Introduction

In large mammals, chronic disease often manifests as having low detectability, moderate impacts on adult mortality and fecundity, and depressed population growth rates that are sensitive to changes in adult survival [1]. Chronic diseases are difficult to detect due to lack of mass mortalities, rapid population declines, or shifts in age structure [1]. Few studies have investigated population-level impacts of chronic diseases in wildlife populations, despite the recently increasing interest and emphasis of population-level effects of wildlife diseases [2]. The dearth of well-studied population-level effects of chronic diseases is worrisome given that research suggests diseases with preclinical stages rather than acute diseases are more likely to influence long-term population-level dynamics [2]. The widespread potential of population-level impacts warrants further research on chronic wildlife diseases [2].

Confounding the issue of investigating chronic diseases is the temptation of, and pressure on, managers to react to newly discovered diseases in ways that may not be optimal. Chronic disease may have minor effects on population vital rates early in a disease epidemic [3]. However, if the disease state shifts from epidemic to endemic, then vital rates may not be affected for years or decades and monitoring must be completed over an extended time-frame. Monitoring is crucial because a firm understanding of the effects of disease on population vital rates is necessary to accurately model disease dynamics and determine suitable management options [3]. Unfortunately, wildlife diseases are challenging to study because of their insidious nature, logistical difficulties, statistical challenges, and high costs [2, 3]. This is the case for chronic wasting disease of white-tailed deer (Odocoileus virginianus), mule deer (Odocoileus hemionus), Rocky Mountain elk (Cervus elaphus nelsoni), and moose (Alces alces shirasi) [4–8].

Chronic wasting disease (CWD) is a uniformly fatal, progressive neurodegenerative transmissible spongiform encephalopathy (TSE) that occurs in wild cervid populations in 21 U.S. states and two Canadian provinces [9]. The TSEs are caused by proteinase-resistant, abnormal isoforms (PrP^res) of normal host cellular proteins (PrP^C) known as prions. The causative agent of CWD is known as PrP^CWD.

There are few studies on population-level effects of CWD on cervid populations. One such study was conducted on a mule deer population near Boulder, Colorado, USA [10]. Deer abundance declined 45% during 1988–2006. It was believed CWD had been endemic since 1985 and was highly prevalent (males = 41%; females = 20%). The decline was attributed to high prevalence of CWD resulting in low overall adult survival (0.72).

Information suggests CWD has potential to cause population declines and possibly localized extinctions at high prevalence; however, this has not been definitively proven or observed. To address if and how CWD negatively impacts deer population dynamics, we intensively monitored a white-tailed deer population in southeastern Wyoming over a protracted time-period (2003–2010) to estimate population vital rates and model the influence of disease on population performance. We hypothesized that demographic rates are altered by CWD to an extent large enough to lower the population growth rate. The specific objectives were to monitor CWD-positive and CWD-negative white-tailed deer in a high-prevalence CWD endemic area throughout their lifespan via radio-telemetry and global positioning system (GPS) collars. We sought to determine the following for the two segments of the population: a) demographic and disease indices including CWD prevalence, causes of mortality, pregnancy and recruitment rates, b) annual survival, and c) finite rate of population growth (λ). These indicators allowed us to determine the magnitude of the effect of CWD on a free-ranging white-tailed deer population.

Materials and Methods

Anesthesia was used on all white-tailed deer that were captured and processed for enrollment into study. Deer were chemically immobilized with 0.03 mg/kg carfentanil and 0.7 mg/kg xylazine. All deer were injected subcutaneously with procaine/benzathine penicillin G combination (25,000 units/kg based on benzathine fraction, Bimeda, Le Sueur, Minnesota, USA) and intramuscularly with 1.5 mg/kg of Banamine (Intervet Inc., Merck Animal Health, Summit, New Jersey, USA). Immobilized deer were reversed with 100 mg naltrexone per 1 mg carfentanil and 2 mg/kg tolazoline and monitored until recovered and ambulatory. All animal procedures were approved through the University of Wyoming (Laramie, Wyoming, USA) Institutional Animal Care and Use Committee (Protocol #A-3216-01). Wyoming Game and Fish Department approved our Chapter 33 Capture Permit to capture the pre-determined number of white-tailed deer annually (Permit #531).

