at least 72 h later and, if stored at 22°C, recovery was possible for up to 48 h [150] . MERS-CoV titre decreased somewhat when recovered from milk at 22°C but pasteurization completely ablated MERS-CoV infectivity [150] . In a subsequent study, MERS-CoV RNA was identified in the milk, nasal secretion and faeces of DCs from Qatar [151] . A single study has examined the ability of MERS-CoV to survive in the environment [150] . Plastic or steel surfaces were inoculated with 10 6 TCID 50 of MERS-CoV at different temperature and relative humidity (RH) and virus recovery was attempted in cell culture. At high ambient temperature (30°C) and low RH (30 %) MERS-CoV remained viable for 24 h [150] . By comparison, a well known and efficently transmitted respiratory virus, influenza A virus, could not be recovered in culture beyond four hours under any conditions [150] . Aerosol experiments found MERS-CoV viability only decreased 7 % at low RH at 20°C. In comparison, influenza A virus decreased by 95 % [150] . MERS-CoV survival is inferior to that previously demonstrated for SARS-CoV [152] . For context, pathogenic bacteria can remain viable and airborne for 45 min in a coughed aerosol and can spread 4 m. MERS-CoV's ability to remain viable over long time periods gives it the capacity to thoroughly contaminate a room's surfaces when occupied by an infected and symptomatic patient [153] . Whether MERS-CoV can remain adrift and infectious for extended periods (truly airborne) remains unknown. Such findings expand our understanding of the possibilities for droplets to transmit respiratory viruses in many settings, including hospital waiting rooms, emergency departments, treatment rooms, open intensive care facilities and private patient rooms. The nature and quality of air exchange, circulation and filtration are important variables in risk measurement and reduction as is the use of negative pressure rooms to contain known cases. Droplet spread between humans is considered the mechanism of human-to-human transmission and the need for droplet precautions was emphasized after the Al-Ahsa hospital, the KSA and the South Korean outbreaks [21, 23, 154, 155] . By extrapolation, aerosol-generating events involving DCs (urination, defecation, and preparation and consumption of DC products) should be factored into risk measurement and reduction efforts and messaged using appropriate context. The provision of evidence supporting the best formulation of personal protective equipment to be worn by HCWs who receive, manage or conduct procedures on infectious cases remains a priority. MERS-CoV was found and characterized because of its apparent association with severe, and therefore more obvious, illness in humans; we were the canaries in the coal mine. Sero-assays and prospective cohort studies have yet to determine the extent to which milder or asymptomatic cases contribute to MERS-CoV transmission chains. However, transmission of MERS-CoV is defined as sporadic (not sustained), intra-familial, often healthcare associated, inefficient and requiring close and prolonged contact [22, 31, 63, 93, 97, 102, 156] In a household study, 14 of 280 (5 %) contacts of 26 MERS-CoV positive index patients were RNA or antibody positive; the rate of general transmission, even in outbreaks is around 3 % [31] . It seems that the majority of human cases of MERS-CoV, even when numbers appear to increase suddenly, do not readily transmit to more than one other human so to date, the localized epidemic of MERS-CoV has not been self-sustaining [157] [158] [159] [160] [161] . That is to say, the basic reproduction number (R 0 ) -the average number of infections caused by one infected individual in a fully susceptible populationhas been close to one throughout various clusters and outbreaks. If R 0 was greater than 1, a sustained increase in case numbers would be expected. Some R o calculations may be affected by incomplete case contact tracing, limited community testing and how a case is defined. That MERS has had a constant presence in the Arabian Peninsula since 2012 is due to ongoing, sporadic spill-over events from DCs amplified by poorly controlled hospital outbreaks. The first known MERS human-to-human transmission event was one characterized by acute LRT disease in a healthcare setting in Jordan. In stark contrast, a sero-survey of HCW who were sometimes in close and prolonged contact with the first, fatal MERS-CoV case in 2012 [162] , found none of the HCW had seroconverted four months later, despite an absence of eye protection and variable compliance with required PPE standards [162] . Early on in the MERS story, samples for testing were mostly collected from patients with severe illness and not those with milder acute respiratory tract infections. Contacts of confirmed MERS cases were often observed for clinical illness, but not tested. These omissions may have confounded our understanding of MERS-CoV transmission and biased early data towards higher numbers of seriously ill and hospitalized patients, inflating the apparent proportion of fatal cases. Case-control studies were not a focus. As testing paradigms changed and contacts were increasingly tested, more asymptomatic and mild infections were recognized [163] . A rise in the cases termed asymptomatic (which enlarge the denominator for calculations of the proportion of fatal cases, defined in [164] ) resulted in a drop in the proportion of fatal cases during the Jeddah-2014 outbreak. Historically, such rises are consistent with changing definitions and laboratory responses and clinical management of a newly discovered virus infection that was first noted only among the severely ill. Upon follow-up, over three-quarters of such MERS-CoV RNA positive people did recall having one or more symptoms at the time, despite being reported as asymptomatic [165] raising some question over the reliability of other reported data. The proportion of fatal MERS cases within the KSA compared to outside the KSA, as well as the age, and sex distribution change in different ways when comparing MERS outbreaks. Approximately 43 % of MERS cases (549 of 1277) in the KSA were fatal betwen 2012 and December 2015 while 21 % (72 of 330) died among those occurring outside of the KSA. The total number of male cases always outnumber females and the proportion of male deaths is always greater than the proportion of females who die. However the proportion of male deaths from total males with MERS is a similar figure to that for females. In the KSA, there is a greater proportion of younger males among cases and deaths than were observed from the 2015 South Korean or the Jeddah-2014 outbreaks (Additional file 2: Figure S2 ). Why these aspects have differed may be due to differences in the time to presentation and diagnosis, the nature and quality of supportive care, the way a person became infected (habits, exposure to a human or zoonotic source, viral load, route of infection) or the extent to which different populations are burdened by underlying diseases [40] . As a group, HCWs comprised 16 % of MERS cases in the KSA and South Korea. It is apparent that the weekly proportion of infected HCWs increases alongside each steep rise in overall detections (Fig. 5) . In May 2013, the WHO published guidelines for IPC during care of probable or confirmed cases of MERS-CoV infection in a healthcare setting [166] . This is explainable because to date, each case rise has been intimately associated with healthcare-facility related outbreaks [118] . These rises in MERS-CoV detections can decrease the average age during each event because HCWs are usually younger than inpatients with MERS. Healthcare facilities have been a regular target for suggested improvements aimed at improving infection prevention and control (IPC) procedures [115, 118] . Most of the analysis of MERS-CoV genetics has been performed using high throughput or "deep" sequencing methods for complete genome deduction [167] [168] [169] . MERS-CoV was the first subject of such widespread use of deep sequencing to study an emerging viral outbreak with global reach. The technique can produce genomic [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] length coverage in a single experiment with highly repetitious measurement of each nucleotide position [52, 140] . Despite assays having been published early on, subgenomic sequencing, once the mainstay of viral outbreak studies, has less often been published during MERS-CoV characterization [48] . As more genomes from both humans and DCs have been characterized, two clades have become apparent; A and B (Fig. 6) . Clade A contains only human-derived MERS-CoV genomes from Jordan, while Clade B comprises the majority of human and camel genomes deduced thus far [168] . Two studies during 2015, one looking at Jeddah-2014 MERS-CoV variants and another looking at a variant exported from South Korea to China, have now identified signs of genetic recombination among MERS-CoV variants. While human and camel whole genome sequences have retained >99 % identity with each other, members of genetically distinct lineages can and do swap genetic material when suitable conditions and coinfections co-occur [170] [171] [172] . Shared identity implies that the major source for human acquisition is the DC, rather than another animal, although more testing of other animal species is needed to confirm that conclusion. Over a month, a DC virus sequenced on different occasions did not change at all indicating a degree of genomic stability in its host, supporting that DCs are the natural, rather than intermediate, host for the MERS-CoV we know today [77] . To date, recombination has been localised to breakpoints near the boundary between ORF1a and ORF1b regions, within the spike gene [170] and in the ORF1b region (Fig. 2) [172] . It is not unexpected that recombination should occur since it is well known among other CoVs [124] and because the majority of MERS-CoV whole genomes collected from samples spanning three years (2012-2015) and from humans, camels and different countries have shown close genetic identity to each other, with just enough subtle variation to support outbreak investigations so long as whole genome sequencing is applied [52, 77, 135, 138, 168, [173] [174] [175] . Changes in genome sequence may herald alterations to virus transmissibility, replication, persistence, lethality or response to future drugs. If we have prior knowledge of the impact of genetic changes because of thorough characterization studies, we can closely Fig. 6 The genetic relationship between MERS-CoV nucleotide sequences (downloaded from GenBank using the listed accession numbers and from virological.org [212] ). This neighbour joining tree was created in MEGA v6 using an alignment of human and DCderived MERS-CoV sequences (Geneious v8.1 [211] ). Clades are indicated next to dark (Clade A) or pale (Clade B) blue vertical bars. Camel icons denote genomes from DCs. Healthcare or community outbreaks are boxed and labelled using previously described schemes [212, 213] monitor the genomic regions and better understand any changes in transmission or disease patterns as they occur. Genetic mutations noted during the largest of human outbreaks, Jeddah-2014, did not impart any major replicative or immunomodulatory changes when compared to earlier viral variants in vitro [156, 176] . However, we understand very little of the phenotypic outcomes that result from subtle genetic change in MERS-CoV genomes. To date no clinical relevance or obvious in vivo changes to viral replication, shedding or transmission has been reported or attributed to mutations or to new recombinant viruses [156] . But vigilance and larger, more contemporary and in vivo studies are needed. Genome sequence located to a distinct clade were identified from an Egyptian DC that was probably imported from Sudan. This does not fit into either of the current clades [125, 168, 177] . A virus sequenced from a Neoromicia capensis bat was more closely related to MERS-CoV than other large bat-derived sequences had been to that point, but the genome of a variant of a MERS-CoV has yet to be discovered and deduced from any bat [125] . Analyses of MERS-CoV genomes have shown that most single nucleotide differences among variants were located in the last third of the genome (Fig. 2) , which encodes the spike protein and accessory proteins [168] . At least nine MERS-CoV genomes contained amino acid substitutions in the receptor binding domain (RBD) of the spike protein and codons 158 (N-terminal region), 460 (RBD), 1020 (in heptad repeat 1), 1202 and 1208 bear investigation as markers of adaptive change [140, 169] . The spike protein had not changed in the recombinant MERS-CoV genome identified in China in 2015 but was reported to have varied at a higher rate than that for complete MERS-CoV genomes, among South Korean variants [172, 178] . This highlights that subgenomic regions may not always contain enough genetic diversity to prove useful for differentiating viral variants. Despite this, one assay amplifying a 615 nucleotide fragment of the spike S2 domain gene for Sanger sequencing agreed with the results generated by the sequencing of a some full genomes and was useful to define additional sequence groupings [177] . Genomic sequence can also be used to define the geographic boundaries of a cluster or outbreak and monitor its progress, based on the similarity of the variants found among infected humans and animals when occurring together, or between different sites and times (Fig. 6 ) [169] . This approach was employed when defining the geographically constrained MERS hospital outbreak in Al-Ahsa, which occurred between 1 st April and 23 rd May 2013, as well as clusters in Buraidah and a community outbreak in Hafr Al-Batin, the KSA. Genomic sequencing identified that approximately 12 MERS-CoV detections from a community outbreak in Hafr Al-Batin between June and August 2013 may have been triggered by an index case becoming infected through DC contact [175] . Sequencing MERS-CoV genomes from the 2013 Al-Ahsa hospital outbreak indicated that multiple viral variants contributed to the cases but that most were similar enough to each other to be consistent with human-tohuman transmission. Molecular epidemiology has revealed otherwise hidden links in transmission chains encompassing a period of up to five months [179] . However, most outbreaks have not continued for longer than two to three months and so opportunities for the virus to adapt further to humans through co-infection and sustained serial passage have been rare [169] . In Riyadh-2014, genetic evidence supported the likelihood of multiple external introductions of virus, implicating a range of healthcare facilities in an event that otherwise looked contiguous [23, 168, 179] . Riyadh is a nexus for camel and human travel and has had more MERS cases than any other region of the KSA to date but also harbours a wide range of MERS-CoV variants [128, 167, 179] . However the South Korean outbreak originated from a single infected person, resulting in three to four generations of cases [180, 181] . Studies of this apparently recombinant viral variant did not find an increased evolutionary rate and no sign of virus adaptation thus the outbreak seems to have been driven by circumstance rather than circumstance together with mutation [181] . For many MERS cases detected outside the Arabian Peninsula, extensive contact tracing has been performed and the results described in detail. Contact tracing is essential to contain the emergence and transmission of a new virus and today it is supported by molecular epidemiology. Although it is an expensive and time consuming process, contact tracing can identify potential new infections and through active or passive monitoring, react more rapidly if disease does develop. Results of contact tracing to date have found that onward transmission among humans is an infrequent event. For example, there were 83 contacts, both symptomatic and asymptomatic, of a case treated in Germany who travelled from the UAE but no sign of virus or antibody were found in any of them [73] . The very first MERS case had made contact with 56 HCWs and 48 others, but none developed any indication of infection [162] . In a study of 123 contacts of a case treated in France, only seven matched the definition for a possible case and were tested; one who had shared a 20 m 2 hospital room while in a bed 1.5 m away from the index case for a prolonged period was positive [26] . None of the contacts of the first two MERS cases imported into the USA in 2014 contained any MERS-CoV footprint [182] and none of the 131 contacts of two travellers returning to the Netherlands developed MERS-CoV antibodies or tested RNA positive [25, 183] . Analyses of public data reveal many likely instances of nosocomial acquisition of infection in the Arabian Peninsula and these data may be accompanied by some details noting contact with a known case or facility. One example identified the likely role of a patient with a subclinical infection, present in a hospital during their admission for other reasons, as the likeliest index case triggering a family cluster [93] . Contact tracing was a significant factor in the termination of a 2015 outbreak involving multiple South Korean hospitals [184] . Such studies demonstrate the necessity of finding and understanding a role for mild and asymptomatic cases, together with restricting close contact or prolonged exposure of infected people to others, especially older family members and friends with underlying disease (Fig. 4c) . The hospital-associated outbreak in Jeddah in 2014 was the largest and most rapid accumulation of MERS-CoV detections to date. The greatest number of MERS-CoV detections of any month on record occurred in Jeddah in April. The outbreak was mostly (>60 % of cases) associated with human-to-human spread within hospital environments and resulted from a lack of, or breakdown in, infection prevention and control [37, 185, 186] . A rise in fatalities followed the rapid increase in case numbers. In 2015 two large outbreaks occurred. South Korea was the site of the first large scale outbreak outside the Arabian Peninsula and produced the first cases in both South Korea and China, occurring between May and July 2015. This was closely followed by a distinct outbreak in Ar Riyad province in the KSA which appeared to come under control in early November. After staying in Bahrain for two weeks, a 68 year old male (68 M) travelled home to South Korea via Qatar, arriving free of symptoms on the 4 th May 2015 [187] . He developed fever, myalgia and a cough nearly a week later (11 th ). He visited a clinic as an outpatient between the 12 th and 15 th of May and was admitted to Hospital A on the 15 th [188] . He was discharged from Hospital A on the 17 th then visited and was admitted to the emergency department of Hospital B on the 18 th . During this second stay, a sputum sample was taken and tested positive for MERS-CoV on the 20 th [187, 188] , triggering transfer to the designated isolation treatment facility. Over a period of 10 days, the index case was seen at three different hospitals, demonstrating a key feature of "hospital shopping" that shaped the South Korean outbreak. Approximately 34 people were infected during this time [187] . In total 186 cases were generated in this outbreak, all linked through a single transmission chain to 68 M; 37 cases died [189] . In South Korea, the national health insurance system provides for relatively low cost medical care, defraying some costs by making family members responsible for a portion of the ministration of the sick, resulting in them sometimes staying for long periods in the rooms that often have more than four beds in them [24] . Other factors thought to have enabled this outbreak included unfamiliarity of local clinicians with MERS, ease with which the public can visit and be treated by tertiary hospitals, the custom of visiting sick friends and relatives in hospitals, the hierarchical nature of Korean society, crowded emergency rooms, poor IPC measures, a lack of negative pressure isolation rooms and poor inter-hospital communication of patient disease histories [24, [190] [191] [192] . All of the reported transmission occurred across three or four generations and apart from one unknown source, were all hospital-acquired [24, 120, 181, [193] [194] [195] . Few clinical details about these cases have been reported to date and detail on transmission and contact tracing is minimal. The hospitals involved were initially not identified, governmental guidance and actions produced confusing messages and there was very limited communication at all early on which resulted in unnecessary concern, distrust and a distinct economic impact [191, [196] [197] [198] . Early in the outbreak, a infected traveller, the son of an identified case in South Korea, passed through Hong Kong on his way to China where he was located, isolated and cared for in China [91, 199, 200] . No contacts became ill. The outbreak was brought under control in late July/ early August [201] after improved IPC measures were employed, strong contact tracing monitoring and quarantine, expanded laboratory testing, hospitals were better secured, specialized personnel were dispatched to manage cases and international cooperation increased [202, 203] . A review of public data showed that, as for MERS in the KSA, older age and the presence of underlying disease were significantly associated with a fatal outcome in South Korea. [40] Even though R 0 is <1, super-spreading events facilitated by circumstances created in healthcare settings and characterized by cluster sizes over 150, such as this one, are not unexpected from MERS-CoV infection [204] . The dynamic of an outbreak depends on the R 0 and an individual's viral shedding patterns, contact type and frequency, hospital procedures and population structure and density [204] . In the region of Ar Riyad, including the capital city of Riyadh, a hospital based cluster began, within a single hospital, from late June 2015 [205] . By mid-September there had been approximately170 cases reported but the outbreak appeared to been brought under control in November. It became apparent early on that MERS-CoV spread relatively ineffectively from human-to-human. Despite ongoing and possibly seasonal introduction of virus to the human population via infected DCs and perhaps other animals yet to be identified, the vast majority of MERS-CoV transmission has occurred from infected to uninfected humans in close and prolonged contact through circumstances created by poor infection control in health care settings. This opportunistic virus has had its greatest impact on those with underlying diseases and such vulnerable people, sometimes suffering multiple comorbidities, have been most often associated with hospitals, creating a perfect storm of exposure, transmission and mortality. It remains unclear if this group are uniquely affected by MERS-CoV or if other respiratory virus infections, including those from HCoVs, produce a similarly serious impact. In South Korea, a single imported case created an outbreak of 185 cases and 36 deaths that had a disproportionate impact on economic performance, community behaviour and trust in government and the health care system. Household human-to human transmission occurs but is also limited. Educational programs will be essential tools for combatting the spread of MERS-CoV both within urban and regional communities and for the health care setting. Vigilance remains important for containment since MERS-CoV is a virus with a genetic makeup that has been observed for only three years and is not stable. Among all humans reported to be infected, nearly 40 % have died. Continued laboratory testing, sequencing, analysis, timely data sharing and clear communication are essential for such vigilance to be effective. Global alignment of case definitions would further aid accurate calculation of a case fatality ratio by including subclinical case numbers. Whole genome sequencing has been used extensively to study MERS-CoV travel and variation and although it remains a tool for experts, it appears to be the best tool for the job. MERS and SARS have some clinical similarities but they also diverge significantly [206] . Defining characteristics include the higher PFC among MERS cases (above 50 % in 2013 and currently at 30-40 %; well above the 9 % of SARS) and the higher association between fatal MERS and older males with underlying comorbidities. For the viruses, MERS-CoV has a broader tropism, grows more rapidly in vitro, more rapidly induces cytopathogenic change, triggers distinct transcriptional responses, makes use of a different receptor, induces a more proinflammatory state and has a delayed innate antiviral response compared to SARS-CoV. There appears to be a 2-3 % prevalence of MERS-CoV in the KSA with a 5 % chance of secondary transmission within the household. There is an increased risk of infection through certain occupations at certain times and a much greater chance for spread to other humans during circumstances created by humans, which drives more effective transmission than any R 0 would predict on face value. Nonetheless, despite multiple mass gatherings that have afforded the virus many millions of opportunities to spread, there have remarkably been no reported outbreaks of MERS or MERS-CoV during or immediately after these events. There is no evidence that MERS-CoV is a virus of pandemic concern. Nonetheless, hospital settings continue to describe MERS cases and outbreaks in the Arabian Peninsula. As long as we facilitate the spread of MERS-CoV among our most vulnerable populations, the world must remain on alert for cases which may be exported more frequently when a host country with infected camel reservoirs is experiencing human clusters or outbreaks. The MERS-CoV appears to be an enzootic virus infecting the DC URT with evidence of recent genetic recombination. It may once have had its origins among bats, but evidence is lacking and the relevance of that to today's ongoing epidemic is academic. Thanks to quick action, the sensitive and rapid molecular diagnostic tools required to achieve rapid and sensitive detection goal have been in place and made widely available since the virus was reported in 2012. RT-PCR testing of LRT samples remains the gold standard for MERS-CoV confirmation. Serological tools continue to emerge but they are in need of further validation using samples from mild and asymptomatic infections and a densely sampled cohort study to follow contacts of new cases may address this need. Similarly, the important question of whether those who do shed MERS-CoV RNA for extended periods are infectious while appearing well, continues to go unanswered. It is even unclear just how many 'asymptomatic' infections have been described and reported correctly which in turn raises questions about the reliability of other clinical data collection to date. While the basic virology of MERS-CoV has advanced over the course of the past three years, understanding what is happening in, and the interplay between, camel, environment and human is still in its infancy. Additional file 1: Figure S1 . The

Where did the first known cases of Middle East respiratory syndrome (MERS) occur?

