Environmental Factors Influencing White-Tailed Deer (Odocoileus virginianus) Exposure to Livestock Pathogens in Wisconsin

White-tailed deer (Odocoileus virginianus) are commonly exposed to disease agents that affect livestock but environmental factors that predispose deer to exposure are unknown for many pathogens. We trapped deer during winter months on two study areas (Northern Forest and Eastern Farmland) in Wisconsin from 2010 to 2013. Deer were tested for exposure to six serovars of Leptospira interrogans (grippotyphosa, icterohaemorrhagiae, canicola, bratislava, pomona, and hardjo), bovine viral diarrhea virus (BVDV-1 and BVDV-2), infectious bovine rhinotracheitis virus (IBR), and parainfluenza 3 virus (PI3). We used logistic regression to model potential intrinsic (e.g., age, sex) and extrinsic (e.g., land type, study site, year, exposure to multiple pathogens) variables we considered biologically meaningful to exposure of deer to livestock pathogens. Deer sampled in 2010–2011 did not demonstrate exposure to BVDV, so we did not test for BVDV in subsequent years. Deer had evidence of exposure to PI3 (24.7%), IBR (7.9%), Leptospira interrogans serovar pomona (11.7%), L. i. bratislava (1.0%), L. i. grippotyphosa (2.5%) and L. i. hardjo (0.3%). Deer did not demonstrate exposure to L. interrogans serovars canicola and icterohaemorrhagiae. For PI3, we found that capture site and year influenced exposure. Fawns (n = 119) were not exposed to L. i. pomona, but land type was an important predictor of exposure to L. i. pomona for older deer. Our results serve as baseline exposure levels of Wisconsin white-tailed deer to livestock pathogens, and helped to identify important factors that explain deer exposure to livestock pathogens.


Introduction
rhinotracheitis. The impact of infection is more significant when coupled with secondary bacterial pneumonia [18,19]. Parainfluenza 3 virus is transmitted through aerosolization and contact with nasal fluids [20]. Little is known about the disease in deer, but it is thought to spread through contact with domesticated cattle. Previous serologic surveys have shown a high seroprevalence against PI3 in many wild ruminant populations, including white-tailed deer [6,19,21,22].
The primary objectives of this study were to 1) determine if Wisconsin white-tailed deer are currently exposed to livestock pathogens and 2) identify environmental variables that influence exposure of white-tailed deer to livestock pathogens in Wisconsin. To our knowledge, serologic surveys for exposure of deer to livestock pathogens have not been conducted in Wisconsin since the 1970's. Additionally, we aimed to identify if deer were commonly exposed to multiple livestock pathogens. Given similar environmental variables in study areas and behavioral characteristics of deer, we expected that some deer would show evidence of exposure to numerous livestock-associated pathogens. To our knowledge, our study is the first to include exposure to pathogens as explanatory variables in models that explain exposure of white-tailed deer to other livestock pathogens. Lastly, we identified environmental characteristics associated with deer exposed to livestock pathogens because how deer contribute to the transmission cycle of many livestock pathogens is largely unknown [5,14].

Materials and Methods
This research was conducted in strict accordance to the 2011 Guidelines of the American Society of Mammalogists for the use of Wild Mammals in Research. The Institutional Animal Care and Use Committee, Research Animal Resources Center, at the University of Wisconsin-Madison (Protocol Number: A01446-0-08-10) specifically approved this study, including all sampling methods. White-tailed deer that were mortally injured during capture were euthanized by gunshot to the head (approved in 2013 American Veterinary Medical Association euthanasia guidelines).
White-tailed deer were trapped concurrently in two study areas in Wisconsin selected to represent regions where habitat type and deer densities differed (Fig 1). We trapped deer on private land, and the owner of each parcel granted us permission to trap. The Wisconsin Department of Natural Resources granted permission to trap deer on state lands and the United States Forest Service granted permission for trapping on the Chequamegon-Nicolet National Forest. Deer trapping on Wisconsin County Forest lands was approved by the Wisconsin County Forest Association. Deer trapping on county land was approved by Rusk County and Sawyer County. We did not trap on tribal lands and endangered species were not captured.
