Introduction

Patterns in genetic diversity and differentiation are the result of a combination of historical and contemporary effects on evolutionary forces including genetic drift, gene flow and natural selection (Allendorf 1983; Avise 2000; Hedrick 2000; Hartl and Clark 2007). Historical events, including range contraction or expansion in conjunction with large-scale landscape changes (e.g. glacial periods) initially shape patterns of genetic diversity (Hewitt 2000); however, contemporary gene flow and anthropogenic habitat changes continue to alter the genetic landscape of species’ ranges and population structure (Lande 1988). Understanding the factors that influence genetic structure and how they are interrelated is crucial to species conservation and management (Lande 1988; Amos and Harwood 1998).

Species with expansive ranges, long-distance dispersal capabilities, and continuous distributions can benefit from analyses of genetic diversity to inform large-scale conservation planning. For example, in most cases species are managed with respect to government-imposed (e.g. state or province) boundaries. However, mobile species do not recognize such artificial boundaries and thus management at these levels may not be appropriate. As such, genetic analyses can clarify the most suitable level for management. Studies of wide-ranging and vagile species are numerous and have focused on a host of vertebrates such as: marine mammals (Mendez et al. 2008), terrestrial mammals (von Holdt et al. 2010), fishes (Ward et al. 1997; Castro et al. 2007), birds (Hull et al. 2008), and reptiles (Shillinger et al. 2008). The goal of such work was to effectively manage genetic diversity given the biology of the species and to consider the conservation implications of such results.

Examinations of genetic diversity and population structure in fields has been useful to large-scale management efforts because some felids are economically important, many have threatened conservation status, and because they function in many ecosystems as apex carnivores (Ray et al. 2005; Roemer et al. 2009). Schwartz et al. (2002) examined large-scale spatial synchrony in population dynamics of Canada lynx (Lynx canadensis). Estimates of gene flow indicated that large numbers of lynx disperse long distances from the core of their range, creating a wave of immigrants that drive cycle-like synchrony in Western lynx populations. To maintain population connectivity, Schwartz et al. (2002) concluded that international efforts would be necessary to appropriately manage Canada lynx populations. Conversely, molecular analyses of Eurasian lynxes (L. lynx) from Finland and the Baltic states indicated that populations from the two regions were genetically distinct and might warrant separate management (Hellborg et al. 2002). In addition, restricted gene flow in jaguars (Panthera onca) due to major geographical barriers like the Amazon River and Darien Straits produced measurable genetic differentiation (Eizirik et al. 2001). Eizirik et al. (2001) recommended that historical barriers to dispersal should be used to define operational conservation units in future management of this species.

Bobcats (L. rufus (Schreber, 1777)) are solitary, highly mobile, polygamous carnivores (Larivière and Walton 1997; Anderson and Lovallo 2003) and are widely distributed in North America, ranging throughout the contiguous U.S. and portions of Canada and Mexico (Lopez-Gonzalez et al. 1998; Anderson and Lovallo 2003). Twelve subspecies are recognized within their distribution (Hall 1981). The classification of subspecies is based mostly on differences in pelage coloration (Young 1958); however, Samson (1979) confirmed these taxa using multivariate analyses of cranial characters. Bobcats are considered habitat generalists, but typically require woody cover for a variety of activities, including dispersal (Anderson and Lovallo 2003).

Bobcats are considered furbearers or game mammals in North America and as such, their management is based primarily on regulations prescribed by state or provincial wildlife conservation agencies. During the last few decades, bobcats have increased in many U.S. states (Woolf and Hubert 1998), and continue to do so (Roberts and Crimmins 2010). The U.S. allows harvest of bobcats in 39 states while others prohibit it due to their relative rarity, especially in the Midwest (spp L. r. rufus) where land-use practices (i.e., intensive row-crop agriculture) have rendered the landscape less suitable for bobcats. In Canada, bobcats are at their northern range limit, but are hunted or trapped in seven of the eight provinces in which they are native. Bobcats are considered common in the southern regions of provinces in Western Canada (e.g. British Columbia, Alberta) and in the Atlantic provinces (e.g. New Brunswick and Nova Scotia). Hunting and trapping has been halted in Quebec since 1991 due to large declines in harvest (Lavoie et al. 2010) but populations have since been found to be stable or increasing (Roberts and Crimmins 2010). In Mexico, five states permit hunting; there is no evidence of declines in central and southern Mexico (Hansen 2007).

By virtue of its high vagility, broad distribution, international economic importance, and role in many North American ecosystems as an apex carnivore, the bobcat serves as an excellent model for exploring questions related to partitioning of genetic diversity, degree of genetic differentiation and how these relate to current management practices. Specifically, management based on patterns in genetic diversity and differentiation (rather than the current management strategy) may be more appropriate since bobcats most likely do not adhere to artificial boundaries such as state and provincial lines. The ability of bobcats to disperse long distances (Knick and Bailey 1986; Kamler et al. 2000; Nielsen and Woolf 2003; Johnson et al. 2010) suggests that gene flow is high and genetic differentiation low. However, some studies indicate that gene flow has been restricted at smaller scales due to natural and anthropogenic barriers (Riley et al. 2006; Millions and Swanson 2007). By understanding range-wide patterns in genetic diversity, recommendations for bobcat management can be made that maintain both historical and present connections in order to preserve the maximal amount of genetic diversity necessary for the species to continue to thrive and adapt.

