By sequencing 1140 bp of the entire mitochondrial cytochrome b gene in 80 individual Z. quitzeoensis specimens from 12 populations (Figure 1 and Additional file 1), 65 haplotypes were detected (Table 1). Fifty four sites were variable (4.73%), 37 were non-synonymous, and 17 were parsimony informative. As expected for a protein-coding gene, third codon positions were the most variable (10) followed by first (5) and second (2). The average ratio of non-synonymous/synonymous substitutions was d_N/d_S = 0.45 (CI: 0.35–0.56) and no evidence of positively selected sites was detected with any of the two methods used. Full details of substitution parameters and evolutionary models are given in Additional file 2.

Weighted parsimony analyses generated 95 equally parsimonious trees (length = 272, CI = 0.871, RI = 0.967). The three methods (NJ, MP and Bayesian) produced largely congruent tree topologies, with proportionally similar bootstrap and posterior probabilities supporting major lineage and population divergences (Figure 2).

Phylogenetic analyses identified two distinct lineages (Figure 2). Lineage I comprised 40 haplotypes distributed in the Lower Lerma and Ameca basins and in Lake Chapala, while Lineage II was composed of 25 haplotypes from the Middle Lerma Basin and from the Cuitzeo and Zacapu Lakes (Figure 2). Mean maximum likelihood and uncorrected p distance between the two lineages were D¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafeiraqKbaebaaaa@2CFC@_ML = 3.05% ± 0.23 and D¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafeiraqKbaebaaaa@2CFC@_p = 2.81% ± 0.21. Both lineages showed a well-defined internal structure (Table 2).

Within Lineage I, we found two main clades. One clade (Ameca Clade) included 18 haplotypes (Hn) from the 3 sampled sites in the Ameca River Basin (Los Veneros, Magdalena and Moloya), and the other clade (Chapala-Lower Lerma Clade: Hn = 22) contained individuals from the localities sampled in the Lower Lerma Basin and Lake Chapala. Distances between Ameca and Chapala-Lower Lerma clades were D¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafeiraqKbaebaaaa@2CFC@_ML = 0.93% ± 0.2 and D¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafeiraqKbaebaaaa@2CFC@_p = 0.92% ± 0.2.

In Lineage II, the first clade comprises the Lake Cuitzeo and Zacapu Lake Basin populations (Cuitzeo-Zacapu Clade: Hn = 20), and the second one, the San Francisco del Rincón population (Middle Lerma Clade: Hn = 5). The D¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafeiraqKbaebaaaa@2CFC@_ML between the two clades was 1.17% and D¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafeiraqKbaebaaaa@2CFC@_p = 1.21 (Table 2).

The statistical parsimony haplotype network for Z. quitzeoensis indicated a similar pattern to that revealed by the phylogenetic analysis and showed the existence of two well-defined evolutionary lineages (Figure 3). The number of mutational steps (21) between lineages exceeded the maximum number of mutational connections justified by the 95% parsimony criterion (14 mutational steps). Thus, each lineage was treated independently in subsequent analyses.

The null hypothesis of the nested contingency analysis of no association between haplotype positions in the cladogram and their geographical locations was rejected in six of the tests performed (P < 0.05) (Figure 3 and Additional file 3).

Allopatric fragmentation could have rendered the geographical pattern of the full cladogram of Z. quitzeoensis. In Lineage I, the five-step clade showed significant values; there was insufficient evidence, however, to discriminate between range expansion, colonization and restricted dispersal or gene flow. Within this lineage, long distance colonization and/or past fragmentation was inferred for Clade 3.6 (Clade Ameca) and long distance colonization associated with subsequent fragmentation followed by range expansion for clades 3.5 (Clade La Luz-Orandino) and 4.2 (Clade Chapala-Lower Lerma) 34.

In Lineage II, the inference key 34 suggested long distance colonization and/or past fragmentation for Clade 4.1 (Lineage II).

The mtDNA Φ_ST values obtained for the comparisons of all the populations ranged from 0.006 to 0.940 (Table 3). Highest significant values were obtained for most comparisons involving the Zacapu population (Φ_ST = 0.518 – 0.940), except for comparisons with two populations from Lake Cuitzeo (Belisario and San Cristóbal). Significant Φ_ST values were also obtained for all the comparisons of Los Veneros with the rest of the populations, except for those involving populations within the same basin (Magdalena and Moloya). The only significant Φ_ST values between samples within the same basin were found in the Lower Lerma River, except for Orandino and La Platanera (Table 3).

