Habitat predictors of genetic diversity for two sympatric wetland‐breeding amphibian species

Abstract Population genetic diversity is widely accepted as important to the conservation and management of wildlife. However, habitat features may differentially affect evolutionary processes that facilitate population genetic diversity among sympatric species. We measured genetic diversity for two pond‐breeding amphibian species (Dwarf salamanders, Eurycea quadridigitata; and Southern Leopard frogs, Lithobates sphenocephalus) to understand how habitat characteristics and spatial scale affect genetic diversity across a landscape. Samples were collected from wetlands on a longleaf pine reserve in Georgia. We genotyped microsatellite loci for both species to assess population structures and determine which habitat features were most closely associated with observed heterozygosity and rarefied allelic richness. Both species exhibited significant population genetic structure; however, structure in Southern Leopard frogs was driven primarily by one outlier site. Dwarf salamander allelic richness was greater at sites with less surrounding road area within 0.5 km and more wetland area within 1.0 and 2.5 km, and heterozygosity was greater at sites with more wetland area within 0.5 km. In contrast, neither measure of Southern Leopard frog genetic diversity was associated with any habitat features at any scale we evaluated. Genetic diversity in the Dwarf salamander was strongly associated with land cover variables up to 2.5 km away from breeding wetlands, and/or results suggest that minimizing roads in wetland buffers may be beneficial to the maintenance of population genetic diversity. This study suggests that patterns of genetic differentiation and genetic diversity have associations with different habitat features across different spatial scales for two syntopic pond‐breeding amphibian species.

in the population. Effective maintenance of genetically diverse populations requires understanding the evolutionary processes responsible for determining gains or losses of genetic diversity. Over an ecological time frame, genetic diversity in populations is gained by gene flow from other populations and lost through genetic drift (Cleary, Fauvelot, Genner, Menken, & Mooers, 2006;Vellend, 2005).
Although natural selection is also a mechanism of evolution, the effect of selection may be difficult to predict (Vellend & Geber, 2005). Therefore, management efforts to maintain or increase genetic diversity in populations should focus on maximizing gene flow by maximizing potential for dispersal and minimizing genetic drift by maximizing effective population sizes. Gene flow is affected by a number of factors, including species' life history traits, vagility, and habitat restrictions (Manel, Schwartz, Luikart, & Taberlet, 2003;Slatkin, 1987;Storfer et al., 2006). For a given species, some habitats may be more suitable for dispersal thereby facilitating gene flow, whereas other habitats may be less suitable and restrict gene flow (Cushman, 2006;Manel et al., 2003).
Few studies have compared landscape genetics of multiple species within the same landscape (Waits, Cushman, & Spear, 2016).
Investigating associations between habitat features and genetic diversity for multiple species within a meta-community may provide insight on the variability in these associations between sympatric species. This study addressed the following objectives for two sympatric, pond-breeding amphibian species that occur in the southeastern United States: assess population structures, identify the habitat features most strongly associated with genetic diversity, and examine the spatial scale at which habitat features are most strongly associated with genetic diversity.
Amphibian species that breed in isolated wetlands are appropriate focal organisms for studying patterns of genetic diversity because many occur as metapopulations due to their relatively limited vagility (Blaustein, Wake, & Sousa, 1994;Gibbs, 1998), high philopatry, and fidelity to breeding sites (see Smith & Green, 2005 for review), and because wetlands are relatively small discrete entities embedded in large matrices of terrestrial habitats (Marsh & Trenham, 2001;Smith & Green, 2005).
Dwarf salamanders (Eurycea quadridigitata; Figure 1a) and Southern Leopard frogs (Lithobates sphenocephalus, formerly Rana sphenocephala; Figure 1b) were selected as model organisms for this study. Both species are widespread and abundant in the southeastern United States (Cash, 2008;Means, 2008), and as adults, both species are strongly associated with wetlands and wetland edges during breeding and nonbreeding seasons. Although they both utilize aquatic habitats for mating and larval development and semi-terrestrial habitats as adults, they differ in vagility and microhabitat requirements.
Dwarf salamanders have limited dispersal ability (Pechmann, Estes, Scott, & Gibbons, 2001) and more specialized microhabitat requirements (Mount, 1975;Petranka, 1998 highly permeable and moist skin, making them extremely vulnerable to dehydration and restricting their activity to brief periods proximate to rain events (Feder, 1983). The larval stage of Dwarf salamanders is 5-6.5 months (Semlitsch, 1980). In Baker County, Georgia, Dwarf salamanders have been found in cypress-gum swamps and grasssedge marsh wetlands, which have longer hydroperiods than cypresssavanna wetlands where Dwarf salamanders were not detected (Liner, 2006). In contrast, Southern Leopard frogs have a greater dispersal ability (Smith & Green, 2005) and breed in a wide variety of wetland types (Liner, 2006). The larval stage of Southern Leopard frogs is around 3 months (Ashton & Ashton, 1988), approximately half that of Dwarf salamanders. Southern Leopard frogs are mediumsized anurans (adults are generally 50-130 mm SVL), and juveniles and adults have lungs and powerful legs. The lower surface area to volume ratio of a larger-bodied species reduces water loss, enabling them to be active for longer periods and in drier conditions compared to Dwarf salamanders (Lindstedt & Boyce, 1985). Southern Leopard frog tadpoles are less susceptible to predation by native fish species than other wetland amphibians (Gregoire & Gunzburger, 2008; but see Werschkul & Christensen, 1977), enabling them to breed in sites with predatory fish (Babbitt, Baber, & Brandt, 2006;Baber, 2001).
Because of differences in dispersal ability and habitat tolerances, Dwarf salamanders were expected to have greater population structure than Southern Leopard frogs (i.e., more discrete populations as a result of lower dispersal rates among wetlands), and Dwarf salamander genetic diversity was expected to be more closely associated with habitat features at smaller spatial scales relative to Southern Leopard frogs (Antonovics, 1976).
F I G U R E 2 Location and landcover (National Land Cover Data, 30-m pixels;Homer et al., 2004) at the Joseph W. Jones Ecological Research Center at Ichauway in Baker, Co., Georgia and the surrounding area, and spatial distribution of sample locations for Dwarf salamanders (Eurycea quadridigitata; yellow circles), Southern Leopard frogs (Lithobates sphenocephalus; purple circles), and both species (yellow and purple circles). The orange star indicates wetland PSK. The black outline is the Ichauway property boundary; blue areas are wetland habitat; green areas are forested upland habitat; beige areas are agricultural land, and red areas are developed land (primarily roads)  Figure 2).

