Background

Pleistocene climatic oscillations have played a major role in structuring current biodiversity patterns and shaping the distribution of species and their genetic diversity both at the regional and global scale [1, 2]. In recent decades, phylogeographic studies have been used to assess the genetic consequences of Pleistocene ice ages on various organisms, highlighting the dynamic nature of species ranges and the role of microevolutionary processes in determining the extent and structure of intraspecific diversity [1, 3, 4].

In Europe, the genetic structure of temperate organisms has been mainly explained by cycles of contraction toward glacial refugia located in southern peninsulas and post-glacial recolonization of northern regions, tracking the expansion of suitable habitats [5, 6] Accordingly, southern peninsular refugia are thought to have played a major role in the long-term maintenance of genetic diversity and differentiation, and relatively few patterns and routes of colonization have been described as paradigms for the post-glacial arrival of species in Northern Europe [6–8].

However, in recent years, multiple evidence from paleoclimatic, palaeontological and phylogeographic studies have identified the existence of glacial refugia in northern regions of Europe, or “northern refugia”, and have highlighted the prominent contribution of these refugia to the present-day genetic diversity of many organisms [9–12]. Mounting examples from plants, insects, mammals, amphibians, and reptiles show that in many temperate species the southern refugia model is insufficient to explain the observed genetic patterns, which would be better explained by a scenario of ice-age survival in a combination of Mediterranean and extra-Mediterranean refugia [13–17]. This changing view with respect to geographical location of refugia has also had direct implications for the inferred scenarios of recolonization for many widespread species from the Western Palaearctic. This is true even in emblematic species, such as the brown bear and the common beech, the phylogeographic patterns of which were previously upheld as paradigms of post-glacial recolonization routes (exclusively) from southern refugia [6, 7]. In these species, the existence of additional extra-Mediterranean refugia has recently become evident through the analysis of fossil DNA and palaeobotanical data, respectively. These evidence suggests that range shifts and expansions in these species could have been much smaller than previously thought [18–21].

Detailed phylogeographic data from widespread Western Palaearctic species are particularly valuable for evaluating the plausibility of a scenario of ice-age survival in Mediterranean and extra-Mediterranean refugia. The common Wall lizard, Podarcis muralis (Laurenti, 1768) is a Western Palaearctic species which provides a suitable case-study for this purpose. Podarcis muralis is a locally abundant lacertid lizard found in a variety of environments across a wide altitudinal range (from sea level to over 2000 m of altitude). It exhibits considerable variation in colour patterns, biometry and pholidosis, which has led to the description of several subspecies [22, 23]. The distribution of P. muralis is unusual relative to that of its congeners and, together with other characteristics, makes this species a useful model for evaluating the relative contribution of southern versus extra-Mediterranean refugia in shaping the current distribution of species and their genetic diversity. Firstly, it covers a wide range across the Western Palaearctic– from North-Eastern Anatolia and the Black Sea coast through all the Mediterranean peninsulas – Iberia, Italy, the Balkans– as well as across extra-Mediterranean regions of France, Germany, Slovenia, Austria and the Central Balkans (Figure 1). Secondly, it is abundant in both Mediterranean and continental climatic conditions and its distribution does not seem to be strongly limited by the presence of other lizards. Throughout the southern peninsulas P. muralis occurs in sympatry with more localized Podarcis species, such as members of the P. hispanica group in the Iberian Peninsula, P. sicula in the Italian Peninsula, and P. erhardii in the Balkans as well as with members of the genera Lacerta, Iberolacerta, and Timon across its whole range. Also, it is a highly successful invasive species in North-Western Europe where around 150 non-native P. muralis populations have been identified [24]. Thus, Mediterranean and extra-Mediterranean areas seem generally suitable for the survival and colonization of this species. Moreover, although little is presently known about the evolutionary history of P. muralis, either a scenario of recent colonization of its northern range from a southern primary range or a long-term history of the species in some of these northern areas can be invoked to explain the observed intraspecific variation. Harris and Arnold [25] hypothesized that P. muralis recently colonized most of its distribution from the Italian Peninsula, based on the observation that it exhibits the greatest morphological and allozyme diversity in Italy [26]. On the other hand, recent genetic assessment identified divergent genetic lineages - likely of Pleistocene origin - in Italy but also in France and the Balkans, disproving a colonization of these two latter areas from Italy [24, 27, 28]. Nevertheless, these studies have been limited to small parts of the range and analyzed only mitochondrial DNA, and particularly partial cytochrome b sequences, thus preventing a full appreciation of the origin of these lineages and of the overall evolutionary history and biogeography of the species.

