Journal of Mammalogy, 93(4):989–1000, 2012
Genetic diversity, population structure, and movements of mountain lions (Puma concolor) in Texas
JOSEPH D. HOLBROOK,* RANDY W. DEYOUNG,JAN E. JANECKA,MICHAEL E. TEWES,RODNEY L. HONEYCUTT, AND JOHN H. YOUNG Caesar Kleberg Wildlife Research Institute, Texas A&M University–Kingsville, Kingsville, TX 78363, USA (JDH, RWD, MET)
Department of Veterinary Integrative Biosciences, College of Veterinary Medicine and Biomedical Sciences, Texas A&M Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021 University, College Station, TX 77843, USA (JEJ) Natural Science Division, Pepperdine University, Malibu, CA 90263, USA (RLH) Texas Parks and Wildlife Department, 4200 Smith School Road, Austin, TX 78612, USA (JHY) Present address of JDH: Department of Fish and Wildlife Sciences, P.O. Box 441136, University of Idaho, Moscow, ID 83844-1136, USA * Correspondent: [email protected]
Knowledge of population boundaries and long-distance movements is important for wildlife conservation. We used genetic tools to investigate genetic diversity, population structure, and movements of mountain lions (Puma concolor) in Texas. We amplified 11 microsatellite loci for 245 individuals collected during 1985–2010 from Texas and New Mexico. Bayesian clustering and values of FST suggested a partitioning of mountain lions into 3 genetically differentiated groups, New Mexico, western Texas, and southern Texas. New Mexico and western Texas exhibited moderate levels of genetic diversity (expected heterozygosity [HE] ¼ 0.61 and 0.58, respectively), whereas diversity in southern Texas was lower (HE ¼ 0.47). Southern Texas displayed elevated genetic structure when compared to western Texas and New Mexico (FST ¼ 0.102–0.148), whereas the comparison between New Mexico and western Texas revealed less subdivision (FST ¼ 0.056). We documented long-distance movement among regions, and New Mexico and western Texas were sources for putative dispersers we sampled outside known populations. Differences in genetic structure and diversity between southern and western Texas support the designation of separate management units. Southern Texas appears isolated and further investigation is needed to determine the current population status. Mountain lion populations in New Mexico and western Texas may be important for future recolonization into portions of the southern United States.
Key words: Bayesian clustering, genetic diversity, genetic structure, long-distance movement, mountain lion, Puma concolor, Texas
Ó 2012 American Society of Mammalogists DOI: 10.1644/11-MAMM-A-326.2
The distribution of mountain lions (Puma concolor) in North was an important industry in Texas during the late 1800s to America has reduced over the last 200 years (Anderson et al. mid-1900s and predator removal was widely practiced, thus 2010; Logan and Sweanor 2001). In Texas, mountain lions are contributing to the decline of mountain lion populations on the periphery of the distribution in both the United States (Lehmann 1969; Wade et al. 1984). Additionally, mountain and North America. Populations were historically distributed lion habitat in central and southern Texas became increasingly throughout the state, but census size and geographic distribu- fragmented during the past century due to agricultural tion have declined over time. Today, breeding populations are practices, urbanization, and energy development. known to persist primarily in the Trans-Pecos and South Texas Mountain lions are not formally managed in Texas, Plain ecoregions of western and southern Texas, respectively receiving the designated status of a nongame or varmint (Schmidly 2004). Similar to many states in the United States, the decline of mountain lions in Texas is attributed to predator control and loss of habitat (Logan and Sweanor 2001). Livestock ranching www.mammalogy.org 989 990 JOURNAL OF MAMMALOGY Vol. 93, No. 4 species since 1970 (Harveson et al. 1996; Russ 1996). main focus was on Texas. New Mexico exhibits a large Furthermore, harvest reporting is not required by law. gradient in elevation and topography with environments Examination of the limited data available suggests that ranging from brush and cacti (Cactus spp.)