DESCRIPTIVE EPIDEMIOLOGY OF DETECTED ANTHRAX OUTBREAKS IN WILD WOOD BISON (BISON BISON ATHABASCAE) IN NORTHERN CANADA, 1962–2008

Anthrax, a notifiable disease per the World Organisation for Animal Health (OIE), is caused by the gram-positive, spore-forming bacterium Bacillus anthracis (Turnbull 2008). Anthrax outbreaks in ruminants are characterized by death within hours to days postexposure; sudden death in herbivores has been identified as one of the first signs of an anthrax outbreak (Gates et al. 2001a; Turnbull 2008). Periodic outbreaks of anthrax have been occurring in northern Canada since 1962 and are the subject of ongoing management and control efforts. Agencies responsible for anthrax risk management continue to seek new insights into more-effective detection and response to outbreaks.

Wild wood bison (Bison bison athabascae) in northern Canada are listed as a threatened species by the Committee on the Status of Endangered Wildlife in Canada (COSEWIC); they are also listed in Appendix II of the Convention on International Trade in Endangered Species (CITES) (Reynolds et al. 2003). In September 2001, the National Wood Bison Recovery Team drafted a National Recovery Plan for the Wood Bison (Gates et al. 2001b). The authors identified information gaps that were deemed pertinent to recovery and included the need to “evaluate the epizootiology of anthrax with a view towards its control in free-ranging bison.”

Since the 1960s, targeted surveillance for anthrax outbreaks has been employed in varying degrees by wildlife managers in Wood Buffalo National Park (WBNP) and the Slave River Lowlands (SRL). Responsibility for anthrax surveillance and outbreak response varies by jurisdiction: Parks Canada (PC) is responsible for WBNP and the Government of the Northwest Territories (GNWT) is responsible for the SRL outside of the Park boundaries. Anthrax Emergency Response Plans (AERP) have been developed by both agencies to manage the potential public health and conservation implications of anthrax in wild wood bison (Moreland 1999; Elkin et al. 2006). Both the GNWT and PC share the goal of identifying the first, and all subsequent, anthrax carcasses during an outbreak. To meet this goal, managers identified a need to assess their ability to detect and respond to carcasses and to anticipate outbreaks.

The cost of anthrax surveillance can be high; in the last decade, the GNWT and PC have spent CDN $15,000–$26,000 each year on aerial surveillance looking for anthrax carcasses. In 2006, the GNWT spent half a million dollars on anthrax carcass surveillance, identification, and the disposal of 28 carcasses.

Descriptive epidemiologic studies of animal anthrax have provided valuable insights into the effects of outbreaks on populations including sex distributions, lengths and locations of outbreaks, and percent mortality (Kellogg et al. 1970; Gainer 1987; Prins and Weyerhaeuser 1987; Gainer and Saunders 1989; Turner et al. 1999; Clegg et al. 2007; Mongoh et al. 2008a; Durrheim et al. 2009). Descriptions of outbreaks in WBNP and the SRL have been separately documented in the literature (Moynihan 1963; Novakowski et al. 1963; Pyper and Willoughby 1964; Cousineau and McClenaghan 1965; Choquette et al. 1972; Broughton 1992; Dragon and Elkin 2001; Nishi et al. 2007) but have not been reviewed together to seek commonalties and differences between outbreaks. Gates et al. (1995) provided the most comprehensive outbreak account in wood bison, but it focused on only one outbreak in the Mackenzie herd. We inventoried and assessed the data available for use in retrospective, descriptive, epidemiologic analysis of anthrax outbreaks in wild wood bison. We provide insight on how these data may be used to inform anthrax surveillance and response as well as recommendations for further disease research and wildlife management.

Data from free-ranging wood bison populations in WBNP and the SRL were included in this study; data from the Mackenzie herd were excluded due to a low number of outbreaks and the geographic separation from WBNP and the SRL. The WBNP and SRL wood bison populations were located in defined geographic regions within northern Alberta and southern Northwest Territories (Fig. 1). Wood Buffalo National Park covers 44,807 km² of northern Alberta and the southern Northwest Territories (PC 2010), and the SRL is entirely within the Northwest Territories.

