Peste Des Petits Ruminants Screening and Diagnostic Tests in African Wildlife in the Context of Rinderpest Eradication (1994–2007)

Peste des petits ruminants (PPR) virus causes a major disease in domestic and wild small ruminants. Understanding the role of wildlife in PPR virus ecology is important for PPR control and its eradication targeted worldwide in 2030. Developing diagnostic tools that provide reliable data for PPR detection in wildlife will help monitor wild populations for PPR and support the eradication program. We analyze a continental-scale dataset from African free-ranging wild ungulates ( n =2570) collected between 1994 and 2007. A Bayesian model estimated the performance of ELISA tests against PPR and rinderpest and their prevalence in African bufalo. Te H-and N-ELISA tests used, not initially developed for wildlife, showed poor sensitivities for the detection of PPR antibodies in African bufalo. Te estimations of PPR antibody prevalence derived from the results of these tests for animals presumably not exposed or potentially exposed to PPR were uncertain. Tus, poor performances of these PPR serological tests in wildlife would not allow robust estimations of PPR antibody prevalence in African bufalo and would be extremely speculative in non-bufalo wild ungulate species. We recommend that current and new tests be validated for wildlife hosts to provide sufcient sensitivity and specifcity of detection and a diagnostic protocol be developed for PPR wildlife research.


Introduction
Peste des petits ruminants virus (PPRV) belongs to the genus Morbillivirus within the Paramyxoviridae family closely related to the now eradicated rinderpest (RP) virus [1]. Te associated disease (PPR) is one of the most serious and widespread pathologies in domestic small ruminants, representing a major threat for the livelihoods of millions of small-scale farmers across Africa and Asia [2]. Although sheep and goats are primarily afected, it seems that the PPRV can also infect a wide range of wildlife and unconventional hosts [3][4][5]. However, there has been very limited evidence so far of disease occurrence of PPR in freeranging wildlife populations in Africa (but see [5,6]). Evidence of PPR disease occurrence in wildlife is limited to African ungulates in captivity [7] and to severe outbreaks in free-ranging ungulates in Asia [8,9]. Te virus is transmitted mainly through direct contact and survives only briefy in the environment: from 3 to 10 days at 37-40 degrees Celsius up to 3 days in European context; more data are lacking for the African continent [10][11][12]. Te severity of the disease in livestock varies according to the virus strain, the host breed and species and traits (e.g., age and immunity status), and the production system (e.g., intensive, semi-extensive, and free-roaming). Globally, PPR mimics the symptomatology of RP with clinical signs similar to other respiratory syndromes and include coughing, nasal and ocular discharges, and more severe symptoms leading to death in the acute form. It can be assumed that all wild ungulates are susceptible to PPR; however, with regard to the display of clinical signs, only some subfamilies of Bovidae, including Caprinae (only Siberian ibex, Capra sibirica) and Antilopinae (currently only two Asian antelopes: saiga (Saiga tatarica) and goitered gazelle (Gazella subgutturosa)), are known so far to express the disease in free-ranging conditions, in a similar manner to domestic goats and sheep [9,13] unlike those kept in captivity, for which clinically diseased animals have been reported in a much wider variety of ungulate subfamilies [4,14,15]. Camelidae can also express the disease [5]. Convalescent and vaccinated small ruminants develop a strong and lifelong immunity and are protected against re-infection. Cattle show subclinical infection with PPRV with little evidence of viremia and no virus excretion, while they do seroconvert [16,17].
Food and Agriculture Organization (FAO) of the United Nations and the World Organization for Animal Health (WOAH, founded as OIE) have identifed PPR as the next disease to be eradicated worldwide [18][19][20] with a target of eradication by 2030 [21]. Tis objective echoes the worldwide eradication of RP in 2011, the frst ever eradicated animal disease [22]. Where and when both viruses were cooccurring, ecological interactions between viruses included cross-immunity eliciting cross-reactive antibodies. Tus, in the last stages of the Global Rinderpest Eradication Program (GREP), diferential serological diagnosis was required to assess progress towards freedom in both domestic and wild animal populations. For this purpose, diferent tools were developed and implemented, mainly competitive enzyme-linked immunosorbent assays (ELISAs) [23][24][25] in association with neutralization assays [26].
PPRV is currently present in the form of four lineages (I-IV) in West, Central, and East Africa, the Middle East, and Asia [1]. Te history of its evolution and geographical spread is not completely understood [27] but East Africa has experienced a recent introduction of virus into naive small ruminant populations with high morbidity and mortality (30 to 70%). PPR was frst reported in Uganda in 2003 and since 2006-2008, Uganda, Kenya, and Tanzania ofcially recognized the infection and have been severely impacted by recurrent outbreaks [28][29][30]. Today, the virus is threatening Southern African countries with Angola, Burundi, and the Democratic Republic of Congo already afected, Mozambique, Rwanda, and Zambia with outbreaks close to international borders however without disease cases, and Malawi, Namibia, and Zimbabwe at a high risk of PPR introduction [31].
Among the potential challenges to successful eradication of PPR worldwide is the understanding of the role of wildlife in PPRV ecology and PPR dynamics [4,32]. Tis is particularly important in the socioecological context of rural Africa and Central-South Asia where domestic stock coexists with a large diversity of wild ungulate species, many of them being susceptible to PPR [33]. Tese contexts of wildlifelivestock interfaces provide ample opportunities for virus sharing between wild and domestic hosts [4,9,30,[34][35][36]. So far, our knowledge about the role of wildlife in PPRV ecology is limited to (i) outbreaks in ex situ populations in zoos or fenced enclosures that provide some indication of species susceptibility (e.g., [37,38]) recently reviewed in Munir [39]; Parida et al. [1]; and Fine et al. [4] and (ii) occasional in situ outbreaks in wild mountain Caprinae (Siberian ibex) and in two antelopes (saiga and goitered gazelle) in Mongolia [8,9,40].
To better control the disease, improve small ruminant economies, and prevent biodiversity decline, it is vital to identify potential maintenance and bridge hosts among wildlife and to improve epidemiosurveillance methodologies and testing systems for wildlife populations. Failure to do so might also ultimately compromise eradication programs. Although it is likely that the majority of infection cycles are maintained within and between domestic livestock, some wildlife species or communities may be able to maintain PPRV for variable periods of time [41]. Wildlife populations could also act as bridge hosts for PPRV, linking otherwise unconnected infected and naive domestic ungulate populations [42]. Finally, wildlife populations (none of them vaccinated) could be used as sentinel populations for PPRV circulation in regions where vaccination programs are undertaken in domestic stock [43], a strategy that was successfully implemented during the RP eradication program [44].
During the rinderpest eradication campaign conducted across Africa between 1994 and 2007, a large number and diversity of wild ungulates were sampled and tested for both RP and PPR viruses: 2570 serum samples were collected from 48 taxa (species and subspecies). Te general objective of the wildlife component of the RP eradication campaign was to clarify the regional RP epidemiological status in Africa, since unequivocal data on wild virus circulation could be obtained from non-vaccinated wildlife sentinels. Because of the cross-reactivity between RP and PPR, samples were tested serologically for antibodies directed against RP antibodies and PPR antibodies. Hence, this continental-scale dataset from African wild ungulates provided an opportunity to explore the role of wildlife in PPR epidemiology since (i) diagnosis on both diseases was advised by international institutions and (ii) wildlife surveillance was used as a tool for RP control.
By using this comprehensive multihost dataset, our goal was to retrospectively estimate the seroprevalence of PPR and RP in African bufalo (Syncerus cafer) and other wildlife species across time and in the sub-Saharan African countries. To achieve this objective, the performance in terms of sensitivity and specifcity of the four ELISA tests implemented was estimated using a Bayesian model where prior distributions for sensitivity and specifcity parameters were specifed based on virus neutralization reference tests results obtained from a subset of the collected samples [45]. We discuss the strengths and limitations of the study knowing it was not designed in terms of (i) spatiotemporal sampling strategy (non-probabilistic sampling or empirical sampling) and (ii) targeted wildlife species and adequacy of the screening tests used to answer questions about PPR prevalence. Indeed, wildlife samples were tested with ELISA tests that were validated only for domestic animals.