Data Analyses

All statistical analyses and regression models were programmed using SAS (SAS Institute, Cary, North Carolina, USA) unless stated otherwise. We wished to determine the influence of covariates, including CWD-status, on the probability of pregnancy, which is an important vital rate to understand the population dynamics as it relates to CWD. We used PROC LOGISTIC to perform a logistic regression analysis [22] on probability of pregnancy (event = 1) given CWD-status (CWD), age, body condition score (BCS) at time of capture, and year of capture (year). No fawn (≤8 months) deer were pregnant, so that age class was excluded from analysis. Single parameter models were generated to begin forward parameter selection; however, none of the parameters were significant. We generated the full model containing all four parameters: (1)

Pregnancy is an important vital rate used in the matrix population model, and thus we needed to determine if pregnancy varied by CWD-status to inform how to use the metric in the population matrix. Due to small sample sizes, we utilized PROC FREQ to perform Fisher’s exact χ² analysis [23] comparing observed proportion of pregnant does to expected proportion of pregnant does by CWD-status for each capture year as an overall test for significance of CWD effects on pregnancy.

We needed to calculate fecundity estimates separately by CWD-status to determine if CWD impacted ability of does to raise fawns (a hypothesis of interest to population dynamics related to CWD) as well as to include as a vital rate in the population model. We performed a 2-group t-test [24] using PROC TTEST to compare average number of fawns per doe between CWD-positive and CWD-negative female deer. Recruitment was analyzed separately between 2008 and 2009. We used these same data with PROC GLIMMIX to perform a mixed model [25] given that some of the fawn counts were from the same does in 2008 and 2009. The model produced a log odds of fawn production given CWD-status, age, year, CWD x age, and CWD x year: (2)

To better understand what factors influence annual survival of white-tailed deer, we analyzed annual survival data using Cox proportional hazards model to examine survival differences given the following covariates: CWD-status, sex, age class, year, and migratory/dispersal status (binary) [26, 27]. We determined mortality dates as the first mortality event recorded by the GPS unit (4 hour delay) or estimated based on carcass condition from the first date of hearing a mortality signal during radio-telemetry for VHF-marked deer (4 or 6 hour delay depending on model). Deer were right censored at the date of the last relocation if lost to follow-up due to transmitter failure, dropped transmitter, or long-range dispersal and we failed to relocate with aerial telemetry. We also right censored deer killed during capture or by poachers, or that survived to the end of the study period. Deer killed legally by hunter harvest were not right censored as hunting was an integral part of the study system. Deer that initially were CWD-negative then tested CWD-positive during subsequent captures were right censored as CWD-negative at the capture date of first CWD-positive test. We tested proportionality of hazards ratios using the TEST option in PROC PHREG [28]. We utilized an Extended Cox model after we determined that proportionality of hazards ratios was not met ( = 9.0252, P = 0.0605). We used PROC PHREG in SAS to evaluate the effects of the above covariates plus age x duration interaction term (to account for lack of proportionality) on annual survival of deer, modeled as duration known alive (duration) x living status (e.g., alive (0) or dead (1); status) [29]. The following was the full model: (3)

We implemented backwards elimination for parameter selection and Akaike’s Information Criteria (AIC) [30] for model selection (supported models were within 2 AIC values of the model with the lowest AIC value; ΔAIC) along with consideration of significant parameters (based on Wald χ² statistic) and our biological knowledge of the system.

We wished to compare subcategory-specific annual survival rates based on capture year_t to capture year_t+1 to better understand the pairwise comparisons of significant factors from the Cox proportional hazards modeling. While the Cox proportional hazards modeling indicates which factors are important and the risk associated with each factor, it does not provide an actual survival estimate to be used for intra- and inter-population comparisons. Therefore, we used Kaplan-Meier survival estimation [31] to generate these metrics with PROC LIFETEST in SAS. We generated survival estimates separately by age and CWD-status for males and females. The χ² Log-Rank test [32] determined differences in annual survival rates by sub-categories using PROC LIFETEST by strata.

We also needed to generate age and CWD-status specific survival estimates by biological year (June 1 –May 31) for does to be used as a vital rate in the matrix population model. For these analyses, we once again used the Kaplan-Meier survival estimator. Biological year of fawns was defined as September 1 –May 31 because fawn recruitment of marked does was determined during the first week in September. However, these fawns were not marked with radio transmitters; we captured fawns when they were much older in February; therefore, we had to estimate fawns survival from September to February. We combined published estimates of fawn survival during this missing time period from Dusek et al. [33] and survival data of fawns tracked on this study from February1 through May 31 with a weighted average to account for differing lengths of time between the two sources of survival estimates to produce one fawn survival estimate.

