Assessment of spatial genetic structure to identify populations at risk for infection of an emerging epizootic disease

Abstract Understanding the geographic extent and connectivity of wildlife populations can provide important insights into the management of disease outbreaks but defining patterns of population structure is difficult for widely distributed species. Landscape genetic analyses are powerful methods for identifying cryptic structure and movement patterns that may be associated with spatial epizootic patterns in such cases. We characterized patterns of population substructure and connectivity using microsatellite genotypes from 2,222 white‐tailed deer (Odocoileus virginianus) in the Mid‐Atlantic region of the United States, a region where chronic wasting disease was first detected in 2009. The goal of this study was to evaluate the juxtaposition between population structure, landscape features that influence gene flow, and current disease management units. Clustering analyses identified four to five subpopulations in this region, the edges of which corresponded to ecophysiographic provinces. Subpopulations were further partitioned into 11 clusters with subtle (F ST ≤ 0.041), but significant genetic differentiation. Genetic differentiation was lower and migration rates were higher among neighboring genetic clusters, indicating an underlying genetic cline. Genetic discontinuities were associated with topographic barriers, however. Resistance surface modeling indicated that gene flow was diffuse in homogenous landscapes, but the direction and extent of gene flow were influenced by forest cover, traffic volume, and elevational relief in subregions heterogeneous for these landscape features. Chronic wasting disease primarily occurred among genetic clusters within a single subpopulation and along corridors of high landscape connectivity. These results may suggest a possible correlation between population substructure, landscape connectivity, and the occurrence of diseases for widespread species. Considering these factors may be useful in delineating effective management units, although only the largest features produced appreciable differences in subpopulation structure. Disease mitigation strategies implemented at the scale of ecophysiographic provinces are likely to be more effective than those implemented at finer scales.


| INTRODUC TI ON
Emerging wildlife diseases are being increasingly recognized as an important threat to the health of wildlife populations (Sutherland et al., 2018). Many species affected by emerging wildlife diseases have experienced or are predicted to experience population declines or extinctions (Edmunds et al., 2016;Scheele et al., 2019;Thogmartin et al., 2013). Due to the pervasive effects of these diseases, preventing the geographic spread of disease to naïve populations is often a priority of efforts focused on managing emerging wildlife diseases (Langwig et al., 2015). Therefore, understanding the factors that influence the distribution of diseases and predicting patterns of future occurrence have become important objectives for managing diseases. Determining the extent and distribution of wildlife populations can inform disease mitigation strategies because population structure is often correlated with the occurrence and prevalence of wildlife diseases (Blanchong et al., 2008;Cullingham, Kyle, Pond, Rees, & White, 2009).
For species characterized by fidelity to specific habitat patches, the extent of population boundaries and the distribution of associated diseases often correspond to discrete habitat edges. For example, the population structure of little brown bats (Myotis lucifugus) and distribution of white-nose syndrome are closely associated with winter hibernation colonies, movement among which is strongly influenced by local topography (Miller-Butterworth, Vonhof, Rosenstern, Turner, & Russell, 2014). Many species, however, are habitat generalists with widespread distributions. In cases where populations are continuously distributed on landscapes, spatial population structure often still exists, although subpopulation boundaries can be difficult to delineate (Vergara et al., 2015). A common practice among wildlife specialists is to define population or management units based on geophysical or political boundaries (Rosenberry & Diefenbach, 2019), which may not be reflective of the underlying disease risk for widespread species. Genetic clustering algorithms have become an important tool for defining population substructure and landscape features associated with genetic discontinuities in mobile and widespread species (Coulon et al., 2006). Algorithms that account for spatial autocorrelation of allele frequencies have been shown to be particularly useful for identifying cryptic subpopulation edges in continuously distributed populations (Safner, Miller, McRae, Fortin, & Manel, 2011). For example, Guillot (2008) was able to detect six spatially distinct subpopulations of wolverines (Gulo gulo) in a widespread population using a Bayesian clustering algorithm with a correlated allele frequency model where previous efforts based on nonspatial models were able to detect only three (Cegelski, Waits, & Anderson, 2003). By treating allele frequencies as correlated across related clusters as opposed to independently distributed, spatially explicit clustering is more likely to represent the true underlying structure of continuously distributed populations (Guillot, 2008).
Thus, clustering methods accounting for genetic autocorrelation are likely to have increased power to detect cryptic subpopulation structure and may outperform other clustering and edge detection methods in identifying features associated with genetic discontinuities (Safner et al., 2011;Vergara et al., 2015). While gene flow may be widespread in continuously distributed populations, disease prevalence rates can vary spatially even with minute deviations from genetic panmixia (Blanchong et al., 2008). Therefore, detection of cryptic genetic discontinuities using spatially explicit clustering methods may provide insights into the potential occurrence and distribution of wildlife diseases affecting widespread species.
The movement of infected individuals among different subpopulations may also influence epizootic patterns at a landscape scale . Dispersal may not occur uniformly and can be influenced by many biological and environmental factors including the permeability of the landscape matrix (Bowler & Benton, 2005). Landscape elements can alter transmission patterns by impeding or directing the movement of infected individuals. For example, the incidence of rabies is concomitant with the permeability of rivers to raccoon (Procyon lotor) gene flow, with certain rivers acting as barriers to dispersal and disease transmission (Cullingham et al., 2009). Connectivity analyses have also been demonstrated to be useful for predicting potential transmission corridors based on correlations between gene flow and landscape composition in areas where diseases are recently emergent (Paquette, Talbot, Garant, Mainguy, & Pelletier, 2014). Considering landscape connectivity jointly with patterns of subpopulation structure may improve efforts to mitigate the geographic diffusion of wildlife diseases at a landscape scale by identifying subpopulations that may be at risk for establishment and features that facilitate or impede the dispersal of potentially infected individuals, and by extension, transmission of disease among subpopulations.