Data Collection.
Te African Wildlife Veterinary Project in the Pan African Rinderpest Campaign (PARC) program was initiated by African Union/IBAR, supported by European Union and implemented by the consortium represented by CIRAD (International Centre for Agricultural Research for Development) and ZSL (Zoological Society of London), and this work was subsequently consolidated with the activities of the Epidemiology Unit of the Program for the Control of Epizootic Diseases (PACE) funded by the European Union and other donors. Data collection on wildlife was conducted under the authority of the African Union (AU IBAR) and countries' agreements within this institution. International wildlife experts based in regional AU IBAR ofces (in Bamako and Nairobi) carried out feld operations in close collaboration with national experts from relevant ministries (mainly agriculture and environment) who facilitated local authorizations and logistics and participated in the data collection.
Wildlife serological data were collected in the frame of the successive PARC and PACE which took place from November 1997 to June 2007. Te dataset included samples taken by the Kenya Wildlife Services Veterinary Unit supported by the PARC program during the widespread rinderpest epidemic in Kenya and Tanzania afecting wildlife. Subsequently, more extensive survey work was undertaken in West, Central, and East Africa within the African Wildlife Veterinary Project (1998)(1999)(2000) and then by the epidemiology unit of the PACE program at AU IBAR based in Nairobi [46]. Te objective of this wildlife surveillance was to sample wildlife populations in key ecosystems and assess their historic RP status (Table 1). Tis was based on (i) ageing the sampled animals, (ii) collecting an age stratifed sample in each population, and thereby (iii) assessing from age structured antibody prevalence the date of the latest likely rinderpest virus circulation in that population. Within the overall wildlife dataset (48 species, n � 2573, 14 countries), we decided to work on the bufalo subdataset (1 species, n � 1211, 10 countries) during the period 1997-2007 because the bufalo sample size was substantial, whereas sample sizes in other species were too small for making proper subdatasets. Te number of sampled individuals per country is presented in Table 2.