Genetic structure in bobcats has been assessed at small scales (Croteau et al. 2010; Janecka et al. 2006, 2007; Millions and Swanson 2006, 2007; Riley et al. 2006); however, range-wide population genetic structure and its relevance to current management policy have not been assessed. In this study, we used microsatellite and mitochondrial DNA to examine genetic structure of bobcats in North America. The objectives were to identify patterns of genetic diversity and genetic differentiation among populations to gain an understanding of how past demographic events and current population isolation have influenced these aspects of population structure. The implications of such research may aid in future management of this wide-ranging species.

Methods

Sample area and DNA extraction

Ear, tongue, or liver tissues were collected from 587 wild bobcats during 1995–2006 (Fig. 1). Tissue samples were collected during field research and harvest in various states/provinces, by several universities, state agencies and individuals, and were stored at room temperature in a 20 % DMSO solution saturated with NaCl until DNA extraction. Liver segments obtained by necropsy were frozen at −70 °C. Total genomic DNA was extracted from these tissues using the DNeasy tissue kit (Qiagen Inc., Valencia, CA).

Fig. 1
figure 1

Map of locations for bobcats sampled from 1995 to 2006 throughout North America. Areas shaded in grey indicate county or state where tissue samples were taken

Full size image

Samples were obtained from 16 U.S. states and two Canadian provinces (Table 1): Arkansas (AR), California (CA), Florida (FL), Illinois (IL), Indiana (IN), Kentucky (KY), Louisiana (LA), Maine (ME), Michigan [Upper and Lower Peninsulas: UP and LP, respectively], Minnesota (MN), Missouri (MO), New Brunswick (NB), North Dakota (ND), Nova Scotia (NS), Nevada (NV), Texas (TX), Wisconsin (WI), and Wyoming (WY). Samples from Michigan were separated into UP and LP because a large water barrier isolates these regions, and because previous studies demonstrated significant genetic differences between these locations (Millions and Swanson 2006, 2007). As a result of this division, our study included 19 populations representing eight of the twelve putative subspecies (Hall 1981). In most cases, county-level location information was provided for the tissue sample, however, given the method of tissue collection, in some cases, only state-level location information was available (e.g. MN, LA). Because bobcats are managed by state or provincial regulations, we initially defined sample populations based upon these state/provincial boundaries and populations were named using the abbreviation for that state/province (except in the case of Michigan wherein populations were named LP and UP).

Table 1 Number of individuals surveyed (n), observed heterozygosity (Ho), expected heterozygosity (He), total allelic richness and regional cluster as indicated by STRUCTURE (Pritchard et al. 2000) for 17 populations of bobcats in North America surveyed at 10 microsatellite loci
Full size table

Marker development and analysis

We used both microsatellite markers and mitochondrial DNA to address our research questions. Ten microsatellite loci developed in other felid species were amplified using polymerase chain reaction (PCR). Six microsatellite primer pairs were originally developed for the domestic cat (Felis catus; FCA023, FCA043, FCA045, FCA077, FCA090, FCA096; Menotti-Raymond and O’Brien 1995; Menotti-Raymond et al. 1999), three were designed for the Canada lynx (LC109, LC110, LC111; Carmichael et al. 2000) and one was designed for the bobcat (BCE5T; Faircloth et al. 2005). See Croteau et al. (2010) for PCR reaction conditions and thermal cycling profiles for all microsatellite primer sets.

PCR products were diluted in loading buffer (1:1) containing deionized formamide, blue dextran EDTA, and ROX-labeled size standard (The Gel Company MRK-400 size standard; The Gel Company, San Francisco, CA), and resolved on a 36-cm-long 5 % denaturing Long Ranger gel (Cambrex Bio Science, East Rutherford, NJ) and run at 2,500 scans/h for 2.5 h using an ABI 377 automated genotyper. Gel images were analyzed using Genescan version 3.1.2 software and alleles were defined using Genotyper version 2.5 genotyping software (PE Applied Biosystems, Applera Corporation, Valencia, CA). Genotypes were typically re-run at least three times to ensure there were no genotyping errors. MICROCHECKER version 2.3.3 (van Oosterhout et al. 2004) was used to check for the occurrence of null alleles.

