Individual Assignment Tests

Because natural population boundaries were not clearly defined for G. berlandieri, clustering analyses were carried out before estimating standard population parameters that required a priori knowledge of population boundaries. Assignment of individuals to groups was performed by using 2 Bayesian clustering software programs with the Markov Chain Monte Carlo estimation. First, the number of genetic clusters (K) was estimated by using STRUCTURE 2.2 (Pritchard et al. 2000). Because shared ancestry was expected in G. berlandieri, the admixture model with correlated allele frequency was used. A range of clusters (K) from 1 to 7 was tested 10 times with 5 × 10⁵ iterations and 5 × 10⁴ burn-in at each K. The number of clusters (K) with the highest mean log-likelihood was selected as the best description of population structure. Second, a spatial clustering program, GENELAND 3.1.2 (Guillot et al. 2008), was used to incorporate geographic locations of samples in inferring the number of groups. The number of clusters (K) was first inferred from a range of K between 1 and 5, by using the Dirichlet model for allele frequency with 5 × 10⁵ iterations and 100 thinning. This process was replicated 10 times to assess the consistency of the modal K. The same operation was carried out at the fixed K estimated by the first analysis with the maximum rate of Poisson process at 100, the maximum number of nuclei at 300 with 10⁴ iterations, and 10 thinning. Consistency of population membership, was verified by iterating the analysis 10 times.

Microsatellite Genetic Variation

Traditional methods to estimate population parameters were performed for the inferred groups determined by the clustering methods described above. Observed (HO) and expected heterozygosity (HE) per locus were calculated for each group by using FSTAT 2.9.3 (Goudet 1995). Significant departure from Hardy-Weinberg equilibrium (HWE) for each locus and presence of linkage disequilibrium between loci in each group were tested by using GENEPOP 4.0 (Rousset 2008). Exact tests for HWE (Guo and Thompson 1992) and G-based likelihood tests for linkage disequilibrium by using the Markov Chain algorithm were carried out with 10⁵ iterations with 10⁴ burn-in to estimate statistical significance. Allelic richness (Ar), the average number of alleles per locus corrected for sample size, was calculated by using FSTAT 2.9.3. The loci that significantly deviated from HWE after sequential Bonferroni correction (Rice 1989) were tested for null alleles by using MICRO-CHECKER (van Oosterhout et al. 2004), and genotypes were adjusted based on the estimated allele frequencies by the Brookfield algorithm (Brookfield 1996).

Population Structure

A genetic distance–based analysis was used to assess the degree of population structure. A pairwise rho (ρ) (Goodman 1997) value corrected for sample size, and allele variance of R_ST (Slatkin 1995) was calculated by using RSTCALC 2.2 (Goodman 1997). Because of the limited sample size and the number of loci as well as possible deviations from the stepwise mutation model (Gaggiotti et al. 1999), F_ST (Weir and Cockerham 1984) under the infinite alleles model (IAM) (Slatkin 1995), which is recommended as more conservative estimate (Gaggiotti et al. 1999), was also calculated by using FSTAT 2.9.3 (Goudet 1995). Both values were estimated with 10,000 permutations. Hierarchical partitioning of a total genetic variation was conducted by using analysis of molecular variance (AMOVA) implemented by ARLEQUIN 3.1 (Excoffier and Schneider 2005). The amount of genetic variation accounted for from the difference between the 2 inferred groups, from among sampling areas within groups, and among individuals within sampling areas were estimated with F_ST and R_ST. Statistical significance was determined by 10,000 permutations.

Comparison Between G. berlandieri and G. agassizii

Genetic variation and the degree of population structure were compared between G. berlandieri and G. agassizii in the Sonoran Desert in Arizona (n  =  154 from 9 sampling areas) (Edwards et al. 2004). The comparable sample size and study range of the sister species of G. berlandieri make this data set from G. agassizii an ideal reference to assess genetic variation in G. berlandieri. Allelic richness (Ar) and HE for the same loci used in this study were compared between the entire sample of G. berlandieri and those of G. agassizii. When assuming the 9 sampling areas as putative populations as described by Edwards et al. (2004), the degree of population structure was estimated with F_ST and ρ. STRUCTURE 2.2 was used to detect larger groups of the populations of G. agassizii with the same parameters described previously.