Testing Methodology.
To look for evidence of RP or PPR infections, competitive ELISA (c-ELISA) tests were applied to all serum samples. For both diseases, the H-and the N-ELISA tests were used based on the use of monoclonal antibodies (Mab) targeting either the hemagglutinin protein or the nucleoprotein prepared at Pirbright [24] and at CIRAD [23,25], respectively. If cross-reactivity occurred with ELISA tests, virus neutralization tests (VNTs) specifc either to RP or PPR were implemented. Although cumbersome and not readily adaptable to large-scale surveys, VNT is the reference test for international trade in the WOAH (founded as OIE) Terrestrial Manual [26,47]. Tus, the status of each sampled animal was derived from the cumulative set of these interpreted results.
In the frst step, the following ELISA-based tests were applied on sera: the RP H-ELISA and the PPR H-and N-ELISA. However, along the project, a lack of sensitivity in cattle became a matter of concern with the RP H-ELISA, although highly specifc, due to the use of crude virus as an antigen and a Mab raised against the vaccine RBOK strain. Te RP N-ELISA is based on the expression of a recombinant N protein as antigen and the related N-based Mab and had been developed earlier in 1992. However, it had been underused because the RP H-ELISA was already commercialized under a kit format. Later on, it was decided to use both RP ELISA tests. For all c-ELISA, the cutof was settled at 50%. As for diferential VNT, threshold was set at 1/10 based on successive 1/2 dilutions of sera from 1/5 to 1/320 (i.e., 1/5, 1/10, 1/20, 1/40, 1/80, 1/160, and 1/320). Te test was considered as conclusively positive for one of the two viruses when the neutralization titre for that virus was at least 1/10 and with a diference of at least two levels with the neutralization titre of the heterologous virus. Te combination of tests implemented is documented in Table 3. All the samples were subjected to the RP H-ELISA and the PPR N-ELISA tests. Some samples were also tested with RP N-ELISA and/or PPR H-ELISA. Finally, a fraction of the samples was also subjected to the diferential VNTs. Tese samples comprised (i) all the samples which were positive (see below) according to the RP H-ELISA, (ii) most of the samples which were positive (see below) according to the PPR N-ELISA, and (iii) some samples which were negative according to both the RP H-ELISA and the PPR N-ELISA.

Approach to Manage the Dataset.
Te test result data were stratifed according to the applied c-ELISA tests and the PPR and RP status according to the diferential VNTs. Te number of samples in each category is presented in Table 3. Te data were also stratifed according to the presumed exposure status of the sampled animal with regard to RP and PPR. Te sampling date, the age of the animal at sampling, and the years of the last RP case report and of the frst PPR case report in domestic ungulates in the country where the sample had been collected were considered. It was considered that, given the sampling date and its age at this date, Transboundary and Emerging Diseases Table 2: Diagnostic results per species: number of animal sampled per species (order/family name are given and tribe for Bovidae family); number of positive ("+")/number of tested for each of the six diagnostic tests.    an animal should not have been exposed to RP viruses if its estimated birth date was more than two years after the last RP case report in domestic ungulates in the country where the sample had been collected. Te 2-year threshold was selected, considered adequate in the absence of virus circulation after the last known outbreak. Conversely, it was considered that, given the sampling date and its age at this date, an animal could have been exposed to RP viruses if it was estimated to be born before the last RP case report in domestic ungulates in the country where the sample had been collected. Te same reasoning applies to PPR. Te distribution of samples in the diferent countries covered by  Te third "+" column represents the number of individuals per species considered positive for each disease according to the following criteria. PPR1: positive for PPR according to the diferential VNT test in a PPR potentially exposed/RP presumably free context. PPR2: positive for PPR according to the PPR VNT test in a PPR potentially exposed/RP presumably free context but diference in titre level with the RP VNT test <2. PPR3: positive for PPR according to the PPR VNT test in a PPR potentially exposed/RP presumably free context but the RP VNT test has not been done. PPR4: positive for PPR according to the diferential VNT test in a PPR potentially exposed and RP potentially exposed or doubtful context. RP1: positive for RP according to the diferential VNT test in a RP potentially exposed/PPR presumably free context. RP2: positive for RP according to the RP VNT test in a RP potentially exposed/PPR presumably free context but diference in titre level with the PPR VNT test <2. RP3: positive for RP according to the RP VNT test in a RP potentially exposed/PPR presumably free context but the PPR VNT test has not been done. RP4: positive for RP according to the diferential VNT test in a RP potentially exposed and PPR potentially exposed or doubtful context. Bold values indicate the number of positive individuals per species for both RP and PPR as international standards for each disease. Bold values in other columns indicate which individuals have been recognised positive and for which test. the database is presented according to the RP and PPR presumed exposition status in Supplementary Materials S2 and S3, respectively. Ageing of bufalo is accurate to within months up to the age of six years old, based on dentition after which time, ageing becomes dependent on horn shape and other less specifc factors [48]. Te samples collected less than 3 years before the frst PPR case report in domestic ungulates were considered as doubtful regarding potential exposure to PPRV, and the samples from animals born less than 3 years after the last RP case report in domestic ungulates were considered as doubtful regarding potential exposure to RP virus. Te 3-year threshold was chosen to take into account the possibility of undetected outbreaks in domestic ruminants and unknown exposure context for wildlife.

Descriptive Analysis.
Te data were explored before elaborating a statistical model to estimate prevalence and test performance parameters. Firstly, the sample distribution of percent inhibition values of the c-ELISA tests to detect RP and PPR antibodies was plotted for diferent potential exposure contexts. Secondly, the outcomes of the diferential VNTs (when available) were considered through the potential exposure contexts of these samples (see above) and to the results of the diferential VNTs available in the same sample cluster (samples collected in the same place on the same date and thus likely to be from the same bufalo population). Tis procedure would allow detecting incoherent diferential VNT results and assess the reliability of this test. Tirdly, the sample distributions of percent inhibition values of the c-ELISA tests for samples with conclusive VNTs were plotted.

