Contacts and foot and mouth disease transmission from wild to domestic bovines in Africa

Abstract

Wildlife is a maintenance host for several significant livestock diseases. Interspecific pathogen transmission may occur in complex socio‐ecological systems at wild‐domestic interfaces that have so far been seldom studied. We investigated the relationship between the dynamics of foot and mouth disease (FMD) in vaccinated and unvaccinated cattle populations with respect to frequency of contacts with African buffalo at different buffalo‐cattle interfaces.

A total of 36 GPS collars were deployed on African buffalo (Syncerus caffer) and cattle (Bos taurus, Bos indicus) to assess contact patterns at the periphery of 3 protected areas in Zimbabwe. Simultaneously, a longitudinal survey of 300 cattle with five repeated sampling sessions on known individuals during 16 months was undertaken. Immunological assays (ELISAs), that allowed tracking the production of antibodies following infection or vaccination, were used to assess serological transitions (i.e., incidence and reversion) in the surveyed cattle. Variation in rates of serological transitions across seasons, sites and as a function of the frequency of contact with buffalo was analyzed using generalized linear mixed models.

The incidence in the cattle populations of FMD antibodies produced following infection varied among sites and as a function of contact rates with African buffalo. The incidence was higher for sites with higher contact rates between the two species. The serological incidence was also related to seasons, being higher during the dry or rainy seasons depending on sites. The reversion rate pattern was the opposite of this incidence rate pattern. Vaccination seemed partly efficient at the individual level, but it did not prevent the diffusion of FMD viruses from the wild reservoir host to the domestic cattle population. Furthermore, antibodies were detected in areas where cattle had not been vaccinated, suggesting that the virus may have spread without being detected in domestic populations.

Access to resources shared by buffalo and livestock, particularly water and grazing areas during the dry season, could partly explain the observed patterns of FMD transmission. We discuss how insights on ecological processes leading to wildlife‐livestock contacts may provide some innovative solutions to improve FMD management, including surveillance, prevention or control of buffalo‐borne outbreaks, by adopting strategies targeting risky areas and periods.

Introduction

Although wildlife reservoirs have been identified for several significant livestock diseases (Bengis et al. 2004) and emerging or re‐emerging infectious diseases with high pandemic potential (Taylor et al. 2001, Patz et al. 2004, Lloyd‐Smith et al. 2009), the circumstances leading to pathogen spillover between sympatric wild and domestic hosts (and potentially humans) often remain poorly documented. Because most wildlife species have a secretive nature, direct or indirect contacts at wildlife‐domestic interfaces are rarely quantified (Patz et al. 2004, Smieszek 2009). Furthermore, contacts between wildlife and livestock result from an intricate interplay between drivers of different natures, including biophysical features (e.g., rainfall, soil type, topography), ecological traits (e.g., foraging and spatial behaviors) and human practices (e.g., livestock husbandry, wildlife management). Multi‐host epidemiological systems at wild‐domestic interfaces are thus particularly complex.

Foot and mouth disease (FMD) provides an appropriate model to study pathogen ecology at wildlife‐domestic interfaces (Brahmbhatt et al. 2012). This disease is due to an aphthovirus that can infect ungulates and is highly contagious with a basic reproductive number varying from 2 to 38 depending on climatic conditions (Haydon et al. 1997). FMD has important economic impacts and constrains beef exporting countries to implement massive control programs (e.g., through culling, zoning and/or vaccination) (Paton et al. 2009). Although FMD is widespread worldwide, only 7 serotypes of the virus have been identified. Following the successful eradication of rinderpest, the World Organization for Animal Health (OIE) and the Food and Agriculture Organization have focused their attention on the control of FMD worldwide. Because of its economic implications for animal production in non‐endemic areas, such as Europe, FMD epidemiology has been well studied in high density domestic populations (Cottam et al. 2008), and in most parts of the world eradication is a reachable objective (Melo et al. 2002). However, in Africa the buffalo (Syncerus caffer) is the main wild local reservoir of FMD (Thomson et al. 2003, Bengis et al. 2004). The mechanisms of transmission between sympatric buffalo and cattle in African tropical areas are still poorly understood. However empirical information (Thomson et al. 2003),serological and genetic results (Vosloo et al. 2002b, Ayebazibwe et al. 2010) and in natura experimens (Dawe et al. 1994) all suggest that foot and mouth virus regularly spills over of from buffalo reservoir to cattle. As this emblematic wild species is also a key species in savannah ecosystems, eradication of FMD from endemic areas is currently impossible, because culling or vaccination of buffalo are not an option. Understanding FMD spillover at wildlife‐domestic interfaces in Africa is therefore crucial to control this important disease and a prerequisite for designing a FMD management strategy at the continental scale.

In southern Africa, 3 main FMD virus sero‐types are frequently detected: SAT1, SAT2 and SAT3 (Southern African Territories) (Rweyemamu et al. 2008, Paton et al. 2009), among which no cross–immunity has been observed. The three SATs circulate in domestic and wild populations. The African buffalo has been shown to carry asymptomatic infections and excrete viral particles, while cattle show symptoms (Vosloo et al. 2002a, Thomson et al. 2003, Grubman and Baxt 2004, Paton et al. 2009, Ayebazibwe et al. 2010).

Although the epidemiology of FMD viruses circulation in the buffalo populations is still far from being fully uncovered, there is evidence that calves have a prominent role in the maintenance and transmission of the virus (Vosloo and Thomson 2004). In southern Africa, buffalo are mostly seasonal breeders (Pienaar 1969), but calves may be born throughout the year, providing a constant supply of susceptible animals and allowing year‐round virus circulation (Hedger 1972). Furthermore, African buffalo have been reported to carry live virus for up to 5 years (Condy et al. 1985).

Cattle are also key hosts in the dynamics of FMD in domestic ungulate populations. Viruses can usually be isolated up to 28 days after infection (Charleston 2011) and the carrier state can last up to 3.5 years (Grubman and Baxt 2004).

The virus is highly contagious in cattle populations and can be transmitted via direct contact between individuals, but also via environmental contamination, although climatic conditions in Africa apparently limit this transmission route (OIE 2009, Lefèvre et al. 2010).

Vaccination programs are regularly implemented in cattle close to FMD‐infected buffalo source populations, but unless serological tests that allow differentiating between viral and vaccinal antibodies are available, such programs compromise the detection of viral circulation in host populations.

