Potential Bacillus anthracis Risk Zones for Male Bison in Southwestern Montana

Abstract

A re-emergence of anthrax, a zoonosis caused by the long-lived spores of Bacillus anthracis, occurred with a multi-species outbreak in southwestern Montana in 2008. It substantially impacted a managed herd of ~3500 free-ranging bison, Bison bison bison, on a large, private ranch southwest of Bozeman, with ~8% mortality and a disproportionate ~28% of mature males killed; a similar high rate occurred in male elk, Cervus canadensis. Grazing herbivores are particularly at risk for anthrax from ingesting spore-contaminated soil and grasses in persistent environmental pathogen reservoirs. We predicted areas of mature male bison habitat preference on the landscape using GPS collar data and a resource selection function model using environmental covariates. We overlaid preferred areas with ecological niche model-based predictions of B. anthracis environmental reservoirs to identify areas of high anthrax risk for male bison. Overlap areas were distributed across the ranch and not confined to pastures associated with the previous outbreak, suggesting that ongoing pasture exclusion alone will not prevent future outbreaks. Such data suggest vaccination campaigns should continue for bison and results can be further used to prioritize carcass surveillance in areas of greatest overlap.

Anthrax is a world-wide zoonosis of concern caused by the spore-forming bacterium Bacillus anthracis (Hugh-Jones and Blackburn 2009; Fasanella et al. 2010). These spores can remain viable in the soil for extended periods, creating long-lasting pathogen reservoirs on the landscape (Hugh-Jones and Blackburn 2009; Turner et al. 2014). Soil is a primary reservoir for spores, which grazing herbivores ingest, along with contaminated grasses, plants, and roots (Fasanella et al. 2010; Ganz et al. 2014). Herbivores may encounter concentrated spores at anthrax-caused carcass sites, where spores have been found two years or longer after animal death (Lindeque and Turnbull 1994; Turner et al. 2014).

A multi-species outbreak of anthrax occurred on a privately owned ranch in southwest Montana in the summer of 2008 (Blackburn et al. 2014; Morris et al. 2015). Bison, Bison bison bison, kept as livestock, and free-ranging male elk, Cervus canadensis, were the two species most affected, with at least two white-tailed deer, Odocoileus virginianus, also killed (Morris et al. 2015). At least 298 bison were lost, with a male skew (~28% of the male population died versus ~8% of the females; Bagamian et al. 2013). Higher male disease burden has been observed in several bison anthrax outbreaks (Broughton 1992; Gates et al. 1995; Dragon et al. 1999; Nishi et al. 2007; Shury et al. 2009; Salb et al. 2014). In this region, anthrax is considered a re-emerging disease, as no cases had been reported in decades (Blackburn et al. 2014).

The study ranch (Figure 1) occupies ~380 km² of the Northern Madison Study Area (NMSA; Atwood 2006), and is managed for bison and wildlife. Fences restrict bison to specific pastures, but do not prevent movement of cervids or predators. Previously, Morris and others (2015) identified areas where male elk likely overlap with potential anthrax risk zones across the NMSA. Our objective was to model and map male bison resource selection across the ranch during the anthrax risk period (June – August) and estimate overlap with potential anthrax risk zones defined from the 2008 case locations and two subsequent cases in 2010 (Blackburn et al. 2014).

Figure 1.

Study area with male bison GPS fix points (four random points per animal per day), the study ranch boundary and pasture fences, including the pastures accessible to the bison during the study period. The shaded areas represent potential Bacillus anthracis risk zones based on an ecological niche model prediction.

As a pilot study, adult male bison (n=3) were fitted with global positioning system (GPS) collars (model GPS3300L; Lotek, Newmarket, ON, Canada) recording fixes every 30 minutes during the 2012 risk period (Table 1). Capture protocols were developed by the ranch wildlife veterinarian (DLH) and approved by the University of Florida Institutional Animal Care and Use Committee (UF IACUC #201207594 to JKB).

Table 1.

GPS start and end dates and number of fixes for each male bison collared during this study.

Bison	Start	End	Number of Points
Bison 1	1 Jun 2012	15 Aug 2012	304
Bison 2	1 Jun 2012	19 Jul 2012	196
Bison 3	1 Jun 2012	3 Jul 2012	132

Predictions of bison habitat preference were created using a mixed-model logistic regression resource selection function (RSF) with environmental variables as fixed effects and a random intercept term for individual bison (Manly et al. 2002; Gillies et al. 2006). To remain consist with the previous work of Morris et al. (2015), we used a use-versus-available framework with available area defined by a minimum convex polygon (100% MCP) around a pool of GPS points selected from a random draw of four points per day per animal. The MCP was first clipped to pastures available to bison. Available points were randomly drawn 5:1 per used point from the final MCP. The mixed model was calculated in R (nloptr package, Johnson 2007; ln_bobyqa optimizer algorithm, Powell 2009; glmer package, Bates et al. 2015). Accuracy was measured through a five-fold cross validation, and both validation and mapping were performed with equal interval binning (Morris et al. 2015, 2016).

