Fatal attraction: vegetation responses to nutrient inputs attract herbivores to infectious anthrax carcass sites

Abstract

Parasites can shape the foraging behaviour of their hosts through cues indicating risk of infection. When cues for risk co-occur with desired traits such as forage quality, individuals face a trade-off between nutrient acquisition and parasite exposure. We evaluated how this trade-off may influence disease transmission in a 3-year experimental study of anthrax in a guild of mammalian herbivores in Etosha National Park, Namibia. At plains zebra (Equus quagga) carcass sites we assessed (i) carcass nutrient effects on soils and grasses, (ii) concentrations of Bacillus anthracis (BA) on grasses and in soils, and (iii) herbivore grazing behaviour, compared with control sites, using motion-sensing camera traps. We found that carcass-mediated nutrient pulses improved soil and vegetation, and that BA is found on grasses up to 2 years after death. Host foraging responses to carcass sites shifted from avoidance to attraction, and ultimately to no preference, with the strength and duration of these behavioural responses varying among herbivore species. Our results demonstrate that animal carcasses alter the environment and attract grazing hosts to parasite aggregations. This attraction may enhance transmission rates, suggesting that hosts are limited in their ability to trade off nutrient intake with parasite avoidance when relying on indirect cues.

1. Introduction

Factors that affect host foraging ecology can be fundamental to disease dynamics, by regulating parasite transmission [1]. For a variety of host–parasite systems, hosts are infected with parasites by ingesting free-living parasite stages along with food or are parasitized while foraging. Where parasites can be detected, these parasite-inhabited locations can be avoided by foraging hosts, and anti-parasite behaviours can have a similar or greater effect on animal foraging patterns than anti-predator behaviours [2–4]. In fact, behavioural avoidance of parasite exposure can be more important than immunity in reducing infection in a population [5]. If parasites cannot be detected directly, as is likely to be the case for microparasites and many macroparasite larvae, hosts must rely on parasite-associated cues to avoid parasite infection. Faecal matter is an important cue for the potential presence of parasites with faecal–oral transmission, and how faeces affects herbivore foraging decisions and parasite risk has been extensively studied (e.g. [6–10]). Given that faeces can have a positive effect on nutrients available to vegetation, herbivores face a trade-off between nutrient acquisition and parasite exposure when foraging near faeces-contaminated patches [9,11]. Analogous to faeces, animal carcasses represent detectable potential sources of parasites and nutrients in the environment, yet no studies have assessed herbivore foraging responses to carcass sites and what role these may have in disease transmission.

Foraging is often heterogeneously distributed across a landscape. At smaller scales, nutrient hotspots—characterized by increased levels of nitrogen, phosphorus and several important minerals [12]—spatially concentrate mammalian herbivores in grazing systems [13]. Such nutrient hotspots have been connected to abiotic heterogeneity, such as volcanic soil and catenal effects [12,14], as well as biotic drivers, such as nutrient concentration by herbivores [15,16]. Animal carcasses are important biotic agents that create localized nutrient pulses while at the same time aggregating parasites in the environment. Carcasses can create nutrient hotspots that can persist for several years, altering soil fertility and vegetation response [17–19]. However, it has yet to be determined how herbivores respond to these hotspots, including whether lingering visual or olfactory cues from a carcass serve as a deterrent to herbivores or if these nutrient-rich sites eventually become preferred foraging locations. In selecting carcass-generated nutrient hotspots, herbivores face a trade-off between increased nutrient intakes and a risk of infection by environmental parasites. Avoidance of or attraction to carcass sites, and the relative time scales of these behaviours, may strongly influence parasite transmission rates to susceptible individuals.

We evaluated herbivore behavioural responses to carcass-mediated nutrient hotspots and how these may affect host–parasite contacts over a 3-year period, in a guild of mammalian herbivores in Etosha National Park (Namibia) that succumb to the bacterial pathogen Bacillus anthracis (BA). BA is the causative agent of anthrax, a virulent disease that can kill herbivorous hosts within two weeks of a lethal exposure [20] (although sublethal exposures do occur [21]). BA is an environmentally transmitted pathogen that forms hardy spores, which can persist for years in the environment [22]. The environmental persistence of BA has long been associated with soil properties, weather and climate characteristics (e.g. [23]), and more recently with interactions with invertebrates and microbiota [24,25]. However, factors affecting parasite survival only represent one part of host–parasite contacts. The other part, which in grazing herbivores has been poorly studied, is how, where and when mammalian hosts contact BA in the environment.