We needed to calculate annual CWD incidence as a vital rate to be used within the matrix population model, but we also needed to perform comparisons on these estimates by sex and age class to inform how to populate the matrix with this metric. We calculated annual CWD incidence using the time-to-event (CWD conversion) Kaplan-Meier estimator [31] with PROC LIFETEST. Incidence was calculated separately by sex and for each age-class. We performed a 2-group t-test [24] to determine if a difference in incidence existed between bucks and does using PROC TTEST (Cochran option).

Our ultimate question of interest was to determine the impact of CWD on the growth rate of this population. We calculated the finite rate of population growth (λ) using a post-breeding, age-structured, female-only 18 x 18 dimension Leslie matrix [34, 35] in MATLAB^® (The MathWorks, Inc., Natick, Massachusetts, USA). Vital rates incorporated were fecundity (average number of fawns per doe in the first week of September) and age-specific pregnancy rates, survival rates, and CWD incidence. All vital rates were estimated separately for CWD-negative and CWD-positive deer except fecundity, which did not differ by CWD-status. The 18 x 18 transition matrix, A, represented the estimated demographic rates of the study population with both CWD-negative and CWD-positive females, and the transition between them due to infection, represented (Fig 1). We calculated the population growth rate as the dominant eigenvalue, λ₁ [35]. To determine the population vital rates that most influenced lambda (i.e., the vital rate that lambda was most sensitive to and would change the most if those vital rates changed), we performed sensitivity analysis of A in MATLAB using the vitalsens.m function developed by Morris and Doak [36] to quantify how sensitive λ was to a change in value of each vital rate. In addition, we determined the elasticity of each vital rate [35].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Leslie Matrix Population Model.

Post-breeding, age-structured, female-dominated 18x18 Leslie Matrix model of white-tailed deer population on Deer Creek drainage SW of Glenrock, WY (2003–2010) that was located within the chronic wasting disease (CWD) endemic area. n_t represents the number of deer in each age class by CWD-status ((-) = PrP^CWD not detected, (+) = PrP^CWD detected). represents estimated survival by age class, i, and CWD-status (- or +), is the estimated fecundity rate and is the age-specific CWD incidence rate.

https://doi.org/10.1371/journal.pone.0161127.g001

We calculated the 95% confidence interval for λ₁ using parametric bootstrapping [36]. Specifically, 2000 each of age and CWD-status specific survival and CWD incidence rates were randomly estimated from the β-distribution using the betaval function and 2000 estimates of the fecundity rate were generated using the stretchbetaval function; both functions were from the popbio package [37] in program R v.3.2.5 [38]. The estimates were based on the mean and bias-corrected variance estimate generated using the Kendall function in the popbio package. We then used the 2000 vital rate estimates to generate 2000 λ₁ estimates with the eigen.analysis function in the popbio package based on the A matrix and we determined the 95% confidence interval from the 0.025 and 0.975 quantiles of the λ₁ distribution [36]. The popbio package reproduces the same results in program R as the MATLAB-based analysis.

To understand the effect of CWD on population vital rates, we created two matrices, Aneg and Apos, taking the form of 9 x 9 transition matrices for CWD-negative and CWD-positive populations respectively. The two matrices differed in that one model assumed 0% CWD-prevalence and one model assumed 100% CWD-prevalence, allowing determination of the magnitude in change of λ (Δλ) due to CWD. We performed a life table response experiment (LTRE) on the transition matrices, Aneg and Apos, using the vitalsens.m function in MATLAB [35] to better understand the influence of CWD on λ₁.

To evaluate the influence of CWD incidence on population growth rate, we varied incidence from 0 to 1.0 by 0.05. This range of incidence rates was inserted in the full 18 x 18 matrix model, keeping all other parameters equal, and calculating λ across the range. State wildlife agencies routinely track CWD by annual prevalence from hunter-harvested or targeted deer and elk. We converted each incidence rate into annual prevalence based on the following equation: (4) where P was prevalence, I was incidence, and was the estimated duration of illness estimated by Kaplan-Meier [31] using PROC LIFETEST. All CWD-positive deer were included in the analysis with the enrollment date set as the date of first positive CWD test and mean time known alive (i.e., ) calculated by Kaplan-Meier analysis [31].