–dominated populations in both western and southern Texas are restricted deserts, pinyon-pine (Pinus spp.) and juniper (Juniperus spp.) by low survival (Harveson 1997; Young et al. 2010) and low woodlands, to upland forests dominated by mixed conifers reproductive rates (Harveson 1997; Pittman et al. 2000). A (Bailey 1980). Western Texas is primarily a desert environment preliminary genetic analysis found that mountain lions in dominated by shrubs, cacti, and grasses with a few isolated southern Texas have low levels of genetic diversity and appear mountain ranges where trees such as oak (Quercus spp.), to be isolated from those in western Texas (Walker et al. 2000). juniper, and pine (Pinus spp.) are abundant (Bailey 1980). Mountain lions occur at low densities, exhibit large home Southern Texas is characterized as an arid environment
ranges, display elusive behavior, inhabit rough terrain, and are exhibiting low elevations, mild topography, and dense woody Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021 cryptically colored (Logan and Sweanor 2001). These factors vegetation (e.g., huisache [Acacia farnesiana], granjeno [Celtis make traditional census methods (e.g., mark–recapture or ehrenbergiana], and honey mesquite [Prosopis glandulosa]) transects) prohibitively expensive and logistically difficult, interspersed with grasses. particularly for surveys at large spatial extents. Additionally, Sample collection and DNA analysis.—We obtained because harvest reporting is not required in Texas, managers mountain lion tissue samples from Texas and New Mexico are unable to use harvest as a demographic index to monitor during 1985–2010. Samples from Texas were donated by population trends (e.g., Anderson and Lindzey 2005). Alter- hunters and trappers, sampled from roadkills, or collected when natives to conventional methods such as genetic tools are marking individuals during previous research (Harveson 1997). useful to circumvent the challenges with monitoring wildlife Tissue samples from New Mexico were provided by the populations (DeYoung and Honeycutt 2005). Accordingly, Museum of Southwestern Biology, Division of Genomic genetic data have been increasingly used to study populations Resources (MSB 58960–58963, 92685, 142863, 142867– of solitary, highly mobile carnivores (e.g., Haag et al. 2010; 142871, 142873, 142878, 142882, 142884–142887, 142890, Spong et al. 2000). For mountain lions, genetic data have been 142891, 142893, 142896, 142901, 142902, 142909–142911, used to delineate population boundaries and management units 142913, 142923, 142928, 145874, and 157080). Tissue was (e.g., Anderson et al. 2004; Ernest et al. 2003; Loxterman frozen, dried, or placed in lysis buffer (Longmire et al. 1997) 2011; McRae et al. 2005). Genetic data also have been used to until DNA extraction. We used a commercial kit (Qiagen identify long-distance movements among populations (Frantz DNeasy tissue kit; Qiagen, Valencia, California) to extract et al. 2006; Wasser et al. 2008), which can provide an objective DNA from all tissue samples. We initiated the polymerase means of establishing habitat priorities and conservation chain reaction to amplify 11 microsatellite loci (FCA008, protocols (Beier 2010; LaRue and Nielsen 2008). Knowledge FCA035, FCA043, FCA077, FCA082, FCA090, FCA096, of source populations and dispersal corridors is important to FCA132, FCA133, FCA176, and FCA205) described by develop management actions that mediate mountain lion- Menotti-Raymond et al. (1999). We amplified all loci human conflicts (Thompson and Jenks 2010), which are individually in 10-ll reaction volumes containing 5 llof increasingly more common in suburban landscapes (Beier AmpliTaq Gold PCR Master Mix (Applied Biosystems, Foster 2010). City, California), 0.24 lM of each primer, and 10–50 ng of Historical persecution, habitat loss, and unregulated harvest DNA. We used a touchdown polymerase chain reaction profile have provoked questions regarding the viability of mountain with thermal conditions consisting of an initial denaturation at lion populations in Texas (Russ 1996). The overall goal of this 948C for 10 min, 