Anthrax is endemic in Namibia and considered part of the natural ecosystem in Etosha National Park, allowing for ecological studies of this host–parasite relationship that could not be conducted in areas where outbreaks are managed and carcass sites decontaminated. We conducted a longitudinal study at marked plains zebra (Equus quagga) carcass sites from 2010 to 2013, to assess the effect and duration of (i) carcass nutrients on soil fertility, grass quality and grass biomass, (ii) BA concentrations on grasses and in soil at anthrax carcass sites, and (iii) herbivore presence and foraging activity at anthrax carcass and non-carcass grassland control sites using motion-sensing camera traps.

2. Material and methods

(a). Study area and carcass site selection

This study was conducted in Etosha National Park, Namibia, a semi-arid savannah with seasonal rains falling primarily between November and April (detailed description of study area in [26]). The years of study represent the range of rainfall values in central Etosha, with one of the highest rainfalls on record (2011: 705 mm), average rainfall years (2010: 390 mm and 2012: 378 mm) and a drought year (2013: 222 mm). Anthrax cases in Etosha are recorded annually, with a peak in the cases occurring towards the end of the rainy season (March–April) [22,26].

We focused on adult zebra (2+ years old) carcass sites, because most anthrax cases observed in Etosha are in zebras [22,26], and to standardize carcass body size and its nutrient influence at observed sites (details in the electronic supplementary material; typical carcass site shown in figure 1). In total, 43 carcass sites were selected to assess carcass nutrients, BA concentrations or herbivore foraging. Carcass sites for the nutrient study (n = 8) were all sites that tested negative for BA, to protect human health during sampling and analysis. Anthrax-positive carcass sites were used for camera traps (n = 13) and for BA sampling (n = 21). These sites were spread over a 300 km² area in central Etosha (electronic supplementary material, figure S1), with the mean minimum distances between neighbouring sites of 4.7 km for the nutrient study, 4.5 km for the camera study and 2.0 km for the BA study.

Figure 1.

A typical zebra anthrax carcass site a few days after death in Etosha National Park, Namibia. Scavengers have consumed soft tissue, denuded the surrounding vegetation and dispersed the remains. Note the digesta pile at the centre, marked with the stake, and the partial skeleton visible in the upper left outside the denuded area.

(i). Carcass effects on soil and grasses

Carcass sites for the nutrient study included six formed in 2010 and one each in 2011 and 2012 (month of death from January to June). Sites were sampled once per year at the end of the growing season (late March/early April) to evaluate the effect of the carcass on soil nutrients and grass quality and biomass. Samples were not collected in the year of death, only the following years, to have one full growing season for grasses to respond to carcass nutrients. Sampling occurred in 2011–2013 and individual sites were sampled up to three times.

Vegetation samples at carcass sites were collected from three sampling zones radiating out from the marking stake: 0–3 m, 3–6 m and 6–9 m (electronic supplementary material, figure S2). The 0–3 m sampling zone represents the location of highest disturbance where most carcass fluids were deposited. Vegetation harvesting for plant biomass was conducted along two transect lines oriented 180° apart. Plant quadrats (80 × 80 cm) were placed in the centre of each sampling zone along the transect lines and all above-ground plant biomass was harvested. Grass biomass estimates for each sampling zone are averages of the two replicate quadrats. Soil cores of 10 cm in depth were collected at the centre of the site near the marking stake (0 m; centre of the digesta pile) and in the centre of each of the plant sampling zones in four directions radiating out from the carcass sites (electronic supplementary material, figure S2). All soil samples from one sampling zone were pooled for analysis. Dry weights of clipped grass samples (leaves, stems and inflorescences) were recorded from each quadrat and grasses were analysed for percentage nitrogen, while soil samples were analysed for percentage nitrogen, pH, organic matter, P, K, Mg, Ca and Na (see the electronic supplementary material for details). Grass nitrogen was only assessed for the 2010 sites, 1 and 2 years after death; grass biomass in the drought year (2013) was insufficient for protein analyses.

