Matching loci surveyed to questions asked in phylogeography

Abstract

Although mitochondrial DNA (mtDNA) has long been used for assessing genetic variation within and between populations, its workhorse role in phylogeography has been criticized owing to its single-locus nature. The only choice for testing mtDNA results is to survey nuclear loci, which brings into contrast the difference in locus effective size and coalescence times. Thus, it remains unclear how erroneous mtDNA-based estimates of species history might be, especially for evolutionary events in the recent past. To test the robustness of mtDNA and nuclear sequences in phylogeography, we provide one of the largest paired comparisons of summary statistics and demographic parameters estimated from mitochondrial, five Z-linked and 10 autosomal genes of 30 avian species co-distributed in the Caucasus and Europe. The results suggest that mtDNA is robust in estimating inter-population divergence but not in intra-population diversity, which is sensitive to population size change. Here, we provide empirical evidence showing that mtDNA was more likely to detect population divergence than any other single locus owing to its smaller N_e and thus faster coalescent time. Therefore, at least in birds, numerous studies that have based their inferences of phylogeographic patterns solely on mtDNA should not be readily dismissed.

1. Introduction

Genetic variation within and between populations has been used as a ‘yardstick’ to measure divergence between populations and to quantify their effective population sizes [1]. Genetic analyses based on mitochondrial DNA (mtDNA) have widely been used during the past three decades to infer animal phylogeography, focusing on geographical distribution of haplotypes and the underlying process affecting their distribution [2]. The high mutation rate, fast coalescence time owing to a small effective size and matrilineal inheritance make mtDNA more likely to track lineage divergence than any single nuclear gene, and is thus a relatively leading molecular indicator of population differentiation [2]. Nevertheless, analyses of mtDNA generate a single-gene tree and any single-locus tree could present a biased view of species history owing to the stochastic nature of the coalescence process or effects of selection [3–5]. In effect, the inferences from the first three decades of mtDNA phylogeography have been called into question [4,6].

Ironically, we do not have a clear idea about how biases affecting the distribution of mtDNA variation might influence conclusions about species history. It has been suggested that mtDNA data are efficient, considering both experimental cost and accuracy, in detecting phylogeographic patterns [2,7], but might be unreliable for quantifying evolutionary processes [6,8]. In general, genetic divergence between populations is less affected by population size change and selection than genetic diversity within populations [9–11]. If so, mtDNA divergence between populations should be a good predictor of actual divergence time. Alternatively, some studies suggest that the relationship between mtDNA and actual divergence are influenced by population size or selection [5,12–14]. If so, mtDNA distance might provide a biased view of population divergence.

The solution favoured in the literature is to analyse variation at multiple nuclear loci. However, the coalescence times of mtDNA and autosomal and sex-linked nuclear genes differ greatly. Given that the average coalescence time of mtDNA is 1 N_e generations, and nuclear loci 3 N_e (sex-linked) or 4 N_e (autosomal), there is a window of time between 1 N_e and 3–4 N_e when individual nuclear loci on average will be unable to detect population differentiation, whereas mtDNA probably can. There is a tendency to use more nuclear loci in hope of detecting all events, at least those greater than 1 N_e in the past. However, the average coalescence time for a series of nuclear loci is still 3–4 N_e, and thus analyses based on multiple nuclear loci are not always better than mtDNA-based ones in resolving recent phylogeographic patterns ([8], but see [15]) and increase experimental cost considerably. In addition, disparate locus effective sizes among mtDNA and nuclear genes can have a confounding effect on tracking demographic history of species [16–18], requiring further examination.

The forest bird community of the Caucasus is isolated by over 700 km from Palaearctic forests by the Black Sea and Sea of Azov on the west, the Caspian Sea on the east, and by arid grasslands and semi-deserts of southeast Europe and the Anatolian Peninsula on the north and south, respectively [19]. We sequenced mtDNA, 10 autosomal and five Z-linked introns from Caucasian and European Russian (European hereafter) populations for 30 forest-dwelling avian species that represent diverse phylogenetic backgrounds. Because these species share these geographical and ecological barriers (C.-M. Hung, S. V. Drovetski, R. M. Zink 2016, unpublished data), they in effect represent replicate tests across a common geographical area. Our goals were to assess the relationship between mtDNA and nuclear DNA (nuDNA) estimates of: (i) intra-population genetic diversity, (ii) inter-population genetic divergence, and (iii) the effect of population size on genetic divergence estimate. This study can help clarify whether mtDNA provides a fair or biased perspective on different aspects of species' history, and what we should take from the mtDNA era of phylogeography.