The control region of the mitochondrial DNA genome was amplified by PCR using two primers developed for this study (CatPro 5′-CAAGGAAGAAGCAACAGC3′ and Cat 12S 5′ TRRAGGGCATTTTCACCG3′) anchoring at positions 16,249 and 1,018, respectively, of the domestic cat (Felis catus) mtDNA genome (GenBank Accession Number U20753; Lopez et al. 1996), and amplifying approximately 1,742 base pairs (bp). A subset of samples was utilized for DNA sequencing, rather than the entire sample set of 587 due to technical and time limitations. Specifically, a small sample of individuals was selected from each population (e.g. state or province) to ensure that a sufficient sample size was obtained and the geographic area of the sample area was accurately represented. However, in some cases sample size was lower than intended due to poor sequence resolution. PCR reactions were performed in 50 μl volumes using Thermoprime ABGene 2X concentrated PCR mastermix (Thermo Fisher Scientific, Rockford, IL), 0.5 μmol of each primer and 30–50 ng of DNA per reaction. Conditions for amplification consisted of a denaturing step of 94 °C for 2 min, followed by 35 cycles of 94 °C for 1 min, 60 °C for 1 min, and 72 °C for 1 min, with a final extension of 72 °C for 5 min. PCR products were purified using QIAquick PCR purification kits (Qiagen Inc., Valencia, CA). Three primers were designed to sequence portions of the first hypervariable segment (HVS-1) and the conserved central portion of the bobcat control region: forward primer LruFwd3 5′-CACTATCAGCACCCAAAGC-3′ mapping to position 16,269 of the domestic cat mitochondrial genome (position 3 in the fragment); reverse primer, LruRev2 5′-GACCCCGCATAGAGAATAAG-3′ mapping to position 160 in domestic cat (position 882 of the fragment); and internal primer LruRevSeq 5′-AGGATTGCTGGTTTCTCG-3′ mapping to position 16,862 in domestic cat (position 574 of the fragment). These primers produced 813 base pairs (bp) of sequence. Cycle sequencing reactions were performed in an Eppendorf Mastercycler using a 96 °C denaturation step for 4 min, followed by 99 cycles of 96 °C for 30 s, 50 °C for 15 s, and 60 °C for 4 min. Fragments were purified using Sephadex G-50 columns, electrophoresed on a 36 cm long, 5 % denaturing Long Ranger gel (Cambrex Bio Science, east Rutherford, NJ), and run at 1,200 scans/h for 7 h using an ABI 377 automated sequencer (Applied Biosystems). Gel images were analyzed using Genescan version 3.1.2 software (PE Applied Biosystems, Applera Corporation, Valencia, CA); sequences were examined and aligned using SEQUENCHER version 4.7 (Gene Codes Corporation).

The portion of the control region that was amplified for this study contains a region of tandem repeats (Hoelzel et al. 1994) that vary in size, number of variable sites, and number of repeats (Lopez et al. 1996; Freeman et al. 2001). Due to the uncertainty of aligning the repeats among individuals, only 464 bp of non-repetitive sequence flanking the repeat region was used for analysis. In cases where alignment was difficult due to poor amplification and/or sequence resolution, individuals were re-sequenced for that fragment and re-aligned. This resulted in the re-sequencing of approximately 50 % of individuals for any given fragment to ensure correct alignment.

Analysis of genetic diversity

Patterns of contemporary genetic variation within each population was assessed for microsatellites by calculating observed and expected heterozygosity in GENETIX 4.05 (Belkhir et al. 2001) and polymorphic characteristics of the microsatellites, such as the number of alleles, allelic richness adjusted for sample size, and allele frequencies, were calculated using FSTAT 2.9.3.2 (Goudet et al. 2002). Linkage disequilibrium (LD) between pairs of loci across all populations was assessed using GENEPOP 4.0 (Rousset 2008). Significance values for deviations from Hardy–Weinberg equilibrium (HWE) were calculated using an exact test (P < 0.05; Guo and Thompson 1992; Raymond and Rousset 1995) as implemented in GENEPOP 4.0 (Rousset 2008). This method can identify significant deviations in HWE resulting in an excess or deficiency of homozygotes as indicated by positive or negative FIS scores. Tests for both LD and HWE were corrected for multiple comparisons using a sequential Bonferroni adjustment (Holm 1979; Rice 1989).

Patterns of historical genetic diversity for sample areas were assessed for the mitochondrial control region using the numbers of variable sites and haplotypes, and transitions and transversions between haplotypes in DnaSP version 4 (Rozas et al. 2003). Haplotype diversity (h) and nucleotide diversity (π) within sample areas were calculated using ARLEQUIN version 3.11 (Excoffier et al. 2005). Both haplotype and nucleotide diversity are useful for inferring population expansion, demographic bottlenecks, Hardy–Weinberg equilibrium, and admixture of samples (Rogers and Harpending 1992; Avise 2000). Matrices of pairwise differences between haplotypes were computed for all analyses of molecular variation.

Analysis of historical demography

NETWORK version 4.5 (Bandelt et al. 1999) was employed to generate a median joining network. We used a network because phylogenetic trees depict sequence haplotypes on terminal branches, with internal branches representing extinct ancestral haplotypes. In intraspecific samples ancestral haplotypes are often still present, and networks are therefore a more accurate depiction of haplotype relationships (Smouse 1998).

Several neutrality statistics were used to test for historical population growth, contraction, or stability. The expansion coefficient (S/d) was calculated following Peck and Congdon (2004). S/d can indicate a population expansion or equilibrium, where large values of S/d suggest recent population expansion and small values suggest stable long-term size (Peck and Congdon 2004). To detect past population expansion, ARLEQUIN version 3.11 (Excoffier et al. 2005) was employed to calculate the Fu’s FS statistic (Fu 1997) and Tajima’s D statistic (Tajima 1989). We used 100,000 simulations of Hudson’s (1990) coalescent simulation algorithm to test the significance of FS and D under the null hypothesis of selective neutrality and population equilibrium. A significantly large negative value for FS suggests an excess of recent mutations and is taken as evidence against neutrality of mutations and population equilibrium (Fu 1997). Population expansions can cause a significant negative departure of Tajima’s D from zero (Tajima 1989). Fu’s FS is thought to be more sensitive to population expansion than Tajima’s D (Fu and Li 1993), but both statistics were implemented to confirm selective neutrality. Fu and Li’s D* and F* statistics were calculated in DnaSP version 4 (Rozas et al. 2003), significant values are associated with selection (Fu and Li 1993). To distinguish between the effects of background selection and population expansion, a comparison of FS to D* and F* statistics was performed. A significant FS with non-significant D* and F* implies expansion whereas the reverse indicates selection (Fu 1997).