| Microsatellite marker development and analysis
DNA was extracted from Southern Leopard frog tissue using silicabinding techniques (Lance et al., 2009) and from Dwarf salamander tissue using phenol chloroform extractions (Sambrook, Fritsch, & Maniatis, 1989 were rerun and analyzed at each locus to estimate genotyping error rates (Table S1).
It was assumed based on the philopatric tendencies of both species (for review see Blaustein et al., 1994;Smith & Green, 2005) that the genetic compositions of breeding assemblages were similar between 2008 and 2009. Sibship among individuals was estimated for each population, in Colony v2.0 (Jones & Wang, 2010). Larval samples are often biased toward particular families as they contain genetic material from successful breeders. Goldberg and Waits (2010b) found that when full siblings were collected from a population, removing all but one individual from analysis led to population and landscape genetic parameter estimates that were more similar to those calculated from adult samples. Therefore, when samples had a probability of full sibship >90%, we removed the individual with the less complete genotype (Tables S1 and S2).

| Population genetic diversity and differentiation
For each population, the genetic diversity parameter rarefied allelic richness (r g ) was calculated in Fstat v2.9.3.2, and H o and F were calculated in Genalex v6.503. For both species, one of the sites (PSK) had significantly lower r g compared to the other sites (  (Raymond & Rousset, 1995;Rousset, 2008). Populations that demonstrate diffusive stepping-stone model dispersal patterns should exhibit strong isolation-by-distance (IBD) population structure (Rousset, 1997