2. Material and methods

(a). Taxon sampling and gene sequencing

For each of the 30 studied species, we analysed five individuals from the Caucasian population and five from the European population for each locus (however, we treated Ficedula semitorquata and F. hypoleuca as the same taxa, and the former as the Caucasian population and the latter as the European population; electronic supplementary material, supplementary methods section, tables S1 and S2). We obtained the complete mitochondrial ND2 gene sequences (1041 bp) for each sample. We sequenced 10 autosomal (MPP, ACL, 00132, 08352, 09385, 13380, 14765 and 17483, RHO and TGFB) [20–23] and five Z-linked introns (MUSK-TL, H. V. Miranda 2012, personal communication; ACO, ABCA1, NNT and MAK10) [23–25] for these samples. For loci, in which we could not sequence at least four individuals for both populations of a species, we removed the data from further analyses. Despite sampling gaps, the data matrix included 368 species × gene comparisons (electronic supplementary material, table S1).

The phases of sequences containing indels were sorted using the program Champuru [26] or manually [8,27,28]. Sites where indels occurred were excluded from further analyses. The allelic states of individuals with multiple heterozygous sites but no indels were resolved using Phase 2.1.1 [29] implemented in DnaSP 5 [30]. The number of interactions was set as 5000, the thinning interval as 10, the number of burn-in iterations as 500 and other variables as default. Homozygous genotypes and genotypes with single heterozygous sites or with indels were included in the Phase analyses to improve their performance [8]. We tested intra-locus recombination based on recombination rates estimated using a Bayesian method implemented in Phase [8,11] (see the electronic supplementary material, supplementary methods section, for details).

(b). Summary statistics and neutrality test

The program MTML-msBayes [31] was used to compute the nucleotide diversity of each population (π) and that averaged over the two populations (π_ave hereafter), nucleotide distance (D) and net nucleotide distance (D_A) [32] between the two populations for each gene. D_A = D − (π_X + π_Y)/2, where D was the nucleotide distance between population X and Y, and π_X and π_Y were the nucleotide diversity of population X and Y, respectively [32]. Thus, the difference between D and D_A was that the latter controlled for intra-population diversity and the former did not. We calculated the F_ST values for each gene between the two populations using the program DnaSP 5.

We used Hudson–Kreitman–Aguade (HKA) tests [9] implemented in the HKA program (http://genfaculty.rutgers.edu/hey/software) to test the neutrality of the mtDNA and nuclear introns. The HKA test examines whether the association between the level of intra-taxon polymorphism and inter-taxon divergence departs from the expectation of an evolutionary neutral model for multiple genes. A few introns that were fixed at some species (electronic supplementary material, table S3) were removed because HKA tests require genetic variation. Given that (i) these fixed introns (noncoding genes) are more likely to represent neutral patterns and (ii) exclusion of them from population genetic analyses might artificially increase estimated nuclear genetic diversity of species, we kept these loci for further analyses.

(c). Comparisons among mitochondrial DNA, autosomal and Z-linked summary statistics

We performed linear regression analyses to compare the π_ave, D, D_A and F_ST values of mtDNA with those of autosomal or Z-linked introns. We also compared the values of individual Z-linked introns and autosomal introns. In addition, there can be a potential problem caused by the confounding effect of inter-intron variations in substitution rates and coalescence times when comparing multiple introns against the same mtDNA or Z-linked intron. Therefore, we also performed another two sets of comparisons based on (i) the summary statistics of each intron separately, and (ii) those ones averaged over each species' (autosomal or Z-linked) introns, respectively.

(d). Comparisons between demographic patterns inferred from mitochondrial DNA and nuclear DNA

To test the relationship between mtDNA summary statistics and the corresponding demographic history of the species, we estimated demographic parameters using a coalescence-based model, Isolation with Migration (IMa) [33], based on all nuclear loci. These nuclear loci were neutral and had no intra-locus recombination (see Results), fitting the assumptions of the IMa model. We used IMa to estimate the migration rates between Caucasian and European populations in both directions (m₁ and m₂, respectively; m = M/µ; M: the proportion of immigrants in a population per generation and µ is the geometric mean of the substitution rates of genes), their and ancestral effective population size parameters (θ₁, θ₂ and θ_A, respectively; θ = 4 N_eµ), and their divergence time (t = Tµ, T: the number of generations or years since divergence). We did not perform such analysis on datasets with only autosomal or only Z-linked introns because most IMa analyses could not return reliable estimates based on the smaller datasets. The Infinite Sites model was used for the nuclear loci except for those not fitting the assumptions of the model, in which case the Hasegawa–Kishino–Yano model was used. Each run involved 1–2 × 10⁷ steps after a burn-in of 10⁶ steps, employing 30–40 Markov chain Monte Carlo chains with geometric heating. Plots of trend lines were examined to assess convergence in parameter estimates. In addition, at least two independent runs were performed to assure convergence. Parameters that did not have a clear peak in their posterior probability distributions or did not converge between independent runs were excluded from further analyses.