| Chronic wasting disease
Chronic wasting disease is an emerging and fatal prion disease that affects ecologically and culturally important members of the Cervidae family (Miller & Williams, 2004), including free-ranging of ecophysiographic provinces are likely to be more effective than those implemented at finer scales.

K E Y W O R D S
chronic wasting disease, disease spread, gene flow, hierarchical genetic structure, landscape genetics, Odocoileus virginianus, white-tailed deer white-tailed deer (Odocoileus virginianus), mule deer (O. hemionus), elk (Cervus canadensis), and moose (Alces alces) populations in North America (Carlson et al., 2018). Of these species, white-tailed deer are often of particular interest since they are the most common and widely distributed cervid in North America (Heffelfinger, 2011), and because they are the primary vector and species affected in areas where chronic wasting disease is emerging, such as eastern and central North America. No effective treatment currently exists, so management strategies typically include surveillance and targeted herd reductions, with the goal of minimizing the geographic diffusion of the disease (Evans, Schuler, & Walter, 2014;. While active management strategies, such as targeted culling efforts, can lead to changes in local prevalence rates Mateus-Pinilla, Weng, Ruiz, Shelton, & Novakofski, 2013), chronic wasting disease has continued to spread geographically at broader scales in most disease foci.
Because chronic wasting disease is spread by direct interactions and indirect contacts through shared environments (Saunders, Bartelt-Hunt, & Bartz, 2012), population structure may influence the transmission of this disease across affected landscapes. Mismatches between the scale of management efforts, the extent of outbreaks, and subpopulation edges may contribute, in part, to the continued diffusion of this disease. White-tailed deer also maintain high rates of dispersal (Long, Diefenbach, Rosenberry, Wallingford, & Grund, 2005;Lutz, Diefenbach, & Rosenberry, 2015), so movement is also likely to influence transmission at broader scales in areas where chronic wasting disease is emerging as well. While barriers to movement are permeable, dispersal and gene flow patterns are altered by anthropogenic and topographic barriers (Blanchong et al., 2008;Long, Diefenbach, Wallingford, & Rosenberry, 2010;Lutz, Diefenbach, & Rosenberry, 2016;Robinson, Samuel, Lopez, & Shelton, 2012), which may in turn slow the geographic diffusion of chronic wasting disease (Hefley, Hooten, Russell, Walsh, & Powell, 2017). Evaluating the genetic structure of white-tailed deer and the extent of connectivity among delineated subpopulations may aid in predicting future epizootic patterns and improve disease mitigation efforts.

