Factors affecting apparent prevalence of tuberculosis and brucellosis in wood bison

Summary

• 1
  Bison (Bison bison) abundance in Wood Buffalo National Park, Canada, declined from in excess of 10 000 bison in the late 1960s to a low of 2200 bison in the late 1990s.
• 2
  Bovine tuberculosis (Mycobacterium bovis) and brucellosis (Brucella abortus), were introduced to Wood Buffalo National Park in the late 1920s. As each of these pathogens have the potential to reduce survival and reproduction in bison, they are suspected to have played a role in the decline in bison abundance.
• 3
  We live-captured bison for disease testing in February and March of 1997, 1998 and 1999. Forty-nine percent tested positive for tuberculosis (i.e. were positive on the caudal fold test and/or fluorescent polarization assay, n = 342). Further, 30·9% of bison were seropositive for brucellosis (i.e. agglutinated in the buffered-plate antigen test and had a titre of 1 : 5 in the complement fixation test or had a titre of ≥ 1 : 10 in the complement fixation test, n = 346). Prevalence for both diseases increased with age and males were more likely to test positive for tuberculosis. Prevalence of either disease did not appear to be directly related to density of bison, as prevalence rates were not greater in the high density Delta population than the lower density Hay Camp and Nyarling River populations.
• 4
  Comparison of our results to previous brucellosis and tuberculosis surveys in Wood Buffalo National Park indicates that prevalence of neither pathogen is a direct function of bison density. These pathogens are endemic within the bison population of the park.

Introduction

Between 1925 and 1929, plains bison (Bison bison bison) were relocated from Wainwright Buffalo Park in central Alberta, Canada, to the newly created Wood Buffalo National Park (WBNP) to relieve overcrowding at Wainwright (the history of the park was reviewed by Carbyn, Oosenbrug & Anions 1993; Fuller 2002). Over 6000 bison were shipped although many were reported to have perished on route or were euthanized upon arrival (Carbyn et al. 1993; p. 27). The plains bison mixed with the 1500 indigenous wood bison (Bison bison athabascae), and by 1934 the population numbered approximately 12 000 bison. Bovine brucellosis (Brucella abortus) and tuberculosis (Mycobacterium bovis) were discovered in WBNP in the 1950s. The two pathogens were probably introduced with the bison from Wainwright National Park.

Bovine tuberculosis and brucellosis each have the potential to reduce survival and fecundity in bison (reviewed by Tessaro 1987, 1989), have significant economic consequences for the domestic cattle and bison ranching industries (e.g. Animal, Plant and Food Health Risk Assessment 1998), and are seen as the main impediment to further recovery of a once endangered species bison in northern Canada (Gates, Chowns & Reynolds 1994; RENEW 2001). Further, the presence of these pathogens may have played a significant role in the decline of the bison population of WBNP dropping from approximately 11 000 bison in 1970 to less than 2500 at the end of the 1990s (e.g. Joly 2001).

The prevalence of these pathogens in bison in WBNP has not been evaluated in a systematic manner. Previously, prevalence levels were determined during bison slaughter operations (Choquette et al. 1961, 1978; Fuller 1962) or on opportunistically sampled bison (Tessaro et al. 1990). These estimates of prevalence were questioned during public hearings on disease management in the late 1980s (e.g. Ferguson 1989); in particular whether these data were representative of the bison populations in WBNP (Ferguson & Burke 1994). Further, if transmission of either pathogen is density-dependent, prevalence may have declined in response to the sustained and substantial decline in numbers. This would be an important factor to consider in managing these pathogens, as if it could be shown that prevalence is density-dependent then partial culls of the population might be a viable management action for disease control. Alternatively, if prevalence is insensitive to bison density, more drastic management actions would be necessary for disease eradication. Dobson & Meagher (1996) proposed a nonlinear, density-dependent relationship between bison numbers and Brucella abortus prevalence. Rodwell, Whyte & Boyce (2001) found that tuberculosis prevalence in African buffalo (Syncerus caffer) did not covary with population size among regions in Kruger National Park, South Africa. It is not known if prevalence of either pathogen is a function of density or population size in bison.