Pairwise F_ST values based on microsatellite data were significant for most of the comparisons, except for Moloya-Magdalena in the Ameca basin and for Belisario-San Cristóbal in the Cuitzeo Lake basin (Table 3). Highest F_ST values were observed for the comparisons between La Luz and San Francisco del Rincón, and La Platanera with La Mintzita (F_ST = 0.455, 0.408 and 0.406 respectively). Although the La Luz and La Platanera populations occur in the same basin (Lower Lerma), their F_ST values were among the highest. Notwithstanding, the other population inhabiting the Lower Lerma Basin (i.e. Orandino) showed significant yet much lower F_ST values when compared with La Luz and La Platanera (F_ST = 0.271 and 0.224 respectively).

The AMOVA performed for the mitochondrial and nuclear data, revealed significant structure among populations (Φ_ST = 0.836, P < 0.001 and F_ST = 0.262, P < 0.001 respectively). For the subsequent analyses, populations were grouped in different hierarchical arrangements according to previous information and to uncover groupings obtained in the previous analyses (e. g. phylogenetic analysis and NCA) and by their biogeographical arrangement (Table 4). Significant values were obtained when the biogeographic arrangement of basins 6 was considered (Φ_CT = 0.787, P < 0.001, F_CT = 0.107, P < 0.05). When the basins where each lineage was found were considered separately, only Lineage I showed significant structure for mtDNA (Table 5). When two gene pools corresponding to the two lineages found were considered, 74.8% (P < 0.001) of the total variance was explained as differences among groups for mtDNA and 11.18% in the case of microsatellite data (P < 0.01). Division of the populations into the groups suggested by the phylogenetic and NCA analysis maximised among-group variance for the mtDNA data (Φ_CT = 0.823, P < 0.001). Thus, the structure previously obtained in the phylogenetic and NCA was statistically supported for the mtDNA, but not for the microsatellite data.

The Mantel test revealed a significant correlation between geographical and mitochondrial genetic distances (r = 0.4123, P = 0.01). Otherwise, non significant correlation was found considering the genetic distances obtained using the microsatellite data (r = 0.1894, P = 0.179) (Figure 4).

The Bayesian structure analysis for the microsatellite data clearly revealed a genetic structure among the specimens analysed. According to the maximum likelihood value, we estimated a number of genetic clusters of K = 6. Estimates of lnPr(X|K) increased rapidly between K = 1 and K ≤ 6, whereas beyond K > 6 lnPr(X|K) started to oscillate. But when we considered Δ K, we obtained clear peak at K = 5 (Figure 5). Conversely, the highest global F_ST value was found for K = 6, indicating that these clusters explained the maximum level of structure in our sample (Additional file 4). However, for the following discussion we adopted the more conservative measure of K = 5, obtained with the correction of Evanno et al. 35. All populations were assigned with high probability (Q = 0.844 – 0.962) to their inferred cluster.

The results derived from the microsatellite NJ tree were highly congruent with those from the Bayesian clustering analysis and most of the phylogenetic mtDNA tree, but with some differences within Linage I (Figure 5). Two well-supported lineages were obtained. In Lineage II, all samples from the Cuitzeo and Zacapu basins grouped together, and San Francisco del Rincón appeared differentiated from these. Otherwise, Lineage I showed a higher degree of structure, with its populations distributed in three different clusters (Figure 5). When considering K = 6, Orandino represented a separate cluster. Slightly different results, regarding the position of the Lower Lerma populations, were rendered by the two types of marker for population relationships (Figure 5). The population assignment test correctly assigned 80.59% of the individuals to their original population. While for the Lineage I populations, 91.5% correct assignments were obtained, in Lineage II, gene flow was found among all populations (72% of the individuals were assigned correctly) except for the site San Francisco del Rincón, whose individuals were all unequivocally assigned.