MERS coronavirus: diagnostics, epidemiology and transmission https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4687373/ SHA: f6fcf1a99cbd073c5821d1c4ffa3f2c6daf8ae29 Authors: Mackay, Ian M.; Arden, Katherine E. Date: 2015-12-22 DOI: 10.1186/s12985-015-0439-5 License: cc-by Abstract: The first known cases of Middle East respiratory syndrome (MERS), associated with infection by a novel coronavirus (CoV), occurred in 2012 in Jordan but were reported retrospectively. The case first to be publicly reported was from Jeddah, in the Kingdom of Saudi Arabia (KSA). Since then, MERS-CoV sequences have been found in a bat and in many dromedary camels (DC). MERS-CoV is enzootic in DC across the Arabian Peninsula and in parts of Africa, causing mild upper respiratory tract illness in its camel reservoir and sporadic, but relatively rare human infections. Precisely how virus transmits to humans remains unknown but close and lengthy exposure appears to be a requirement. The KSA is the focal point of MERS, with the majority of human cases. In humans, MERS is mostly known as a lower respiratory tract (LRT) disease involving fever, cough, breathing difficulties and pneumonia that may progress to acute respiratory distress syndrome, multiorgan failure and death in 20 % to 40 % of those infected. However, MERS-CoV has also been detected in mild and influenza-like illnesses and in those with no signs or symptoms. Older males most obviously suffer severe disease and MERS patients often have comorbidities. Compared to severe acute respiratory syndrome (SARS), another sometimes- fatal zoonotic coronavirus disease that has since disappeared, MERS progresses more rapidly to respiratory failure and acute kidney injury (it also has an affinity for growth in kidney cells under laboratory conditions), is more frequently reported in patients with underlying disease and is more often fatal. Most human cases of MERS have been linked to lapses in infection prevention and control (IPC) in healthcare settings, with approximately 20 % of all virus detections reported among healthcare workers (HCWs) and higher exposures in those with occupations that bring them into close contact with camels. Sero-surveys have found widespread evidence of past infection in adult camels and limited past exposure among humans. Sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics have been available almost from the start of the emergence of MERS. While the basic virology of MERS-CoV has advanced over the past three years, understanding of the interplay between camel, environment, and human remains limited. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1186/s12985-015-0439-5) contains supplementary material, which is available to authorized users. Text: An email from Dr Ali Mohamed Zaki, an Egyptian virologist working at the Dr Soliman Fakeeh Hospital in Jeddah in the Kingdom of Saudi Arabia (KSA) announced the first culture of a new coronavirus to the world. The email was published on the website of the professional emerging diseases (ProMED) network on 20 th September 2012 [1] (Fig. 1) and described the first reported case, a 60 year old man from Bisha in the KSA. This information led to the rapid discovery of a second case of the virus, this time in an ill patient in the United Kingdom, who had been transferred from Qatar for care [2] . The new virus was initially called novel coronavirus (nCoV) and subsequentlty entitled the Middle East respiratoy syndrome coronavirus (MERS-CoV). As of 2 nd of September 2015, there have been 1,493 detections of viral RNA or virus-specific antibodies across 26 countries (Additional file 1: Figure S1 ) confirmed by the World Health Organization (WHO), with over a third of the positive people dying (at least 527, 35 %) [3] . Since that first report, a slow discovery process over the following two to three years revealed a virus that had infected over 90 % of adult dromedary camels (DC; Camelus dromedarius) in the KSA [4] , also DCs across the Arabian Peninsula and parts of Africa that are a source of DC imports for the KSA [5] . To date, MERS-CoV has not been detected in DCs tested in zoos or herds from other parts of the world [6] [7] [8] [9] . Occasionally, virus is transmitted from infected DCs to exposed humans. Subsequent transmission to other humans requires relatively close and prolonged exposure [10] . The first viral isolate was patented and concerns were raised that this would restrict access to both the virus and to viral diagnostics [11, 12] . However, sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics were quickly described and virus was made freely available subject to routine biosafety considerations [13] . Subsequent epidemiology and research has identified the cell receptor as exopeptidase dipeptidyl peptidase 4 (DPP4; also called CD26); that MERS-CoV has a broad tropism, replicating better in some cells lines and eliciting a more proinflammatory response than SARS-CoV; is widespread in DCs; has the potential to infect other animals and that MERS kills its human host more often than SARS did (20-40 % versus 9 % for SARS [14] ) [15] [16] [17] [18] [19] . In humans, overt disease was given the name Middle East respiratory syndrome, with the acronym MERS. From intermittent animal-to-human spill-over events, the MERS-CoV spreads sporadically among people, causing more severe disease among older adults, especially males, with pre-existing diseases. The spread of MERS-CoV among humans has often been associated with outbreaks in hospitals, with around 20 % of all cases to date involving healthcare workers (HCWs). Although DCs appear to suffer the equivalent of a 'common cold' from MERS-CoV infection, in humans, the virus can be a more serious and opportunistic pathogen associated with the death of up to 40 % of reported cases. It has yet to be established whether infections thought to have been acquired from an animal source produce a more severe outcome than those spread between humans [20] . Studies have established that the mean incubation period for MERS is five to six days, ranging from two to 16 days, with 13 to 14 days between when illness begins in one person and subsequently spreads to another [21] [22] [23] [24] . Among those with progressive illness, the median time to death is 11 to 13 days, ranging from five to 27 days [23, 24] . Fever and gastrointestinal symptoms may form a prodrome, after which symptoms decline, only to be followed by a more severe systemic and respiratory syndrome [25, 26] . The first WHO case definition [27] defined probable cases of MERS based on the presence of febrile illness, cough and requirement for hospitalization with suspicion of lower respiratory tract (LRT) involvement. It also included roles for contact with a probable or confirmed case or for travel or residence within the Arabian Peninsula. If strictly adhered to, only the severe syndrome would be subject to laboratory testing, which was the paradigm early on [21] . From July 2013, the revised WHO case definition included the importance of seeking out and understanding the role of asymptomatic cases and from June 2014, the WHO definition more clearly stated that a confirmed case included any person whose sample was RT-PCR positive for MERS-CoV, or who produced a seroconversion, irrespective of clinical signs and symptoms. [28] [29] [30] Apart from the WHO and the KSA Ministry of Health reports, asymptomatic or subclinical cases of MERS-CoV infection were documented in the scientific literature although not always as often as occurred early on [31, 32] . The KSA definition of a case became more strict on 13 th May 2014, relying on the presence of both clinical features and laboratory confirmation [33] . Testing of asymptomatic people was recommended against from December 2014 [34] , reinforced by a case definition released by the KSA Ministry of Health in June 2015 [35] . The KSA has been the source of 79 % of human cases. Severe MERS is notable for its impact among older men with comorbid diseases including diabetes mellitus, cirrhosis and various lung, renal and cardiac conditions [36] [37] [38] . Interestingly in June 2015, an outbreak in South Korea followed a similar distribution [39, 40] . Among laboratory confirmed cases, fever, cough and upper respiratory tract (URT) signs and symptoms usually occur first, followed within a week by progressive LRT distress and lymphopaenia [37] . Patients often present to a hospital with pneumonia, or worse, and secondary bacterial infections have been reported [37, 41] . Disease can progress to acute respiratory distress syndrome and multiorgan system failure [37] . MERS has reportedly killed approximately 35 % of all reported cases, 42 % of cases in the KSA, yet only 19 % of cases in South Korea, where mortality ranged from 7 % among younger age groups to 40 % among those aged 60 years and above [42] ; all may be inflated values with asymptomatic or mild infections sometimes not sought or not reported [34] . General supportive care is key to managing severe cases [43] . Children under the age of 14 years are rarely reported to be positive for MERS-CoV, comprising only 1.1 % (n = 16) of total reported cases. Between 1 st September 2012 and 2 nd December 2013, a study described the then tally of paediatric cases in the KSA, which stood at 11 (two to 16 years of age; median 13 years); nine were asymptomatic (72 %) and one infant died [44] . In Amman, Jordan, 1,005 samples from hospitalized children under the age of two years with fever and/or respiratory signs and symptoms were tested but none were positive for MERS-CoV RNA, despite being collected at a similar time to the first known outbreak of MERS-CoV in the neighbouring town of Al-Zarqa [45] . A second trimester stillbirth occurred in a pregnant woman during an acute respiratory illness and while not RT-rtPCR positive, the mother did subsequently develop antibodies to MERS-CoV, suggestive of recent infection [46] . Her exposure history to a MERS-CoV RT-rtPCR positive relative and an antibody-reactive husband, her incubation period and her symptom history met the WHO criteria for being a probable MERS-CoV case [46] . Diagnostic methods were published within days of the ProMED email announcing the first MERS case [47] , including several now gold standard in-house RT-rtPCR assays (Fig. 2 ) as well as virus culture in Vero and LLC-MK2 cells [18, 47, 48] . A colorectal adenocarcinoma (Caco-2) epithelial cell line has since been recommended for isolation of infections MERS-CoV [49] . We previously [18] .). Open reading frames are indicated as yellow rectangles bracketed by terminal untranslated regions (UTR; grey rectangles). FS-frame-shift. Predicted regions encompassing recombination break-points are indicated by orange pills. Created using Geneious v8.1 [211] and annotated using Adobe Illustrator. Beneath this is a schematic depicting the location of RT-PCR primers (blue arrows indicate direction) and oligoprobes (green rectangles) used in the earliest RT-rtPCR screening assays and conventional, semi-nested (three primers) RT-PCR confirmatory sequencing assays [47, 48] . Publication order is noted by first [27 th September 2012; red] and second [6 th December 2012; orange] coloured rectangles; both from Corman et al. [47, 48] Those assays recommended by the WHO are highlighted underneath by yellow dots [53] . The NSeq reverse primer has consistently contained one sequence mismatch with some MERS-CoV variants. An altered version of that from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] reviewed the broad tropism of MERS-CoV [5] . However, as is well described, cell culture is a slow, specialised and insensitive method [50] while PCR-based techniques are the preferred method for MERS-CoV detection. The first open reading frames (ORF 1a and 1b; Fig. 2 ) have become a key diagnostic and taxonomic target for CoV species identification. With less than 80 % identity between the amino acid sequence of MERS ORF 1ab and betacoronavirus relatives, Tylonycteris bat HKU4 and Pipistrellus bat HKU5, it can be concluded that it is a novel and distinct virus. MERS-CoV is predicted to encode ten open reading frames with 5' and 3' untranslated regions [51] . The structural proteins include the spike (S), envelope (E), membrane (M) and nucleocapsid (N) [52] . The products of ORF1a and ORF1b are predicted to encode nonstructural proteins. The majority of specimen testing to date has employed validated RT-rtPCR assays shown to be sensitive and specific [47, 48, 53] . The RealStar® kit uses these WHOrecommended assays [54] . The target sequences of these screening assays have not changed among genomes examined until at least mid-2015 (IMM observation). Other RT-rtPCR assays have been developed and validated for use as laboratory-based diagnostic tools [55] [56] [57] . Additionally, loop-mediated [58, 59] or recombinase polymerase [60] isothermal assays have been designed for field deployment. The detection of MERS-CoV antigen has not been common to date but the combination of short turnaround time from test to result, high throughput and identification of viral proteins makes this an attractive option. Detection of viral proteins rather than viral RNA indicates the likely presence of infectious virus. The first rapid immunochromatographic tool described could detect recombinant MERS-CoV nucleocapsid protein from DC nasal swabs with 94 % sensitivity and 100 % specificity compared to RT-rtPCR [61] . A different approach used a monoclonal antibody-based capture ELISA targeting the MERS-CoV nucleocapsid protein with a sensitivity of 10 3 TCID 50 and 100 % specificity [62] . Demonstration of a seroconversion to a MERS-CoV infection meets the current WHO definition of a case so optimized and thoroughly validated sero-assays employed alongside good clinical histories are useful to both identify prior MERS-CoV infection and help support transmission studies. Because serology testing is, by its nature, retrospective, it is usual to detect a viral footprint, in the form of antibodies, in the absence of any signs or symptoms of disease and often in the absence of any viral RNA [63] . Strategic, widespread sero-surveys of humans using samples collected after 2012 are infrequent. Much of the Arabian Peninsula and all of the Horn of Africa lack baseline data describing the proportion of the community who may have been infected by a MERS-CoV. However, sero-surveys have had widespread use in elucidating the role of DCs as a transmission source for MERS-CoV. Because of the identity shared between DC and human MERS-CoV (see Molecular epidemiology: using genomes to understand outbreaks), serological assays for DC sero-surveys should be transferrable to human screening with minimal re-configuration. Also, no diagnostically relevant variation in neutralization activity have been found from among a range of circulating tested MERS-CoV isolates and sera, so whole virus or specific protein-based sero-assays should perform equivalently in detecting serological responses to the single MERS-CoV serotype [49] . The development of robust serological assays requires reliable panels of wellcharacterized animal or human sera, including those positive for antibodies specific to MERS-CoV, as well as to likely sources of cross-reaction [64] . Obtaining these materials was problematic and slowed the development and commercialization of antibody detection assays for human testing [64] . A number of commercial ELISA kits, immunofluorescent assays (IFA) kits, recombinant proteins and monoclonal antibodies have been released [31, [65] [66] [67] [68] . Initially, conventional IFAs were used for human sero-surveys. These relied on MERS-CoV-infected cell culture as an antigen source, detecting the presence of human anti-MERS-CoV IgG, IgM or neutralizing antibodies in human samples [18, 48, 69] . No sign of MERS-CoV antibodies was found among 2,400 sera from patients visiting Hospital in Jeddah, from 2010 through 2012, prior to the description of MERS-CoV [18] . Nor did IFA methods detect any sign of prior MERS-CoV infection among a small sample of 130 healthy blood donors from another Hospital in Jeddah (collected between Jan and Dec 2012) [70] . Of 226 slaughterhouse workers, only eight (3.5 %) were positive by IFA, and those sera could not be confirmed by virus neutralization (NT) test. The study indicated that HCoV-HKU1 was a likely source of crossreactive antigen in the whole virus IFA [70] . Whole virus MERS-CoV IFA also suffered from some cross-reactivity with convalescent SARS patient sera and this could not be resolved by an NT test which was also cross-reactive [71] . IFA using recombinant proteins instead of whole-virus IFA, has been shown to be a more specific tool [31] . Since asymptomatic zoonoses have been posited [72] , an absence of antibodies to MERS-CoV among some humans who have regular and close contact with camels may reflect the rarity of actively infected animals at butcheries, a limited transmission risk associated with slaughtering DCs [70] , a pre-existing cross-protective immune status or some other factor(s) resulting in a low risk of disease and concurrent seroconversion developing after exposure in this group. IFA using recombinant proteins instead. Some sero-assays have bypassed the risks of working with infectious virus by creating transfected cells expressing recombinant portions of the MERS-CoV nucleocapsid and spike proteins [48, 73] , or using a recombinant lentivirus expressing MERS-CoV spike protein and luciferase [74, 75] . A pseudo particle neutralization (ppNT) assay has seen widespread used in animal studies and was at least as sensitive as the traditional microneutralization (MNT) test. [10, 74, [76] [77] [78] ] Studies using small sample numbers and ppNT found no evidence of MERS-CoV neutralizing antibody in sera from 158 children with LRT infections between May 2010 and May 2011, 110 sera from 19 to 52 year old male blood donors and 300 selfidentified animal workers from the Jazan Region of the KSA during 2012 [79, 80] . Similarly, a study of four herdsmen in contact with an infected DC herd in Al-Ahsa, eight people who had intermittent contact with the herd, 30 veterinary surgeons and support staff who were not exposed to the herd, three unprotected abattoir workers in Al-Ahsa and 146 controls who were not exposed to DCs in any professional role, found none with serological evidence of past MERS-CoV infection using the ppNT assay [10] . A delay in the neutralizing antibody response to MERS-CoV infection was associated with increased disease severity in South Korea cases with most responses detectable by week three of illness while others, even though disease was severe, did not respond for four or more weeks [81] . The implications for our ability to detect any response in mild or asymptomatic cases was not explored but may be a signifcant factor in understanding exposure in the wider community. A Jordanian outbreak of acute LRT disease in a hospital in 2012 was retrospectively found to be associated with MERS-CoV infection, initially using RT-rtPCR, but subsequently, and on a larger scale, through positivity by ELISA and IFA or MNT test. [46, 82, 83] This outbreak predated the first case of MERS in the KSA. The ELISA used a recombinant nucleocapsid protein from the group 2 betacoronavirus bat-CoV HKU5 to identify antibodies against the equivalent crossreactive MERS-CoV protein [71] . It was validated using 545 sera collected from people with prior HCoV-OC43, HCoV-229E, SARS-CoV, HCoV-NL63, HRV, HMPV or influenza A(H1N1) infections but was reportedly less specific than the recombinant IFA discussed above. It was still considered an applicable tool for screening large sample numbers [82] . A protein microarray expressing the S1 protein subunit has also been validated and widely used for DC testing [5, 84] . Detection of MERS-CoV infection using ELISA or S1 subunit protein microarray [84] is usually followed by confirmatory IFA and/ or a plaque-reduction neutralization (PRNT) [69, 70, 85] or MNT test. [74, 85, 86] This confirmatory process aims toensure the antibodies detected are able to specifically neutralize the intended virus and are not more broadly reactive to other coronaviruses found in DCs (bovine CoV, BCoV) or humans (HCoV-OC43, HCoV-229E, HCoV-NL63, HCoV-HKU1, SARS-CoV). In the largest study of human sera, a tiered diagnostic process assigned both recombinant IFA and recombinant ELISA positive sera to 'stage 1' seropositivity. A stage 2 seropositive result additionally required a suitably titred PRNT result [87] . The study found 15 sera collected in 2012 to 2013 from 10,009 (0.2 %) people in 13 KSA provinces contained MERS-CoV antibodies, but significantly higher proportions in occurred in camel shepherds (two of 87; 2.3 %) and slaughterhouse workers (five of 140; 3.6 %) [87] . Contemporary surveys are needed. MERS-CoV does not appear to be easily transmitted from DCs to humans, or perhaps it is [72] , but generally does not trigger a detectable immune response if only mild disease or asymptomatic infection results. Serology assays are in need of further validation in this area so care is required when moving newly developed diagnostic serology algorithms from a research setting to one that informs public health decisions. This was reinforced when a false positive US case, purported to have been infected after a handshake and two face-to-face meetings, did not withstand further confirmatory analysis using a more specific, NT assay and was subsequently retracted [88, 89] . The WHO recommends sampling from the LRT for MERS-CoV RT-rtPCR testing, especially when sample collection is delayed by a week or more after onset of symptoms. [53] LRT samples are also best for attempting isolation of infectious virus, although the success of culture is reduced when disease persists [49] . Recommended sample types include bronchoalveolar lavage (BAL), tracheal/tracheobronchial aspirate, pleural fluid and sputum [53, 90] . Fresh samples yield better diagnostic results than refrigerated material [69] and if delays in testing of ≥72 h are likely, samples (except for blood) should be frozen at −70°C [90] . If available, lung biopsy or autopsy tissues can also be tested [53] . The URT is a less invasive and more convenient sampling site however, and an oropharyngeal and throat swab or a nasopharyngeal aspirate/wash are recommended when URT sampling is to be conducted [90] . Paired sera, collected two to three weeks apart are preferable for serological testing while a single sample is suggested to be sufficient if collected two weeks after onset of disease or a single serum collected during the first 10-12 days if conducting RT-rtPCR [53, 90] . Human urine and stool have been found to contain MERS-CoV RNA 12 to 26 days after symptom onset [25, 69, 91] and are listed as samples that should be considered [53, 90] . In two cases that arrived in the Netherlands, urine was RT-rtPCR negative but faeces was weakly positive and sera were RT-rtPCR positive for five days or more [25] . The finding of MERS-CoV viral RNA in serum provides an avenue for retrospective PCR-based studies if respiratory samples are unavailable [83] . RNAaemia may also correlate with disease severity; signs of virus were cleared from the serum of a recovered patient, yet lingered until the death of another [92] . Clinically suspected MERS cases may return negative results by RT-rtPCR. Data have shown one or more negative URT samples may be contradicted by further URT sampling or the use of LRT samples, which is preferred [2, 43, 93] . Higher viral loads occur in the LRT compared to the URT. [22, 69, 88, 94] This fits with the observation that the majority of disease symptoms are reported to manifest as systemic and LRT disease [21] . However, on occasion, even LRT specimens from MERS cases may initially be negative, only to later become positive by RT-PCR [95] . This may be due to poor sampling when a cough is absent or non-productive or because the viral load is low [95] . Despite this both the largest human MERS-CoV studies [32, [96] [97] [98] and smaller ones [22, 25, 99] , use samples from the URT. It is then noteworthy that one study reported an association between higher loads in the URT and worse clinical outcome including intensive care and death [94] . At writing, no human data exist to define whether the virus replicates solely or preferentially in the LRT or URT, or replicates in other human tissues in vivo although MERS-CoV RNA has been detected from both the URT and LRT in a macaque monkey model [100] .The distribution of DPP4 in the human upper airways is also not well described. Individual human case studies report long periods of viral shedding, sometimes intermittently and not necessarily linked to the presence of disease symptoms. [25, 69, 99, 101] In one instance, a HCW shed viral RNA for 42 days in the absence of disease [99] . It is an area of high priority to better understand whether such cases are able to infect others. Over three quarters of MERS cases shed viral RNA in their LRT specimens (tracheal aspirates and sputum) for at least 30 days, while only 30 % of contacts were still shedding RNA in their URT specimens [91, 102] . In the only study to examine the effect of sample type on molecular analysis, 64 nasopharyngeal aspirates (NPA; an URT sample), 30 tracheal aspirates, 13 sputa and three BAL were examined. The tracheal aspirates and BAL returned the highest viral load values followed by NPA and sputum. Unsurprisingly, higher viral loads generally paralleled whole genome sequencing and culture success and, in NPA testing, were significantly correlated with severe disease and death [49, 94, 103] . This study demonstrated the importance of LRT sampling for whole genome sequencing. When tested, samples positive for MERS-CoV are often negative for other pathogens [2, 25, 93, 104] . However, many studies make no mention of additional testing for endemic human respiratory viruses [21, 23, 73, 105] . When viruses are sought, they have included human herpesvirus (HHV), rhinoviruses (HRV), enteroviruses (EV), respiratory syncytial virus (RSV), parainfluenzavirus types 1, 2 and 3 (PIVs),influenzaviruses (IFVs), endemic HCoVs, adenoviruses (AdVs) metapneumovirus (MPV) and influenza A\H1N1 virus; co-detections with MERS-CoV have been found on occasion [2, 22, 37, 69, 97] . Bacterial testing is sometimes included (for example, for Legionella and Pneumococcus) but the impact of bacterial co-presence is also unclear [22, [104] [105] [106] . Further testing of the LRT sample from the first MERS case used IFA to screen for some viruses (negative for IFV, PIVs, RSV and AdVs) and RT-PCR for others (negative for AdV, EVs, MPV and HHVs) [18] . RT-PCR also detected MERS-CoV. The WHO strongly recommends testing for other respiratory pathogens [53] but with this recommendation often discounted, there are limited data to address the occurrence and impact of co-infections or alternative viral diagnoses among both MERS cases and their contacts. Little is known of other causes of MERS-like pneumonia in the KSA or of the general burden of disease due to the known classical respiratory viruses. Testing of adult pilgrims performing the Hajj in 2012 to 2014 has not detected any MERS-CoV. In 2012, nasal swabs from 154 pilgrims collected prior to leaving for or departing from the KSA were tested [47] . In 2013, testing was significantly scaled up with 5,235 nasopharyngeal swabs from 3,210 incoming pilgrims and 2,025 swabs from outgoing pilgrims tested [98] . It should be noted that most pilgrims arrived from MERS-free countries. A further 114 swabs were taken from pilgrims with influenza-like illness [96, 107] . In earlier Hajj gatherings, it was found that influenza viruses circulated widely, whilst other viruses, often rhinoviruses, circulated more selectively, interpreted as indicating their importation along with foreign pilgrims. [107] [108] [109] Over time, increased influenza vaccination has been credited for a fall in the prevalence of influenza like illnesses among Hajj pilgrims. [110] A LRT sample is often not collected for these studies [98, 107, 109] , so false negative findings are a possibility although little is known about the initial site of MERS-CoV infection and replication; it may have been assumed it was the LRT because disease was first noticed there but the URT may be the site of the earliest replication. In Jeddah between March and July 2014 (hereafter called the Jeddah-2014 outbreak; Fig. 3 ), there was a rapid increase in MERS cases, accompanied by intense screening; approximately 5,000 samples from in and around the region were tested in a month yielding around 140 MERS-CoV detections (~3 % prevalence) [111] . Among 5,065 individuals sampled and tested across the KSA between October 2012 and September 2013,108 (2.1 %) detections were made in a hospital-centric population which included hospitalized cases (n = 2,908; 57.4 %), their families (n = 462; 9.1 %) and associated HCWs (n = 1,695; 33.5 %) [32] . Among the detections, 19 (17.8 %) were HCWs and 10 (9.3 %) were family contacts [32] . The 2-3 % prevalence of active MERS-CoV infections is not dissimilar to the hospital-based prevalence of other human CoVs. [112] However, the proportion of deaths among those infected with MERS-CoV is much higher than that known for the HCoVs NL63, HKU1, 229E or OC43 in other countries, and even above that for SARS-CoV; it is not a virus that could reasonably be described as a "storm in a teacup". It is the low transmission rate that has prevented worldwide spread, despite many "opportunities". Very early in the MERS outbreak, some animals were highly regarded as either the reservoir or intermediate host(s) of MERS-CoV with three of the first five cases having contact with DCs [73, 113, 114] . Today, animal MERS-CoV infections must be reported to the world organization for animal health as an emerging disease [115] . A summary of the first MERS cases reported by the WHO defined animal contact with humans as being direct and within 10 days prior to symptom onset [20] . This definition made no specific allowance for acquisition from DCs through a droplet-based route, which is very likely route for acquisition of a virus that initially and predominantly causes respiratory disease [23] . Camels are known to produce high levels of MERS-CoV RNA in their URT and lungs [116] . Providing support for a droplet transmission route and perhaps indicating the presence of RNA in smaller, drier droplet nuclei, MERS-CoV RNA was identified in a high volume air sample collected from a barn housing an infected DC [117] . The precise source from which humans acquire MERS-CoV remains poorly studied but it seems likely that animal and human behavioural factors may play roles (Fig. 3) [118] . These factors may prove important for human cases who do not describe any DC contact [119] nor any contact with a confirmed case. Whether the WHO definition of animal contact is sufficient to identify exposure to this respiratory virus remains unclear. Wording focuses on consumption of DC products but does not specifically ascribe risk to a droplet route for acquisition of MERS-CoV from DC [120] . Some MERS patients are listed in WHO disease notices as being in proximity to DCs or farms, but the individuals have not described coming into contact with the animals. No alternative path for acquiring infection is reported in many of these instances. What constitutes a definition of "contact" during these interviews has been defined for one study [72] . Despite this lack of clarity, the WHO consider that evidence linking MERS-CoV transmission between DCs to humans is irrefutable (Fig. 4) [120] . The possibility that bats were an animal host of MERS-CoV was initially widely discussed because of the existing diversity of coronaviruses known to reside among them [121] [122] [123] [124] . Conclusive evidence supporting bats as a source for human infections by MERS-CoV has yet to be found, but bats do appear to host ancestral representatives [53, 125] . However, these are not variants of the same virus nor always within the same phylogenetic lineage as MERS-CoV; they are each a genetically distinct virus. Bat-to-human infection by MERS-CoV is a purely speculative event. The only piece of MERS-CoV-specific evidence pointing to bats originates from amplification of a 190 nt fragment of the RNAdependent RNA polymerase gene of the MERS-CoV genome, identified in a faecal pellet from an insectivorous Emballonuridae bat, Taphozous perforatus found in Bisha, the KSA [121] . While very short, the sequence of the fragment defined it as a diagnostic discovery. Subsequently a link to DCs was reported [85] and that link has matured into a verified association [38, 126] (Fig. 4) . (See figure on previous page.) Fig. 3 Monthly detections of MERS-CoV (blue bars) and of cases who died (red bars) with some dates of interest marked for 2012 to 4 th September 2015. An approximation of when DC calving season [128] and when recently born DCs are weaned is indicated. Spring (green) and summer (orange) in the Arabian Peninsula are also shaded. Note the left-hand y-axis scale for 2014 and 2015 which is greater than for 2012/13. Sources of these public data include the WHO, Ministries of Health and FluTrackers [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] . Modified and reprinted from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] DCs, which make up 95 % of all camels, have a central presence in the Arabian Peninsula where human-DC contact ranges from little to close [119] . Contact may be commonplace and could occur in variety of ways (Fig. 4a) . There are several large well-attended festivals, races, sales and parades which feature DCs and DCs are also kept and bred close to populated areas in the KSA [127, 128] . DC milk and meat are widely consumed and the older DC is an animal of ritual significance after the Hajj pilgrimage [129] . However, MERS-CoV infection frequency is reportedly much lower than is the widespread and frequent habit of eating, drinking and preparing DC products. Daily ingestion of fresh unpasteurized DC milk is common among the desert Bedouin and many others in the KSA. DC urine is also consumed or used for supposed health benefits. Despite camel butchery being a local occupation, neither butchers nor other at-risk groups are identifiable among MERS cases; this may simply be a reporting issue rather than an unexplainable absence of MERS. A small case-control study published in 2015 identified direct DC contact, and not ingestion of products, to be associated with onset of MERS [38] . The first sero-survey of livestock living in the Middle East region was conducted during 2012-2013 [85] . DCs were sampled from a mostly Canary Island-born herd and from Omani DCs (originally imported from the Horn of Africa) [85] . A neutralising antibody assay found only 10 % of strongly seropositive Canary