FMD ecology at buffalo‐cattle interfaces is thus likely to depend on three important processes: inter‐species contacts, within‐species viral circulation and vaccination‐induced immunity in domestic populations. To our knowledge, empirical investigation of the interplay of these processes in situ is lacking, partly because of the difficulty to assess the intensity and frequency of host contacts.

In this article, the transmission of FMD from African buffalo to cattle (Bos taurus/Bos indicus) is investigated at the edge of three protected areas in Zimbabwe. The economic breakdown in Zimbabwe between 2000 and 2008 and the resulting decrease in the ability of governmental veterinary services to operate efficiently have led to a resurgence of FMD outbreaks at the country level, with recurrent epidemic peaks during dry seasons in districts close to national parks (Governmental Veterinary Services, personal communication). Outbreak descriptions report a high proportion of mild and sub‐clinical cases (Kennedy et al. 1984), suggesting that the virus can spread for relatively long periods of time while remaining undetected (Jori et al. 2009). These reports point at the buffalo‐cattle interface as the source of primary FMD outbreaks. Ring FMD vaccination around outbreaks is the current control option for the Government Veterinary Services in Zimbabwe. However, the extent to which spatial and seasonal variation in contact rates between buffalo and cattle drive FMD incidence patterns is still unknown, and so is the impact of vaccination campaigns on the circulation of the virus.

In order to address these questions, we undertook simultaneously: (1) a telemetry‐based movement survey of sympatric buffalo and cattle populations to describe patterns of inter‐species contacts; and (2) a longitudinal serological survey at the individual level to describe the dynamics of FMD infection and immunity responses in cattle populations. In addition, we considered the role of cattle vaccination on FMD epidemiological dynamics in cattle populations. We predicted that FMD prevalence and incidence in cattle would increase with increasing levels of interaction with buffalo. We also expected that the vaccination should decrease the level of natural incidence and increase the level of immunity at the population level.

Methods

Study sites

The study was carried out between April 2010 and September 2011, in three Zimbabwean sites located in two TransFrontier Conservation Areas (TFCAs) including interfaces between wildlife and domestic animals: (1) Malipati Communal Land (CL)/Gonarezhou National Park (NP), (2) Pesvi CL/Kruger NP both within the Great Limpopo (GL‐TFCA), (3) Dete CL/Hwange NP within the KAvango‐ZAmbezi (KAZA‐TFCA) (Fig. 1). Telemetry and serological protocols were conducted simultaneously in these three sites. Unfenced or unmaintained fenced (and therefore largely permeable) interfaces in the three sites could in principle not prevent wild and domestic ungulate movement across the land‐use boundaries. In parallel, serological surveys in cattle populations living far from wildlife interface were implemented in two sites as ‘negative controls': Tinde (30 km from Hwange NP) and Choumpani (20 km from Gonarezhou NP). Seasonal climatic patterns are relatively similar in the five sites with a rainy season from December to March followed by a cold dry season between April and July and a hot dry season from August to November (according to precipitation data available from world.climate.com). A detailed description of the characteristics of these three interfaces is provided in Appendix A and partially in Caron et al. (in press).

Study sites and experimental design at protected areas borders in Zimbabwe (southern Africa) ; with the number of cattle sampled every 4 months (left column) and the number of GPS collars deployed on cattle and buffalo on each side of the 3 interfaces studied (right column).

Human activities in the communal areas outside national parks essentially consist of subsistence farming with small‐scale livestock production. Small herds are bred extensively with on average 12 heads of cattle and small ruminants (goats and a few sheep) per herder, according to the district veterinary services of Chiredzi and Cross Dete.

Cattle vaccination strategies differed among sites during the study period. There was no vaccination in the KAZA‐TFCA (Dete‐Hwange and Tinde), whereas three sessions took place in the GL‐TFCA (Pesvi‐Kruger, Malipati‐Gonarezhou and Choumpani). According to district veterinary services, the proportion of cattle vaccinated was lower in Pesvi‐Kruger than in Malipati‐Gonarezhou and Choumpani (Appendix B). The vaccine used is trivalent (Aflovax) and contains all three SATs.

GPS tracking protocol

We used Global Positioning System (GPS) collars to monitor the movements and contacts between selected cattle and buffalo herds simultaneously in the three sites. A total of 36 GPS UHF collars (manufactured by African Wildlife Tracking) were fitted on adult female cattle and buffalo (out of which three broke down) (Fig. 1). The GPS were scheduled to acquire one location per hour. The data acquisition periods extended from April 2010 to August‐September 2011 in Dete‐Hwange, from March 2010 to May 2011 in Malipati‐Gonarezhou and from June 2010 to July 2011 in Pesvi‐Kruger. One collar was fitted per herd for the cattle and 2 to 4 for the buffalo. In order to track the movements of each cattle herd, the dominant female, according to their owner's perception, was fitted with a collar. Cattle herds were usually confined at night in ‘kraals' (enclosures). During the day, herds were commonly oriented by a herder to grazing areas and watering points where they were left free to forage and drink. Wildlife census data (Park and Wildlife Management Authority [PWMA], Cirad and CNRS, unpublished data [Dunham 2012]), field expertise from PWMA rangers and flyover from SANParks (South Africa National Parks) personnel for North Kruger during the capture, allowed assessment of the number of buffalo herds in areas and confirmed that most buffalo herds potentially in contact with the GPS‐equipped cattle herds included individuals with GPS collars (Appendix A). Buffalo were immobilized using standard procedures by helicopter (de Garine‐Wichatitsky et al. 2010). After sample collection, buffalo were woken‐up using chemical antidote.

Potential infecting contacts between cattle and buffalo

The transmission of the virus is complex. A direct pathway is well documented in the literature with infection by oral inhalation of viral particles following close contact (OIE 2009, Charbonnier and Launois 2011). An indirect pathway also seems to be possible but is less documented in Africa. It is believed that wind transmission played a role in the FMD epidemics in non‐endemic areas (Sorensen et al. 1992, Mikkelsen et al. 2003). However, this type of transmission would only be possible under wet climatic conditions and would thus be unlikely (or possible only during rare climatic events during the cold dry season) for the southern African strains found in regions where semi‐arid climatic conditions prevail. Indirect transmission through the ground or water could nonetheless be possible, even in semi‐arid regions. This possibility is not well established, but supported by some experimental data. It has indeed been shown that the virus can survive in the ground from 10 days at 37°C to 365 days at 4°C (Lefèvre et al. 2010, Charbonnier and Launois 2011).