Figure 1
figure 1

Sampling localities, phylogenetic relationships, and distribution of mitochondrial haplotypes and haploclades. Maximum Likelihood (ML) trees depicting the phylogenetic relationships among 75 combined mitochondrial sequences (cytb and nd4) of Podarcis muralis from 52 localities (A) and among 185 cytb sequences of Podarcis muralis obtained in this study (N=82) and from GenBank, (N=103) (see Additional file 1) (B). ML bootstrap values over 1000 replicates are reported in correspondence to the nodes; black stars indicate BA posterior probabilities = 1.00; Podarcis liolepis was used as an outgroup. In the combined cytb/nd4 ML tree specimen codes are reported (see Table 1 for details) and main mitochondrial clades numbered from 1 to 17; in the cytb ML tree tip nodes are collapsed and main mitochondrial clades are numbered from 1 to 23 according to the combined cytb/nd4 ML tree and named according to previous studies. Samples used in this study and geographic distribution of mitochondrial clades are reported in the map: big circles show the 53 localities sampled in this study and triangles show the geographic origin for the cytb sequences obtained from GenBank, coloured according to mitochondrial clades defined by the ML trees (black triangles indicate admixed populations for which the frequency of haplotypes belonging to each clades is shown by the pie diagrams; grey circles indicate samples for which only nuclear sequences were available).

Full size image

In this study, we analysed DNA sequences data from 86 individuals from 52 localities across the species range, including extensive sampling in the Mediterranean peninsulas where native diversity is expected to be highest. We used two partial mitochondrial genes (cytb and nd4), and three partial nuclear genes (acm4, mc1r, and pdc), all of which have been successfully used for assessing variation within other species of lacertids (e.g. [29, 30]). Furthermore, by comparing 103 published cytochrome b sequences from previous studies with our assessment, we placed these in a more robust phylogenetic framework, and further used them to confirm that all known diversity was included in this study. This comprehensive data set enables us to evaluate whether the patterns of genetic variation observed in this species reflect a recent expansion from a single or a few Pleistocene southern glacial refugia or, alternatively, from a combination of Mediterranean and extra-Mediterranean refugia. Under the first scenario we expect to identify endemic lineages in one or more southern refugia, to observe only southern-derived haplotypes in northern areas, and an overall pattern of ‘southern richness vs northern purity’ of genetic diversity [1, 2, 6]. Alternatively, if P. muralis also survived in northern glacial refugia, we predict an occurrence of lineages endemic to northern regions, with an ancient (Pleistocene) divergence from southern lineages, and an absence of a northern purity pattern of genetic diversity. The main goal of our range-wide multilocus approach is, therefore, to identify the biogeographic and evolutionary processes shaping the genetic diversity of P. muralis and to discuss, in the framework of European phylogeography, how this species came to have such an unusual biogeographic pattern.

Methods

Data collection

We sampled a total of 86 individuals from 52 localities spanning the range of Podarcis muralis with particular attention to the Iberian, Italian, and Balkan peninsulas. Three specimens of two Iberian Podarcis, P. liolepis (Boulenger, 1905) and P. bocagei (Seoane, 1884), and the Italian wall lizard P. sicula (Rafinesque-Schmaltz, 1810), were sampled and designated as outgroups. We used multiple outgroups because the phylogenetic position of P. muralis within the genus Podarcis is not well resolved, yet this species seems more closely related to Iberian Podarcis or P. sicula[31, 32]. Additionally, partial cytb sequences for 103 individuals were obtained from GenBank and included in phylogenetic analyses. Specimens included in molecular analyses, together with collection locality details, are reported in Table 1 and Figure 1. Accession numbers for sequences retrieved from GenBank are reported in Additional file 1.