Island [5] . b Camel-to-human infections appear to be infrequent, while human-to-human spread of infection is regularly facilitated by poor IPC in healthcare settings where transmission is amplified, accounting for the bulk of cases. There are human MERS cases that do not fall into either category of source and it is unclear if these acquired infection through some entirely separate route, or from cases that escaped diagnosis. c Hypothetical ways in which subclinical (when infection may not meet a previously defined clinical threshold of signs and/or symptoms) or asymptomatic (no obvious signs or measured, noticed or recalled symptoms of illness) MERS-CoV infection may be implicated in transmission DC sera could neutralise MERS-CoV while all Omani DC sera had high levels of specific MERS-CoV neutralizing antibody [85] . This indicated that DCs had in the past been infected by MERS-CoV, or a very similar virus. Since this study, a host of peer-reviewed reports have looked at both DCs and other animals, and the possibility that they may host MERS-CoV infection. Seropositive DCs have been found throughout the Arabian Peninsula including Oman, the KSA, Qatar, Jordan, the United Arab Emirates (UAE), Kuwait as well as Sudan, Somalia, Egypt, Tunisia, Nigeria, Kenya and Ethiopia in Africa and the Canary Islands [85, [130] [131] [132] [133] [134] . Other animals tested include sheep, cows, pigs, horses, donkeys, mules, birds, water buffalo, goats, Bactrian camels, llamas and guanaco (south American camelids) but none had detectable neutralising antibody against MERS-CoV [4, 74, 78, 85, 86, 135, 136] . No virology or serology studies of human samples from areas in Africa where there are camels with a history of MERS-CoV have been reported to date. However,an absence of unexplained pneumonia that may be attributable to MERS-CoV infection may not signal the absence of virus among humans in each country but simply reflect a lack of expensive epidemiology studies conducted by resource-poor countries. It is thus unclear whether MERS-CoV, or an antigenically related CoV, is an unrecognized pathogen in these regions, perhaps circulating for even longer than it has been known in the Arabian Peninsula [133] . MERS-CoV RNA has also been detected in DC samples, and recovery of infectious virus has also been achieved from DC samples [4, 77, 117, 132, [137] [138] [139] [140] [141] . From some of these, full or majority length genomes of MERS-CoV have been sequenced [77, 137, 138] . DC versions of MERS-CoV were found to be as similar to each other, as were variants detected from different humans over time and across distance. Antibody screening assays have also detected crossreactive antibodies in sera. These were identified as such by screening sera against similar viruses, for example BCoV or HCoV-OC43 (as an antigenic facsimile for BCoV). It is possible that other MERS-CoV-like viruses also reside within DCs, but this does not detract from the definitive finding of MERS-CoV genetic sequences in both DCs and humans [117, 142, 143] . Screening studies have shown that juvenile DCs are more often positive for virus or viral RNA while older DCs are more likely to be seropositive and RNA or virus negative [76, 77, 144] . In adult DCs, MERS-CoV RNA has been detected among animals with pre-existing antibody, suggesting re-infection is possible [77, 144] . Viral loads among positive DCs can be very high [4, 76, 77, 139, 144] and DCs have been found positive both when ill with URT respiratory signs [77, 117, 142, 145] or when apparently healthy [137] . These findings indicate DCs host natural MERS-CoV infections. Furthermore, stored DC sera have revealed signs of MERS-CoV in DCs which date back over three decades (the earliest collected in 1983) [4, 133, 135] . Older sera have not been tested and so precisely how long DCs have been afflicted by MERS-CoV, whether the virus is enzootic among them, introduced to them decades or centuries ago from bats in Africa or the Arabian Peninsula, or they are the subject of regular but short-lived viral incursions from an as yet unknown host, cannot be answered. Researchers sought to determine a direction for infection; were DCs transmitting virus to humans or were humans infecting DCs? At a Qatari site, a farm owner and his employee became ill in mid-October 2013 and tested positive for MERS-CoV RNA in a sputum and throat swab sample, respectively. RT-rtPCRs found MERS-CoV RNA in 11 of 14 positive DC nasal swabs at the farm; six (43 %) positive by two or more assays [138] . The results indicated a recent outbreak had occurred in this herd; the first indication of MERS-CoV RNA found within DCs with a temporal association to human infections. Three positive DC samples were confirmed by sequencing a 358 nt portion of the spike gene; these sequences were identical to each other, again with close homology to other human and DC MERS-CoV sequences [138] . The DCs and human contacts yielded ORF1a and ORF4b sequences differing by only a single nucleotide each, clustering closely with the Hafr-Al-Batin_1_2013 variant [138] . Subsequent case studies found evidence of a concurrent human and DC infection and the direction of that infection was inferred to be from the ill DCs and to their human owners [117, 142, 146] . Partial genome sequences indicated that a human and a MERS-CoV RT-rtPCR positive DC had been infected by a variant of the same virus, harbouring the same distinct pattern of nucleotide polymorphisms. [142] All nine DC in the owner's herd, serially sampled, reacted in a recombinant S1 antigen ELISA, with the two animals that had been RT-rtPCR positive showing a small, verifiable rise in antibody titre [142] . A rise in titre theoretically begins 10 to 21 days after DC infection [142] . The authors suggested that the rise in titre in DC sera which occurred alongside a declining RNA load, while the patient was actively ill and hospitalized, indicated that the DCs were infected first followed by the owner [117, 142] . BCoV antibodies were also present, and rising in one of the two RT-rtPCR positive animals but no animal's antibodies could neutralise BCoV infection [142] . Camel calving season occurs in the winter months (between late October and late February; Fig. 3 ) and this may be a time when there is increased risk to humans of spill-over due to new infections among naïve DC populations [128] . What role maternal camel antibody might play in delaying infection of calves remains unknown [128, 142] . Juvenile DCs appear to host active infection more often than adult DCs and thus the sacrificial slaughter of DCs, which must be five years of age or older (termed a thane), may not be accompanied by significant risk of exposure to infection. In contrast to earlier results, slaughterhouse workers who kill both younger and older DCs, may be an occupational group with significantly higher incidence of seropositivity to MERS-CoV when animals have active MERS-CoV infections [129, 139, [147] [148] [149] . Expanded virological investigations of African DCs may lead to more seropositive animals and geographic areas in which humans may be at risk. It is possible that there are areas where humans already harbour MERS-CoV infections that have not been identified because of an absence of laboratory surveillance. Virological investigations of bats may lead to findings of ancestral viruses and viral 'missing links' and identifying any other animal sources of zoonotic spread is important to inform options for reducing human exposures [56, 76] . Infectious MERS-CoV added to DC, goat or cow milk and stored at 4°C could be recovered at least 72 h later and, if stored at 22°C, recovery was possible for up to 48 h [150] . MERS-CoV titre decreased somewhat when recovered from milk at 22°C but pasteurization completely ablated MERS-CoV infectivity [150] . In a subsequent study, MERS-CoV RNA was identified in the milk, nasal secretion and faeces of DCs from Qatar [151] . A single study has examined the ability of MERS-CoV to survive in the environment [150] . Plastic or steel surfaces were inoculated with 10 6 TCID 50 of MERS-CoV at different temperature and relative humidity (RH) and virus recovery was attempted in cell culture. At high ambient temperature (30°C) and low RH (30 %) MERS-CoV remained viable for 24 h [150] . By comparison, a well known and efficently transmitted respiratory virus, influenza A virus, could not be recovered in culture beyond four hours under any conditions [150] . Aerosol experiments found MERS-CoV viability only decreased 7 % at low RH at 20°C. In comparison, influenza A virus decreased by 95 % [150] . MERS-CoV survival is inferior to that previously demonstrated for SARS-CoV [152] . For context, pathogenic bacteria can remain viable and airborne for 45 min in a coughed aerosol and can spread 4 m. MERS-CoV's ability to remain viable over long time periods gives it the capacity to thoroughly contaminate a room's surfaces when occupied by an infected and symptomatic patient [153] . Whether MERS-CoV can remain adrift and infectious for extended periods (truly airborne) remains unknown. Such findings expand our understanding of the possibilities for droplets to transmit respiratory viruses in many settings, including hospital waiting rooms, emergency departments, treatment rooms, open intensive care facilities and private patient rooms. The nature and quality of air exchange, circulation and filtration are important variables in risk measurement and reduction as is the use of negative pressure rooms to contain known cases. Droplet spread between humans is considered the mechanism of human-to-human transmission and the need for droplet precautions was emphasized after the Al-Ahsa hospital, the KSA and the South Korean outbreaks [21, 23, 154, 155] . By extrapolation, aerosol-generating events involving DCs (urination, defecation, and preparation and consumption of DC products) should be factored into risk measurement and reduction efforts and messaged using appropriate context. The provision of evidence supporting the best formulation of personal protective equipment to be worn by HCWs who receive, manage or conduct procedures on infectious cases remains a priority. MERS-CoV was found and characterized because of its apparent association with severe, and therefore more obvious, illness in humans; we were the canaries in the coal mine. Sero-assays and prospective cohort studies have yet to determine the extent to which milder or asymptomatic cases contribute to MERS-CoV transmission chains. However, transmission of MERS-CoV is defined as sporadic (not sustained), intra-familial, often healthcare associated, inefficient and requiring close and prolonged contact [22, 31, 63, 93, 97, 102, 156] In a household study, 14 of 280 (5 %) contacts of 26 MERS-CoV positive index patients were RNA or antibody positive; the rate of general transmission, even in outbreaks is around 3 % [31] . It seems that the majority of human cases of MERS-CoV, even when numbers appear to increase suddenly, do not readily transmit to more than one other human so to date, the localized epidemic of MERS-CoV has not been self-sustaining [157] [158] [159] [160] [161] . That is to say, the basic reproduction number (R 0 ) -the average number of infections caused by one infected individual in a fully susceptible populationhas been close to one throughout various clusters and outbreaks. If R 0 was greater than 1, a sustained increase in case numbers would be expected. Some R o calculations may be affected by incomplete case contact tracing, limited community testing and how a case is defined. That MERS has had a constant presence in the Arabian Peninsula since 2012 is due to ongoing, sporadic spill-over events from DCs amplified by poorly controlled hospital outbreaks. The first known MERS human-to-human transmission event was one characterized by acute LRT disease in a healthcare setting in Jordan. In stark contrast, a sero-survey of HCW who were sometimes in close and prolonged contact with the first, fatal MERS-CoV case in 2012 [162] , found none of the HCW had seroconverted four months later, despite an absence of eye protection and variable compliance with required PPE standards [162] . Early on in the MERS story, samples for testing were mostly collected from patients with severe illness and not those with milder acute respiratory tract infections. Contacts of confirmed MERS cases were often observed for clinical illness, but not tested. These omissions may have confounded our understanding of MERS-CoV transmission and biased early data towards higher numbers of seriously ill and hospitalized patients, inflating the apparent proportion of fatal cases. Case-control studies were not a focus. As testing paradigms changed and contacts were increasingly tested, more asymptomatic and mild infections were recognized [163] . A rise in the cases termed asymptomatic (which enlarge the denominator for calculations of the proportion of fatal cases, defined in [164] ) resulted in a drop in the proportion of fatal cases during the Jeddah-2014 outbreak. Historically, such rises are consistent with changing definitions and laboratory responses and clinical management of a newly discovered virus infection that was first noted only among the severely ill. Upon follow-up, over three-quarters of such MERS-CoV RNA positive people did recall having one or more symptoms at the time, despite being reported as asymptomatic [165] raising some question over the reliability of other reported data. The proportion of fatal MERS cases within the KSA compared to outside the KSA, as well as the age, and sex distribution change in different ways when comparing MERS outbreaks. Approximately 43 % of MERS cases (549 of 1277) in the KSA were fatal betwen 2012 and December 2015 while 21 % (72 of 330) died among those occurring outside of the KSA. The total number of male cases always outnumber females and the proportion of male deaths is always greater than the proportion of females who die. However the proportion of male deaths from total males with MERS is a similar figure to that for females. In the KSA, there is a greater proportion of younger males among cases and deaths than were observed from the 2015 South Korean or the Jeddah-2014 outbreaks (Additional file 2: Figure S2 ). Why these aspects have differed may be due to differences in the time to presentation and diagnosis, the nature and quality of supportive care, the way a person became infected (habits, exposure to a human or zoonotic source, viral load, route of infection) or the extent to which different populations are burdened by underlying diseases [40] . As a group, HCWs comprised 16 % of MERS cases in the KSA and South Korea. It is apparent that the weekly proportion of infected HCWs increases alongside each steep rise in overall detections (Fig. 5) . In May 2013, the WHO published guidelines for IPC during care of probable or confirmed cases of MERS-CoV infection in a healthcare setting [166] . This is explainable because to date, each case rise has been intimately associated with healthcare-facility related outbreaks [118] . These rises in MERS-CoV detections can decrease the average age during each event because HCWs are usually younger than inpatients with MERS. Healthcare facilities have been a regular target for suggested improvements aimed at improving infection prevention and control (IPC) procedures [115, 118] . Most of the analysis of MERS-CoV genetics has been performed using high throughput or "deep" sequencing methods for complete genome deduction [167] [168] [169] . MERS-CoV was the first subject of such widespread use of deep sequencing to study an emerging viral outbreak with global reach. The technique can produce genomic [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] length coverage in a single experiment with highly repetitious measurement of each nucleotide position [52, 140] . Despite assays having been published early on, subgenomic sequencing, once the mainstay of viral outbreak studies, has less often been published during MERS-CoV characterization [48] . As more genomes from both humans and DCs have been characterized, two clades have become apparent; A and B (Fig. 6) . Clade A contains only human-derived MERS-CoV genomes from Jordan, while Clade B comprises the majority of human and camel genomes deduced thus far [168] . Two studies during 2015, one looking at Jeddah-2014 MERS-CoV variants and another looking at a variant exported from South Korea to China, have now identified signs of genetic recombination among MERS-CoV variants. While human and camel whole genome sequences have retained >99 % identity with each other, members of genetically distinct lineages can and do swap genetic material when suitable conditions and coinfections co-occur [170] [171] [172] . Shared identity implies that the major source for human acquisition is the DC, rather than another animal, although more testing of other animal species is needed to confirm that conclusion. Over a month, a DC virus sequenced on different occasions did not change at all indicating a degree of genomic stability in its host, supporting that DCs are the natural, rather than intermediate, host for the MERS-CoV we know today [77] . To date, recombination has been localised to breakpoints near the boundary between ORF1a and ORF1b regions, within the spike gene [170] and in the ORF1b region (Fig. 2) [172] . It is not unexpected that recombination should occur since it is well known among other CoVs [124] and because the majority of MERS-CoV whole genomes collected from samples spanning three years (2012-2015) and from humans, camels and different countries have shown close genetic identity to each other, with just enough subtle variation to support outbreak investigations so long as whole genome sequencing is applied [52, 77, 135, 138, 168, [173] [174] [175] . Changes in genome sequence may herald alterations to virus transmissibility, replication, persistence, lethality or response to future drugs. If we have prior knowledge of the impact of genetic changes because of thorough characterization studies, we can closely Fig. 6 The genetic relationship between MERS-CoV nucleotide sequences (downloaded from GenBank using the listed accession numbers and from virological.org [212] ). This neighbour joining tree was created in MEGA v6 using an alignment of human and DCderived MERS-CoV sequences (Geneious v8.1 [211] ). Clades are indicated next to dark (Clade A) or pale (Clade B) blue vertical bars. Camel icons denote genomes from DCs. Healthcare or community outbreaks are boxed and labelled using previously described schemes [212, 213] monitor the genomic regions and better understand any changes in transmission or disease patterns as they occur. Genetic mutations noted during the largest of human outbreaks, Jeddah-2014, did not impart any major replicative or immunomodulatory changes when compared to earlier viral variants in vitro [156, 176] . However, we understand very little of the phenotypic outcomes that result from subtle genetic change in MERS-CoV genomes. To date no clinical relevance or obvious in vivo changes to viral replication, shedding or transmission has been reported or attributed to mutations or to new recombinant viruses [156] . But vigilance and larger, more contemporary and in vivo studies are needed. Genome sequence located to a distinct clade were identified from an Egyptian DC that was probably imported from Sudan. This does not fit into either of the current clades [125, 168, 177] . A virus sequenced from a Neoromicia capensis bat was more closely related to MERS-CoV than other large bat-derived sequences had been to that point, but the genome of a variant of a MERS-CoV has yet to be discovered and deduced from any bat [125] . Analyses of MERS-CoV genomes have shown that most single nucleotide differences among variants were located in the last third of the genome (Fig. 2) , which encodes the spike protein and accessory proteins [168] . At least nine MERS-CoV genomes contained amino acid substitutions in the receptor binding domain (RBD) of the spike protein and codons 158 (N-terminal region), 460 (RBD), 1020 (in heptad repeat 1), 1202 and 1208 bear investigation as markers of adaptive change [140, 169] . The spike protein had not changed in the recombinant MERS-CoV genome identified in China in 2015 but was reported to have varied at a higher rate than that for complete MERS-CoV genomes, among South Korean variants [172, 178] . This highlights that subgenomic regions may not always contain enough genetic diversity to prove useful for differentiating viral variants. Despite this, one assay amplifying a 615 nucleotide fragment of the spike S2 domain gene for Sanger sequencing agreed with the results generated by the sequencing of a some full genomes and was useful to define additional sequence groupings [177] . Genomic sequence can also be used to define the geographic boundaries of a cluster or outbreak and monitor its progress, based on the similarity of the variants found among infected humans and animals when occurring together, or between different sites and times (Fig. 6 ) [169] . This approach was employed when defining the geographically constrained MERS hospital outbreak in Al-Ahsa, which occurred between 1 st April and 23 rd May 2013, as well as clusters in Buraidah and a community outbreak in Hafr Al-Batin, the KSA. Genomic sequencing identified that approximately 12 MERS-CoV detections from a community outbreak in Hafr Al-Batin between June and August 2013 may have been triggered by an index case becoming infected through DC contact [175] . Sequencing MERS-CoV genomes from the 2013 Al-Ahsa hospital outbreak indicated that multiple viral variants contributed to the cases but that most were similar enough to each other to be consistent with human-tohuman transmission. Molecular epidemiology has revealed otherwise hidden links in transmission chains encompassing a period of up to five months [179] . However, most outbreaks have not continued for longer than two to three months and so opportunities for the virus to adapt further to humans through co-infection and sustained serial passage have been rare [169] . In Riyadh-2014, genetic evidence supported the likelihood of multiple external introductions of virus, implicating a range of healthcare facilities in an event that otherwise looked contiguous [23, 168, 179] . Riyadh is a nexus for camel and human travel and has had more MERS cases than any other region of the KSA to date but also harbours a wide range of MERS-CoV variants [128, 167, 179] . However the South Korean outbreak originated from a single infected person, resulting in three to four generations of cases [180, 181] . Studies of this apparently recombinant viral variant did not find an increased evolutionary rate and no sign of virus adaptation thus the outbreak seems to have been driven by circumstance rather than circumstance together with mutation [181] . For many MERS cases detected outside the Arabian Peninsula, extensive contact tracing has been performed and the results described in detail. Contact tracing is essential to contain the emergence and transmission of a new virus and today it is supported by molecular epidemiology. Although it is an expensive and time consuming process, contact tracing can identify potential new infections and through active or passive monitoring, react more rapidly if disease does develop. Results of contact tracing to date have found that onward transmission among humans is an infrequent event. For example, there were 83 contacts, both symptomatic and asymptomatic, of a case treated in Germany who travelled from the UAE but no sign of virus or antibody were found in any of them [73] . The very first MERS case had made contact with 56 HCWs and 48 others, but none developed any indication of infection [162] . In a study of 123 contacts of a case treated in France, only seven matched the definition for a possible case and were tested; one who had shared a 20 m 2 hospital room while in a bed 1.5 m away from the index case for a prolonged period was positive [26] . None of the contacts of the first two MERS cases imported into the USA in 2014 contained any MERS-CoV footprint [182] and none of the 131 contacts of two travellers returning to the Netherlands developed MERS-CoV antibodies or tested RNA positive [25, 183] . Analyses of public data reveal many likely instances of nosocomial acquisition of infection in the Arabian Peninsula and these data may be accompanied by some details noting contact with a known case or facility. One example identified the likely role of a patient with a subclinical infection, present in a hospital during their admission for other reasons, as the likeliest index case triggering a family cluster [93] . Contact tracing was a significant factor in the termination of a 2015 outbreak involving multiple South Korean hospitals [184] . Such studies demonstrate the necessity of finding and understanding a role for mild and asymptomatic cases, together with restricting close contact or prolonged exposure of infected people to others, especially older family members and friends with underlying disease (Fig. 4c) . The hospital-associated outbreak in Jeddah in 2014 was the largest and most rapid accumulation of MERS-CoV detections to date. The greatest number of MERS-CoV detections of any month on record occurred in Jeddah in April. The outbreak was mostly (>60 % of cases) associated with human-to-human spread within hospital environments and resulted from a lack of, or breakdown in, infection prevention and control [37, 185, 186] . A rise in fatalities followed the rapid increase in case numbers. In 2015 two large outbreaks occurred. South Korea was the site of the first large scale outbreak outside the Arabian Peninsula and produced the first cases in both South Korea and China, occurring between May and July 2015. This was closely followed by a distinct outbreak in Ar Riyad province in the KSA which appeared to come under control in early November. After staying in Bahrain for two weeks, a 68 year old male (68 M) travelled home to South Korea via Qatar, arriving free of symptoms on the 4 th May 2015 [187] . He developed fever, myalgia and a cough nearly a week later (11 th ). He visited a clinic as an outpatient between the 12 th and 15 th of May and was admitted to Hospital A on the 15 th [188] . He was discharged from Hospital A on the 17 th then visited and was admitted to the emergency department of Hospital B on the 18 th . During this second stay, a sputum sample was taken and tested positive for MERS-CoV on the 20 th [187, 188] , triggering transfer to the designated isolation treatment facility. Over a period of 10 days, the index case was seen at three different hospitals, demonstrating a key feature of "hospital shopping" that shaped the South Korean outbreak. Approximately 34 people were infected during this time [187] . In total 186 cases were generated in this outbreak, all linked through a single transmission chain to 68 M; 37 cases died [189] . In South Korea, the national health insurance system provides for relatively low cost medical care, defraying some costs by making family members responsible for a portion of the ministration of the sick, resulting in them sometimes staying for long periods in the rooms that often have more than four beds in them [24] . Other factors thought to have enabled this outbreak included unfamiliarity of local clinicians with MERS, ease with which the public can visit and be treated by tertiary hospitals, the custom of visiting sick friends and relatives in hospitals, the hierarchical nature of Korean society, crowded emergency rooms, poor IPC measures, a lack of negative pressure isolation rooms and poor inter-hospital communication of patient disease histories [24, [190] [191] [192] . All of the reported transmission occurred across three or four generations and apart from one unknown source, were all hospital-acquired [24, 120, 181, [193] [194] [195] . Few clinical details about these cases have been reported to date and detail on transmission and contact tracing is minimal. The hospitals involved were initially not identified, governmental guidance and actions produced confusing messages and there was very limited communication at all early on which resulted in unnecessary concern, distrust and a distinct economic impact [191, [196] [197] [198] . Early in the outbreak, a infected traveller, the son of an identified case in South Korea, passed through Hong Kong on his way to China where he was located, isolated and cared for in China [91, 199, 200] . No contacts became ill. The outbreak was brought under control in late July/ early August [201] after improved IPC measures were employed, strong contact tracing monitoring and quarantine, expanded laboratory testing, hospitals were better secured, specialized personnel were dispatched to manage cases and international cooperation increased [202, 203] . A review of public data showed that, as for MERS in the KSA, older age and the presence of underlying disease were significantly associated with a fatal outcome in South Korea. [40] Even though R 0 is <1, super-spreading events facilitated by circumstances created in healthcare settings and characterized by cluster sizes over 150, such as this one, are not unexpected from MERS-CoV infection [204] . The dynamic of an outbreak depends on the R 0 and an individual's viral shedding patterns, contact type and frequency, hospital procedures and population structure and density [204] . In the region of Ar Riyad, including the capital city of Riyadh, a hospital based cluster began, within a single hospital, from late June 2015 [205] . By mid-September there had been approximately170 cases reported but the outbreak appeared to been brought under control in November. It became apparent early on that MERS-CoV spread relatively ineffectively from human-to-human. Despite ongoing and possibly seasonal introduction of virus to the human population via infected DCs and perhaps other animals yet to be identified, the vast majority of MERS-CoV transmission has occurred from infected to uninfected humans in close and prolonged contact through circumstances created by poor infection control in health care settings. This opportunistic virus has had its greatest impact on those with underlying diseases and such vulnerable people, sometimes suffering multiple comorbidities, have been most often associated with hospitals, creating a perfect storm of exposure, transmission and mortality. It remains unclear if this group are uniquely affected by MERS-CoV or if other respiratory virus infections, including those from HCoVs, produce a similarly serious impact. In South Korea, a single imported case created an outbreak of 185 cases and 36 deaths that had a disproportionate impact on economic performance, community behaviour and trust in government and the health care system. Household human-to human transmission occurs but is also limited. Educational programs will be essential tools for combatting the spread of MERS-CoV both within urban and regional communities and for the health care setting. Vigilance remains important for containment since MERS-CoV is a virus with a genetic makeup that has been observed for only three years and is not stable. Among all humans reported to be infected, nearly 40 % have died. Continued laboratory testing, sequencing, analysis, timely data sharing and clear communication are essential for such vigilance to be effective. Global alignment of case definitions would further aid accurate calculation of a case fatality ratio by including subclinical case numbers. Whole genome sequencing has been used extensively to study MERS-CoV travel and variation and although it remains a tool for experts, it appears to be the best tool for the job. MERS and SARS have some clinical similarities but they also diverge significantly [206] . Defining characteristics include the higher PFC among MERS cases (above 50 % in 2013 and currently at 30-40 %; well above the 9 % of SARS) and the higher association between fatal MERS and older males with underlying comorbidities. For the viruses, MERS-CoV has a broader tropism, grows more rapidly in vitro, more rapidly induces cytopathogenic change, triggers distinct transcriptional responses, makes use of a different receptor, induces a more proinflammatory state and has a delayed innate antiviral response compared to SARS-CoV. There appears to be a 2-3 % prevalence of MERS-CoV in the KSA with a 5 % chance of secondary transmission within the household. There is an increased risk of infection through certain occupations at certain times and a much greater chance for spread to other humans during circumstances created by humans, which drives more effective transmission than any R 0 would predict on face value. Nonetheless, despite multiple mass gatherings that have afforded the virus many millions of opportunities to spread, there have remarkably been no reported outbreaks of MERS or MERS-CoV during or immediately after these events. There is no evidence that MERS-CoV is a virus of pandemic concern. Nonetheless, hospital settings continue to describe MERS cases and outbreaks in the Arabian Peninsula. As long as we facilitate the spread of MERS-CoV among our most vulnerable populations, the world must remain on alert for cases which may be exported more frequently when a host country with infected camel reservoirs is experiencing human clusters or outbreaks. The MERS-CoV appears to be an enzootic virus infecting the DC URT with evidence of recent genetic recombination. It may once have had its origins among bats, but evidence is lacking and the relevance of that to today's ongoing epidemic is academic. Thanks to quick action, the sensitive and rapid molecular diagnostic tools required to achieve rapid and sensitive detection goal have been in place and made widely available since the virus was reported in 2012. RT-PCR testing of LRT samples remains the gold standard for MERS-CoV confirmation. Serological tools continue to emerge but they are in need of further validation using samples from mild and asymptomatic infections and a densely sampled cohort study to follow contacts of new cases may address this need. Similarly, the important question of whether those who do shed MERS-CoV RNA for extended periods are infectious while appearing well, continues to go unanswered. It is even unclear just how many 'asymptomatic' infections have been described and reported correctly which in turn raises questions about the reliability of other clinical data collection to date. While the basic virology of MERS-CoV has advanced over the course of the past three years, understanding what is happening in, and the interplay between, camel, environment and human is still in its infancy. Additional file 1: Figure S1 . The

Where was the the case first to be publicly reported was from ?