We performed linear regression analyses to compare mtDNA π with IMa θ, and mtDNA D and D_A with IMa t. Because F_ST can be used to estimate the level of gene flow between populations and the two parameters are supposed to be negatively correlated [34], we also performed linear regression analyses to compare mtDNA F_ST with IMa m₁ + m₂.

In addition, we converted the µ-scaled demographic parameters of IMa to real values (i.e. N_e in the unit of individual, T in the unit of year and M in the unit of per generation) and compared them with the corresponding mtDNA summary statistics as above. To convert the scaled parameters, we calculated the geometric mean of the substitution rates of these nuclear loci by multiplying the sequence lengths by 1.35 × 10⁻⁹ and 1.62 × 10⁻⁹ substitutions site⁻¹ yr⁻¹ for autosomal and Z-linked introns, respectively [35]. We set the generation times to 2 years by multiplying the ages of sexual maturity by two [36] of these small passerine species that generally reach sexual maturity at an age of 1 year.

(e). Comparisons between mitochondrial DNA networks and nuclear-based demographic parameters

Network 4.5.1.6 (fluxus-engineering.com) was used to generate a minimum spanning haplotype networks for the ND2 gene to reveal the patterns of mtDNA divergence between Caucasian and European populations. We then compared the estimated IMa parameters (t and m₁ + m₂) of species with the three different types of ND2 networks to determine if they were consistent with inferences from nuclear loci.

(f). Effect of N_e on estimate of genetic divergence

According to the nearly neutral theory [37], small populations tend to have higher substitution rates for coding genes than large populations owing to relaxation of purifying selection in the former [17]. Thus, we examined the effect of N_e on the inter-population genetic distances (as a proxy of substitution rates) of mtDNA (coding genes), autosomal and Z-linked introns (noncoding genes). If relaxed purifying selection had significant effect in our data, we expected a negative relationship between species' N_e and the genetic distance ratios of mtDNA to autosomal introns or mtDNA to Z-linked introns. We compared the average IMa-estimated N_e of the two populations with mtDNA/autosomal and mtDNA/Z-linked D as well as D_A ratios.

3. Results

(a). Summary statistics and neutrality tests

The π values showed similar ranges of variation between the Caucasian and European populations across the 30 species for each of the three types of genes, mtDNA, Z-linked and autosomal introns (electronic supplementary material, tables S4–S7). The D, D_A and F_ST values for mtDNA were generally larger than those averaged over autosomal or Z-linked introns (electronic supplementary material, tables S4, S8–S10 and supplementary results section).

The multilocus HKA tests could not reject the neutral model for the mtDNA and autosomal and Z-linked introns of the 30 species (electronic supplementary material, table S3). No intra-locus recombination was detected in the nuclear introns of the 30 species (data not shown).

(b). Comparisons among mitochondrial DNA, autosomal and Z-linked summary statistics

MtDNA D_A and F_ST values were positively correlated with autosomal D_A (p < 0.0001, r² = 0.2303) and F_ST values (p < 0.0001, r² = 0.1818) and Z-linked D_A (p < 0.0001, r² = 0.2163) and F_ST values (p < 0.0001, r² = 0.2607; figure 1), respectively. MtDNA D values were positively correlated with Z-linked D values (p = 0.0006, r² = 0.1099) but not significantly correlated with autosomal D values (p = 0.207, r² = 0.0068; electronic supplementary material, figure S1). MtDNA π_ave values were not significantly correlated with autosomal π_ave values (p = 0.12, r² = 0.001) although they were correlated with Z-linked π_ave values (p = 0.03, r² = 0.0448; electronic supplementary material, figure S1).

Figure 1.

Relationships of net genetic distance (D_A) and F_ST values between mtDNA and nuDNA. The correlation between the D_A values of ND2 and those of (a) autosomal and (b) Z-linked introns for 29 taxon pairs; (c,d) those of the F_ST values (***p < 0.001).