We were interested in the population age structure to determine if the population was shifted to a young age structure. We calculated the dominant right eigenvector (w₁), which gives the stable population structure, using function eigenall.m in MATLAB [35] to determine the stable population age structure. Given the following equation: (5) where w_i’s are age-specific contributions to population growth. When one sums the age specific w_i’s and then divides each w_i by the total, the proportion of the population in each age class is determined [35]. The age structure of both male and female deer by CWD-status was determined using this method.

Results

During the study period (January 2003–February 2010), 112 deer were captured as fawns (≤8 months-old; female: male = 57: 55) and 63 deer captured originally as adults (≥1.5 years-old; female: male = 27: 36). All deer were recaptured annually. Overall CWD prevalence during the study period (last known CWD-status of each individual deer) was 35.4% (n = 161). Prevalence was higher in does (42%, n = 81) than bucks (28.8%, n = 80, = 6.608, P = 0.01). Average annual CWD prevalence (based on annual tonsil biopsies) was 23.8% (n = 345) overall, 24.3% (n = 202) for does, and 23.1% (n = 143) for bucks.

There were 118 mortalities (CWD-negative = 64, CWD-positive = 50, CWD-unknown = 4) during the study period (S1 Table). Hunter harvest was the most common cause of mortality (n = 46) and more CWD-positive deer (n = 19) were harvested than expected based on average annual CWD prevalence (41.3% vs. 23.8%, = 8.876, P = 0.029). Bucks were more common (76%) than does in the harvest. There were 20 capture-related mortalities, representing 4.2% of all captures (n = 476). This is an overestimate of capture-related mortality because many non-target deer were captured and released without injury during Clover trapping. Seventeen deer (female: male = 12: 5) died of clinical CWD; does comprised 71% of clinical cases, but made up only 48% of the study population.

Age, CWD-status, year, and body condition score did not influence pregnancy. Average proportion pregnant for CWD-negative deer was 0.95 (n = 109, 95% C.I. = 0.92–0.99) and for CWD-positive deer was 0.92 (n = 38, 95% C.I. = 0.84–1.0). There was not a statistically significant difference detected in proportion pregnant by CWD-status annually or across all years combined ( = 0.601, P = 0.438).

Average number of fawns per doe was 0.74 (95% C.I. = 0.47–1.00). There was no statistically significant difference detected in the average number of fawns per doe by CWD-status in 2008 (CWD-negative = 0.56, CWD-positive = 0.67, t₁₇ = -0.23, P = 0.819) or 2009 (CWD-negative = 0.90, CWD-positive = 1.00, t₁₃ = -0.22, P = 0.829).

Six Cox proportional hazards models were evaluated using AIC, statistical significance of parameter estimates, and biological knowledge of the system. The full model included the following parameters: CWD, Sex, CWD x Sex, Age, Age x Time, Migration, Dispersal, and Year; the top model included CWD, Sex, Age, Age x Time, and Dispersal. The most significant parameter was CWD, which had the highest hazard ratio. The CWD-positive deer were 4.51 times more likely to die annually than CWD-negative deer (β₁ = 1.51, = 44.62, P <0.001, 95%, C.I. = 2.9–7.0). Bucks were 1.70 times more likely to die than does (β₂ = 0.532, = 5.17, P = 0.023, 95% C.I. = 1.08–2.69). Deer that did not disperse were 1.61 times more likely to die than deer that did disperse; however, the result was not statistically significant (β₅ = -0.493, = 1.118, P = 0.290, 95% C.I. = 0.656–4.08). Age and age over time did not affect survival probability.

Kaplan-Meier survival log rank tests were performed on all ages combined (overall) and by each age class (Table 1). Survival comparisons were all statistically different except CWD-positive females vs. males ( = 2.73, P = 0.098), indicating sex did not significantly affect survival of CWD-positive deer. Of the five significant comparisons, CWD-positive deer had lower survival than CWD-negative deer and males had lower survival than females. Log rank tests were highly significant by CWD-status and less-so by sex, which indicated CWD was a greater indicator of annual survival. There were no clear trends by age class.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Annual Survival Rate Comparisons.