Herein, we present the results of tuberculosis and brucellosis testing of bison in WBNP during February and March of 1997–1999. Our objective was to examine age- and sex-related trends in prevalence, as well as to determine if prevalence varied among three populations of various sizes within WBNP as an indication of a possible relationship between density and prevalence. Further, we discuss previous estimates of prevalence in an alternate examination of possible pathogen–density relationships.

Methods

Wood Buffalo National Park is approximately 44 000 km² in size, and straddles the border between the Province of Alberta and the North-west Territories (60° N, 112° W). Aspects of the WBNP environment relevant to bison ecology are described in detail in Carbyn et al. (1993).

Bison in WBNP are spatially distributed as a metapopulation, with five main populations (sensuWells & Richmond 1995): Garden River, Delta, Hay Camp, Nyarling River and Little Buffalo (Joly 2001). Bison were captured for testing in the Nyarling River (females only), Hay Camp and Delta populations. Bison densities at the start of the study varied dramatically among our three study areas: Nyarling River, 0·011 bison km⁻¹; Hay Camp, 0·124 bison km⁻¹; and Delta, 0·162 bison km⁻¹ (Joly 2001). Bison capture and handling protocols were described in detail in Joly (2001). Briefly, adult males too large to be slung under the helicopter were darted from the air using carfentanil and xylazine hydrochloride (Haigh & Gates 1995). Bulls were re-immobilized after 72 h to determine the results of the caudal fold test for tuberculosis. In February and March of 1997 and 1998, female and juvenile bison were captured with a net-gun fired from a Hughes 500D helicopter (Helicopter Wildlife Management, Salt Lake City, Utah). Bison were held in captivity for 72 h, after which the results of the caudal fold test were determined. Adult (> 3 years) and sub-adult (1–2 years) animals were kept in individual pens, while calves (< l year) were paired to reduce isolation stress. A long-acting neuroleptic, Clopixol-Acuphase, was administered to keep the bison calm during captivity (Ebedes 1992). Second-cut alfalfa hay or alfalfa cubes were provided for food, and fresh snow was provided for water. After 72 h in the corral, the bison were immobilized with carfentanil and xylazine hydrochloride in order to read the results of the caudal fold test for tuberculosis. In 1998, the xylazine was reversed using tolazoline. Naltrexone hydrochloride was used to reverse the carfentanil, and bison were released in a manner that encouraged individuals to leave as a group. In 1999, female bison were captured with a net gun fired from a Hughes 500D helicopter (Helicopter Wildlife Management and Helicopter Capture Services, Salt Lake City, Utah). Handling (in the same manner as described above) was done at the capture site and bison were released immediately after handling and radio-collaring. Bison were relocated in 3 days and recaptured by net gunning to determine the results of the caudal fold test.

Blood samples were taken from each bison from the caudal or carotid vein during the first handling period. Blood was collected in serum separator tubes to facilitate clotting, and prevented from freezing. Serum was removed by centrifuge within 12 h, and serum samples were frozen until serological tests were conducted.

The standard procedure for Brucella abortus testing in cattle is to use the buffered plate antigen test (BPAT) as a screening test, and the complement fixation test (CFT) as the confirmatory test for sera that agglutinate in the BPAT (Nielsen et al. 1996). Testing was done at the Animal Disease Research Institute, Lethbridge, Alberta (Animal and Plant Health, Canadian Food Inspection Agency). Sera that met either of the following criteria are referred to as B. abortus-positive (Joly, Leighton & Messier 1998): (a) agglutinated in the BPAT and agglutinated in the CFT at dilutions of ≥ 1 : 5 or (b) agglutinated in the CFT at dilutions of > 1 : 5 . We refer to the proportion of a given subset of bison (i.e. sex, age and population) that is B. abortus-positive as B. abortus-prevalence. This is equivalent to ‘apparent prevalence’ as defined by Greiner & Gardner (2000, p. 17): the probability that a randomly selected unit of analysis has a positive test result.