Table 1 Geographical location and codes for the studied specimens of Podarcis muralis and the outgroup species P. liolepis
Full size table

Total genomic DNA was extracted from alcohol-preserved tail muscle collected from live specimens following standard high-salt protocols [33]. Portions of five genes were amplified by polymerase chain reaction: two mitochondrial genes, cytochrome b (cytb) and NADH dehydrogenase subunit 4 with flanking tRNAs (nd4), and three nuclear genes, Melanocortin receptor 1 (mc1r), acetylcholinergic receptor M4 (acm4), and Phosducin (pdc). The primers used for amplification and respective references are reported in Additional file 2. For pdc we used primers newly developed for this study. Amplifications were carried out in 25 μL volumes, containing 1X PCR buffer (50 mm Tris–HCl, 50 mm NaCl, pH 8.5); 3 mM MgCl2; 0.6 mM each dNTP, 2U of GoTaq DNA polymerase (Promega), 0.4 μM each primer and approximately 50 ng of genomic DNA. Amplification conditions consisted of a preliminary denaturation step at 92°C for 3 minutes, followed by 16 touchdown cycles with 30 seconds at 92°C, annealing temperature decreasing 0.5°C per cycle from 60°C to 52°C (30 seconds) and extension for 1 minute at 72°C. 20 more cycles similar to these but with annealing temperatures stable at 52°C followed. A final extension was carried out at 72°C for 15 minutes.

Data analysis

Electropherograms were checked and consensus sequences were aligned in GENEIOUS 6.0 (http://www.geneious.com) using the Geneious Alignment algorithm. We discarded the hypothesis of pseudogenes occurring in our mitochondrial sequence dataset by confirming: (i) the absence of stop codons in protein-coding fragments, (ii) the consistency of the base composition pattern with a mitochondrial origin (less than 5% of guanines in the third position, see [34]), and (iii) the overall similarity with the reference mitochondrial genome of P. muralis (Genbank accession number FJ460597). Haplotype phase of nuclear genes was determined using PHASE version 2.1.1. [35]. Estimations were carried out under a model with recombination (−MR0 option), with the initial 1000 iterations discarded as burn-in, 1 as thinning interval and 1000 post-burnin iterations. Three independent runs were conducted for each gene. After monitoring the goodness of fit for each run according to the program’s manual, we accepted haplotype reconstructions which both i) yielded the same result in each of the three runs, and ii) had associated phase posterior probabilities of at least 0.75 in the average of the three runs. For each nuclear gene, the possible occurrence of recombination events was assessed using the Pairwise Homoplasy Index (PHI) test [36] implemented in SPLITSTREE 4.11 [37].

In order to assess patterns of genetic diversity and evaluate whether southern peninsulas harboured most of the diversity of P. muralis, we estimated the overall genetic diversity of the species and its distribution across populations and regions. For each gene we computed the following summary statistics of genetic diversity: number of segregating sites S, nucleotide diversity π, number of haplotypes H, and haplotype diversity Hd, both overall and for specific groups or clades defined within the species (see Results and Discussion) with DNASP 5 [38]. Comparisons of diversity measures can be greatly affected by different sample sizes among groups. In order to account for this, we followed a resampling approach: given the lowest sample size (except 0 or 1) amongst the various groups being compared (s), for each group with sample size > s we resampled 100 sets of s sequences, calculated diversity measures for each of these 100 sets and took the average for each diversity measure. This procedure allows a straightforward comparison of diversity measures derived from groups or clades with distinct sample sizes. Resampling was performed with the aid of a series of scripts written in Python 2.7.1 (available from the authors upon request) and taking advantage of DNASP “batch mode” calculations. Average uncorrected genetic distances between mitochondrial clades were assessed using MEGA 5 [39]. Furtehrmore, to assess genetic differentiation between clades we calculated FST (based on p-distance) and its significance by performing 1000 permutations in ARLEQUIN 3.5.1.2 [40].