20 cycles of 948C for 30 s, 628C for 30 s, study was to use genetic data to provide insights into 618C for 30 s, 608C for 30 s, and 728C for 60 s, followed by 30 population structure and movements of mountain lions in cycles of 948C for 30 s, 558C for 90 s, and 728C for 60 s, with a Texas and adjacent populations in New Mexico. Our specific final extension of 608C for 10 min. We used electrophoresis on objectives were to estimate genetic diversity, characterize an ABI 3130xl DNA analyzer (Applied Biosystems) for population genetic structure, identify dispersal among known fragment separation, and determined genotypes with populations, and assign population of origin to putative long- GeneMapper version 4.0 (Applied Biosystems). All sample distance dispersers sampled outside known populations. sets injected on the DNA analyzer had a positive and negative Information from this study will expand on previous genetic control. We randomly selected and reanalyzed 10% of analyses of Texas mountain lions, provide much-needed individuals to calculate the genotyping error rate. information on the current status of mountain lions in Texas, Genetic diversity and Hardy–Weinberg equilibrium.—We and identify source populations that are providing dispersers computed estimates of genetic diversity for the overall sample, into historical range. as well as for each of the 3 geographic regions (Fig. 1). Geographic regions were designated based on ecoregion and adequate sample size. We used the computer program Arlequin MATERIALS AND METHODS version 3.5 (Excoffier and Lischer 2010) to estimate mean Study area.—We conducted this study throughout New observed heterozygosity (HO), expected heterozygosity (HE— Mexico, western Texas, and southern Texas (Fig. 1), but our Nei 1987), and number of alleles per locus (A). We estimated August 2012 HOLBROOK ET AL.—GENETICS OF MOUNTAIN LIONS IN TEXAS 991 Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021
FIG.1.—Distribution of mountain lions (Puma concolor)sampled(n ¼ 245) throughout Texas and New Mexico during 1985–2010. Triangles represent individuals sampled from known populations (n ¼ 237), and stars indicate potential dispersers sampled outside known populations (n ¼ 8). Information for each disperser is associated with the numbers next to stars: 1 ¼ PC001, Bossier City, Louisiana; 2 ¼ PC003, Kerr County, Texas; 3 ¼ PC004, Fisher County, Texas; 4 ¼ PC042, Deaf Smith County, Texas; 5 ¼ PC123, Edwards County, Texas; 6 ¼ PC163, Real County, Texas; 7 ¼ PC164, Kimble County, Texas; and 8 ¼ PC165, Sutton County, Texas. We grouped individuals from known populations based on ecoregion and sample size for analyses (dotted lines separate groups): New Mexico (n ¼ 31), western Texas (n ¼ 178), and southern Texas (n ¼ 28). mean allelic richness (Ar) using HP-RARE version 1.0 1984) among all county pairs; FST is the proportion of genetic (Hurlbert 1971; Kalinowski 2005). We tested Hardy– diversity explained by allele frequency differences among Weinberg expectations using FIS (Weir and Cockerham groupings (Holsinger and Weir 2009). We used regression to 1984), and assessed statistical significance (2-sided) by test for a relationship between the estimates of FST/(1 FST) comparing the observed value against a null value derived and Euclidean distance (Rousset 1997). We computed the from 1,023 permutations of alleles among individuals. We standard error of the slope by jackknifing over loci and computed and tested FIS using Arlequin version 3.5 (Excoffier assessed statistical significance by permuting locations among and Lischer 2010). groups 1,000 times, which is equivalent to a Mantel test (Hardy Genetic associations with distance.