| Habitat and land cover characterization
Habitat and land cover variables were characterized for each sample site. arCMap 9 (ESRI 2009) was used to create circle buffers around the center of each wetland (Piha et al., 2007). Radii sizes were based on approximate spatial scales found to be significantly associated with amphibian diversity and abundance in previous studies ( forest, forest; agriculture, ag; and wetlands, wtlnd) within the buffers.
Given the rural location of the study site, devel was a general indicator of road density. Genetic diversity parameters for both focal species were expected to be negatively associated with devel, which may be partial barriers to dispersal and a source of mortality (Carr & Fahrig, 2001;Gibbs, 1998;Vos & Chardon, 1998). Forest is necessary for upland habitat and dispersal in many other pond-breeding amphibian species (Guerry & Hunter, 2002;Trenham & Shaffer, 2005), and therefore, genetic diversity parameters for both species were expected to be positively associated with forest cover. Genetic diversity parameters were expected to be negatively associated with ag, as agricultural landscapes may be partial barriers to amphibian dispersal because of the potential for water loss (Rothermel & Semlitsch, 2002).
Both focal species are generally associated with aquatic habitats yearround (Bonett & Chippindale, 2011;Cash, 2008;Means, 2008) and breed in isolated wetlands, and were therefore expected to have genetic diversity parameters positively associated with wtlnd. Percent area variables were arcsine square root transformed to meet assumptions of statistical distribution normality. Landscape variables at specific spatial scales are from here on referred to by a subscript (e.g., devel 1.0 km refers to development within the 1.0 km buffer).
Local habitat variables believed to be of biological relevance to both species were wetland area, isolation, and hydroperiod (area, iso, and hydro, respectively). Neutral genetic theory suggests that populations at larger wetlands should be larger and more genetically diverse as a result of greater carrying capacity (Antonovics, 1976).
Neutral genetic theory also suggests that populations that are less isolated should be larger and more genetically diverse as a result of greater immigration rates than populations that are more isolated (Antonovics, 1976 (Bonett & Chippindale, 2011;Means, 2008). However, fish may be predators of Dwarf salamander larvae (Liner, 2006) Snodgrass et al. (2000, and Dwarf salamanders may therefore occur more often in wetlands with hydroperiods that are prohibitively short for establishment of fish populations. Southern Leopard frog larvae are unpalatable to local fish species and therefore may thrive in wetlands with longer hydroperiods (Babbitt et al., 2006;Baber, 2001). With the exception of PSK, a hardwood depression, all Dwarf salamanders were collected from cypressgum swamps, whereas Southern Leopard frogs were collected from all three wetland types. However, there was no difference in Southern Leopard frog allelic richness or heterozygosity among wetland types ( Figure S1). Therefore, wetland type was not included as a predictor around Ichauway (Kirkman et al., 2012). Hydroperiod was calculated as the average number of days over a calendar year that a wetland was at least 25% full (Kirkman et al., 2012). All predictor variables were tested for statistical normality with the Shapiro-Wilk normality test in r v3.2.3 (Table S4).

| Model selection and model averaging
We used multiple linear regression and an information theoretic approach to model genetic diversity parameters as a function of habitat features. The information theoretic approach has become more common in landscape genetic studies to investigate relationships be- value corrected for small sample size (AICc; Burnham & Anderson, 2002). As many predictor variables were highly correlated (R 2 > .5), we used condition number (CN) to determine the degree to which multicollinearity was an issue within models (Lazaridis, 2007 accordance with Burnham and Anderson (2004). Predictor variables in the top and composite models were considered statistically significant when the 95% confidence intervals did not cross zero.
Top models of genetic diversity for the Dwarf salamander were the same for analyses with and without PSK; however, results differed for analyses with and without PSK for the Southern Leopard frog (Table 2 and Table S5). Composite models from model averaging differed for both genetic diversity parameters for both species between analyses with and without PSK (Table 3 and Table S6). Additionally, patterns of spatial autocorrelation often differed between analyses with and without PSK, with inclusion of PSK resulting in stronger spatial autocorrelation for a number of predictor variables ( Figure S2). Model selection and averaging results are therefore presented and discussed for analyses without PSK, and results from analyses with PSK are available in Tables S5 and S6.

| RESULTS
After removing full siblings from the analysis, the Dwarf salamander sample size ranged from 27 to 31 per wetland and the Southern Leopard frog sample size ranged from 15 to 30 (Table 1). Full siblings were removed to obtain genetic parameter estimates that were not biased toward particular families (Goldberg & Waits, 2010b). The number of alleles per locus ranged from 3 to 27 for the Dwarf salamander, and 7-29 for the Southern Leopard frog (Table S3); and among loci, the mean number of alleles across populations ranged from 2.7 to 12.9 for the Dwarf salamanders, and 4.9-16.5 for Southern Leopard frogs (Table S3). After Bonferroni corrections (Weir, 1990), there was no evidence of linkage disequilibrium for Dwarf salamander loci and the possibility of slight linkage disequilibrium between the two Southern Leopard frog loci, or inbreeding or null alleles. The high number of loci out of HWE for Dwarf salamanders (5 of 12 loci, Table S3) and Southern Leopard frogs (8 of 11 loci, Table S3) suggested null alleles or inbreeding, as indicated by the positive F values (Table S3). All loci had null allele frequency estimates greater than zero in at least two populations (Table S3). Null alleles can artificially reduce r g and H o .  Figure S2.

