Pathogens can regulate or decimate free-ranging wildlife populations. Wild turkeys (Meleagris gallopavo), which are widespread across the United States, southern Canada, and northern and central Mexico, are a prized upland gamebird that has experienced dramatic population growth and range expansion as the result of reintroduction campaigns. While increased abundance may promote disease transmission, little is known about the effects of pathogen infections on demographic metrics in wild turkeys. Lymphoproliferative disease virus (LPDV) and reticuloendotheliosis virus (REV) are oncogenic retroviruses that infect poultry and wild turkeys and can result in disease and mortality, though most infected individuals appear asymptomatic. We investigated whether retroviral infections influence wild turkey fitness by evaluating effects on female survival and several reproduction metrics. We live-captured 163 female wild turkeys throughout central Maine, USA, during three winters, from 2018–2020. We collected blood for LPDV and REV molecular diagnostics and attached a GPS or VHF transmitter to monitor survival and nesting. Infection with REV was associated with nearly half the cumulative annual survival probability, while LPDV-infected hens laid an average of 1.4 fewer eggs per clutch. We detected no effects of retroviral infection on nest initiation, nesting propensity, or hatch rate, and coinfection was not associated with any measured demographic metric. These findings demonstrate that retroviral infections can negatively affect survival and clutch size in female wild turkeys even in the absence of overt disease, highlighting the importance of considering pathogen effects when evaluating the population dynamics of free-ranging wildlife.

Field methods

We captured 163 live female turkeys over three winter seasons (Jan–Mar) using rocket or drop nets from 29 capture sites, located mostly in central and southern Maine (Figure 1). For all captured birds, we recorded the year of capture, body mass, and determined sex. We also determined age by observing primary wing feathers, which are consistent in length with the middle feathers of the tail fan for adults but are not uniform in length compared to the middle feathers of the tail fan for sub-adults (less than one year old; Dickson 1992). Each captured wild turkey was fitted with one of three unique transmitter models: (1) an 80g VHF backpack-style harness transmitter (n = 91; Advanced Telemetry Systems, Isanti, Minnesota), (2) a 90g GPS backpack-style harness transmitter with a built in VHF component (n = 46; Lotek Wireless Fish and Wildlife Monitoring, Newmarket, Ontario, CA), or (3) a 12g VHF necklace transmitter (n = 26; Advanced telemetry Systems, Isanti, Minnesota). For molecular diagnostics of LPDV and REV, whole blood was drawn from the brachial vein into an EDTA tube (1–5 mL; n =129) or from a foot venipuncture into a heparin-treated capillary tube and stored in queen’s lysis buffer (~1mL, n = 34). All capture, handling, and sampling of wild turkeys was approved by the University of Maine Institutional Animal Care and Use Committee (IACUC Protocol # A2017_11_03).

LPDV and REV molecular diagnostics

We used a molecular approach to determine the LPDV and REV proviral infection status of all sampled individuals. From the majority of blood samples (n = 127), we isolated the buffy coat layer by centrifuging for 15 minutes at 2500 RPM. In some cases (n = 36), when blood volume was too low for buffy coat optimization or when blood was collected via capillary tubes, we used whole blood. We extracted genomic DNA from both buffy coat and whole blood using Qiagen DNeasy Blood and Tissue Kits (Qiagen, Valencia, CA), following the manufacturer’s instructions. For each extraction, we included a negative control and quantified DNA concentration using a NanoDrop One Spectrophotometer (Thermo Fisher Scientific, Wilmington, DE) or Qubit Fluorometer (Thermo Fisher Scientific, Waltham, MA). We determined retroviral infection status by PCR amplification of both a 413 base pair region of the LPDV gag gene and a 580 base pair region of the REV pol gene, following protocols described in Shea et al. (2022).

Monitoring survival and reproduction

We used both GPS and VHF transmitters to monitor the survival of 163 individuals and 111 nests. We tracked all available individuals from their respective date of capture (first capture occurred on February 3^rd 2018) through the end of the study (November 14^th 2020). For VHF-marked birds, we attempted to locate birds approximately once a week to record locations using a hand-held three-element directional antenna, where the live/dead status of each bird was determined based on the speed of the transmitter signal. We assessed the survival of GPS-marked individuals according to location information, with the date of mortality inferred according to sequential points at a single location. We recorded locations every hour during daylight (shifted periodically) from November through July, with an additional overnight location to record roosting sites. We downloaded data directly from transmitters regularly and uploaded information to Movebank (Kays et al. 2022). If either a GPS- or VHF-tagged bird was suspected dead, the transmitter was approached to confirm mortality status. All birds were monitored with increased frequency for two weeks immediately following capture to assess the potential for capture-related mortality, and birds that died during this time were censored.