—We implemented 2 and Vekemans 2002). approaches to evaluate genetic associations across a gradient of Second, we used spatial autocorrelation to explore the spatial Euclidean distance. First, we spatially grouped individuals by extent of population structure. At the individual level, county, or combined geographically proximate individuals autocorrelation analyses describe the correlation between from .1 county to maintain n 5 (New Mexico: Bernalillo– average gene frequencies of a pair of individuals (Hardy and San Miguel, Dona Ana, Grant–Catron, Hidalgo, and Socorro– Vekemans 1999; Scribner et al. 2005). We used Moran’s I Sierra–Lincoln; Texas: LaSalle–McMullen–Kleberg–Live (Hardy and Vekemans 1999) as the measure of autocorrelation Oak, Maverick–Kinney–Webb, Brewster–Pecos, Culberson– because of its robust performance (Epperson 2004). We Hudspeth, Jeff Davis–Reeves, Presidio, and Terrell–Val computed mean Moran’s I-values for all pairs of individuals Verde). We computed pairwise FST (Weir and Cockerham within 15 Euclidean distance classes. We used 15 classes with 992 JOURNAL OF MAMMALOGY Vol. 93, No. 4 a large and equal number of pairs (i.e., 1,864) to ensure precise genetic clusters, with 8 independent runs for each K to evaluate estimates of Moran’s I, and low coefficient of variation within consistency. We calculated the arithmetic mean and standard each class (Hardy and Vekemans 2009). We tested the deviation of the log probability of the data (Ln P(D)) across statistical significance (2-sided) of Moran’s I for each distance runs for each K to identify the plateau and determine the class by comparing observed values to a randomized value optimal number of clusters (Pritchard et al. 2007). We also computed from 1,000 permutations of individual locations. We calculated the DK statistic (Evanno et al. 2005) and used calculated the standard error of Moran’s I by jackknifing over ancestry proportions (i.e., q-values) as an index (Pritchard et al. loci. We used the program SPAGeDi version 1.3 (Hardy and 2007) to inform the selection of K. Vekemans 2002) to perform regression and spatial autocorre- Assigningorigintodispersers.—We used Bayesian lation analyses. clustering and assignment tests to determine origin for
Genetic structure.—To further examine population genetic dispersers among known populations, and potential dispersers Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021 structure, we implemented traditional genetic differentiation sampled outside known populations. We used mean q-values methods as well as Bayesian clustering. We used the county (over 8 runs) from the structure analyses and geographic groupings and the 3 regions to compute pairwise and overall locations to identify long-distance movement among known FST (Weir and Cockerham 1984) using the computer program populations. We defined long distance as movement from Arlequin version 3.5 (Excoffier and Lischer 2010). We tested western Texas to southern Texas or vice versa, and southern statistical significance (2-sided) by comparing the observed Texas to New Mexico or vice versa (i.e., 200 km). Similar to value to a null value derived from 1,023 permutations of previous studies (e.g., Latch et al. 2006, 2008), we considered genotypes among groups (i.e., counties or regions). individuals residents of a cluster if q . 0.75 and admixed if q ¼ We applied 2 Bayesian clustering algorithms that incorpo- 0.25–0.75. rate spatial locations. We employed the algorithm implemented Next, we used 3 Bayesian assignment methods to assign in Geneland (Guillot et al. 2005a, 2005b) version 3.2.4 using origin to the potential dispersers sampled outside known program R version 2.11.1 (R Development Core Team 2011). populations. When reference populations are known a priori, This model uses a Markov chain Monte Carlo approach to infer assignment methods provide a more explicit way to discern genetic discontinuities among georeferenced genotypes. We genetic origins. In all analyses we considered the 3 regions evaluated 1–8 possible genetic clusters (K), with 8 independent (western Texas, southern Texas, and New Mexico) as reference runs for each K. We implemented the spatial model with 10 km populations, and the potential dispersers as unknowns. First, of uncertainty, because spatial coordinates of some sample we used the modified assignment approach of Rannala and locations were approximate (i.e., not taken with a global Mountain (1997) in GeneClass version 2 (Piry et al. 2004). positioning system). We assumed allele frequencies to be This approach provides likelihood ratio scores for each correlated and used 100,000 Markov chain Monte Carlo unknown individual to each reference population (Piry et al. iterations while recording 1,000 (thinning ¼ 100). We selected 2004), and we used scores . 