Phylogenetic relationships among mitochondrial haplotypes were inferred using Maximum Likelihood (ML) and Bayesian (BA) methods. We tested three outgroups, P. liolepis, P. bocagei, and P. sicula (both combined and separately) through preliminary ML and BA tree searches. The outgroup effect on the ML and BA topologies was limited and differences between trees calculated using different outgroups were limited to unsupported nodes (not shown). Thus, we present and discuss phylogenetic analyses using P. liolepis as an outgroup (Genbank Accession numbers for P. bocagei cytb/nd4: DQ081139/DQ081153; for P. sicula cytb/nd4: KF372034/KF372035; for P. liolepis see Table 1). The ML search and the model selection were carried out in TREEFINDER version October 2008 [41]. We selected the best partitioned model of nucleotide substitution under the corrected Akaike’s Information Criterion (AICc) by comparing the AICc scores of seven different partition schemes (under a fixed ML topology): (i) by gene fragments (cytb/nd4+tRNAs), (ii) by genes (cytb/nd4/tRNAs); (iii) by coding regions (cytb+nd4/tRNAs), (iv-v) by codon position alone (two schemes:1st/2nd/3rd and 1st+2nd/3rd), and (vi-vii) by codon position, in combination with gene partitions (for a total of seven or five partitions, depending on the codon partitioning scheme). The optimal model had seven partitions, three codon partitions for each coding gene and one partition for tRNAs (cytb: HKY for the 1st, the 2nd, and the 3rd codon positions; nd4: TN+ G for the the 1st, and HKY+G for the 2nd and the 3rd codon positions; tRNAs: HKY+G, each model with frequency and rate heterogeneity parameters optimisation). We performed a global tree search using 100 random start trees generated through equidistant random walks of random nearest-neighbour-interchanges (NNI) starting from the centre tree obtained by a simple ML search. Support for the nodes (BP) was evaluated with 1000 bootstrap replicates. Additionally, in order to evaluate the total mitochondrial variation known for this species, we carried out a ML analysis following the same procedure as above (the best model in this case was the HKY with a single partition), using a dataset including both the cytb sequences generated in this study (N=82) and those available in GenBank for native populations of P. muralis (N=103). This dataset provides even greater coverage of the whole species range (see Figure 1). The obtained phylogenetic trees were visualised and edited using FigTree v.1.3.1 (available at http://tree.bio.ed.ac.uk/software/figtree/). The geographic pattern of haplogroup distribution and the relationships between haplotypes help to disentangle the two alternative scenarios predicted by the southern refugia model or by a model of Mediterranean and extra-Mediterranean refugia. For this purpose, we mapped the distribution of haplogroups identified by the phylogenetic trees across our range-wide sampling.

Bayesian analyses were carried out using the BEAST v.1.7.4 package [42] for two purposes: i) to estimate a phylogeny of the combined mitochondrial DNA dataset based on Bayesian inference; ii) to obtain estimates of the mtDNA time to the most recent common ancestor (TMRCA) of P. muralis and groups detected within the species. The choice of an appropriate tree prior for this dataset is somewhat problematic: on one hand, the bulk of our data is intraspecific, hence a “speciation” prior such as the Yule process may not be adequate. On the other hand, because the presence of an outgroup is mandatory and due to the high levels of geographic substructure detected in our focal species (see Results), our data are not totally amenable to analyses under a basic coalescent process. Therefore, we performed these analyses under both a Yule process and a coalescent process with constant size, to evaluate the effect of prior selection on our estimates of phylogeny and TMRCAs. To avoid overparametrization, we implemented a model with a simpler partition scheme, defined by gene fragments (two-partitions), and selected under AICc in TREEFINDER. The best evolutionary model for our dataset assumed the HKY+G and the TIM2+G+I substitution models for the cytb and nd4 fragments, respectively. The TIM2 model was manually specified within the BEAUti input file [42]. To infer mtDNA TMRCAs, we assumed a relaxed molecular clock for both cytb and nd4, applying an uncorrelated lognormal model [43]. We used the available rates proposed for Podarcis for the longer fragment, nd4, while providing a large uninformative prior for the cytb substitution rate (parameter ucld.mean). In the case of nd4 we assumed that differences accumulate at a mean rate of 2.26% per million years, obtained as the average between divergence rates proposed for Podarcis for this fragment (ranging from 1.74% to 2.78% per million years; [44]). In order to incorporate this uncertainty into TMRCA estimates, we defined a normal prior distribution on the mutation rate with mean 1.13%/MY and with 95% of the probability distribution encompassed between the two aforementioned limits (0.87 and 1.39%, respectively). For each tree prior, BEAST was run twice, with 100 million iterations per run, sampling every 10000 steps. TRACER v. 1.5 (available at http://beast.bio.ed.ac.uk/Tracer) was used to visualize the results and assess if the effective sample size of estimated parameters was satisfactory. We discarded the initial 10% of sampled trees as burn-in. Upon confirmation of convergence, the two outputs for each tree prior were combined and subsequent data were retrieved from the combined data set.