The Northern Forest study area (5,905 km 2 ) included Sawyer and Rusk counties, and 29% was publicly owned (e.g., county, state, or federal) and 5% was under tribal ownership. An additional 10% of the study area was private land enrolled in Open Managed Forest Law, so the public had access for hunting. Much of the landscape (78%) consisted of forested or shrubland habitat, including northern hardwoods (Acer spp., Betula spp., Tilia americana, Quercus spp., Fagus americana, Tsuga canadensis, Pinus spp., Abies balsamia, Populus spp., Picea glauca), tamarack-black spruce (Larix laricina, Picea mariana) bogs, northern white cedar (Thuja occidentalis) swamps, and alder (Alnus spp.) swamps. Primary land use in the area was forestry, and topography was moderately rolling hills [23]. Soils included stony glacial till, pitted outwash sands, and peat. Deer habitat quality has declined since the 1930s with the maturation of forests. Deer population goals in the Northern Forest currently average approximately 70% of carrying capacity [23].
The Eastern Farmland study area (6,003 km 2 ) included Shawano, Waupaca, and Outagamie counties, and 2% of the land area was publicly owned, and 0.7% was private land enrolled in Open Managed Forest Law. Thirty-five percent of the landscape was forested, including floodplain forests and lowland hardwood swamps (Acer spp., Quercus spp., Ulmus rubra, Fraxinus spp. Tilia americana, Salix spp.), and cattail (Typha spp.) or sedge (Carex spp.) wetlands. Primary land use throughout the area was row crop agriculture, and topography was dominated by gently rolling hills [23]. Deer population goals in the Eastern Farmland have increased  substantially since the early 1960s and current goals average 20-40% of estimated carrying capacity [23].
We collected blood samples from deer via jugular venipuncture and separated sera from cells in the field via centrifugation [5]. For recaptured deer, we took blood only upon their initial capture. Sera were sent to the Minnesota Veterinary Diagnostic Laboratory and were tested for antibody against six serovars (grippotyphosa, icterohaemorrhagiae, canicola, bratislava, pomona, and hardjo) of Leptospira interrogans using microscopic agglutination, against BVDV and IBR using serum neutralization, and against PI3 using hemagglutination inhibition [7,11,30]. These tests use either bacterial antigen or respective virus which will react with antibodies if present. The initial dilution of sera for serum neutralization tests for IBR and BVDV antibody was 1:8 and sera were diluted two-fold until 1:8,192. For BVD virus, Singer strain (genotype 1) and strain 125 (genotype 2) were used and for IBR strain Colorado was used. The initial dilution of sera for the hemagglutination inhibition test for PI3 antibody was 1:10 and sera were diluted two-fold until 1:640. The Reisinger SF-4 strain was used for the test. Viruses were obtained from the National Veterinary Services Laboratory. An antibody titer ! 1:100 was considered positive for exposure to Leptospira interrogans, a titer of 1:10 was considered positive for exposure to PI3, and titers ! 1:8 were considered evidence of exposure to IBR and BVDV [11,16,17,31].

Analyses
An information-theoretic modeling approach was used to identify relationships between variables and white-tailed deer exposure to livestock pathogens. We conducted logistic regression analyses on exposure to PI3 and L. i. pomona (S1 File). We posited biologically plausible intrinsic (e.g., age, sex) and extrinsic (e.g., land type, study site, year, evidence of exposure to multiple infectious diseases) variables we considered influential in deer exposure to infectious diseases (Table 1). Prior to modeling, we screened all predictor variables for collinearity using Pearson's correlation coefficients (r > |0.5|) and used quantile plots to evaluate assumptions of normality; we used one variable from a set of collinear variables for modeling [32]. For PI3 and L. i. pomona, we specified 15 models: a global model containing 7 variables and a subset of a priori models. We also developed models that included the interaction of variables (e.g., site, year, land type) to account for multiple factors potentially influencing disease exposure simultaneously. Logistic regressions were used to analyze models that represented competing hypotheses explaining exposure to PI3 or L. i. pomona. However, all possible combinations of variables were not considered as this inflates the number of models beyond what can be reliably analyzed [33].
We used Akaike's Information Criterion (AIC) to select models that best described the data and used Akaike weights (w i ) as a measure of relative support for each model [33]. Because the number of deer tested was small relative to the number of variables (K) in several models (i.e., n/K < 40), we used AIC corrected for small sample size (AIC c ) for model selection [33]. We used formulas presented in Burnham and Anderson (2002) to calculate AIC c from the log-likelihoods for each model [32]. We compared AIC c values to select the most parsimonious model and considered models differing by 2 ΔAIC c from the selected model as competitive models [33,34]. We examined models 2 ΔAIC c from the best model to ascertain if they differed by 1 parameter from the best model and had essentially the same maximized loglikelihood [34]. In this case, models with additional parameters were unsupported and noncompetitive and considered potential models only because they contributed an additional parameter [33,34]. To account for model selection uncertainty among competitive models, we used model averaging to generate unconditional parameter estimates; parameter estimates were unconditional in the sense that they were conditioned on a suite of models weighted by their respective AIC c scores rather than on a single model whose probability of being the highest-ranked model may be only marginally higher than next-ranked models [32,33].