Overall gene diversity was H_d = 0.987 and nucleotide diversity π = 0.0172. Most haplotypes were found at single sites. Only two haplotypes were shared among individuals from different localities. One of them was also the most common haplotype overall, found in 8 individuals from the sites Zacapu, San Cristóbal and Belisario (ZAC-SCR-BEL 59), and the other one was found in 3 individuals from Magdalena and Moloya (MAG-MOL 2) (Figures 2 and 3). The rest of the haplotypes were shared among individuals within the same localities.

All the populations analysed showed a remarkable number of haplotypes (Table 1). The level of gene diversity within populations was H¯ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGafmisaGKbaebaaaa@2D06@_d = 0.987 (0.714 – 1.00). However, the nucleotide diversity exhibited by most of the tested populations (π = 0.0015 – 0.0049) was low. Mean pairwise nucleotide diversity (k) ranged from 1.71 to 5.69 within populations (average number of nucleotide differences for the whole sample k = 19.79; Table 1). The mismatch distribution (MMD) for all the data set was bimodal (not shown), one of the peaks represents the differences between lineages, and the other the differences among individuals within lineages. The two lineages and the different clades obtained in the previous analyses were tested independently. The MMD for Lineage I was unimodal and bimodal for Lineage II (Figure 6), which is expected when populations are geographically subdivided 36. In both cases, raggedness indices (r) were not significant, thus not rejecting the null hypothesis of stationarity. Conversely, Tajima's D-statistic and Fu's statistic (F_s) were significantly negative and supported the expansion model for the different groups tested (Table 5). Both tests show a significant excess of the number of segregating sites and singletons compared to the average pairwise sequence divergence. These tests indicate that the different groups analyzed are in mutation-migration-drift genetic disequilibrium with respect to mtDNA alleles. The different results obtained by these three demographic tests might be due to the low power of mismatch distribution based tests (e.g. raggedness), compared with methods based on the mutation frequency (e.g. Tajima's D) or haplotype distribution (e.g. Fu's F). In a variety of cases and for large samples sizes, Fu's F has been proved to be the most powerful test to detect population growth 37. Values of τ differed between lineages and among clades. For Lineage I, the mean of the mismatch distribution was τ = 5.647 and the estimated time since population growth was 1.5 × 10⁶ years before present. Within Lineage I, the estimated time since population expansion for the two clades obtained (Ameca and Chapala-Lower Lerma) were ca. 1.4 × 10⁶ (τ = 5.08) and ca. 1.08 × 10⁶ (τ = 5.647), respectively. In Lineage II, we estimated ca. 860,000 years since population expansion according to the value of τ = 3.09. For the clades within Lineage II, the estimates of time since population expansion were ca. 790,000 years for Cuitzeo-Zacapu (τ = 2.83) and ca. 550,000 years for San Francisco del Rincón (τ = 1.95).

The trees obtained by the different methods (NJ, MP and BI) using both types of markers (cytochrome b and microsatellites) distinguished two independent lineages, supporting the conclusions of previous studies 21 in which two divergent groups were identified within Z. quitzeoensis. Lineage I inhabits areas west of the Middle Lerma, including the Lower Lerma, Ameca and the Chapala Lake basins. Lineage II occurs at sites east of the Angulo River and the Middle Lerma Basin, including the Cuitzeo and Zacapu lakes. Our NCA also supported this conclusion, indicating the two main lineages were not nested together; moreover, the 21 mutation steps between the two lineages exceeded the 95% parsimony limits (14 mutation steps). This result suggests allopatric fragmentation between the two clades.

The formation of the two lineages of Z. quitzeoensis was dated at ca. 3.3 Mya based on the molecular clock calibration of 0.9% divergence per million years 21. These two lineages could be the result of a dispersal process of the ancestral population from the Lower Lerma-Chapala area to the Middle Lerma-Cuitzeo-Zacapu area (Figure 1), followed by an isolation event. This hypothesis is based on the facts that: Lineage I is the most widely distributed group (distributed in Chapala-Ameca-Lower Lerma drainage), and it shows, overall, a higher genetic diversity than Lineage II (Hd = 0.991, S = 76 and Hd = 0.946, S = 54, respectively).