Bayesian Model.
A subset of the African bufalo serological data was selected to ft the Bayesian model. It included only the samples which had been collected either in a RP potentially exposed and PPR presumably free context or in a RP presumably free and PPR potentially exposed context. Tis resulted in a reduction in sample size from 1211 to 768. Te objective of the analysis was to get estimations of PPR and RP serological prevalence in African bufalo in contrasted contexts regarding potential exposure to the virological agents. Te stratifcation of the data according to the exposure status allowed generating categories among which serological prevalence should difer. Tis is an important condition in order to be able to assess the performances of serological tests in the absence of gold standard. Moreover, it could allow evaluating whether PPR would circulate in African bufalo before the emergence of the disease in domestic ruminants and whether rinderpest would still circulate in African bufalo after its eradication in domestic ruminants. Te following assumptions were made: (i) It was assumed that because of cross-protection between PPR and RP and because the selected samples had been collected in contexts in which only one of the diseases was believed to circulate (see above), no individual can have both antibodies against PPR and against RP. Hence, there were three possible serological statuses: seropositive for PPR and seronegative for RP, seropositive for RP and seronegative for PPR, and seronegative for both PPR and RP. For any potential exposure group, the frequencies of these three possible states summed to 1. Tere were thus 2 4 (i.e., 16) possible outcomes when all the tests had been applied, 2 3 (i.e., 8) when only three tests had been applied, and 2 2 (i.e., 4) when only two tests had been applied. Te frequencies of these outcomes were considered as realizations of multinomial distributions with 16, 9, or 4 probability parameters. Each multinomial probability parameter was a function of RP and PPR prevalence (prev_PPR, prev_RP) and of sensitivities and specifcities of the c-ELISA tests. To account for cross-reactivity, two sensitivity parameters were considered for each c-ELISA test: Se_PPR was the probability of a positive test outcome for an individual for which true status was seropositive for PPR while Se_RP was the probability of a positive test outcome for an individual for which true status was seropositive for RP. For each test specifcity, Sp was defned as the probability of a negative test result for an individual for which true status was negative for both PPR and RP. Te relationships between the probabilities of the multinomial distribution and these parameters are provided in Supplementary Material S1.

Prior Distributions.
Test sensitivities and specifcities prior distributions were determined based on the comparison of the c-ELISA test outcomes with the outcomes of the diferential virus neutralization test in samples for which this latest test had been conclusive. Concerning estimation of prevalence, for each potential exposure category, a noninformative Dirichlet prior distribution (i.e., Dirichlet  (1,1,1)) was used to account for the constraint that the sum of prevalence should be below 1 (this constraint is the consequence of the cross-immunity assumption which implies that an individual is either positive for RP and negative for PPR, or positive for PPR and negative for RP, or negative for both PPR and RP).

Model Implementation.
Te model was implemented in OpenBUGS [49]. It was ftted to all data strata simultaneously. Tree Monte Carlo Markov Chains were simulated using the Metropolis-Hastings algorithm [50]. Chain mixing, unimodality of posterior distributions, and Transboundary and Emerging Diseases Gelman-Rubin convergence diagnostic were checked to assess model convergence [51].

Determination of α and β Parameters.
Te α and β parameters of a beta distribution for probability parameters p (here sensitivities and specifcities) can be thought of as the number of positive and negative outcomes generated by a binomial process of parameters p, α + β. Te numbers of positive and negative c-ELISA test results for samples with conclusive VNT outcome were thus used to parameterize the prior distributions of c-ELISA test performance parameters. c-ELISA test outcomes of the samples considered as positive for PPR according to the diferential VNT were used to evaluate sensitivity against PPR parameters. c-ELISA test outcomes of the samples considered as positive for RP according to the diferential VNT were used to evaluate sensitivity against RP parameters. For specifcity parameters, the c-ELISA test outcomes of the samples considered as negative for both RP and PPR according to the diferential VNT were used. All α, β pairs were set so that α + β was never larger than 10. Te number of test results in the data ranged between 491 and 743, depending on ELISA test considered (Table 3) which is much larger than the sum of parameters used for the beta prior distributions indicated above. Consequently, prior distributions were not overly informative, as can be confrmed by considering the width of the 95% credible intervals of prior distributions (Figures 1-4).
For sensitivity parameters, α was chosen as roughly proportional to the number of positive c-ELISA outcomes and β as roughly proportional to the number of negative c-ELISA outcomes. For specifcity parameters, α was chosen as roughly proportional to the number of negative c-ELISA outcomes and β as roughly proportional to the number of positive c-ELISA outcomes.