| Objectives
Here, we have evaluated the spatial population structure and genetic connectivity of white-tailed deer in a large area of the Mid-Atlantic region of the United States where chronic wasting disease is an emerging wildlife pathogen. We also assessed the relative resistance of landscape variables hypothesized to affect gene flow in order to identify potential disease transmission corridors. We hypothesized that patterns of population structure would be subtle and gene flow would be widespread. Despite this, we predicted that rivers, topography, the availability of forest cover, and highways with high volume traffic would modulate the extent and directionality of deer gene flow (Blanchong et al., 2008;Kelly et al., 2014;Locher, Scribner, Moore, Murphy, & Kanefsky, 2015;Long et al., 2010;Robinson et al., 2012). We also predicted that subpopulations would be arranged hierarchically because dispersal barriers were expected to be permeable to deer movement.

| Study area and sample collection
From 2013 to 2017, we collected tissue samples (n = 2,222), consisting primarily of connective or muscle tissue biopsies, from an area encompassing 82,000 km 2 , which included samples from Pennsylvania, Virginia, and Maryland ( Figure 1). The sampling region spans three ecophysiographic provinces (Piedmont, Ridge-and-Valley, and Appalachian Plateau) that are topographically distinct and separated by major topographic escarpments ( Figure 1). This region is heterogeneous for features predicted to influence the genetic structure of white-tailed deer and patterns of gene flow, including forest cover, topographic complexity, major highways, and rivers ( Figure S1). Samples were collected in conjunction with disease surveillance efforts of state agencies and included samples from hunter harvest, vehicle mortality, and targeted removal. Additional samples were collected from captured deer in northern and central Pennsylvania with protocols approved by The Pennsylvania State University (IACUC protocol 47,054). Locations were recorded as either the centroid of the municipal township, hunting management unit, or 2.59 km 2 sampling grid cell, or as explicit spatial coordinates, depending on the collection method and agency. Both male and female deer were included in the sample (males = 50.7%, females = 47.9%; unknown = 1.4%), since both sexes are capable of dispersing (Long et al., 2005;Lutz et al., 2015). Tissue samples were suspended in 95% ethanol and stored in a −20°C freezer until DNA extraction.
Locality data for chronic wasting disease cases detected from 2009 to 2017 were also recorded for comparison to patterns of population structure and gene flow. Chronic wasting disease was first recorded in the sampling region in northern Virginia in 2009, but cases were found as early as 2005 in an adjacent state (West Virginia). Since then, additional cases have been detected in free-ranging herds in western Maryland and central Pennsylvania, but region-wide prevalence rates in free-ranging populations are still estimated to be low (≤1%; Evans, Kirchgessner, Eyler, Ryan, & Walter, 2016;Evans et al., 2014). Disease management areas were established in response to the detection of chronic wasting disease in free-ranging populations in these states, and for surveillance in the proximity of infected cap-

| DNA extraction and microsatellite genotyping
We isolated DNA using the animal tissue protocol for the QIAGEN DNeasy blood and tissue extraction kits (QIAGEN). Tissue digestions were incubated for a minimum of 4 hr to ensure samples were completely lysed, and DNA elutions were carried out with a single 150 µl volume of elution buffer to maximize DNA concentration. We quantified the concentration of extracted DNA (ng/µl) using a NanoDrop spectrophotometer (Thermo Fisher Scientific) and diluted samples to approximately 20 ng/µl. All samples were genotyped using 11 microsatellite loci previously shown to be effective for genotyping in this region (Miller, Edson, Pietrandrea, Miller-Butterworth, & Walter, 2019). The resulting amplicons were analyzed on an Applied Biosystems genetic analyzer (model 3730 XL) at the Penn State Genomics Core Facility. One negative control (deonized H 2 O) was included on each plate in order to ensure amplicons were not contaminated by external sources of DNA and at least one previously genotyped sample was included as well to confirm reproducibility. We used GeneMarker (Softgenetics) to determine allele identity. Alleles were binned using the MsatAllele R package (version 1.5; Alberto, 2009), which can account for imperfect repeat motifs known to impact several of these loci (Miller et al., 2019).
We tested for the presence of genotyping errors, null alleles, and deviations from equilibrium assumptions to determine data quality (Appendix S1).