We determined associations between response and predictor variables using odds ratios (OR); OR for predictor variables are the relative amount by which the odds of the outcome increase (OR>1.0) or decrease (OR<1.0) with each unit increase in the predictor variable [35].
Further, we determined predictive capabilities of models with area under the receiver operating characteristic (ROC) curve. We considered ROC values ! 0.8 excellent discrimination, 0.7-0.8 acceptable discrimination, 0.5-0.7 as low discrimination, and values 0.5 indicated that model predictive capabilities were no better than random [36,37,38].

Results
During the 2010-2011 trapping season, deer (n = 133) did not show evidence of exposure to BVDV-1 or BVDV-2. As a result, we did not test for exposure to BVDV in subsequent years. Most animals had relatively low titers to the pathogens tested, but several deer had higher titers to PI3. For PI3, 14 deer had titers of 1:10, 33 deer had titers of 1:20, 8 deer had titers of 1:40, 10 deer each had titers of 1:80 and 1:160, and 3 deer had titers of 1:320. For L. i. pomona, 13 deer had titers of 1:100, 17 deer had titers of 1:200, 5 deer had titers of 1:400 and 2 deer had titers of 1:800. Table 2 shows seroprevalence of PI3, L. i. pomona, and IBR with reference to variables tested in Table 1 (S1 File). We tested 315 deer sera for exposure to IBR and 25 (7.9%) were exposed. We tested 316 sera for exposure to Leptospira interrogans serovar pomona and 37 (11.7%) showed evidence of exposure. Only 3 of 315 (1.0%) deer had evidence of exposure to L. i. bratislava, 8 of 316 (2.5%) had evidence of exposure to L. i. grippotyphosa, and 1 of 316 (0.3%) had evidence of exposure to L. i. hardjo. Deer did not demonstrate exposure to L. interrogans serovars canicola and icterohaemorrhagiae. We tested 316 sera for exposure to PI3 and 78 (24.7%) showed evidence of exposure. Logistic regression models explaining exposure of white-tailed deer to PI3 are listed in Table 3 in order of their ranking. Of 15 logistic regression models explaining exposure to PI3, a model containing CS, CY, and the interaction between CS and CY was selected as the highest ranked model and parameter estimates (β) for CS 1, and the CS 1 Ã CY 1 did not have 95% confidence intervals that included zero (Table 4). Support for this model was substantial (w i = 0.998) and predictive capability of the model was acceptable (ROC = 0.70); all other models were noncompetitive (ΔAIC c ! 12.49, w i < 0.02). The logistic equation for this model was logit (μ) = -0.300-2.374 (CS 1)-0.761 (CY 1)-0.799 (CY 2) site) + 2.857 (CS 1 Ã CY 1)-0.377 (CS 1 Ã CY 2; Table 5).
Fawns (n = 119) did not demonstrate evidence of exposure to L. i. pomona, thus we omitted this cohort from our analyses. Logistic regression models explaining exposure of white-tailed deer to L. i. pomona are listed in Table 4 in order of their ranking Of 15 logistic regression models explaining exposure to L. i. pomona, a model containing land type was the highestranked model, though support for this model was not substantial (w i = 0.278) and predictive capability of the model was low (ROC = 0.64). Several other models were < 4.0 ΔAIC units from the highest-ranked model, thus we generated parameter estimates across competitive models. Model-averaged parameters estimated for the top model suggested that exposure to L. i. pomona was influenced by land type (β = -1.002, 95% confidence interval = -1.971 to-0.034). However, parameter estimates for the remaining covariates (IBR, PI3, CS, age, CY, sex) were inconsistent, 95% confidence intervals always overlapped zero, and P-values based on permutation tests for these covariates were not significant (P ! 0.07), suggesting these factors had little effect on exposure of white-tailed deer to L. i. pomona. The logistic equation for the highest-ranked model was logit(μ) = 1.954-1.002 (LT; Table 5). Probability of exposure to parainfluenza 3 virus was 0.917 (OR = 0.093, 95% CI = 0.020-0.437) lower in the Northern Forest than the Eastern Farmland study site (Table 5). Conversely, probability of exposure to parainfluenza 3 was 17.41 (OR = 17.406, 95% CI = 3.138-96.562) times greater in the Northern Forest than the Eastern Farmland study site during 2011 (Table 5). For the remaining categorical variables in the highest-ranked parainfluenza 3 exposure model, odds ratio 95% confidence intervals overlapped each other, indicating that these were poor predictors of exposure to parainfluenza 3 (Table 5). Moreover, probability of exposure to L. i. pomona decreased by 0.37 (OR = 0.367, 95% CI = 0.105-0.629) between private and public land (Table 5).   Table 5. Parameter estimates (β), standard error (SE), odds ratio, and odds ratio 95% confidence intervals (lower CI and upper CI) for the best-approximating models in candidate sets evaluated for exposure of white-tailed deer to PI3 and L. i. pomona in northern and east-central Wisconsin, USA, 2010-2013.