A mismatch distribution and raggedness statistic (rg) were calculated in ARLEQUIN version 3.11 (Excoffier et al. 2005). Mismatch distributions describe the frequency of pairwise nucleotide differences among individuals. A unimodal distribution approximating a Poisson curve indicates a population that has undergone recent expansion (Rogers and Harpending 1992; Harpending et al. 1993). A multimodal distribution is expected in a population at mutation-drift equilibrium (Slatkin and Hudson 1991). Furthermore, population growth will generate waves in the distribution that are seldom seen in equilibrium populations (Rogers and Harpending 1992). The value of rg also allows differentiation of populations that are expanding from those at equilibrium. Large values of rg indicate populations that are stationary with multimodal mismatch distributions (Harpending et al. 1993). In ARLEQUIN version 3.11, the validity of the expansion model is tested by means of the sudden expansion model using the sum of squared deviations (SSD) between the observed and the expected mismatch as a test statistic. A P value is obtained by calculating the number of simulated SSD that are ≥ the observed SSD, with a P value ≤0.05 taken as evidence of departure from the expansion model. The significance test for rg is similar to that for the mismatch distribution.

Analysis of population structure

Current population structure was examined using STRUCTURE (version 2.3.1; Pritchard et al. 2000, 2009) to infer the number of genetic clusters (K) in our sample and to assign individuals to these clusters. STRUCTURE employs a Bayesian clustering algorithm to determine the posterior probability of the data for a given number of genetic groups and finds the optimal K assuming Hardy–Weinberg and linkage equilibrium.

We estimated K in STRUCTURE by performing ten independent runs for each K (K = 1–18), using 100,000 burn-in steps to ensure likelihood convergence, and 750,000 Markov Chain Monte Carlo (MCMC) repetitions. We did not include prior population delineation information and assumed correlated allele frequencies and population admixture. The posterior probability was then calculated for each value of K using the log-likelihood of the data, and we chose the optimal K using an ad hoc statistic ∆K, based on the second order rate of change in the log probability of the data between successive K values (Evanno et al. 2005). We assessed the proportion of membership/assignment of each individual to each genetic cluster using CLUMPP (Jakobbson and Rosenberg 2007), for the ∆K that best explained the data.

Current population differentiation (FST) was measured by Weir and Cockerham’s (1984) unbiased estimator θ, and determined for pairs of populations and STRUCTURE identified clusters (based on microsatellite data) using GENEPOP 4.0 (Rousset 2008). Analysis of Molecular Variation (AMOVA) of mitochondrial control region sequences was implemented to determine the extent of historical population structure in bobcats using ARLEQUIN version 3.11 (Excoffier et al. 2005) to calculate pairwise ϕ-statistics (Excoffier et al. 1992) between populations and to quantify the percentage of variation within and among those populations. The topology of the network and STRUCTURE identified clusters were used in combination to group populations together for AMOVA analyses. The groupings that maximize the among-population variance component (ϕST) and are statistically significant indicate the most parsimonious geographical subdivisions (Eizirik et al. 2001; Liu et al. 2007). The significance of ϕST was tested using 10,000 permutations (ϕST: permuting haplotypes among populations). Pairwise tests of FST and ϕST were corrected for multiple comparisons using sequential Bonferroni (Holm 1979; Rice 1989). Mantel tests (Mantel 1967) were used to test for isolation by distance (IBD) between STRUCTURE-identified clusters using ARLEQUIN version 3.11 (Excoffier et al. 2005). Geographic distances in kilometers (km) were converted to FST/(1 − FST) in GENEPOP 4.0 using 100,000 permutations. Geographic distances were measured between sample locations using ArcView GIS 3.3 (Environmental Systems Research Institute Inc. 2002). When the county from which a bobcat was sampled was unknown, the centroid of the state/province was used for that locale; where the county location was known, the centroid of the counties/townships sampled was used to calculate an overall centroid representing all of the samples collected from that area. These provided a reference point for measuring geographic distance. Given that IBD is used here for inter-population analyses, both methods should result in an accurate and reliable point for calculations of geographic distance between sample areas.

Principal Coordinate Analysis (PCA) as implemented in GENALEX 6 (Peakall and Smouse 2006) was used to further identify major patterns of genetic differentiation in the microsatellites. Pairwise FST between populations was selected to form the basis of the PCA and was calculated using those methods described in GENALEX 6 (Peakall and Smouse 2006).

Results

Patterns of genetic diversity

Multilocus genotypes for 581 individuals (of 587 total individuals) were obtained representing 17 sample areas (Table 1). Approximately 4.4 % of genotypes were unobtainable within that sample. All microsatellite loci were polymorphic, with an average of 16 alleles per locus, ranging from 11 to 28 alleles. Observed heterozygosity ranged from 0.53 to 0.79 (Table 1). Allelic richness values averaged 4.5, with LP and ME exhibiting the lowest richness (also corresponding to the lowest levels of observed heterozygosity) and Western populations exhibiting the highest values (Table 1). Of 170 pairwise tests for deviations from Hardy–Weinberg equilibrium, two were significant after Bonferroni correction (adjusted α = 0.00029), both occurring in the IL population and both indicating a slight homozygote excess (FIS = 0.018 for LC109; FIS = 0.086 for BCE5T). After Bonferroni correction, two significant linkages between pairs of loci across all populations were found of the 45 pairwise comparisons performed (P < 0.001). Examining LD on a location-by-location basis detected one significant score in IL, and three in ND (α = 0.00029). MICROCHECKER did not detect any null alleles.

