Advertisement

Landscape Ecology

, Volume 34, Issue 5, pp 1097–1115 | Cite as

Connectivity of mule deer (Odocoileus hemionus) populations in a highly fragmented urban landscape

  • Devaughn L. FraserEmail author
  • Kirsten Ironside
  • Robert K. Wayne
  • Erin E. Boydston
Research Article

Abstract

Context

Urbanization is a substantial force shaping the genetic and demographic structure of natural populations. Urban development and major highways can limit animal movements, and thus gene flow, even in highly mobile species. Characterizing varying species responses to human activity and fragmentation is important for maintaining genetic continuity in wild animals and for preserving biodiversity. As one of the only common and wide-ranging large wild herbivores in much of urban North America, deer play an important ecological role in urban ecosystems, yet the genetic impacts of development on deer are not well known.

Objectives

We assessed genetic connectivity for mule deer to understand their genetic response to habitat fragmentation, due to development and highway barriers, in an increasingly urbanized landscape.

Methods

Using non-invasive sampling across a broad region of southern California, we investigated genetic structure among several natural areas that were separated by major highways and applied least-cost path modelling to determine if landscape context and highway attributes influence genetic distance for mule deer.

Results

We observed significant yet variable differentiation between subregions. We show that genetic structure corresponds with highway boundaries in certain habitat patches, and that particular landscape configurations more greatly limit gene flow between patches.

Conclusions

As a large and highly mobile species generally considered to be well adapted to human activity, mule deer nonetheless showed genetic impacts of intensive urbanization. Because of this potential vulnerability, mule deer and other ungulates may require further consideration for effective habitat management and maintenance of landscape connectivity in human-dominated landscapes.

Keywords

Habitat fragmentation Highways Roads Gene flow Urbanization Ungulates Least-cost paths Resource selection 

Introduction

Habitat fragmentation is a major threat to biodiversity worldwide (Brook et al. 2008; Mora and Sale 2011) that can have demographic and genetic consequences for natural populations. Populations may decline due to resource limitation alone (Herkert 1994; Newman et al. 2013). However, factors that reduce genetic diversity such as impeded movement between habitat patches (Sato et al. 2014; Barr et al. 2015) and inbreeding (Bouzat et al. 1998; Johnson et al. 2010), can limit the adaptive potential to environmental change and have direct fitness effects through inbreeding depression (Eldridge et al. 1999; Reviewed in Keller and Waller 2002; Wilson et al. 2016). Large multilane highways can be a significant impediment to gene flow (Keller and Largiader 2003; Riley et al. 2006; Holderegger and Di Giulio 2010; Munshi-South and Kharchenko 2010; Frantz et al. 2012), and an important source of mortality (Vickers et al. 2015) for a variety of species. Cumulatively, such impacts can reduce long-term population viability.

The genetic impacts of fragmentation on medium and large mammalian carnivores have been of particular conservation focus (Riley et al. 2006; Lee et al. 2012; Ruell et al. 2012; Benson et al. 2016a; McClure et al. 2017) as these species are typified by life history traits that intersect poorly with human development (e.g., large home ranges, territorial, low population densities, high maternal investment for few offspring, human conflict). Comparatively fewer studies exist which specifically examine the genetics of large ungulates with respect to fragmentation, especially in the context of urban development. Large ungulates possess intrinsically different life history traits than carnivores and so may differ in their genetic response to habitat fragmentation. Some studies suggest that ungulates are highly sensitive to fragmentation, and more specifically, to anthropogenic barriers (Wang and Schreiber 2001; Coulon et al. 2004; Epps et al. 2005; Frantz et al. 2012; Ito et al. 2013). Indeed, roadways were shown to influence genetic connectivity of mule deer in San Diego, CA (Mitelberg and Vandergast 2016) and white-tailed deer in Wisconsin and Illinois (Robinson et al. 2012). Other studies, however, suggest that deer can adapt to urban settings (Nicholson et al. 1997; Harveson et al. 2007; Blanchong et al. 2013). Characteristics of their social structure, specifically male-biased dispersal and non-random mating across matrilineal groups, may buffer them from genetic impacts of fragmentation observed in other wide-ranging species by maintaining high levels of genetic diversity partitioned among social groups (Blanchong et al. 2013; Crawford et al. 2018). However, landscape features may play an important role in determining social relationships in deer (Koen et al. 2017) and urban development may alter how genetic diversity is partitioned by increasing overlap between social groups (Crawford et al. 2018). Understanding factors influencing the genetic responses of large, highly social and mobile species such as deer is essential for the maintenance of genetic variability in rapidly expanding urban landscapes, especially where large wild spaces are intersected by major highways. The aim of this study was to determine if major highways and the surrounding urban matrix act as significant dispersal barriers for mule deer in southern California.

Southern California serves as a powerful model for assessing the effects of expanding urban development and highway infrastructure on wild populations (Hunter et al. 2003; Vandergast et al. 2007). California is among the most heavily human populated of 25 biodiversity hotspots across the globe, which directly contributes to species imperilment (Cincotta et al. 2000). The juxtaposition of dense human populations and large undeveloped natural areas is common across the state, particularly in the Los Angeles Basin. Vast networks of major highways are a defining characteristic of southern California landscapes, and multiple studies demonstrate the genetic and ecological impacts of fragmentation due to roads on medium and large carnivores across the area (Riley et al. 2006; Ernest et al. 2014; Poessel et al. 2014; Benson et al. 2016a). For example, mountain lion (Puma concolor) populations in both the Santa Monica Mountains (Los Angeles and Ventura Counties) and the Santa Ana Mountains (Orange, San Bernardino, and Riverside Counties) exhibited extremely low genetic diversity and high levels of inbreeding, attributable to the isolation and mortality imposed by highways (Ernest et al. 2014; Riley et al. 2014; Vickers et al. 2015; Gustafson et al. 2017). The long-term viability for these mountain lion populations was predicted to be highly contingent on maintaining immigration from surrounding populations (Riley et al. 2014; Ernest et al. 2014; Vickers et al. 2015; Benson et al. 2016a; Gustafson et al. 2017). Similarly, definitive population structure was observed among both bobcats (Lynx rufus) and coyotes (Canis latrans) sampled across two highways transecting the Santa Monica Mountains. High genetic differentiation was also observed between bobcats sampled in the Santa Ana Mountains and the San Joaquin Hills (Lee et al. 2012). For territorial species, highways may impose social barriers to reproduction for dispersing individuals because territory boundaries may coincide with freeways, thus limiting gene flow despite physical movement across them (Riley et al. 2006).

Relative to carnivores, little is known about the effects of habitat fragmentation on the movement, demography and population genetic structure of the only common native ungulate, the mule deer (Odocoileus hemionus), in southern California. Camera trapping studies suggest that deer movement may be more restricted by highways than other species, as highway underpasses were used less frequently by deer than bobcats, coyotes, and raccoons (Ng et al. 2004; Alonso et al. 2014). Deer appear to have highly specific requirements for utilizing these structures, such as adequate height of the underpass and suitable adjacent habitat (Ng et al. 2004), and thus may have fewer opportunities to cross highways. Further, deer movement across highways is increasingly limited by higher traffic volumes (Coe et al. 2015). Therefore, deer may be more impacted by the isolating effects of highways. In contrast, deer are less restricted by territoriality than carnivores, and so may have more opportunities for reproduction, and hence increase their gene flow, if they successfully traverse roadway barriers. Given the role large herbivores play in structuring ecosystems (Hobbs 1996; Rooney and Waller 2003) and their importance as prey for large carnivores such as mountain lions (Benson et al. 2016b), understanding factors that influence local population dynamics and movement for deer is important in conservation planning.

In this study, we conducted the first assessment of mule deer genetic connectivity in several mountainous areas in southern California. We sampled across a large geographic region with multiple subregions of appropriate deer habitat that bordered highways and development. The natural topography in the absence of development between adjacent subregions was not likely prohibitive to deer movements, and in some cases, highways formed the only definitive break between otherwise continuous subregions (e.g. Fig. S1). We predicted that if highways form barriers to gene flow, then population structure would align with highways and associated urban development that prevented or inhibited deer from successfully crossing. Further, we expected lower pairwise relatedness between individuals in different subregions than between individuals in the same subregion when controlling for distance. Finally, we expected that habitat features and the presence of intervening highways would be the strongest predictors of genetic distances between individual deer.