GPS locations of collared buffalo and cattle were used to document potentially infective contacts between the two species. We considered that a potentially infective contact occurred whenever a cattle location was recorded within 300 m of a buffalo location less than 15 days after the buffalo location had been recorded. The underlying assumption for the time window was that the virus can survive in the environment for 15 days (and we therefore adopted a conservative approach given the semi‐arid climatic conditions of the study sites). For the spatial window, we considered that (1) the buffalo tracked is part of a herd, (2) the GPS precision is imperfect (i.e., precision range of 15–30 m from manufacturer data) and (3) tracked individuals move during the one hour period between two recorded locations. Moreover, for the purpose of drawing unbiased contact rate comparisons, we took into account the variation among sites and seasons in GPS acquisition rates and number of GPS collars deployed on buffalo. We divided the number of contacts recorded for a given cattle during a given season by the product of the number of locations recorded for that cattle by the number of buffalo locations recorded in the same study site. We thus expressed contact rates in terms of number of contacts relatively to the number of cattle‐buffalo location pairs. Because this rate was extremely low and contrasted (from 1.1 × 10⁻⁴ to 6.6 × 10⁻⁸), these contact rate estimations were multiplied by 10⁸ and then log transformed. The resulting index is hereafter referred to as ‘buffalo contact index'.

Home range (HR) calculation

Annual home ranges were computed (up to the 0.95 isopleth) for each collared individual using the movement‐based kernel density estimation method and volume overlap between wild and domestic animals was calculated (Benhamou and Cornelis 2010). More detailed information is provided in Appendix C.

Epidemiological survey

Longitudinal survey in livestock and snapshot sampling in wildlife

In each cattle herd tracked with GPS, a serological survey was conducted on 10 randomly chosen ear‐tagged adults (males and females) in the owner's herd (Fig. 1). In addition, 50 heads of cattle in each TransFrontier Conservation Area (TFCA) were serologically monitored as ‘control' with no buffalo/cattle contact in Tinde and Choumpani. 310 cattle were blood sampled every 4 months, during a 16 months period from April 2010 to August 2011 (i.e., each cattle was sampled up to 5 times over this period). Four seasons were covered: cold‐dry 1 (April 2010 to July 2010), hot‐dry (August 2010 to November 2010), rainy (December 2010 to March 2011) and cold‐dry2 (April 2011 to July 2011). Logistical constraints prevented the full coverage of the planned sampling sessions. Sessions are missing for 2 sites: Malipati‐Gonarezhou for the cold‐dry 2 period and Pesvi‐Kruger for the rainy period. Furthermore, all cattle in each herd could not be sampled at each sampling occasions (some had died, had been sold or could not be found).

The GPS‐tracked buffalo were sampled for serological analyses during the fitting and/or removal of collars (i.e., 1 or 2 sampling). FMD data obtained from samples collected during other projects of the veterinary services were also opportunistically used.

Serological diagnostic

Serological testing was conducted in the Governmental Veterinary Services (GVS) of Zimbabwe. Two different ELISA tests were applied on each sample to detect FMD antibodies in livestock: an ELISA commercial test (PrioCHECK FMDV NS) and a SATs liquid phase blocking ELISA test.

The ELISA commercial test detects the antibodies produced following natural viral replication, irrespective of the virus serotype, but not vaccinal antibodies (Moonen et al. 2004). The specificity and sensitivity of the test are estimated at 98.1% and 97.2%, respectively (Brocchi et al. 2006).

The SATs blocking ELISA test can detect both antibodies produced following natural infection by the serotypes circulating in southern Africa (SAT 1/2/3) and antibodies produced following vaccination. The antigens used for the blocking ELISA were SAT1 (Bot 1/68); SAT2 (Zim 5/81); SAT3 (Zim 4/81). The sensitivity of the liquid phase blocking ELISA for SATs tests are for SAT1: 92.4%; SAT2: 83.4%; SAT3: 80.3%. The specificities are 100%, 99.8% and 100%, for SAT1: SAT2 and SAT3, respectively (Sorensen et al. 1992).

For buffalo, SATs serology was performed at the ARC‐OVI (Agricultural Research Council‐Onderstepoort Veterinary Institute) laboratory in South Africa. Liquid phase blocking ELISA was used to detect SAT1, SAT2 and SAT3 structural proteins antibodies. GVS and ARC‐OVI used the SADC (Southern African Development Community) harmonized Liquid Phase Blocking ELISA serological tests (ISO 17025 accredited protocol).

Laboratory cross validation SATs test

Fifty five samples (30 positives and 25 negatives for NSP) analyzed at GVS of Zimbabwe were also analyzed for SAT1, SAT2 and SAT3 by the regional OIE (world organization for animal health) reference laboratory for FMD (i.e., ARC‐ OVI, South Africa). The Cohen's kappa statistic (Dohoo et al. 2010) was used to evaluate the agreement between the results of the tests performed in the two laboratories. The results of the NSP tests performed by GVS of Zimbabwe were not cross validated because it was assumed that their validity was guaranteed by the precise protocol instructions provided by the manufacturer of this commercial kit.

Antibody dynamics and interpretation of serological data

Little is known about the production and persistence of antibodies produced following natural FMD infection or vaccination in wild and domestic hosts in southern Africa. In the literature, the domestic host virus incubation period varies between 2 and 14 days, the first clinical signs could appear between 2 and 10 days and the virus excretion could spread over from 2 to 28 days. The natural antibodies appear after 7 or 14 days and they could persist in the host for almost 3 years. As for the vaccinal antibodies, they may persist on average for one year but re‐booster vaccination sessions every 6 months are recommended in endemic areas to effectively protect host individuals (Thomson et al. 2003, Grubman and Baxt 2004, Cloete et al. 2008, Paton et al. 2009, Charleston 2011).

We assumed that: (1) a transition from the NSP seronegative status to the NSP seropositive status between two sampling sessions reflected the occurrence of a natural infection event during this period; (2) a transition from the SATs seronegative status to the SATs seropositive status between two sampling sessions reflected the occurrence of either natural infection or vaccination during this period; (3) for SATs and NSP a transition from the seropositive status to the seronegative status between two sampling sessions reflected a decrease of respectively vaccinal or natural antibodies below a threshold where immunity is inefficient at protecting against an infection; and (4) an individual in the SATs seropositive status at a given sampling session should be better immunized against infection until the next sampling session than a SATs seronegative individual.