MERS coronavirus: diagnostics, epidemiology and transmission https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4687373/ SHA: f6fcf1a99cbd073c5821d1c4ffa3f2c6daf8ae29 Authors: Mackay, Ian M.; Arden, Katherine E. Date: 2015-12-22 DOI: 10.1186/s12985-015-0439-5 License: cc-by Abstract: The first known cases of Middle East respiratory syndrome (MERS), associated with infection by a novel coronavirus (CoV), occurred in 2012 in Jordan but were reported retrospectively. The case first to be publicly reported was from Jeddah, in the Kingdom of Saudi Arabia (KSA). Since then, MERS-CoV sequences have been found in a bat and in many dromedary camels (DC). MERS-CoV is enzootic in DC across the Arabian Peninsula and in parts of Africa, causing mild upper respiratory tract illness in its camel reservoir and sporadic, but relatively rare human infections. Precisely how virus transmits to humans remains unknown but close and lengthy exposure appears to be a requirement. The KSA is the focal point of MERS, with the majority of human cases. In humans, MERS is mostly known as a lower respiratory tract (LRT) disease involving fever, cough, breathing difficulties and pneumonia that may progress to acute respiratory distress syndrome, multiorgan failure and death in 20 % to 40 % of those infected. However, MERS-CoV has also been detected in mild and influenza-like illnesses and in those with no signs or symptoms. Older males most obviously suffer severe disease and MERS patients often have comorbidities. Compared to severe acute respiratory syndrome (SARS), another sometimes- fatal zoonotic coronavirus disease that has since disappeared, MERS progresses more rapidly to respiratory failure and acute kidney injury (it also has an affinity for growth in kidney cells under laboratory conditions), is more frequently reported in patients with underlying disease and is more often fatal. Most human cases of MERS have been linked to lapses in infection prevention and control (IPC) in healthcare settings, with approximately 20 % of all virus detections reported among healthcare workers (HCWs) and higher exposures in those with occupations that bring them into close contact with camels. Sero-surveys have found widespread evidence of past infection in adult camels and limited past exposure among humans. Sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics have been available almost from the start of the emergence of MERS. While the basic virology of MERS-CoV has advanced over the past three years, understanding of the interplay between camel, environment, and human remains limited. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1186/s12985-015-0439-5) contains supplementary material, which is available to authorized users. Text: An email from Dr Ali Mohamed Zaki, an Egyptian virologist working at the Dr Soliman Fakeeh Hospital in Jeddah in the Kingdom of Saudi Arabia (KSA) announced the first culture of a new coronavirus to the world. The email was published on the website of the professional emerging diseases (ProMED) network on 20 th September 2012 [1] (Fig. 1) and described the first reported case, a 60 year old man from Bisha in the KSA. This information led to the rapid discovery of a second case of the virus, this time in an ill patient in the United Kingdom, who had been transferred from Qatar for care [2] . The new virus was initially called novel coronavirus (nCoV) and subsequentlty entitled the Middle East respiratoy syndrome coronavirus (MERS-CoV). As of 2 nd of September 2015, there have been 1,493 detections of viral RNA or virus-specific antibodies across 26 countries (Additional file 1: Figure S1 ) confirmed by the World Health Organization (WHO), with over a third of the positive people dying (at least 527, 35 %) [3] . Since that first report, a slow discovery process over the following two to three years revealed a virus that had infected over 90 % of adult dromedary camels (DC; Camelus dromedarius) in the KSA [4] , also DCs across the Arabian Peninsula and parts of Africa that are a source of DC imports for the KSA [5] . To date, MERS-CoV has not been detected in DCs tested in zoos or herds from other parts of the world [6] [7] [8] [9] . Occasionally, virus is transmitted from infected DCs to exposed humans. Subsequent transmission to other humans requires relatively close and prolonged exposure [10] . The first viral isolate was patented and concerns were raised that this would restrict access to both the virus and to viral diagnostics [11, 12] . However, sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics were quickly described and virus was made freely available subject to routine biosafety considerations [13] . Subsequent epidemiology and research has identified the cell receptor as exopeptidase dipeptidyl peptidase 4 (DPP4; also called CD26); that MERS-CoV has a broad tropism, replicating better in some cells lines and eliciting a more proinflammatory response than SARS-CoV; is widespread in DCs; has the potential to infect other animals and that MERS kills its human host more often than SARS did (20-40 % versus 9 % for SARS [14] ) [15] [16] [17] [18] [19] . In humans, overt disease was given the name Middle East respiratory syndrome, with the acronym MERS. From intermittent animal-to-human spill-over events, the MERS-CoV spreads sporadically among people, causing more severe disease among older adults, especially males, with pre-existing diseases. The spread of MERS-CoV among humans has often been associated with outbreaks in hospitals, with around 20 % of all cases to date involving healthcare workers (HCWs). Although DCs appear to suffer the equivalent of a 'common cold' from MERS-CoV infection, in humans, the virus can be a more serious and opportunistic pathogen associated with the death of up to 40 % of reported cases. It has yet to be established whether infections thought to have been acquired from an animal source produce a more severe outcome than those spread between humans [20] . Studies have established that the mean incubation period for MERS is five to six days, ranging from two to 16 days, with 13 to 14 days between when illness begins in one person and subsequently spreads to another [21] [22] [23] [24] . Among those with progressive illness, the median time to death is 11 to 13 days, ranging from five to 27 days [23, 24] . Fever and gastrointestinal symptoms may form a prodrome, after which symptoms decline, only to be followed by a more severe systemic and respiratory syndrome [25, 26] . The first WHO case definition [27] defined probable cases of MERS based on the presence of febrile illness, cough and requirement for hospitalization with suspicion of lower respiratory tract (LRT) involvement. It also included roles for contact with a probable or confirmed case or for travel or residence within the Arabian Peninsula. If strictly adhered to, only the severe syndrome would be subject to laboratory testing, which was the paradigm early on [21] . From July 2013, the revised WHO case definition included the importance of seeking out and understanding the role of asymptomatic cases and from June 2014, the WHO definition more clearly stated that a confirmed case included any person whose sample was RT-PCR positive for MERS-CoV, or who produced a seroconversion, irrespective of clinical signs and symptoms. [28] [29] [30] Apart from the WHO and the KSA Ministry of Health reports, asymptomatic or subclinical cases of MERS-CoV infection were documented in the scientific literature although not always as often as occurred early on [31, 32] . The KSA definition of a case became more strict on 13 th May 2014, relying on the presence of both clinical features and laboratory confirmation [33] . Testing of asymptomatic people was recommended against from December 2014 [34] , reinforced by a case definition released by the KSA Ministry of Health in June 2015 [35] . The KSA has been the source of 79 % of human cases. Severe MERS is notable for its impact among older men with comorbid diseases including diabetes mellitus, cirrhosis and various lung, renal and cardiac conditions [36] [37] [38] . Interestingly in June 2015, an outbreak in South Korea followed a similar distribution [39, 40] . Among laboratory confirmed cases, fever, cough and upper respiratory tract (URT) signs and symptoms usually occur first, followed within a week by progressive LRT distress and lymphopaenia [37] . Patients often present to a hospital with pneumonia, or worse, and secondary bacterial infections have been reported [37, 41] . Disease can progress to acute respiratory distress syndrome and multiorgan system failure [37] . MERS has reportedly killed approximately 35 % of all reported cases, 42 % of cases in the KSA, yet only 19 % of cases in South Korea, where mortality ranged from 7 % among younger age groups to 40 % among those aged 60 years and above [42] ; all may be inflated values with asymptomatic or mild infections sometimes not sought or not reported [34] . General supportive care is key to managing severe cases [43] . Children under the age of 14 years are rarely reported to be positive for MERS-CoV, comprising only 1.1 % (n = 16) of total reported cases. Between 1 st September 2012 and 2 nd December 2013, a study described the then tally of paediatric cases in the KSA, which stood at 11 (two to 16 years of age; median 13 years); nine were asymptomatic (72 %) and one infant died [44] . In Amman, Jordan, 1,005 samples from hospitalized children under the age of two years with fever and/or respiratory signs and symptoms were tested but none were positive for MERS-CoV RNA, despite being collected at a similar time to the first known outbreak of MERS-CoV in the neighbouring town of Al-Zarqa [45] . A second trimester stillbirth occurred in a pregnant woman during an acute respiratory illness and while not RT-rtPCR positive, the mother did subsequently develop antibodies to MERS-CoV, suggestive of recent infection [46] . Her exposure history to a MERS-CoV RT-rtPCR positive relative and an antibody-reactive husband, her incubation period and her symptom history met the WHO criteria for being a probable MERS-CoV case [46] . Diagnostic methods were published within days of the ProMED email announcing the first MERS case [47] , including several now gold standard in-house RT-rtPCR assays (Fig. 2 ) as well as virus culture in Vero and LLC-MK2 cells [18, 47, 48] . A colorectal adenocarcinoma (Caco-2) epithelial cell line has since been recommended for isolation of infections MERS-CoV [49] . We previously [18] .). Open reading frames are indicated as yellow rectangles bracketed by terminal untranslated regions (UTR; grey rectangles). FS-frame-shift. Predicted regions encompassing recombination break-points are indicated by orange pills. Created using Geneious v8.1 [211] and annotated using Adobe Illustrator. Beneath this is a schematic depicting the location of RT-PCR primers (blue arrows indicate direction) and oligoprobes (green rectangles) used in the earliest RT-rtPCR screening assays and conventional, semi-nested (three primers) RT-PCR confirmatory sequencing assays [47, 48] . Publication order is noted by first [27 th September 2012; red] and second [6 th December 2012; orange] coloured rectangles; both from Corman et al. [47, 48] Those assays recommended by the WHO are highlighted underneath by yellow dots [53] . The NSeq reverse primer has consistently contained one sequence mismatch with some MERS-CoV variants. An altered version of that from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] reviewed the broad tropism of MERS-CoV [5] . However, as is well described, cell culture is a slow, specialised and insensitive method [50] while PCR-based techniques are the preferred method for MERS-CoV detection. The first open reading frames (ORF 1a and 1b; Fig. 2 ) have become a key diagnostic and taxonomic target for CoV species identification. With less than 80 % identity between the amino acid sequence of MERS ORF 1ab and betacoronavirus relatives, Tylonycteris bat HKU4 and Pipistrellus bat HKU5, it can be concluded that it is a novel and distinct virus. MERS-CoV is predicted to encode ten open reading frames with 5' and 3' untranslated regions [51] . The structural proteins include the spike (S), envelope (E), membrane (M) and nucleocapsid (N) [52] . The products of ORF1a and ORF1b are predicted to encode nonstructural proteins. The majority of specimen testing to date has employed validated RT-rtPCR assays shown to be sensitive and specific [47, 48, 53] . The RealStar® kit uses these WHOrecommended assays [54] . The target sequences of these screening assays have not changed among genomes examined until at least mid-2015 (IMM observation). Other RT-rtPCR assays have been developed and validated for use as laboratory-based diagnostic tools [55] [56] [57] . Additionally, loop-mediated [58, 59] or recombinase polymerase [60] isothermal assays have been designed for field deployment. The detection of MERS-CoV antigen has not been common to date but the combination of short turnaround time from test to result, high throughput and identification of viral proteins makes this an attractive option. Detection of viral proteins rather than viral RNA indicates the likely presence of infectious virus. The first rapid immunochromatographic tool described could detect recombinant MERS-CoV nucleocapsid protein from DC nasal swabs with 94 % sensitivity and 100 % specificity compared to RT-rtPCR [61] . A different approach used a monoclonal antibody-based capture ELISA targeting the MERS-CoV nucleocapsid protein with a sensitivity of 10 3 TCID 50 and 100 % specificity [62] . Demonstration of a seroconversion to a MERS-CoV infection meets the current WHO definition of a case so optimized and thoroughly validated sero-assays employed alongside good clinical histories are useful to both identify prior MERS-CoV infection and help support transmission studies. Because serology testing is, by its nature, retrospective, it is usual to detect a viral footprint, in the form of antibodies, in the absence of any signs or symptoms of disease and often in the absence of any viral RNA [63] . Strategic, widespread sero-surveys of humans using samples collected after 2012 are infrequent. Much of the Arabian Peninsula and all of the Horn of Africa lack baseline data describing the proportion of the community who may have been infected by a MERS-CoV. However, sero-surveys have had widespread use in elucidating the role of DCs as a transmission source for MERS-CoV. Because of the identity shared between DC and human MERS-CoV (see Molecular epidemiology: using genomes to understand outbreaks), serological assays for DC sero-surveys should be transferrable to human screening with minimal re-configuration. Also, no diagnostically relevant variation in neutralization activity have been found from among a range of circulating tested MERS-CoV isolates and sera, so whole virus or specific protein-based sero-assays should perform equivalently in detecting serological responses to the single MERS-CoV serotype [49] . The development of robust serological assays requires reliable panels of wellcharacterized animal or human sera, including those positive for antibodies specific to MERS-CoV, as well as to likely sources of cross-reaction [64] . Obtaining these materials was problematic and slowed the development and commercialization of antibody detection assays for human testing [64] . A number of commercial ELISA kits, immunofluorescent assays (IFA) kits, recombinant proteins and monoclonal antibodies have been released [31, [65] [66] [67] [68] . Initially, conventional IFAs were used for human sero-surveys. These relied on MERS-CoV-infected cell culture as an antigen source, detecting the presence of human anti-MERS-CoV IgG, IgM or neutralizing antibodies in human samples [18, 48, 69] . No sign of MERS-CoV antibodies was found among 2,400 sera from patients visiting Hospital in Jeddah, from 2010 through 2012, prior to the description of MERS-CoV [18] . Nor did IFA methods detect any sign of prior MERS-CoV infection among a small sample of 130 healthy blood donors from another Hospital in Jeddah (collected between Jan and Dec 2012) [70] . Of 226 slaughterhouse workers, only eight (3.5 %) were positive by IFA, and those sera could not be confirmed by virus neutralization (NT) test. The study indicated that HCoV-HKU1 was a likely source of crossreactive antigen in the whole virus IFA [70] . Whole virus MERS-CoV IFA also suffered from some cross-reactivity with convalescent SARS patient sera and this could not be resolved by an NT test which was also cross-reactive [71] . IFA using recombinant proteins instead of whole-virus IFA, has been shown to be a more specific tool [31] . Since asymptomatic zoonoses have been posited [72] , an absence of antibodies to MERS-CoV among some humans who have regular and close contact with camels may reflect the rarity of actively infected animals at butcheries, a limited transmission risk associated with slaughtering DCs [70] , a pre-existing cross-protective immune status or some other factor(s) resulting in a low risk of disease and concurrent seroconversion developing after exposure in this group. IFA using recombinant proteins instead. Some sero-assays have bypassed the risks of working with infectious virus by creating transfected cells expressing recombinant portions of the MERS-CoV nucleocapsid and spike proteins [48, 73] , or using a recombinant lentivirus expressing MERS-CoV spike protein and luciferase [74, 75] . A pseudo particle neutralization (ppNT) assay has seen widespread used in animal studies and was at least as sensitive as the traditional microneutralization (MNT) test. [10, 74, [76] [77] [78] ] Studies using small sample numbers and ppNT found no evidence of MERS-CoV neutralizing antibody in sera from 158 children with LRT infections between May 2010 and May 2011, 110 sera from 19 to 52 year old male blood donors and 300 selfidentified animal workers from the Jazan Region of the KSA during 2012 [79, 80] . Similarly, a study of four herdsmen in contact with an infected DC herd in Al-Ahsa, eight people who had intermittent contact with the herd, 30 veterinary surgeons and support staff who were not exposed to the herd, three unprotected abattoir workers in Al-Ahsa and 146 controls who were not exposed to DCs in any professional role, found none with serological evidence of past MERS-CoV infection using the ppNT assay [10] . A delay in the neutralizing antibody response to MERS-CoV infection was associated with increased disease severity in South Korea cases with most responses detectable by week three of illness while others, even though disease was severe, did not respond for four or more weeks [81] . The implications for our ability to detect any response in mild or asymptomatic cases was not explored but may be a signifcant factor in understanding exposure in the wider community. A Jordanian outbreak of acute LRT disease in a hospital in 2012 was retrospectively found to be associated with MERS-CoV infection, initially using RT-rtPCR, but subsequently, and on a larger scale, through positivity by ELISA and IFA or MNT test. [46, 82, 83] This outbreak predated the first case of MERS in the KSA. The ELISA used a recombinant nucleocapsid protein from the group 2 betacoronavirus bat-CoV HKU5 to identify antibodies against the equivalent crossreactive MERS-CoV protein [71] . It was validated using 545 sera collected from people with prior HCoV-OC43, HCoV-229E, SARS-CoV, HCoV-NL63, HRV, HMPV or influenza A(H1N1) infections but was reportedly less specific than the recombinant IFA discussed above. It was still considered an applicable tool for screening large sample numbers [82] . A protein microarray expressing the S1 protein subunit has also been validated and widely used for DC testing [5, 84] . Detection of MERS-CoV infection using ELISA or S1 subunit protein microarray [84] is usually followed by confirmatory IFA and/ or a plaque-reduction neutralization (PRNT) [69, 70, 85] or MNT test. [74, 85, 86] This confirmatory process aims toensure the antibodies detected are able to specifically neutralize the intended virus and are not more broadly reactive to other coronaviruses found in DCs (bovine CoV, BCoV) or humans (HCoV-OC43, HCoV-229E, HCoV-NL63, HCoV-HKU1, SARS-CoV). In the largest study of human sera, a tiered diagnostic process assigned both recombinant IFA and recombinant ELISA positive sera to 'stage 1' seropositivity. A stage 2 seropositive result additionally required a suitably titred PRNT result [87] . The study found 15 sera collected in 2012 to 2013 from 10,009 (0.2 %) people in 13 KSA provinces contained MERS-CoV antibodies, but significantly higher proportions in occurred in camel shepherds (two of 87; 2.3 %) and slaughterhouse workers (five of 140; 3.6 %) [87] . Contemporary surveys are needed. MERS-CoV does not appear to be easily transmitted from DCs to humans, or perhaps it is [72] , but generally does not trigger a detectable immune response if only mild disease or asymptomatic infection results. Serology assays are in need of further validation in this area so care is required when moving newly developed diagnostic serology algorithms from a research setting to one that informs public health decisions. This was reinforced when a false positive US case, purported to have been infected after a handshake and two face-to-face meetings, did not withstand further confirmatory analysis using a more specific, NT assay and was subsequently retracted [88, 89] . The WHO recommends sampling from the LRT for MERS-CoV RT-rtPCR testing, especially when sample collection is delayed by a week or more after onset of symptoms. [53] LRT samples are also best for attempting isolation of infectious virus, although the success of culture is reduced when disease persists [49] . Recommended sample types include bronchoalveolar lavage (BAL), tracheal/tracheobronchial aspirate, pleural fluid and sputum [53, 90] . Fresh samples yield better diagnostic results than refrigerated material [69] and if delays in testing of ≥72 h are likely, samples (except for blood) should be frozen at −70°C [90] . If available, lung biopsy or autopsy tissues can also be tested [53] . The URT is a less invasive and more convenient sampling site however, and an oropharyngeal and throat swab or a nasopharyngeal aspirate/wash are recommended when URT sampling is to be conducted [90] . Paired sera, collected two to three weeks apart are preferable for serological testing while a single sample is suggested to be sufficient if collected two weeks after onset of disease or a single serum collected during the first 10-12 days if conducting RT-rtPCR [53, 90] . Human urine and stool have been found to contain MERS-CoV RNA 12 to 26 days after symptom onset [25, 69, 91] and are listed as samples that should be considered [53, 90] . In two cases that arrived in the Netherlands, urine was RT-rtPCR negative but faeces was weakly positive and sera were RT-rtPCR positive for five days or more [25] . The finding of MERS-CoV viral RNA in serum provides an avenue for retrospective PCR-based studies if respiratory samples are unavailable [83] . RNAaemia may also correlate with disease severity; signs of virus were cleared from the serum of a recovered patient, yet lingered until the death of another [92] . Clinically suspected MERS cases may return negative results by RT-rtPCR. Data have shown one or more negative URT samples may be contradicted by further URT sampling or the use of LRT samples, which is preferred [2, 43, 93] . Higher viral loads occur in the LRT compared to the URT. [22, 69, 88, 94] This fits with the observation that the majority of disease symptoms are reported to manifest as systemic and LRT disease [21] . However, on occasion, even LRT specimens from MERS cases may initially be negative, only to later become positive by RT-PCR [95] . This may be due to poor sampling when a cough is absent or non-productive or because the viral load is low [95] . Despite this both the largest human MERS-CoV studies [32, [96] [97] [98] and smaller ones [22, 25, 99] , use samples from the URT. It is then noteworthy that one study reported an association between higher loads in the URT and worse clinical outcome including intensive care and death [94] . At writing, no human data exist to define whether the virus replicates solely or preferentially in the LRT or URT, or replicates in other human tissues in vivo although MERS-CoV RNA has been detected from both the URT and LRT in a macaque monkey model [100] .The distribution of DPP4 in the human upper airways is also not well described. Individual human case studies report long periods of viral shedding, sometimes intermittently and not necessarily linked to the presence of disease symptoms. [25, 69, 99, 101] In one instance, a HCW shed viral RNA for 42 days in the absence of disease [99] . It is an area of high priority to better understand whether such cases are able to infect others. Over three quarters of MERS cases shed viral RNA in their LRT specimens (tracheal aspirates and sputum) for at least 30 days, while only 30 % of contacts were still shedding RNA in their URT specimens [91, 102] . In the only study to examine the effect of sample type on molecular analysis, 64 nasopharyngeal aspirates (NPA; an URT sample), 30 tracheal aspirates, 13 sputa and three BAL were examined. The tracheal aspirates and BAL returned the highest viral load values followed by NPA and sputum. Unsurprisingly, higher viral loads generally paralleled whole genome sequencing and culture success and, in NPA testing, were significantly correlated with severe disease and death [49, 94, 103] . This study demonstrated the importance of LRT sampling for whole genome sequencing. When tested, samples positive for MERS-CoV are often negative for other pathogens [2, 25, 93, 104] . However, many studies make no mention of additional testing for endemic human respiratory viruses [21, 23, 73, 105] . When viruses are sought, they have included human herpesvirus (HHV), rhinoviruses (HRV), enteroviruses (EV), respiratory syncytial virus (RSV), parainfluenzavirus types 1, 2 and 3 (PIVs),influenzaviruses (IFVs), endemic HCoVs, adenoviruses (AdVs) metapneumovirus (MPV) and influenza A\H1N1 virus; co-detections with MERS-CoV have been found on occasion [2, 22, 37, 69, 97] . Bacterial testing is sometimes included (for example, for Legionella and Pneumococcus) but the impact of bacterial co-presence is also unclear [22, [104] [105] [106] . Further testing of the LRT sample from the first MERS case used IFA to screen for some viruses (negative for IFV, PIVs, RSV and AdVs) and RT-PCR for others (negative for AdV, EVs, MPV and HHVs) [18] . RT-PCR also detected MERS-CoV. The WHO strongly recommends testing for other respiratory pathogens [53] but with this recommendation often discounted, there are limited data to address the occurrence and impact of co-infections or alternative viral diagnoses among both MERS cases and their contacts. Little is known of other causes of MERS-like pneumonia in the KSA or of the general burden of disease due to the known classical respiratory viruses. Testing of adult pilgrims performing the Hajj in 2012 to 2014 has not detected any MERS-CoV. In 2012, nasal swabs from 154 pilgrims collected prior to leaving for or departing from the KSA were tested [47] . In 2013, testing was significantly scaled up with 5,235 nasopharyngeal swabs from 3,210 incoming pilgrims and 2,025 swabs from outgoing pilgrims tested [98] . It should be noted that most pilgrims arrived from MERS-free countries. A further 114 swabs were taken from pilgrims with influenza-like illness [96, 107] . In earlier Hajj gatherings, it was found that influenza viruses circulated widely, whilst other viruses, often rhinoviruses, circulated more selectively, interpreted as indicating their importation along with foreign pilgrims. [107] [108] [109] Over time, increased influenza vaccination has been