Materials and methods

Study area

The study area consisted of two main regions, abbreviated as LA (including areas in both Los Angeles and Ventura Counties) and OC (including areas in San Bernardino, Riverside, and Orange Counties), that were further divided into eight subregions based on geographical features (Fig. 1a), urban development (Fig. 1b) and highways separating them (Fig. 1c). Sampling locations (subregions) in the LA Region were the Santa Monica Mountains (SMM) which are bordered to the north and east by two highways and to the south by the Pacific Ocean; the Santa Susana Mountains and Simi Hills (SIMI) situated directly north of the Santa Monica Mountains across California state highway CA-101; the Hollywood Hills including Griffith Park (HH) which are located at the eastern end of the Santa Monica Mountains Ecoregion and directly east of the Santa Monica Mountains across Interstate Highway I-405; and the Verdugo Mountains (VM) which lie just northeast of the Hollywood Hills across two highways (I-5 and CA-134) and a highly developed urban matrix (Fig. 1d). The Verdugo Mountains and the adjacent San Gabriel Mountains are intersected by I-210 to the east.
Fig. 1

Maps depicting landscape topography (a), urbanization (b), and major highways and roads (c) across the study area. Locations of genetic samples gathered across the subregions and other deer occurrence records used to estimate habitat selection are shown for the LA (d) and OC (e) regions. *SMM Santa Monica Mountains, SIMI Simi Hills, SJH San Joaquin Hills, CH Chino Hills, PB Prado Basin; SAM Santa Ana Mountains, HH Hollywood Hills + Griffith Park, VM Verdugo Mountains

The OC Region included the Santa Ana Mountains (SAM) which run north–south and are bordered to the north by California state highway CA-91; the Chino Hills and the Puente-Chino Hills Corridor (CH) located directly north of the Santa Ana Mountains across CA-91; Prado Basin (Prado) which lies directly east of the Chino Hills across California state highway CA-71; and San Joaquin Hills (SJH) which are southwest of the southern end of the Santa Ana Mountains and separated by interstate highway I-5 as well as a large urban matrix, and are bordered to the south by the Pacific Ocean (Fig. 1e). VM and HH were the closest parts of the LA Region to the OC Region, but were separated from CH by continuous urban matrix interspersed with small natural fragments and multiple highways, including I-10 and CA-60. The size and landscape context varied across each of these subregions, as did the size, traffic volume and degree of associated urban development for each of the highways (Fig. 1; Table S1). Natural habitat areas in all subregions were characterized primarily by a mixture of chaparral/sage-scrub and oak woodland/grasslands.

Sample collection

Scat and opportunistic tissue samples from animals found dead were collected from November 2014–June 2015, by researchers, project partners, and volunteers. Collectors followed a protocol to prioritize collection from deer pellet piles that appeared fresh and consolidated, and hence most likely from a single individual (Online Appendix I). Search efforts were targeted within state and national parks and other public open spaces, including natural vegetation areas abutting highways, developed areas, and non-natural open spaces such as golf courses. In many instances, however, search efforts were random in conjunction with volunteer collector activities. All subregions were sampled repeatedly over the course of 7 months to reduce potential temporal effects associated with seasonal deer movements or scat detectability. Our primary goal was to obtain a minimum of 20 unique genotypes per subregion, which was obtained for all but one subregion (VM). Site information, GPS coordinates, date, time, and initials of the collector were recorded for each sample. Samples were collected and placed in paper envelopes or suspended in 95% ethanol and transported to UCLA for processing. Scat samples were assigned a quality score based on appearance and received further processing or were archived and stored in the lab. A total of 648 (29 tissue; 619 scat) samples were collected, rated on a quality scale of 1–5 (scat), stored, and catalogued. DNA extractions were performed on 538 (19 tissue; 509 scat) of these samples (quality ≥ 3). Here and throughout, the term ‘sample’ refers to the scat or tissue from which DNA was extracted, whereas genotypes refer to the samples for which PCR amplifications were successful across a minimum of 10 loci. Finally, we refer to the genotypes included in the final analyses as individuals, because at this stage each genotype corresponds to an individual deer as genotypes of replicate samples were removed from these analyses, as explained in “Mapping, recaptures, and null alleles” section.

DNA extraction and processing

DNA was extracted and PCR amplified as described by Mitelberg (2010). Briefly, 10–12 scat pellets from a sample were suspended in 1 × PBS solution (pH 7.4) and vortexed vigorously to suspend deer epithelial cells from the surface of the pellets. The PBS suspension was then centrifuged at high speed and excess PBS was pipetted off, for a final volume of ~ 1 mL. DNA was extracted from this suspension using the DNeasy Blood and Tissue kit (Qiagen, USA) according to the manufacturer’s protocol. Two elutions were performed on every sample. All sixteen markers were amplified in a single 5 µl reaction following the protocol designed by Mitelberg (2010). Specifically, 1.5 µl DNA were added to 2.5 µl Qiagen Hotstart taq mastermix (Qiagen Multiplex Kits, Qiagen, USA), 0.365 µl primer mix, 0.635 µl H20. Thermocycling conditions were as follows: Initial denaturation at 95 °C for 15 min, followed by 37 cycles of 72 °C for 90 s, 59 °C for 60 s and a final extension at 68 °C for 30 min. PCR products were run on an ABI Prism DNA analyzer and visualized in GeneMapper v 2.7 (Applied Biosystems, Foster City, California).

To ensure accurate genotyping, all samples were genotyped from a minimum of three separate PCR reactions. Heterozygote calls per locus were made only when both alleles were observed at least twice and at least once as a single genotype across the three reactions. Homozygote calls required that the single allele be observed three times across the three reactions. If spurious alleles were observed in any of the reactions, a fourth PCR reaction was run. In some instances, additional reactions were run using the second DNA elution or at single markers to fill in missing loci. We selected to analyze only samples that were genotyped at a minimum of 10 of the 15 loci (n = 328), as this cutoff gave a genotyping success rate > 50% across samples and exceeded the minimum number of markers as recommended by Kolodziej et al. (2012). Support for this cut-off was verified through our analysis to identify replicates in Cervus v. 3.0.7 (Kalinowski et al. 2007) as explained below. All markers had greater than seventy-five percent genotyping success across all samples.

Analysis

Mapping, recaptures, and null alleles

Collection locations of sufficiently genotyped samples were mapped from GPS coordinates recorded in the field and assigned a location attribute according to the subregion in which they were found. Genotypes originating from the same individual but different samples were identified using Cervus v. 3.0.7 (Kalinowski et al. 2007). Genotypes that matched at a minimum of eight loci with no mismatches were assumed to be from the same individual deer. We iteratively tested the minimum number of loci with no mismatches that could distinguish individuals. No further replicates could be identified between 8 down to 5 loci, thereby confirming our selection of a minimum of 10 loci genotyped per sample. If samples with matching genotypes were collected on different days or were separated by more than 2 km, we considered them independent recapture events of the individual. Otherwise, samples yielding genotypes from the same individual were considered redundant, and only the most complete genotype was retained. If the number of loci was equal, we selected the genotype corresponding to the first sample collected for further analyses. Euclidean distances between each recapture were calculated in ArcMAP v. 10.3.1 (Esri, Redlands, California) and tabulated along with the time (in days) between sampling. Microsatellite data were checked using MICROCHECKER v. 2.0 (Van Oosterhout et al. 2004), with 10,000 randomizations implemented for each run. We ran the program iteratively with various population groupings to assess the presence of null alleles.

Genetic diversity, effective population size, and population structure

We made a priori population assignments to each individual corresponding to the subregion in which they were sampled. We calculated summary statistics of genetic diversity for each subregion using GenAlEx v. 6.502 (Peakall and Smouse 2012). We calculated pairwise genetic distances, measured as Jost’s D (Dest), between subregions using the MMOD package v 1.3.2 (Winter 2012) in R v 3.3.0. Jost’s D is suggested to perform better with highly polymorphic loci than traditional measures of genetic distance (Fst, Gst), which can give paradoxical results under conditions of strong differentiation or high diversity, as they strictly employ measures of heterozygosity and do not account for the identity and distribution of individual alleles (Jost 2008). We then used hclust from the statistics package in R (R Core Team 2011) to generate a dendrogram from the pairwise genetic distance matrix. We estimated significance for each pairwise comparison separately from 10,000 permutations of genotype and subregion. We estimated effective population size (Ne) for each subregion in NeEstimator v. 2.1 (Do et al. 2014) using the linkage disequilibrium method (Waples and Do, 2008). We used a critical value for the minimum allele frequency of 0.02 and omitted 2 loci with greater than 15% missing data (H and L) to get a numeric estimate of Ne for each subregion.