https://doi.org/10.1371/journal.pone.0161127.t001

There was a large difference in annual survival of CWD-positive deer (0.396) compared to CWD-negative deer (0.801, = 39.70, P = <0.001, Fig 2A). Kaplan-Meier survival curves indicated a slow, steady drop in survival of CWD-positive does (Fig 2B). Survival estimates were similar during most of the year for CWD-positive and CWD-negative bucks and then dropped slowly for CWD-negative bucks but precipitously for CWD-positive bucks between weeks 35–40, which coincided with the 6-week hunting season. Fewer CWD-negative bucks survived annually (73%) than CWD-negative does (85%; Fig 2D). The CWD-positive deer were the only group that did not differ significantly by sex, but both survival rates were extremely low (Fig 2E, Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Kaplan-Meier Annual Survival Rate Curves.

Survival rate curves of segments of the white-tailed deer study population that was captured, tested for CWD by tonsil biopsy, marked with radio transmitters, and followed by radio telemetry SW of Glenrock, WY (2003–2010).

https://doi.org/10.1371/journal.pone.0161127.g002

Annual CWD incidence increased more rapidly in bucks, reaching peak in the first year of life (51%—indicated by CWD-positive test as a 1.5 years-old (yearling)), declining slightly in the second year, returning to near peak incidence during the third year, and then declining steadily to 0 by the 6^th year (Fig 3). Incidence increased slower in females, but reached a higher peak than males (65%) during the 5^th year, then also dropping to 0 during the 6^th year. Incidence was not significantly different between sexes (t₄ = -1.26, P = 0.277).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Annual Incidence Rates.

Chronic wasting disease (CWD) annual incidence rate by sex and age class of white-tailed deer captured, CWD-tested annually, radio-collared, and monitored by radio-telemetry SW of Glenrock, WY (2003–2010). The CWD incidence was calculated by Kaplan-Meier time to event analysis.

https://doi.org/10.1371/journal.pone.0161127.g003

The dominant eigenvalue, λ₁, of our 18 x 18 matrix model was 0.896 (0.859–0.980), which indicates 10.4% annual decline from 2003–2010, assuming a stable age distribution. A λ₁ of 0.896 is not sustainable (t_0.5 = 5 years, t_extinction = 48 years). To determine magnitude of Δλ₁ due to CWD, λ₁ was determined for a subpopulation of CWD-negative (Aneg) and CWD-positive (Apos) deer, which were 1.07 and 0.681 respectively. The results suggest CWD significantly depressed λ₁ in the study population.

Sensitivity analysis indicated λ₁ was most sensitive to changes in survival of CWD-negative fawns (0.280) and yearlings (0.269) and, to a lesser extent, 2.5 years-olds (0.154). Fecundity was not important; however, λ₁ was slightly sensitive to changes in fecundity of yearlings (0.156). Changes to vital rates in older age classes did not significantly affect λ₁. Elasticity results were similar to sensitivity analysis, except in the case of fecundity, which had small values (≤0.06) for every age class (S2 Table).

Survival of yearlings and 2.5 years-olds was most severely reduced by CWD and had the greatest impact on lowering λ₁. These results, combined with sensitivity analysis, suggest that survival overall across younger age cohorts is influencing λ₁. The LTRE indicated that survival of yearlings and 2.5 years-olds contributed most to the change in λ₁ (Δ λ₁) cause by CWD, with Δ λ₁ of 0.144 and 0.173 respectively. No other vital rate caused a Δ λ₁ ≥0.025. The LTRE was similar to sensitivity analysis except in the case of fawn survival, which contributed 0 to the treatment effect because survival of CWD-negative and CWD-positive fawns was equal. Overall treatment effect of CWD was 0.348. The magnitude of the negative effect on λ₁ by CWD infection was 0.382. Survival was recalculated for each age class with all hunter-related mortalities right censored from Kaplan-Meier analysis. With hunting-related mortality censored, the resulting vital rates estimated a λ₁ of 1.00.

By varying incidence rates, recalculating λ₁ for each incidence rate and plotting λ₁ by incidence, it was determined λ₁ dropped below 1.0 at an annual incidence rate of 0.26. Transforming incidence into prevalence and then plotting the λ₁ values with the new prevalence values estimated λ₁ was <1.0 at 0.27 (27%, Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Lambda by Prevalence.