Statistical modeling

Frequencies of status transitions in the different sites and seasons

As a first descriptive step, we assessed the frequency in the sampled cattle of the different epidemiological transitions as a function of season, site and test (SATs vs. NSP) (Fig. 2). The different transitions are (1) no antibody acquired (−/−): the individual was recorded negative both at (t) and at (t + 1); (2) antibodies acquired (−/+): the individual was recorded negative at (t) and positive at (t + 1); (3) antibodies persisted: (+/+) the individual was recorded positive at both (t) and (t + 1); and (4) antibodies lost: (+/−) the individual was recorded positive at (t) and negative at (t + 1).

Epidemiological status and transitions histories for FMD in cattle. ELISA NSP detected natural antibodies from infection and ELISA SATs detected natural and vaccine antibodies produced from infection and/or vaccination.

Models for depicting incidence and reversion of serological antibody status

As a second analytical step, statistical models were used to depict variations in NSP and SATs serological incidence and reversion rates. In the statistical models, the dependent variable was binary: the serological status of an individual either changed or remained unchanged during a time interval. The statistical models used were thus generalized linear mixed models (GLMM) with binomial error structures. Because the same individual could produce more than one pair of successive serological records and because individuals were aggregated in herds, independence of statistical units was questionable. Individual and/or herd random effects were thus included in the models whenever pseudo‐replication at the individual and/or herd levels was observed. Finally, because not all cattle have been sampled at each of the 5 sampling sessions, the time between two sampling sessions varied in our longitudinal surveys from 4 to 8 months. It was necessary to account in statistical models for this variation in the length of the time interval over which serological transitions were observed. To do so, a complementary log‐log link function was used and the log of the length of the time interval was included as an offset in all the models presented below (Fortin et al. 2008). Goodness of fit was assessed through Pearson overdispersion test (Bolker et al. 2009).

Models were first built to depict spatial and temporal variations in serological transition rates. In these models the fixed effects of season and site were used. Alternative models were built where the effect of site was substituted by the effects of area (KAZA‐TFCA vs GL‐TFCA) and distance to the national parks (close vs. far‐control). In addition, the fixed effects of age and the random effects of individual and/or herd were considered in these models. Selection among models including different combinations of these factors and interactions was performed using Akaike Information Criterion (Burnham and Anderson 2002, Bolker et al. 2009).

Finally, the effects of the factor that could underlie the pattern of variation in serological incidence and/or reversion rates were tested. We focused particularly on the effects of vaccination, of potential immunity and of the rate of contacts with buffalo estimated with GPS telemetry data. The effects of the contact rate with buffalo were only assessed in sites close to national parks, because no GPS collars have been fitted on cattle far from national parks where contact rates with buffalo were assumed to be null. The effects of vaccination and potential immunity were assessed only in the GL‐TFCA where cattle have been vaccinated during the study period. The models used to assess the influence of these factors also included the fixed effects of age and the random effects of individual and/or herd. Likelihood ratio tests were used to assess the statistical significance of these effects (Bolker et al. 2009).

All the statistical analyses were performed using the R (R Development Core Team 2011) and 2012 RStudio, Inc. free softwares.

Results

Host contacts

Hosts spatial distribution and contacts

Home range estimations by the movement kernel method showed that cattle and buffalo shared a fraction of their habitat during the study period (Figs. 3–5). The gradient of spatial overlap ranged from 1.4% for Dete‐Hwange to 2.8% for Malipati‐Gonarezhou and 16.9% for Pesvi‐Kruger. The locations of contacts between cattle and buffalo are indicated on the map (Figs. 3–5). More than 70% of contacts (on average for the 3 sites) occurred less than 500 m from water pans or river beds while cattle spent on average 17% of their time close to water.

Cattle and buffalo home ranges (red‐ brown & red‐yellow respectively) and contacts per sites at the interface of Hwange‐Dete national parks (NP) in southern Africa. The location of contacts events between cattle and buffalo is represented by pink stars (i.e., a cattle position recorded within 300 m of a buffalo position less than 15 days after the buffalo position had been recorded).

Cattle and buffalo home ranges (red‐brown and red‐yellow, respectively) and contacts per sites at the interface of Kruger‐Pesvi national parks (NP) in southern Africa. The location of contacts events between cattle and buffalo is represented by pink stars (i.e., a cattle position recorded within 300 m of a buffalo position less than 15 days after the buffalo position had been recorded).

Contact rates per site and season

The contact rate varied according to site and season and, within a site and season, among individual cattle (Fig. 6). For 39 out of 70 cattle/season combinations, no contacts with buffalo were recorded. Non null contact rates ranged from 6.24 × 10⁻⁸ (buffalo contact index: 1.83) to 28172 × 10⁻⁸ (buffalo contact index: 10.25) contacts per pair of buffalo/cattle positions (Fig. 6). Irrespective of the season, the average over cattle of the contact index was highest in Pesvi‐Kruger, slightly lower in Malipati‐Gonarezhou and much lower in Dete‐Hwange (Fig. 7). Seasonal variation was relatively consistent across sites. In the three sites the average contact index was relatively low during the rainy season and relatively high during the hot and dry season (Fig. 7).

Contacts rate between cattle and buffalo: frequency of “buffalo contact index” depending on individual cattle.

Contacts rate between cattle and buffalo: “buffalo contact index” average depending on sites and seasons.

Serological analysis and modeling

Laboratory cross validation

With Kappa statistics estimated at 0.6 and 0.61 for the blocking ELISA tests for SAT1 and SAT3 respectively, there was a substantial agreement between the GVS and the OVI laboratories for these two tests (n = 54/test, p‐value < 0.001). We thus considered the SAT1 and SAT3 test results reliable and appropriate for being used in our models. However, the Kappa statistic for the SAT2 blocking ELISA tests was inferior to 0.2, reflecting poor agreement between laboratories. SAT2 results were thus not used in subsequent analyses. SAT1 and SAT3 results were combined by considering that a positive sample for either SAT1 or SAT3 could be considered as positive for the SATs test.

FMD serology on buffalo

The SATs serological prevalence estimated from buffalo captured in national parks by the OVI was 94.7% for Gonarezhou (n = 38, October 2008) (Caron et al., in press); 95.7% (n = 47, June 2010) and 100 % (n = 67, July 2011) for Kruger (unpublished data) and 80% for Hwange (n = 15, December 2009) (Wildlife Unit records, Harare, Zimbabwe).