credited for a fall in the prevalence of influenza like illnesses among Hajj pilgrims. [110] A LRT sample is often not collected for these studies [98, 107, 109] , so false negative findings are a possibility although little is known about the initial site of MERS-CoV infection and replication; it may have been assumed it was the LRT because disease was first noticed there but the URT may be the site of the earliest replication. In Jeddah between March and July 2014 (hereafter called the Jeddah-2014 outbreak; Fig. 3 ), there was a rapid increase in MERS cases, accompanied by intense screening; approximately 5,000 samples from in and around the region were tested in a month yielding around 140 MERS-CoV detections (~3 % prevalence) [111] . Among 5,065 individuals sampled and tested across the KSA between October 2012 and September 2013,108 (2.1 %) detections were made in a hospital-centric population which included hospitalized cases (n = 2,908; 57.4 %), their families (n = 462; 9.1 %) and associated HCWs (n = 1,695; 33.5 %) [32] . Among the detections, 19 (17.8 %) were HCWs and 10 (9.3 %) were family contacts [32] . The 2-3 % prevalence of active MERS-CoV infections is not dissimilar to the hospital-based prevalence of other human CoVs. [112] However, the proportion of deaths among those infected with MERS-CoV is much higher than that known for the HCoVs NL63, HKU1, 229E or OC43 in other countries, and even above that for SARS-CoV; it is not a virus that could reasonably be described as a "storm in a teacup". It is the low transmission rate that has prevented worldwide spread, despite many "opportunities". Very early in the MERS outbreak, some animals were highly regarded as either the reservoir or intermediate host(s) of MERS-CoV with three of the first five cases having contact with DCs [73, 113, 114] . Today, animal MERS-CoV infections must be reported to the world organization for animal health as an emerging disease [115] . A summary of the first MERS cases reported by the WHO defined animal contact with humans as being direct and within 10 days prior to symptom onset [20] . This definition made no specific allowance for acquisition from DCs through a droplet-based route, which is very likely route for acquisition of a virus that initially and predominantly causes respiratory disease [23] . Camels are known to produce high levels of MERS-CoV RNA in their URT and lungs [116] . Providing support for a droplet transmission route and perhaps indicating the presence of RNA in smaller, drier droplet nuclei, MERS-CoV RNA was identified in a high volume air sample collected from a barn housing an infected DC [117] . The precise source from which humans acquire MERS-CoV remains poorly studied but it seems likely that animal and human behavioural factors may play roles (Fig. 3) [118] . These factors may prove important for human cases who do not describe any DC contact [119] nor any contact with a confirmed case. Whether the WHO definition of animal contact is sufficient to identify exposure to this respiratory virus remains unclear. Wording focuses on consumption of DC products but does not specifically ascribe risk to a droplet route for acquisition of MERS-CoV from DC [120] . Some MERS patients are listed in WHO disease notices as being in proximity to DCs or farms, but the individuals have not described coming into contact with the animals. No alternative path for acquiring infection is reported in many of these instances. What constitutes a definition of "contact" during these interviews has been defined for one study [72] . Despite this lack of clarity, the WHO consider that evidence linking MERS-CoV transmission between DCs to humans is irrefutable (Fig. 4) [120] . The possibility that bats were an animal host of MERS-CoV was initially widely discussed because of the existing diversity of coronaviruses known to reside among them [121] [122] [123] [124] . Conclusive evidence supporting bats as a source for human infections by MERS-CoV has yet to be found, but bats do appear to host ancestral representatives [53, 125] . However, these are not variants of the same virus nor always within the same phylogenetic lineage as MERS-CoV; they are each a genetically distinct virus. Bat-to-human infection by MERS-CoV is a purely speculative event. The only piece of MERS-CoV-specific evidence pointing to bats originates from amplification of a 190 nt fragment of the RNAdependent RNA polymerase gene of the MERS-CoV genome, identified in a faecal pellet from an insectivorous Emballonuridae bat, Taphozous perforatus found in Bisha, the KSA [121] . While very short, the sequence of the fragment defined it as a diagnostic discovery. Subsequently a link to DCs was reported [85] and that link has matured into a verified association [38, 126] (Fig. 4) . (See figure on previous page.) Fig. 3 Monthly detections of MERS-CoV (blue bars) and of cases who died (red bars) with some dates of interest marked for 2012 to 4 th September 2015. An approximation of when DC calving season [128] and when recently born DCs are weaned is indicated. Spring (green) and summer (orange) in the Arabian Peninsula are also shaded. Note the left-hand y-axis scale for 2014 and 2015 which is greater than for 2012/13. Sources of these public data include the WHO, Ministries of Health and FluTrackers [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] . Modified and reprinted from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] DCs, which make up 95 % of all camels, have a central presence in the Arabian Peninsula where human-DC contact ranges from little to close [119] . Contact may be commonplace and could occur in variety of ways (Fig. 4a) . There are several large well-attended festivals, races, sales and parades which feature DCs and DCs are also kept and bred close to populated areas in the KSA [127, 128] . DC milk and meat are widely consumed and the older DC is an animal of ritual significance after the Hajj pilgrimage [129] . However, MERS-CoV infection frequency is reportedly much lower than is the widespread and frequent habit of eating, drinking and preparing DC products. Daily ingestion of fresh unpasteurized DC milk is common among the desert Bedouin and many others in the KSA. DC urine is also consumed or used for supposed health benefits. Despite camel butchery being a local occupation, neither butchers nor other at-risk groups are identifiable among MERS cases; this may simply be a reporting issue rather than an unexplainable absence of MERS. A small case-control study published in 2015 identified direct DC contact, and not ingestion of products, to be associated with onset of MERS [38] . The first sero-survey of livestock living in the Middle East region was conducted during 2012-2013 [85] . DCs were sampled from a mostly Canary Island-born herd and from Omani DCs (originally imported from the Horn of Africa) [85] . A neutralising antibody assay found only 10 % of strongly seropositive Canary Island [5] . b Camel-to-human infections appear to be infrequent, while human-to-human spread of infection is regularly facilitated by poor IPC in healthcare settings where transmission is amplified, accounting for the bulk of cases. There are human MERS cases that do not fall into either category of source and it is unclear if these acquired infection through some entirely separate route, or from cases that escaped diagnosis. c Hypothetical ways in which subclinical (when infection may not meet a previously defined clinical threshold of signs and/or symptoms) or asymptomatic (no obvious signs or measured, noticed or recalled symptoms of illness) MERS-CoV infection may be implicated in transmission DC sera could neutralise MERS-CoV while all Omani DC sera had high levels of specific MERS-CoV neutralizing antibody [85] . This indicated that DCs had in the past been infected by MERS-CoV, or a very similar virus. Since this study, a host of peer-reviewed reports have looked at both DCs and other animals, and the possibility that they may host MERS-CoV infection. Seropositive DCs have been found throughout the Arabian Peninsula including Oman, the KSA, Qatar, Jordan, the United Arab Emirates (UAE), Kuwait as well as Sudan, Somalia, Egypt, Tunisia, Nigeria, Kenya and Ethiopia in Africa and the Canary Islands [85, [130] [131] [132] [133] [134] . Other animals tested include sheep, cows, pigs, horses, donkeys, mules, birds, water buffalo, goats, Bactrian camels, llamas and guanaco (south American camelids) but none had detectable neutralising antibody against MERS-CoV [4, 74, 78, 85, 86, 135, 136] . No virology or serology studies of human samples from areas in Africa where there are camels with a history of MERS-CoV have been reported to date. However,an absence of unexplained pneumonia that may be attributable to MERS-CoV infection may not signal the absence of virus among humans in each country but simply reflect a lack of expensive epidemiology studies conducted by resource-poor countries. It is thus unclear whether MERS-CoV, or an antigenically related CoV, is an unrecognized pathogen in these regions, perhaps circulating for even longer than it has been known in the Arabian Peninsula [133] . MERS-CoV RNA has also been detected in DC samples, and recovery of infectious virus has also been achieved from DC samples [4, 77, 117, 132, [137] [138] [139] [140] [141] . From some of these, full or majority length genomes of MERS-CoV have been sequenced [77, 137, 138] . DC versions of MERS-CoV were found to be as similar to each other, as were variants detected from different humans over time and across distance. Antibody screening assays have also detected crossreactive antibodies in sera. These were identified as such by screening sera against similar viruses, for example BCoV or HCoV-OC43 (as an antigenic facsimile for BCoV). It is possible that other MERS-CoV-like viruses also reside within DCs, but this does not detract from the definitive finding of MERS-CoV genetic sequences in both DCs and humans [117, 142, 143] . Screening studies have shown that juvenile DCs are more often positive for virus or viral RNA while older DCs are more likely to be seropositive and RNA or virus negative [76, 77, 144] . In adult DCs, MERS-CoV RNA has been detected among animals with pre-existing antibody, suggesting re-infection is possible [77, 144] . Viral loads among positive DCs can be very high [4, 76, 77, 139, 144] and DCs have been found positive both when ill with URT respiratory signs [77, 117, 142, 145] or when apparently healthy [137] . These findings indicate DCs host natural MERS-CoV infections. Furthermore, stored DC sera have revealed signs of MERS-CoV in DCs which date back over three decades (the earliest collected in 1983) [4, 133, 135] . Older sera have not been tested and so precisely how long DCs have been afflicted by MERS-CoV, whether the virus is enzootic among them, introduced to them decades or centuries ago from bats in Africa or the Arabian Peninsula, or they are the subject of regular but short-lived viral incursions from an as yet unknown host, cannot be answered. Researchers sought to determine a direction for infection; were DCs transmitting virus to humans or were humans infecting DCs? At a Qatari site, a farm owner and his employee became ill in mid-October 2013 and tested positive for MERS-CoV RNA in a sputum and throat swab sample, respectively. RT-rtPCRs found MERS-CoV RNA in 11 of 14 positive DC nasal swabs at the farm; six (43 %) positive by two or more assays [138] . The results indicated a recent outbreak had occurred in this herd; the first indication of MERS-CoV RNA found within DCs with a temporal association to human infections. Three positive DC samples were confirmed by sequencing a 358 nt portion of the spike gene; these sequences were identical to each other, again with close homology to other human and DC MERS-CoV sequences [138] . The DCs and human contacts yielded ORF1a and ORF4b sequences differing by only a single nucleotide each, clustering closely with the Hafr-Al-Batin_1_2013 variant [138] . Subsequent case studies found evidence of a concurrent human and DC infection and the direction of that infection was inferred to be from the ill DCs and to their human owners [117, 142, 146] . Partial genome sequences indicated that a human and a MERS-CoV RT-rtPCR positive DC had been infected by a variant of the same virus, harbouring the same distinct pattern of nucleotide polymorphisms. [142] All nine DC in the owner's herd, serially sampled, reacted in a recombinant S1 antigen ELISA, with the two animals that had been RT-rtPCR positive showing a small, verifiable rise in antibody titre [142] . A rise in titre theoretically begins 10 to 21 days after DC infection [142] . The authors suggested that the rise in titre in DC sera which occurred alongside a declining RNA load, while the patient was actively ill and hospitalized, indicated that the DCs were infected first followed by the owner [117, 142] . BCoV antibodies were also present, and rising in one of the two RT-rtPCR positive animals but no animal's antibodies could neutralise BCoV infection [142] . Camel calving season occurs in the winter months (between late October and late February; Fig. 3 ) and this may be a time when there is increased risk to humans of spill-over due to new infections among naïve DC populations [128] . What role maternal camel antibody might play in delaying infection of calves remains unknown [128, 142] . Juvenile DCs appear to host active infection more often than adult DCs and thus the sacrificial slaughter of DCs, which must be five years of age or older (termed a thane), may not be accompanied by significant risk of exposure to infection. In contrast to earlier results, slaughterhouse workers who kill both younger and older DCs, may be an occupational group with significantly higher incidence of seropositivity to MERS-CoV when animals have active MERS-CoV infections [129, 139, [147] [148] [149] . Expanded virological investigations of African DCs may lead to more seropositive animals and geographic areas in which humans may be at risk. It is possible that there are areas where humans already harbour MERS-CoV infections that have not been identified because of an absence of laboratory surveillance. Virological investigations of bats may lead to findings of ancestral viruses and viral 'missing links' and identifying any other animal sources of zoonotic spread is important to inform options for reducing human exposures [56, 76] . Infectious MERS-CoV added to DC, goat or cow milk and stored at 4°C could be recovered at least 72 h later and, if stored at 22°C, recovery was possible for up to 48 h [150] . MERS-CoV titre decreased somewhat when recovered from milk at 22°C but pasteurization completely ablated MERS-CoV infectivity [150] . In a subsequent study, MERS-CoV RNA was identified in the milk, nasal secretion and faeces of DCs from Qatar [151] . A single study has examined the ability of MERS-CoV to survive in the environment [150] . Plastic or steel surfaces were inoculated with 10 6 TCID 50 of MERS-CoV at different temperature and relative humidity (RH) and virus recovery was attempted in cell culture. At high ambient temperature (30°C) and low RH (30 %) MERS-CoV remained viable for 24 h [150] . By comparison, a well known and efficently transmitted respiratory virus, influenza A virus, could not be recovered in culture beyond four hours under any conditions [150] . Aerosol experiments found MERS-CoV viability only decreased 7 % at low RH at 20°C. In comparison, influenza A virus decreased by 95 % [150] . MERS-CoV survival is inferior to that previously demonstrated for SARS-CoV [152] . For context, pathogenic bacteria can remain viable and airborne for 45 min in a coughed aerosol and can spread 4 m. MERS-CoV's ability to remain viable over long time periods gives it the capacity to thoroughly contaminate a room's surfaces when occupied by an infected and symptomatic patient [153] . Whether MERS-CoV can remain adrift and infectious for extended periods (truly airborne) remains unknown. Such findings expand our understanding of the possibilities for droplets to transmit respiratory viruses in many settings, including hospital waiting rooms, emergency departments, treatment rooms, open intensive care facilities and private patient rooms. The nature and quality of air exchange, circulation and filtration are important variables in risk measurement and reduction as is the use of negative pressure rooms to contain known cases. Droplet spread between humans is considered the mechanism of human-to-human transmission and the need for droplet precautions was emphasized after the Al-Ahsa hospital, the KSA and the South Korean outbreaks [21, 23, 154, 155] . By extrapolation, aerosol-generating events involving DCs (urination, defecation, and preparation and consumption of DC products) should be factored into risk measurement and reduction efforts and messaged using appropriate context. The provision of evidence supporting the best formulation of personal protective equipment to be worn by HCWs who receive, manage or conduct procedures on infectious cases remains a priority. MERS-CoV was found and characterized because of its apparent association with severe, and therefore more obvious, illness in humans; we were the canaries in the coal mine. Sero-assays and prospective cohort studies have yet to determine the extent to which milder or asymptomatic cases contribute to MERS-CoV transmission chains. However, transmission of MERS-CoV is defined as sporadic (not sustained), intra-familial, often healthcare associated, inefficient and requiring close and prolonged contact [22, 31, 63, 93, 97, 102, 156] In a household study, 14 of 280 (5 %) contacts of 26 MERS-CoV positive index patients were RNA or antibody positive; the rate of general transmission, even in outbreaks is around 3 % [31] . It seems that the majority of human cases of MERS-CoV, even when numbers appear to increase suddenly, do not readily transmit to more than one other human so to date, the localized epidemic of MERS-CoV has not been self-sustaining [157] [158] [159] [160] [161] . That is to say, the basic reproduction number (R 0 ) -the average number of infections caused by one infected individual in a fully susceptible populationhas been close to one throughout various clusters and outbreaks. If R 0 was greater than 1, a sustained increase in case numbers would be expected. Some R o calculations may be affected by incomplete case contact tracing, limited community testing and how a case is defined. That MERS has had a constant presence in the Arabian Peninsula since 2012 is due to ongoing, sporadic spill-over events from DCs amplified by poorly controlled hospital outbreaks. The first known MERS human-to-human transmission event was one characterized by acute LRT disease in a healthcare setting in Jordan. In stark contrast, a sero-survey of HCW who were sometimes in close and prolonged contact with the first, fatal MERS-CoV case in 2012 [162] , found none of the HCW had seroconverted four months later, despite an absence of eye protection and variable compliance with required PPE standards [162] . Early on in the MERS story, samples for testing were mostly collected from patients with severe illness and not those with milder acute respiratory tract infections. Contacts of confirmed MERS cases were often observed for clinical illness, but not tested. These omissions may have confounded our understanding of MERS-CoV transmission and biased early data towards higher numbers of seriously ill and hospitalized patients, inflating the apparent proportion of fatal cases. Case-control studies were not a focus. As testing paradigms changed and contacts were increasingly tested, more asymptomatic and mild infections were recognized [163] . A rise in the cases termed asymptomatic (which enlarge the denominator for calculations of the proportion of fatal cases, defined in [164] ) resulted in a drop in the proportion of fatal cases during the Jeddah-2014 outbreak. Historically, such rises are consistent with changing definitions and laboratory responses and clinical management of a newly discovered virus infection that was first noted only among the severely ill. Upon follow-up, over three-quarters of such MERS-CoV RNA positive people did recall having one or more symptoms at the time, despite being reported as asymptomatic [165] raising some question over the reliability of other reported data. The proportion of fatal MERS cases within the KSA compared to outside the KSA, as well as the age, and sex distribution change in different ways when comparing MERS outbreaks. Approximately 43 % of MERS cases (549 of 1277) in the KSA were fatal betwen 2012 and December 2015 while 21 % (72 of 330) died among those occurring outside of the KSA. The total number of male cases always outnumber females and the proportion of male deaths is always greater than the proportion of females who die. However the proportion of male deaths from total males with MERS is a similar figure to that for females. In the KSA, there is a greater proportion of younger males among cases and deaths than were observed from the 2015 South Korean or the Jeddah-2014 outbreaks (Additional file 2: Figure S2 ). Why these aspects have differed may be due to differences in the time to presentation and diagnosis, the nature and quality of supportive care, the way a person became infected (habits, exposure to a human or zoonotic source, viral load, route of infection) or the extent to which different populations are burdened by underlying diseases [40] . As a group, HCWs comprised 16 % of MERS cases in the KSA and South Korea. It is apparent that the weekly proportion of infected HCWs increases alongside each steep rise in overall detections (Fig. 5) . In May 2013, the WHO published guidelines for IPC during care of probable or confirmed cases of MERS-CoV infection in a healthcare setting [166] . This is explainable because to date, each case rise has been intimately associated with healthcare-facility related outbreaks [118] . These rises in MERS-CoV detections can decrease the average age during each event because HCWs are usually younger than inpatients with MERS. Healthcare facilities have been a regular target for suggested improvements aimed at improving infection prevention and control (IPC) procedures [115, 118] . Most of the analysis of MERS-CoV genetics has been performed using high throughput or "deep" sequencing methods for complete genome deduction [167] [168] [169] . MERS-CoV was the first subject of such widespread use of deep sequencing to study an emerging viral outbreak with global reach. The technique can produce genomic [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] length coverage in a single experiment with highly repetitious measurement of each nucleotide position [52, 140] . Despite assays having been published early on, subgenomic sequencing, once the mainstay of viral outbreak studies, has less often been published during MERS-CoV characterization [48] . As more genomes from both humans and DCs have been characterized, two clades have become apparent; A and B (Fig. 6) . Clade A contains only human-derived MERS-CoV genomes from Jordan, while Clade B comprises the majority of human and camel genomes deduced thus far [168] . Two studies during 2015, one looking at Jeddah-2014 MERS-CoV variants and another looking at a variant exported from South Korea to China, have now identified signs of genetic recombination among MERS-CoV variants. While human and camel whole genome sequences have retained >99 % identity with each other, members of genetically distinct lineages can and do swap genetic material when suitable conditions and coinfections co-occur [170] [171] [172] . Shared identity implies that the major source for human acquisition is the DC, rather than another animal, although more testing of other animal species is needed to confirm that conclusion. Over a month, a DC virus sequenced on different occasions did not change at all indicating a degree of genomic stability in its host, supporting that DCs are the natural, rather than intermediate, host for the MERS-CoV we know today [77] . To date, recombination has been localised to breakpoints near the boundary between ORF1a and ORF1b regions, within the spike gene [170] and in the ORF1b region (Fig. 2) [172] . It is not unexpected that recombination should occur since it is well known among other CoVs [124] and because the majority of MERS-CoV whole genomes collected from samples spanning three years (2012-2015) and from humans, camels and different countries have shown close genetic identity to each other, with just enough subtle variation to support outbreak investigations so long as whole genome sequencing is applied [52, 77, 135, 138, 168, [173] [174] [175] . Changes in genome sequence may herald alterations to virus transmissibility, replication, persistence, lethality or response to future drugs. If we have prior knowledge of the impact of genetic changes because of thorough characterization studies, we can closely Fig. 6 The genetic relationship between MERS-CoV nucleotide sequences (downloaded from GenBank using the listed accession numbers and from virological.org [212] ). This neighbour joining tree was created in MEGA v6 using an alignment of human and DCderived MERS-CoV sequences (Geneious v8.1 [211] ). Clades are indicated next to dark (Clade A) or pale (Clade B) blue vertical bars. Camel icons denote genomes from DCs. Healthcare or community outbreaks are boxed and labelled using previously described schemes [212, 213] monitor the genomic regions and better understand any changes in transmission or disease patterns as they occur. Genetic mutations noted during the largest of human outbreaks, Jeddah-2014, did not impart any major replicative or immunomodulatory changes when compared to earlier viral variants in vitro [156, 176] . However, we understand very little of the phenotypic outcomes that result from subtle genetic change in MERS-CoV genomes. To date no clinical relevance or obvious in vivo changes to viral replication, shedding or transmission has been reported or attributed to mutations or to new recombinant viruses [156] . But vigilance and larger, more contemporary and in vivo studies are needed. Genome sequence located to a distinct clade were identified from an Egyptian DC that was probably imported from Sudan. This does not fit into either of the current clades [125, 168, 177] . A virus sequenced from a Neoromicia capensis bat was more closely related to MERS-CoV than other large bat-derived sequences had been to that point, but the genome of a variant of a MERS-CoV has yet to be discovered and deduced from any bat [125] . Analyses of MERS-CoV genomes have shown that most single nucleotide differences among variants were located in the last third of the genome (Fig. 2) , which encodes the spike protein and accessory proteins [168] . At least nine MERS-CoV genomes contained amino acid substitutions in the receptor binding domain (RBD) of the spike protein and codons 158 (N-terminal region), 460 (RBD), 1020 (in heptad repeat 1), 1202 and 1208 bear investigation as markers of adaptive change [140, 169] . The spike protein had not changed in the recombinant MERS-CoV genome identified in China in 2015 but was reported to have varied at a higher rate than that for complete MERS-CoV genomes, among South Korean variants [172, 178] . This highlights that subgenomic regions may not always contain enough genetic diversity to prove useful for differentiating viral variants. Despite this, one assay amplifying a 615 nucleotide fragment of the spike S2 domain gene for Sanger sequencing agreed with the results generated by the sequencing of a some full genomes and was useful to define additional sequence groupings [177] . Genomic sequence can also be used to define the geographic boundaries of a cluster or outbreak and monitor its progress, based on the similarity of the variants found