Curvilinear relationship between increasing chronic wasting disease (CWD) prevalence and decreasing lambda (λ₁) simulated from vital rates of a white-tailed deer population captured, CWD-tested annually, marked with radio transmitters, and monitored by radio-telemetry SW of Glenrock, WY (2003–2010). The curve was generated by holding all population vital rates constant, but varying incidence up and down from the population incidence by intervals of 0.05, re-running the Leslie matrix population model with the constant vital rates and altered incidence value populating the transition matrix, A, and calculating lambda. The incidence rates were then converted into prevalence estimates to be more useful to wildlife managers because state wildlife agencies collect surveillance data in prevalence proportions, not incidence rates. The solid horizontal line at λ₁ = 1.0 represents the threshold at which population growth begins to decline (λ₁< 1.0) and the dark dashed line is the simulated population growth rate with accompanying 95% confidence intervals (lighter double dashed lines).

https://doi.org/10.1371/journal.pone.0161127.g004

The dominant right eigenvector, w₁, determined the proportion of CWD-negative deer was highest in the fawn and yearling age classes and continued a constant, steep downward slope until the proportion was 0.005 in 6.5 years-old deer (Fig 5). Conversely, proportion of CWD-positive deer was lowest in fawns and yearlings, climbed to approximately 0.15 in 2.5 years-old deer and then plateaued (Fig 5). The age structure of all deer combined showed the majority of the population was found in the first three age classes followed by a rapid decline in older age classes (Fig 5), indicating the age structure was shifted to the left and dominated by young, immature, and sub-prime-aged deer.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Population Age Structure.

Proportional age structure of females separate by chronic wasting disease (CWD)-status and combined CWD-negative and CWD-positive deer of white-tailed deer captured, CWD-tested annually, marked with radio transmitter, and monitored by radio-telemetry SW of Glenrock, WY (2003–2010). Age structure was calculated from the dominant right eigenvector, w₁, of the Leslie matrix population model of the transition matrix, A. The CWD-negative age structure was the result of assuming 0% CWD incidence modeled with a 9x9 ANeg transition matrix, the CWD-positive age structure was the result of assuming 100% CWD incidence with a 9x9 APos transition matrix, and the combined age structure was the result of modeling what is occurring in this population currently based on the 18x18 combined transition matrix, A; this result represents the actual population age structure.

https://doi.org/10.1371/journal.pone.0161127.g005

Discussion

The difference in survival by CWD-status and the high proportion of CWD-positive deer in this population help explain the declining population trend (λ₁ = 0.896). The CWD-positive deer were 4.5 times more likely to die annually than CWD-negative deer. These results support concerns of wildlife managers, wildlife disease experts, and conservationists that this endemic (chronic) disease can negatively impact deer population sustainability at high disease prevalence. The sensitivity analysis and LTRE indicated survival of fawns, yearlings, and 2.5 year-old CWD-negative deer were primarily responsible for the reduction in λ₁ caused by CWD. It is likely that CWD and hunter harvest, the main causes of mortality, have produced the young age structure observed in this population. At the current λ₁, this population is not sustainable with possible extinction in 48 years at current levels of mortality and fecundity given the worst-case scenario of frequency dependent transmission [39] and no immigration or genetic selection for less susceptible genotypes for CWD [40].

Our estimate of λ is the lowest reported for a free-ranging cervid population with endemic CWD. Dulberger et al. [41] reported a λ of 0.97 (95% credible interval = 0.82–1.09) in a CWD-endemic mule deer population in Colorado, and λ = 1.0 has been reported for CWD-endemic elk populations in South Dakota and Colorado [42, 43]. These values were not particularly worrisome as λ either overlapped 1.0 given the credible interval or was equal to 1.0, indicating stable populations. It is particularly concerning how low our λ₁ value was given that the study species was white-tailed deer, which have a higher lifetime reproductive potential than the other three CWD susceptible species.

Hunter harvest often is a major cause of mortality in white-tailed deer, which are the most common and wide-spread big game species in North America. We demonstrated that CWD-positive adults were over-represented in hunter harvest, and others [44] have suggested CWD-positive mule deer also are more vulnerable to hunter harvest. The behavioral shifts, including movement patterns, changes in breeding behavior during harvest, decreased reaction time to stimuli, and changes in habitat type used by CWD-positive mule deer may have caused biased harvest proportions. Conversely, Grear et al. [45] found no difference in harvest susceptibility between CWD-negative and CWD-positive white-tailed deer in Wisconsin, perhaps due to relatively low CWD prevalence (6.3% in adults). It is probable that the behavioral changes suggested by Conner et al. [44] affect CWD-positive deer susceptibility to harvest. Captive CWD-positive deer often show altered response to human activity [4], including an apparent lack of recognition of human presence. Activity analysis suggested CWD-positive bucks did not participate in the rut at the same level as CWD-negative bucks; the rut coincided with the hunting season [11]. Our data support the notion that CWD-positive bucks were less aware of the rut and the hunting season and were more susceptible to being shot by a hunter.