Tests for LD and HWE were performed within each cluster. Tests for LD yielded a few significant values (e.g. in the West between FCA023/LC110, and in the East between FCA043/FCA096 and FCA077/LC110; α = 0.00111). However no pattern in the significant LD scores for specific loci was revealed. Some significant values in all clusters were found when testing for HWE. For example in the East two loci were out of HWE (all homozygous excess), in the Midwest two loci were out of HWE (one homozygous excess, one deficiency), and one locus was out of HWE in the West (homozygous excess). No pattern of significant HWE scores was evident. Significance for HWE was tested at α = 0.00166 as per the Bonferroni correction.

Of the 587 individuals, 185 were sequenced for a portion of the mitochondrial control region, and represented 19 sample areas (Table 2). Thirty-eight unique haplotypes (GenBank Accession Numbers FJ204789–FJ204826) were identified, with 35 of the 464 sites polymorphic (7.3 %) and 26 of which were parsimony informative. Of the 35 pairwise substitutions, we found 31 transitions, three transversions, and one insertion-deletion. Among all samples, h was 0.817 (±0.025) and π was 0.0077. The mean number of nucleotide differences between haplotypes was 3.79 and ranged from 2.44 to 7.56 (Table 2).

Table 2 Population locations, sample sizes (n), number of haplotypes (H#), haplotype diversity (h), percent nucleotide diversity (π), and mean population pairwise differences (mean pop pair diffs) for mtDNA control region variation within 19 areas in 185 bobcats sampled from 1995 to 2006 throughout North America
Full size table

Historical demography

The haplotype network illustrated a clear distinction between haplotypes found in most Western bobcats and those from the rest of the range. The branching pattern in Eastern/Midwestern haplotypes approximated a star-like shape, with a common haplotype having several rare haplotypes that differed from it by a single substitution. Western bobcats were separated by several mutational differences from Eastern/Midwestern bobcats with a common haplotype (Fig. 2). Overall, the network approached a “dumbbell-like” shape, which has been suggested to occur in a species that has experienced two population expansions (Avise 2000). Haplotype 1 was most common in the central portion of the range (n = 73 individuals) and included 79.5 % of individuals from the Midwestern populations (IL, IN, KY, LP, MN, MO, UP and WI), 1.4 % of individuals from the Western populations (ND), 13.6 % of individuals from the Eastern populations (ME and NS), and the remaining 5.5 % from other populations (AR, LA and FL). Interestingly, 58 % of the individuals possessing haplotype 2 (the second most numerous haplotype) were from Eastern populations (ME, NB, and NS) and haplotype 26 was the most common Western haplotype appearing in eight individuals from three Western populations. Also, some of the one-step branches derived from haplotype 1 led to haplotypes that occurred in widely dispersed populations. For example, haplotype 3 was found in FL, IL, LA, MO, MN, and TX.

Fig. 2
figure 2

Median joining network depicting relationships among 38 mitochondrial control region haplotypes among 19 areas in 185 bobcats sampled from 1995 to 2006 throughout North America. Haplotypes are represented by circles and areas of circles are proportional to haplotype frequency. Cross marks on branches define the number of substitutions between connected haplotypes, and numbers indicate haplotype

Full size image

The network suggested two separate population expansions occurred (Avise 2000), one comprising Western haplotypes (“Left Side”) and another comprising all other haplotypes (mostly Midwestern/Eastern but some western samples; “Right Side”). For clarification the “Right Side” includes those Western samples associated with the right side of the network. The mismatch distributions for those haplotypes on the “Left” and “Right” sides of the network fit the sudden expansion model (P = 0.500 and P = 0.557 respectively). Also, the rg index was low for each mismatch distribution (rg = 0.045, P = 0.448 for “Left Side”; rg = 0.032, P = 0.783 for “Right Side”). Tajima’s D was negative for both areas but not significantly different from 0 (D = - 0.52, P = 0.342 for “Left Side”; D = - 1.39, P = 0.063 for “Right Side”), whereas Fu’s FS was significantly negative in both cases (FS = −7.50, P < 0.000001 for “Left Side”; FS = −27.1, P < 0.000001 for Right Side”) indicating a departure from population stability and further supporting the hypothesis of recent expansion. Furthermore, Fu and Li’s F* and D* were not significant for either side (F* = 0.125, P > 0.10, D* = 0.253, P > 0.10 for “Left Side”; F* = −2.29; 0.10 > P > 0.05, D* = −2.18, 0.10 > P > 0.05, for “Right Side”), suggesting that selection was not responsible for the observed pattern of demography. The expansion coefficient was moderate for the “Left Side” (S/d = 4.51) and large for the “Right Side” (S/d = 11.11).