For the purposes of this analysis, the designation of WBNP included all of the herds within the boundaries of the WBNP, while the SRL included bison inhabiting the Hook Lake and Grand Detour regions. The herds or subpopulations were located approximately in either WBNP (Nyarling, Pine Lake/Hay Camp, Garden River, Delta) or the SRL (Hook Lake), with the exception of the Little Buffalo River herd in the northeast corner of the WBNP which also formed part of the bison groups found in the Grand Detour region of the SRL. For the purposes of tabulating carcass data, Nyarling, Little Buffalo, and Hay Camp were designated as Park Central, and Garden River and the Delta were designated as Park South. All wood bison in this study were part of herds in which bovine tuberculosis (Mycobacterium bovis) and bovine brucellosis (Brucella abortus) are endemic (Tessaro et al. 1990; Joly 2001).

Data sources included text books, scientific papers, grey literature (including internal memos and reports of the management agencies), formal and informal meeting minutes and presentations, and raw data in the form of typed or handwritten reports, tables or spreadsheets, and global positioning system (GPS) points and maps. We used PubMED and Google Scholar to search the scientific literature. The keywords “anthrax and wood bison” and “B. anthracis and wood bison” were used and relevant papers were identified. Those references regarding anthrax and wood bison, cited in these papers but which did not show up on the PubMED and Google scholar searches, were also used. Internal reports and raw data were generated and compiled by employees of PC, the Canadian Wildlife Service, Health of Animals Branch of Agriculture Canada (now the Canadian Food Inspection Agency), and the GNWT. While most of the raw data were recorded at the time of collection, reports were generated weeks to months later. Paper copies of internal reports were found in the PC library in WBNP and GNWT offices in Fort Smith. Where possible, the consistency of raw data and published data was compared. In the instances where data inconsistencies existed, the raw data (spreadsheet, internal report, handwritten note of direct observation, etc.) were used preferentially.

Each AERP calls for periodic surveillance flights from June to August–September, typically by fixed-wing aircraft, over locations of past outbreaks and areas the bison were known to inhabit during the surveillance period. Blood or tissue samples/swabs collected from bison carcasses found during that time were sent to a federal Canadian Food Inspection Agency laboratory for B. anthracis culture confirmation (Moreland 1999; Elkin 2006). Records indicated that dead bison were found at other times of the year on flights conducted for other purposes; suspected causes of death were often recorded for those cases.

An outbreak was defined as the detection of at least one dead animal positive for anthrax, as confirmed by culture. Records indicated that once an outbreak was confirmed within an area, typically a herd home range, subsequent dead bison found within the area of the outbreak were counted toward the total number of anthrax carcasses reported at the end of the outbreak. Not all carcasses were tested. The majority of carcasses included in the final mortality number were found during the summer. In 1963, 1967, 1978, and 2001, carcasses were found in September and October, after the leaves had fallen but before snow accumulation. These carcasses were recorded by outbreak personnel as anthrax deaths based on estimated time of death and proximity to summer outbreak sites. If there was an outbreak in the SRL and a carcass was found in WBNP, the WBNP carcass was tested and not just assumed to be anthrax and vice versa. Carcasses found in different regions of the same jurisdiction (e.g., Garden River and Delta regions of WBNP) were initially tested to confirm an outbreak in that region.

Start dates for outbreaks were determined by information reported in field notes such as estimated time since death at first discovery (e.g., fresh, dead 3 days). If no such notes were present, the start day was assigned to the first day that a carcass was found. The date of the end of an outbreak was set by the day the last fresh carcass was found. Date of death was estimated by personnel based on carcass decomposition at the time of identification, presence of maggots, and time since last surveillance when the location was confidently surveyed.

In WBNP, bison segregation counts were performed annually by PC staff from 1984–2008 (excluding 1994, 1998, 1999, 2000, and 2003) according to methods outlined by Bradley and Zimmer (2004). These counts provided the numbers of calves, yearlings, and total bulls per 100 cows. Prior to 1992, all of the segregation counts were done in the Delta region of the Park. While these population data did not include all possible years, the overall sex distribution appeared to be representative of wood bison in the area based on other observations (Carbyn et al. 1993; Gates et al. 1995). The numbers of animals reported for each year reflected the combined numbers calculated for each of the five subpopulations in the Park: Little Buffalo, Nyarling, Pine Lake/Hay Camp, Garden River, and Delta. For reasons not recorded, not all of the subpopulations were represented each year. Population numbers for the SRL were not available for all years 1962–2008.