FMD serology on cattle: description of sample prevalence

NSP and SATs prevalence in samples collected in the GL‐TFCA (Malipati‐Gonarezhou/Pesvi‐Kruger and Choumpani) and particularly in sites close to the national parks (Malipati‐Gonarezhou and Pesvi‐Kruger) was higher than in samples obtained in the KAZA‐TFCA (Dete‐Hwange and Tinde). In all the sites, SATs prevalence was higher than NSP prevalence (Table 1).

NSP and SATs FMD serological prevalence estimations (with 95% confidence intervals) in the study cattle populations.

Frequencies of all possible status transitions in different sites and seasons

NSP serological status showed a much higher stability in the KAZA‐TFCA (Dete‐Hwange and Tinde), where the seronegative status (−/−) was by far the most frequently observed transition, than in the GL‐TFCA (Malipati‐Gonarezhou, Pesvi‐Kruger and Choumpani) where antibody acquisition (−/+), antibody persistence (+/+) and antibody loss (+/−) was observed at substantial frequencies (Fig. 8). Most individuals in the KAZA‐TFCA thus remained in the NSP sero‐negative status during the study period while antibody acquisition (−/+) was observed in a substantial fraction of the GL‐TFCA cattle population (Fig. 8). Antibody loss (+/−) was relatively frequent in both areas, suggesting that NSP antibodies were not very long lasting. In both areas there was a clear seasonal pattern in the relative frequency of the different transitions. Antibody acquisition (−/+) and antibody persistence (+/+) occurred mainly during the hot‐dry and rainy seasons while the frequencies of antibody loss (+/−) and stability of the seronegative status (−/−) reached their maximum during the cold‐dry seasons (Fig. 8).

Frequencies of all possible status transitions in different sites and seasons: (1) no antibody acquired: (−/−) an individual was recorded negative both at (t) and at (t + 1), (2) antibodies acquired: (−/+) an individual was recorded negative at (t) and positive at (t + 1), (3) antibodies persisted: (+/+) an individual was recorded positive at both (t) and (t + 1) and (4) antibodies lost: (+/−) an individual was recorded positive at (t) and negative at (t + 1) for NSP test.

For SATs serological transitions (Fig. 9), the contrast between the two areas was striking. In the KAZA‐TFCA, the most frequently observed transition was stability of the seronegative status (−/−) while in the GL‐TFCA antibody persistence (+/+) was clearly the dominant transition. Moreover, SATs serological status seemed more stable in the GL‐TFCA, with SATs antibody persisting over the study period in most individuals, than in the KAZA‐TFCA where antibody acquisition (−/+) and loss (+/−) were frequently observed. Unlike for NSP serological transitions, no clear seasonal pattern was observed in the relative frequency of the different SATs transitions.

Frequencies of all possible status transitions in different sites and seasons: (1) no antibody acquired: (−/−) an individual was recorded negative both at (t) and at (t + 1), (2) antibodies acquired: (−/+) an individual was recorded negative at (t) and positive at (t + 1), (3) antibodies persisted: (+/+) an individual was recorded positive at both (t) and (t + 1) and (4) antibodies lost: (+/−) an individual was recorded positive at (t) and negative at (t + 1) for SATs test.

Modeling variation in NSP serological incidence and reversion according to sites and seasons

According to AIC selection, NSP spatial and temporal incidence rate variation was best described with a model including the random effect of individual, the fixed effects of age, area, season, distance to national parks and the interaction between area and season (Fig. 10 and all statistical detailed are given in Appendix D: Tables D1 and D2). According to this model, incidence increased with age, was slightly higher in sites close to national parks and was much higher in the GL‐TFCA, where it frequently reached 0.2–0.3, than in the KAZA‐TFCA where it was most often <0.15. The seasonal variation pattern differed slightly between the two areas. In the GL‐TFCA, incidence was maximum, around 0.2–0.3, during the hot‐dry season, slightly lower, around 0.15–0.20, during the rainy season and much lower, around 0–0.5, during the cold‐dry seasons. In the KAZA‐TFCA incidence was also minimum, close to 0.0 during the cold‐dry seasons but was higher, around 0.10, during the rainy season than during the hot‐dry season (around 0.05).

FMD Serological incidence and reversion of natural and vaccinal antibodies depending on sites and seasons effects for NSP test.

The model selected to describe spatial and temporal variation in NSP reversion rate included the random effect of individual, the fixed effects of area, season, distance to the national park and the interaction between distance to national parks and season (Fig. 10 and all statistical detailed are given in Appendix D: Tables D3 and D4). According to this model, reversion was much lower in the GL‐TFCA with values often <0.2 than in the KAZA‐TFCA with values always >0.6 and much lower in sites close to national parks. The seasonal variation pattern differed slightly between sites close to national parks and sites far from national parks. In all sites, reversion was highest during the cold‐dry seasons, particularly during the second year. However, reversion was lowest during the rainy season in sites far from national parks while it was lowest during the hot‐dry season in sites close to national parks.

Variation in SATs serological incidence and reversion according to sites and seasons

The model selected for depicting SATs spatial and temporal incidence rate variation included the random effect of herds and not of individuals because most models including an individual random effect failed to converge (Fig. 11 and all statistical details are given in Appendix D: Tables D1 and D2). The fixed effects included in the selected model were age, area, season, distance to national parks and the interaction between season and distance to national parks. According to this model, incidence rate increased with age and was much higher in the GL‐TFCA with most incidence rates close to 1 compared to the KAZA‐TFCA where incidence was most often <0.2. In sites close to national parks, incidence rates showed little seasonal variation while in sites far from national parks incidence rates were lower during the cold‐dry seasons than during the other seasons.

FMD Serological incidence and reversion of natural and vaccinal antibodies depending on sites and seasons effects for SATs test.

The model selected to describe spatial and temporal variation in SATs reversion rate included the random effect of herd rather and of individual for the same reason as is noted above. The fixed effects included in the selected models were age, area, distance to national parks, season, the interaction between distance to national parks and season and the interaction between area and distance to national parks (Fig. 11 and all statistical details are given in Appendix D: Tables D3 and D4). According to this model, reversion rate decreased with age and was much lower in the GL‐TFCA where it never exceeded 0.2 compared to the KAZA‐TFCA where it ranged between 0.25 and 1.0. Reversion rate was lower in sites close to national parks, particularly in the KAZA‐TFCA. Reversion showed little seasonal variation, being slightly higher during the first cold‐dry season at the beginning of the study period than during the other seasons.