The dispersion event we refer to, from the Chapala-Lower Lerma to the Middle Lerma-Cuitzeo-Zacapu area, could have been promoted by a period of high precipitation and humidity occurred in Central Mexico in the early Pliocene (5.2 – 3.6 Mya). This could have caused an increase of the water bodies level in this area causing them to get in contact, as was previously proposed for other lakes in Central Mexico 3839. The subsequent isolation of the ancestor of the two lineages that induced allopatric fragmentation, could have been the result of the end of the humid period, and/or the formation of a biogeographical barrier promoted by the geologic activity of the Penjamillo Graben 4041, the Chapala-Tula fault or the activity and formation of the Corredor Tarasco volcanic field, which commenced during the Late Miocene-Early Pliocene 40 (Figure 1). This climatic change and the high tectonic activity have been proposed as the causes for the isolation of the palaeolakes along the TMVB during the Late Miocene-Pliocene (Figure 1) 38.

Since recent geological times, some of the regions where these two lineages occur have been connected, and at present constitute one single hydrologic system (Lerma River). However, our findings of ancestral isolation between Lower Lerma-Chapala and Middle Lerma-Zacapu-Cuitzeo are supported by at least one pair of sister species with the same cladogenetic pattern, Skiffia lermae-S. multipunctata, dated around 3.2 Mya 21. This indicates that the same biogeographic event could have promoted the isolation of the two divergent groups and consequently they could be considered as two ESUs 4243. Furthermore, considering the morphological differences between groups, and pending of more detailed morphometric studies, they could be considered as two species.

The results of all our analyses revealed significant genetic structure and differentiation among populations between and within the two lineages, but with certain differences as indicated by the two types of molecular markers.

Mitochondrial and nuclear markers reveal different parts of the evolutionary history of the organisms due to their different inheritance modes and mutation rates. Even though mitochondrial markers coalesce faster than nuclear markers due to the smaller effective size and the lack of recombination, microsatellites have a much faster mutation rate which makes them useful to detect more recent processes 5051.

In general, the relationships inferred with the mitochondrial and microsatellite data set were congruent, but some differences were found. Whereas both markers retrieved the existence of two main clades, the relationships within them were not the same (Figure 5). Discrepancies were observed in Lineage I for the comparisons of the Chapala-Lower Lerma populations. In this case, both makers showed significant differences among geographically close populations. High and significant pairwise F_ST values were obtained for most of the comparisons of populations from the Lower Lerma basin (Table 3), which also were assigned to different genetic clusters by the Bayesian clustering analysis and microsatellite NJ tree, but formed a single clade according to mtDNA (Figure 5). Some of these localities have recently suffered a severe reduction in the size of their water bodies, and become isolated with a consequent increase of the effects of genetic drift, resulting in genetic differentiation. Moreover, it has been suggested that the Lower Lerma populations suffered a more intense genetic drift than other populations within Z. quitzeoensis 49. Similarly the non-significant correlation between geographic and genetic distances based on the microsatellite data might be explained by the stochastic processes that are recently shaping the genetic structure of Z. quitzeoensis, and consequently the high genetic differences between nearby localities.

The differences found in the genetic diversity values were much higher for the microsatellite than the mitochondrial data, which showed a high diversity for all populations (Table 1 and Additional file 5). The populations of Zacapu and San Francisco del Rincón, which showed allelic richness values of only 2.93 and 2.97, still retained haplotypic diversities of 0.917 and 0.857. Thus, eventhough samples sizes are, in some cases, low and unequal we might deduce that the loss of diversity is occurring at a different rate for the two types of markers 52. Additionally, none of the Tajima's D values, either for single populations or lineages, were significantly positive, which would be indicative of a bottleneck 53, whereas seven out of the 10 populations analyzed showed a significant signature for a recent bottleneck, with at least one of the three microsatellite mutational models (Additional file 5).

The importance of fish diversity in Central Mexico has been largely recognized 54 but very few efforts have been made to establish a basis for their conservation. Further, no studies have addressed the evolutionary processes underlying such diversity targeted at maintaining these processes. Our results, derived from both phylogeographic and population analyses, could prompt certain conservation management strategies. First, the two reciprocally monophyletic lineages of Z. quitzeoensis, support the idea that future conservation plans should be aimed at managing the populations of both lineages independently and, furthermore, be considered as two ESUs 4243.