We tested for Mycobacterium bovis using the caudal fold test with (PPD) tuberculin (Thoen et al. 1988; Monaghan et al. 1994). Bison were injected with 0·1 mL of (PPD) tuberculin intradermally in the caudal fold. A veterinarian (T. Shury) accredited by the Canadian Food Inspection Agency conducted the tuberculosis test. He inspected the injection site after 72 h to determine pathogen exposure status (Monaghan et al. 1994). Bison were classified as positive, negative or suspicious reactors, based on the degree of reaction at the injection site. As recently infected animals or those in the latter stages of infection may not react to the injection of tuberculin (e.g. Griffin & Buchan 1994), we supplemented the caudal fold test with the fluorescent polarization assay for Mycobacterium bovis (Lin et al. 1996). Lin et al. (1996) did not provide interpretation criteria for the fluorescent polarization assay. To determine the diagnostic threshold for the fluorescent polarization assay, we examined the millipolarization units (mp) values for 85 bison sera from Elk Island National Park, Canada (Om Surujballi, Canadian Food Inspection Agency, Nepean, ON, unpublished data), a known tuberculosis-free herd. We rearranged the formula for comparison of a single observation against a sample distribution (equation 9·5 in Sokal & Rohlf 1995) to determine the maximum mp value that would be consistent with a distribution of negative samples (one-tailed alpha = 0·05). The mean mp for these bison was 166·5 (SE 0·48, n = 85), and less than 5% of a distribution of negative bison would have a mp value greater than 174 mp. We assumed that sera from bison with mp values greater than this threshold had a positive status. Bison that tested positive on either the caudal fold or the fluorescent polarization tests were considered positive and are referred to as M. bovis-positive. We refer to the proportion of a given subset of bison (i.e. sex, age and population) that is M. bovis-positive as M. bovis-prevalence. As above, this definition is synonymous with apparent prevalence as defined by Greiner & Gardner (2000).

The variables age, sex and population (Nyarling River, Hay Camp or Delta) were examined as possible factors associated with a positive status for each disease using logistic regression analysis (Sokal & Rohlf 1995; pp. 767–776). Model selection was based on small sample size corrected Akaike Information Criteria (AIC_c), and parameter estimates were averaged among all possible models weighted by Akaike weights (Burnham & Anderson 1998, pp. 119–140). Analysis was first conducted using age as a continuous variable; however, when preliminary analysis indicated nonlinearity in the relationship we grouped bison into the following age classes and entered age as a categorical variable in the model: < l, 1, 2, 3–5, 6–10 and 11–19. We found in a preliminary analysis an interaction between age and sex for both pathogens, therefore we elected to conduct analyses on each sex separately. We did not include in the analysis test results from the second capture if a bison was caught in more than one year.

Results

Overall, 30·9% (107/346) bison were B. abortus-positive. Sera that were anti-complementary (i.e. that reacted to the addition of complement in the absence of antigen) in the complement fixation test were excluded (n = 11). Inclusion of these sera drastically alters the sensitivity and specificity of this test, depending on whether they are treated as positive or negative (see Gall et al. 2000).

Initial analysis indicated that B. abortus-prevalence was a function of sex and age, where prevalence increased with age and females were more likely to test positive than males (male vs. female odds ratio, 0·70; 95% confidence interval (CI), 0·52–0·95). Sex-specific analyses indicated that age was a good predictor of B. abortus-prevalence in males (Table 1). Few males < 2 years of age were B. abortus-positive compared to females of the same age, yet after 2 years of age B. abortus-prevalence rose dramatically among males (Fig. 1). There appeared to be a bimodal relationship between age and B. abortus-prevalence in males; males 2–5 years of age had high B. abortus-prevalence, then B. abortus-prevalence declined in 6–10 year olds before increasing again in males > 11 years of age (Table 1; Fig. 1).