Effect of contacts with buffalo on serological incidence and reversion in sites close to national parks

Models including the random effect of individuals, the fixed effects of age and the effect of the ‘buffalo contact index', in sites close to national parks were used to test the effect of contacts with buffalo. NSP and SATs incidence rate increased with increasing ‘buffalo contact index' (LRT p‐values: 0.018 for NSP and <0.0001 for SATs). Conversely, according to models with the same structure, NSP and SATs reversion rates decreased with increasing ‘buffalo contact index' (LRT p‐values: 0.006 for NSP and 0.0002 for SATs) (Fig. 12 and Table 2). Changing the size of the temporal or of the spatial dimension of the window used to define contacts did not change the outcome of the test of the effect of the contact index on epidemiological transition rates. Indeed, a sensitivity analysis presented in Appendix E, shows that contact indices computed using two‐, three‐ or four‐fold narrower temporal or spatial windows influenced serological transition rates significantly and with consistent signs.

FMD Serological incidence and reversion probabilities of foot and mouth antibodies in cattle populations depending on the buffalo contact rate for the two models developed (NSP for ‘natural antibodies' and SATs for ‘natural' and ‘vaccinal' antibodies).

Likelihood ratio test (LRT) of the factors underlying variation in FMD serological incidence and reversion.

Influence of vaccination sessions on NSP serological incidence and reversion in the GL‐TFCA

Models were built for NSP serological incidence and reversion in vaccinated sites (GL‐TFCA and Choumpani) that included the random effect of individuals, the fixed effects of age and the effect of a variable ‘vaccination session'. This variable distinguished site × period combinations when and where a vaccination session was undertaken from combinations when and where no vaccination session occurred. NSP incidence rate in the GL‐TFCA and Choumpani was higher when and where vaccination took place (LRT p‐value: 0.009). Conversely, NSP reversion rate in the GL‐TFCA and Choumpani was lower when and where vaccination took place (LRT p‐value: 0.01) (Table 2).

Influence of SATs serological status on NSP serological incidence and reversion in the GL‐TFCA

Models were built for NSP serological incidence and reversion in vaccinated sites (GL‐TFCA and Choumpani) that included the random effect of individuals, the fixed effects of age and the effect of a variable ‘presence of vaccinal antibodies' (SATs (+) at t). This variable distinguished individuals where FMD antibodies had been detected by the SATs test at the beginning of a time period from individuals where such antibodies had been detected. SATs seropositive individuals showed lower NSP incidence (LRT p‐value: 0.01) than SATs seronegative individuals. The SATs serological status had no influence on NSP serological reversion (Table 2).

Discussion

Diseases transmission at wild‐domestic interfaces is an important epidemiological issue on most continents. As identified in studies on brucellosis circulation between elk and cattle in North America (Proffitt et al. 2011) and on tuberculosis circulation between badger and cattle in England (Donnelly et al. 2006, Woodroffe et al. 2009), characterizing and quantifying the ‘infectious' contacts between domestic and wild hosts seem to be crucial to understand the complex dynamics of multi‐hosts systems. By combining telemetric and epidemiological approaches on sympatric cattle and buffalo, our study provides strong evidence that contact rate with wild buffalo significantly influences FMD dynamics in cattle populations living at the periphery of conservation areas in Zimbabwe.

We demonstrated an association between the probability of a cattle sero‐converting or ‐reverting over a 4 months period and the frequency with which the herd to which that cattle belong used space occupied by buffalo less than two weeks before. As direct contacts were almost never recorded, our results suggest that the survival of FMD viruses in the environment is high enough for delayed contacts to generate virus spillover from buffalo to cattle. Furthermore, in spite of strong serological evidence for infection events, no clinical signs for FMD were detected in the some monitored cattle populations during the course of our study which tends to support the idea of a silent virus circulation in some areas of southern Africa. Finally vaccination campaigns, undertaken in cattle populations living close to wildlife monitored in our study, seemed not reach a large enough coverage to prevent introduction of FMD viruses from wildlife, and particularly buffalo, and subsequent viral circulation in cattle populations. The main results of our study are synthesized in (Table 3).

Results guide for discussion.

Buffalo‐cattle contact patterns

Simultaneous radiotracking of sympatric cattle and buffalo revealed significant differences in contacts patterns between sites, occurring more frequently in the GL‐TFCA sites compared to the KAZA‐TFCA sites. Although the permeability of the boundaries was similar in the 3 interface sites, very different contact rates were observed. However, the seasonal patterns of contacts were similar among sub regions and sites: lowest contacts during the rainy season, increasing during the dry season, usually reaching a maximum during the hot‐dry season. This seasonal pattern suggests that contacts between cattle and buffalo are driven by resource availability, a limiting factor during dry seasons. African wild ungulates such as buffalo usually avoid cattle and humans (e.g., Kock 2003), but the scarcity of water and key forage resources in our study sites during the dry season may have reduced the possibilities to do so. A major difference is observed in terms of water distribution among study sites (Figs. 3–5), with river systems for Pesvi‐Kruger and Malipati‐Gonarezhou (GL‐TFCA), and scattered water pans for Dete‐Hwange (KAZA‐TFCA). This contrast in water distribution pattern could explain the difference in contact rates between our two study areas. Indeed, being restricted to living close to rivers may lead to a strong competition for water access, whereas scattered water distribution may reduce the interaction probability.

It should be noted that the intent of this study was not to estimate the total number of contacts between the selected cattle herds and all individual buffalo. This would have required enormous resources to equip with GPS collars all individual buffalo likely to come into contact with the cattle herds monitored. Instead, we took advantage of the gregarious habits of buffalo that move in cohesive herds (Sinclair 1977). Fitting 2–3 GPS collars on adult females belonging to known buffalo herd in the study areas allowed us to capture the movements of the herds likely to come into contact with cattle herds. In addition, buffalo annual home range overlaps between females collared in the same herd ranged between 60 and 88%, indicating that individual females remained associated with the same herd during the course of our study, although occasional fusion‐fission events of groups of female buffalo have been described in southern Africa (Cross et al. 2005). We assumed that the herds containing adult female individuals represented the largest demographic compartment in resident buffalo populations, although small bachelor herds (Estes 1993) may have occasionally roamed in the area. The total number of individual contacts recorded is probably underestimated, as cattle may have come into contact with several individuals from the same herd or with bachelor herds, but the biases are unlikely to have been heterogeneous across sites and seasons, and the comparison of the trends thus remains appropriate.