Gene Flow

Contemporary gene flow was analyzed by assessing isolation by distance within and between the inferred groups. Pairwise genetic distance (F_ST/1 – F_ST ) and geographic distance were compared among sampling areas across the groups and within groups by using the Mantel test with 10,000 permutations for significance implemented by GENEPOP 4.0 (Rousset 2008). Migration rate and its direction between the groups were estimated by using 3 methods, each with different assumptions. First, indirect estimate of gene flow (Nm) by using the private allele method (Slatkin and Barton 1989) was calculated by GENEPOP 4.0. The indirect estimate of Nm assumes equilibrium populations with constant sizes connected by symmetrical migration (Wright 1969). Because natural populations do not always satisfy these assumptions (Paetkau et al. 2004), migration rates were also estimated by the more flexible method implemented by a Bayesian software program, BAYESASS 1.3 (Wilson and Rannala 2003). BAYESASS does not require the assumption of HWE, and it estimates more contemporary migration (past few generations) than the Nm. In addition, the program allows estimation of asymmetrical migration rates, because only 2 groups were inferred for G. berlandieri. Migration rates were estimated with 5 × 10⁶ iterations, 10⁵ burn-in, and thinning of 2000. Default parameters were used for delta values for allele frequency, migration rate, and inbreeding coefficient. Lastly, detection of contemporary migrants for the last few generations was performed with GENECLASS 2 (Piry et al. 2004), including the 3 tortoises found outside the species range. The frequency-based method and the Bayesian method (Rannala and Mountain 1997) were used to assess the consistency in their estimation of migrants. The ratio of the likelihood of an individual assigned to the group where it was sampled to the maximum likelihood assigned to any group was simulated 10,000 times by Monte-Carlo resampling (Paetkau et al. 2004).

Microsatellite Genotyping

Of 155 individuals tested for microsatellite genotyping, 17 individuals, for which fewer than 3 loci amplified, were eliminated, which resulted in 138 individuals included for further analyses (Table 1). Of 16 microsatellite loci tested, 11 loci were polymorphic (Goag4, Goag6, Goag7, Goag32, GP15, GP30, GP55, GP61, GP81, GP102, and Cm58) (Table 2). One locus did not amplify (Goag3), and 4 loci were monomorphic (Goag5, GP19, GP26, and GP96). These 5 loci were excluded from further analyses. Three loci (GP55, GP61, and GP102) exhibited inconsistent stutter peaks and weak amplification of larger alleles. Because of the potential cross-species amplification problems, subsequent analyses were conducted with only 8 loci, excluding these 3 poorly amplifying loci. Because some of these loci provided imperfect repeats in G. berlandieri, all loci had alleles that required sequence verification to clearly define homology.

Table 2 Comparison of the microsatellite motifs observed in the samples of the Texas tortoise (Gopherus berlandieri) in this study with those described in the original species from which the microsatellite markers were isolated: the desert tortoise (Gopherus agassizii) (Edwards et al. 2003), the gopher tortoise (Gopherus polyphemus) (Schwartz et al. 2003), and the green sea turtle (Chelonia mydas) (FitzSimmons et al. 1995).

View Fullscreen

Individual Assignment Tests

The assignment tests implemented by STRUCTURE 2.2 reached the highest mean likelihood at K  =  2. Plotted individual mean proportions of genotype assigned to either cluster suggested the group boundary weakly separates DUVAL to the north and the rest to the south (Fig. 3). The southern clustering was more uniform than the northern clustering, which contained several individuals that showed greater proportion of genotype assigned to the southern group. The first procedure to infer the number of groups implemented by GENELAND 3.1.2 also provided the highest posterior probability of population membership at K  =  2. The second procedure of 10 runs at K  =  2 produced consistent results. The group boundary was estimated at just south of DUVAL, which separated a northern and southern cluster within Texas (Fig. 2). The population structure estimates performed by STRUCTURE and GENELAND agreed on 2 groups with the weakly defined boundary at southern Duval Co. Therefore, the following analyses treated a cluster that contained SA, FRIO, EZ, WZ, CWMA, and DUVAL as the northern group and a cluster containing UWRGV, STARR, UERGV, and BR as the southern group.