(ii). Bacillus anthracis concentrations at carcass sites

Twenty-one anthrax-positive adult zebra carcass sites were marked between 2010 and 2012, and a subset of these sites was sampled in 2012 (n = 7 sites) and 2013 (n = 19 sites) to assess BA concentrations. Sites were considered anthrax-positive if diagnostic blood swabs from the carcass tested positive for BA (determined through culture with isolates confirmed by PCR [27]). As herbivores could potentially ingest BA at a carcass site by grazing on grass leaves, by inadvertently consuming grass roots when grazing or by ingesting soil for trace minerals [26], we sampled the above-ground grass parts, grass roots and the soil surrounding the plants separately to assess if levels of BA contamination varied among these different locations in the environment. Each carcass site was sampled between mid-February and early April. The dominant grass species within 1 m of the marking stake was sampled, and three replicates of this species were collected for analysis at each site. Plants were dug from the soil and carefully shaken to remove soil clods, the roots were clipped and placed in a 50 ml centrifuge tube, and the above-ground parts were placed in a separate 50 ml tube. A sample of the soil surrounding each plant was collected. Samples were refrigerated prior to analysis, with plant samples cultured for BA within 2 days and soil samples within a month of collection. The grass species collected included Enneapogon desvauxii, Chloris virgata, Eragrostis nindensis, Monelytrum luederitzianum and Aristida adscensionis, all of which are consumed by zebras [28].

The concentration of BA in all samples was assessed via bacterial culture in serial dilution on PLET agar using standard soil protocols [20] (details in electronic supplementary material). Colonies of BA were identified morphologically; in the cases where the identification was uncertain, the colony was streaked on blood agar for confirmation tests to determine if the colony was non-haemolytic, had penicillin sensitivity and was lysed by gamma phage [20]. The number of colony-forming units (CFU) was estimated relative to the sample weight (CFU g⁻¹), as plant samples were not a standard weight (all soil samples were 5 g).

(iii). Herbivore use of carcass sites

We used motion-triggered camera traps to monitor herbivore activity in 13 anthrax carcass sites and 13 matched control sites. Controls were placed 100 m from each carcass, with the direction oriented to retain similar landscape features to the carcass site. The location of each anthrax carcass site was marked with rocks placed 2.5 m from the centre in four directions (electronic supplementary material, figure S3) defining a standardized grazing ‘patch’ and providing depth of field for evaluating animal presence from photographs. Control sites were likewise marked with rocks to monitor the same-sized grazing patch as for carcass sites. Each camera was placed 12 m from the centre of the site and 1.2 m above the ground. The cameras were programmed to take photos continuously without delay between sequential triggers of the motion sensor, with 10 photos taken at 1 s intervals for each trigger (using either Reconyx RC55 Rapidfire or PC 800 Hyperfire cameras, with the same model used for matched carcass–control pairs). We focused this study on five herbivore species most commonly observed as anthrax cases in Etosha (table 1).

Table 1.

Anthrax cases in Etosha National Park, 1968–2011, and population estimates from an aerial survey, 2012.

scientific name	common name	anthrax confirmed	anthrax suspected	total	% of all confirmed cases	population estimate (95% CI)
Equus quagga	plains zebra	1586	346	1932	52.3	16 174 (13 310–19 038)
Connochaetes taurinus	blue wildebeest	660	224	884	21.8	2482 (1622–3342)
Antidorcas marsupialis	springbok	384	70	454	12.7	12 267 (10 110–14 424)a
Loxodonta africana	African elephant	297	225	522	9.8	2810 (2042–3578)
Oryx gazella	gemsbok	38	9	47	1.3	5298 (4356–6240)
—	all other species	67	38	105	2.2	—

^aThe springbok population estimate is a considerable underestimate, apparent in the number of visitations by species at camera sites.