Moreover, within each lineage, we found genetic structure for the two markers, supporting previously identified Operational Conservation Units (OCU's) for Zoogoneticus 49. When the species under study shows high genetic structure, ideally all the populations should be protected since they contain unique portions of the total variation of the species. This may increase the adaptation and the survival possibilities of the species as a whole.

Fin clips were obtained from individuals of Zoogoneticus quitzeoensis. Specimens were collected from 12 sites distributed across six regions along the Mesa Central of Mexico (Figure 1 and Additional file 1), representing most of the distribution range of the species. The fishes were sampled using minnow traps, seine nets and by electrofishing. Tissue was preserved in 96% ethanol and most fishes were returned to the water unharmed. A few fish specimens were incorporated into the Goodeid Conservation Program of the Universidad Michoacana de San Nicolás de Hidalgo. Because of the endangered status of Z. quitzeoensis 27 and its scarcity at all the collection sites, sample sizes ranged from 7 to 20 individuals per population (Additional file 1). Similar sample sizes have been used successfully in similar phylogeograpic studies of freshwater fauna 55565758596061. However, it has long been proved that phylogenetic and population genetic inferences are sensitive to the number of taxa included. The number of specimens and populations needed to resolve their relationships depend on the amount of polymorphism relative to the extent of divergence. Thus, it would be appropriate to use small samples per groups when most of the variation occurs among groups 62. However, in such cases other inferences should be taken with caution as some estimates, for example those based on genetic diversity, could be biased. Sampling at each site was conducted during the same field season. Total genomic DNA was extracted according to standard CTAB and phenol-chloroform extraction procedures 63.

In order to compare the genetic structure between mitochondrial and nuclear markers, we used a microsatellite data set from a previous study about the effects of human impacts on the genetic variability of Z. quitzeoensis 49. However, not all the populations included here for the mtDNA study were available at that time, and therefore the number of analysed populations differs between marker types.

Two overlapping fragments of the cytochrome b gene (1140 bp total) were amplified via polymerase chain reaction (PCR) from 80 specimens distributed in 12 sampling sites. The primers used were those used in Machordom & Doadrio 64. The amplification process was conducted using the conditions described elsewhere 21. PCR products were sequenced in an ABI PRISM 3700 DNA analyser. Chromatograms and alignments were visually checked and verified. All sequences were deposited in GenBank under accession numbers: EU679420-EU679499.

We used phylogenetic tree-building algorithms to infer the phylogenetic relationships among sequence haplotypes. Maximum Parsimony Analysis (MP) was performed by heuristic searching with the tree-bisection-reconnection (TBR) branch swapping algorithm and random stepwise sequence addition using 10 replicates. Two different weights were given to the characters; first all characters were equally weighted, and second, transversions (Tv) were assigned eight times the weight of transitions (Ti) according to the empirically determined Tv/Ti ratio for Zoogoneticus obtained in PAUP 4.0b10 65. The robustness of the MP topologies was assessed by bootstrapping with 1000 replicates (full heuristic search) of 10 random stepwise addition replicates each. The model of DNA substitution that best fitted the data set was selected using MODELTEST 3.7 66 using the Bayesian information criterion (BIC) for each codon position and the Akaike information criterion (AIC). A Neighbour-Joining (NJ) phylogram was obtained using maximum-likelihood distances according to the model selected by AIC. Bootstrap values for this analysis were obtained from 1000 replications. All phylogenetic analyses were performed using PAUP 4.0b10. Bayesian analysis was conducted with MrBayes 3.1.1 67. By simulating a Markov Chain Monte Carlo reaction for 2 × 10⁶ cycles and using the substitution model obtained for each codon position selected by BIC criterion, 20,000 trees were generated, 1000 of which were burned and discarded. Posterior clade probabilities were used to assess node support. To identify ancestral and derived haplotypes, the trees were rooted using Zoogoneticus tequila, the sister species of Z. quitzeoensis, as outgroup, and a molecular clock of 0.9% per million years was applied to the pairwise uncorrected p genetic distances 21.