Descriptive Analysis.
Te sampling operations were undertaken in 14 countries within 3 regions of Africa, west, central, and east, including 14 countries and over a 14-year period (1994-2007) (Table 1). Overall, 2570 free-ranging individuals of 48 African wild animal taxa (including species and subspecies) were sampled and tested ( Table 2). Te data were also stratifed according to the presumed exposure context of the sampled animal with regard to RP and PPR (based on WOAH-founded as OIE-reports and publications). Considering the progressive eradication of RP and the presence or emergence of PPR during the study period, the estimated age of African bufalo (n � 1211, the most represented species in the sample) was considered in relation to the last known and ofcial RP outbreak (Supplementary Material S2) and their sampling time in relation to the frst PPR outbreak in the country (Supplementary Material S3). Sampled individuals were tested using 6 serological tests (virus neutralization test, hereafter referred to as VNT-the gold standard, H-and N-ELISA) for both RP and PPR. However, the 6 tests were not applied systematically to each sample because the objectives of the RP eradication campaign varied during the 14 years of the data collection, and some ELISA tests were either required or not during distinct    (Table 3). All serum tested for the ELISA tests targeting PPRV antibodies were negative for species belonging to the Aepycerotinae (n = 58), Antilopinae (n = 158), Cephalophinae (n = 20) Bovidae subfamilies, and the Girafdae (n = 99), Suidae (n = 289) and Felidae (n = 6) families, either (i) because no survivor was found after an established infection (however, no PPR outbreak in these subfamilies was ever observed in the wild), (ii) because behaviour, ecology, and size of the population did not create the opportunity for infection, (iii) because of the lack of receptivity to PPR infection, or (iv) because the ELISA tests were not adapted. Positive results for the ELISA tests targeting PPR antibodies have been obtained in this study in species belonging to the Reduncinae (i.e., waterbuck, 1/77), Hippotraginae (roan antelope, 1/39), Alcelaphinae (i.e., Western hartebeest, 1/5: lelwel hartebeest, 1/75; tiang, 1/23), and Bovinae (i.e., bushbuck, 2/22; African bufalo, 56/1211) subfamilies of Bovidae.
Te sample distributions of percent inhibition of the 4 c-ELISA tests in the diferent potential exposure contexts revealed an unexpected pattern for the PPR H-ELISA test and to a lesser extent for the PPR N-ELISA test ( Figure 5). Indeed, higher percent inhibition values for the H-ELISA test towards PPR were observed in samples from animals presumed to have not been exposed to PPR (sampled more than two years before the frst PPR case report) as compared to samples from animals that could have been exposed to PPR (sampled after the frst PPR case report). As for the N-ELISA test towards PPR, the distributions of percent inhibition in the two potential exposure contexts were fairly similar. Sample distributions of percent inhibition for the c-ELISA tests towards RP were more in agreement with the potential exposure status of the samples.
Among the 80 samples which had been subjected to the diferential VNTs and for which this test had been conclusive (i.e., to be PPR positive, the PPR titre with cross neutralization was at least two levels above the RP virus titre), 9 presented an outcome of this test that was incompatible with the potential exposure status of the sampled animal or with conclusive diferential VNTs obtained in animals from the same cluster. Te results of the diferential VNTs for these samples were subsequently considered as inconclusive. Furthermore, because the results of the diferential VNTs could thus not be considered as fully reliable, this test was not considered as gold standard. Nonetheless, it was considered as reliable enough to be used to specify prior distributions for the c-ELISA tests in the Bayesian model. Te 71 samples with reliable VNT results (because they were compatible with the potential exposure context of the sampled animal and with the outcomes of the diferential VNTs for other animals in the same cluster) were used to plot the distributions of percent inhibition of the c-ELISA tests for samples of diferent serological status according to the diferential VNT results ( Figure 6). For the PPR H-and RP H-ELISA, the only samples with percent inhibition >50% (and thus considered positive according to these tests) were samples positive for RP according to the diferential VNTs. For the PPR N-ELISA and the RP N-ELISA, the samples with percent inhibition >50% (and thus considered positive according to these tests) included both samples positive for RP and samples positive for PPR according to the VNTs. RP N-ELISA was the only c-ELISA test that produced positive results among samples that were negative for both PPR and RP according to the diferential VNTs. Tese patterns suggest serious defciencies in the performances of the Prevalence of RP in RP potentially exposed contexts Prevalence of RP in RP presumably free contexts Prevalence of PPR in PPR potentially exposed contexts Prevalence of PPR in PPR presumably free contexts c-ELISA tests, especially regarding the sensitivity towards PPR of the tests targeting PPR antibodies and regarding cross-reactivity. Te frequency tables of outcomes of the c-ELISA tests (positive or negative) by serological status according to the diferential VNTs (positive for PPR or positive for RP or negative) were used to defne the prior distributions of the c-ELISA tests' performance parameters (Table 4). Te 2.5% and 97.5% quantiles of the resulting prior distribution are plotted along the 95% credible intervals obtained from the posterior distribution in Figures 1-4.

Tests' Performance.
Test performance parameter estimations are presented in Table 5 and Figures 1-3. Te approach used generated uncertain estimations of sensitivity towards PPR for the two c-ELISA tests targeting PPR antibodies. However, 95% credible intervals for these two tests were below 0.5, refecting poor sensitivity. By contrast, the estimations of sensitivity towards RP for the two c-ELISA tests targeting RP antibodies were medium and high for the H-ELISA and N-ELISA tests, respectively, with reasonable uncertainty levels. Sensitivity estimations also revealed severe cross-reactivity issues with high estimations for sensitivity towards RP for the H-ELISA test targeting PPR and for sensitivity towards PPR for the N-ELISA test targeting RP as well as medium estimation for sensitivity towards RP for the N-ELISA test targeting PPR. Te RP H-ELISA was the only test for which the estimation of sensitivity towards the non-targeted disease's antibodies was low. Specifcity (defned in this specifc context as the probability of a negative outcome for a sample that is indeed negative for both PPR and RP antibodies) estimations were high for all the c-ELISA tests (although slightly lower for the N-ELISA test targeting RP). Table 6 and Figure 4. Certainly, due to the poor performances of the c-ELISA tests targeting PPR antibodies, the PPR seroprevalence estimation for individuals that should not have been exposed to the PPRV was extremely uncertain. Te estimation of PPR seroprevalence for individuals that could have been exposed to the PPRV was less uncertain and low (upper bound of the 95% credible interval at 0.13), suggesting limited circulation of PPRV in bufalo sampled across various populations in West and East Africa.