Cameras were mounted at carcass/control sites for 11–26 months (mean 20 months), monitoring different locations between March 2010 and March 2013, including eight carcass sites formed in 2010, three from 2011 and two from 2012. After removing non-informative data (from cameras knocked down by animals, battery or memory card failure, or human error) there were 10 996 days of observations. These 26 cameras were triggered a total of 119 226 times and 6.5% of these were by the five herbivore species within the 20 m² patches. In total, within the patches there were 11 783 triggers of springbok, 5196 of zebra, 1043 of wildebeest, 455 of gemsbok and eight of elephant. Given the paucity of data for elephants and that none were observed foraging within the patches, this species was not considered further.

To quantify herbivore grazing events in carcass and control patches, we evaluated photographs at the level of the individual trigger, counting the number of individuals observed in the patch and the number grazing within a 10-photo series (all behavioural assessments were done by W.C.T.). An individual was recorded as grazing if a bite was observed, or from a series of photographs with the individual's head down and moving at grass height as if biting, plucking and chewing. We therefore focus on site use, looking at animal presence and grazing events observed at carcass and control patches. We did not attempt to quantify potential exposures for individual animals, owing to the difficulty of following an individual within a group through sequential triggers. Owing to the length of the study, the number of people involved and the amount of image data produced, we built custom software for efficient data management, annotation and extraction [29] (described in the electronic supplementary material).

(b). Data analysis

We used linear mixed models (LMMs) to assess factors affecting soil P, N, pH, Na, K, Mg, Ca and organic matter. As fixed effects we included the effect of the carcass (from sampling distance) and site age (1, 2 or 3 years post-mortem) as continuous variables, and sampling year as a categorical variable, while controlling for variation among sites using carcass site as a random effect. Sampling year and the interaction between distance and site age were initially included in all models, but excluded from the final model if non-significant. One exception to these methods is for soil P, where the carcass exhibited a strong effect on P evident only at the 0 m sampling distance, and dropping to presumably baseline levels for all other distances (1.5–7.5 m). Therefore for the P model, we compared sampling distance as a categorical variable comparing the inner versus the midpoint of the other sampling distances (0 m versus 4.5 m). We log-transformed soil N, P, Na, K, Mg and Ca measurements to reduce heteroscedasticity in model residuals.

Factors affecting grass biomass at the negative carcass sites were assessed using a LMM, with fixed effects of site age and sampling distance (1.5, 4.5 and 7.5 m) as continuous variables, sampling year as a categorical variable, and the interaction between distance and site age. The random effects included carcass site and a variable to account for heterogeneity in the variance of biomass estimates among sampling years [30]. The effect of the carcass on grass N 1 and 2 years after death was evaluated using LMMs, with sampling distance as a categorical fixed effect and carcass site as a random effect.

Grass and soil sampling at anthrax-positive carcass sites found BA in at least one sample at 20 of 21 sites tested (the negative site was excluded from analysis). The three replicate samples from each carcass site were averaged by sample type for analysis. The concentration of BA spores (CFU g⁻¹) found on above-ground grass components and roots were each compared with the background levels in soil using Wilcoxon rank sum tests (n = 28 paired samples; some sites were sampled in both years). We then evaluated whether CFU g⁻¹ decreased with the age of the site after death (1–3 years) for grass, root and soil samples using Kruskal–Wallis rank sum tests as the count data were highly overdispersed.

The herbivore site use data were analysed based on herbivore presence (N individuals present per camera trigger) and grazing (G individuals per trigger seen grazing) summed daily per site per patch type (carcass versus control) for the four herbivore species separately. If one camera from a site was not collecting data for a period of time, data from the paired camera during that time interval was excluded to avoid biasing comparisons by patch type. Because the distance between control and carcass patches is much smaller than the daily distance covered by all species, each daily observation was regarded as independent of the previous day's observations (see electronic supplementary material, figure S4 for autocorrelation structure).