| Genetic diversity parameters
T A B L E 3 Model-averaged estimates of local-and landscape-scale predictor variables of allelic richness (r g ) and observed heterozygosity (H o ) in the Dwarf salamander (Eurycea quadridigitata) and the Southern Leopard frog (Lithobates sphenocephalus) for populations from Ichauway, located in southwest Georgia, USA. Estimates were calculated based on models in the confidence set (i.e., all models with a ΔAICc ≤ 2) Both species exhibited IBD, indicating that populations closer in proximity to each other were more closely related each other than those further apart. The strong correlation between genetic and geographic distance for Southern Leopard frogs was driven by PSK, as the R 2 value decreased from .797 (p = .005) to .217 (p = .041) after PSK was removed ( Figure 3). In contrast, the strength of Dwarf salamander IBD correlations increased after PSK was removed (with PSK:

| Population structure and isolation by distance
R 2 = .492, p < .001; without PSK: R 2 = .573, p = .003; Figure 3), indicating this pattern was not caused by a single outlying population for Dwarf salamanders.

| Summary of local and landscape characteristics
On average, forest was the most dominant and devel was the least dominant landscape type surrounding study wetlands for both species ( Figure S3). Excluding PSK, Dwarf salamander study wetlands were on average larger than Southern Leopard frog study wetlands (mean = 4.11 ha, SD = 1.96 ha; mean = 1.79 ha, SD = 1.81 ha; respectively), more isolated (mean = −6.66, SD = 1.82; mean = −8.90, SD = 3.03; respectively), and had longer hydroperiods (mean = 193.76 days, SD = 21.62 days; mean = 148.93 days, SD = 41.22 days; respectively). When PSK was not included, most predictor variables were spatially autocorrelated at Southern Leopard frog sites, but were not spatially autocorrelated at Dwarf salamander sites ( Figure S2).
F I G U R E 3 Graphs of isolation by distance for Dwarf salamander and Southern Leopard frog populations. Significant correlations between genetic distance (F ST ) and geographic distance indicate populations exhibit stepping-stone dispersal, whereby populations that are closer together are more closely related

| Top models
Multicollinearity was not an issue for any of the top models of genetic diversity for either species (CN < 2 for all models). Top models of Dwarf salamander genetic diversity suggested stronger associations (greater R 2 values) with habitat variables than the top models of Southern Leopard frog genetic diversity (Table 3). Surrounding road and wetland areas were the best predictors of Dwarf salamander genetic diversity. The top model of Dwarf salamander r g was devel 0.5 km and wtlnd 1.0 km , and the top model for Dwarf salamander H o was wtlnd 0.5 km (Table 3). Predictor variables in both Dwarf salamander top models were significant (i.e., 95% confidence intervals did not cross zero), and both models were more strongly correlated with the respective diversity parameter than top models of Southern Leopard frog genetic diversity (Table 3). Top models for Southern Leopard frog genetic diversity were ag 2.5 km for r g , and devel 2.5 km for H o ( Table 3).
The low R 2 values and lack of statistical significance of variables in the top models of Southern Leopard frog genetic diversity parameters (Table 3) suggested a lack or very weak relationship with the predictor variables of this study, or insufficient variance in parameter values to detect a relationship. The only predictor variables contained in any of the top models that did not exhibit significant spatial autocorrelation were wtlnd 0.5 km and devel 0.5 km for the Dwarf salamander, suggesting that the relationships between the measures of genetic diversity and other predictor variables in the associated top models may have been statistical artifacts due to spatial autocorrelation.

| Composite models
Predictor variables in the composite models of Dwarf salamander r g and H o were identical to those in the respective top models, with the addition of wtlnd 2.5 km for r g (Table 3). All associations between Dwarf salamander genetic diversity and wtlnd variables were positive, whereas devel 0.5 km was negatively associated with r g (Table 3).
Neither Southern Leopard frog genetic diversity parameter was significantly associated with any habitat features in the composite models (Table 3).

| DISCUSSION
This study investigated the population structures of, and habitat features and spatial scales associated with genetic diversity patterns of two sympatric, pond-breeding amphibian species that occur in the southeastern United States. Results indicated that the two species exhibited large differences in population structure and habitat features associated with genetic diversity parameters.