We used nine environmental variables (Table 2), those in the elk study (Morris et al. 2015) plus distance to streams and time-averaged NDVI. All variables were resampled to 30m to project the model onto the landscape (Morris et al. 2015). Bison have shown preferences for water (mixed), lower elevations, gentler slopes, different landcover types, and roads (Fischer and Gates 2005; Bruggeman et al. 2007; Fortin et al. 2009; Allred et al. 2011).

Table 2:

Environmental covariates used in the resource selection function (RSF) model used to predict bison habitat use. All covariates were converted to 30m raster layers to map RSF outputs. Coefficients and standard error for fixed effect variables in the final resource selection function (RSF) model developed to predict bison habitat use during the anthrax risk period in Southwestern Montana.

Variable	Description	Data Source	Coefficient Estimate	Std. Error
Fixed effects Intercept	from best model		−1.97	0.204
Aspect	Categorical – southerly (134°–224°)	Derived from Digital Elevation Model (DEM)	0.124	0.169
Elev	Elevation	DEM from United States Geological Survey (USGS) National Elevation Dataset (NED)	0.313	0.089
Forest	Categorical – forest vs. non-forest	2010 Montana Spatial Data Infrastructure (MSDI) land cover dataset	−3.357	0.569
NDVI	MODIS derived Normalized Difference Vegetation Index, Average from 1 Jun – 3 Aug 2012, 250 meters	Reverb|ECHO NASA	0.494	0.074
RD_PRIM	Distance to Primary Roads	Derived in (Morris et al. 2015)	0.103	0.067
RD_SEC	Distance to Secondary Roads	Derived in (Morris et al. 2015)	−0.744	0.07
RD_TERT	Distance to Tertiary Roads	Derived in (Morris et al. 2015)	−0.295	0.058
Slope	Slope	Derived from DEM	−0.791	0.085
Stream	Distance to Streams	Derived in (Morris et al. 2015)	0.075	0.062

We defined potential anthrax risk zones using a previously published ecological niche model (ENM) for this region (Morris et al. 2015). That study provided three cutoffs for risk. We chose the conservative definition, or most limited estimate, of B. anthracis presence, and clipped the estimate to ranch boundaries (Figure 1). Briefly, risk zones were defined from a presence-only genetic algorithm (GARP) using bison, elk, and white-tailed deer cases from the 2008 outbreak, two bison cases in 2010, and covariates associated with anthrax. The final model set had a total omission of 3.4% or was 96.6% accurate at predicting actual cases. Full details are available in Morris et al. (2015).

All nine environmental variables were significant for estimating bison resource selection (p ≤ 0.001, Table 2) and the final model was highly accurate with an average Spearman’s rank correlation coefficient from the cross fold validation of 0.90. The coefficients indicated that mature male bison on the ranch selected areas dominated by non-forest areas, gentler slopes, close proximity to secondary and tertiary roads, higher NDVI values, southerly aspect, distances farther from primary roads and streams, and higher elevations. The random effect for individual bison was significant, indicating variation between the collared males (intercepts of 0.36, 0.05, and −0.41, respectively).

To illustrated areas of most likely selection by male bison, we mapped the top five RSF bins (preferred habitat areas) onto the landscape. Preferred areas were in the central and northeast portions of the ranch (Figure 2A). The preferred areas overlapped with potential risk zones across 118 km², ~30% of the ranch. Overlap areas were spread across the ranch and not confined to specific pastures (Figure 2B). In an effort to prevent disease, bison grazing access has been restricted during the seasonal risk period, and bison remain entirely excluded from pastures exhibiting the highest mortality in 2008 (Morris et al. 2015). However, our results suggest are risk areas available to male bison despite exclusion efforts and this practice alone may not prevent future outbreaks.

Figure 2.

Predictions of male bison habitat preference (top 5 equal interval bins) from a resource selection function (RSF) model (A) and overlap of RSF-based outputs with potential seasonal anthrax risk zones (B).

We defined the anthrax risk season as June through August based on the 2008 and 2010 cases. This period overlaps bison rut, wherein sexually mature male bison are competing against each other (Dragon et al. 1999). Rut may alter the bison space preference and behaviors, e.g. wallowing (Dragon et al. 1999); stress from rut may increase infection risk and movement away from the herd after rut may make finding dead males difficult. While our sample size was limited to three individual males, model validation indicated high accuracy for predicting habitat selection during the seasonal anthrax risk period, despite variation in these three individuals. Future work should increase the number of collared bison. Overall, surveillance is advised, particularly in areas of high bison/anthrax risk overlap, including diagnostic testing for any bison deaths in summer months. Likewise, disease risk extends beyond current excluded pastures, indicating vaccination should continue. Nearby ranches should also perform testing for unusual livestock deaths in summer months and consider vaccination.