Table 1.  Relationship between age and Brucella abortus prevalence in males (n = 94). Odds ratios refer the increase in probability of testing positive for brucellosis relative to the youngest age class (e.g. male bison in the 3–5 year age class are 40·26 times more likely to test positive for brucellosis than male bison in the < 1-year age class)

Age class	Odds ratio	95% CI	Wald statistic*	P
*d.f. = 1.
< 2 years	−	Reference category	−	–
2 years	48·22	4·08–569·81	9·46	0·002
3–5 years	40·26	3·89–416·72	9·60	0·002
6–10 years	4·82	0·269–86·29	1·14	0·29
11–14 years	45·00	3·73–543·1	8·97	0·003

Figure 1. Apparent prevalence of Brucella abortus in male bison from the Delta (dark bar) and Hay Camp (shaded bar) bison populations of Wood Buffalo National Park, Canada (1997–1999). Sample size is indicated at the top of each bar.

B. abortus-prevalence in females increased with age in the Delta population, but this effect was not detectable in either the Hay Camp or the Nyarling River populations (Table 2). Female bison in the Hay Camp region were 1·5 times more likely to be B. abortus-positive than in the Delta after controlling for the effects of age (95% CI; 1·1–2·0; Fig. 2). As we did not test females < 2 years of age in the Nyarling River population, and only three 2-year olds were tested there, we repeated the analysis including only females > 2 years of age and in all three populations. After controlling for age, females in the Nyarling River were more likely to be B. abortus-positive than those in the Delta population (odds ratio, 1·7, 95% CI, 1·0–2·7) but not the Hay Camp population (odds ratio, 1·0; 95% CI, 0·6–1·6).

Table 2.  Relationship of age to Brucella abortus seroprevalence for females in the Delta and Hay Camp populations (n = 215). There was no effect of age on brucellosis seroprevalence for females 2 years and older in the Nyarling River population (Wald Statistic = 5·18, d.f. = 3, P = 0·15)

Age class	Odds ratio (95% CI)*
	Delta	Hay Camp
* Odds ratios refer to the increase in odds of a female bison in a particular age class testing positive for brucellosis relative to the youngest age class (< l year for Hay Camp and < 2 years for Delta). † Delta 1-year age class includes 12 seronegative females < 1 year of age as none in that age class tested positive.
< 1 year	−	Reference category
1 year†	Reference category	2·25 (0·32–15·76)
2 years	1·92 (0·24–15·26)	2·81 (0·50–15·77)
3–5 years	8·33 (1·23–56·67)	1·61 (0·26–10·14)
6–10 years	7·29 (1·31–40·53)	2·86 (0·47–17·35)
> 11 years	4·50 (0·88–22·95)	5·32 (0·94–30·0)

Figure 2. Apparent prevalence of Brucella abortus in female bison from the Delta (dark bar) and Hay Camp (clear bar), and Nyarling River (shaded bar) bison populations of Wood Buffalo National Park, Canada (1997–1999). Sample size is indicated at the top of each bar.

Forty-nine percent of bison tested in late winters 1997–1999 tested positive on either the caudal fold test or the fluorescent polarization assay (n = 342). Mycobacterium bovis-prevalence was a function of age, and it increased faster for males than females (model-averaged odds ratios (95% CI): sex, 1·16 (0·95–1·43); age, 1·25 (1·14–1·36); sex × age, 0·89 (0·81–0·97)). Sex-specific analyses indicated that age was the predominant factor associated with M. bovis-prevalence in males, with the probability of testing positive for tuberculosis increasing 1·26 times for each additional year of age (95% CI, 1·03–1·54; Table 3). There was some indication that M. bovis-prevalence may increase with age faster in the Hay Camp area, and that males in general may have a higher M . bovis-prevalence in the Hay Camp population, although neither odds ratio differed from one (Table 3). The most parsimonious model predicted that male M. bovis-prevalence increased to almost 70% by the age of 9 years (Fig. 3).