Prevalence Results for African Bufalo. Seroprevalence estimations are presented in
As expected, the RP seroprevalence estimation for individuals that could have been exposed to the RP virus (born before the last RP outbreak report in the domestic compartment) was larger than the RP seroprevalence estimation for individuals that should not have been exposed to the RP virus (which was close to 0). However, even in the former situation, RP seroprevalence estimation was low (upper bound of the 95% credible interval at 0.24), suggesting limited circulation of RP virus in bufalo sampled across various populations in West and East Africa.

Discussion
Tis study analyzes the largest free-ranging wildlife dataset ever explored for PPRV. Te results are important considering the current plan to eradicate PPR in Africa, the lack of knowledge about the potential role of African wildlife in PPR epidemiology [4], and the efort and cost it would mean Should not have been exposed to the target disease Could have been exposed to the target disease Figure 5: Distribution of percent inhibition values of c-ELISA tests depending on the exposure status of the animals. For PPR, the animals that should not have been exposed to the virus are those that have been sampled at least three years before the frst PPR outbreak record in the country while the animals that could have been exposed to the virus are those that have been sampled after the frst PPR outbreak record in the country. For RP, the animals that should not have been exposed to the virus are those that are born (based on determination of age at sampling) at least three years after the last RP outbreak record in the country while the animals that could have been exposed to the virus are those that are born before the last RP outbreak record in the country.
to collect a similar continental dataset. Tis dataset was however not initially designed to estimate PPR seroprevalence. It also sufers from common biases associated with sampling free-ranging wildlife (e.g., small sample size, diversity of species, sample representativeness of animal populations, and diagnostic tests not developed or adapted for wildlife species) and other biases such as diferent laboratory standards and quality of the cold chain during transport from the feld to the laboratory. However, all tests were performed in reference laboratories and all the wildlife experts involved in data collection-co-authors of this article-have dedicated great eforts to make sure the samples would reach the laboratory in adequate cold chain conditions. We used a Bayesian modelling approach to cope with these issues in assessing the performance of the tests before inferring any epidemiological outcome. Te performances of the sensitivity of c-ELISA tests designed for the detection of PPR in domestic animals and used for wildlife during the RP eradication campaign were poor (Table 5 and Figures 1 and 2): frst because their sensitivity towards PPR antibodies was low and second because their sensitivity towards RP antibodies was at the same level or even higher than their sensitivity towards PPR. Te application of these tests could thus result in missing many PPR positive individuals and, in contexts where both PPR and RP circulate, in qualifying as PPR positive individuals that are indeed RP positive. c-ELISA tests designed for the detection of RP performed better in terms of sensitivity towards the targeted antibodies (i.e., RP antibodies), especially for the N-ELISA test, but also presented crossreactivity issues (particularly for N-ELISA test). Te only parameter that refected good performance for all tests (although slightly poorer for the N-ELISA test targeting RP) was specifcity defned as the probability of a negative outcome for a sample negative to both PPR and RP antibodies (Table 5 and Figure 3). As a consequence, the estimation of PPR seroprevalence in wildlife species was only possible for the species with the largest sample size, the African bufalo, and yet with the abovementioned uncertainty, it was impossible to conclude whether or not PPR circulated in African bufalo populations before or after its detection in domestic ungulate populations. Due to the high viral load shared during PPR outbreaks, one might think that PPR ELISAs would perform well during epizootics; however, so far in Africa, PPR spillover from sheep and goats to free-ranging wildlife does not appear to lead to clinical syndromes and much virus expression [30,36,52]. Tis difers from RP where a range of wildlife species expressed disease clinically and viral spread was recorded in their populations. As expected, RP seroprevalence estimations indicate that RP disappeared in African bufalo after eradication of the disease in the domestic compartment. PPR and RP serological test results are provided for a wide range of wildlife taxa (n � 48) to inform future research (Table 2). PPR seropositive samples were identifed in several taxa belonging to various Bovidae subfamilies, with little data available from suids, perissodactyls, and elephants.
In this study, the c-ELISAs implemented were considered at the time as highly accurate, standardized, and robust, able to measure the immune response due to infection and or vaccination in the respective domestic hosts. Tese ELISA tests were validated comparatively to VNT, the WOAH (founded as OIE) gold standard test with potentiality to replace it. However, cross-reaction among morbilliviruses is one of the main constraints for achieving a reliable diagnosis. Te problem of diferential diagnosis is particularly acute with PPR and RP, both viruses overlapping in host range as well as in geographical distribution during RP seromonitoring activities in Africa. Terefore, the PPR and RP VNTs used as a diferential test by titrating samples in parallel played a critical role in RP serological surveillance and eradication programs to ensure distinction of the homologous from the heterologous immune response in domestic as well as in wild population [44,53]. Here, 9 out of 80 VNT tests (11.3%) had to be discarded because of inconsistencies between presence and absence of disease in the area (based on ofcial declarations but with 2-year bufer before frst or after last declaration) or between VNT results from other individuals in the same herd (e.g., in Kenya in 2001, 7/8 individuals from the same herd were confrmed RP cases and 1/8 was initially a confrmed PPR case, which is   highly unlikely given cross-reactivity between viruses). Tose results call for a re-evaluation on the reliability of virus neutralization tests to be used as gold standards.
Te estimated low sensitivity of c-ELISA tests for the African bufalo is a matter of concern and questions its use in cattle which are closely related bovines. Despite the relative phylogenetic closeness between wild and domestic ungulates, the ELISA tests did not perform as well with wildlife as in livestock. Our results indicated that the RP H-ELISA was the test with the best sensitivity for PPR ( Figure 6 and Table 5). However, since the study took place, the PPR H-ELISA has been removed from the market and the N-ELISA test has evolved. In addition, now that RP is eradicated, cross-reactivity due to this important virus afecting a wide range of ungulates is now excluded. An increase in PPR tests' performance can be expected to improve PPR monitoring in wildlife. Lessening the cross-reactions with other morbilliviruses (e.g., canine distemper virus) in the development of tests designated for the diferential serological diagnosis of PPR in domestic as well as in wild population would be highly benefcial. Te resulting tools will help to improve our knowledge in the ecology and evolution of PPR viruses and our understanding of the geographical distribution and spread of the disease in specifc areas as well as the determinants and drivers of PPR at the interface of populations of domestic and wild animals. Promising methodology for increasing PPR test specifcity was developed. Tey rely on short synthetic peptides representing a single epitope as alternative antigens to recombinant proteins or to the whole microorganism [54][55][56][57][58]. More recently, novel neutralization assays based on pseudotyped heterologous viral species expressing the surface glycoprotein(s) of individual morbilliviruses virus were developed [59]. In this regard, conventional VNT will still be needed to validate new tests to evaluate their diagnostic potential in unexplored populations, camel and diferent wild species, shown to be susceptible to PPRV [3,60,61]. Finally, it needs to be noted that during this study, VNT was only applied when c-ELISAs were positive. Tis means that there is a bias in the calculation of the relative sensitivity and specifcity observed when compiling the results. If VNT had been systematically applied, the level of knowledge of the specifcity would have been higher. Indeed, in the dataset, there was no serum which tested negative both for ELISA and VNT. More sera of this kind would increase the specifcity ratio.