85% to indicate assignment. K using the mode of the maximized posterior probability. Second, we employed the assignment methods implemented in Second, we applied the Bayesian algorithm described by structure version 2.2 (Falush et al. 2003) and BAPS version 5 Corander et al. (2003) using BAPS version 5. This approach (Corander et al. 2003). For both analyses we assumed that K ¼ uses stochastic optimization to infer the posterior mode of 3, corresponding to the regional reference populations. We genetic structure in the data. We implemented the spatial employed the USEPOPINFO option in structure (Falush et al. clustering of individuals (Corander et al. 2008b) and explored 2003), and executed 100,000 Markov chain Monte Carlo burn- K ¼ 1–8, with 8 independent runs for each K. We selected K in and 500,000 data collecting repetitions. We assumed allele based on the partitioning of individuals that maximized the log frequencies were correlated, no admixture, and updated marginal likelihood. The optimal partition of individuals in frequencies with only reference populations. Because results Geneland and BAPS should minimize Hardy–Weinberg and of this analysis can be sensitive to the a priori assigned linkage disequilibrium within clusters. migration rate (MIGPRIOR), we analyzed the data using a We also employed a nonspatial Bayesian clustering range of values (i.e., 0.001–0.10) as suggested by Pritchard et algorithm implemented in the computer program structure al. (2007). The choice of MIGPRIOR did not substantially version 2.2 (Pritchard et al. 2000). This model uses a Markov influence results, thus we only present results using MIGP- chain Monte Carlo to infer genetic clusters and assign RIOR ¼ 0.05 (default value). Because we incorporated prior individuals to ancestral populations while minimizing Hardy– population information (USEPOPINFO) and assumed no Weinberg and linkage disequilibrium within clusters (Pritchard admixture among reference populations, more certainty is et al. 2000). The algorithm also estimates ancestry proportions associated with assignments compared to admixture analyses. (q-values) to each genetic cluster for each individual. We Thus, we used a more stringent q-value (q . 0.85) to indicate selected the admixture model and assumed allele frequencies genetic assignment (Frantz et al. 2006). Finally, we employed were correlated (Falush et al. 2003). We performed 100,000 the trained clustering methodology (Corander et al. 2006, Markov chain Monte Carlo burn-in repetitions to reduce initial 2008a) in BAPS (Corander et al. 2003). We explored the configuration effects, followed by 500,000 Markov chain assignment of each unknown individual to regional reference Monte Carlo repetitions of data collection. We explored 1–8 populations, one-by-one. To evaluate the strength of assign- August 2012 HOLBROOK ET AL.—GENETICS OF MOUNTAIN LIONS IN TEXAS 993 ment to each cluster we multiplied by 2 the absolute value of change in the log marginal likelihood (i.e., Bayes factor— Corander et al. 2009) of individual i being assigned to the alternative cluster j. Values of 0 indicate the assigned reference population, and movement to another population with a change 6 suggests substantial support for assignment (Kass and Raftery 1995). A change 2 indicates poor assignment support.
RESULTS Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021 Genetic data.—We successfully genotyped 245 mountain lions (57% males, 39% females, and 4% had no sex information) at 11 microsatellite loci (Fig. 1). A total of 237 genotypes were obtained from known populations in Texas and New Mexico, and 8 were from presumed long-distance dispersers; 1 each from Kerr, Fisher, Deaf Smith, Edwards, FIG.2.