| Subpopulation structure
The Bayesian clustering method implemented in the Geneland R package (version 4.6; Guillot, Mortier, & Estoup, 2005) was used to identify the number of genetic clusters (K) and delineate population structure. Geneland was chosen as the basis for analysis because it is suggested to be better able to identify cryptic genetic discontinuities than alternative clustering algorithms (e.g., TESS, BAPS) and edge detection methods (e.g., Wombling) in scenarios where gene flow is widespread and dispersal barriers are permeable to movement (Safner et al., 2011), patterns which were predicted of white-tailed deer in this region based on previous movement studies (Long et al., 2005(Long et al., , 2010Lutz et al., 2015Lutz et al., , 2016. We estimated K using the correlated allele frequencies model and a spatial uncertainty term set to 7.071 km, which corresponded to the axial edge of a 50 km 2 square centered on the sample coordinates. We determined the number of genetic clusters by evaluating 20 levels of K using five independent runs with the following parameters: of 1,000 iterations. Cluster assignment was determined using the single run with the highest average log posterior density (Guillot, 2008).
Patterns of genetic diversity were summarized for each inferred genetic cluster using GenAlEx (version 6.5; Peakall & Smouse, 2006. Specifically, we calculated the average number of alleles per locus (N A ), observed heterozygosity (H O ), unbiased expected heterozygosity (H E ), and the number of private alleles (P A ) for each genetic cluster. Genetic connectivity among inferred genetic clusters was described using migration estimates from BayesAss (version 1.3; Wilson & Rannala, 2003). Migration rates were estimated using the following MCMC parameters: (a) a single chain of 21,000,000 iterations with a burn-in period of 2,000,000 and thinning every 2,000 steps, (b) a prior on the mixing parameter for allele frequencies of 0.2, (c) a prior on the mixing parameter for inbreeding coefficients of 0.2, and (d) a prior on the mixing parameter for migration rates of 0.05. The chosen prior values produced acceptance rates of proposed changes between 20% and 40%, which is within the suggested guidelines for adequate mixing (Wilson & Rannala, 2003). We ran ten independent replicates and selected the run that minimized the Bayesian deviance criterion for evaluation (Meirmans, 2014).
Pairwise estimates of genetic differentiation among genetic clusters were estimated using Weir and Cockerham's (1984) F ST estimator in FSTAT (version 2.9.3.2; Goudet, 1995). Weir and Cockerham's estimator, hereafter simply referred to as F ST , provides corrections for multiple loci with more than two alleles each (Weir & Cockerham, 1984), making it an appropriate choice for hypervariable loci like microsatellites. We used 5,500 permutations of the data to calculate p-values. We determined significant deviations from panmixia using a Holm-Bonferroni procedure to correct for multiple comparisons (Holm, 1979). We created a subpopulation dendrogram to evaluate hierarchical relationships among inferred genetic clusters using the single-linkage method and pairwise F ST values as a measure of genetic distance. An analysis of molecular variance was also used to determine the hierarchical partitioning of genetic variance at three scales: (a) among individuals within genetic clusters, (b) among genetic clusters within larger subpopulation units determined from the hierarchical clustering analysis, and (c) among all subpopulation units. Hierarchical F-statistics and covariance components were calculated using the poppr R package (version 2.7.1; Kamvar, Tabima, & Grünwald, 2014).
A spatial principal components analysis, carried out in the adegenet R package (version 2.1.1; Jombart, 2008), was used to identify genetic clusters independent of the Geneland analysis for comparison. The connectivity network was defined using a maximum distance of 30 km ("neighborhood by distance option"), which corresponds to a dispersal probability of <5% for white-tailed deer in this region (Diefenbach, Long, Rosenberry, Wallingford, & Smith, 2008).
The number of clusters and cluster membership was identified using two methods. We plotted the lag scores for the first two principal components to visually represent the distribution of local genetic variability. K-means clustering was also used to determine the most likely number of clusters based on lag scores. The optimum number of genetic partitions was chosen using the "elbow method" (Ketchen & Shook, 1996).