Statistical analyses were performed on data from the years between 1962 and 2008 using Stata 10 (StataCorp LP, College Station, Texas, USA). A test for random order was used to test the hypothesis that outbreaks were mutually independent. Given the small number of outbreaks and nonparametric distribution of carcass numbers, a Mann-Whitney U-test was used to test for the effect of outbreak location and month on total mortalities. A Mann-Whitney U-test was also used to test the association between the number of carcasses detected in an outbreak, the month the outbreak started and the length of the outbreak, and location of the outbreak. When testing the associations involving lengths of outbreaks, the outbreak in WBNP in 1968 and the outbreaks in the SRL in 1978 and 2001 were omitted due to a lack of information regarding the start dates. A Spearman's rank correlation was used to investigate the statistical dependence between outbreak length and numbers of carcasses found as well as the relationship between anthrax deaths and population numbers. Odds ratios were used to compare bulls' and cows' probability of dying in an anthrax outbreak compared to their sex distribution in a population.

Anthrax carcass sex was determined for the following years: WBNP in 1978, 1991, 2000, 2001, 2007 and SRL in 2001, 2006. In some of these outbreaks (1991, 2000, 2007), not all of the carcasses were identified by sex, usually because they were in a relatively inaccessible location (e.g., a bog). We assumed there would be little or no differential bias with regard to unidentified carcasses being more of one sex or the other; therefore, the percentages of each sex identified were calculated using the number of identified carcasses, not the total number of carcasses in the outbreak.

Between 1962 and 2008, there were 12 detected outbreak years in wild wood bison in WBNP and SRL. There was one outbreak in the SRL in 2010. A test for random order failed to reject the null hypothesis that their distribution over time was serially independent (P = 0.06; Table 1). Seven anthrax outbreaks were recorded in the SLR and eight in WBNP between 1962 and 2008. Between 1962 and 2008, there was no association between population numbers and anthrax deaths in WBNP (Spearman's rho = 0.1708, P = 0.686) while there was a positive relationship in the SRL (Spearman's rho = 0.9487, P = 0.014). No more than 10% of the population died during any outbreak year (Fig. 2).

Between 1962 and 2008, 1,515 deaths attributed to anthrax were detected between June and October. Although targeted surveillance for anthrax only took place during summer months, bison deaths were detected at other times of the year by personnel on flights for other reasons. Anthrax was not reported as the cause of death in any of these cases. Death by predation was often recorded for individual animals and drowning by falling through thin ice was recorded as the cause of death for large groups of bison.

The average number of carcasses found each outbreak year was 126 (median 94; min 1, max 363). The 2010 SRL outbreak resulted in 47 carcasses being detected. The greatest number of deaths in one outbreak occurred in the SRL (1964, n = 303), but subsequent outbreaks in that region were either lower (Hook Lake) or no anthrax deaths were detected for 30 yr after (Grand Detour). The total number of deaths attributed to anthrax was greatest in the first 3 yr of outbreaks in SRL and have varied by location (Fig. 3). There was not a statistically significant difference between the number of carcasses found and the location of their detection (P = 0.13), but the average carcass numbers varied widely: Hook Lake average – 61 (median 28; min 12, max 281), Grand Detour average – 176 (median 242; min 27, max 259), Park Central average – 48 (median 48; min 32, max 63), Park South average – 65 (median 91; min 1, max 120).

Outbreaks were only detected in June, July, and August. The average length of an outbreak was 25.5 days (median 22.5; min 16, max 44; Fig. 4). The length of an outbreak and the numbers of carcasses found were independent (rho = 0.16, P = 0.63). The length of an outbreak was not associated with the location at which it started (P = 0.17) or the month in which it started (P = 0.81). The numbers of carcasses were associated with the month an outbreak started (P = 0.03). Outbreaks with July start dates had an average of 180 (median 120; min 32, max 363) carcasses found compared with a June average of 43 (median 33; min 15, max 91).

For two outbreaks, the dates of death of individual animals were recorded with sufficient detail and number to allow analysis: these were the 2006 and 2010 outbreaks in SRL – Hook Lake. In 2006, the majority of the recorded deaths occurred in a 2-day window near the start of the outbreak (Fig. 5). In 2010, deaths were more normally distributed (Fig. 6).