We used two complementary approaches to assess the distribution of genetic diversity in mule deer with respect to major highways. First, population subdivision across the study area was assessed using a Bayesian clustering algorithm implemented in the program STRUCTURE v. 2.3.4 (Pritchard et al. 2000). We ran a total of 15 iterations per K value, where K represents the number of distinct genetic clusters, and assessed the data across a range of K = 1–10. Proportional assignments were made for each individual to each of K clusters. We used an admixture model, and ran each simulation for 106 iterations of the MCMC and 2 × 105 burn in. We ran STRUCTURE results through Structure Harvester (Earl 2012) to determine optimal values for K based on the Evanno et al. (2005) method.

Second, we applied discriminant function analysis of principal components (DAPC) using the poppr package v. 2.8.1 (Kamvar et al. 2014) in R. This method applies a k-means clustering algorithm to multi-locus genetic data that has been transformed using principal components analysis. The clustering is followed by a multivariate discriminant function analysis, which minimizes within group variance and maximizes between group variance in the PCA transformed data. This method is less restricted than STRUCTURE by assumptions of Hardy–Weinberg equilibrium and linkage disequilibrium and is useful when weak structure is likely (Jombart et al. 2010). We retained the first 45 principal components, which explained approximately 90% of the data. For both STRUCTURE and DAPC, we calculated the evenness of genetic clusters in each subregion using the Shannon diversity index (Shannon and Weaver 1949) as a means to rank relative admixture, and conversely genetic isolation, across subregions.

Least-cost path analysis

We developed a cost-surface using a suite of candidate landscape covariates, described below, and a competing model framework, Akaike Information Criterion (AIC) (Burnham and Anderson 2002). Mule deer occurrence was established from a combination of the samples collected for genetic analyses in this study, as well as 93 iNaturalist records of mule deer, 91 remote camera detection locations (where mule deer were detected 1 or more times at a site), and 10 locations of mule deer samples collected from Pease et al. (2009), resulting in a total of 829 unique locations of mule deer occurrence in the study area (Fig. 1a). Due to few data sources for true absence records, we employed a balanced, stratified random sampling approach on which to compare the presence data in a logistic regression frame work. We generated 829 paired random points within a distance of 31 km of a presence record to sample the range of values of landscape features in the areas where observations were collected (Fig. 1a). Candidate explanatory variables of mule deer occurrence included topographic features, current land-uses, and vegetative characteristics (Table S2). While genetic responses may lag with respect to changes in the landscape and thus may more likely reflect previous land-use conditions, we selected to use current land-use for two reasons. First, each of the focal highways have been in place for a minimum of 30 years, or approximately 6 generations, with traffic volumes increasing marginally since 1993, or approximately 4 generations (Table S1). Second, because urbanization is an ongoing process, current land-use may provide stronger insights as to the direction of future genetic responses in mule deer.

Topographic variables included a terrain ruggedness measure termed unevenness, a terrain view-shed index known as openness (Yokoyama et al. 2002), steepness of slope, general topographic curvature, and site exposure index. Unevenness was calculated using ArcGIS 10.3.1’s Spatial Analyst extension (Esri 2014) focal cell statistics tool to measure the variation, using standard deviation, of general curvature for a 150 m radius. General curvature was also calculated using ArcGIS 10.3.1’s Spatial Analyst extension (Esri 2014) using a 30-meter resolution digital elevation model (DEM) (U.S. Geological Survey 2009). This DEM was also used to calculate slope in degrees and site exposure index, a measure of the effects of solar radiation and aspect, using the Surface Gradient and Geomorphometric Modeling toolbox (Evans et al. 2014) in ArcGIS (Esri 2014). Land-use and vegetation characteristics were represented using a road density metric and the LANDFIRE program’s datasets for existing vegetation and height (LANDFIRE 2011; Rollins 2009). USGS transportation road segment data for California were obtained from the National Map (https://viewer.nationalmap.gov/ Accessed June 15, 2017) and line density of all classes of roads were calculated using an analysis window of 120 m.

Existing vegetation types (LF EVT) was represented using the attribute values of red, green, and blue, which are a color map with values scaled between 0 and 1, that represent particular classes of vegetation types using the Society of American Foresters classification system. High values of red (LF EVT red) represent barren, developed, and disturbed areas of introduced herbaceous and grassland vegetation types. High values of green (LF EVT green) are associated with native grasslands, woodlands, forests, and riparian vegetation types. High values of blue (LF EVT blue) are associated with developed areas such as agricultural fields and irrigated nonnative vegetation. LANDFIRE’s existing vegetation height classes were used to create a continuous measure of vegetation height by taking the median value of a class, resulting in a range of vegetation height between 0 and 37.5 m for the study area.

We competed six models using combinations of the candidate variables, equally weighted throughout, and selected the best model with the lowest AIC value (Table S3). This model was used to calculate the inverse of the dependent variable (the probability of mule deer presence), to describe the cost of movement and habitat connectivity for mule deer across the study area. A high probability of modeled mule deer-use of the landscape, assumed to represent high quality habitat, is therefore assumed to have low cost of movement. In contrast, modelled low quality habitat is estimated to have a high cost of movement. Pairwise least-cost paths, both the geographic distance and the accumulated cost of the least-cost path, were calculated using a Python v 2.7 tool (https://www.arcgis.com/home/item.html?id=bbc7ae14015747318e06dda0f6c5bddf Accessed July 12, 2017) for use in ArcGIS 10.1 (Esri 2012). We measured the Euclidean distance (m) and the habitat accumulated cost (HAC) which measures the sum of the resistance along the least-cost path (Etherington and Holland 2013).

To explore whether or not high traffic volume, multilane freeway systems are a feature on the landscape that are costly to mule deer movement, we created a second cost-surface. Using 2014 Caltrans data on annual traffic volumes, measured as the total annual traffic volume divided by 365 days (http://www.dot.ca.gov/hq/tsip/gis/datalibrary/ Accessed August 10, 2017), we scaled the observed 0–377,000 Ahead Annual Average Daily Traffic (ADDT) volume between 0 and 1, by dividing by the maximum value. Using a spatial join to attribute the ADDT point values to the Caltrans freeway line segments, we then created a cost raster based on the rescaled ADDT and added it to the habitat-based cost-surface. This resulted in a second cost-surface, ranging in cost values 0–2, that results in the possibility of different least-cost paths and higher accumulated cost paths for pairwise comparisons that are separated spatially by one or more freeway systems. We termed this variable as the freeway accumulated cost (FAC). The ADDT metric only applies to California highways and not urban surface streets, making interactive effects between road density and ADDT minimal (i.e. low correlation coefficient).

Landscape determinants of individual pairwise

We used measures of genetic distance between individuals to test the hypothesis that highways form a barrier to gene flow in mule deer. We first estimated individual pairwise genetic distance in GenAlEx v. 6.502 (Peakall and Smouse 2012), which is calculated as the inverse of the relatedness between two samples based on genetic congruity. For clarity, we refer throughout to this measure of genetic distance as inverse relatedness (IR), so as not to be confused with subregion-level genetic distance as measured by Dest. We compared mean IR for within- versus between subregion comparisons and expected lower relatedness for between-subregion comparisons. Significance was determined by bootstrapping a null distribution of differences in mean IR. We subtracted the means of 500 randomly sampled individual pairwise relatedness values from the mean of 500 different randomly sampled pairwise relatedness values that were drawn from the entire dataset (both with-in and between subregions combined) with replacement. We ran the bootstrap 10,000 times to assess significance. We constrained the data to include only pairwise comparisons that fell within the overlapping range of Euclidean distances for pairs sampled within versus between subregions, such that very large distances observed between geographically distant subregions were removed, as were very small distances that only occurred between samples collected within a single subregion.

To determine if habitat quality and presence of highways better explained inverse relatedness (IR) in mule deer relative to a standard isolation-by-distance model, we employed a comparative modeling approach to identify the strongest predictor of IR. We ran generalized linear mixed effects models that incorporated a single predictor at a time using individual as a random effect to account for non-independence of the data (Clarke et al. 2002; van Strien et al. 2012). We tested three spatially explicit hypotheses related to this overarching theory; mule deer genetic relatedness is a function of nearness (e.g. Euclidean distance), versus a function of habitat connectivity, versus a function of habitat connectivity intersected by freeway systems. We ranked these models according to their coefficients to identify the strongest predictor and employed a bootstrapping approach based on AIC values to assess significance as described in Jaquiéry et al. (2011). We ran 10,000 iterations, sampling the entire dataset with replacement and using the same samples for all four models during each iteration. We limited our data to include only geographically adjacent subregions and assigned significance if the 95% confidence interval (one-tailed) of the ΔAIC between the focal and the base model did not overlap with zero (Jaquiéry et al. 2011).