among infected humans and animals when occurring together, or between different sites and times (Fig. 6 ) [169] . This approach was employed when defining the geographically constrained MERS hospital outbreak in Al-Ahsa, which occurred between 1 st April and 23 rd May 2013, as well as clusters in Buraidah and a community outbreak in Hafr Al-Batin, the KSA. Genomic sequencing identified that approximately 12 MERS-CoV detections from a community outbreak in Hafr Al-Batin between June and August 2013 may have been triggered by an index case becoming infected through DC contact [175] . Sequencing MERS-CoV genomes from the 2013 Al-Ahsa hospital outbreak indicated that multiple viral variants contributed to the cases but that most were similar enough to each other to be consistent with human-tohuman transmission. Molecular epidemiology has revealed otherwise hidden links in transmission chains encompassing a period of up to five months [179] . However, most outbreaks have not continued for longer than two to three months and so opportunities for the virus to adapt further to humans through co-infection and sustained serial passage have been rare [169] . In Riyadh-2014, genetic evidence supported the likelihood of multiple external introductions of virus, implicating a range of healthcare facilities in an event that otherwise looked contiguous [23, 168, 179] . Riyadh is a nexus for camel and human travel and has had more MERS cases than any other region of the KSA to date but also harbours a wide range of MERS-CoV variants [128, 167, 179] . However the South Korean outbreak originated from a single infected person, resulting in three to four generations of cases [180, 181] . Studies of this apparently recombinant viral variant did not find an increased evolutionary rate and no sign of virus adaptation thus the outbreak seems to have been driven by circumstance rather than circumstance together with mutation [181] . For many MERS cases detected outside the Arabian Peninsula, extensive contact tracing has been performed and the results described in detail. Contact tracing is essential to contain the emergence and transmission of a new virus and today it is supported by molecular epidemiology. Although it is an expensive and time consuming process, contact tracing can identify potential new infections and through active or passive monitoring, react more rapidly if disease does develop. Results of contact tracing to date have found that onward transmission among humans is an infrequent event. For example, there were 83 contacts, both symptomatic and asymptomatic, of a case treated in Germany who travelled from the UAE but no sign of virus or antibody were found in any of them [73] . The very first MERS case had made contact with 56 HCWs and 48 others, but none developed any indication of infection [162] . In a study of 123 contacts of a case treated in France, only seven matched the definition for a possible case and were tested; one who had shared a 20 m 2 hospital room while in a bed 1.5 m away from the index case for a prolonged period was positive [26] . None of the contacts of the first two MERS cases imported into the USA in 2014 contained any MERS-CoV footprint [182] and none of the 131 contacts of two travellers returning to the Netherlands developed MERS-CoV antibodies or tested RNA positive [25, 183] . Analyses of public data reveal many likely instances of nosocomial acquisition of infection in the Arabian Peninsula and these data may be accompanied by some details noting contact with a known case or facility. One example identified the likely role of a patient with a subclinical infection, present in a hospital during their admission for other reasons, as the likeliest index case triggering a family cluster [93] . Contact tracing was a significant factor in the termination of a 2015 outbreak involving multiple South Korean hospitals [184] . Such studies demonstrate the necessity of finding and understanding a role for mild and asymptomatic cases, together with restricting close contact or prolonged exposure of infected people to others, especially older family members and friends with underlying disease (Fig. 4c) . The hospital-associated outbreak in Jeddah in 2014 was the largest and most rapid accumulation of MERS-CoV detections to date. The greatest number of MERS-CoV detections of any month on record occurred in Jeddah in April. The outbreak was mostly (>60 % of cases) associated with human-to-human spread within hospital environments and resulted from a lack of, or breakdown in, infection prevention and control [37, 185, 186] . A rise in fatalities followed the rapid increase in case numbers. In 2015 two large outbreaks occurred. South Korea was the site of the first large scale outbreak outside the Arabian Peninsula and produced the first cases in both South Korea and China, occurring between May and July 2015. This was closely followed by a distinct outbreak in Ar Riyad province in the KSA which appeared to come under control in early November. After staying in Bahrain for two weeks, a 68 year old male (68 M) travelled home to South Korea via Qatar, arriving free of symptoms on the 4 th May 2015 [187] . He developed fever, myalgia and a cough nearly a week later (11 th ). He visited a clinic as an outpatient between the 12 th and 15 th of May and was admitted to Hospital A on the 15 th [188] . He was discharged from Hospital A on the 17 th then visited and was admitted to the emergency department of Hospital B on the 18 th . During this second stay, a sputum sample was taken and tested positive for MERS-CoV on the 20 th [187, 188] , triggering transfer to the designated isolation treatment facility. Over a period of 10 days, the index case was seen at three different hospitals, demonstrating a key feature of "hospital shopping" that shaped the South Korean outbreak. Approximately 34 people were infected during this time [187] . In total 186 cases were generated in this outbreak, all linked through a single transmission chain to 68 M; 37 cases died [189] . In South Korea, the national health insurance system provides for relatively low cost medical care, defraying some costs by making family members responsible for a portion of the ministration of the sick, resulting in them sometimes staying for long periods in the rooms that often have more than four beds in them [24] . Other factors thought to have enabled this outbreak included unfamiliarity of local clinicians with MERS, ease with which the public can visit and be treated by tertiary hospitals, the custom of visiting sick friends and relatives in hospitals, the hierarchical nature of Korean society, crowded emergency rooms, poor IPC measures, a lack of negative pressure isolation rooms and poor inter-hospital communication of patient disease histories [24, [190] [191] [192] . All of the reported transmission occurred across three or four generations and apart from one unknown source, were all hospital-acquired [24, 120, 181, [193] [194] [195] . Few clinical details about these cases have been reported to date and detail on transmission and contact tracing is minimal. The hospitals involved were initially not identified, governmental guidance and actions produced confusing messages and there was very limited communication at all early on which resulted in unnecessary concern, distrust and a distinct economic impact [191, [196] [197] [198] . Early in the outbreak, a infected traveller, the son of an identified case in South Korea, passed through Hong Kong on his way to China where he was located, isolated and cared for in China [91, 199, 200] . No contacts became ill. The outbreak was brought under control in late July/ early August [201] after improved IPC measures were employed, strong contact tracing monitoring and quarantine, expanded laboratory testing, hospitals were better secured, specialized personnel were dispatched to manage cases and international cooperation increased [202, 203] . A review of public data showed that, as for MERS in the KSA, older age and the presence of underlying disease were significantly associated with a fatal outcome in South Korea. [40] Even though R 0 is <1, super-spreading events facilitated by circumstances created in healthcare settings and characterized by cluster sizes over 150, such as this one, are not unexpected from MERS-CoV infection [204] . The dynamic of an outbreak depends on the R 0 and an individual's viral shedding patterns, contact type and frequency, hospital procedures and population structure and density [204] . In the region of Ar Riyad, including the capital city of Riyadh, a hospital based cluster began, within a single hospital, from late June 2015 [205] . By mid-September there had been approximately170 cases reported but the outbreak appeared to been brought under control in November. It became apparent early on that MERS-CoV spread relatively ineffectively from human-to-human. Despite ongoing and possibly seasonal introduction of virus to the human population via infected DCs and perhaps other animals yet to be identified, the vast majority of MERS-CoV transmission has occurred from infected to uninfected humans in close and prolonged contact through circumstances created by poor infection control in health care settings. This opportunistic virus has had its greatest impact on those with underlying diseases and such vulnerable people, sometimes suffering multiple comorbidities, have been most often associated with hospitals, creating a perfect storm of exposure, transmission and mortality. It remains unclear if this group are uniquely affected by MERS-CoV or if other respiratory virus infections, including those from HCoVs, produce a similarly serious impact. In South Korea, a single imported case created an outbreak of 185 cases and 36 deaths that had a disproportionate impact on economic performance, community behaviour and trust in government and the health care system. Household human-to human transmission occurs but is also limited. Educational programs will be essential tools for combatting the spread of MERS-CoV both within urban and regional communities and for the health care setting. Vigilance remains important for containment since MERS-CoV is a virus with a genetic makeup that has been observed for only three years and is not stable. Among all humans reported to be infected, nearly 40 % have died. Continued laboratory testing, sequencing, analysis, timely data sharing and clear communication are essential for such vigilance to be effective. Global alignment of case definitions would further aid accurate calculation of a case fatality ratio by including subclinical case numbers. Whole genome sequencing has been used extensively to study MERS-CoV travel and variation and although it remains a tool for experts, it appears to be the best tool for the job. MERS and SARS have some clinical similarities but they also diverge significantly [206] . Defining characteristics include the higher PFC among MERS cases (above 50 % in 2013 and currently at 30-40 %; well above the 9 % of SARS) and the higher association between fatal MERS and older males with underlying comorbidities. For the viruses, MERS-CoV has a broader tropism, grows more rapidly in vitro, more rapidly induces cytopathogenic change, triggers distinct transcriptional responses, makes use of a different receptor, induces a more proinflammatory state and has a delayed innate antiviral response compared to SARS-CoV. There appears to be a 2-3 % prevalence of MERS-CoV in the KSA with a 5 % chance of secondary transmission within the household. There is an increased risk of infection through certain occupations at certain times and a much greater chance for spread to other humans during circumstances created by humans, which drives more effective transmission than any R 0 would predict on face value. Nonetheless, despite multiple mass gatherings that have afforded the virus many millions of opportunities to spread, there have remarkably been no reported outbreaks of MERS or MERS-CoV during or immediately after these events. There is no evidence that MERS-CoV is a virus of pandemic concern. Nonetheless, hospital settings continue to describe MERS cases and outbreaks in the Arabian Peninsula. As long as we facilitate the spread of MERS-CoV among our most vulnerable populations, the world must remain on alert for cases which may be exported more frequently when a host country with infected camel reservoirs is experiencing human clusters or outbreaks. The MERS-CoV appears to be an enzootic virus infecting the DC URT with evidence of recent genetic recombination. It may once have had its origins among bats, but evidence is lacking and the relevance of that to today's ongoing epidemic is academic. Thanks to quick action, the sensitive and rapid molecular diagnostic tools required to achieve rapid and sensitive detection goal have been in place and made widely available since the virus was reported in 2012. RT-PCR testing of LRT samples remains the gold standard for MERS-CoV confirmation. Serological tools continue to emerge but they are in need of further validation using samples from mild and asymptomatic infections and a densely sampled cohort study to follow contacts of new cases may address this need. Similarly, the important question of whether those who do shed MERS-CoV RNA for extended periods are infectious while appearing well, continues to go unanswered. It is even unclear just how many 'asymptomatic' infections have been described and reported correctly which in turn raises questions about the reliability of other clinical data collection to date. While the basic virology of MERS-CoV has advanced over the course of the past three years, understanding what is happening in, and the interplay between, camel, environment and human is still in its infancy. Additional file 1: Figure S1 . The

In what animals MERS-CoV sequences have been found ?

MERS coronavirus: diagnostics, epidemiology and transmission https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4687373/ SHA: f6fcf1a99cbd073c5821d1c4ffa3f2c6daf8ae29 Authors: Mackay, Ian M.; Arden, Katherine E. Date: 2015-12-22 DOI: 10.1186/s12985-015-0439-5 License: cc-by Abstract: The first known cases of Middle East respiratory syndrome (MERS), associated with infection by a novel coronavirus (CoV), occurred in 2012 in Jordan but were reported retrospectively. The case first to be publicly reported was from Jeddah, in the Kingdom of Saudi Arabia (KSA). Since then, MERS-CoV sequences have been found in a bat and in many dromedary camels (DC). MERS-CoV is enzootic in DC across the Arabian Peninsula and in parts of Africa, causing mild upper respiratory tract illness in its camel reservoir and sporadic, but relatively rare human infections. Precisely how virus transmits to humans remains unknown but close and lengthy exposure appears to be a requirement. The KSA is the focal point of MERS, with the majority of human cases. In humans, MERS is mostly known as a lower respiratory tract (LRT) disease involving fever, cough, breathing difficulties and pneumonia that may progress to acute respiratory distress syndrome, multiorgan failure and death in 20 % to 40 % of those infected. However, MERS-CoV has also been detected in mild and influenza-like illnesses and in those with no signs or symptoms. Older males most obviously suffer severe disease and MERS patients often have comorbidities. Compared to severe acute respiratory syndrome (SARS), another sometimes- fatal zoonotic coronavirus disease that has since disappeared, MERS progresses more rapidly to respiratory failure and acute kidney injury (it also has an affinity for growth in kidney cells under laboratory conditions), is more frequently reported in patients with underlying disease and is more often fatal. Most human cases of MERS have been linked to lapses in infection prevention and control (IPC) in healthcare settings, with approximately 20 % of all virus detections reported among healthcare workers (HCWs) and higher exposures in those with occupations that bring them into close contact with camels. Sero-surveys have found widespread evidence of past infection in adult camels and limited past exposure among humans. Sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics have been available almost from the start of the emergence of MERS. While the basic virology of MERS-CoV has advanced over the past three years, understanding of the interplay between camel, environment, and human remains limited. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1186/s12985-015-0439-5) contains supplementary material, which is available to authorized users. Text: An email from Dr Ali Mohamed Zaki, an Egyptian virologist working at the Dr Soliman Fakeeh Hospital in Jeddah in the Kingdom of Saudi Arabia (KSA) announced the first culture of a new coronavirus to the world. The email was published on the website of the professional emerging diseases (ProMED) network on 20 th September 2012 [1] (Fig. 1) and described the first reported case, a 60 year old man from Bisha in the KSA. This information led to the rapid discovery of a second case of the virus, this time in an ill patient in the United Kingdom, who had been transferred from Qatar for care [2] . The new virus was initially called novel coronavirus (nCoV) and subsequentlty entitled the Middle East respiratoy syndrome coronavirus (MERS-CoV). As of 2 nd of September 2015, there have been 1,493 detections of viral RNA or virus-specific antibodies across 26 countries (Additional file 1: Figure S1 ) confirmed by the World Health Organization (WHO), with over a third of the positive people dying (at least 527, 35 %) [3] . Since that first report, a slow discovery process over the following two to three years revealed a virus that had infected over 90 % of adult dromedary camels (DC; Camelus dromedarius) in the KSA [4] , also DCs across the Arabian Peninsula and parts of Africa that are a source of DC imports for the KSA [5] . To date, MERS-CoV has not been detected in DCs tested in zoos or herds from other parts of the world [6] [7] [8] [9] . Occasionally, virus is transmitted from infected DCs to exposed humans. Subsequent transmission to other humans requires relatively close and prolonged exposure [10] . The first viral isolate was patented and concerns were raised that this would restrict access to both the virus and to viral diagnostics [11, 12] . However, sensitive, validated reverse transcriptase real-time polymerase chain reaction (RT-rtPCR)-based diagnostics were quickly described and virus was made freely available subject to routine biosafety considerations [13] . Subsequent epidemiology and research has identified the cell receptor as exopeptidase dipeptidyl peptidase 4 (DPP4; also called CD26); that MERS-CoV has a broad tropism, replicating better in some cells lines and eliciting a more proinflammatory response than SARS-CoV; is widespread in DCs; has the potential to infect other animals and that MERS kills its human host more often than SARS did (20-40 % versus 9 % for SARS [14] ) [15] [16] [17] [18] [19] . In humans, overt disease was given the name Middle East respiratory syndrome, with the acronym MERS. From intermittent animal-to-human spill-over events, the MERS-CoV spreads sporadically among people, causing more severe disease among older adults, especially males, with pre-existing diseases. The spread of MERS-CoV among humans has often been associated with outbreaks in hospitals, with around 20 % of all cases to date involving healthcare workers (HCWs). Although DCs appear to suffer the equivalent of a 'common cold' from MERS-CoV infection, in humans, the virus can be a more serious and opportunistic pathogen associated with the death of up to 40 % of reported cases. It has yet to be established whether infections thought to have been acquired from an animal source produce a more severe outcome than those spread between humans [20] . Studies have established that the mean incubation period for MERS is five to six days, ranging from two to 16 days, with 13 to 14 days between when illness begins in one person and subsequently spreads to another [21] [22] [23] [24] . Among those with progressive illness, the median time to death is 11 to 13 days, ranging from five to 27 days [23, 24] . Fever and gastrointestinal symptoms may form a prodrome, after which symptoms decline, only to be followed by a more severe systemic and respiratory syndrome [25, 26] . The first WHO case definition [27] defined probable cases of MERS based on the presence of febrile illness, cough and requirement for hospitalization with suspicion of lower respiratory tract (LRT) involvement. It also included roles for contact with a probable or confirmed case or for travel or residence within the Arabian Peninsula. If strictly adhered to, only the severe syndrome would be subject to laboratory testing, which was the paradigm early on [21] . From July 2013, the revised WHO case definition included the importance of seeking out and understanding the role of asymptomatic cases and from June 2014, the WHO definition more clearly stated that a confirmed case included any person whose sample was RT-PCR positive for MERS-CoV, or who produced a seroconversion, irrespective of clinical signs and symptoms. [28] [29] [30] Apart from the WHO and the KSA Ministry of Health reports, asymptomatic or subclinical cases of MERS-CoV infection were documented in the scientific literature although not always as often as occurred early on [31, 32] . The KSA definition of a case became more strict on 13 th May 2014, relying on the presence of both clinical features and laboratory confirmation [33] . Testing of asymptomatic people was recommended against from December 2014 [34] , reinforced by a case definition released by the KSA Ministry of Health in June 2015 [35] . The KSA has been the source of 79 % of human cases. Severe MERS is notable for its impact among older men with comorbid diseases including diabetes mellitus, cirrhosis and various lung, renal and cardiac conditions [36] [37] [38] . Interestingly in June 2015, an outbreak in South Korea followed a similar distribution [39, 40] . Among laboratory confirmed cases, fever, cough and upper respiratory tract (URT) signs and symptoms usually occur first, followed within a week by progressive LRT distress and lymphopaenia [37] . Patients often present to a hospital with pneumonia, or worse, and secondary bacterial infections have been reported [37, 41] . Disease can progress to acute respiratory distress syndrome and multiorgan system failure [37] . MERS has reportedly killed approximately 35 % of all reported cases, 42 % of cases in the KSA, yet only 19 % of cases in South Korea, where mortality ranged from 7 % among younger age groups to 40 % among those aged 60 years and above [42] ; all may be inflated values with asymptomatic or mild infections sometimes not sought or not reported [34] . General supportive care is key to managing severe cases [43] . Children under the age of 14 years are rarely reported to be positive for MERS-CoV, comprising only 1.1 % (n = 16) of total reported cases. Between 1 st September 2012 and 2 nd December 2013, a study described the then tally of paediatric cases in the KSA, which stood at 11 (two to 16 years of age; median 13 years); nine were asymptomatic (72 %) and one infant died [44] . In Amman, Jordan, 1,005 samples from hospitalized children under the age of two years with fever and/or respiratory signs and symptoms were tested but none were positive for MERS-CoV RNA, despite being collected at a similar time to the first known outbreak of MERS-CoV in the neighbouring town of Al-Zarqa [45] . A second trimester stillbirth occurred in a pregnant woman during an acute respiratory illness and while not RT-rtPCR positive, the mother did subsequently develop antibodies to MERS-CoV, suggestive of recent infection [46] . Her exposure history to a MERS-CoV RT-rtPCR positive relative and an antibody-reactive husband, her incubation period and her symptom history met the WHO criteria for being a probable MERS-CoV case [46] . Diagnostic methods were published within days of the ProMED email announcing the first MERS case [47] , including several now gold standard in-house RT-rtPCR assays (Fig. 2 ) as well as virus culture in Vero and LLC-MK2 cells [18, 47, 48] . A colorectal adenocarcinoma (Caco-2) epithelial cell line has since been recommended for isolation of infections MERS-CoV [49] . We previously [18] .). Open reading frames are indicated as yellow rectangles bracketed by terminal untranslated regions (UTR; grey rectangles). FS-frame-shift. Predicted regions encompassing recombination break-points are indicated by orange pills. Created using Geneious v8.1 [211] and annotated using Adobe Illustrator. Beneath this is a schematic depicting the location of RT-PCR primers (blue arrows indicate direction) and oligoprobes (green rectangles) used in the earliest RT-rtPCR screening assays and conventional, semi-nested (three primers) RT-PCR confirmatory sequencing assays [47, 48] . Publication order is noted by first [27 th September 2012; red] and second [6 th December 2012; orange] coloured rectangles; both from Corman et al. [47, 48] Those assays recommended by the WHO are highlighted underneath by yellow dots [53] . The NSeq reverse primer has consistently contained one sequence mismatch with some MERS-CoV variants. An altered version of that from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] reviewed the broad tropism of MERS-CoV [5] . However, as is well described, cell culture is a slow, specialised and insensitive method [50] while PCR-based techniques are the preferred method for MERS-CoV detection. The first open reading frames (ORF 1a and 1b; Fig. 2 ) have become a key diagnostic and taxonomic target for CoV species identification. With less than 80 % identity between the amino acid sequence of MERS ORF 1ab and betacoronavirus relatives, Tylonycteris bat HKU4 and Pipistrellus bat HKU5, it can be concluded that it is a novel and distinct virus. MERS-CoV is predicted to encode ten open reading frames with 5' and 3' untranslated regions [51] . The structural proteins include the spike (S), envelope (E), membrane (M) and nucleocapsid (N) [52] . The products of ORF1a and ORF1b are predicted to encode nonstructural proteins. The majority of specimen testing to date has employed validated RT-rtPCR assays shown to be sensitive and specific [47, 48, 53] . The RealStar® kit uses these WHOrecommended assays [54] . The target sequences of these screening assays have not changed among genomes examined until at least mid-2015 (IMM observation). Other RT-rtPCR assays have been developed and validated for use as laboratory-based diagnostic tools [55] [56] [57] . Additionally, loop-mediated [58, 59] or recombinase polymerase [60] isothermal assays have been designed for field deployment. The detection of MERS-CoV antigen has not been common to date but the combination of short turnaround time from test to result, high throughput and identification of viral proteins makes this an attractive option. Detection of viral proteins rather than viral RNA indicates the likely presence of infectious virus. The first rapid immunochromatographic tool described could detect recombinant MERS-CoV nucleocapsid protein from DC nasal swabs with 94 % sensitivity and 100 % specificity compared to RT-rtPCR [61] . A different approach used a monoclonal antibody-based capture ELISA targeting the MERS-CoV nucleocapsid protein with a sensitivity of 10 3 TCID 50 and 100 % specificity [62] . Demonstration of a seroconversion to a MERS-CoV infection meets the current WHO definition of a case so optimized and thoroughly validated sero-assays employed alongside good clinical histories are useful to both identify prior MERS-CoV infection and help support transmission studies. Because serology testing is, by its nature, retrospective, it is usual to detect a viral footprint, in the form of antibodies, in the absence