Using available population estimates for WBNP, population sex ratios varied widely from year to year. On average, bulls (subadult and mature) made up approximately 22% of the population (median 23%; min 4%, max 38%), cows approximately 54% (median 53%; min 39%, max 74%), with the rest of the population being made up of varying numbers of calves and yearlings. In comparison, bulls made up almost 80% of anthrax outbreak deaths while cows accounted for less than 20% (Fig. 7). The odds of a bull being the victim of an anthrax death was 10 times greater than that of a cow compared to the general population ratios (95% CI, 5–21). The bars in Figure 7 indicate that the number of bulls have ranged from just over 50% to 100% of detected anthrax carcasses. Even in the example of WBNP 2001, where anthrax deaths were split more evenly between the sexes, this mortality was skewed in favor of males when compared to the average sex distribution in the population. In the SRL 2010 outbreak, officials counted the deaths of 30 adult males, nine adult females, one subadult male, one subadult female, three yearling females, and three of unknown sex and age.

Our analysis identified four main trends: bulls were at higher risk for death from anthrax, the numbers of animals found dead per outbreak have declined over the past four decades, outbreaks varied in duration and magnitude in a manner unrelated to population size, and the length of an outbreak was not a determinant of the number of dead bison found. Each finding reinforces anecdotal and previously published observations and beliefs about B. anthracis infection patterns and dynamics in northern wood bison. Based on our selected threshold for statistical significance of P = 0.05, we could not reject the hypothesis that outbreaks were serially independent. We believe that it is reasonable to reject the hypothesis that the genesis of an anthrax outbreak is a random process, given the lack of knowledge regarding the contribution of component causes, including weather, to the genesis of an anthrax outbreak (Salb 2011) and because our calculated P-value of 0.06 may have been influenced by our sample size.

Variation in recorded outbreak lengths might be attributed to a number of factors. First, this variation may have reflected differences in the ability of surveillance efforts to find all carcasses as well as errors in estimating the time since death for detected carcasses. Second, it may have reflected a possible misclassification bias that arose when animals were assumed to have died of anthrax, without confirmatory testing, if found spatially or temporally related to another confirmed anthrax death. Third, it may have reflected differences in exposure. Primary exposure to anthrax spores could be hypothesized to have occurred in several ways: 1) spores were in one location and all animals were exposed at one time; then either the animals moved on and were subsequently not exposed further, or an event (rain, etc.) removed the spores from that location; 2) animals were exposed to spores in multiple areas at one time; or 3) spores were in a single location that persists for a prolonged period and animals were continuously exposed by visiting that area (Lindeque and Turnbull 1994; Hugh-Jones and de Vos 2002). Finally, environmental factors and animal behavior could combine to affect outbreak dynamics. Grazing patterns may have differed between the sexes and at different times of the year (Larter and Gates 1991; Komers et al. 1993). Sedge and reed grasses may establish and become locally extirpated within 3 yr, while the presence and absence of water in low-lying areas may be influenced by winter precipitation and spring flooding (Carbyn et al. 1993). This variation complicates decisions on how long to maintain aerial surveillance during an outbreak. Based on averages, at least 3 wk of surveillance should be conducted, but it may need to be longer because some outbreaks lasted more than 6 wk.

Bison deaths clustering at the beginning of an outbreak (Fig. 4) may suggest a point source for the outbreak, then either a removal of this source or a lack of further contact by bison with this source. A more normal distribution of bison deaths during the course of an outbreak (Fig. 5) may suggest a different pattern of exposure to anthrax spores. Pinpointing bison movements before, during, and after an outbreak may help identify sources of spores spatially, how many animals exposed to the source contract the disease, and how long a source(s) continues to be a problem.

From outbreak data analyzed 1962–2008, there was no association observed between the length of an outbreak and the number of mortalities in northern wood bison. However, larger numbers of dead bison were found in July than in June. Both of these observations were also supported by the 2010 SRL outbreak. Based on a review of historical surveillance documents (Salb 2011), there was no known bias between the quality or quantity of surveillance in June versus July; however, it remains unclear how seasonal effects (e.g., locations of bison, congregation of animals within flight range, or other factors) or detection bias (e.g., bulls being easier to see from the air due to their size) may affect carcass detection. Bison behavior may have differed from June to July, thereby increasing exposure to spores; however, given our lack of knowledge regarding how bison acquire infections, it is not possible to evaluate this hypothesis.