Table 3.  Comparison of models of Mycobacterium bovis prevalence in male bison in Wood Buffalo National Park (n = 97). The χ² and P-value refer to the likelihood ratio goodness of fit test. Relative AlC_c is presented as Δ_i, and the Akaike weight (ω_i) refers to the probability that the model is the Kullback–Liebler best model, given the data (see Burnharn & Anderson 1998)

Model*	d.f.	χ^2	P	Δi	ωi
* Model-averaged odds ratios (95% CI): age, 1·26 (1·03–1·54); population, 1·02 (0·98–1·06); age × population 1·01 (0·99–1·03).
Age	1	33·06	< 0·001	0·00	0·52
Age, age × population	2	33·24	< 0·001	1·95	0·20
Age, population	2	33·21	< 0·001	1·98	0·19
Age, population, age × population	3	33·87	< 0·001	3·49	0·09
Age × population	1	2·15	0·14	30·91	< 0·001
Population Population, age × population	1 2	0·91  2·15	0·34     0·34	32·15 33·04	< 0·001< 0·001

Figure 3. Apparent prevalence of Mycobacterium bovis in male bison from the Delta (dark bar) and Hay Camp (shaded bar) bison populations of Wood Buffalo National Park, Canada (1997–1999). Sample size is indicated at the top of each bar.

The most parsimonious model of M. bovis-prevalence in females indicated that prevalence varied as a function of age, population, and an interaction between age and population (Table 4). Mycobacterium bovis-prevalence in Hay Camp females was higher than in Nyarling River, which in turn was higher than in Delta (Table 4). Prevalence increased with age for females, with the probability of testing positive increasing 1·1 times with each year of age (95% CI, 1·00–1·2); however, this effect of age was not similar among the populations (Table 4). We repeated the analysis on separate populations to facilitate interpretation of these data, including age as a categorical variable (Fig. 4). M. bovis-prevalence in females rose with age in both Delta and Hay Camp populations (Hay Camp: Wald statistic = 14·3, d.f. = 5, P = 0·01; Delta: Wald statistic = 19·51, d.f. = 5, P = 0·002). Hay Camp population females in the 11–19 year age class had lower M. bovis-prevalence than those in the 6–10 year age class (Wald statistic = 7·26, d.f. = 1, P = 0·007); whereas the opposite was true for the Delta population (Fig. 4). We were unable to detect an effect of age on M. bovis-prevalence in the Nyarling River population for females 2 years of age or older (Fig. 4; Wald statistic = 3·79, d.f. = 3, P = 0·28).

Table 4.  Comparison of models of Mycobacterium bovis prevalence in female bison in Wood Buffalo National Park (n = 260). The χ² and P-value refer to the likelihood ratio goodness of fit test. Relative AIC_c is presented as Δ_i and the Akaike weight (ω_i) refers to the probability that the model is the Kullback–Liebler best model, given the data (see Burnham & Anderson 1998)

Model*	d.f.	χ^2	P	Δi	ωi
* Model-averaged odds ratios (95% CI): age, 1·09 (1·02–1·16); population: Delta vs. Nyarling River, 0·48 (0·23–0·97; i.e. Delta females were 0·48 times as likely to be positive than Nyarling River); Hay Camp vs. Nyarling River, 2·29 (1·18–4·45); age × Delta vs. Nyarling River 1·09 (1·00–1·19; i.e. probability of infection increased 1·09 times faster with age in the Delta than Nyarling River); age × Hay Camp vs. Nyarling River, 0·96 (0·88–1·04).
Age, population, age × population	3	31·46	< 0·001	0·00	0·72
Population, age × population	2	26·19	< 0·001	3·20	0·14
Age, population	2	23·92	< 0·001	3·47	0·13
Age	1	13·79	< 0·001	9·55	0·01
Age, age × population	2	16·14	0·001	11·25	0·002
Age × population	1	9·76	0·01	15·59	< 0·001
Population	1	7·45	0·02	17·90	< 0·001

Figure 4. Apparent prevalence of Mycobacterium bovis in female bison from the Delta (dark bar) and Hay Camp (clear bar), and Nyarling River (shaded bar) bison populations of Wood Buffalo National Park, Canada (1997–1999). Sample size is indicated at the top of each bar.