—Regression of linearized FST and geographic distance for Kimble, Sutton, and Real counties, Texas, and Bossier City, mountain lions (Puma concolor) sampled from known populations (n Louisiana. Positive and negative polymerase chain reaction ¼ 237) during 1985–2010. Samples were grouped by county in Texas controls were consistent and did not exhibit any contamination. and New Mexico while maintaining n 5. We highlighted comparisons with different symbols to display important trends within Our genotyping error rate was ,1%. each geographic region. Dark diamonds represent comparisons within Estimates of mean HO,HE, A, and Ar were moderate for western Texas and New Mexico. Open circles indicate comparisons of New Mexico and western Texas as well as the total sample the western county grouping in southern Texas (i.e., Maverick– (Table 1). However, point estimates of genetic diversity for Kinney–Webb) to other groups. Open triangles signify comparisons of southern Texas were 10–25% lower than for western Texas and the eastern county group in southern Texas (i.e., LaSalle–McMullen– New Mexico. Hardy–Weinberg equilibrium tests indicated that Kleberg–Live Oak) to other groups; the comparison within southern the 3 regions did not statistically deviate from expectations; Texas (i.e., western and eastern county comparison) also is included however, the total sample exhibited substantial deviation (see arrow). The slope was significant (P , 0.001) based on 1,000 (Table 1). permutations of locations among groups, and the SE was derived by Genetic associations with distance.—We observed a jackknifing over loci. positive relationship (slope ¼ 0.0000002, SE ¼ 0.00000007, P , 0.001) between genetic and Euclidean distance (Fig. 2), value for 2nd cousins. We observed negative autocorrelation which indicated a significant pattern of isolation by distance. between distance classes 10–15 (~370–820 km), substantiating Euclidian distance accounted for about half of the variation in the presence of an isolation-by-distance pattern. Together, genetic distance (R2 ¼ 0.49). Within the southern Texas region, regression and spatial autocorrelation provided evidence for an most pairwise comparisons involving the western county group isolation-by-distance cline, regional-level genetic structure, and (Maverick–Kinney–Webb) followed an isolation-by-distance genetic association among individuals at distances , 50 km. pattern. However, comparisons including the eastern county Genetic structure.—We observed significant genetic group (LaSalle–McMullen–Kleberg–Live Oak) produced differentiation among county groupings (FST ¼ 0.031, P , greater FST values than the predicted relationship (Fig. 2). 0.001) and the 3 regions (FST ¼ 0.074, P , 0.001), indicating Analyses of spatial autocorrelation (Fig. 3) indicated moderate levels of genetic structure. The regional division statistically significant positive autocorrelation for the first 9 accounted for more genetic variation than the county distance classes (~20–250 km), with the exception of class 5 groupings, suggesting that the differences among regions (~105 km). Moran’s I-values in the 1st (~20 km) and 2nd were greater than within regions. Fifty-six of 66 pairwise (~40 km) distance class were greater or equal to the expected comparisons among samples grouped by county were
TABLE 1.—Mean estimates (over 11 loci) of observed (HO) and expected (HE) heterozygosity, number of alleles per locus (A), and allelic richness (Ar) for known populations of mountain lions (Puma concolor) in Texas and New Mexico sampled during 1985–2010. Standard deviations (SDs) are in parentheses, and n indicates sample size. Estimates of FIS indicate tests for Hardy–Weinberg equilibrium, and P-values are based on 1,023 permutations of alleles among individuals. We used the computer program Arlequin 3.5 (Excoffier and Lischer 2010) to compute HO,HE, A, and FIS, and HP-RARE 1.0 (Kalinowski 2005) to compute Ar.
Region n HO HE AAr FIS New Mexico 31 0.57 (0.21) 0.61 (0.22) 4.55 (1.64) 4.43 (1.59) 0.07, P ¼ 0.08 Western Texas 178 0.56 (0.22) 0.58 (0.23) 5.09 (1.81) 4.23 (1.39) 0.02, P ¼ 0.76 Southern Texas 28 0.45 (0.25) 0.47 (0.25) 3.91 (1.64) 3.85 (1.60) 0.02, P ¼ 0.38 Total 237 0.55 (0.21) 0.59 (0.23) 5.55 (1.92) 5.53 (1.90) 0.11, P , 0.01 994 JOURNAL OF MAMMALOGY Vol. 93, No. 4