View Figure

• Download as Powerpoint
• Download Figure

Microsatellite Genetic Variation

A wide range of microsatellite polymorphism was observed across loci in both the northern and southern group. Observed heterozygosity (Ho) per locus ranged from 0.138 to 0.792, with the mean of 0.474 in the northern group, and from 0.162 to 0.798, with the mean of 0.525 in the southern group (Table 3). Allelic richness (Ar) ranged from 2.0 to 6.7 in the northern group and 2.0 to 9.5 in the southern group. Private alleles accounted for 8% of the total alleles observed in the northern group and 21% in the southern group (Table 3). The tests of HWE revealed that 3 loci (Goag6, GP30, and GP81) in the northern group and 2 loci (Goag6 and GP81) in the southern group did not conform to HWE after sequential Bonferroni correction (Rice 1989). These loci exhibited significant heterozygote deficits. The 2 loci (GP30 and GP81) were detected as null alleles. Linkage disequilibrium was detected in a different pair of loci in each group, but none of them were significant after sequential Bonferroni correction.

Table 3 Summary statistics of microsatellite genetic diversity in the northern and southern group of the Texas tortoise (Gopherus berlandieri): the observed number of alleles (A), allelic richness (Ar), observed heterozygosity (Ho), and expected heterozygosity (HE) for each microsatellite.^a

View Fullscreen

Population Structure

Genetic distance–based estimates of population structure yielded Fst value of 0.083 and ρ value of 0.095 with statistical significance (p < 0.001 for both). The AMOVA based on Fst estimated that 8.2% of the total genetic variation was accounted for between groups and 90.8% within groups, and the remaining 1% within site. The AMOVA based on Rst showed 9.1% of the total genetic variation between groups and 85.7% within groups, the remaining within sites.

Comparison Between G. berlandieri and G. agassizii

Genetic variation of G. berlandieri was assessed in comparison with the Sonoran populations of G. agassizii (Edwards et al. 2004). Similar overall HE was shown in both species but twice as high Ar in Sonoran G. agassizii (0.557 and 0.621 for HE, and 5.6 and 11.1 for Ar, respectively) (Fig. 4). The Fst and ρ value of G. agassizii were 0.032 and 0.026, both of which are lower than those of G. berlandieri. The STRUCTURE analysis to determine the number of groups of populations generated the highest mean likelihood at K  =  1.

View Figure

• Download as Powerpoint
• Download Figure

Gene Flow

The test of isolation by distance showed a significant correlation between genetic distance and geographic distance across groups (r  =  0.71, p  =  0.0002) (Fig. 5). Isolation by distance among sampling areas within each group was not significant in either group. The indirect estimate of Nm was 3.4 migrants per generation. Contemporary migration rate estimated by the Bayesian approach was 4.1% per generation from the northern to the southern group (95% CI  =  0.04–3.70), and 1.1% from the southern to the northern group (95% CI  =  0.12–13.4). Detection of contemporary migrants revealed 2 individuals, one from SA and another from DUVAL, that were statistically highly significant (p < 0.01). In addition, one individual from EZ and another from CWMA were detected as migrants with significance (p < 0.05). Of 3 individuals found outside the species range, one individual from ANWR was detected as a migrant (p < 0.025). Significant levels for the detected migrants were consistent in both Bayesian and frequency methods.

View Figure

• Download as Powerpoint
• Download Figure

Microsatellite Genetic Variation

Moderate mean heterozygosity in both northern and southern groups demonstrated the appreciable amount of genetic variation present in G. berlandieri. The southern group appears more genetically diverse, because it displayed higher Ar, HE, and a higher number of private alleles than the northern group whose sample size was more than twice as large as that of the southern group. Studies conducted within the range of the southern group reported generally higher tortoise density than that of the northern group (e.g., Auffenberg and Weaver 1969; Kazmaier et al. 2001c). The southern region may have been better able to support a historically larger size across a more contiguous habitat, which we speculate is likely to have facilitated preservation of greater genetic variation.