Evidence of positive selection was sought using a codon-based approach as implemented in Datamonkey 68. This method does not need to assume equal synonymous substitution rates throughout the sequence and allows to choose the most appropriate model for nucleotide substitution. We used the single likelihood ancestor counting (SLAC) and fixed effects likelihood (FEL) approaches 69 using a P value of 0.1. In both cases, ambiguities in the consensus sequence were averaged in the analysis.

We constructed a 95% statistical parsimonious un-rooted haplotype network using TCS 1.18 70. As a complementary method to those performed before, and considering the limitations and drawbacks of Nested Clade Analysis 4871, we tested geographical association among haplotypes and clades, based on the most parsimonious haplotype network and followed by a nested cladistic analysis (NCA), as described by Templeton 3472. To test for significant associations between clades and geographical sites, nested contingency analysis 73 was conducted by the program GEODIS 2.2 74. The AUTOINFER 1.0 75 software package was used to infer the most suitable population structure model and historical scenario for the observed geographical associations.

Pairwise Φ_ST values were calculated among all geographic populations as an estimate of genetic differentiation. To assess the significance of genetic differentiation at different hierarchical levels, an analysis of molecular variance (AMOVA) was performed as described by Excoffier et al. 76. Populations were initially grouped according to previous biogeographical information 6. In subsequent analyses, we considered the information obtained in both the phylogenetic and the NCA analyses, but other hierarchical arrangements were also tested. Statistical significance was assessed using 20,000 permutations. A Mantel test (100,000 permutations, 77) served to evaluate correlations between linear geographic distances and genetic distances. All analyses were performed using ARLEQUIN 3.1. 78.

Population genetic statistics, such as the number of polymorphic sites (S), haplotype diversity (H_d, 79), nucleotide diversity (π, 79) and the average number of pairwise nucleotide differences (k, 80) were calculated using DnaSP 4.0 81.

To investigate the demographic history of the groups identified in the phylogenetic analyses and through the AMOVA results, a mismatch distribution analysis (MMD) of pairwise substitution differences among haplotypes was performed for the whole data set and the lineages obtained. Deviations from the constant population size model were further tested using the Harpending's raggedness index (r) 82. To test for deviations from neutrality we used Tajima's D 53 and Fu's Fs 83 tests as implemented in DnaSP 4.0 81. We used the MMD age expansion parameter (τ) to date the onset of population expansion 84. This was done by calculating τ using the equation τ = 2μt, were μ is the sum of per-nucleotide mutation rates in the DNA region under study (0.9% PMY; 21) and t is time in generations (0.5 for goodeines).

We examined a previous data set, consisting of 135 individuals of Z. quitzeoensis from 10 populations, genotyped for five microsatellite loci (Additional file 5) 49 to look for differences in the allele frequencies of the populations by estimating F_ST between all sample pairs, according to Weir & Cockerham 85, using ARLEQUIN 3.1. The significance of these estimates was assessed using 10,000 data permutations corrected by Bonferroni adjustment 86.

Geographical and phylogenetical genetic variation were compared among populations and clades by AMOVA. We assessed genetic isolation by distance 87, testing for independence between F_ST estimates and geographical distances using a Mantel test 77; regression matrices of F_ST/1-F_ST values versus the linear distance between sample pairs. All these analyses are implemented in ARLEQUIN 3.1.

To determine relationships among the sampled populations a neighbour-joining tree was created using the POPTREE program 88 based on D_A modified Cavalli-Sforza distances 89 with 5000 bootstrap replications.

Because of the uncertainty that the geographical assignment of individuals to populations could represent biologically significant entities, a Bayesian clustering method was conducted as implemented in the program STRUCTURE 2.1 90. We performed a series of independent runs from K = 1 to 8 populations assuming correlated allele frequencies and an admixture model. For each value of K, the MCMC scheme was run with a burn-in period of 5 × 10⁵ steps and chain length of 5 × 10⁶. Multiple runs were performed for each K to assess convergence of the results. Mean log probabilities were used to calculate ΔK (i.e. a quantity based on the second-order rate of change of the log probability of data between successive K values), to find the true K following the method of Evanno et al. 35. Global F_ST values were calculated for each K to find out which of the structures inferred explained the highest percentage of genetic variation. To assess the level of admixture and gene flow between the central Mexican Basins, we ran a population assignment test using GeneClass 2.0 91.