Population structure

Determining K from STRUCTURE using the ad hoc statistic ∆K clearly indicated that K is equal to 3 (Fig. 3). At K = 3, mean α varied very little between runs (mean α = 0.141–0.147) and there was clear asymmetric assignment of individuals to each cluster (Pritchard et al. 2009). Assignments made by CLUMPP to the three clusters designated by STRUCTURE included one cluster containing 95, 92, 90, and 71 % from WY, NV, ND, and TX respectively, another cluster containing 92, 91, 77, 71, 50, and 36 % of individuals from MO, KY, IN, IL, UP and WI respectively and a third cluster containing 100, 100, 92, 96, 89, 65, 59, and 36 % of individuals from NS, FL, ME, LP, NB, MN, LA and WI respectively (Fig. 4 and Online Resource 1). Percentages are based on a majority-rule assignment. These three clusters will be referred to as West which includes sample areas WY, NV, ND, and TX, Midwest which includes sample areas MO, KY, IN, IL, UP, and WI and East which includes sample areas NS, FL, ME, LP, NB, MN, and LA based on the origin of the majority of individuals assigning to each. Given the equal assignment of WI individuals to two clusters, upon looking further at the data, WI was placed in the Midwest cluster because individuals more strongly assigned to this cluster than to the East cluster. In a few cases, a large percentage of individuals within a given population assigned to two or more regional clusters. For example, 41 % of individuals from LA also assigned to Midwest, 40 % of individuals from UP also assigned to East, 35 % of individuals from MN also assigned to Midwest, and 27 % of individuals from WI also assigned to the West (Online Resource 1). For all populations with a large number of individuals assigning to other clusters, there was no spatial segregation of those individuals. However, this could not be determined for MN or LA due to the lack of specific location information for those populations.

Fig. 3
figure 3

Number of populations (K) estimated using STRUCTURE (Pritchard et al. 2000) for bobcats sampled from 1995 to 2006 throughout North America. Determination of the number of populations (K) based on the rate of change (ΔK) in the log probability of the data (as determined by STRUCTURE) between successive K values. The rate of change in log-likelihood values (ΔK) was estimated using the method of Evanno et al. (2005) where the maximum ΔK indicates the most likely number of populations

Full size image
Fig. 4
figure 4

Percentage of individuals assigned to each of the three clusters using majority-rule consensus and microsatellite data, as identified by STRUCTURE and determined by the method of Evanno et al. (2005) for bobcats from each population sampled from 1995 to 2006 throughout North America

Full size image

Figure 3 illustrates moderately high ∆K, at K = 4, 6, 10 and 12, indicating potential hierarchical structuring (Pritchard et al. 2009). Given this possibility, subsequent STRUCTURE runs using the identical parameters as the previous runs were performed within each of the three clusters to elucidate whether these clusters could be further subdivided. We used the ad hoc statistic ∆K to reveal the number of clusters. These analyses subdivided the West cluster into two clusters; however, all populations assigned to each of the two clusters with almost equal frequency. The Midwest was subdivided into five clusters, with KY and MO assigned to one cluster (II), and WI and UP assigned to one cluster (IV). IL and IN had weak assignment probabilities and assigned to the same four clusters (I, II, IV, V). The East was subdivided into two clusters with FL, LA, LP, and MN assigning to one cluster and ME, NB and NS assigning to the other cluster.

FST between the West and Midwest was 0.0542, West and East 0.0862, and Midwest and East 0.0373. FST between the three clusters was moderate and significant (P < 0.000001). Mantel tests did not support significant isolation by distance within the East cluster (P = 0.417), Midwest cluster (P = 0.68), or West cluster (P = 0.074). For comparison within clusters, population pairwise FST between the 17 populations ranged from 0 to 0.159. Using the Bonferroni correction (adjusted α = 0.00037) six comparisons were not significant (ME and NB; WI and MN, WI and FL, WI and TX, WI and UP, and MN and LA; Online Resource 2). Population pair-wise FST values were used for the PCA plot and illustrated a clear separation between STRUCTURE identified clusters. WY, NV, and ND from the West cluster grouped closely together with TX plotted further away but in the same general area. Sample areas from the Midwest cluster grouped tightly together with the exception of WI. Sample areas in the East cluster had a much broader distribution than the other clusters, with those areas further East grouping together (Fig. 5). Overall, the first and second axes explained 38 and 20 % of genetic variation respectively.

Fig. 5
figure 5

PCA plot indicating genetic distances between sample areas of bobcats sampled from 1995 to 2006 throughout North America. PCA is based on pair-wise FST values using microsatellite data. Circles indicate those sample areas that clustered to West, triangles those that clustered to East, and diamonds those that clustered to the Midwest. Overall, the first and second axes explained 38 and 20 % of genetic variation respectively

Full size image

Pairwise comparisons of ϕST among sampled areas ranged from −0.414 to 0.964, with 25.1 % of those significant after Bonferroni correction (adjusted α = 0.00029; Online Resource 3). The highest significant ϕST values occurred between Western populations (CA, WY, NV, and ND) and all other populations. Hierarchical analysis of genetic variation using AMOVA revealed significant genetic structure among all populations (i.e. no grouping of populations) with ϕST of 0.347 (P < 0.00001; Table 3). When groups of populations were used in AMOVA analyses, the grouping based on the topology of the network maximized ϕST [i.e. Western populations (NV, WY, CA, ND) versus the remaining populations; ϕST = 0.546; Table 3]. AMOVA using three groups of populations (West versus Midwest versus East as revealed by STRUCTURE) resulted in a ϕST of 0.329 (P < 0.00001). AMOVA using four, five and six groups of populations did not produce higher ϕST values (Table 3). A Mantel test detected significant correlation between genetic and geographic distance when comparing all sample sites using ϕST (P = 0.0001, r = 0.510).