Numerous studies of turtles and tortoises, including threatened and endangered species, reported moderate-to-high genetic diversity for microsatellite loci (e.g., Sites et al. 1999; Ciofi et al. 2002; Cunningham et al. 2002; Edwards et al. 2004; Tessier et al. 2005; Paquette et al. 2007). Although many investigators have expressed a concern for population bottlenecks, the detection of the loss of genetic variation as a result of modern anthropogenic impacts may be difficult to capture from genetic data (Sumner et al. 2004; de Thoisy et al. 2006). This is especially true for tortoises as genetic drift proceeds slowly in populations of long-lived organisms (Tessier et al. 2005). Genetic diversity, therefore, may not serve as an efficient proxy for modern population declines in chelonians.

Population Structure and Gene Flow

The Texas populations of G. berlandieri showed a statistically significant but limited amount of population structure, weakly resolving into a northern and southern group, with a boundary at southern Duval Co. (Fig. 2). This weak population structure indicates either recent divergence or persistent gene flow between the 2 groups. Significant isolation by distance and the moderate level of Nm indicated a dominant role for gene flow in the observed amount of structure, which prevented differentiation. Furthermore, the greater heterogeneous proportions of genotypes detected in the northern group suggests that population admixture may be taking place more extensively in the northern group, which indicated more frequent migration from the southern into the northern group. Detection of contemporary migrants from the northern group and the absence of migrants from the southern group also support such asymmetrical migration. In contrast, estimation of contemporary migration rates by BAYESASS suggested a higher migration rate in the reverse direction. The estimated contemporary rates may not be reliable when populations are not sufficiently differentiated (Wilson and Rannala 2003) as implicated by a wide range of confidence intervals. A final assessment of the direction of migration, therefore, needs further examination.

Gopherus berlandieri in Texas and G. agassizii in the Sonoran Desert both face increasing habitat loss and population fragmentation (Edwards et al. 2004). The comparative analyses of genetic variation and its distribution in these closely related species may better inform conservation of these species. The Sonoran populations of G. agassizii (Edwards et al. 2004) in Arizona appear genetically more diverse but less structured than G. berlandieri. Edwards et al. (2004) proposed that a lack of significant population structure was a result of ongoing gene flow. Although the Sonoran populations of G. agassizii seem to undergo slightly more frequent gene flow, strong isolation by distance may be maintaining a weak population structure in both species.

Phylogeography

This study hypothesized a potential role of the Nueces River basin in shaping genetic structure of G. berlandieri. The estimated population structure boundary at southern Duval Co. is outside of the modern Nueces River basin, and approximately 50 km south of the main river system. In addition, the northern group includes sampling areas just west (WZ) and south (CWMA) of the Nueces River (Fig. 2). Therefore, the current genetic evidence of weak population structure did not support historical geographic isolation by the Nueces River.

Conservation of G. berlandieri

Sound management strategies require understanding of cohesive forces, whether ecological or evolutionary, that preserve the integrity of a population. Populations may deserve ESU or MU status when significant genetic differentiation is observed. Defining conservation units can be difficult when populations are weakly differentiated, and the extent of migration is complicated by translocation. Further evidence from ecological and genetic data will be required to determine whether the northern and southern groups of G. berlandieri should be managed as separate MUs. The 2 groups appear to have been experiencing gene flow in recent history. However, assessments of current gene flow in G. berlandieri may be complicated by human translocation. For organisms with limited vagility, human translocation will easily surpass the normal dispersal patterns. Gopherus berlandieri may exhibit such a problem because anecdotal evidence suggests that tortoise translocations are common. Among the detected migrants, actual translocation is quite likely for at least 2 tortoises: one from SA, which is the northern extreme of the species range, and another from ANWR, which is located outside the described species range (Fig. 2). As for two other tortoises found outside the species range (i.e., HAYS and second individual from ANWR; Fig. 2), whether they represent relocated tortoises from the vicinity within the species range or a northerly expanding population remains unclear. Human-mediated release has been addressed in studies of other tortoise species (e.g., Schwartz and Karl 2005; Paquette et al. 2007) and is generally considered to have a negative impact that can homogenize genetic differences between populations (Schwartz and Karl 2005). Translocation of tortoises should be minimized in the management strategy for G. berlandieri, not only to maintain the genetic differences observed between the groups but also to avoid the potential spread of disease such as URTD. Finally, given continuing habitat alteration and fragmentation in southern Texas, the connectivity between the groups should be specifically addressed to assure the long-term persistence of G. berlandieri.