Detected anthrax deaths never comprised more than 10% of the local population, which was similar to the outbreak in the Mackenzie herd where 8% of the population died (Gates et al. 1995). Wildlife losses to anthrax have been most frequently documented in Africa and North America, most often reporting total losses, but infrequently associating the number of deaths with population numbers (Moynihan 1963; Dragon et al. 1999; Nishi et al. 2002; Clegg et al. 2007; Mongoh et al. 2008b). In an anthrax epidemic in white-tailed deer (Odocoileus virginianus), 60–90% of the herd was estimated to have died (Kellogg et al. 1970). An outbreak in impala (Aepyceros melampus) in a park in Tanzania in 1984 lasted almost a year and killed approximately 90% of the population (Prins and Weyerhaeuser 1987). Susceptible species in other locations have also suffered mortalities >90% following outbreaks (Clegg et al. 2007).

The sex bias of anthrax mortalities toward males is not unique to wood bison in our study area. Gates et al. (1995) observed a similar trend in the 1993 outbreak in wood bison in the Mackenzie herd. Pienaar (1967) noted that, typically, the majority of South African species that died from anthrax were male, as did Lindeque and Turnbull (1994) and Clegg et al. (2007). Lindeque and Turnbull (1994) did not note any striking difference in behavior, distribution, condition, or diet between male and female springbok (Antidorcas marsupialis), blue wildebeest (Connochaetes taurinus), and elephant (Loxodonta africana) populations during the peak anthrax outbreak periods in Etosha National Park. There is no known bias in surveillance that would facilitate finding one sex or another apart from the larger size of bulls. There was no appreciable difference between sex distributions in outbreaks where sex of all carcasses was determined versus only some carcasses.

The low number of outcomes reduced statistical power; confidence in associations identified in this study will be strengthened in the future with continued research and analysis. Variables that may contribute to the genesis, propagation, length, and number of deaths may be many but are scantily elucidated in the literature (Kellogg et al. 1970; Gainer 1987; Prins and Weyerhaeuser 1987; Turner et al. 1999; Clegg et al. 2007; Mongoh et al. 2008a; Durrheim et al. 2009). In this study, variables such as location (WBNP vs. SRL) and month an outbreak started are proxy variables and were chosen because they could be identified and measured. Factors represented by location that may contribute to outbreak epidemiology may include differences in local habitat, jurisdictional response to previous outbreaks, animal movement patterns, herd immunity, and previous numbers of undetected outbreaks. Month could also be a proxy variable with factors contributing to outbreaks being weather, height of food plants, numbers of biting insects, animal movement patterns, and body condition (Carbyn et al. 1993; Komers et al. 1993; Turner et al. 1999; Turnbull, 2008; Blackburn and Hugh-Jones 2009; Salb 2011). Future research should include identification of all possible contributors to outbreak genesis and the consistent data collection of these variables for systematic analysis.

This summation of past outbreaks generated several recommendations for both management and future anthrax outbreak research. Surveillance for early detection of anthrax in the NWT must be capable of detecting solitary, isolated bison and of tracking herds across the landscape. A more detailed spatial analysis directed at identifying key geographic areas to target surveillance, and the generation of more hypotheses on environmental determinants of risk, would allow better management and control of anthrax in wood bison. This would require detailed data on landscape features and careful consideration of the appropriate scale for analysis. Placement of GPS collars on bison would also elucidate previous locations of bison prior to outbreaks, potentially narrowing down a source location for exposure to spores. Testing all carcasses would remove any bias regarding numbers of anthrax cases. More-detailed and consistently recorded outbreak data to describe the epidemiology of the disease in bison will allow future exploration of existing and new hypotheses regarding disease genesis.

This project was funded by the Centre for Coastal Health, National Science and Engineering Research Council, and the Government of the Northwest Territories. We gratefully acknowledge everyone in PC Wood Buffalo National Park, the GNWT, and the Canadian Cooperative Wildlife Health Centre for their help in gathering the data for this analysis. We especially appreciate the assistance of Adrian D'hont, GNWT.