of any signs or symptoms of disease and often in the absence of any viral RNA [63] . Strategic, widespread sero-surveys of humans using samples collected after 2012 are infrequent. Much of the Arabian Peninsula and all of the Horn of Africa lack baseline data describing the proportion of the community who may have been infected by a MERS-CoV. However, sero-surveys have had widespread use in elucidating the role of DCs as a transmission source for MERS-CoV. Because of the identity shared between DC and human MERS-CoV (see Molecular epidemiology: using genomes to understand outbreaks), serological assays for DC sero-surveys should be transferrable to human screening with minimal re-configuration. Also, no diagnostically relevant variation in neutralization activity have been found from among a range of circulating tested MERS-CoV isolates and sera, so whole virus or specific protein-based sero-assays should perform equivalently in detecting serological responses to the single MERS-CoV serotype [49] . The development of robust serological assays requires reliable panels of wellcharacterized animal or human sera, including those positive for antibodies specific to MERS-CoV, as well as to likely sources of cross-reaction [64] . Obtaining these materials was problematic and slowed the development and commercialization of antibody detection assays for human testing [64] . A number of commercial ELISA kits, immunofluorescent assays (IFA) kits, recombinant proteins and monoclonal antibodies have been released [31, [65] [66] [67] [68] . Initially, conventional IFAs were used for human sero-surveys. These relied on MERS-CoV-infected cell culture as an antigen source, detecting the presence of human anti-MERS-CoV IgG, IgM or neutralizing antibodies in human samples [18, 48, 69] . No sign of MERS-CoV antibodies was found among 2,400 sera from patients visiting Hospital in Jeddah, from 2010 through 2012, prior to the description of MERS-CoV [18] . Nor did IFA methods detect any sign of prior MERS-CoV infection among a small sample of 130 healthy blood donors from another Hospital in Jeddah (collected between Jan and Dec 2012) [70] . Of 226 slaughterhouse workers, only eight (3.5 %) were positive by IFA, and those sera could not be confirmed by virus neutralization (NT) test. The study indicated that HCoV-HKU1 was a likely source of crossreactive antigen in the whole virus IFA [70] . Whole virus MERS-CoV IFA also suffered from some cross-reactivity with convalescent SARS patient sera and this could not be resolved by an NT test which was also cross-reactive [71] . IFA using recombinant proteins instead of whole-virus IFA, has been shown to be a more specific tool [31] . Since asymptomatic zoonoses have been posited [72] , an absence of antibodies to MERS-CoV among some humans who have regular and close contact with camels may reflect the rarity of actively infected animals at butcheries, a limited transmission risk associated with slaughtering DCs [70] , a pre-existing cross-protective immune status or some other factor(s) resulting in a low risk of disease and concurrent seroconversion developing after exposure in this group. IFA using recombinant proteins instead. Some sero-assays have bypassed the risks of working with infectious virus by creating transfected cells expressing recombinant portions of the MERS-CoV nucleocapsid and spike proteins [48, 73] , or using a recombinant lentivirus expressing MERS-CoV spike protein and luciferase [74, 75] . A pseudo particle neutralization (ppNT) assay has seen widespread used in animal studies and was at least as sensitive as the traditional microneutralization (MNT) test. [10, 74, [76] [77] [78] ] Studies using small sample numbers and ppNT found no evidence of MERS-CoV neutralizing antibody in sera from 158 children with LRT infections between May 2010 and May 2011, 110 sera from 19 to 52 year old male blood donors and 300 selfidentified animal workers from the Jazan Region of the KSA during 2012 [79, 80] . Similarly, a study of four herdsmen in contact with an infected DC herd in Al-Ahsa, eight people who had intermittent contact with the herd, 30 veterinary surgeons and support staff who were not exposed to the herd, three unprotected abattoir workers in Al-Ahsa and 146 controls who were not exposed to DCs in any professional role, found none with serological evidence of past MERS-CoV infection using the ppNT assay [10] . A delay in the neutralizing antibody response to MERS-CoV infection was associated with increased disease severity in South Korea cases with most responses detectable by week three of illness while others, even though disease was severe, did not respond for four or more weeks [81] . The implications for our ability to detect any response in mild or asymptomatic cases was not explored but may be a signifcant factor in understanding exposure in the wider community. A Jordanian outbreak of acute LRT disease in a hospital in 2012 was retrospectively found to be associated with MERS-CoV infection, initially using RT-rtPCR, but subsequently, and on a larger scale, through positivity by ELISA and IFA or MNT test. [46, 82, 83] This outbreak predated the first case of MERS in the KSA. The ELISA used a recombinant nucleocapsid protein from the group 2 betacoronavirus bat-CoV HKU5 to identify antibodies against the equivalent crossreactive MERS-CoV protein [71] . It was validated using 545 sera collected from people with prior HCoV-OC43, HCoV-229E, SARS-CoV, HCoV-NL63, HRV, HMPV or influenza A(H1N1) infections but was reportedly less specific than the recombinant IFA discussed above. It was still considered an applicable tool for screening large sample numbers [82] . A protein microarray expressing the S1 protein subunit has also been validated and widely used for DC testing [5, 84] . Detection of MERS-CoV infection using ELISA or S1 subunit protein microarray [84] is usually followed by confirmatory IFA and/ or a plaque-reduction neutralization (PRNT) [69, 70, 85] or MNT test. [74, 85, 86] This confirmatory process aims toensure the antibodies detected are able to specifically neutralize the intended virus and are not more broadly reactive to other coronaviruses found in DCs (bovine CoV, BCoV) or humans (HCoV-OC43, HCoV-229E, HCoV-NL63, HCoV-HKU1, SARS-CoV). In the largest study of human sera, a tiered diagnostic process assigned both recombinant IFA and recombinant ELISA positive sera to 'stage 1' seropositivity. A stage 2 seropositive result additionally required a suitably titred PRNT result [87] . The study found 15 sera collected in 2012 to 2013 from 10,009 (0.2 %) people in 13 KSA provinces contained MERS-CoV antibodies, but significantly higher proportions in occurred in camel shepherds (two of 87; 2.3 %) and slaughterhouse workers (five of 140; 3.6 %) [87] . Contemporary surveys are needed. MERS-CoV does not appear to be easily transmitted from DCs to humans, or perhaps it is [72] , but generally does not trigger a detectable immune response if only mild disease or asymptomatic infection results. Serology assays are in need of further validation in this area so care is required when moving newly developed diagnostic serology algorithms from a research setting to one that informs public health decisions. This was reinforced when a false positive US case, purported to have been infected after a handshake and two face-to-face meetings, did not withstand further confirmatory analysis using a more specific, NT assay and was subsequently retracted [88, 89] . The WHO recommends sampling from the LRT for MERS-CoV RT-rtPCR testing, especially when sample collection is delayed by a week or more after onset of symptoms. [53] LRT samples are also best for attempting isolation of infectious virus, although the success of culture is reduced when disease persists [49] . Recommended sample types include bronchoalveolar lavage (BAL), tracheal/tracheobronchial aspirate, pleural fluid and sputum [53, 90] . Fresh samples yield better diagnostic results than refrigerated material [69] and if delays in testing of ≥72 h are likely, samples (except for blood) should be frozen at −70°C [90] . If available, lung biopsy or autopsy tissues can also be tested [53] . The URT is a less invasive and more convenient sampling site however, and an oropharyngeal and throat swab or a nasopharyngeal aspirate/wash are recommended when URT sampling is to be conducted [90] . Paired sera, collected two to three weeks apart are preferable for serological testing while a single sample is suggested to be sufficient if collected two weeks after onset of disease or a single serum collected during the first 10-12 days if conducting RT-rtPCR [53, 90] . Human urine and stool have been found to contain MERS-CoV RNA 12 to 26 days after symptom onset [25, 69, 91] and are listed as samples that should be considered [53, 90] . In two cases that arrived in the Netherlands, urine was RT-rtPCR negative but faeces was weakly positive and sera were RT-rtPCR positive for five days or more [25] . The finding of MERS-CoV viral RNA in serum provides an avenue for retrospective PCR-based studies if respiratory samples are unavailable [83] . RNAaemia may also correlate with disease severity; signs of virus were cleared from the serum of a recovered patient, yet lingered until the death of another [92] . Clinically suspected MERS cases may return negative results by RT-rtPCR. Data have shown one or more negative URT samples may be contradicted by further URT sampling or the use of LRT samples, which is preferred [2, 43, 93] . Higher viral loads occur in the LRT compared to the URT. [22, 69, 88, 94] This fits with the observation that the majority of disease symptoms are reported to manifest as systemic and LRT disease [21] . However, on occasion, even LRT specimens from MERS cases may initially be negative, only to later become positive by RT-PCR [95] . This may be due to poor sampling when a cough is absent or non-productive or because the viral load is low [95] . Despite this both the largest human MERS-CoV studies [32, [96] [97] [98] and smaller ones [22, 25, 99] , use samples from the URT. It is then noteworthy that one study reported an association between higher loads in the URT and worse clinical outcome including intensive care and death [94] . At writing, no human data exist to define whether the virus replicates solely or preferentially in the LRT or URT, or replicates in other human tissues in vivo although MERS-CoV RNA has been detected from both the URT and LRT in a macaque monkey model [100] .The distribution of DPP4 in the human upper airways is also not well described. Individual human case studies report long periods of viral shedding, sometimes intermittently and not necessarily linked to the presence of disease symptoms. [25, 69, 99, 101] In one instance, a HCW shed viral RNA for 42 days in the absence of disease [99] . It is an area of high priority to better understand whether such cases are able to infect others. Over three quarters of MERS cases shed viral RNA in their LRT specimens (tracheal aspirates and sputum) for at least 30 days, while only 30 % of contacts were still shedding RNA in their URT specimens [91, 102] . In the only study to examine the effect of sample type on molecular analysis, 64 nasopharyngeal aspirates (NPA; an URT sample), 30 tracheal aspirates, 13 sputa and three BAL were examined. The tracheal aspirates and BAL returned the highest viral load values followed by NPA and sputum. Unsurprisingly, higher viral loads generally paralleled whole genome sequencing and culture success and, in NPA testing, were significantly correlated with severe disease and death [49, 94, 103] . This study demonstrated the importance of LRT sampling for whole genome sequencing. When tested, samples positive for MERS-CoV are often negative for other pathogens [2, 25, 93, 104] . However, many studies make no mention of additional testing for endemic human respiratory viruses [21, 23, 73, 105] . When viruses are sought, they have included human herpesvirus (HHV), rhinoviruses (HRV), enteroviruses (EV), respiratory syncytial virus (RSV), parainfluenzavirus types 1, 2 and 3 (PIVs),influenzaviruses (IFVs), endemic HCoVs, adenoviruses (AdVs) metapneumovirus (MPV) and influenza A\H1N1 virus; co-detections with MERS-CoV have been found on occasion [2, 22, 37, 69, 97] . Bacterial testing is sometimes included (for example, for Legionella and Pneumococcus) but the impact of bacterial co-presence is also unclear [22, [104] [105] [106] . Further testing of the LRT sample from the first MERS case used IFA to screen for some viruses (negative for IFV, PIVs, RSV and AdVs) and RT-PCR for others (negative for AdV, EVs, MPV and HHVs) [18] . RT-PCR also detected MERS-CoV. The WHO strongly recommends testing for other respiratory pathogens [53] but with this recommendation often discounted, there are limited data to address the occurrence and impact of co-infections or alternative viral diagnoses among both MERS cases and their contacts. Little is known of other causes of MERS-like pneumonia in the KSA or of the general burden of disease due to the known classical respiratory viruses. Testing of adult pilgrims performing the Hajj in 2012 to 2014 has not detected any MERS-CoV. In 2012, nasal swabs from 154 pilgrims collected prior to leaving for or departing from the KSA were tested [47] . In 2013, testing was significantly scaled up with 5,235 nasopharyngeal swabs from 3,210 incoming pilgrims and 2,025 swabs from outgoing pilgrims tested [98] . It should be noted that most pilgrims arrived from MERS-free countries. A further 114 swabs were taken from pilgrims with influenza-like illness [96, 107] . In earlier Hajj gatherings, it was found that influenza viruses circulated widely, whilst other viruses, often rhinoviruses, circulated more selectively, interpreted as indicating their importation along with foreign pilgrims. [107] [108] [109] Over time, increased influenza vaccination has been credited for a fall in the prevalence of influenza like illnesses among Hajj pilgrims. [110] A LRT sample is often not collected for these studies [98, 107, 109] , so false negative findings are a possibility although little is known about the initial site of MERS-CoV infection and replication; it may have been assumed it was the LRT because disease was first noticed there but the URT may be the site of the earliest replication. In Jeddah between March and July 2014 (hereafter called the Jeddah-2014 outbreak; Fig. 3 ), there was a rapid increase in MERS cases, accompanied by intense screening; approximately 5,000 samples from in and around the region were tested in a month yielding around 140 MERS-CoV detections (~3 % prevalence) [111] . Among 5,065 individuals sampled and tested across the KSA between October 2012 and September 2013,108 (2.1 %) detections were made in a hospital-centric population which included hospitalized cases (n = 2,908; 57.4 %), their families (n = 462; 9.1 %) and associated HCWs (n = 1,695; 33.5 %) [32] . Among the detections, 19 (17.8 %) were HCWs and 10 (9.3 %) were family contacts [32] . The 2-3 % prevalence of active MERS-CoV infections is not dissimilar to the hospital-based prevalence of other human CoVs. [112] However, the proportion of deaths among those infected with MERS-CoV is much higher than that known for the HCoVs NL63, HKU1, 229E or OC43 in other countries, and even above that for SARS-CoV; it is not a virus that could reasonably be described as a "storm in a teacup". It is the low transmission rate that has prevented worldwide spread, despite many "opportunities". Very early in the MERS outbreak, some animals were highly regarded as either the reservoir or intermediate host(s) of MERS-CoV with three of the first five cases having contact with DCs [73, 113, 114] . Today, animal MERS-CoV infections must be reported to the world organization for animal health as an emerging disease [115] . A summary of the first MERS cases reported by the WHO defined animal contact with humans as being direct and within 10 days prior to symptom onset [20] . This definition made no specific allowance for acquisition from DCs through a droplet-based route, which is very likely route for acquisition of a virus that initially and predominantly causes respiratory disease [23] . Camels are known to produce high levels of MERS-CoV RNA in their URT and lungs [116] . Providing support for a droplet transmission route and perhaps indicating the presence of RNA in smaller, drier droplet nuclei, MERS-CoV RNA was identified in a high volume air sample collected from a barn housing an infected DC [117] . The precise source from which humans acquire MERS-CoV remains poorly studied but it seems likely that animal and human behavioural factors may play roles (Fig. 3) [118] . These factors may prove important for human cases who do not describe any DC contact [119] nor any contact with a confirmed case. Whether the WHO definition of animal contact is sufficient to identify exposure to this respiratory virus remains unclear. Wording focuses on consumption of DC products but does not specifically ascribe risk to a droplet route for acquisition of MERS-CoV from DC [120] . Some MERS patients are listed in WHO disease notices as being in proximity to DCs or farms, but the individuals have not described coming into contact with the animals. No alternative path for acquiring infection is reported in many of these instances. What constitutes a definition of "contact" during these interviews has been defined for one study [72] . Despite this lack of clarity, the WHO consider that evidence linking MERS-CoV transmission between DCs to humans is irrefutable (Fig. 4) [120] . The possibility that bats were an animal host of MERS-CoV was initially widely discussed because of the existing diversity of coronaviruses known to reside among them [121] [122] [123] [124] . Conclusive evidence supporting bats as a source for human infections by MERS-CoV has yet to be found, but bats do appear to host ancestral representatives [53, 125] . However, these are not variants of the same virus nor always within the same phylogenetic lineage as MERS-CoV; they are each a genetically distinct virus. Bat-to-human infection by MERS-CoV is a purely speculative event. The only piece of MERS-CoV-specific evidence pointing to bats originates from amplification of a 190 nt fragment of the RNAdependent RNA polymerase gene of the MERS-CoV genome, identified in a faecal pellet from an insectivorous Emballonuridae bat, Taphozous perforatus found in Bisha, the KSA [121] . While very short, the sequence of the fragment defined it as a diagnostic discovery. Subsequently a link to DCs was reported [85] and that link has matured into a verified association [38, 126] (Fig. 4) . (See figure on previous page.) Fig. 3 Monthly detections of MERS-CoV (blue bars) and of cases who died (red bars) with some dates of interest marked for 2012 to 4 th September 2015. An approximation of when DC calving season [128] and when recently born DCs are weaned is indicated. Spring (green) and summer (orange) in the Arabian Peninsula are also shaded. Note the left-hand y-axis scale for 2014 and 2015 which is greater than for 2012/13. Sources of these public data include the WHO, Ministries of Health and FluTrackers [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] . Modified and reprinted from Mackay IM, Arden KE. Middle East respiratory syndrome: An emerging coronavirus infection tracked by the crowd. Virus Res 2015 Vol 202:60-88 with permission from Elsevier [5] DCs, which make up 95 % of all camels, have a central presence in the Arabian Peninsula where human-DC contact ranges from little to close [119] . Contact may be commonplace and could occur in variety of ways (Fig. 4a) . There are several large well-attended festivals, races, sales and parades which feature DCs and DCs are also kept and bred close to populated areas in the KSA [127, 128] . DC milk and meat are widely consumed and the older DC is an animal of ritual significance after the Hajj pilgrimage [129] . However, MERS-CoV infection frequency is reportedly much lower than is the widespread and frequent habit of eating, drinking and preparing DC products. Daily ingestion of fresh unpasteurized DC milk is common among the desert Bedouin and many others in the KSA. DC urine is also consumed or used for supposed health benefits. Despite camel butchery being a local occupation, neither butchers nor other at-risk groups are identifiable among MERS cases; this may simply be a reporting issue rather than an unexplainable absence of MERS. A small case-control study published in 2015 identified direct DC contact, and not ingestion of products, to be associated with onset of MERS [38] . The first sero-survey of livestock living in the Middle East region was conducted during 2012-2013 [85] . DCs were sampled from a mostly Canary Island-born herd and from Omani DCs (originally imported from the Horn of Africa) [85] . A neutralising antibody assay found only 10 % of strongly seropositive Canary Island [5] . b Camel-to-human infections appear to be infrequent, while human-to-human spread of infection is regularly facilitated by poor IPC in healthcare settings where transmission is amplified, accounting for the bulk of cases. There are human MERS cases that do not fall into either category of source and it is unclear if these acquired infection through some entirely separate route, or from cases that escaped diagnosis. c Hypothetical ways in which subclinical (when infection may not meet a previously defined clinical threshold of signs and/or symptoms) or asymptomatic (no obvious signs or measured, noticed or recalled symptoms of illness) MERS-CoV infection may be implicated in transmission DC sera could neutralise MERS-CoV while all Omani DC sera had high levels of specific MERS-CoV neutralizing antibody [85] . This indicated that DCs had in the past been infected by MERS-CoV, or a very similar virus. Since this study, a host of peer-reviewed reports have looked at both DCs and other animals, and the possibility that they may host MERS-CoV infection. Seropositive DCs have been found throughout the Arabian Peninsula including Oman, the KSA, Qatar, Jordan, the United Arab Emirates (UAE), Kuwait as well as Sudan, Somalia, Egypt, Tunisia, Nigeria, Kenya and Ethiopia in Africa and the Canary Islands [85, [130] [131] [132] [133] [134] . Other animals tested include sheep, cows, pigs, horses, donkeys, mules, birds, water buffalo, goats, Bactrian camels, llamas and guanaco (south American camelids) but none had detectable neutralising antibody against MERS-CoV [4, 74, 78, 85, 86, 135, 136] . No virology or serology studies of human samples from areas in Africa where there are camels with a history of MERS-CoV have been reported to date. However,an absence of unexplained pneumonia that may be attributable to MERS-CoV infection may not signal the absence of virus among humans in each country but simply reflect a lack of expensive epidemiology studies conducted by resource-poor countries. It is thus unclear whether MERS-CoV, or an antigenically related CoV, is an unrecognized pathogen in these regions, perhaps circulating for even longer than it has been known in the Arabian Peninsula [133] . MERS-CoV RNA has also been detected in DC samples, and recovery of infectious virus has also been achieved from DC samples [4, 77, 117, 132, [137] [138] [139] [140] [141] . From some of these, full or majority length genomes of MERS-CoV have been sequenced [77, 137, 138] . DC versions of MERS-CoV were found to be as similar to each other, as were variants detected from different humans over time and across distance. Antibody screening assays have also detected crossreactive antibodies in sera. These were identified as such by screening sera against similar viruses, for example BCoV or HCoV-OC43 (as an antigenic facsimile for BCoV). It is possible that other MERS-CoV-like viruses also reside within DCs, but this does not detract from the definitive finding of MERS-CoV genetic sequences in both DCs and humans [117, 142, 143] . Screening studies have shown that juvenile DCs are more often positive for virus or viral RNA while older DCs are more likely to be seropositive and RNA or virus negative [76, 77, 144] . In adult DCs, MERS-CoV RNA has been detected among animals with pre-existing antibody, suggesting re-infection is possible [77, 144] . Viral loads among positive DCs can be very high [4, 76, 77, 139, 144] and DCs have been found positive both when ill with URT respiratory signs [77, 117, 142, 145] or when apparently healthy [137] . These findings indicate DCs host natural MERS-CoV infections. Furthermore, stored DC sera have revealed signs of MERS-CoV in DCs which date back over three decades (the earliest collected in 1983) [4, 133, 135] . Older sera have not been tested and so precisely how long DCs have been afflicted by MERS-CoV, whether the virus is enzootic among them, introduced to them decades or centuries ago from bats in Africa or the Arabian Peninsula, or they are the subject of regular but short-lived viral incursions from an as yet unknown host, cannot be answered. Researchers sought to determine a direction for infection; were DCs transmitting virus to humans or were humans infecting DCs? At a Qatari site, a farm owner and his employee became ill in mid-October 2013 and tested positive for MERS-CoV RNA in a sputum and throat swab sample, respectively. RT-rtPCRs found MERS-CoV RNA in 11 of 14 positive DC nasal swabs at the farm; six (43 %) positive by two or more assays [138] . The results indicated a recent outbreak had occurred in this herd; the first indication of MERS-CoV RNA found within DCs with a temporal association to human infections. Three positive DC samples were confirmed by sequencing a 358 nt portion of the spike gene; these sequences were identical to each other, again with close homology to other human and DC MERS-CoV sequences [138] . The DCs and human contacts yielded ORF1a and ORF4b sequences differing by only a single nucleotide each, clustering closely with the Hafr-Al-Batin_1_2013 variant [138] . Subsequent case studies found evidence of a concurrent human and DC infection and the direction of that infection was inferred to be from the ill DCs and to their human owners [117, 142, 146] . Partial genome sequences indicated that a human and a MERS-CoV RT-rtPCR positive DC had been infected by a variant of the same virus, harbouring the same distinct pattern of nucleotide polymorphisms. [142] All nine DC in the owner's herd, serially sampled, reacted in a recombinant S1 antigen ELISA, with the two animals that had been RT-rtPCR positive showing a small, verifiable rise in antibody titre [142] . A rise in titre theoretically begins 10 to 21 days after DC infection [142] . The authors suggested that the rise in titre in DC sera which occurred alongside a declining RNA load, while the patient was actively ill and hospitalized, indicated that the DCs were infected first followed by the owner [117, 142] . BCoV antibodies were also present, and rising in one of the two RT-rtPCR positive animals but no animal's antibodies could neutralise BCoV infection [142] . Camel calving season occurs in the winter months (between late October and late February; Fig. 3 ) and this may be a time when there is increased risk to humans of spill-over due to new infections among naïve DC populations [128] . What role maternal camel antibody might play in delaying infection of calves remains unknown [128, 142] . Juvenile DCs appear to host active infection more often than adult DCs and thus the sacrificial slaughter of DCs, which must be five years of age or older (termed a thane), may not be accompanied by significant risk of exposure to infection. In contrast to earlier results, slaughterhouse workers who kill both younger and older DCs, may be an occupational group with significantly higher incidence of seropositivity to MERS-CoV when animals have active MERS-CoV infections [129, 