Among the five Z-linked introns, for which their summary statistics were individually compared with those of autosomal introns, all five loci were significantly correlated in D_A values (p < 0.0001, r² = 0.077–0.2893) and F_ST values (p < 0.05, r² = 0.0243–0.5951), one locus was significantly correlated in D values (ACO: p = 0.0014, r² = 0.0598) and two loci were significantly correlated in π_ave values (ACO: p = 0.003, r² = 0.0753; NNT: p = 0.014, r² = 0.0556; electronic supplementary material, figure S2).

The major patterns remained when we compared the π_ave, D, D_A and F_ST values of mtDNA or individual Z-linked introns: (i) with those of each intron separately (electronic supplementary material, figures S3, S4, and table S11), and (ii) with those ones averaged over each species' (autosomal or Z-linked) introns (electronic supplementary material, figures S5 and S6). Nevertheless, the r² values of linear regression based on these two sets of data (see the electronic supplementary material, supplementary results section, for details) were generally larger than those based on the full dataset.

(c). Mitochondrial DNA summary statistics, networks and nuclear-based demographic parameters

Both of the mtDNA D and D_A values were positively correlated with the IMa t values (D: p = 0.0177, r² = 0.23; D_A: p = 0.0339, r² = 0.1887; figure 2). The F_ST values of mtDNA were negatively correlated with the IMa m₁ + m₂ values (p = 0.0434, r² = 0.2605; figure 2). By contrast, the mtDNA π values were not significantly correlated with the IMa θ values for each population (Caucasus: p = 0.327, r² = 0.037; Europe: p = 0.698, r² = 0.0061). Similar results were obtained when the comparisons were based on the converted IMa parameters (electronic supplementary material, figure S7 and supplementary results section).

Figure 2.

Relationships between (a) mtDNA D and IMa t, (b) mtDNA D_A and IMa t, and (c) mtNDA F_ST and IMa m₁ + m₂ (*p < 0.05).

The ND2 haplotype networks showed three major types of structures. One had no clear structure (n = 15), the second had two clear clusters corresponding to the Caucasian and European populations, respectively, and separated from each other by eight to 50 mutation steps (n = 11), and the third had two distinct clusters separated from each other by eight to 17 mutation steps, one of which contained mixed haplotypes from the two populations, implying introgression between divergent populations (n = 3; electronic supplementary material, figure S8).

The ND2 haplotype networks with two distinct Caucasian and European clusters had the lowest mean m₁ + m₂ values (6.2091 ± 3.831 s.e., n = 9), and the unstructured networks (25.5817 ± 15.39, n = 6) and the networks with introgression signals (24.375, n = 1) had around four times higher mean m₁ + m₂ values (electronic supplementary material, table S12). The unstructured networks had the smallest mean t-values (0.3358 ± 0.0748, n = 13), and the geographically structured networks (0.7773 ± 0.1304, n = 8) and the networks with introgression signals (1.2902 ± 0.4314, n = 3) had around two to four times higher mean t-values (electronic supplementary material, table S12). The same conclusions were reached in the comparisons based on converted IMa parameters (i.e. M₁ + M₂ and T; data not shown).

(d). Effect of N_e on estimate of genetic divergence

There was no correlation between IMa N_e values and mtDNA/autosomal (p = 0.462, r² = 0.0227) or mtDNA/Z-linked D ratios (p = 0.621, r² = 0.0104; electronic supplementary material, figure S9). There was also no correlation between the IMa N_e values and mtDNA/autosomal (p = 0.534, r² = 0.0163) or mtDNA/Z-linked D_A ratios (p = 0.0689, r² = 0.1367; electronic supplementary material, figure S9).

4. Discussion

MtDNA has been one of the most widely used genetic markers in numerous biological fields such as phylogeography, systematics and conservation genetics [38–40]. Thousands of studies have been based solely on mtDNA and the question has arisen as to whether the results of such studies can be accepted or whether they must be tested by corresponding analyses of nuclear loci. In our opinion, a relatively few instances of discordance between mitochondrial and nuclear genes has been emphasized [41–43], despite more frequent cases of congruence between the two types of markers [2,8,44–46]. Our systematic examinations suggest that mtDNA can be an efficient estimator of population divergence, but not effective population size, which requires multiple loci for accurate estimate [2,8].