Results

Mapping, recaptures, and null alleles

Cervus identified a total of 265 unique genotypes representing individual deer. For 40 individual deer, two to five samples were collected per individual deer. For most of these individuals (32 of 40), the additional samples were considered redundant because they were collected in close spatial proximity, on the same day, or both. This resulted in removal of 54 genotypes that were redundant. For the other 8 individuals, there was at least one additional sample designated as a recapture event, resulting in the removal of 9 genotypes. Of these recaptures, 2 were samples of the same individual collected on opposite sides of a highway: across CA-71 (between CH and PB) and across I-405 (between HH and SMM). Another recapture from CH was a sample found in a CA-71 underpass, thereby providing at least two instances of movement between Prado and CH. Direct evidence of movement between Prado and CH through genetic analysis and camera-trap data (Alonso et al. 2014), as well as high relatedness to CH deer (7/9 deer sampled in Prado showed first order relationships with deer sampled in CH) prompted us to include Prado samples with CH in subsequent analysis as a single subregion. The average Euclidean distance between samples for the 8 recaptured individuals was 1.84 km (s.d. = 0.96 km; max = 3.02 km), while the average Euclidean distance between 54 redundant genotypes for 32 individuals was 0.09 km (s.d. = 0.16 km; max = 0.75 km).

Null allele analysis showed locus B to have a very high null allele frequency, as observed in previous studies of mule deer genetics in southern California (Pease et al. 2009; Mitelberg 2010), and was thus removed from the dataset for subsequent analyses. Several other loci exhibited evidence of null alleles depending on how subregions were grouped (data not shown), likely indicating genetic structure and not necessarily the presence of true null alleles. We therefore proceeded in the analysis with the fourteen remaining loci.

Genetic diversity, effective population size, and population structure

Observed heterozygosity averaged across all loci for each subregion ranged from 0.49 ± SE: 0.046–0.69 ± SE:0.072 (Table 1). The lowest observed heterozygosity was in SMM followed by SIMI. All loci were polymorphic in all populations except for SJH for which locus L was monomorphic. The fixation index was lowest in SAM (0.012, SE 0.032) and was highest in SIMI (0.135, SE 0.047). The largest effective population sizes (Ne) were observed in SAM (Ne = 548.2; 95% CI 89.1–inf) followed by VM (Ne = 84; 95% CI 6.6–inf). The smallest effective population size was observed in SJH (Ne = 16.8; 95% CI 10.5–30.8) followed by HH (21.4; 95% CI 14.1–36) (Table 1).
Table 1

Effective population size, observed heterozygosity and expected heterozygosity across 14 loci for each subregion

Subregion (N)

Ne (95% CI)

Ho (SE)

He (SE)

CH (73)

69.6 (48–112.6)

0.51 (0.06)

0.52 (0.05)

SAM (45)

inf (235.9–inf)

0.55 (0.06)

0.56 (0.06)

SJH (24)

16.6 (10.4–29.9)

0.55 (0.06)

0.55 (0.06)

VM (8)

79 (6.9–inf)

0.69 (0.07)

0.59 (0.05)

HH (30)

22.9 (15.2–38.8)

0.54 (0.05)

0.59 (0.05)

SMM (52)

63.9 (40.5–124.8)

0.49 (0.05)

0.56 (0.06)

SIMI (33)

32.9 (20.9–62.3)

0.5 (0.06)

0.57 (0.05)

Ne effective population size, Ho observed heterozygosity, He expected heterozygosity

Global Dest was 0.072 (95% CI 0.057–0.092). Measures of pairwise Dest were significant at p ≤ 0.005 after permutation between all subregions, and ranged from 0.017, between SMM and SIMI, to 0.138, between SJH and SMM (Table 2). The highest average pairwise Dest among subregions, 0.11, was observed in SJH, followed by HH; the lowest average Dest of 0.05 was observed in SAM. Hierarchical clustering of subregion pairwise Dest showed VM clustering as sister group to CH/SAM, with all three subregions clustering as sister to the LA Region, and SJH forming an outgroup to the LA and OC Region groupings. Within the LA Region, SIMI and SMM cluster together with HH as an outgroup (Fig. 2).
Table 2

Pairwise and average genetic distances (Dest) between 7 mountainous subregions separated by major highways in southern California

Subregion

CH

SAM

SJH

VM

HH

SMM

SIMI

Mean dest

CH

0.0001

***

0.0028

***

***

***

0.065

SAM

0.02

***

0.0008

***

***

***

0.05

SJH

0.13

0.07

0.0005

***

***

***

0.11

VM

0.05

0.05

0.09

0.0004

0.0001

0.0061

0.067

HH

0.07

0.05

0.13

0.07

***

***

0.072

SMM

0.06

0.06

0.14

0.08

0.05

0.005

0.068

SIMI

0.06

0.05

0.12

0.06

0.06

0.02

0.062

Below diagonal: genetic distance (Dest); above diagonal: p value

*** Indicates p < 0.0001 after 10,000 bootstrap iterations

Fig. 2

Posterior assignments for individual deer to each of five genetic clusters identified in STRUCTURE. Subregions are arranged and grouped according to the pairwise genetic distance (Dest) matrix as shown by the dendrogram

STRUCTURE analysis on all unique individuals (n = 265) indicated that K = 5 was the optimal number of genetic clusters (Fig. 2; Table 3) while DAPC analysis suggested that K = 6 was optimal. Admixture was present in all subregions in both analyses; however, average proportional assignments to each cluster varied by subregion. The highest posterior assignments to a single cluster ranged from 32-65 percent and from 27 to 64 percent, in STRUCTURE and DAPC, respectively (Table 3). In both analyses, the highest assignment to a single cluster occurred in SJH. Evenness of cluster assignments also varied by subregion and by analysis. In both analyses, SJH showed the lowest evenness.
Table 3

Average proportional assignment of individuals to K clusters and relative evenness across clusters by subregion for both STRUCTURE (A) and DAPC (B)

 

Clust 1

Clust 2

Clust 3

Clust 4

Clust 5

Rank H’/Hmax

STRUCTURE (K = 5)

     

 

 CH

0.41a

0.09

0.36

0.06

0.07

4

 SAM

0.21

0.24

0.32a

0.08

0.15

2

 SJH

0.07

0.65a

0.12

0.08

0.07

7

 VM

0.13

0.21

0.33a

0.18

0.15

1

 HH

0.09

0.08

0.12

0.15

0.55a

5

 SMM

0.11

0.09

0.10

0.44a

0.26

3

 SIMI

0.11

0.07

0.07

0.58a

0.17

6

 

Clust 1

Clust 2

Clust 3

Clust 4

Clust 5

Clust 6

Rank H’/Hmax

DAPC (K = 6)

       

 CH

0.36a

0.20

0.02

0.06

0.28

0.08

3

 SAM

0.23

0.27a

0.11

0.10

0.23

0.07

1

 SJH

0.03

0.02

0.64a

0.23

0.05

0.03

7

 VM

0.17

0.03

0.12

0.60a

0.04

0.04

6

 HH

0.09

0.03

0.00

0.51a

0.15

0.22

5

 SMM

0.04

0.04

0.02

0.16

0.33

0.41a

4

 SIMI

0.06

0.11

0.07

0.18

0.24

0.33a

2

Shannon–Weaver calculations of evenness are shown as rankings for each subregion, with 1 suggesting the most even distribution of genetic clusters and 7 showing the lowest evenness across genetic clusters

aHighest proportional assignment for each subregion

Least-cost path analysis

Several topographic features, vegetative, and anthropogenic factors were found to be associated with the likelihood of mule deer occurrence (Fig. 3; Tables S2, S3) on the landscape and were used to create a resistance surface (Fig. 4). Terrain unevenness and openness both had strong quadratic relationships suggesting a preference for mid-range values within the modeled region. LANDFIRE’s green attribute associated with several native vegetation types had a positive relationship, while steep slopes, high road densities, and non-native vegetation (LANDFIRES’s red and blue values) resulted in a negative response (Fig. 3a). Some responses were modified by an interaction term (Fig. 3b). The response became negative to site exposure index (SEI) on steeper slopes (high values of SEI are southern aspects receiving more solar input), and to LANDFIRE’s red values in areas of higher road densities. Though the response to LANDFIRE’s green metric was positive, the response was even stronger at higher road densities. Generally, resistance costs surrounding highway underpasses were greater between subregions with higher pairwise Dest (Fig. 4; Table 2). The correlation between the road density raster and the ADDT raster was 0.079.
Fig. 3