admixture (q ¼ 0.25–0.75) than the other regions: K ¼ 2— New Mexico (29%), southern Texas (18%), western Texas (34%); K ¼ 3—New Mexico (32%), southern Texas (21%), western Texas (66%); and K ¼ 4—New Mexico (23%), southern Texas (18%), western Texas (51%). We mapped individuals assuming K ¼ 4 to explore if there was an obvious biological or terrain feature that may explain the additional cluster in western Texas. We included individuals in 1 of the 2 clusters within western Texas if q was .0.65. There was no clear biological interpretation of the additional cluster in
western Texas. Incoherent clustering has been documented in Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021 clumped and opportunistic sampling designs (McRae et al. 2005; Schwartz and McKelvey 2009), as well as in data exhibiting isolation by distance (Frantz et al. 2009), both of which are characteristic of our data. However, we conducted exploratory analyses by separating males and females to determine if dispersal differences were responsible for the FIG.3.—Mean autocorrelation coefficients (Moran’s I)and Euclidean distance among pairs of individuals using 15 distance additional cluster in western Texas. For both males (n ¼ 98) classes for known mountain lion (Puma concolor) populations (n ¼ and females (n ¼ 73) results indicated K ¼ 1, but when 237) in Texas and New Mexico sampled during 1985–2010. Dark combined, Ln P(D) and DK suggested K ¼ 2. Accordingly, we circles represent observed Moran’s I-values, and open circles represent explored genetic differentiation between sexes in western null values based on 1,023 permutations of individual locations. Error Texas, which proved to be low (FST ¼ 0.005, P . 0.05). We bars indicate 6 1 SE, and were computed by jackknifing over loci. were unable to identify biological support for the additional cluster in western Texas. Therefore, we concluded that our statistically .0.0 (Table 2). Estimates of FST between Texas sample was composed of only 3 genetic clusters, a solution and New Mexico were moderate to high (0.056–0.279), and supported by FST analyses and 2 of 3 clustering algorithms. generally increased with Euclidean distance. All estimates The partition of individuals from Geneland, BAPS, and including the LaSalle–McMullen–Kleberg–Live Oak group structure with K ¼ 3 suggested the clusters generally were notably high (FST ¼ 0.124–0.279). Pairwise FST-values corresponded to the 3 regions of New Mexico, western Texas, among regions were positive, and revealed that the greatest and southern Texas. structure was associated with southern Texas (New Mexico– Assigning origin to dispersers.—We identified long-distance southern Texas FST ¼ 0.148, P , 0.001; western Texas– movements among known populations using mean q-values southern Texas FST ¼ 0.102, P , 0.001; and New Mexico– from structure, assuming K ¼ 3. Two adult males sampled in western Texas FST ¼ 0.056, P , 0.001). Jeff Davis and Brewster counties, western Texas, exhibited Of the 8 runs using Geneland, the maximized posterior ancestry to southern Texas (PC040—q ¼ 0.793, SD ¼ 0.010; probability of K occurred at 3 for 6 runs, and at 5 for 2 runs. and MLA19—q ¼ 0.933, SD ¼ 0.003). In addition, 2 males and However, the maximized probability for all 6 runs at K ¼ 3 was 1 adult female sampled in Maverick, LaSalle, and Kinney higher than at K ¼ 5. Therefore, we inferred the optimal counties, southern Texas, exhibited ancestry to New Mexico number of clusters to be 3. The clusters of individuals and (PC007—q ¼ 0.780, SD ¼ 0.005; PC189—q ¼ 0.794, SD ¼ membership probability suggested by Geneland corresponded 0.005; and PC121—q ¼ 0.753, SD ¼ 0.008). to the 3 regions, New Mexico, western Texas, and southern Before implementing assignment tests, it is important to Texas. Similarly, the BAPS results indicated that the log ensure that a sufficient number of loci and individuals have marginal likelihoods for the 10 best-visited partitions were been sampled from reference populations (Manel et al. 2002). maximized at K ¼ 3, providing a posterior probability of 1 for Reasonable levels of genetic diversity and differentiation also K ¼ 3. The clustering of individuals from BAPS (Fig. 4) are required. Our reference populations were composed of 28– approximately corresponded to New Mexico, western Texas, 178 individuals genotyped at 11 loci with reasonable levels of and southern Texas, corroborating the results from Geneland. HE and FST providing adequate power to assign origins (Latch The structure results were less clear than those from et al. 2006; Manel et al. 2002). Results from GeneClass, Geneland and BAPS. The mean Ln P(D) appeared to reach a structure, and BAPS were consistent and implied strong plateau at K ¼ 2orK ¼ 3, peaked at K ¼ 4, and declined and genetic assignments for 6 of the 8 potential dispersers (Table became more variable at K . 