Table 3 Number of groups defined, source of variation, percent variation, values of ϕST, and p values (P) determined by Analysis of Molecular Variation (AMOVA), for the mtDNA control region among 19 areas in 185 bobcats sampled from 1995 to 2006 throughout North America. Population acronyms are listed in the text
Full size table

Discussion

Both microsatellites and mitochondrial DNA confirmed little genetic structure over a broad range in bobcats. However, significant regional genetic differentiation was revealed between sample areas in the West and East/Midwest using mitochondrial DNA, and the West, Midwest and East using microsatellites. The regional population structure in bobcats was most likely established by historical processes that include separation via regional barriers and repeated expansions and contractions in population size resulting in population bottlenecks and genetic drift. The phylogeographic structure determined by this study suggests that bobcats as a species are mobile with no long-standing barriers to gene flow (Avise 2000) within regions. However, over time contemporary processes may have acted upon the historical genetic structure to initiate differentiation between the Midwest and East.

Population history and structure

It is evident from the “dumbbell” shaped network that bobcats have experienced two population expansions (Avise 2000). Smaller π estimates in Midwestern and Eastern North America compared to Western areas indicate that expansion took place in the West first with subsequent expansion in the Midwest and East (Zink et al. 2000). It is plausible that during the Pleistocene, Western populations of bobcats existed south of glacial ice and subsequently expanded into central and Eastern areas of the U.S. following glacier retreat. Hewitt (1996) suggested that organisms that are mobile and capable of long-distance dispersal are best able to track climate changes. Given the mobility of bobcats, it is probable that during periods of suitable climate conditions they gradually increased their distribution by dispersing into new areas. The glacial and interglacial cycles that occurred during the Pleistocene would have resulted in continual expansions and contractions of populations in the Midwestern and Eastern US. As a result, multiple extirpations and founder events would have occurred in northern-most Midwestern and Eastern areas because they were the last areas to be exposed due to glacial retreat. The continual cycle of colonization and extinction would ultimately lead to differentiation from the West. Additionally, it appears that dispersal back to the West after colonization of the Midwest/East was limited, suggesting that barriers to dispersal may have formed over time. This is supported by the moderate microsatellite FST values between the West and Midwest (0.0542) and West and East (0.0862), and the few to no individuals assigning to the West from either the East or Midwest (outside of WI; Fig. 4; Online Resource 2). It is likely that the Great Plains or large rivers (e.g. Mississippi) reinforced regional differentiation by limiting dispersal back into the West. Additionally, positive isolation by distance was evident over the entire study area signifying that gene flow is increasingly less likely at great distances. The additive effect of geographic barriers and large distances between populations could have further enhanced the patterns of population structure found here.

The analyses of microsatellite data corroborate the control region population structure analyses to a degree, identifying the West as discrete but also distinguishing between Midwest and East. This result indicates that current processes are likely maintaining the historically present geographic structure (i.e. structure between the West and Midwest/East); however, the microsatellite data also indicate subdivision within the Midwest/East. The increased structure observed with microsatellite data is in contrast to what one would expect given the philopatric nature of female bobcats (Croteau et al. 2010; Janecka et al. 2007) and maternal inheritance of mtDNA. This result might reflect increased power in the microsatellites in detecting population structure or incomplete lineage sorting in mtDNA. Alternatively, it is likely that habitat fragmentation in the North American interior, and recent population recovery affect genetic differentiation between and within the East, Midwest and West. First, woody cover is a crucial habitat feature for bobcats (Anderson 1987; Rolley 1987; Nielsen and Woolf 2002; Woolf et al. 2002; Anderson and Lovallo 2003), and the patchy distribution of woody cover coupled with widespread agricultural development in the Midwest may sufficiently impede dispersal and gene flow between regions and specific sample areas. Multi-scale changes in habitat composition, landscape characteristics, and metropolitan areas have been recognized as impediments to dispersal in other mobile carnivores (e.g. cougars: Ernest et al. 2003; McRae et al. 2005 and wolverines, Gulo gulo: Kyle and Strobeck 2001). Secondly, noticeable decreases of bobcats range-wide, and most significantly in the Midwest, resulting from increased agricultural land use, recreational hunting and trapping, and predator control in the twentieth century (Brown and Yeager 1943; Deems and Pursley 1978; Erickson et al. 1981; Hamilton and Fox 1987; Woolf and Hubert 1998) and subsequent increases in density, might have impacted allele frequencies. Following bobcat decreases, they were protected under federal legislation (i.e. the Convention on International Trade in Endangered Species (CITES) and the U.S. Endangered Species Act (ESA 1973)) and bobcat numbers rebounded (in the Midwest especially; Woolf and Hubert 1998), and continue to do so (Roberts and Crimmins 2010). Expansions of small, isolated bobcat populations may have shifted allele frequencies and contributed to genetic differentiation between and within the Midwest and East. Nevertheless, the Midwest and East are not entirely distinct from one another. Assignment of individuals to other clusters, and weak assignment probabilities imply that there is some gene flow between the Midwest and East. Bobcat life history indicates that they readily disperse and undertake long-distance dispersal (Knick and Bailey 1986; Kamler et al. 2000; Nielsen and Woolf 2003; Johnson et al. 2010), suggesting that gene flow between the Midwest and East is plausible.

Patterns of genetic diversity

Microsatellite diversity was moderate to high among bobcat sample areas, and although the diversity measures here are not directly comparable to other studies due to differences in microsatellites used, number of loci and species, our estimates are similar or greater than other bobcat studies and mobile North American medium-sized cats (e.g. Schwartz et al. 2002; Riley et al. 2006; Millions and Swanson 2007; Campbell and Strobeck 2006). On the other hand, average mtDNA diversity was low in comparison to other medium-sized cats (423 base pair fragment control region analyzed in ocelots (Leopardus pardalis) and margays (L. wiedii; Eizirik et al. 1998)).