Female wild turkeys were monitored from April 15^th to July 30^th, each year of the study, for suspected nesting behavior. Particularly, we investigated the following metrics to evaluate the effects of pathogen infection on individual fitness: weekly survival rate, daily nest survival rate (DNSR), clutch size (number of eggs laid), nest initiation (Julian day of first egg laid), nesting propensity (rate at which a female nested if she was available to do so), and hatch rate proportion of eggs available that hatched). We monitored tagged hens for suspected nesting behavior from April 15 to July 30. Locations of VHF-marked individuals were collected at least twice a week via short-distance triangulation. If a hen was found alive in the same location during two successive visits, she was assumed to be on a nest. After 2 weeks, we approached the hen’s location and flushed her to confirm nesting and locate the nest. We delayed flushing by this amount of time to decrease the chance of nest abandonment, which may be greatest when females are flushed during egg‐laying or early incubation (Götmark 1992). Nonetheless, nest abandonment occurred on eight occasions following flushing (further detailed in Gonnerman et al. 2022). We then floated 3-4 eggs to determine the incubation stage, estimate the initiation date of the nest (Westerkov 1950), and to predict a hatch date. Counting eggs during incubation reduced the likelihood of predation and underestimates. Furthermore, we continued to monitor the nest at least once a week, with a goal of 3 visits per week when possible. We increased visits around the suspected hatch date to better determine the actual hatch date. Once a hen was suspected to have left the nest, we approached the nest to assess its fate as hatched or failed.

Location data from GPS-marked hens was downloaded weekly, and point locations were reviewed in Google Earth. If we observed that a hen was making repeated visits to a single location around the same time of day, or had settled in a location she had previously visited regularly, we assumed she was nesting. Once the hen began regular movements or discontinued regular daily visits in the case of failure during the laying phase, we visited the suspected nest site to verify the nest and its fate.

Encounter history

We compiled weekly status (live/dead) for each GPS- and VHF-marked wild turkey to develop a weekly encounter history for an individual, which included the week the turkey was captured, the last week it was found alive, the last week it was checked, and its final status at the end of the monitoring period. We increased monitoring frequency during the nesting season and similarly created an encounter history for DNSR. A sub-adult at capture remained a sub-adult through its first nesting season and was considered an adult starting August 1^st of the year of capture, to differentiate first-time breeders.

Demographic statistical analyses

We evaluated the relationships between proviral infection status (REV, LPDV, coinfection), as independent variables, and several fitness metrics, including weekly survival rate, DNSR, clutch size for first and second nesting attempts, nest initiation for first and second nesting attempts, hatch rate, and nesting propensity for first and second nesting attempts. All analyses were conducted in RStudio (RStudio 2021) using Program R (R Core Team 2021), and we used the AICccmodavg package (Mazerolle 2020) to employ a tiered AICc model selection approach. First, we considered non-pathogen factors that could explain variation in fitness metrics appropriate to each specific analysis; this included season, turkey age, and transmitter type when evaluating effects on survival, and turkey age, transmitter type, nest age, nest attempt, nest initiation, and/or nest year depending on the response variable (DNSR, clutch size, nest initiation, hatch rate, or nesting propensity), when evaluating effects on reproduction. While we measured turkey weight at the time of capture, this was not included in analyses, given that weight at capture is likely not relevant in subsequent years for survival or reproduction, and given our model structure. If there was support (<2 delta AICc) for any models containing non-pathogen variables, the variables were included in a baseline model (only non-pathogen variables) and all other pathogen models. For the second AICc model selection (i.e, the pathogen model), we evaluated the baseline and pathogen models against an intercept-only null model. For pathogen models, we included LPDV or REV infection status as a binary variable (0 = PCR negative, 1 = PCR positive), and a 4-level coinfection categorical variable (represented as “coinf” in supplemental materials) with a level for each of the following: uninfected, infected with LPDV, infected with REV, infected with both LPDV and REV. We also included an age model when sample size allowed (with an age variable added to the baseline null model) because demographic estimates have been previously demonstrated to vary based on age in wild turkeys (Lehman et al. 2008; Pollentier et al. 2014). Lastly, age was also considered as a predictor in two independent models per pathogen variable, one that included an additive age term and one that included an age interaction term with each pathogen variable because our previous work revealed that adults are more likely to be infected than sub-adults (Shea et al. 2022), and age-specific variation in pathogen effects can result in disproportional consequences in population growth.

Burnham and Anderson (2003) advised against using multiple notations of significance (e.g., reporting p values when using AICc model selection); therefore, we interpreted the significance of variables contained in supported models of the second AICc model selection by evaluating coefficients and their 95% confidence intervals (significance = confidence interval not overlapping zero). In linear models, AIC comparisons also account for an additional parameter estimating residual variance.