4 (Fig. 5). The DK statistic of 3). PC001 and PC042 were strongly assigned to New Mexico. Evanno et al. (2005) provided moderate support for K ¼ 2 and This is particularly interesting because PC001 was a male K ¼ 3, but high support for K ¼ 4 (Fig. 5). Ancestry proportions sampled in Bossier City, Louisiana, .800 km from New (q-values) for most individuals at K ¼ 2, K ¼ 3, and K ¼ 4 Mexico (Fig. 1). The assignment for PC0042 is not surprising maintained high values, indicating support for all 3 scenarios. because this male was sampled ,10 km from New Mexico. However, at K ¼ 2–4, western Texas displayed greater PC004, PC123, PC163, and PC165 all exhibited strong August 2012 HOLBROOK ET AL.—GENETICS OF MOUNTAIN LIONS IN TEXAS 995
assignments to western Texas. These assignments are reason- able because PC004 was a male sampled in north-central Terrell–
Val Verde Texas, and PC123 (male), PC163 (female), and PC165 (male) were sampled in central Texas (Fig. 1). — 0.000
NS — Unfortunately, we were unable to assign 2 dispersers. PC003 was moderately or weakly assigned to all reference populations by GeneClass and structure, suggesting admixed ancestry.
— 0.011 0.015 However, BAPS provided essentially no support to any
Reeves Presidio reference population (i.e., a change , 3 in 2 times the log Jeff Davis– marginal likelihood), indicating that PC003 could be from an er 2010) to compute estimates.
unsampled source. Further, by all methodologies PC164 was Downloaded from https://academic.oup.com/jmammal/article/93/4/989/969801 by guest on 02 October 2021 — 0.069 0.049 0.067 moderately and weakly assigned to western Texas and New Hudspeth Culberson– Mexico with essentially no support for southern Texas. PC164 appeared to be a product of combined ancestry from New — 0.070 0.016 0.002 0.004 NS NS Mexico and western Texas. Assignments from GeneClass, Pecos
Brewster– structure, and BAPS indicated that long-distance movements have occurred across our sampling area. 0.05) estimates based on 1,023 permutations of genotypes , — 0.078 0.103 0.089 0.083 0.077 P
Webb DISCUSSION Genetic structure in populations of mountain lions is low in Maverick–Kinney– relatively continuous habitats (Anderson et al. 2004; Culver et al. 2000; Sinclair et al. 2001), but can be highly structured in fragmented habitats (Ernest et al. 2003). Associations of ) in Texas and New Mexico sampled during 1985–2010. Individuals were grouped
— 0.159 0.124genetic 0.212 0.158 and 0.144 geographic 0.173 distance in our study indicated that genetic structure was present at both the local and regional scales. For the local scale (,50 km), autocorrelation analyses Kleberg–Live Oak LaSalle–McMullen– suggested that sampled mountain lions exhibited relatively high genetic associations, particularly for a territorial species. Puma concolor Female philopatry (Logan and Sweanor 2001, 2010; Sweanor — 0.279 0.069 0.098 0.102 0.118 0.083 0.088
Lincoln et al. 2000), high sampling effort (e.g., hunting and trapping) at local scales (Schwartz and McKelvey 2009), habitat loss, or a Socorro–Sierra– combination could have contributed to the nonindependence among proximate individuals. At the regional scale, autocor- relation and regression analyses identified a significant for mountain lions ( ) in the subdiagonals indicate statistically significant (
ST isolation-by-distance pattern, suggesting a decrease in genetic F New Mexico Texas Catron Hidalgo Grant– similarity with increasing geographic distance. This pattern is consistent with other continuous (Anderson et al. 2004; NS Ana
Dona Sinclair et al. 2001) as well as structured (Ernest et al. 2003; Loxterman 2011; McRae et al. 2005) mountain lion popula-
5. Black circles ( tions.