Influence of buffalo‐cattle contact rates on natural dynamics of FMD in cattle

The observed spatial and temporal variation in buffalo‐cattle contact frequency coincided with differences in FMD dynamics in cattle populations. This supports the hypothesis that buffalo act as the main reservoir of virus (Vosloo et al. 2002b, Thomson and Bastos 2004), which is likely to be responsible for primary outbreaks in the cattle populations. Our results show that the incidence of NSP antibodies in cattle herds, which reveal the natural FMD infections, was higher in the GL‐TFCA sites compared to the KAZA‐TFCA sites. Within each study site, NSP incidence was significantly higher at sites located close to national parks, compared to others. Lower incidence in sites far from national parks is not a demonstration of a cause to effect relationship between proximity to buffalo populations and FMD incidence in cattle because none of the other environmental variables potentially influencing FMD epidemiology was under our control. However, it supports the hypothesis that national parks harbor FMD wild host reservoir populations.

Finally in GL‐TFCA, incidence peaked when contact frequency was the highest, i.e., during the hot‐dry season. The NSP reversion pattern was also compatible with the hypothesis that contacts with buffalo trigger FMD infection in cattle. It was higher, in the KAZA‐TFCA, where contact rates with buffalo were relatively lower than in the GL‐TFCA. In a given area, it was also higher in site located far from national park.

Immunity induced by the production of antibodies following infection

NSP reversion rates were quite high, especially in sites where FMD virus circulation was less intense and where multiple exposures to the virus were less likely to occur during any given time interval. In comparison, SAT reversion in sites without vaccination were lower than for NSP; this may be due to differential temporal persistence of antibodies against an epitope or another. Furthermore, all our models showed a positive age effect on the probability of being seropositive for NSP that would not be expected due to the high reversion rate. One possible interpretation of this age effect would be that individuals that have been frequently infected by FMD viruses develop faster and stronger immune response after multiple exposures (Anderson and May 1991).

Influence of vaccination on FMD dynamics in cattle

Vaccination against FMD, performed in the GL‐TFCA sites by the Government Veterinary Services during the course of this study, seemed to provide some protection against infection at individual cattle level. Cattle, in which SATs antibodies had been detected at the beginning of a time interval, had a significantly lower NSP serological incidence than SATs negative cattle. However, we also found relatively high NSP prevalence and incidence rates in the vaccinated area (GL‐TFCA), which suggest that the protection induced by the vaccination strategy was not completely efficient. This may be explained by the fact that the vaccine used does not fully protect against the strains currently circulating in this endemic region. Cattle that have recovered from infection with one FMD virus type are susceptible to re‐infection with any of the other 6 virus types (Vosloo and Thomson 2004). Moreover, the percentage of vaccinated hosts in the population might not be sufficient, below the vaccination coverage of 50–90% recommended by (Keeling et al. 2003), in order to reduce the length of the outbreak by breaking chains of transmission. This could be particularly the case for the Pesvi site where, according to the veterinary services, only 33% to 62% of the cattle have been vaccinated during the three vaccination sessions organized in the course of our study period (Appendix B). The frequency of vaccination campaigns (once a year) was also below recommendations for prophylactic control of FMD (Parida 2009), as vaccines offer only short‐lived protection in endemic regions (from 3 to 6 months). Finally, the positive significant relationship between natural serological incidence (NSP) and the vaccination campaigns implemented (between two sampling sessions) indicates that these campaigns were organized when the virus was already circulating in the cattle populations. The low reversion rates (mostly <0.20) estimated in the vaccinated area (GL‐TFCA) does not fully support the hypothesis of a short duration of vaccinal immunity. However, SATs serological results are difficult to interpret since the SATs test detects both vaccinal and natural antibodies. The relatively low reversion rates estimated in the GL‐TFCA could thus be the result of the concomitant viral and vaccinal exposures creating a booster effect.

Uncertainty on serological states

The FMD serological diagnostic tests have to be used carefully when trying to infer the history of exposure of individuals. Indeed, in the sites of the KAZA‐TFCA where we would expect matching NSP and SAT serological results because of the absence of vaccination, NSP prevalence was lower than SATs prevalence (Table 2). This difference could result from a relative lack of sensitivity of the NSP test and/or of specificity of the SATs test, which could be due to differences in the dynamics of the specific immune responses involved in each test. The NSP test, commercialized internationally, could indeed be less sensitive than the SATs test developed for the purpose of detecting the strains circulating in southern Africa. Our prevalence, incidence and reversion estimations are thus certainly biased to an unknown extent. However, we assume that although the bias may differ between the two tests, it should be homogeneous for each test across sites, seasons, age classes and contact rate levels. Under such a homogeneity assumption, comparisons among sites and seasons and across ages and contact rate indices can be conducted. One potential approach to unravel epidemiological processes while taking account of various sources of uncertainty (duration of immunity following vaccination or natural infection, sensitivity and specificity of the two serological tests) would be to use state‐space modeling. State‐space modeling would allow estimating probabilities of transition between epidemiological states (susceptible, infected and recovered) along with parameters such as sensitivities and specificities that link observations to epidemiological status. However, the state‐space modeling approach would require statistical developments that are well beyond the scope of the present study (McClintock et al. 2010).

Implications for the management of FMD at the wildlife‐livestock interface

Control of FMD could arguably be considered as one of the major constraints to the coexistence of livestock and wildlife in southern Africa. Although strict land policies, animal movement controls and fencing can largely resolve the problem if strictly applied (Thomson 1995), the un‐sustainability of these control options and the indirect costs induced have raised concern for decades among conservationists (e.g., Taylor and Martin 1987). Even the best maintained fences cannot restrain the movements of all wild animals (Sutmoller et al. 2000, Dion et al. 2011), and these events may result in outbreaks of infectious diseases in neighboring livestock, including FMD (Hargreaves et al. 2004). In many areas in southern Africa fences, separating livestock populations from wild reservoirs of FMD, are absent or no longer functional because they have not been properly maintained and/or because they have been destroyed by wildlife or people (e.g., Ferguson and Hanks 2010).