To analyse the multivariate and presumably nonlinear relationships between site use, time and patch type, we used a statistical nonlinear regression model known as a generalized additive model (GAM) with a non-parametric smoother function [31]. We looked for differences in mean site use, differences in seasonality and differences in trends over time since host death. In the nomenclature below, we let f(X|Y) represent the penalized non-parametrical smoother function estimated for a continuous variable X given a categorical variable Y. In our model below, we use N_t and G_t to represent the number of animals present and the number grazing, respectively, in site S on day t; T to represent the day since start of the first camera series, to account for the effects of events in time that occur across sites (e.g. a drought); J to represent the Julian day depending on site S (i.e. site-specific seasonality); A to represent the time since animal death depending on type of patch; and C to represent a binary variable of patch type (carcass or control). The model also includes constants a as a site-specific intercept and b as the effect of C on the fitted relationship. To account for overdispersion, a quasi-Poisson error family with a logarithmic link was used when modelling N or G, and quasi-binomial error family with a logit link in the logistic regression of G|N, essentially allowing an independent variance to mean ratio in the models. Thus, the models for zebra and springbok were of the form

2.1

and

2.2

where

2.3

and

2.4

The value of the site-specific constants a and the effect of patch type b differ between equations (2.1), (2.3) and (2.4). The error terms ɛ denote quasi-Poisson-distributed (equations (2.1) and (2.4)) or quasi-binomial (equation (2.3)) error families. In equation (2.3), the N term is square-root-transformed to stabilize variance. For wildebeest and gemsbok the smaller number of observations meant that the interaction between Julian date and site was unreliable (f₂, f₆ and f₉), and thus the effect of Julian date was fitted as one effect without the interaction with site. The spatial autocorrelation was surprisingly low, as shown by the non-parametric correlograms (electronic supplementary material, figure S4) with a max ρ < 0.19 for N, and no significant spatial autocorrelation whatsoever for a difference measure of site-specific differences, .

3. Results

(a). Carcass effects on soil and grasses

Soil P was significantly higher at the inner distance (0 m) where the carcass was opened than at the outer distances and did not decline significantly over the 3 years of study (distance: p < 0.0001; site age: p = 0.6201; figure 2a). Soil nitrogen decreased significantly with distance from the carcass centre, a pattern that persisted for 3 years (distance: p < 0.0001; site age: p = 0.1026; figure 2b). Soil pH increased with distance from the carcass centre (p = 0.0096; figure 2c), with no effect of site age (p = 0.6458). Soil Na decreased significantly with distance from the carcass centre, an effect only observed 1 year after death (distance: p = 0.0075; site age: p = 0.0001; distance × site age: p = 0.0063; figure 2d). Soil K significantly decreased with distance from the site centre (distance: p = 0.0156); however, this effect was not strongly evident due to variation in K estimates among sites and years (site age: p = 0.0212; 2012 versus 2011: p < 0.0001; 2013 versus 2011: p < 0.0001; electronic supplementary material, figure S5). There was no detected effect of the carcass on soil Ca, Mg or organic matter (see electronic supplementary material, figure S5; full statistical results of soil analyses in electronic supplementary material, table S1). No effect of Ca in particular is probably due to removal of bones from the site, so that the nutrient influx came primarily from body fluids.

Figure 2.

Effects of zebra carcasses on average soil concentrations of (a) phosphorus, (b) nitrogen, (c) pH and (d) sodium at four sampling distances (0, 1.5, 4.5 and 7.5 m), and on (e) grass biomass and (f) nitrogen from three sampling zones (0–3, 3–6 and 6–9 m).

Grass biomass at carcass sites was significantly higher at the inner (1.5 m) distance although this effect was only apparent 1 year after death (distance: p = 0.0231; site age: p = 0.2504; distance × site age: p = 0.0245; figure 2e). There was no significant difference in grass biomass estimates recorded in an above-average and an average rainfall year (2011 versus 2012: p = 0.6444). However, there was a highly significant effect of year on grass biomass owing to a drought in 2013 (2011 versus 2013: p < 0.0001; figure 2e). Grass nitrogen was significantly higher 1 year after death at the inner distance (1.5 m versus 4.5 m distances: p = 0.0520; 1.5 m versus 7.5 m distances: p = 0.0491; figure 2f), although by 2 years after death, there was no statistically significant effect of the carcass on grass nitrogen (full statistical results of grass analyses in electronic supplementary material, table S2).