Discussion
In the white-tailed deer sampled, we identified antibodies against several pathogens that can affect livestock and white-tailed deer, and in some cases, humans [7,8,17,19]. Serologic evidence of exposure to PI3 and L. i. pomona occurred in over 10% of deer sampled, but contrary to Wolf et al. (2008), who sampled deer in adjacent Minnesota, we did not identify BVDV antibody in deer captured in two areas in Wisconsin [10]. Wolf et al. (2008) sampled 114 deer and found 20-39% seroprevalence, thus it is reasonable to assume that, if present, we would have detected antibodies given our sampling intensity (n = 133 deer) [10]. Similar serologic investigations have found low seroprevalence in white-tailed deer. Passler et al. (2008) identified antibody in 2 of 165 (1.2%) white-tailed deer in Alabama [39]. Additionally, Sadi et al. (1991) did not identify antibody in 396 deer sera from Anticosti Island, Quebec and speculated that the deer had not been exposed because domestic ruminants had not been on the island for over 50 years [17]. Cantu [40]. White-tailed deer may be exposed only sporadically to BVDV depending upon site and/or year and we may not have sampled during a year with obvious exposure. Cattle have been shown to experimentally transmit BVDV to white-tailed deer in close proximity, and fawns born to adults exposed to infected cattle were persistently infected with BVDV [41], so perhaps deer are not coming in contact with infected cattle on our study sites. White-tailed deer are commonly exposed to PI3, and we found that 25% of the deer sera from Wisconsin had antibodies against PI3, similar to that reported previously for mule deer trapped on national parks (32%) [42]. In Minnesota, Ingebrigtsen et al. (1986) found that 20% of the white-tailed deer tested were seropositive for PI3 [6]. Seroprevalences documented in this study and from Minnesota are lower than what has been reported in other areas. Seroprevalence of 82 to 84% has been reported for white-tailed deer in Quebec in the 1980's [17], and seroprevalence of 95% was reported for mule deer in Idaho in 1975 [43]. Exposure to PI3 may depend upon local environmental factors, including deer and/or cattle density.
Deer in the Eastern Farmland study site were more commonly exposed to PI3 (30%) than those captured in the Northern Forest (18%), and both the number of cattle and the number of deer are higher in the eastern study area. In December 2012, over 103,000 cattle were registered with the National Agriculture Statistics Service (NASS; [44]) from Waupaca, Outagamie, and Shawano counties, which includes the Eastern Farmland study area (Fig 1). However, approximately 18,000 cattle were registered with NASS in December 2012 in Rusk and Sawyer counties where Northern Forest deer were captured [44]. Parainfluenza 3 is likely transmitted to deer from cattle, however the disease also could be maintained in deer with no contact with cattle [18,42]. A previous study has shown higher seroprevalence for livestock diseases in deer herds with higher densities [5]. In 2012, fall deer densities for the three deer management units closest to Shiocton, Wisconsin (e.g., where Eastern Farmland deer were trapped) were 267, 197, and 194 deer per square km [45]. Fall deer densities in 2012 in deer management units near Winter, Wisconsin (e.g., where Northern Forest deer were trapped) were 73, 73, and 47 deer per square km [45]. Though speculative, increased seroprevalence of PI3 in the Eastern Farmland could be attributed to higher deer and/or cattle density, suggesting that host density affects exposure to livestock pathogens.