As for other morbilliviruses (e.g., measles virus (Keeling and Grenfell, 1997) and RP (Rossiter and James, 1989)), the high and long-lasting immunity to PPR infection in recovering animals suggests that large populations are required for maintenance, with sufcient infux of new susceptible hosts, especially young animals. In smaller populations, epidemics may gradually decline until new virus is reintroduced. Te high-density domestic populations are therefore considered the most likely source of infection for wildlife [9,36]. However, some of the regions of Africa sampled in this analysis are characterized by the presence of relatively large wildlife communities (e.g., East African ecosystems such as Greater Serengeti and South Sudan grasslands, where large populations of wild ungulates migrate seasonally) that cohabitated in close proximity with large livestock populations without fencing [6,30,36].
Bufaloes were considered as a priority species, including all subspecies, for surveillance of RP. Tey are phylogenetically close to cattle, they share their susceptibility and sensitivity to RP virus, and they produce antibody detectable by standard cattle serological tests. Te large distribution of bufalo was also adequate as a general sentinel and surveillance population for RP in Africa. Similarly, as the related water bufalo in Asia is subclinically infected by PPR and mounts an immune response showing high prevalence of PPR antibodies in endemic situations [64,65], the African bufalo could also theoretically be used as a sentinel species for PPR. However, the PPR seroprevalence estimations in African bufalo reported here are not very useful in assessing whether this species plays an important role in the maintenance of PPR and should be monitored in PPR surveillance programs. Indeed, the estimations obtained are very imprecise and do not follow the expected pattern as for the comparison of contexts where the sampled animals could vs. should not have been exposed to PPR (Table 6 and Figure 4). Te latter challenges the strong assumption that countries would have been free from PPR prior to the frst PPR outbreak report in the domestic compartment. However, the estimation of PPR seroprevalence reported here for contexts where African bufalo could have been exposed is very low (95% credible interval [0.001; 0.13], Table 6 and Figure 4). Such low prevalence could result from the fact that many of the bufalo samples were collected in East African areas where, at the time of sampling, PPR had not yet or had only recently been detected in domestic animals. Te epidemiological status of bufalo populations regarding PPR could be diferent now that PPR circulation has been going on for a longer time in the domestic compartment [30,52]. Moreover, Bufalo may be a dead-end host species for PPRV, as is the case for cattle [30,36], in which case PPRV transmission within bufalo populations or from bufalo to domestic ungulates would not be possible.
Te presence of PPR seropositive healthy animals is an indication of infection and recovery of animals while high prevalence in a herd suggests viral circulation within a species' population. Te results presented in Table 2 could suggest a large host species range for PPR in wildlife and inform future PPR surveillance in natural ecosystems notably in species from Alcelaphinae, Reduncinae, and Hippotraginae subfamilies as recently reported [4,9]. However, sample sizes for these species were too small to estimate sensitivity and specifcity of the serological tests in these species so that interpretations of the test results are somewhat speculative and have to be considered with caution. Moreover, future investigations could widen the range of species targeted, including for instance migratory ungulates (e.g., Tomson's gazelle, Eudorcas thomsonii in the Serengeti ecosystem, Tanzania and Kenya; Mongalla gazelle, Eudorcas albonotata in the Sudd ecosystem, South Sudan).
Many questions remain regarding population/community size thresholds and determinants of PPR maintenance in natural ecosystems comprising a wide range of susceptible Transboundary and Emerging Diseases 13 hosts, including wild and domestic sympatric populations. Like measles [66] and canine distemper viruses [67], PPR might be maintained through the interaction of multiple interconnected susceptible wild and domestic host communities, acting as one meta-community, each community experiencing intermittent but non-simultaneous PPR epizootics. Te socioecological context of wild populations/ communities could be a better predictor of its capacity to maintain the virus than its strict species diversity. Furthermore, a range of wildlife species could also link or bridge distant domestic populations and be involved in the spread of PPR across geographical space, without necessarily being able to maintain PPR on the long run [4,42,68]. A PPR maintenance model for the domestic compartment has recently been developed and should be adapted to the wildlife compartment and wild/domestic integrated compartments [69]. Te RP virus emerged two thousand years ago, the PPR virus two hundred years ago, and the canine distemper virus fve hundred years ago in both the terrestrial and marine environments and the measles virus, a human virus, most likely originated from the RP virus. Te evolutionary history of morbilliviruses points at plausible emergence of new viruses, given the current host range and geographic coverage of these viruses. Today, at the beginning of a massive international efort to eradicate PPR globally, the role of African and Asian wildlife in PPR epidemiology is still largely unknown despite recent proofs of ongoing circulation [4]. Considering sheep and goats as the primary hosts for PPRV, the African wild ungulate community is characterized by the near absence of wild sheep and goat species on the continent (except for the walia ibex, Capra walie, the Nubian ibex, Capra nubiana, and the Barbary sheep, Ammotragus lervia, with highly fragmented ranges restricted to isolated mountain massifs), in contrast to the Asian ungulate community, with large populations of numerous taxa spread over immense chains of mountains, a fact that could explain the variability in susceptibility to the disease observed between the continents. Given the results presented here, we recommend (i) longitudinal studies in carefully selected isolated wildlife populations to test the hypothesis of a potential maintenance role of wildlife populations or communities alone and of wildlife populations exposed to wildlife/livestock interfaces (i.e., to explore the maintenance community hypothesis); (ii) virus maintenance modelling in wild and mixed (wild and domestic) host populations to explore host population/ community threshold for PPR maintenance; and (iii) the development of a wildlife protocol using new serological tests and re-evaluating the performance of the PPR N-ELISA for wildlife in the current context (i.e., evolution of the test since the study and in the absence of RP). Tis protocol will need to be validated for key wildlife species (e.g., in Africa, the African bufalo and some selected antelope species such as Grant and Tompson's gazelles). Te current candidate tests for this wildlife protocol are pseudotype assays (e.g., [70]), the luciferase immunoprecipitation system (LIPS) [71], a new b-ELISA developed on the African continent [72], and non-invasive PPR diagnostic tests under development for wildlife species [73].