(a). Affinities among mitochondrial DNA, autosomal and Z-linked markers in summary statistics

In general, genetic diversity depends on effective population size and substitution rate, whereas genetic distance depends on divergence time and substitution rate. Our tests control for effective population size and divergence time because we compared the same species' mtDNA and nuDNA, which share the same species history. Therefore, the departures from a perfect correlation (i.e. r² = 1) might reflect variations in the relative substitution rates among loci or stochasticity in the coalescence process. This argument is consistent with the observation of reduced r² values in regression analyses based on full datasets compared with those ones based on each intron separately or on average values for each species' introns. That is, inter-intron variations in substitution rates and coalescence times have greater influence on the former than the latter. Nevertheless, the mtDNA D_A and F_ST values were consistently and positively correlated with those of nuDNA regardless of the intron datasets used. This suggests that estimates of relative population divergent level based on mtDNA can be considered statistically accurate, although the correlation between mtDNA and nuDNA is not close to perfect.

Furthermore, the estimates of F_ST from mtDNA are generally larger than those from autosomal and Z-linked introns in this study (figure 3), almost certainly because of the smaller N_e and thus faster coalescent time of mtDNA [47]. Zink & Barrowclough [2] compared mtDNA and nuDNA estimates of F_ST to ascertain how well they corresponded to four potential outcomes (figure 3) and found that only ca 5% suggested greater nuDNA differentiation than mtDNA (category C). In general, our analyses of F_ST values suggested that mtDNA was a leading indicator of population differentiation relative to nuclear loci in spite of ca 2% of data points (5 out of 225 for mtDNA versus autosomal introns; 2 out of 93 for mtDNA versus Z-linked introns) occurring in category C (figure 3). In addition, comparisons between mtDNA and autosomal introns have more instances of higher mtDNA differentiation (category B) than those between mtDNA and Z-linked introns (89 out of 225 versus 25 out of 93, p = 0.02, Fisher's exact test). Notably, 25% (6 out of 24) of the IMa-estimated divergence times fall between 1 N_e and 3 N_e and 42% (10 out of 24) between 1 N_e and 4 N_e (electronic supplementary material, table S13), consistent with the proportions of data points occurring in category B for the comparisons of mtDNA versus Z-linked introns and mtDNA versus autosomal introns, respectively. Of course, there is variation around the theoretical expectations; nevertheless, these results suggest that lineage separations which fall between 1 N_e and 3 N_e or 4 N_e will be difficult for most nuclear loci to recover (but see [15]).

Figure 3.

Comparisons of divergence (a) between mitochondrial and autosomal estimates and (b) between mitochondrial and Z-linked estimates of F_ST values. In general, the area below the diagonal indicates cases where mtDNA is more sensitive than nuDNA to population structure, whereas the area above the diagonal indicates cases where nuDNA is more sensitive than mtDNA. Four categories of phylogeographic results delimited arbitrarily at an F_ST value of 0.2 [2] are shown with background shading. Categories A and D indicate consistent phylogeographic patterns between mtDNA and nuDNA, where both show absence or presence of population structure, respectively; category B suggests consistent results with mtDNA as relatively leading indicators because of the small effective population size and fast coalescent times; category C suggests conflicting results between mtDNA and nuDNA, where the latter shows population structure but the former does not.

By contrast, intra-population genetic diversity can also be strongly influenced by natural selection and changes in population size [11,48,49]. The neutrality tests reveal no evidence of selection on the mtDNA and nuDNA in this study. However, genes with different N_e may respond to population size dynamics with differing degrees of change [16,50]. For example, genes with smaller N_e (mtDNA in this case) may respond to the same population bottleneck with more dramatic change in genetic variation than genes with larger N_e (nuDNA in this case) [41,50]. Because the 30 temperate zone avian species are likely to have experienced frequent population fluctuations over multiple glacial–interglacial cycles [51,52], the discrepancies between their mtDNA and nuDNA population genetic variation could be expected. This could explain the lack of significant correlation between mtDNA and autosomal introns in π_ave and D values (the latter are confounded by intra-population genetic diversity). It is also possible that the small sample sizes for mtDNA and nuDNA result in inaccurate estimates of genetic diversity [53], causing no correlation between mtDNA and autosomal π_ave values, or reducing the ability to detect small but significant relationships. Nevertheless, the potential bias should lead to conservative, not exaggerated, effects on mtDNA's performance.