Evaluating pathogen effects on the weekly survival rate

We modeled weekly survival probability for females using the nest survival model in the RMark package (Laake 2013) in the program R (R Core Team 2021). This known-fate model type is commonly used for survival analyses for radio-marked individuals and has been applied to turkeys in this manner previously (Collier et al. 2009), among other species. We chose this approach because it allowed for irregular monitoring of individuals, which best fit our study design by accommodating both staggered entry and exit of individuals continually throughout the study. Furthermore, we generally followed turkeys to death or the end of the study. An incomplete history likely represents a bird which either left the study area or whose transmitter stopped working. We flew planes with telemetry searching well beyond our study area to ensure it was not the former, and truncated histories if we subsequently and reasonably assumed the latter. We exponentiated the weekly survival rate across 52 weeks to obtain an annual survival probability. For the first AICc model selection, we evaluated turkey age, season (winter = Jan–Mar, spring = Apr–Jun, summer = Jul–Sep, and fall = Oct–Dec), and transmitter type (as either backpack-harness style where we combined GPS and VHF models, or necklace style) as predictors of survival, comparing these models against an intercept only null model. Since the model containing season better predicted survival than the null model, and Shea et al. (2022) also found evidence that LPDV infection varied seasonally, we hypothesized there might also be an interaction effect between season and infection status on survival. Therefore, for the second (pathogen) AICc model selection, in addition to comparing the age models specified above (interaction and additive term with each pathogen variable), we also included three additional season models, each with an interaction term between season and one of the pathogen variables (LPDV, REV, coinfection).

Evaluating pathogen effects on reproduction

We modeled DNSR using the nest survival model in the RMark package (Laake 2013). The variables included in the initial non-pathogen AICc model selection were turkey age (at nesting), nest age (days), nest attempt (first or second), Julian day of nest initiation, and nest year. We also evaluated transmitter type since, when gathering nesting data, transmitter types (GPS vs. VHF) were expected to result in different levels of disturbance to hens. Lastly, we calculated overall nest success, defined as the percentage of initiated nests that survived to hatching.

We used AICc model selection to determine the most appropriate models for describing the impacts of pathogen infection on clutch size (number of eggs laid) and nest initiation (Julian day of first egg laid). Clutch size reportedly varies based on nest attempt (Roberts et al. 1995); thus, we conducted preliminary linear regression analyses and confirmed that nest attempt was a significant predictor of clutch size in our study. Nest initiation inherently varies based on nest attempt, as second nesting attempts chronologically follow first nest attempts. Therefore, we subset our data to analyze the first and second nest attempts separately for clutch size and nest initiation (the third nest attempt was excluded from the analyses due to a sample size of one). The variables included in the initial non-pathogen AICc linear model selection for both clutch size and nest initiation were turkey age (at nesting) and year (of nesting). For clutch size, we also included two models containing either the Julian day of nest initiation or the Julian day quadratic term. Clutch size (count data) was evaluated for residual normality. Residuals met the assumption of normality based on the Shapiro–Wilk test (p > 0.05) and inspection of Normal Q–Q Plots. For nest initiation of the second nesting attempt, the 4-level coinfection categorical variable was not assessed in the second (pathogen) model selection due to small sample size. To further validate the clutch size results and assess the biological relevance, we used a simulation model to evaluate how recruitment would respond to a single loss in egg production by allowing clutch size to vary based on the middle 50% of our clutch sizes and allowing recruitment rate to vary (Table S1).

We examined the relationship between pathogen infection status and nesting propensity, defined as the rate at which a female nested if she was available to do so, for both first and second nesting attempts. We used AIC generalized linear model selection with a binomial distribution to determine if hen age or nest year affected nesting propensity (0 = did not nest, 1 = nested) for the initial non-pathogen model. To determine nesting propensity for first and second nests, we excluded individuals with VHF transmitters due to the higher potential for missed nests. We included any hen that was alive (available to nest) on the average Julian day of nest initiation, specifically for each year and nesting attempt. For second nests, individuals were considered available to nest only if they had failed their previous attempt and were alive on the estimated average Julian day of nest initiation for that attempt in a given year.  When analyzing nesting propensity for the first nesting attempt, the 4-level coinfection categorical variable was not assessed in the second (pathogen) model selection due to small sample size. Similarly, only univariate pathogen models were assessed in the second (pathogen) model selection for the second nest attempt due to limited sample size.

Lastly, we assessed the effect of pathogen infection on the hatch rate of successful nests for both VHF- and GPS-marked birds. Hatch rate is defined as the ratio of the number of eggs that hatched out of the total number of eggs available to hatch (clutch size). We included hen age, nest year, nest initiation (two models containing either Julian day of nest initiation or the Julian day quadratic term), and nest attempt in the initial (non-pathogen) linear model selection. Due to the small sample size, we only assessed univariate pathogen variables as an additive variable in the second model selection.