| Landscape genetics
We evaluated the relationship between genetic connectivity within and among genetic clusters using a resistance surface modeling approach in order to identify potential corridors of movement and disease transmission. Samples were regrouped using a 25 × 25 km sampling grid matched to the extent of each identified genetic cluster in order to maintain a more equal sampling distribution conducive to landscape genetic analyses, while at the same time preserving cluster assignments. While two genetic clustering methods were used, we chose to use the method that produced the finest-scale genetic partitions for landscape genetic analyses since these edges would be more reflective of subtle landscape barriers. Any grid that incorporated <20 samples was excluded from further analyses. Pairwise Survey). Rasters were scaled to a pixel size of 6 km 2 , which corresponds to the average home range size of white-tailed deer in this region (Evans et al., 2014).
Resistance values were chosen concurrently for all rasters using a genetic optimization algorithm implemented in the ResistanceGA R package (version 4.0-5; Peterman, 2018). A mutation rate of 0.125 and a crossover rate of 0.850 were used to generate resistance values for each iteration. Deer gene flow was modeled using random-walk commute distances for each realization of the resistance surface. We used the log-likelihood of a maximum-likelihood population effects model to evaluate the correlation between genetic distance and resistance distance at each iteration as the selection rule. The exploration operators included a maximum of 1,000 iterations, a convergence rule of termination after 50 iterations without improvement, and a joint maximum resistance of 5,000 per pixel. Since the optimization algorithm is a stochastic process, we produced five replicate runs of the optimization procedure and evaluated the model with the highest log-likelihood. Because white-tailed deer population structure is predicted to display an underlying pattern of isolation-by-distance (Kelly et al., 2014;Locher et al., 2015;Robinson et al., 2012), we also evaluated a distance-only model for comparison. Models were ranked based on Akaike's information criterion corrected for small sample size (AICc). Current density maps were also produced to display connectivity corridors using CIRCUITSCAPE (version 4.0;McRae, Dickson, Keitt, & Shah, 2008).

| Population structure
The Geneland algorithm identified 10-13 subpopulations across five independent runs. The value of K for the iteration with the highest  Table 1; Table S2). The proportion of resident individuals approached the lower bounds for this parameter for three populations (<0.73; Meirmans, 2014). High immigration rates from neighboring genetic clusters were causative in all cases (Table S2). Thus, migration rates were interpreted as relative measures of genetic connectivity rather than absolute estimates.
Genetic clusters exhibited a nested pattern using a hierarchical clustering analysis based on genetic distance among inferred clusters. Four hierarchical groups were identified (hereafter referred to as subpopulations, Figure 2). Demarcation of these larger subpopulations generally coincided with the boundaries of ecophysiographic provinces. Inferred migration and gene flow were greatest among genetic clusters in the central Ridge-and-Valley population (Table S2)

| Landscape genetics
Subpartitioning each of the 11 genetic clusters using a 25 × 25 km sampling grid resulted in 34 sampling units for landscape genetics analyses with sample sizes ranging from 20 to 324 (total sample size = 1,796; Figure S1). The Geneland clusters were chosen as a basis for this analysis because they represent the finest-scale TA B L E 1 Genetic summary statistics and sample sizes (N) for 11 white-tailed deer genetic clusters in the Mid-Atlantic region of the United States inferred from Geneland. Genetic cluster designations correspond to Figure 1. Measures of genetic diversity include allelic richness (N A ), observed heterozygosity (H O ), expected heterozygosity (H E ), and the number of private alleles (P A ). Recent immigration rates (m imm ) were used to define general patterns of genetic connectivity    (Figures 1 and 4). Forty-five deer positive for chronic F I G U R E 2 Hierarchical clustering of inferred white-tailed deer subpopulations in the Mid-Atlantic region of the United States. The length of each branch corresponds to the degree of genetic divergence (F ST ) among subpopulations wasting disease occurred within areas corresponding to the highest predicted probability of deer movement (95th to 100th current density quantiles; Figure 4b).