Response curves for mule deer occurrence data versus and equal number of randomly stratified locations. Factors exhibiting responses (quadratic and linear) include terrain unevenness and openness, and 4 LANDFIRE attributes (a). Factors modified by an interaction term include site exposure index (SEI) × slope, and 2 LANDFIRE attributes (red and green) × road density

Fig. 4

Cost-surface used to estimate mule deer habitat connectivity for the study area (a). Using a resource selection function (RSF), we estimated high quality habitat would have little cost for movement for mule deer and landscape features where deer were not observed would have a high cost of movement on a scale of 0–1 (the inverse probability of deer presence). To explore whether or not high traffic volume may pose a barrier to movement, we added additional costs to the locations of freeway systems based on the Ahead Annual Average Daily Traffic (ADDT) volume rescaled to values between 0 and 1 and added it to the habitat base surface. These areas of increased cost to traverse are shown for a portion of the US-101 (b), I-405 (c), the CA-91 and CA-71 intersection (d), and the I-5 (e). Points indicate crossings available to deer (gray circles = overpasses; black circles = underpasses; red star = Liberty Canyon underpass)

Determinants of individual pairwise relatedness

Mean inverse relatedness (IR) between adjacent subregions was 16.54 (95% CI 16.18–16.83) and within subregions was 15.93 (95% CI 15.59–16.22), for a small yet significant difference of 0.608 (p = 0.007) across comparable Euclidean distances. The most highly ranked and significant predictor of IR based on linear coefficients was the freeway accumulation cost (FAC), which was significant as the 95% confidence interval did not overlap with zero (Table 4; Fig. S2). HAC was also significant. However, while the coefficient for FAC as a predictor of IR was greater than for HAC, the difference in the power of the models to explain IR according to ΔAIC and was not was not significant according to our bootstrapping results. Collectively, these analyses suggest that both highways and internal patch structure may have an important influence on genetic connectivity for mule deer.
Table 4

Generalized linear models for predictors of individual pairwise genetic distance, IR, between deer sampled in adjacent subregions

Model

Rank

β

Df

AIC

ΔAIC

95% CId

IRa ~ euclidean + (1 | Individual)

3

0.434

4

72,998

22

− 12.18

IR ~ HACtb + (1 | Individual)

2

0.488

4

72,977

1

− 0.55

IR ~ FACc + (1 | Individual)

1

0.49

4

72,976

0

aInverse Relatedness (IR)

bHabitat Accumulated Cost

cFreeway Accumulated Cost

dOne-tailed after 10,000 bootstraps

Discussion

Roads and urban development can pose dispersal challenges for wildlife, particularly large and highly mobile species such as ungulates, whose home ranges and dispersal movements are likely to be intersected by highways. Our resource selection models showed a strong negative response of mule deer presence with road density and developed areas. Accordingly, we found some evidence of genetic structure among subregions as well as a subtle effect of highways on individual pairwise genetic distance. Interestingly, however, our resource selection models also showed an interaction between road density and landcover, such that the positive response to high quality vegetation becomes stronger with higher road density, suggesting that deer respond differently to roads depending on abutting habitat and are potentially more likely to use more variable habitats in the absence of high road density. Therefore, not all highways and landscape configurations are equal in the context of wildlife movement. In conjunction with studies focused on carnivores, our results highlight the need to account for behavioral differences in focal species and underscore the importance of specific features of natural habitat, as well as highway and urban infrastructure barriers when planning for habitat connectivity in fragmented landscapes.

Genetic structuring in North American mule deer is suggested to result primarily from historical processes such as glaciation (Latch et al. 2009; Pease et al. 2009; Latch et al. 2014) and contemporary ecological processes (Pease et al. 2009). Broadly, mule deer in California partition into five distinct genetic clusters attributed mainly to ecological factors related to climate and elevation, with deer in our study area belonging to a single genetic cluster (Pease et al. 2009). Yet our results reveal more subtle patterns of genetic structure at both regional and local scales. Regionally, mule deer in our study area appeared to cluster into three main groups: (1) LA Region (SIMI, SMM, and HH); (2) CH, SAM, and VM; and (3) SJH (Fig. 2). However, in both k-means clustering approaches, no single cluster was exclusive to any single region or subregion; and all clusters were represented in all or nearly all subregions (Table 3). The latter evidence suggests high levels of historic connectivity between subregions likely facilitated by the geography of the area (i.e. movement along mountainous corridors) and isolation by distance commonly observed in this species (Latch et al. 2014). This contrasts with our measures of genetic differentiation between subregions, all of which were statistically significant and substantially higher than has been observed in other ungulate populations, including white-tailed deer (Odocoileus virginianus), across distances exceeding 1000 km (Cullingham et al. 2011; Mager et al. 2013).

Although highways are often constructed along natural breaks in landscape topography, we argue that the genetic differentiation between subregions in our study area is unlikely to result from the historic influence of natural breaks in the landscape versus artificial barriers for several reasons. First, shared ancestry as demonstrated by both STRUCTURE and DAPC results suggest historic admixture across the entire study region, and the small distances between subregions are unlikely to preclude deer movement in the absence of a strong barrier (e.g. Fig. S1). Additionally, the portion of the Santa Monica Mountains Transverse Range extending into the Hollywood Hills and located east of I-405, is otherwise geographically continuous with the rest of the Santa Monica Mountains apart from the presence of I-405. Some exceptions may be VM/HH and SJH/SAM, as these mountainous subregions are separated by large natural intervening valleys which are now largely urbanized.

At more local scales, several interesting patterns emerged. First, although VM is geographically more proximate to the LA Region and to HH specifically (Fig. 1), both pairwise Dest and STRUCTURE results suggested that deer sampled in VM were more genetically similar to OC Region deer (Fig. 2; Table 2), specifically CH and SAM. Further, although represented by only 8 individuals, VM had the second highest estimated Ne and the highest observed heterozygosity, suggesting connectivity with a larger area of suitable deer habitat. In this case the connectivity was likely with the San Gabriel Mountains which are substantially larger than SAM or SMM and for which there is a large underpass across I-210 that is well buffered by natural vegetation and limited human development. In contrast, the matrix between VM and HH is highly developed and requires traversing a wide intersection of two highways (CA-134 and I-5). While high development also exists between VM and CH, we hypothesize that large population sizes accommodated by large open areas in both the San Gabriel and the Santa Ana Mountains, as well as probable long-distance movements of deer as documented in nearby surrounding areas (Nicholson et al. 1997), have minimized the effects of genetic drift that would otherwise lead to higher differentiation as observed in other subregions (i.e. HH and SJH). Future work in VM that includes sampling deer in the San Gabriel Mountains and more data on the historical connection between them might show the connectivity requirements for a relatively small mountain range to support such a high Ne.

Our results also highlight a potentially important management unit for mule deer in southern California. Deer from SJH showed the highest genetic differentiation from other subregions, as well as the lowest effective population size. Specifically, average pairwise Dest between SJH and all other subregions was over 60% greater than the next highest average Dest (observed in HH), comparable to values observed for populations of white-tailed deer separated by the Rocky Mountains in Canada (Cullingham et al. 2011). Further, in both STRUCTURE and DAPC analyses, SJH had the highest incidence of assignment to a single cluster and hence the lowest evenness in genetic diversity of all subregions considered (Table 3). A vast urban matrix and major highway (I-5) separate SJH and SAM subregions, and similar genetic results for this area have been found in bobcats (Lee et al. 2012; Ruell et al. 2012) and other species (Vandergast et al. 2007). Unfortunately, current analysis prevents us from distinguishing whether this level of isolation is directly due to habitat fragmentation, in which case management approaches focused on increased connectivity would be useful; or whether this is a historical pattern resulting from natural landscape and ecological processes, in which case management that preserves this unique population by monitoring population size would be desirable. Enhancing sample sizes for this region and improving genetic diagnostics through sequencing to accurately measure coalescence time and other properties of genomic isolation such as inbreeding could help determine the most appropriate management approach for this genetically unique population.

The pattern for SJH in the OC Region is paralleled to a lesser magnitude by HH in the LA Region, with HH having the second highest average Dest and the third lowest evenness in both STRUCTURE and DAPC analyses (Table 3). HH is separated from SMM by I-405, one of the busiest highways in the country that spans 12 lanes between these subregions. Ongoing camera monitoring of four potential crossings along this highway segment indeed demonstrated that deer are less likely to utilize crossings than other species, although some movement of deer under I-405 was observed at a single underpass (J. Brown, NPS-unpublished data).