| Dwarf salamander habitat associations and isolation by distance
While IBD analysis suggested that much of the population structure of Dwarf salamanders and some of the structure of Southern Leopard frog populations could be explained by distance from other wet- However, Dwarf salamander genetic diversity was positively associated with the percent area of wetland in surrounding buffers at all three spatial scales included in our study. Given that the buffers may encompass the study wetlands themselves, as well as surrounding wetlands within the radius, these measures inherently describe a combination of wetland area and isolation. Previous studies on amphibians have suggested that wetlands within several hundred meters of each other, without significant intervening barriers to dispersal, may serve effectively as single populations (Petranka, Smith, & Floyd Scott, 2004;Veysey et al., 2011;Zamudio & Wieczorek, 2007). This may be the case for Dwarf salamander populations at Ichauway, potentially due to frequent dispersal among proximal wetlands. Similar relationships with wetland connectivity and amphibian diversity have been found in previous studies. For example, wetland presence, the number of wetlands, wetland density, and wetland area in the surrounding landscape have been found to be positively associated with amphibian species richness (Brodman et al., 2003;Houlahan & Findlay, 2003), abundance and density (Brodman et al., 2003;Hecnar & M'Closkey, 1998;Peterman, Anderson, Drake, Ousterhout, & Semlitsch, 2013), and lower levels of inbreeding (Scribner et al., 2001).
The negative association between Dwarf salamander allelic richness and percent area of roads within 0.5 km suggests that roads can have a negative effect on amphibian genetic diversity. Other studies have shown that roads are barriers to amphibian dispersal (Carr & Fahrig, 2001;Gibbs & Shriver, 2005;Reh & Seitz, 1990) and decrease abundance, species, and genetic diversity (Houlahan & Findlay, 2003;Kuhn, 1987;Reh & Seitz, 1990 were more strongly associated with Dwarf salamander genetic diversity than were local features (i.e., area, iso, hydro). Similar results were found for spotted salamander abundance (Veysey et al., 2011).
The positive relationship between Dwarf salamander allelic richness and percent wetland area within 2.5 km suggests that some degree of gene flow may occur among populations at this scale, enabling the introduction or reintroduction of alleles lost through drift. Houlahan and Findlay (2003) found a similar spatial scale to be associated with species diversity, with positive correlations between amphibian species richness and proportion of wetlands and forest cover at distances between 2.0 and 3.0 km from breeding wetlands, suggesting this spatial scale may be evolutionarily and ecologically relevant for multiple species of pond-breeding amphibians.

| Effect of drought
Below-average rainfall in southwest Georgia in 2006, 2007, and 2008 (total rainfall deviated from average −10.2, −43.2, and −7.6 cm, respectively, in Baker County, Georgia (University of Georgia 2017) likely influenced how the focal species utilized the landscape prior to and during sample collection. Piha et al. (2007) found that regionalscale variables were better predictors of Common frog (Rana temporaria) egg mass abundance after a period of drought, compared to the stronger relationship between landscape-scale variables and egg mass abundance during normal weather conditions. Walls, Barichivich, Brown, Scott, and Hossack (2013) found that Mole salamander (Ambystoma talpoideum) occupancy rates of ponds decreased by more than 50% and local extinction rates increased in association with a 2-year drought, potentially due to drought-induced pond drying.
Because drought rendered a number of potential study sites dry or unoccupied, the wetlands included in this study were inherently larger or had significantly longer hydroperiods than wetlands that were not included, which may have influenced the statistical relationship between genetic diversity and variables associated with wetland size and hydroperiod, as well as other habitat associations. Partially filled wetlands may have reduced the number of individuals that bred in them, thereby increasing the chance of full siblings being collected.

| Conservation and management implications
The maintenance of population genetic diversity is widely accepted as important to the conservation and management of wildlife; however, it is an often-overlooked component in biodiversity conservation management (Taberlet et al., 2012). This study indicates that sympatric pond-breeding amphibian species may be differentially affected by habitat alterations. For example, targeted landscape management may be used to assist with maintaining genetically variable populations of Dwarf salamanders, and gene flow in Dwarf salamanders may be more greatly affected by habitat alterations than in Southern Leopard frogs.
The low degree of population structure in the Southern Leopard frog and lack of associations between habitat features and genetic diversity parameters suggest that at the spatial scale of our study targeted landscape management may not be an necessary for maintaining or facilitating gene flow, and targeted landscape management for other species may not greatly affect rates of gene flow for Southern Leopard frogs.
Though studies have called for an integrated management of wetland complexes and intervening terrestrial habitats to conserve amphibians (Semlitsch, 2000(Semlitsch, , 2002, there is still a general tendency to manage amphibian breeding habitats as isolated units with limited buffer areas.
Potentially negative impacts of even unpaved roads around wetlands on amphibian populations may also be underappreciated. Lastly, results from this study suggest that genetic diversity of sympatric amphibian species may be differentially affected by habitat types surrounding breeding wetlands, with some species more sensitive to differences in habitat types than others. Habitat management efforts focused on maintenance of genetic diversity in populations may be most effective when targeting species with stronger associations to habitat.

CONFLICT OF INTEREST
None declared.

DATA ACCESSIBILITY
Microsatellite data for this manuscript are archived in the online Supplementary Information.