Discussion

disease testing

We determined the diagnostic threshold for tuberculosis in the fluorescent polarization assay by examining the distribution of test results for a tuberculosis-free bison population. Although we could evaluate the specificity of this test in diagnosing tuberculosis in bison as the sample of known-negative bison was used to derive the threshold, none of the known tuberculosis-free Elk Island National Park sera were greater than this threshold, nor were any of the control bison in Lin et al. (1996). This suggests the fluorescent polarization assay has a high specificity in bison, although we recognize the circular reasoning in this statement. The sensitivity of the fluorescent polarization assay in bison is unknown and will be assessed in future work (Om Surujballi, Canadian Food Inspection Agency, Nepean, ON, personal communication). We note that the diagnostic threshold based on serological methods should be based on knowledge of the distribution of results for both infected and non-infected animals. Future research addressing this shortcoming is necessary. There are no controlled studies evaluating the sensitivity or specificity of the caudal fold test with PPD tuberculin in bison, although Thoen et al. (1988) found that bison reacted similarly to the test as domestic cattle. Tessaro (1989) showed the caudal fold test using ‘old type’ (OT) tuberculin has a sensitivity of 66·7% and specificity of 89·6% in bison. The sensitivity and specificity of OT tuberculin and (PPI) tuberculin in domestic cattle were similar when compared at similar dosages (OT vs. PPI) tuberculin: sensitivity, 84% vs. 85%; specificity 99% vs. 98%; recalculated from Legg & Maunder 1940; Francis et al. 1978). The buffered plate antigen test for brucellosis has a reported specificity in bison of 91·7% and a sensitivity of 92·1%, and the complement fixation test has a specificity of 95·5% and sensitivity of 89·5% in bison (Gall et al. 2000). Despite these inherent test errors, we believe our data provide the best possible estimates of prevalence given the state of pathogen testing of bison, thus these results are comparable to other studies.

brucella abortus prevalence

Few males < 2 years of age were B. abortus-positive compared to females of the same age, yet after 2 years of age B. abortus-prevalence rose dramatically among males. The most likely route of transmission of brucellosis is contact with an infected foetus and/or foetal material (Cheville, McCullough & Paulson 1998), and so the rate of contact with parturient or aborting females is likely to be a determinant of prevalence. Young males (< 2 years) plains bison initiate few interactions with females older than 2 years of age (Rothstein & Griswold 1991), which may reduce their exposure to the pathogen. However, the most frequent visitors to a parturient female and newborn calf are 2-year old bulls who sniff and lick placental material and consequently are exposed to large numbers of bacteria (Rhyan 2000).

We found a decline in B. abortus-prevalence for males in the 6–10 year age class. Perhaps this pattern is the result of a decline in antibody titre to non-detectable levels in this age class after initial exposure as young–mature bull. We can offer some support for this decline in antibody as complement fixation titres for 40/45 bison that were B. abortus-positive at first-capture declined in the 1, 2 or 3-year interval between collar deployment and collar removal and the remainder stayed stable. The increase in B. abortus-prevalence in the 11–14 year age class may be related to the development of arthritic lesions associated with brucellosis (e.g. Tessaro, Forbes & Turcotte 1990). Specifically, we propose that bulls in this age class have been infected since first infection as a young bull; however, the antibody titres declined in intermediate ages until development of brucellar arthritis late in life. We qualify this speculation by noting the relatively small sample sizes involved.

mycobacterium bovis prevalence

Mycobacterium bovis-prevalence increased with age in both males and females in the Hay Camp and Delta populations. This is consistent with previous data for infected bison herds in and around WBNP (Fuller 1962; Novakowski 1965). The inability to detect an increase in prevalence with age in the Nyarling River sample is likely to be a consequence of smaller sample sizes and having sampled primarily adult females. We found that males may have had a higher prevalence than females, and that prevalence increased at a higher rate with age in males than females. We hypothesize that high prevalence in males is related to long periods of nose-to-nose contact in aggressive encounters during the rut. This result is important in regards to disease containment programs, as mature males with little access to females during the rut have larger home ranges, and are often associated with range expansion (e.g. Gates & Larter 1990; Larter & Gates 1994). These males had relatively high prevalence in this study (40–70%; Fig. 3). If males that wander come into contact with pathogen-naïve populations during periods of nomadism or emigration excursions, they are likely to transmit tuberculosis to males in these populations.