The lowest divergence was observed between CH and SAM and SMM and SIMI in the OC and LA Regions, respectively, reflecting either higher rates of gene flow between the subregions or slower rates of genetic drift due to population size. Although CH and SAM showed low levels of genetic distance (Dest) across CA-91, the presence of highways and intervening development may later show genetic impacts. The expansion of CA-91 is relatively recent (27% increase in ADDT between 1993 and 2014; Table A1), and monitoring of several underpasses along CA-91 by remotely-triggered cameras showed only limited use by deer of a single underpass connecting CH to SAM (Boydston and Crooks 2013). This is particularly relevant given that our resistance-surfaces were calculated from current land-use and not historic land-use. Although current patterns of genetic differentiation in mule deer may better reflect previous land-use patterns, the process of continuing development and increasing human population in southern California will likely intensify observed differentiation in the absence of appropriate habitat connectivity planning. Low differentiation between SIMI and SMM may also result from the presence of at least one potential movement corridor located at Liberty Canyon (Fig. 4) and low relative traffic volumes along US-101. The existing underpass at this location has a road through it for vehicles but also has dimensions adequate for deer to use and levels of adjacent natural habitat associated with deer use of underpasses (Ng et al. 2004). This corridor has also been identified as an important connectivity unit by the California Essential Habitat Connectivity Assessment (Spencer et al. Spencer 2010) and as a proposed location for a wildlife crossing to improve habitat connectivity for mountain lions and other wildlife in this study area. Similarly, CA-71 did not appear to inhibit deer movements, given that several recaptured individuals were identified directly within underpasses and on either side of this highway. This highway has several large underpasses, some built with wildlife-friendly features as mitigation for highway widening (Alonso et al. 2014), thus highlighting the benefits of planning for wildlife connectivity during urban infrastructure expansion.

Our results suggest that I-405 and I-5 and the associated urban matrix are strong candidate barriers to gene flow for mule deer in our study area. Although we did not detect strong population structure across CA-91 or CA-101, the patterns of increasing volume of traffic and expanding development occurring in these systems suggests these highways may become problematic for genetic exchange in mule deer in the future. In sum, this study reveals several important areas and highways associated with limited connectivity for wildlife in our study area. More broadly, our results highlight the importance of landscape context in shaping the genetic responses of wildlife to urban development and fragmentation. Further, studying urban adapted species with respect to habitat fragmentation is important for maintaining biodiversity in urban landscapes, particularly in the case of large mammals that can move large distances and may encounter highways more frequently in their movements. Our study suggests that deer and other large ungulates may serve as important sentinels for the genetic impacts of fragmentation on wildlife. The relative ease of non-invasive sampling for this and similar species makes them an ideal system for rapid assessment of gene-flow, while detection of reduced gene flow in this species most likely indicates reduced gene flow in other species. Conversely, maintaining connectivity focused on deer is likely to enhance connectivity for a variety of other species.

Our study has several limitations that require mention. First, our sample collection was largely opportunistic or biased toward public open spaces. Future research adopting a systematic repeat-transect approach wherein individuals could be sampled repeatedly over time would enable a spatial mark-recapture model that could more accurately estimate population size and home range sizes. Currently, little is known about the true population size, home range sizes and movements of deer in this study area. Second, while our permutation and mixed effects modeling approaches enabled us to handle pairwise comparisons and account for different movement behaviors across individuals, we recognize that measures of genetic distance between individuals potentially overlooks the influence of social behavior on gene flow and that measures of genetic distance at the population level are more ideal. Future efforts to incorporate a greater number of subregion replicates in a variety of urbanized landscapes would improve statistical power to disentangle the effects of habitat characteristics and major roadways on gene flow in mule deer. Finally, the genetic markers used in study are limited in their ability to resolve the basis for the genetic distinctiveness of the San Joaquin Hills mule deer population. Genomic approaches could provide more precise estimates of coalescence between populations and provide invaluable insights into local adaptation, historical gene flow, and levels of inbreeding in these areas.

Conclusion

As mule deer are not territorial, individuals that disperse successfully across highways may be more likely to contribute genetically to the adjoining population compared to carnivores. However, as human populations continue to expand in urban areas with associated increase in transportation needs and infrastructure, effective migration opportunities may decrease. Establishment of at least one adequate crossing structure on a highway that divides habitat areas may be enough to maintain geneflow in mule deer and potentially other species. Our findings underscore the continued relevancy of wildlife crossings and connectivity across urban matrices to land and wildlife managers in areas where connectivity is increasingly limited by continued urbanization. Future work that addresses seasonal movements, survival and reproductive success linked to habitat quality and urban development for mule deer and other highly mobile species will provide a stronger framework for designing wildlife corridors where urban associated barriers may impose direct costs in terms of demographic and genetic viability.

Notes

Acknowledgements

Several organizations provided volunteer and staff assistance in collecting samples and accessing lands, including Santa Monica Mountains National Recreation Area National Park Service, Irvine Ranch Conservancy, OC Parks, California State Parks, Orange County Trackers, and Ventura County Wildlife Trackers. We thank M.B. Bourne, C. Shew, and J. Curti for their diligent work in the lab. We thank A. Vandergast and A. Mitelberg for consultation on laboratory protocol and analysis. We thank M. Peters for his excellent computer programming and GIS skills that aided the computation of least-cost paths. We also thank L. Lyren and A. Vandergast for conceptual input, and additionally thank L. Lyren for excellent field work and organization launching this project. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Funding

This work was supported by Army Corps of Engineers, United States Geological Survey Ecosystems Mission Area, and the UC Catalyst Grant Program (#CA-16-376437).

Supplementary material

10980_2019_824_MOESM1_ESM.docx (3.1 mb)
Supplementary material 1 (DOCX 3177 kb)