Current FMD control measures in southern Africa also strongly rely on vaccination of livestock populations at risk of spill‐over infection from wildlife populations, especially from buffalo. This prophylactic strategy at the wildlife‐livestock interface based on vaccination suffers from several technical and practical limitations. First, the current vaccine used in southern African region where FMD is endemic may not provide protection against strains that circulate in the buffalo populations. This would require frequent isolations of strains from wild and cattle populations, to adapt local vaccines, a strategy not currently implemented. Second, the immunity provided to vaccinated cattle only lasts for 3–6 months. The frequency of vaccination campaigns could thus be increased. Developing countries such as Zimbabwe, which also faced severe economic difficulties during the past decade, lack the appropriate financial resources to purchase a sufficient number of vaccine doses to immunize a sufficient proportion of the cattle population at risk of infection from buffalo reservoirs to prevent outbreaks (Keeling et al. 2003). Instead, the strategy of “ring vaccination” adopted by default aims at containing outbreaks so that they do not spread to other areas of the country. However, our results suggest that in some areas the virus circulates in cattle without provoking detectable symptoms and therefore without triggering vaccination campaigns. Improving surveillance strategies, for instance through the use of NSP serological diagnostic on randomly selected animals, in all risky areas (i.e., close to reservoir population) and during risky periods (i.e., dry season) could be an important step towards FMD control.

In this paper, we investigated wildlife‐livestock contacts and FMD circulation at unfenced interfaces in Zimbabwe, included in Transfrontier Conservation Areas. We focused on the potential role of the buffalo reservoir population in the introduction of the virus in cattle. Using repeated serology at the individual level and generalized linear modeling of serological status transitions, we found that the dynamics of the FMD infections in cattle was strongly related to spatio‐temporal variations of contact rates with buffalo. Identifying the determinants and intensity of the contact rates is essential to understand, and possibly control, the natural dynamic of the disease. Our results indicate that interspecific contacts, possibly leading to disease transmission, are relatively rare and localized events. We argue that this provides opportunities to improve FMD management, by manipulating access to key resources (forage or water) or adapting livestock and/or wildlife management practices in order to reduce the frequency of buffalo‐cattle contacts. Variation in resource distribution (between seasons, interannual), livestock and wildlife management strategies and predation risks are likely to be important drivers of cattle‐buffalo interactions. Further investigation is required to adequately test their relative contribution to disease transmission at the wildlife‐livestock interface (Caron et al. in press). Furthermore, transboundary animal diseases involving wild species threaten the success of Transfrontier Conservation Areas by triggering competing claims over the management of human/wildlife interfaces between the veterinary and conservation sectors, as well as the often neglected but crucial local community component (de Garine‐Wichatitksy et al. 2012). FMD management and control in these areas should be negotiated and accepted by all stakeholders, instead of being regarded as an exclusive veterinary issue, imposing international and governmental priorities without considering local dynamics and context (Osofsky et al. 2005, Michel et al. 2006). Finding novel solutions and interventions to control disease transmission at the wildlife‐livestock interface is the contemporary challenge to the veterinary community, disease biologists, community representative and wildlife managers alike (Kock 2003).

Acknowledgments

The authors thank Roger F, Monicat F, Jori F, Zisadza P, Hofmeyr M, Lebel S, Leger L, Tran A, Gely M, Butete B, Mandina C, Kameva L, SANParks, Governmental Veterinary Services of Zimbabwe, the Park and Wildlife Management Authority of Zimbabwe. We thank all the cattle owners involved in the project for their great collaboration. The project was conducted within the framework of the Research Platform “Production and Conservation in Partnership” (RP‐PCP) and as part of the AHEAD initiative. It was funded by Cirad, the French Ministry of Foreign Affairs (Harare, SCAC funds 2009‐2012), the EU (PARSEL project), the Zimbabwean office of the Food and Agriculture Organisation (FAO) and the French ANR (FEAR project ANR‐08‐BLAN‐0022).

Appendix A

Table A1. Comparison of the different study sites in national parks (NP).

Appendix B

Table B1. Vaccination sessions and percentage of cattle vaccinated per site. Data recorded by the veterinary services in Chiredzi. The FMD vaccines used were sourced from the Botswana Vaccine Institute (BVI).

Appendix C

Home range computation and volume overlap calculation

Movement‐based methods have been shown to estimate more reliable space use than classical location‐based approaches (Cumming and Cornelis 2012). The extent to which the collared individuals shared space was quantified using the volume of intersection statistics (Fieberg and Kochanny 2005). Analyses were performed using R software (R Development Core Team 2011) and RStudio 2012 free software. Geographic coordinates were projected using a rgdal package. Home ranges and home range overlaps were computed using adehabitat HR and raster packages. Maps were designed using ArcGis (version 9.3.1; ESRI, Redlands, California).

Appendix D

Table D1. Variance components for serological incidence.

Table D2. Selection among alternative GLMM to describe serological incidence. Abbreviations are: site = Dete, Tinde, Malipati, Pesvi, Choumpani; season = cold‐dry1, hot‐dry, rainy, cold‐dry2; area = KAZA‐TFCA, GL‐TFCA; control = Tinde, Choumpani. Random effects: individuals for NSP and herds for SAT.

Table D3. Model selection serological reversions.

Table D4. Model selection serological reversion: GLMM. Random effect: individuals for NSP and herds for SATs.

Appendix E

We computed the contact indices for each cattle at each season using different time and space windows. The contact indices obtained with time windows of 7.5 days, 3.7 days, 1.8 days and 0.9 days were highly correlated with contact indices obtained with time windows of 15 days. Similarly, the contact indices obtained with space windows of 150meters, 75 meters and 30meters were highly correlated with contact indices obtained with a space windows of 300 meters. So, reducing the length of the time or the space window is unlikely to have any influence on the effect of contact index on serological rates.

Fig. E1. Correlation between contact indices computed with varying spatial window lengths for a fixed temporal window of 0–15 days.

Fig. E2. Variation over seasons and sites of the average of contact indices computed with varying spatial window lengths for a fixed temporal window of 0–15 days.

Fig. E3. Correlation between contact indices computed with varying temporal window lengths for a fixed spatial window of 0–300 m.

Fig. E4. Variation over seasons and sites of the average of contact indices computed with varying temporal window lengths for a fixed spatial window of 0–300 m.

Table E1. To strengthen our point that our results are little sensitive to the length of the spatial and temporal window, we did a sensitivity analysis, testing the influence of contact rates on sero‐conversion using contact rates computed with varying spatial and temporal windows.

Fig. Fig. 8. Cattle and buffalo home ranges (red‐brown and red‐yellow, respectively) and contacts per sites at the interface of Gonarezhou‐Malipati national parks (NP) in southern Africa. The location of contacts events between cattle and buffalo is represented by pink stars (i.e., a cattle position recorded within 300 m of a buffalo position less than 15 days after the buffalo position had been recorded).