139, [147] [148] [149] . Expanded virological investigations of African DCs may lead to more seropositive animals and geographic areas in which humans may be at risk. It is possible that there are areas where humans already harbour MERS-CoV infections that have not been identified because of an absence of laboratory surveillance. Virological investigations of bats may lead to findings of ancestral viruses and viral 'missing links' and identifying any other animal sources of zoonotic spread is important to inform options for reducing human exposures [56, 76] . Infectious MERS-CoV added to DC, goat or cow milk and stored at 4°C could be recovered at least 72 h later and, if stored at 22°C, recovery was possible for up to 48 h [150] . MERS-CoV titre decreased somewhat when recovered from milk at 22°C but pasteurization completely ablated MERS-CoV infectivity [150] . In a subsequent study, MERS-CoV RNA was identified in the milk, nasal secretion and faeces of DCs from Qatar [151] . A single study has examined the ability of MERS-CoV to survive in the environment [150] . Plastic or steel surfaces were inoculated with 10 6 TCID 50 of MERS-CoV at different temperature and relative humidity (RH) and virus recovery was attempted in cell culture. At high ambient temperature (30°C) and low RH (30 %) MERS-CoV remained viable for 24 h [150] . By comparison, a well known and efficently transmitted respiratory virus, influenza A virus, could not be recovered in culture beyond four hours under any conditions [150] . Aerosol experiments found MERS-CoV viability only decreased 7 % at low RH at 20°C. In comparison, influenza A virus decreased by 95 % [150] . MERS-CoV survival is inferior to that previously demonstrated for SARS-CoV [152] . For context, pathogenic bacteria can remain viable and airborne for 45 min in a coughed aerosol and can spread 4 m. MERS-CoV's ability to remain viable over long time periods gives it the capacity to thoroughly contaminate a room's surfaces when occupied by an infected and symptomatic patient [153] . Whether MERS-CoV can remain adrift and infectious for extended periods (truly airborne) remains unknown. Such findings expand our understanding of the possibilities for droplets to transmit respiratory viruses in many settings, including hospital waiting rooms, emergency departments, treatment rooms, open intensive care facilities and private patient rooms. The nature and quality of air exchange, circulation and filtration are important variables in risk measurement and reduction as is the use of negative pressure rooms to contain known cases. Droplet spread between humans is considered the mechanism of human-to-human transmission and the need for droplet precautions was emphasized after the Al-Ahsa hospital, the KSA and the South Korean outbreaks [21, 23, 154, 155] . By extrapolation, aerosol-generating events involving DCs (urination, defecation, and preparation and consumption of DC products) should be factored into risk measurement and reduction efforts and messaged using appropriate context. The provision of evidence supporting the best formulation of personal protective equipment to be worn by HCWs who receive, manage or conduct procedures on infectious cases remains a priority. MERS-CoV was found and characterized because of its apparent association with severe, and therefore more obvious, illness in humans; we were the canaries in the coal mine. Sero-assays and prospective cohort studies have yet to determine the extent to which milder or asymptomatic cases contribute to MERS-CoV transmission chains. However, transmission of MERS-CoV is defined as sporadic (not sustained), intra-familial, often healthcare associated, inefficient and requiring close and prolonged contact [22, 31, 63, 93, 97, 102, 156] In a household study, 14 of 280 (5 %) contacts of 26 MERS-CoV positive index patients were RNA or antibody positive; the rate of general transmission, even in outbreaks is around 3 % [31] . It seems that the majority of human cases of MERS-CoV, even when numbers appear to increase suddenly, do not readily transmit to more than one other human so to date, the localized epidemic of MERS-CoV has not been self-sustaining [157] [158] [159] [160] [161] . That is to say, the basic reproduction number (R 0 ) -the average number of infections caused by one infected individual in a fully susceptible populationhas been close to one throughout various clusters and outbreaks. If R 0 was greater than 1, a sustained increase in case numbers would be expected. Some R o calculations may be affected by incomplete case contact tracing, limited community testing and how a case is defined. That MERS has had a constant presence in the Arabian Peninsula since 2012 is due to ongoing, sporadic spill-over events from DCs amplified by poorly controlled hospital outbreaks. The first known MERS human-to-human transmission event was one characterized by acute LRT disease in a healthcare setting in Jordan. In stark contrast, a sero-survey of HCW who were sometimes in close and prolonged contact with the first, fatal MERS-CoV case in 2012 [162] , found none of the HCW had seroconverted four months later, despite an absence of eye protection and variable compliance with required PPE standards [162] . Early on in the MERS story, samples for testing were mostly collected from patients with severe illness and not those with milder acute respiratory tract infections. Contacts of confirmed MERS cases were often observed for clinical illness, but not tested. These omissions may have confounded our understanding of MERS-CoV transmission and biased early data towards higher numbers of seriously ill and hospitalized patients, inflating the apparent proportion of fatal cases. Case-control studies were not a focus. As testing paradigms changed and contacts were increasingly tested, more asymptomatic and mild infections were recognized [163] . A rise in the cases termed asymptomatic (which enlarge the denominator for calculations of the proportion of fatal cases, defined in [164] ) resulted in a drop in the proportion of fatal cases during the Jeddah-2014 outbreak. Historically, such rises are consistent with changing definitions and laboratory responses and clinical management of a newly discovered virus infection that was first noted only among the severely ill. Upon follow-up, over three-quarters of such MERS-CoV RNA positive people did recall having one or more symptoms at the time, despite being reported as asymptomatic [165] raising some question over the reliability of other reported data. The proportion of fatal MERS cases within the KSA compared to outside the KSA, as well as the age, and sex distribution change in different ways when comparing MERS outbreaks. Approximately 43 % of MERS cases (549 of 1277) in the KSA were fatal betwen 2012 and December 2015 while 21 % (72 of 330) died among those occurring outside of the KSA. The total number of male cases always outnumber females and the proportion of male deaths is always greater than the proportion of females who die. However the proportion of male deaths from total males with MERS is a similar figure to that for females. In the KSA, there is a greater proportion of younger males among cases and deaths than were observed from the 2015 South Korean or the Jeddah-2014 outbreaks (Additional file 2: Figure S2 ). Why these aspects have differed may be due to differences in the time to presentation and diagnosis, the nature and quality of supportive care, the way a person became infected (habits, exposure to a human or zoonotic source, viral load, route of infection) or the extent to which different populations are burdened by underlying diseases [40] . As a group, HCWs comprised 16 % of MERS cases in the KSA and South Korea. It is apparent that the weekly proportion of infected HCWs increases alongside each steep rise in overall detections (Fig. 5) . In May 2013, the WHO published guidelines for IPC during care of probable or confirmed cases of MERS-CoV infection in a healthcare setting [166] . This is explainable because to date, each case rise has been intimately associated with healthcare-facility related outbreaks [118] . These rises in MERS-CoV detections can decrease the average age during each event because HCWs are usually younger than inpatients with MERS. Healthcare facilities have been a regular target for suggested improvements aimed at improving infection prevention and control (IPC) procedures [115, 118] . Most of the analysis of MERS-CoV genetics has been performed using high throughput or "deep" sequencing methods for complete genome deduction [167] [168] [169] . MERS-CoV was the first subject of such widespread use of deep sequencing to study an emerging viral outbreak with global reach. The technique can produce genomic [207] [208] [209] . Earlier and subsequent versions of this chart are maintained on a personal blog [210] length coverage in a single experiment with highly repetitious measurement of each nucleotide position [52, 140] . Despite assays having been published early on, subgenomic sequencing, once the mainstay of viral outbreak studies, has less often been published during MERS-CoV characterization [48] . As more genomes from both humans and DCs have been characterized, two clades have become apparent; A and B (Fig. 6) . Clade A contains only human-derived MERS-CoV genomes from Jordan, while Clade B comprises the majority of human and camel genomes deduced thus far [168] . Two studies during 2015, one looking at Jeddah-2014 MERS-CoV variants and another looking at a variant exported from South Korea to China, have now identified signs of genetic recombination among MERS-CoV variants. While human and camel whole genome sequences have retained >99 % identity with each other, members of genetically distinct lineages can and do swap genetic material when suitable conditions and coinfections co-occur [170] [171] [172] . Shared identity implies that the major source for human acquisition is the DC, rather than another animal, although more testing of other animal species is needed to confirm that conclusion. Over a month, a DC virus sequenced on different occasions did not change at all indicating a degree of genomic stability in its host, supporting that DCs are the natural, rather than intermediate, host for the MERS-CoV we know today [77] . To date, recombination has been localised to breakpoints near the boundary between ORF1a and ORF1b regions, within the spike gene [170] and in the ORF1b region (Fig. 2) [172] . It is not unexpected that recombination should occur since it is well known among other CoVs [124] and because the majority of MERS-CoV whole genomes collected from samples spanning three years (2012-2015) and from humans, camels and different countries have shown close genetic identity to each other, with just enough subtle variation to support outbreak investigations so long as whole genome sequencing is applied [52, 77, 135, 138, 168, [173] [174] [175] . Changes in genome sequence may herald alterations to virus transmissibility, replication, persistence, lethality or response to future drugs. If we have prior knowledge of the impact of genetic changes because of thorough characterization studies, we can closely Fig. 6 The genetic relationship between MERS-CoV nucleotide sequences (downloaded from GenBank using the listed accession numbers and from virological.org [212] ). This neighbour joining tree was created in MEGA v6 using an alignment of human and DCderived MERS-CoV sequences (Geneious v8.1 [211] ). Clades are indicated next to dark (Clade A) or pale (Clade B) blue vertical bars. Camel icons denote genomes from DCs. Healthcare or community outbreaks are boxed and labelled using previously described schemes [212, 213] monitor the genomic regions and better understand any changes in transmission or disease patterns as they occur. Genetic mutations noted during the largest of human outbreaks, Jeddah-2014, did not impart any major replicative or immunomodulatory changes when compared to earlier viral variants in vitro [156, 176] . However, we understand very little of the phenotypic outcomes that result from subtle genetic change in MERS-CoV genomes. To date no clinical relevance or obvious in vivo changes to viral replication, shedding or transmission has been reported or attributed to mutations or to new recombinant viruses [156] . But vigilance and larger, more contemporary and in vivo studies are needed. Genome sequence located to a distinct clade were identified from an Egyptian DC that was probably imported from Sudan. This does not fit into either of the current clades [125, 168, 177] . A virus sequenced from a Neoromicia capensis bat was more closely related to MERS-CoV than other large bat-derived sequences had been to that point, but the genome of a variant of a MERS-CoV has yet to be discovered and deduced from any bat [125] . Analyses of MERS-CoV genomes have shown that most single nucleotide differences among variants were located in the last third of the genome (Fig. 2) , which encodes the spike protein and accessory proteins [168] . At least nine MERS-CoV genomes contained amino acid substitutions in the receptor binding domain (RBD) of the spike protein and codons 158 (N-terminal region), 460 (RBD), 1020 (in heptad repeat 1), 1202 and 1208 bear investigation as markers of adaptive change [140, 169] . The spike protein had not changed in the recombinant MERS-CoV genome identified in China in 2015 but was reported to have varied at a higher rate than that for complete MERS-CoV genomes, among South Korean variants [172, 178] . This highlights that subgenomic regions may not always contain enough genetic diversity to prove useful for differentiating viral variants. Despite this, one assay amplifying a 615 nucleotide fragment of the spike S2 domain gene for Sanger sequencing agreed with the results generated by the sequencing of a some full genomes and was useful to define additional sequence groupings [177] . Genomic sequence can also be used to define the geographic boundaries of a cluster or outbreak and monitor its progress, based on the similarity of the variants found among infected humans and animals when occurring together, or between different sites and times (Fig. 6 ) [169] . This approach was employed when defining the geographically constrained MERS hospital outbreak in Al-Ahsa, which occurred between 1 st April and 23 rd May 2013, as well as clusters in Buraidah and a community outbreak in Hafr Al-Batin, the KSA. Genomic sequencing identified that approximately 12 MERS-CoV detections from a community outbreak in Hafr Al-Batin between June and August 2013 may have been triggered by an index case becoming infected through DC contact [175] . Sequencing MERS-CoV genomes from the 2013 Al-Ahsa hospital outbreak indicated that multiple viral variants contributed to the cases but that most were similar enough to each other to be consistent with human-tohuman transmission. Molecular epidemiology has revealed otherwise hidden links in transmission chains encompassing a period of up to five months [179] . However, most outbreaks have not continued for longer than two to three months and so opportunities for the virus to adapt further to humans through co-infection and sustained serial passage have been rare [169] . In Riyadh-2014, genetic evidence supported the likelihood of multiple external introductions of virus, implicating a range of healthcare facilities in an event that otherwise looked contiguous [23, 168, 179] . Riyadh is a nexus for camel and human travel and has had more MERS cases than any other region of the KSA to date but also harbours a wide range of MERS-CoV variants [128, 167, 179] . However the South Korean outbreak originated from a single infected person, resulting in three to four generations of cases [180, 181] . Studies of this apparently recombinant viral variant did not find an increased evolutionary rate and no sign of virus adaptation thus the outbreak seems to have been driven by circumstance rather than circumstance together with mutation [181] . For many MERS cases detected outside the Arabian Peninsula, extensive contact tracing has been performed and the results described in detail. Contact tracing is essential to contain the emergence and transmission of a new virus and today it is supported by molecular epidemiology. Although it is an expensive and time consuming process, contact tracing can identify potential new infections and through active or passive monitoring, react more rapidly if disease does develop. Results of contact tracing to date have found that onward transmission among humans is an infrequent event. For example, there were 83 contacts, both symptomatic and asymptomatic, of a case treated in Germany who travelled from the UAE but no sign of virus or antibody were found in any of them [73] . The very first MERS case had made contact with 56 HCWs and 48 others, but none developed any indication of infection [162] . In a study of 123 contacts of a case treated in France, only seven matched the definition for a possible case and were tested; one who had shared a 20 m 2 hospital room while in a bed 1.5 m away from the index case for a prolonged period was positive [26] . None of the contacts of the first two MERS cases imported into the USA in 2014 contained any MERS-CoV footprint [182] and none of the 131 contacts of two travellers returning to the Netherlands developed MERS-CoV antibodies or tested RNA positive [25, 183] . Analyses of public data reveal many likely instances of nosocomial acquisition of infection in the Arabian Peninsula and these data may be accompanied by some details noting contact with a known case or facility. One example identified the likely role of a patient with a subclinical infection, present in a hospital during their admission for other reasons, as the likeliest index case triggering a family cluster [93] . Contact tracing was a significant factor in the termination of a 2015 outbreak involving multiple South Korean hospitals [184] . Such studies demonstrate the necessity of finding and understanding a role for mild and asymptomatic cases, together with restricting close contact or prolonged exposure of infected people to others, especially older family members and friends with underlying disease (Fig. 4c) . The hospital-associated outbreak in Jeddah in 2014 was the largest and most rapid accumulation of MERS-CoV detections to date. The greatest number of MERS-CoV detections of any month on record occurred in Jeddah in April. The outbreak was mostly (>60 % of cases) associated with human-to-human spread within hospital environments and resulted from a lack of, or breakdown in, infection prevention and control [37, 185, 186] . A rise in fatalities followed the rapid increase in case numbers. In 2015 two large outbreaks occurred. South Korea was the site of the first large scale outbreak outside the Arabian Peninsula and produced the first cases in both South Korea and China, occurring between May and July 2015. This was closely followed by a distinct outbreak in Ar Riyad province in the KSA which appeared to come under control in early November. After staying in Bahrain for two weeks, a 68 year old male (68 M) travelled home to South Korea via Qatar, arriving free of symptoms on the 4 th May 2015 [187] . He developed fever, myalgia and a cough nearly a week later (11 th ). He visited a clinic as an outpatient between the 12 th and 15 th of May and was admitted to Hospital A on the 15 th [188] . He was discharged from Hospital A on the 17 th then visited and was admitted to the emergency department of Hospital B on the 18 th . During this second stay, a sputum sample was taken and tested positive for MERS-CoV on the 20 th [187, 188] , triggering transfer to the designated isolation treatment facility. Over a period of 10 days, the index case was seen at three different hospitals, demonstrating a key feature of "hospital shopping" that shaped the South Korean outbreak. Approximately 34 people were infected during this time [187] . In total 186 cases were generated in this outbreak, all linked through a single transmission chain to 68 M; 37 cases died [189] . In South Korea, the national health insurance system provides for relatively low cost medical care, defraying some costs by making family members responsible for a portion of the ministration of the sick, resulting in them sometimes staying for long periods in the rooms that often have more than four beds in them [24] . Other factors thought to have enabled this outbreak included unfamiliarity of local clinicians with MERS, ease with which the public can visit and be treated by tertiary hospitals, the custom of visiting sick friends and relatives in hospitals, the hierarchical nature of Korean society, crowded emergency rooms, poor IPC measures, a lack of negative pressure isolation rooms and poor inter-hospital communication of patient disease histories [24, [190] [191] [192] . All of the reported transmission occurred across three or four generations and apart from one unknown source, were all hospital-acquired [24, 120, 181, [193] [194] [195] . Few clinical details about these cases have been reported to date and detail on transmission and contact tracing is minimal. The hospitals involved were initially not identified, governmental guidance and actions produced confusing messages and there was very limited communication at all early on which resulted in unnecessary concern, distrust and a distinct economic impact [191, [196] [197] [198] . Early in the outbreak, a infected traveller, the son of an identified case in South Korea, passed through Hong Kong on his way to China where he was located, isolated and cared for in China [91, 199, 200] . No contacts became ill. The outbreak was brought under control in late July/ early August [201] after improved IPC measures were employed, strong contact tracing monitoring and quarantine, expanded laboratory testing, hospitals were better secured, specialized personnel were dispatched to manage cases and international cooperation increased [202, 203] . A review of public data showed that, as for MERS in the KSA, older age and the presence of underlying disease were significantly associated with a fatal outcome in South Korea. [40] Even though R 0 is <1, super-spreading events facilitated by circumstances created in healthcare settings and characterized by cluster sizes over 150, such as this one, are not unexpected from MERS-CoV infection [204] . The dynamic of an outbreak depends on the R 0 and an individual's viral shedding patterns, contact type and frequency, hospital procedures and population structure and density [204] . In the region of Ar Riyad, including the capital city of Riyadh, a hospital based cluster began, within a single hospital, from late June 2015 [205] . By mid-September there had been approximately170 cases reported but the outbreak appeared to been brought under control in November. It became apparent early on that MERS-CoV spread relatively ineffectively from human-to-human. Despite ongoing and possibly seasonal introduction of virus to the human population via infected DCs and perhaps other animals yet to be identified, the vast majority of MERS-CoV transmission has occurred from infected to uninfected humans in close and prolonged contact through circumstances created by poor infection control in health care settings. This opportunistic virus has had its greatest impact on those with underlying diseases and such vulnerable people, sometimes suffering multiple comorbidities, have been most often associated with hospitals, creating a perfect storm of exposure, transmission and mortality. It remains unclear if this group are uniquely affected by MERS-CoV or if other respiratory virus infections, including those from HCoVs, produce a similarly serious impact. In South Korea, a single imported case created an outbreak of 185 cases and 36 deaths that had a disproportionate impact on economic performance, community behaviour and trust in government and the health care system. Household human-to human transmission occurs but is also limited. Educational programs will be essential tools for combatting the spread of MERS-CoV both within urban and regional communities and for the health care setting. Vigilance remains important for containment since MERS-CoV is a virus with a genetic makeup that has been observed for only three years and is not stable. Among all humans reported to be infected, nearly 40 % have died. Continued laboratory testing, sequencing, analysis, timely data sharing and clear communication are essential for such vigilance to be effective. Global alignment of case definitions would further aid accurate calculation of a case fatality ratio by including subclinical case numbers. Whole genome sequencing has been used extensively to study MERS-CoV travel and variation and although it remains a tool for experts, it appears to be the best tool for the job. MERS and SARS have some clinical similarities but they also diverge significantly [206] . Defining characteristics include the higher PFC among MERS cases (above 50 % in 2013 and currently at 30-40 %; well above the 9 % of SARS) and the higher association between fatal MERS and older males with underlying comorbidities. For the viruses, MERS-CoV has a broader tropism, grows more rapidly in vitro, more rapidly induces cytopathogenic change, triggers distinct transcriptional responses, makes use of a different receptor, induces a more proinflammatory state and has a delayed innate antiviral response compared to SARS-CoV. There appears to be a 2-3 % prevalence of MERS-CoV in the KSA with a 5 % chance of secondary transmission within the household. There is an increased risk of infection through certain occupations at certain times and a much greater chance for spread to other humans during circumstances created by humans, which drives more effective transmission than any R 0 would predict on face value. Nonetheless, despite multiple mass gatherings that have afforded the virus many millions of opportunities to spread, there have remarkably been no reported outbreaks of MERS or MERS-CoV during or immediately after these events. There is no evidence that MERS-CoV is a virus of pandemic concern. Nonetheless, hospital settings continue to describe MERS cases and outbreaks in the Arabian Peninsula. As long as we facilitate the spread of MERS-CoV among our most vulnerable populations, the world must remain on alert for cases which may be exported more frequently when a host country with infected camel reservoirs is experiencing human clusters or outbreaks. The MERS-CoV appears to be an enzootic virus infecting the DC URT with evidence of recent genetic recombination. It may once have had its origins among bats, but evidence is lacking and the relevance of that to today's ongoing epidemic is academic. Thanks to quick action, the sensitive and rapid molecular diagnostic tools required to achieve rapid and sensitive detection goal have been in place and made widely available since the virus was reported in 2012. RT-PCR testing of LRT samples remains the gold standard for MERS-CoV confirmation. Serological tools continue to emerge but they are in need of further validation using samples from mild and asymptomatic infections and a densely sampled cohort study to follow contacts of new cases may address this need. Similarly, the important question of whether those who do shed MERS-CoV RNA for extended periods are infectious while appearing well, continues to go unanswered. It is even unclear just how many 'asymptomatic' infections have been described and reported correctly which in turn raises questions about the reliability of other clinical data collection to date. While the basic virology of MERS-CoV has advanced over the course of the past three years, understanding what is happening in, and the interplay between, camel, environment and human is still in its infancy. Additional file 1: Figure S1 . The

Where is MERS-CoV is enzootic in DC?