In the Hay Camp population, M. bovis-prevalence in females declined in females 11 years of age relative to younger age classes. Rodwell, Whyte & Boyce (2001) suggested a decline in prevalence of M. bovis in aged African buffalo relative to younger individuals due to a decline in relative survival in tuberculosis-positive buffalo in older age classes. We believe this interpretation also explains the similar results for the Hay Camp population. It is puzzling that we did not see the same pattern among Delta females. Carbyn et al. (1993, pp. 184–186) found that bison in this age class were disproportionately represented in wolf kills in the Hay Camp area (≥ 11 year-old bison formed 20·8% of wolf kills vs. 10·8% of the total population). This trend was not as striking in the Delta area (≥ 11 year-old bison formed 14·0% of wolf kills vs. 10·8% of the total population). If wolves in the Hay Camp area target aged bison, then it is logical to hypothesize that infected bison in this age class would be at greatest risk of predation, resulting in a decline in tuberculosis prevalence relative to younger age classes. An alternate explanation is that environmental conditions in the Delta area are such that general body condition of bison is higher (D. O. Joly and F. Messier, unpublished data), and so perhaps progression of the disease is slowed by the immune system.

density relationships

Due to differing testing methods, the present results are not directly comparable to previous pathogen surveys in Wood Buffalo National Park. However, Table 5 indicates that pathogen levels have not declined in the last 40 years for either pathogen despite a major decline in population size. The decline in bison abundance for the Delta population has been substantially greater in magnitude than elsewhere in WBNP (Carbyn et al. 1998). Therefore, if prevalence was related to bison abundance, we would expect that the difference in prevalence between the Delta and Hay Camp populations would change as well. We found that M. bovis-prevalence was greater for females in the Hay Camp population relative to the Delta population. Historical data from WBNP indicate that the Hay Camp population may have had a higher prevalence of M. bovis than the pre-decline Delta population: 38·4% of 1567 bison slaughtered in Hay Camp area from 1950 to 1966 had tuberculous lesions compared to only 28% of 722 bison slaughtered in the Delta population from 1951 to 1970 (χ² = 22·2, d.f. = 1, P < 0·001; data from Carbyn et al. 1993, p. 30). Note that these inferences require the assumption that the historical data are representative of the bison population at the time, and the age structures of the samples were similar between the Hay Camp and Delta populations. Mycobacterium bovis prevalence in the Hay Camp population was 8% higher than the Delta population in 1997–1999, which is comparable to the 10% difference seen in 1950–1970. This stability in relative prevalence suggests that transmission of tuberculosis may be a nonlinear function of density, where a decline in transmission does not occur until very low densities. This could be a consequence of the gregarious nature of bison, and may also explain the lack of a relationship between density and tuberculosis prevalence in African buffalo (Rodwell et al. 2001). More evidence for a nonlinear density-transmission relationship is found in the fact that Mycobacterium bovis prevalence did not differ among young females in the Delta and Nyarling River populations (< 6 years; Fig. 4), despite the Delta population being on average 5 times larger during recent years. As the number of bison counted in the latter population has not exceeded 236 bison since 1990, we suggest that tuberculosis will persist at very low densities.