Our findings indicate that seroprevalence for PI3 varies with year. Previous studies have documented significant variation in PI3 seroprevalence in free-ranging ungulates by year [46].
We found similar seroprevalence in 2011 (41/134 = 31%) and 2013 (22/78 = 28%), but seroprevalence was lower in 2012 (15/104 = 14%), but prevalence changed dramatically with year in the Northern Forest site. In 2011, prevalence was 36%, but dropped to 2% and 7% in 2012 and 2013, respectively, so the interaction between year and site was significant, indicating that seroprevalence varies independently on each site on a yearly basis. Myers et al. (2015) also found a difference in seroprevalence for PI3 with study site in the same year for mule deer [40]. Since both deer and cattle have been shown to transmit PI3 [20], perhaps deer and/or cattle are maintaining PI3 in the populations separately between the two study sites. Additional research is needed to better understand how cattle and/or deer density affect seroprevalence of PI3 in deer.
We found that 12% of the white-tailed deer sampled in Wisconsin had antibody against L. i. pomona. In Washington, Myers et al. (2015) found antibody against Leptospira interrogans serovars grippotyphosa (13%), bratislava (4%), and canicola (1%), but not serovar pomona [40]. However, Roug et al. (2012) found that 6.5% of the mule deer sera tested were seropositive for Leptospira antibody (all serovars combined) [47]. In Minnesota, only 3% of the white-tailed deer sera tested in 1976 were seropositive for all serovars of Leptospira interrogans combined [6]. From 1984 through 1989, Goyal et al. (1992) tested white-tailed deer sera for antibody against 6 serovars of L. interrogans in Minnesota [48]. Deer sera were not seropositive from 1984-1985, but all sera collected from 1986 through 1988 were positive for antibody against L. i. pomona and/or L. i. bratislava, suggesting variability in exposure by year [48]. Fournier et al. (1986) found that 7% of the white-tailed deer sera tested in Ohio had antibody against at least one serovar of Leptospira, and found that a higher proportion of cattle were seropositive from counties where seropositive deer were identified than from counties where deer were seronegative [11].
In our study, exposure of white-tailed deer to L. i. pomona was positively associated with public land. Leptospira interrogans serovar pomona is maintained in deer and other maintenance hosts, including wildlife and domestic animals. Myers et al. (2015) suggested that mule deer may be exposed to L. interrogans from contact with cattle that commonly share their range [40]. They also propose that mule deer could serve as maintenance hosts for the bacterium [40]. Cirone et al. (1978) found that over 75% of raccoons (Procyon lotor), coyotes (Canis latrans), and striped skunks (Mephitis mephitis) tested in California were seropositive for L. i. pomona antibody [49]. In the same study, Cirone et al. (1978) suggested that high seroprevalence in white-tailed deer was related to the site where deer were sampled which varied with access to contaminated water [49]. All black rats (Rattus rattus) sampled from a ring-necked pheasant (Phasianus colchicus) farm with access to fresh tapwater were seronegative, whereas Norway rats (R. norvegicus) sampled from sites with access to only stagnant water were seropositive for leprospiral antibodies [49]. White-tailed deer are positively associated with water, especially when moisture content of foods is low [50]. Additionally, home ranges for whitetailed deer in the midwestern United States often contain ponds, lakes, or other waterways [51]. At this time, we do not know how deer are exposed to L. i. pomona, but contaminated water sources on public land may play a role, specifically if maintenance hosts and white-tailed deer occur at higher densities at these water sources, and deer density has been proposed as a factor influencing seroprevalence of L. i. pomona [11].
We did not find evidence that individual white-tailed deer were exposed to numerous livestock pathogens during the same year and environmental variables that explained exposure to PI3 and L. i. pomona differed within our study. Leptospira sp. is transmitted primarily through contaminated water and soil [7] and PI3 is primarily transmitted by aerosolization [20]. We found that site explained exposure to PI3 potentially because densities of deer and cattle are higher in the Eastern study area, and PI3 is transmitted from animal to animal. At this time, we do not have information regarding water sources or contamination associated with stagnant water on public land, so we can only speculate as to the cause of the relationships we identified for L. i. pomona seroprevalence. Future research focusing on potential risk factors for exposure to L. i. pomona is warranted, including contaminated water sources and contact with cattle.
Supporting Information S1 File. Raw data used for models explaining exposure of white-tailed deer in Wisconsin to parainfluenza 3 and Leptospira interrogans pomona. (XLSX)