Data Availability
Te datasets generated and analyzed during the current study are available from the corresponding author on reasonable request. Te Bayesian model is available in supplementary information.

Ethical Approval
All wild animals captured and sampled between 1994 and 2007 in this study were handled by professional veterinarians (the main ones being co-authors of this article) respecting the codes of conduct and principles of wildlife handling in order to ensure the minimum and shortest stress applied on the animal.

Conflicts of Interest
Te authors declare that they have no conficts of interest.

Acknowledgments
Tis retrospective study was supported by a grant from the European Commission Animal Health and Welfare European Research Area Network for the IUEPPR Project "Improved Understanding of Epidemiology of PPR" in the framework of ANIHWA 2013 and by a grant (SI2.756606) from the European Commission Directorate General for Health and Food Safety awarded to the European Union Reference Laboratory for Peste des petits ruminants (EURL-PPR). Te study relies on extensive collaborations and contributions from multiple collaborators across countries, from feld assistants to protected areas and veterinary services managers in the context of the African Wildlife Veterinary Program within the PARC and PACE programs. Open-access funding was enabled and organized by COUPERIN CY23.

Supplementary Materials
Supplementary Material 1: Bayesian model code. Supplementary Material 2: distribution of number of sampled bufalo (n = 1211) with regard to their birth year, the country where they were sampled, and the year of the last RP outbreak record. It was considered that, given the sampling date and its age at this date, an animal should not have been "exposed" to RP viruses if its estimated birth date was more than two years after the last RP case report in domestic ungulates in the country where the sample had been collected. When within the 2-year period, the exposure was considered "doubtful." Vertical bars indicate the year of the last reported RP outbreak in each country. Dashed vertical bars mean that the last RP outbreak report occurred before the indicated year. Supplementary Material 3: distribution of number of sampled bufalo (n = 1211) with regard to the sampling year, the country where they were sampled, and the year of the frst PPR outbreak record. It was considered that, given the sampling date, an animal could have been exposed to PPR virus if it was estimated to be born after the frst PPR case report in domestic ungulates in the country where the sample had been collected. If its sampling date was less than two years before frst PPR case report in the country, the exposure context was considered "doubtful." Vertical bars indicate the year of the frst PPR outbreak in each country. Dashed vertical bars mean that the frst PPR case report occurred before the indicated year. (Supplementary Materials)