For the following analyses, we considered as clades groups of haplotypes which are reciprocally monophyletic at the mitochondrial loci with the ML and BA tree reconstructions; displaying geographical contiguity; and divergent by more than 1% with its closer relative in line with previous studies ([24, 27, 28]; see also [45]). While we acknowledge that such criteria could be considered arbitrary, to further test that these groups correspond to evolutionary significant lineages, ESU sensu[46], we confirmed they show significant divergence of allele frequencies at nuclear loci as assessed by FST statistics (see above).

To assess how genetic variance at the combined cytb/nd4 sequences was hierarchically distributed among and within mitochondrial clades, we performed a spatial analysis of molecular variance (SAMOVA) with SAMOVA 1.0 [47]. We pooled samples from neighbouring localities in accordance with the mitochondrial clades previously identified. Geographic coordinates for each clade were calculated as the geographic centroid of member localities. Clades represented by less than two individuals were not included in the analysis. SAMOVA was run for 10,000 iterations from each of 100 random initial conditions, and testing the predefined number of groups (K) from 2 to 14.

Phylogenetic relationships among nuclear haplotypes were inferred by median joining networks [48] using the software NETWORK 4.6.1.0 (available at http://www.fluxus-engineering.com/sharenet.htm).

Finally, we specifically tested the hypothesis of a recent expansion of P. muralis from southern glacial refugia against the alternative scenario of Pleistocene survival of the species in both southern and northern refugia. The first hypothesis implies that northern localities belong to the same genetic pool as the southern region from which they originated. As such, sequence divergence between northern and southern localities will be, in most cases, significantly lower than the divergence observed between clades and possibly similar to intraclade divergence. By contrast, the latter hypothesis implies that northern regions have hosted surviving populations which have evolved independently from southern ones. Thus, we expect the mean sequence divergence between northern and southern localities to be similar to mean divergence among clades. This rationale provides a statistical framework to explicitly compare the two competing hypotheses. A requisite of this test is that mean between-clade sequence divergence is significantly higher than within-clade divergence. Therefore, as a first step, we performed a permutation test to evaluate if between-clade p-distances (D bc ) were significantly higher than within-clade distances (D wc ). For this purpose, we randomly shuffled individuals among clades 1000 times in order to obtain the empirical distribution of the difference D bc – D wc . The corresponding p-value was then obtained as the percentage of times for which the observed D bc – D wc was smaller than random. This test was performed for the combined mitochondrial DNA data set, as well as for the three nuclear loci in separate, always using the units defined on the basis of the mtDNA phylogeny. In continuation, for each northern locality and for each locus, we evaluated if the sequence divergence from southern localities fell within the “between-clade” or the “within-clade” level of differentiation. We classified localities as “northern” or “southern” according to biogeographic barriers separating central Europe from major glacial refugia in southern peninsulas as defined by previous literature [1, 4, 5, 7, 15]: localities north or east to the Pyrenees (locality 1), north to the Appennine chain (localities 25–26), west or east to the Alps (localities 20–23 and 35–36, respectively), and in the central Balkans (localities 37–39) were therefore considered as “northern”. For each northern locality we performed two different tests: first, we examined its mean p-distance from all southern localities (localities 2–16, 18, 27–34, and 40–46). Second, to account for the possibility that northern localities may have been preferably colonized from adjacent refugia, we also examined the mean p-distance of each northern locality from contiguous southern regions only (locality 1 vs localities 2–16, and 18; localities 20–26 vs localities 27–34; localities 37–39 vs. localities 40–46; localities 35–36 were compared either with localities 27–34 or with localities 40–46, given their intermediate geographic location). These comparisons were performed according to the following permutation procedure: individuals were shuffled among localities at random across 1000 permutation cycles. In each permutation, we calculated the mean p-distance between each northern locality and the set of southern localities being compared (D i , where i is the code of the northern locality), the mean between-clade distance (D bc ) and the mean within-clade distance (D wc ). For each northern locality i, we then assessed whether the difference between D bc and D i (D bc – D i ) was equal or higher than the distance we observed in the real data. The same was done for the difference between D i and D wc (D i - D wc ). The proportion of permutations verifying each of these conditions was used as a p-value to assess if the observed differences were significant or could be due to chance alone.