| Population structure and connectivity
White-tailed deer are common and widespread in the Mid-Atlantic region, but subtle patterns of population substructure were ob- where subpopulation edges were more closely associated with anthropogenic barriers, habitat configuration, and rivers (Blanchong et al., 2008;Kelly et al., 2014;Locher et al., 2015;Robinson et al., 2012). Features included in the landscape resistance model, including forest cover, topographical complexity, and roads with high traffic volume, have been previously demonstrated to influence the individual movement patterns of white-tailed deer in this region (Long et al., 2005(Long et al., , 2010Lutz et al., 2016). Our results further demonstrate that these landscape features also affect broader patterns of population connectivity. Despite the relationship between genetic discontinuities and topography, areas with <20% forest cover were the most resistant to white-tailed deer gene flow. Landscape resistance associated with reduced forest cover was highest in the southeastern portion of the study domain, an area with higher human densities and/or more widespread agricultural production relative to other areas. While less resistant to gene flow, high volume traffic roads were also concentrated in these same areas. Human development, agriculture, and roads have all been shown previously to reduce gene flow in anthropogenically modified landscapes (Blanchong, Sorin, & Scribner, 2013;Kelly et al., 2014;Locher et al., 2015;Robinson et al., 2012).Therefore, it is likely that human densities and anthropogenic development (e.g., impervious surface, highways) are combining to reduce deer movement across these open areas in an otherwise predominately forested landscape. The effects of topography on landscape resistance were also less than forest cover but more widely distributed across the region ( Figure S1).
Therefore, topography is likely having broader effects on regional subpopulation structure, while human barriers have more intense but localized effects.
Features contributing to landscape resistance were responsible for directing patterns of gene flow (Figure 4b). Areas with the highest probability of deer gene flow occurred along forested corridors adjacent to areas of greater elevational relief, roads with high traffic volumes, and/or open areas. Previous studies also demonstrate that landscape barriers change the direction of individual deer movement (Long et al., 2010;Lutz et al., 2016), and our results highlight that landscape barriers can affect the directionality of regional connectivity patterns. Although large streams had very low resistance values, gene flow pathways occurred proximally to streams. Streams While population connectivity was influenced by landscape features, genetic differentiation was low at all hierarchical scales.
While deer commonly terminate dispersal near topographic and anthropogenic barriers, radiomarked deer have been documented crossing them (Long et al., 2010;Lutz et al., 2016). Therefore, our results support the hypothesis and expand on the scale of assessment that landscape features are not barriers to movement but decrease dispersal to produce subtle changes to populations structure.