References

  1. Alonso RS, Lyren LM, Boydston EE, Haas CD, Crooks KR (2014) An evaluation of a road expansion and wildlife connectivity mitigation project on a southern California freeway. Southwest Nat 59:181–187CrossRefGoogle Scholar
  2. Barr KR, Kus BE, Preston KL, Howell S, Perkins E, Vandergast AG (2015) Habitat fragmentation in coastal southern California disrupts genetic connectivity in the cactus wren (Campylorhynchus brunneicapillus). Mol Ecol 24(10):2349–2363CrossRefPubMedGoogle Scholar
  3. Benson JF, Mahoney PJ, Sikich JA, Serieys LE, Pollinger JP, Ernest HB, and Riley SP (2016a) Interactions between demography, genetics, and landscape connectivity increase extinction probability for a small population of large carnivores in a major metropolitan area. In: Proceedings of Royal Society B, vol. 283, no. 1837. The Royal Society, p 20160957Google Scholar
  4. Benson JF, Sikich JA, Riley SPD (2016b) Individual and population level resource selection patterns of mountain lions preying on mule deer along an urban-wildland gradient. PLoS ONE 11(7):e0158006CrossRefPubMedPubMedCentralGoogle Scholar
  5. Blanchong JA, Sorin AB, Scribner KT (2013) Genetic diversity and population structure in urban white-tailed deer. J Wildl Manage 77:855–862.  https://doi.org/10.1002/jwmg.521 CrossRefGoogle Scholar
  6. Bouzat JL, Cheng HH, Lewin HA, Westemeier RL, Brawn JD, Paige KN (1998) Genetic evaluation of a demographic bottleneck in the greater prairie chicken. Conserv Biol 12(4):836–843CrossRefGoogle Scholar
  7. Boydston EE, Crooks KR (2013) Movement patterns of bobcats and coyotes after widening of CA-71 near CA-91 in Southern California: final report. In: U.S. Geological Survey, Sacramento, p 194Google Scholar
  8. Brook BW, Sodhi NS, Bradshaw CJ (2008) Synergies among extinction drivers under global change. Trends Ecol Evol 23(8):453–460CrossRefPubMedGoogle Scholar
  9. Burnham, K. P., Anderson, D. R., 2002. Model selection and multimodel inference: a practical information-theoretic approach (2nd ed.). Springer-Verlag, ISBN 0-387-95364-7Google Scholar
  10. Cincotta RP, Wisnewski J, Engelman R (2000) Human population in the biodiversity hotspots. Nature 404(6781):990CrossRefPubMedGoogle Scholar
  11. Clarke RT, Rothery Peter, Raybould AF (2002) Confidence limits for regression relationships between distance matrices: estimating gene flow with distance. J Agric, Biol, Environ Stat 7:361CrossRefGoogle Scholar
  12. Coe PK, Nielson RM, Jackson DH, Cupples JB, Seidel NE, Johnson BK, Gregory SC, Bjornstrom GA, Larkins AN, Speten DA (2015) Identifying migration corridors of mule deer threatened by highway development. Wildl Soc Bull 39(2):256–267CrossRefGoogle Scholar
  13. Coulon A, Cosson JF, Angibault JM, Cargnelutti B, Galan M, Morellet N, Petit E, Aulagnier S, Hewison AJM (2004) Landscape connectivity influences gene flow in a roe deer population inhabiting a fragmented landscape: an individual-based approach. Mol Ecol 13(9):2841–2850CrossRefPubMedGoogle Scholar
  14. Crawford JC, Quinn AD, Williams DM, Rudolph BA, Scribner KT, Porter WF (2018) Fine-scale spatial genetic structure of deer in a suburban landscape. J Wildl Manage 82(3):596–607CrossRefGoogle Scholar
  15. Cullingham CI, Merrill EH, Pybus MJ, Bollinger TK, Wilson GA, Coltman DW (2011) Broad and fine-scale genetic analysis of white-tailed deer populations: estimating the relative risk of chronic wasting disease spread. Evol Appl 4(1):116–131CrossRefPubMedGoogle Scholar
  16. Do C, Waples RS, Peel D, Macbeth GM, Tillett BJ, Ovenden JR (2014) NeEstimator V2: re-implementation of software for the estimation of contemporary effective population size (Ne) from genetic data. Mol Ecol Res. 14:209–214CrossRefGoogle Scholar
  17. Earl DA (2012) STRUCTURE HARVESTER: a website and program for visualizing STRUCTURE output and implementing the Evanno method. Conserv Genet Res 4(2):359–361CrossRefGoogle Scholar
  18. Eldridge MD, King JM, Loupis AK, Spencer P, Taylor AC, Pope LC, Hall GP (1999) Unprecedented low levels of genetic variation and inbreeding depression in an island population of the black-footed Rock-Wallaby. Conserv Biol 13(3):531–541CrossRefGoogle Scholar
  19. Epps CW, Palsbøll PJ, Wehausen JD, Roderick GK, Ramey RR, McCullough DR (2005) Highways block gene flow and cause a rapid decline in genetic diversity of desert bighorn sheep. Ecol Lett 8(10):1029–1038CrossRefGoogle Scholar
  20. Ernest HB, Vickers TW, Morrison SA, Buchalski MR, Boyce WM (2014) Fractured genetic connectivity threatens a southern California puma (Puma concolor) population. PLoS ONE 9(10):e107985CrossRefPubMedPubMedCentralGoogle Scholar
  21. Esri (2012) ArcGIS Desktop Release 10.1. Environmental Systems Resource Institute, Redlands. http://www.esri.com/
  22. Esri (2014) ArcGIS Desktop Release 10.3. Environmental Systems Resource Institute, Redlands. http://www.esri.com/
  23. Etherington TR, Holland EP (2013) Least-cost path length versus accumulated-cost as connectivity measures. Landscape Ecol 28:1223–1229CrossRefGoogle Scholar
  24. Evanno G, Regnaut S, Goudet J (2005) Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study. Mol Ecol 14(8):2611–2620CrossRefGoogle Scholar
  25. Evans J, Oakleaf J, Cushman S and Theobald D (2014) An ArcGIS toolbox for surface gradient and geomorphometric modeling, version 2.0-0, Laramie. http://evansmurphywix.com/evansspatial
  26. Frantz AC, Bertouille S, Eloy MC, Licoppe A, Chaumont F, Flamand MC (2012) Comparative landscape genetic analyses show a Belgian motorway to be a gene flow barrier for red deer (Cervus elaphus), but not wild boars (Sus scrofa). Mol Ecol 21(14):3445–3457CrossRefPubMedGoogle Scholar
  27. Gustafson KD, Vickers TW, Boyce WM, Ernest HB (2017) A single migrant enhances the genetic diversity of an inbred puma population. R Soc Open Sci 4(5):170115CrossRefPubMedPubMedCentralGoogle Scholar
  28. Harveson PM, Lopez RR, Collier BA, Silvy NJ (2007) Impacts of urbanization on Florida Key deer behavior and population dynamics. Biol Conserv 134(3):321–331CrossRefGoogle Scholar
  29. Herkert JR (1994) The effects of habitat fragmentation on midwestern grassland bird communities. Ecol Appl 4(3):461–471CrossRefGoogle Scholar
  30. Hobbs NT (1996) Modification of ecosystems by ungulates. J Wildl Manage 60:695–713CrossRefGoogle Scholar
  31. Holderegger R, Di Giulio M (2010) The genetic effects of roads: a review of empirical evidence. Basic Appl Ecol 11:522–531CrossRefGoogle Scholar
  32. Hunter RD, Fisher RN, Crooks KR (2003) Landscape-level connectivity in coastal southern California, USA, as assessed through carnivore habitat suitability. Nat Areas J 23(4):302–314Google Scholar
  33. Ito TY, Lhagvasuren B, Tsunekawa A, Shinoda M, Takatsuki S, Buuveibaatar B, Chimeddorj B (2013) Fragmentation of the habitat of wild ungulates by anthropogenic barriers in Mongolia. PLoS ONE 8(2):e56995CrossRefPubMedPubMedCentralGoogle Scholar
  34. Jaquiéry J, Broquet T, Hirzel AH, Yearsley J, Perrin N (2011) Inferring landscape effects on dispersal from genetic distances: how far can we go? Mol Ecol 20(4):692–705CrossRefPubMedGoogle Scholar
  35. Johnson WE, Onorato DP, Roelke ME, Land ED, Cunningham M, Belden RC, McBride R, Jansen D, Lotz M, Shindle D, Howard J (2010) Genetic restoration of the Florida panther. Science 329(5999):1641–1645CrossRefPubMedGoogle Scholar
  36. Jombart T, Devillard S, Balloux F (2010) Discriminant analysis of principal components: a new method for the analysis of genetically structured populations. BMC Genet 11(1):94CrossRefPubMedPubMedCentralGoogle Scholar
  37. Jost L (2008) GST and its relatives do not measure differentiation. Mol Ecol 17(18):4015–4026CrossRefPubMedGoogle Scholar
  38. Kalinowski ST, Taper ML, Marshall TC (2007) Revising how the computer program CERVUS accommodates genotyping error increases success in paternity assignment. Mol Ecol 16:1099–1106CrossRefPubMedPubMedCentralGoogle Scholar
  39. Kamvar ZN, Tabima JF, Grünwald NJ (2014) Poppr: an R package for genetic analysis of populations with clonal, partially clonal, and/or sexual reproduction. PeerJ 2:e281.  https://doi.org/10.7717/peerj.281 CrossRefPubMedPubMedCentralGoogle Scholar
  40. Keller I, Largiader CR (2003) Recent habitat fragmentation caused by major roads leads to reduction of gene flow and loss of genetic variability in ground beetles. Proc R Soc Lond B 270:1513CrossRefGoogle Scholar
  41. Keller LF, Waller DM (2002) Inbreeding effects in wild populations. Trends Ecol Evol 17(5):230–241CrossRefGoogle Scholar
  42. Koen EL, Tosa MI, Nielsen CK, Schauber EM (2017) Does landscape connectivity shape local and global social network structure in white-tailed deer. PLoS ONE 12(3):e0173570CrossRefPubMedPubMedCentralGoogle Scholar