Results

We obtained 82 cytb mitochondrial sequences of 411 base pairs (bp), 79 nd4 mitochondrial sequences of 873 bp (75 individuals sequenced for both fragments), 156 sequences for the nuclear gene fragment mc1r, and 154 for acm4 and pdc with 688bp, 423 bp, and 342 bp, respectively. GenBank accession numbers for the sequences generated in this study are reported in Table 1. Multiple sequence alignment did not require gap positions in any of the studied genes. The phi test did not find statistically significant evidence for recombination within nuclear gene fragments (p > 0.05). Sample size and localities, and clade assignment are reported in Table 2.

Table 2 Summary statistics of genetic diversity for each gene, clade, and region
Full size table

The phylogenetic relationships inferred from the ML tree based on the combined mitochondrial sequences showed 17 well supported clades (BP≥90) with a geographic coherence (Figure 1A). The Bayesian tree showed an identical topology at supported nodes (not shown) with the same 17 clades supported with a posterior probability of 1.00 (Figure 1A), independently of the choice of tree prior. Clade 1 grouped haplotypes from Northwest France and Southeast Pyrenees (Spain). Haplotypes included in clades 2 and 3 occurred in populations from Northern and Central Spain, respectively. Clade 6 is restricted to Provence, France. Five different clades occurred in Italy: clade 5 in Northern Italy beneath the Alpine arch; clades 9 and 10 in central Italy; and clades 7 and 8 in the Calabrian Peninsula. Slovenian populations clustered in clade 11 and samples from Serbia in clade 4. In the Balkan Peninsula, clades 12, 13, and 14 were distributed in Thessaly, Central Greece, and Peloponnese regions, respectively. Clade 16 is restricted to Samothraki Island and clades 15 and 17 are distributed in Turkish Thrace and North-Western Anatolia, respectively. Clades 12 and 16 were formed by single haplotypes, and as such they cannot be strictly considered as clades. However, we use the term ‘clade’ for simplicity, given their divergence from other clades, and, in the case of clade 12, also because this is indeed formed by multiple haplotypes in the analyses including Genbank sequences (see below and Figure 1B). The average uncorrected genetic distance between mitochondrial clades was 4.5 and 3.1% for cytb and nd4 respectively (ranging from 1 to 7.4% for cytb and 0.9 to 4.7% for nd4) and both loci show significant FST values (0.86 and 0.83 for cytb and nd4 respectively; P=0.001). Clades 15, 16, and 17 showed the lowest pairwise distance values at both cytb (1%) and nd4 (0.9%), while the minimum value of pairwise genetic distances across other clades was 1.8% for both genes. Phylogenetic relationships between Iberian clades West of the Pyrenees (clades 2 and 3), between southern Italian clades (clades 7 and 8), between Slovenian and Balkan clades (clades 11 and 12), and among clades from Turkey and Samothraki (clades 15–17) were well supported in both ML and BA trees, while the relationships between clades 1–6 was supported in BA but not in ML analyses (BP<60). The relationships between the remaining clades were not resolved.

The ML analysis combining the 84 cytb sequences generated in this study and 103 sequences published from previous studies showed 23 well supported clades (Figure 1B). These included the 17 clades identified based on the combined cytb/nd4 ML tree and are numbered according to the combined ML tree (Figure 1A) and named according to previous studies [22, 26, 27]. Previous studies had identified 12 of these mitochondrial mtDNA lineages – eight across Italy (“1Aa”, “1Ac”, “1Ad”, “1Ae”, “1B”, “2a”, “2b”, and “2c”; [26]), one from the island of Elba [27], and one from each of Western France, Eastern France, and the Balkans [23]. We separated the clades “1A” and “1B” listed in [26] into their constituent sub-clades, since clade “1A” was not supported by phylogenetic analyses and clade “1B” from southern Italy is composed of distinct units (from Calabria, Apulia and Campania), which are as differentiated as the other clades. In our assessment, ten previously undetected clades were identified (clades 2,3, 8, and 11–17 in Figure 1A). Additionally, a new lineage was identified in East Macedonia (clade 18) that had not been previously described as a group despite being comprised of sequences retrieved from GenBank.