Over-representation of CWD-positive deer in the hunter harvest suggests behavior is altered by CWD prior to clinically recognizable CWD infection. Rather than thinking of CWD as a strictly pre-clinical disease followed by a short, obvious clinical stage of disease, we believe CWD infection should be envisioned as a slow, progressive decline in health and alteration of normal behavior, which ends with clinically recognizable disease. Given the relatively short clinical stage of CWD and the limited hunting season, it is hard to believe CWD-positive deer would be more susceptible to harvest if this slow alteration in health and behavior does not occur. Further, the majority of hunters do not intentionally harvest emaciated or sick animals.

There was a discrepancy in sex ratio of deer that died of clinical CWD (female: male = 12: 5). The high proportion of bucks in the harvest (76%) and over-representation of CWD-positive deer compared to CWD-negative deer may explain why females comprised 71% of clinical CWD cases. Data suggest CWD-positive bucks were harvested at a higher rate than expected and prior to reaching terminal stages of disease while the low harvest rate of does facilitated disease progression to clinical CWD. Females lived longer (137.2 weeks) after testing positive for CWD than bucks (107.4 weeks), which supports this argument. Also, the matriarchal social structure of females may explain why CWD incidence was higher in females and a more steady progression than males. Males were removed earlier in disease progression and had less time to spread disease directly to susceptible bucks in their bachelor herds throughout most of the year. Meanwhile, females progressed to clinical CWD, presumably shedding infectious prions into the environment and transmitting prions directly to susceptible females in their familial groups early in infection [46] and throughout most of the year. It is known that CWD prevalence is not spatially homogenous [47–50]. White-tailed deer are highly faithful to small home ranges in the Rocky Mountain West [11]. Prolonged prion shedding by CWD-positive does within their home range, including favored bedding locations, accompanied by communal grooming and shared home ranges with females provided opportunity for disease transmission through time.

Our study finding of higher incidence in does than bucks contradicts other reported studies that documented higher incidence in bucks than does (e.g., [45, 50, 51]). Presumably in hunted populations, bucks were the favored hunted sex as well. We believe that this discrepancy may be a function of the riparian habitat concentrating white-tailed deer and thus environmental contamination and allowing for the proposed role of does in the transmission of CWD in our study system. It is possible that in the future, when other habitats, such as winter lots in Wisconsin (where CWD has not been endemic for as long as Wyoming) have had similar time to become equivalently contaminated, does may become similarly important to transmission and incidence may increase in does in these population. In other words, perhaps our study population is an indicator of things to come, where initially bucks experience higher incidence until a threshold is met when does experience higher CWD incidence. This scenario assumes concentrated environmental contamination, however. For wide-ranging and dispersed populations, bucks may always experience higher incidence than females.

It is important to note that hunters may have had a bias in regards to harvesting collared deer. It is possible that hunters avoided shooting collared does in lieu of harvesting an uncollared doe to avoid altering the study results and to not have to deal with the hassle of returning a collar. Hunters targeting bucks may not have had such concerns if the antler size was large enough. If this was the case, then we may have over-emphasized the ratio of bucks to does in the harvest ratio. We believe this bias was relatively minor, at least within the main study site that encompassed the majority of the winter range, because hunters were forced to use one hunting outfitter on the VR Ranch and after conversations with this outfitter, they at least claimed to not be biased for or against harvesting collared animals.

Pregnancy and recruitment results indicate CWD does not compromise reproduction in female white-tailed deer. Blanchong et al. [52] also determined pre-clinical CWD did not negatively impact female reproduction in Wisconsin white-tailed deer. No difference in pregnancy indicates does participate in the rut regardless of CWD infection-status. It was not possible to determine if there was a difference in pregnancy and recruitment between pre-clinical and clinical CWD-positive does. However, it was common during the study to find one or two near-term fetuses in clinical-CWD female carcasses during the third trimester (Cornish and Edmunds, unpublished data). It is likely that fetuses exacerbate emaciation and hasten the death of does with terminal CWD. Our findings suggest does in pre-clinical disease give birth to fawns and are as successful at raising fawns to early September as CWD-negative does. Equal reproduction by CWD-positive does should dampen somewhat the negative effects of CWD on deer populations. Future research on neonate and young fawn survival is warranted, specifically to address the ability of CWD-positive white-tailed deer does to raise young to the age of population recruitment.