Significant HWE scores indicating homozygote excesses in the microsatellite data were evident in a few cases within clusters. However, given that there was no consistency to the pattern of loci out of HWE, we do not think the assumption of random mating was violated. Loci out of HWE could also be a result of sex-biased dispersal behavior in bobcats (Croteau et al. 2010; Janecka et al. 2007), further fine scale sub-structuring within clusters, a clumped sampling scheme, not sampling all individuals that would be contained within the cluster and specifying a cluster to a sample area even though the assignment values were weak.

The relatively high levels of genetic variation historically present and maintained throughout the Western portion of bobcat range suggest that bobcats have persisted at large effective population sizes in this region and gene flow is ongoing. Extensive gene flow within this area will counteract the negative effects of genetic drift and potentially introduce new alleles (Hartl and Clark 2007). Low values of nucleotide and haplotype diversity in the Midwest and East suggest that populations have historically experienced small effective population sizes (Avise 2000). Additionally, small values of π can be attributed to stochastic events (e.g. drastic changes in the environment and/or founder and bottleneck effects) that occur during expansion, which can lead to a reduction in genetic variation within the newly occupied range (Zink et al. 2000). The current low levels of microsatellite variation in the East are most likely a result of population isolation and genetic drift. Bobcats have thriving populations in the East and Midwest, as evidenced by hunting and trapping seasons and growing populations (Woolf and Hubert 1998; Roberts and Crimmins 2010); however, the ability of bobcats to easily disperse between the East and other portions of bobcats’ range may be hindered due to barriers such as urban development and agricultural land use in the Midwest. Since Eastern populations are at the range boundary, it is likely that anthropogenic land use, and possibly geographic distance, play a role in the degree of connectivity of Eastern populations to other populations at the range core.

Conservation implications

The potential of a species to evolve, adapt and respond to environmental change is primarily dependent upon genetic diversity (Frankham et al. 2002). The World Conservation Union (IUCN) recognized that the preservation of genetic diversity is one of three priorities essential to conserving biodiversity as a whole (McNeely et al. 1990). Furthermore, genetic diversity has shown to be essential to ecosystem recovery and function (Reusch et al. 2005; Frankham 2005). Given this, identifying both historical and contemporary patterns of genetic diversity is a vital component of managing species and populations for conservation.

Understanding what drives patterns of genetic diversity and population structure is especially important in populations of wide-ranging species because lack of gene flow between populations within the range can lead to isolation and small population size which will eventually result in inbreeding, loss of genetic variability and higher extinction risk (Frankam et al. 2002). From our results, we can infer that bobcats are maintaining gene flow, and thus genetic diversity. This result is no surprise given their dispersal capabilities (Knick and Bailey 1986; Kamler et al. 2000; Nielsen and Woolf 2003; Johnson et al. 2010). Gene flow is most likely influencing the growth and stability of populations as documented by Roberts and Crimmins (2010).

Despite high abundance and improved status of the bobcat throughout much of its range (Woolf and Hubert 1998; Roberts and Crimmins 2010), impending changes in the composition and structure of bobcat habitat are not likely to be conducive to gene flow. These habitat changes are already at work as evidenced by the microsatellite analyses. Furthermore, closer inspection of FST results show that bobcats exhibit moderate genetic differentiation (Hartl and Clark 2007) between many samples areas (47 % of pairwise FST estimates ≥0.05; Online Resource 2), demonstrating that several populations may be experiencing some level of isolation. However, the clumped sampling scheme may artificially inflate FST results (Schwartz and McKelvey 2009). Regardless, these results coupled with STRUCTURE analyses from within clusters and some loci out of HWE support a finer level of substructuring. It has been suggested that the most significant barriers to gene flow in wildlife populations are anthropogenic in nature, in the form of roads, agriculture, and human settlement (Kyle et al. 2000; Kyle and Strobeck 2001). As human populations increase and expand into exurban areas, and if former bobcat habitat is converted to row-crop agriculture for biofuel production (Bies 2006; U.S. Department of Agriculture 2007), fragmentation and loss of habitat will increase, resulting in reduced gene flow and increased subdivision of bobcat populations. Therefore, it will become necessary to preserve contiguous tracts of habitat to allow for bobcat dispersal to maintain historical connections that facilitate gene flow and maintain present levels of genetic diversity.

Of more immediate conservation concern is how to properly manage bobcat populations given our findings. Although smaller-scale studies can better the understanding of local patterns of population structure, large-scale studies are important for assessing the range-wide distribution of genetic diversity of a species, and can provide vital information for conservation planning. Bobcats historically have been managed by state and provincial conservation agencies on a state-by-state basis. These boundaries mean little to bobcats given the size of their home ranges and dispersal capabilities. Thus, referring to regional differences among genetic variation may be more appropriate for determining the most suitable strategy for bobcat conservation. We suggest that wildlife biologists and agency administrators consider implementing regional (i.e., multi-state) management plans for bobcats that maintain historical connections and mitigate limitations to gene flow within regions to prevent loss of genetic diversity. Furthermore, future management of bobcats would benefit from a more rigorous landscape genetics approach that pinpoints specific anthropogenic or natural barriers to gene flow and the effects that landscape composition, configuration and matrix quality have on spatial genetic variation. This approach would allow biologists to identify areas that are isolated, habitat/landscape variables that impede gene flow and the potential for corridor development to ensure effective dispersal.