| Chronic wasting disease management
Chronic wasting disease management zones are often defined by a combination of previously defined wildlife management units, political boundaries, and/or inferred dispersal barriers. Since the diffusion of chronic wasting disease can be facilitated by animal F I G U R E 5 Relationship between pixel values of the raw landscape covariate rasters and the optimized resistance values included in the composite resistance raster used to infer patterns of gene flow for white-tailed deer. (a) forest cover, (b) highway traffic volume, and (c) topographic heterogeneity movement , understanding of population substructure and dispersal behaviors may improve disease mitigation strategies when compared to undirected management efforts (Blanchong et al., 2008). Even those strategies that incorporate landscape features associated with potential population discontinuities may be ineffective if the permeability of such barriers is not addressed. For example, a common practice of regional disease management strategies is to define surveillance areas in part by state roads, which had little influence on subpopulation structure and gene flow here and in a previous study (Robinson et al., 2012).
Defining future disease management efforts by features associated with genetic discontinuities at a regional scale may be more effective. In the Mid-Atlantic region, this corresponds to major geophysical escarpments concomitant with ecophysiographic boundaries and the extent of hierarchical subpopulation units and large, deforested areas associated with human development.
Based on our results, genetic clusters or subpopulation units that are separated by such features in this region are hypothesized to be less at risk for dispersal-mediated disease transmission relative to those that maintain higher rates of connectivity. Differences in landscape features associated with subpopulation structure described here when compared to other studies suggest that landscape-scale correlations are likely to be context specific, however (Blanchong et al., 2008;Kelly et al., 2014;Robinson et al., 2012). While defining population substructure can aid in defining the extent of disease management efforts, landscape connectivity models may help to identify potential transmission corridors between infected and naïve groups (Paquette et al., 2014). Many deer with chronic wasting disease occurred within corridors with high predicted probability of movement ( Figure 4b). This could suggest that landscape characteristics dictating deer movement may be correlated with the transmission and occurrence of chronic wasting disease in this region. Therefore, landscape resistance models may improve efforts to forecast the continued spread of chronic wasting disease in this and other regions. Gene flow occurred along pathways paralleling resistant features in heterogeneous landscapes in the Mid-Atlantic region. We predict that chronic wasting disease may be more likely to spread along these connectivity corridors, given previous relationships between white-tailed deer movement and chronic wasting disease spread . Disease diffusion models also suggest the axial spread of chronic wasting disease along landscape features resistant to deer movement in other regions (Hefley et al., 2017). The proximity of connectivity corridors to forested riparian cover also supports previous predictions that these features are likely to play an important role in deer movement and may affect transmission dynamics of chronic wasting disease in heterogeneous environments (Nobert, Merrill, Pybus, Bollinger, & Hwang, 2016;Walter et al., 2011). Based on our results, we also predict that disease transmission may be more diffusive in homogeneous landscapes, such as those in the Allegheny Plateau, due to the undirected nature of deer movement.

| Future directions
Our results help to elucidate patterns of spatial substructure and genetic connectivity in a continuously distributed population of white-tailed deer where chronic wasting disease is emerging.
Population-level trends may help to describe the relative risk of transmission and provide important insights into future epizootic patterns. Genetic samples from deer with chronic wasting disease were limited in this current study, however, and we did not have access to spatial records of chronic wasting disease from neighboring states with active infection, such as West Virginia. Continued genetic sampling and disease surveillance will clarify genetic cluster membership in areas where samples were currently unavailable and improve efforts to assess the association between subpopulation structure and chronic wasting disease occurrence. Integrating measures of population substructure and connectivity into diffusion models, such as those used in Hefley et al. (2017) to predict the temporal spread of chronic wasting disease, represent a critical extension of the current study for validating the inferred correlation between genetic connectivity, landscape resistance, and transmission risk. Diffusion models will also allow for the quantitative assessment of chronic wasting disease transmission at a landscape scale.
Individual-based genetic analyses, such as ancestry and assignment analyses, will also allow for testing specific hypotheses regarding the origin of chronic wasting disease cases. For example, Miller and Walter (In press) used simulated reference clusters based on empirical genotypes from wild and captive deer in order to evaluate the influence of captive egression on chronic wasting disease occurrence in free-ranging populations. Similar analyses would likely be useful in further determining the role dispersal plays in patterns of chronic wasting disease epizootiology. Describing the number of genetic clusters and degree of population connectivity was an important first step in optimizing individual-based admixture and assignment analyses.

ACK N OWLED G M ENTS
We wish to acknowledge J. Edson and P. Pietrandrea for their assistance in the laboratory. We also wish to acknowledge the numer- Neither agency was involved in the analysis or interpretation of data, or in the preparation of this paper. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

CO N FLI C T O F I NTE R E S T
None declared. Writing-review & editing (equal).

DATA AVA I L A B I L I T Y S TAT E M E N T
Microsatellite genotypes, locality information, and raster data will be available on Zenodo following a 1-year embargo period from the date of publication (https://doi.org/10.5281/zenodo.3675373).