Pregnancy-specific protein B (PSPB) is not 100% accurate; Duquette et al. [53] documented 5 cases where white-tailed deer were found to be pregnant by trans-abdominal ultrasound but were deemed nonpregnant by PSPB. However, overall they found strong agreement between the two methods and recommended using either depending on the study objectives. We feel comfortable that the PSPB was an appropriate test, but it is possible that we underestimated pregnancy rates and therefore overestimated λ₁, which already is extremely low for a white-tailed deer population. Considering the high pregnancy rates reported in this study, the impact on λ₁ from inaccurate tests likely was minimal.

The modeling exercise that determined λ₁ can be expected to be less than 1.0 (assuming other vital rates remain constant) at a prevalence of 27% suggests that as CWD in a population approaches these values, wildlife managers may choose to switch their objectives from lowering CWD prevalence by decreasing deer density to one of maintaining a sustainable population. The hunting-free Leslie matrix indicated removing additive hunting mortality in female deer resulted in a sustainable population. Therefore, it is recommended that at high CWD prevalence, hunting of does should be limited or ceased if the objective is to maintain population numbers. Currently this is a rare situation in most CWD endemic areas due to the relatively short period of time CWD has been present in most locations; this population should serve as an indication of what can happen at high prevalence when CWD has been endemic for an extended time period. Through time as prevalence rises in other endemic populations, more managers will be forced to make these choices if more effective management strategies or treatments are not developed. This recommendation is contingent on continued surveillance and monitoring of CWD in deer and elk populations in endemic areas as well as few or only minor public health concerns regarding CWD transmission to humans or livestock. Furthermore, if it becomes possible to accurately target and remove CWD-positive deer in a cost-effective manner, this management approach should be implemented in these populations where non-targeted culling is likely to be detrimental to population sustainability.

This population highlights the potential long-term negative outcome of endemic CWD to population sustainability and stresses the importance of preventing CWD from becoming endemic in a population, rather than attempting to manage it after the fact. Therefore, as previously suggested [43], the best management strategy remains minimizing movement of CWD to new areas.

Supporting Information

Acknowledgments

The authors thank personnel from the Wyoming Game and Fish Department (WGFD) and the Wyoming State Veterinary Laboratory (WSVL), Laramie, Wyoming, USA for assistance in the field and laboratory. In particular, thank you to G. Boyd, H. Edwards, D. Lutz, J. Binfet, S. Edberg, S. Talbott, C. Hansen, J. Seeman, B. Kroger, S. Dauwalter, K. Lutz, C. Havlik, D. Beach, S. Blajszak, L. Posusta, R. Lebert, B. Vandeberg, A. Kerr, J. Galles, J. Lund, L. Cope, C. Tate, and J. Jennings-Gaines from the WGFD. We thank E. Anderson, J. Herreman, C. Quitmeyer, B. Parrie, B. Bonner, J. Brown, A. Dougherty, B. Morris, L. Meadows, L. Woods, B. Howell, P. Jaeger, M. DeVivo, M. Thelen, M. Holloran, S. Slater, C. Anderson, H. Sawyer, S. Becker, C. Schmerge, N. Miller, M. MacGregor, A. Kaiser, A. Payne, C. Lane, J. Marquard, L. Plemmons, A. Roosa, K. Hamond, and W. MacDonald at the WSVL, University of Wyoming, and other entities. Veterinary externs to thank include T. Burgess, M. Nagendran, K. Troxell, K. Buckle, A. Cartoceti, L. Coffee, L. Last, K. Theis, E. Casey, and A. Hoffman. We also are indebted to landowners that provided access for our project. This manuscript was greatly improved by the helpful comments of R. Monello, three anonymous reviewers, and our Associate Editor. Funding was provided by Morris Animal Foundation, the U.S. Geological Survey, WGFD, White-tails Unlimited, and National Cattleman’s Beef Association. Use of any trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.