(b). Bacillus anthracis concentrations at carcass sites

Bacillus anthracis was found in all three sample types: in soil, on grass roots and on above-ground grass components. Across all sites irrespective of age, there were significantly lower concentrations of BA on grasses above ground than in the soil (p = 0.0027), but no significant difference in concentrations between grass roots and soil (p = 0.5651). When comparing how BA concentrations varied with site age, the concentrations of BA present on grass roots or in soil did not differ significantly among the site ages sampled (p = 0.1337; soil: p = 0.2507; figure 3). However, the concentrations of BA on the above-ground component of grasses did decrease significantly from younger to older sites (p = 0.0029; figure 3). Therefore, 1 year after death there was no significant difference in the concentration of spores among sample types (p = 0.9680), but by 2 and 3 years after death, spore concentrations significantly differed among the sample types (2 years after death: p = 0.0307; 3 years after death: p = 0.0082). This pattern was driven by the decline in BA spore concentrations on the above-ground component over time, which dropped to near zero on 3-year-old sites (figure 3).

Figure 3.

Concentrations of Bacillus anthracis on grass, root and surrounding soil at carcass sites (a) 1 year, (b) 2 years and (c) 3 years after animal death. Here ‘grass’ refers to the above-ground component. Concentrations are quantified as colony forming units (CFU) g⁻¹ of sample material and averaged for replicate samples from each site.

(c). Herbivore use of carcass sites

Four species were present in the camera-monitored patches in sufficient numbers to be included in analyses: springbok (N = 17 516 animal events), zebra (N = 7452), wildebeest (N = 1611) and gemsbok (N = 595). These animal events (N) are the number of individuals observed within a patch during a single trigger of the motion sensor, summed over all triggers.

Multivariate nonlinear models indicated some initial avoidance of the carcass patches after host death, especially for zebra, but the effect was small and transient (figure 4a–d). However, with time the situation reversed, and zebra and springbok exhibited clear and significant preferences for grazing in carcass patches compared with control patches (p < 0.001; figure 4e,f,i,j), resulting in a disproportionate tendency to forage in the potentially infectious patches in the first year after death of the focal animal. The results for wildebeest are more ambiguous, as fewer data existed; however, a greater tendency towards grazing in carcass patches was apparent (p < 0.01; figure 4g,k). The results suggest that for zebra, springbok or wildebeest encountering a site where a zebra has died within the last year, an animal is up to four times more likely to graze at the potentially infectious carcass patch than at a random grassland patch nearby (figure 4e–g). Gemsbok showed no clear foraging preferences; if anything, they displayed an avoidance of carcass patches (though with the smaller amount of data this trend is uncertain: p < 0.1; figure 4l). Carcass and control patches seemingly became indistinguishable again for grazers from 1.5 to 2.5 years after death of the focal animal (figure 4). This seems to match the time scale of carcass effects detected in grass biomass and nitrogen, and early preference for grazing at carcass sites would significantly increase the odds of anthrax transmission from grazing in the first year after death. Full GAM results are in the electronic supplementary material, appendix S.

Figure 4.

The effect of anthrax carcasses on site visitations and grazing behaviour for zebra, springbok, wildebeest and gemsbok. The upper panels (a–d) show the proportion of site visitations that occur at carcass versus control patches over time since animal death (equation (2.1)). The middle panels (e–h) show the ratio of foraging probability at carcass versus control patches (equations (2.2) and (2.3)), given that an animal is present in the patch (i.e. P(G_a|N_a)/P(G_c|N_c), where presence in anthrax carcass versus control patches is denoted by N_a and N_c, and grazing events in anthrax carcass versus control patches by G_a and G_c, respectively). The lower panels (i–l) show the proportion of all observed grazing events that occur in carcass patches (equation (2.4)). Across all plots, red lines show the mean predicted effect of days after host death on visitation or grazing behaviour between carcass and control patches as predicted by the GAMs. The greyscale density clouds indicate a range of these predicted model effects, from bootstrap data drawn independently from the observed ranges of the variables to give an idea of the relative magnitude of the effects (details in electronic supplementary material). The blue circles and lines show the raw response variable data aggregated at 3-day (light blue) and 7-day (dark blue) intervals with smoothing lines. The horizontal grey lines indicate the no-difference line between carcass and control patches: values above this line indicate a preference for (a–d) visiting or (e–l) grazing at carcass patches, below indicate a preference for control patches.