Table 5.  Apparent prevalence of Brucella abortus and Mycobacterium bovis from various surveys in the Greater Wood Buffalo National Park Ecosystem. Sample size is indicated in parentheses, and population size data are from Joly (2001)

Year(s)	n	Mycobacterium bovis (%)	Brucella abortus (%)	References
1952–1956	> 10 000	40	–	Fuller (1962)
1959	> 10 000	13·5	–	Choquette et al. (1961)
1959–1974	> 10 000	–	31	Choquette et al. (1978)
1983–1985	4 662	21	25	Tessaro et al. (1990)
1997–1999	2 172	49	31	This study

Dobson & Meagher (1996) proposed a nonlinear relationship between B. abortus-prevalence and density. They suggested that a minimum population size of 200 was required to sustain B. abortus in bison populations; however, on visual examination the figure they present to support this number does not include any populations less than 200. The degree of transmission of B. abortus is likely a function of group size during the third trimester of gestation, as this is the period in which most brucellosis-induced abortions occur (Cheville et al. 1998). We determined the typical group size (Jarman 1982) of bison during bison total count surveys conducted by WBNP staff in February and March during each year from 1981 to 1999 (excluding 1986 and 1993) for each of the Hay Camp, Delta and Nyarling River populations (Joly 2001), and used an analysis of covariance (Sokal & Rohlf 1995, pp. 499–521) to test the effect of population numbers and population on typical group size (Fig. 5; note that data from Nyarling River were very variable pre-1990 and so these data were not included). Typical group size was related to numbers of bison (F_1,39 = 60·2, P < 0·001), but this relationship did not vary among populations (F_2,39 = 1·34, P = 0·27). Based on this relationship, we predict that typical group sizes in the Delta population during 1950–1970 was in excess of 100 bison, 3–5 times that seen during the 1997–1999 surveys. Forty-four percent of bison tested in the Delta population from 1950 to 1970 were B. abortus-positive, significantly more than the 19%B. abortus-prevalence in the Hay Camp population over the same period (χ² = 120·78, d.f. = 1, P < 0·001; data from Carbyn et al. 1993; p. 30). In contrast, our data show that B. abortus-prevalence in the Hay Camp population is presently greater than in the Delta population (1997–1999, Fig. 4). Thus, a decline in B. abortus-prevalence may be associated with a decline in typical group size. However, it is important to note that these results may be a function of different age and sex compositions in the 1950–1970 samples. Further, the number of bison counted each year in the Nyarling River population since 1990 ranges from 112 to 236 bison (Joly 2001), not biologically different from the minimum threshold for B. abortus persistence of 200 bison suggested by Dobson & Meagher (1996) and prevalence was not lower there than in the higher density Delta population. Movement of B. abortus-positive bison from the Hay Camp and Delta to the Nyarling River population is likely to be too low (Joly 2001) to account for B. abortus-prevalence in the latter population, and so we conclude that brucellosis is self-sustaining in this population. Clearly, the minimum number of bison required for brucellosis to persist is very low, and is likely to be significantly lower than that proposed by Dobson & Meagher (1996). We conclude that despite some evidence for density dependence in prevalence, brucellosis will persist indefinitely in the Wood Buffalo National Park bison metapopulation in the absence of management intervention.

Figure 5. Typical bison group sizes seen on late winter surveys in the Delta, Hay Camp and Nyarling River bison populations of Wood Buffalo National Park, Canada (1981–1999).

Acknowledgements

This research was funded primarily through the Bison Research and Containment Program of Heritage Canada, under the advice of the Research Advisory Committee of the Bison Research and Containment Program. The Northern Scientific Training Program, the Government of the Northwest Territories and the University of Saskatchewan provided additional funding. D.O.J. was supported through a University of Saskatchewan Graduate Scholarship and a Natural Sciences and Engineering Research Council postgraduate scholarship. We gratefully thank the Canadian Food Inspection Agency (Lethbridge, AB, and Nepean, ON) for conducting disease testing. Thanks to Tim Evans, Trevor Evans, Michelle MacDonald and, particularly, Amanda Plante for technical assistance. Residents of Fort Chipewyan and Fort Smith provided support throughout the study, and the crews of Helicopter Wildlife Management, Helicopter Capture Services, and Big River Air risked their lives so that we may conduct this study.