We also found that Z-linked introns are positively correlated with mtDNA in π_ave and D values even though less significant than in D_A and F_ST values. These results suggest that mtDNA and Z-linked introns have more similar coalescent times and respond to population size change in a more similar way than do mtDNA and autosomal introns. The phenomenon is expected given that effective size difference between mtDNA and Z-linked introns (1 : 3) is smaller than those between mtDNA and autosomal introns (1 : 4).

When comparing the values of individual Z-linked introns (n = 5) against all autosomal introns, we find positive correlations in all cases of D_A and F_ST values but only two cases for π_ave values (electronic supplementary material, figure S2). The results suggest that mtDNA can estimate population divergence as well as, if not better than, any single Z-linked intron [44,54]. In addition, Z-linked introns seem slightly better than mtDNA in estimating population diversity that, however, cannot be estimated accurately by most (or any) single locus, further confirmed by the one-to-one gene comparisons (electronic supplementary material, figure S4).

(b). Association between mitochondrial DNA phylogeographic patterns and nuclear DNA coalescence history

The usefulness of mtDNA as a genetic marker is further confirmed when we compare the summary statistics of mtDNA with IMa coalescence model parameters based on all nuclear introns. Both of the D and D_A values of mtDNA are positively correlated with the IMa divergence times, although the π values of mtDNA are not correlated with the IMa effective population sizes. Interestingly, there is significantly negative correlation between the mtDNA F_ST values and the IMa migration rates.

The power of mtDNA in inferring population divergence and gene flow is also illustrated by the comparison between mtDNA haplotype networks and IMa divergence times and migration rates. The unstructured networks have the smallest average IMa divergence time and the geographically structured networks have the smallest average IMa migration rate, as one might expect. The results indicate that the mtDNA-based phylogeographic patterns can reflect the nuDNA-based phylogeographic history.

(c). Little effect of effective population size on genetic divergence

The nearly neutral theory predicts that slightly deleterious mutations are more likely to drift to fixation than to be removed by purifying selection in small populations, leading to higher substitution rates [37]. Smith & Klicka [14] show that estimated divergence times among populations of a songbird (Cardinalis cardinalis) are significantly longer when inferred via mtDNA comparisons than by multiple nuclear introns in small populations, but not in large populations.

By contrast, there is no evidence showing that mtDNA genetic distances between populations are impacted by population size in this study. The results suggest that the effect of relaxed purifying selection is not large enough to cause significant change in mtDNA substitution rate even comparing species with fairly different population sizes. However, it is also possible that the divergence times between populations are too recent to reflect the substitution rate changes in mtDNA. If so, studies focusing on taxonomic levels above species might reveal a stronger effect of effective population sizes on estimates of genetic distances between taxa [55]. Nevertheless, the results suggest that such effect may not be a major concern for phylogeography, which focuses on the species level or below.

5. Conclusion

We do not doubt that next-generation sequencing approaches will revolutionize phylogeography. However, our goal here is to provide a retrospective analysis on the mtDNA era of phylogeography and estimate the credibility of nearly three decades of mtDNA studies. Many studies have found that the results based on multiple nuclear genes or genome-wide data are often consistent with those based on one single mtDNA gene data in phylogeographic patterns or species delimitation ([8,11,44–46,56]; but see [5,41,42]). Here, we provide empirical data confirming the assertion that mtDNA is a useful marker for detecting recent population differentiation because of its fast coalescent time at least in avian taxa (figure 3). Our results suggest that numerous studies which have based their inferences of population divergence patterns solely on mtDNA [57–60] are probably reliable. However, our experimental design based on two-population comparison is relatively simple. For more complex scenarios such as multiple populations with mtDNA introgression, the chance of mtDNA–nuDNA conflict is expected to increase and datasets combining mtDNA and nuDNA will be required for accurate inferences of complex history [42]. How complex scenarios bias mtDNA-based inference warrants further study.

Data accessibility

DNA sequences deposited in GenBank with accessions: KU366711-KU366810; KU551341-KU551858; KU716149-KU719477; KU857048-KU859397.

Authors' contributions

All authors participated in research design and wrote the manuscript. C.M.H. and S.V.D. carried out the molecular laboratory work. S.V.D. collected samples and requested tissue samples from museums. C.M.H. carried out data analysis. R.M.Z. coordinated the study.

Competing interests

We have no competing interests.

Funding

Support came from NSF (DEB 0919494) to R.M.Z. and from Fundos FEDER funds through the COMPETE program and the Fundação para a Ciência e a Tecnologia, Portugal (FCOMP-01-0124-FEDER- 008941; FCT reference PTDC/BIA- BEC/103435/2008) to S.V.D.