In evaluating the potential biological impact of these results for our understanding of BA transmission, it is important to consider the variation among sites in their attractiveness to herbivores, because it could be argued that a preference for grazing at carcass patches might be most substantial in marginal, nutrient-poor areas where little grazing occurs anyway. If strong attraction was only observed at sites with few herbivore visitations, this would indeed lessen the importance of foraging preference for carcass sites in BA transmission. However, there were no significant correlations between the abundance of animals (N) visiting camera sites and the relative differences among sites in grazing preference between carcass and control patches (Spearman's ρ < 0.3, p > 0.1). While there was site-specific variation in grazing preference between the carcass and control patches, this was independent from the average number of animals present per day. Thus, the preference for carcass patches seems likely to be representative and valid for Etosha's grassland plains, and therefore likely to be ecologically significant for BA transmission in this system.

The number of visitations and grazing events were also assessed in relation to seasonality (by Julian date) and distance to water, but no predictive models (GAMs) of visitation rates could be built using distance to water as a meaningful predictor. While strong seasonal effects in site use are evident (electronic supplementary material, figure S6), the differences in seasonality among sites are not easily explained by the shortest distance (or weighted sum of distances) to surface water.

4. Discussion

The aims of our study were to determine (i) whether herbivores could be exposed to BA when grazing at anthrax carcass sites, (ii) whether herbivores preferentially feed at or avoid anthrax carcass sites, (iii) the duration of processes identified under (i) and (ii), and (iv) the potential importance of carcass sites for foraging-based disease transmission in grazing herbivores. We found that nutrients from carcasses positively alter the environment in ways that are attractive to some, but not all herbivore species, and that anthrax carcass sites represent a significant risk of exposure to BA for grazing herbivores. The time scales of peak site infectivity and attractiveness coincide in the first year after death—evidence that selective foraging may be particularly important in sustaining BA transmission in endemic anthrax areas. When host foraging responses to parasite-associated cues vary with time, the success of the avoidance period in reducing or preventing infection depends upon the lifespan of the parasite in the environment. Here we describe an infection-avoidance mismatch for herbivores foraging at anthrax carcass sites, but this has also been observed for other parasite-associated cues. For example, herbivores most strongly avoid foraging near fresh faeces [8,32], but it can take weeks before nematode eggs in faeces develop into infectious larvae and migrate onto surrounding vegetation [8]. Therefore, if hosts rely on indirect cues to assess temporal changes in parasite risk, trading off parasite avoidance with food acquisition in high-parasite/high-quality foraging locations may be an imperfect strategy.

Animal carcasses represent a significant and localized pulse of nutrients into the environment, which can lead to increased herbivory on plants fertilized by carcasses (e.g. cicada carcasses [33]). Previous studies of ungulate carcass effects on soil and vegetation have been conducted only in temperate or arctic environments [17–19,34], and Towne [17] suggested that abundant scavenger communities would reduce the nutrient effect of carcasses in tropical systems. In our subtropical system, however, we found that carcasses created clear and biologically significant effects on soil P, N and pH for at least 3 years post-mortem, with shorter-term effects on soil Na, K, grass biomass and grass N, despite very swift consumption of soft tissue by scavengers [35]. After body fluids have entered the environment, scavenger removal of the carcass remains may actually enhance the chances of subsequent herbivory, because the presence of skin and bones on the site might provide a visual cue that deters herbivory, or at least prevent vegetation re-growth beneath the remains.