  43. Kolodziej K, Theissinger K, Brün J, Schulz HK, Schulz R (2012) Determination of the minimum number of microsatellite markers for individual genotyping in wild boar (Sus scrofa) using a test with close relatives. Eur J Wildl Res 58(3):621–628CrossRefGoogle Scholar
  44. LANDFIRE (2011) LANDFIRE 1.1.0—existing vegetation height and type layers, LANDFIRE (ed). U.S. Department of Agriculture and U.S. Department of the InteriorGoogle Scholar
  45. Latch EK, Heffelfinger JR, Fike JA, Rhodes J (2009) Species-wide phylogeography of North American mule deer (Odocoileus hemionus): cryptic glacial refugia and postglacial recolonization. Mol Ecol 18(8):1730–1745CrossRefPubMedGoogle Scholar
  46. Latch EK, Reding DM, Heffelfinger JR, Alcalá-Galván CH, Rhodes OE (2014) Range-wide analysis of genetic structure in a widespread, highly mobile species (Odocoileus hemionus) reveals the importance of historical biogeography. Mol Ecol 23(13):3171–3190CrossRefPubMedGoogle Scholar
  47. Lee JS, Ruell EW, Boydston EE, Lyren LM, Alonso RS, Troyer JL, Crooks KR, VandeWoude SUE (2012) Gene flow and pathogen transmission among bobcats (Lynx rufus) in a fragmented urban landscape. Mol Ecol 21(7):1617–1631CrossRefPubMedGoogle Scholar
  48. Mager KH, Colson KE, Hundertmark KJ (2013) High genetic connectivity and introgression from domestic reindeer characterize northern Alaska caribou herds. Conserv Genet 14(6):1111–1123CrossRefGoogle Scholar
  49. McClure ML, Dickson BG, Nicholson KL (2017) Modeling connectivity to identify current and future anthropogenic barriers to movement of large carnivores: a case study in the American Southwest. Ecol Evol 7(11):3762–3772CrossRefPubMedPubMedCentralGoogle Scholar
  50. Mitelberg A (2010) Social structure and genetic connectivity in the San Diego southern mule deer. Master’s thesis, San Diego State University. San Diego, CaliforniaGoogle Scholar
  51. Mitelberg A, Vandergast AG (2016) Non-invasive genetic sampling of southern mule deer (Odocoileus hemionus fuliginatus) reveals limited movement across California State route 67 in San Diego county. West Wildl 3:8–18Google Scholar
  52. Mora C, Sale PF (2011) Ongoing global biodiversity loss and the need to move beyond protected areas: a review of the technical and practical shortcomings of protected areas on land and sea. Mar Ecol Prog Ser 434:251–266CrossRefGoogle Scholar
  53. Munshi-South J, Kharchenko K (2010) Rapid, pervasive genetic differentiation of urban white-footed mouse (Peromyscus leucopus) populations in New York City. Mol Ecol 19(19):4242–4254CrossRefPubMedGoogle Scholar
  54. Newman BJ, Ladd P, Brundrett M, Dixon KW (2013) Effects of habitat fragmentation on plant reproductive success and population viability at the landscape and habitat scale. Biol Cons 159:16–23CrossRefGoogle Scholar
  55. Ng SJ, Dole JW, Sauvajot RM, Riley SP, Valone TJ (2004) Use of highway undercrossings by wildlife in southern California. Biol Conserv 115(3):499–507CrossRefGoogle Scholar
  56. Nicholson MC, Bowyer RT, Kie JG (1997) Habitat selection and survival of mule deer: tradeoffs associated with migration. J Mamm 78(2):483–504CrossRefGoogle Scholar
  57. Peakall R, Smouse PE (2012) GenAlEx 6.5: genetic analysis in Excel. Population genetic software for teaching and research-an update. Bioinformatics 28:2537–2539CrossRefPubMedPubMedCentralGoogle Scholar
  58. Pease KM, Freedman AH, Pollinger JP, McCormack JE, Buermann W, Rodzen J, Banks J, Meredith E, Bleich VC, Schaefer RJ, Jones K (2009) Landscape genetics of California mule deer (Odocoileus hemionus): the roles of ecological and historical factors in generating differentiation. Mol Ecol 18(9):1848–1862CrossRefPubMedGoogle Scholar
  59. Poessel SA, Burdett CL, Boydston EE, Lyren LM, Alonso RS, Fisher RN, Crooks KR (2014) Roads influence movement and home ranges of a fragmentation-sensitive carnivore, the bobcat, in an urban landscape. Biol Conserv 180:224–232CrossRefGoogle Scholar
  60. Pritchard JK, Stephens M, Donnelly P (2000) Inference of population structure using multilocus genotype data. Genetics 155(2):945–959PubMedPubMedCentralGoogle Scholar
  61. R Core Team (2011) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna. ISBN 3-900051-07-0. http://www.R-project.org/
  62. Riley SP, Pollinger JP, Sauvajot RM, York EC, Bromley C, Fuller TK, Wayne RK (2006) FAST-TRACK: a southern California freeway is a physical and social barrier to gene flow in carnivores. Mol Ecol 15(7):1733–1741CrossRefPubMedGoogle Scholar
  63. Riley SP, Serieys LE, Pollinger JP, Sikich JA, Dalbeck L, Wayne RK, Ernest HB (2014) Individual behaviors dominate the dynamics of an urban mountain lion population isolated by roads. Curr Biol 24(17):1989–1994CrossRefPubMedGoogle Scholar
  64. Robinson SJ, Samuel MD, Lopez DL, Shelton P (2012) The walk is never random: subtle landscape effects shape gene flow in a continuous white-tailed deer population in the Midwestern United States. Mol Ecol 21:4190–4205CrossRefPubMedGoogle Scholar
  65. Rollins MG (2009) LANDFIRE: a nationally consistent vegetation, wildland fire, and fuel assessment. Int J Wildl Fire 18:235–249CrossRefGoogle Scholar
  66. Rooney TP, Waller DM (2003) Direct and indirect effects of white-tailed deer in forest ecosystems. For Ecol Manage 181(1):165–176CrossRefGoogle Scholar
  67. Ruell EW, Riley SPD, Douglas MR, Antolin MF, Pollinger JR, Tracey JA, Lyren LM, Boydston EE, Fisher RN, Crooks KR (2012) Urban habitat fragmentation and genetic population structure of bobcats in coastal southern California. Am Midl Nat 168(2):265–280CrossRefGoogle Scholar
  68. Sato JJ, Kawakami T, Tasaka Y, Tamenishi M, Yamaguchi Y (2014) A few decades of habitat fragmentation has reduced population genetic diversity: a case study of landscape genetics of the large Japanese field mouse, Apodemus speciosus. Mamm Study 39(1):1–10CrossRefGoogle Scholar
  69. Shannon CE, Weaver W (1949) The mathematical theory of communication. University Illinois Press, Urbana, p 17Google Scholar
  70. Spencer W, Beier P, Penrod K, Winters P, Rustigian-Romsos JS, Parisi M and Pettler A (2010) California essential habitat connectivity project: a strategy for conserving a connected California. Prepared for California Department of Transportation, California Department of Fish and Game, and Federal Highways AdministrationGoogle Scholar
  71. U.S. Geological Survey (2009) National Elevation Dataset (NED)–Raster digital data, U.S. Geological Survey (ed). The National Map, Sioux FallsGoogle Scholar
  72. Van Oosterhout C, Hutchinson WF, Wills DP, Shipley P (2004) MICRO-CHECKER: software for identifying and correcting genotyping errors in microsatellite data. Mol Ecol Res 4(3):535–538Google Scholar
  73. Van Strien MJ, Keller D, Holderegger R (2012) A new analytical approach to landscape genetic modelling: least-cost transect analysis and linear mixed models. Mol Ecol 21:4010–4023CrossRefGoogle Scholar
  74. Vandergast AG, Bohonak AJ, Weissman DB, Fisher RN (2007) Understanding the genetic effects of recent habitat fragmentation in the context of evolutionary history: phylogeography and landscape genetics of a southern California endemic Jerusalem cricket (Orthoptera: Stenopelmatidae: Stenopelmatus). Mol Ecol 16(5):977–992CrossRefPubMedGoogle Scholar
  75. Vickers TW, Sanchez JN, Johnson CK, Morrison SA, Botta R, Smith T, Cohen BS, Huber PR, Ernest HB, Boyce WM (2015) Survival and mortality of pumas (Puma concolor) in a fragmented, urbanizing landscape. PLoS ONE 10(7):e0131490CrossRefPubMedPubMedCentralGoogle Scholar
  76. Wang M, Schreiber A (2001) The impact of habitat fragmentation and social structure on the population genetics of roe deer (Capreolus capreolus L.) in Central Europe. Heredity 86(6):703–715CrossRefPubMedGoogle Scholar
  77. Waples RS, Do CHI (2008) LDNE: a program for estimating effective population size from data on linkage disequilibrium. Mol Ecol Res 8(4):753–756CrossRefGoogle Scholar
  78. Wilson, MC, Chen, XY, Corlett, RT, Didham, RK, Ding, P, Holt, RD, Holyoak, M, Hu, G, Hughes, AC, Jiang, L and Laurance, WF (2016) Habitat fragmentation and biodiversity conservation: key findings and future challenges. 219–227Google Scholar
  79. Winter DJ (2012) MMOD: an R library for the calculation of population differentiation statistics. Mol Ecol Res 12(6):1158–1160CrossRefGoogle Scholar
  80. Yokoyama R, Shirasawa M, Pike RJ (2002) Visualizing topography by openness: a new application of image processing to digital elevation models. Photogramm Eng Remote Sens 68:257–266Google Scholar

Copyright information

© This is a U.S. government work and its text is not subject to copyright protection in the United States; however, its text may be subject to foreign copyright protection 2019

Authors and Affiliations

  1. 1.Department of Ecology and Evolutionary BiologyUniversity of CaliforniaLos AngelesUSA
  2. 2.U.S. Geological SurveySouthwest Biological Science CenterFlagstaffUSA
  3. 3.U.S. Geological SurveyThousand OaksUSA

Personalised recommendations