Experimental lethal doses for ingested anthrax are high in herbivores (around 10⁷–10⁸ spores [20]), therefore transmission depends on BA aggregating in the environment and the behaviour of potential hosts at these aggregations. BA spores are known to be concentrated in soils at anthrax carcass sites and can persist there for several years [22,36], patterns that are not affected by scavenger presence or exclusion [37]. Despite the hypothesized importance of grazing in BA transmission [20,38], this is the first study to examine levels of BA contamination on grasses in the natural environment. We found that the above-ground component of grass holds BA spores, and 1 year after death the concentrations on grasses were as high as in soil. However, although spore concentrations remained high in soil and on grass roots over the 3 years of study, by the second year after death concentrations on grasses above ground were significantly lower, and by the third year they were near zero. This indicates that herbivores can be exposed to relatively high concentrations of BA from grazing at carcass sites in the first year when nutrient-rich vegetation has regenerated. If herbivores consume grass roots as well as above-ground components when grazing, they can also be exposed to high BA spore concentrations throughout the second year. This opens the door to species-specific and seasonal differences in exposure risk to BA based on variation in foraging ecology [26,28].

The camera trap study provides further support that transmission of BA is foraging-based and species-specific. Forage selection is hierarchical, from large-scale seasonal movements through daily movements for water, forage, rest and predator avoidance, down to step-by-step decisions about potential food items within sensory range. Our results suggest that the forage selection processes that cause overrepresentation of grazing in infectious patches take place on the very fine scale of step-by-step decisions. There were signs of early avoidance of carcass nutrient patches after death, the length of which is likely to depend upon when the first rainfall (and thus plant regrowth) occurs relative to the time of death. However, should a herbivore encounter a recently created carcass patch, it is still more likely to graze there than at a nearby grassland patch, a pattern that was particularly strong for zebra. From around 10 months onwards herbivores were as likely to visit carcass patches as control patches, yet were still more likely to graze when visiting carcass patches. After approximately 2 years, attraction to the carcass patches disappeared, and the patch types seem indistinguishable to grazers. Attraction strength varied among species, with zebra, springbok and wildebeest showing periodically strong preferences for grazing at anthrax carcass patches (up to four times the frequency of grazing events in carcass patches relative to control patches in the first year). Similarly, we did not detect any significant preference for carcass patches by gemsbok, which compared with the other herbivore species examined are rarely found as anthrax mortalities. In future research, we will evaluate transmission probabilities, and the timing and intensity of anthrax outbreaks among herbivore species.

Southern Africa has the greatest genetic diversity of BA and is hypothesized to be its geographical origin [39]. Therefore, the environmental conditions in our study area may represent the environment in which the bacterium evolved much of its current life history. A recent experiment demonstrates that BA spores in soil enhance grass seedling establishment, and that even small additions of blood increase grass height [40]. This may indicate a BA–grass mutualism whereby BA benefits from a quick regeneration of nutrient-rich grasses at carcass sites to attract herbivorous hosts. Although not evaluated in this study, exsanguination from anthrax carcasses [41] could lead to a greater localized nutrient release than from non-anthrax carcasses, where blood coagulates in tissues that are subsequently consumed and dispersed by scavengers. From an evolutionary perspective, the life-history strategy of environmental transmission by BA essentially releases its virulence evolution from the constraint of host preservation. On the contrary, killing a host as quickly as possible may increase transmission probabilities, by avoiding a prolonged period of non-transmission in a weakened host. Thus, beyond its ability to produce fulminant infections in hosts, BA seem highly adapted to a life-history strategy of exploiting grazing herbivores.

In conclusion, we find that host foraging responses to carcass sites are dynamic, changing with time from avoidance to attraction to no preference, with the strength and duration of these behavioural states varying among herbivore species. The avoidance period may be sufficient to reduce contact with parasites that only persist for short times in the environment. However, for long-persisting disease agents (such as BA, prions and some coccidia), the ability to survive beyond the initial period of carcass site avoidance may lead to more transmission events than would be expected by chance. These results demonstrate how host foraging ecology and behaviour can affect host–parasite contacts and, ultimately, transmission of environmental parasites in a multi-host natural disease system.

Data accessibility

Data from this study have been made publically available at the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.m49k6.

Funding statement

Funding was provided by NSF OISE-1103054 (to W.C.T.), NIH GM083863 (to W.M.G.) and CEES funds (to W.C.T.).