MtDNA TMRCA estimates within P. muralis were roughly similar independently of the choice of tree prior in BEAST analyses (Additional file 3). The TMRCA of P. muralis was estimated at 2.47-2.67 million years ago, thus coinciding with the Plio-Pleistocene boundary. The TMRCA estimates of the various clades bound the period from Early Pleistocene to Late Pleistocene.

SAMOVA analyses showed a lack of genetic structure above the clade level (Additional file 4). No clade grouping explained the genetic variation observed in P. muralis significantly better than other grouping options, as indicated by the uniform increase of FCT values toward higher values of K.

The phylogenetic networks depicting the relationships between acm4, mc1r, and pdc haplotypes are shown in Figure 2. The acm4 and mc1r networks showed a pattern of low divergence and a lack of apparent phylogeographic structure, although both show significant FST values (0.37 and 0.24 respectively; P=0.001) between mtDNA-defined clades. In both loci the most common haplotypes were found in most localities (with the exception of localities 47 and 53 for acm4, while locality 46 is not represented in the dataset of these two genes, see Table 1 and 2) and derived haplotypes divergent by one or two mutational steps had a more restricted distribution. On the other hand, evidence of geographic association of nuclear haplotypes was more prominent for the pdc locus (Figure 2), leading to a higher and also significant FST value (FST = 0.56; P=0.001). The structure of the network showed two most frequent haplotypes, each exhibiting derived variants: the haplogroup placed in the left side of the network was restricted to the Balkans, Turkey, Slovenia, and Provence (France), while the haplogroup on the right side included all the haplotypes from Italy. Haplotypes sampled in the Iberian Peninsula and Eastern France belonged to both haplogroups.

Figure 2
figure 2

Phylogenetic relationships among nuclear haplotypes. Haplotype median joining networks based on sequences of the nuclear genes acm4, mc1r, and pdc. Haplotypes are represented by circles with area proportional to their frequency and coloured according to mitochondrial clades as defined by the Maximum Likelihood tree (see Figure 1).

Full size image

Estimates of genetic diversity S, π, H, and Hd for each gene, clade, and geographic region are given in Table 2. Nucleotide and haplotype diversity of mitochondrial genes was higher in Northern, Central, and Southern Italy (clades 5, 9, and 8, respectively) and in clade 14 from Southern mainland Greece and the Peloponnese. At the nuclear gene level, an overall pattern of low genetic diversity was found. Higher diversity values were observed in populations from Northern Italy (clade 5), whereas no clear trend was apparent across putative southern refugial areas.

Permutation tests showed that between-clade p-distance is significantly higher than within-clade distances for all loci (D bc /D wc for mitochondrial DNA: 0.0347/0.0052; for acm4: 0.0051/0.0022; for mc1r: 0.0029/0.0013; for pdc: 0.0097/0.0043; P=0.001; see Additional file 5). This allowed us to perform tests aiming at evaluating whether the genetic distance between northern and southern localities is consistent with the distance typically observed either between or within clades. Results from these analyses showed that mean pairwise distances between northern and southern localities are significantly different from within-clade differentiation, but not significantly different from between-clade distance (except locality 1), independently of the locus used for these comparisons (for more details see Additional file 5). Similar results were obtained when comparing northern localities to the sub-set of contiguous southern locations, although in this case this pattern is strong at mitochondrial but less evident at nuclear loci. These results support the hypothesis that populations of P. muralis from northern regions correspond to distinct evolutionary units from those occurring in southern peninsulas, and therefore that they have likely originated in situ rather than as a result of postglacial expansions of southern populations.

Discussion

The common wall lizard Podarcis muralis exhibited a complex phylogeographic pattern with multiple divergent mtDNA clades across its range. In total 23 allo- parapatric clades were identified, most of them with a restricted distribution in the Iberian, Italian, and Balkan peninsulas, as well as in Western and Eastern France, Slovenia and Austria, Central Balkans, and North-Western Anatolia. The detection of previously unidentified lineages was evidently due to the extended sampling scheme of this study as compared to earlier assessments, which were either focussed on a regional scale [27, 28], or did not sample all clades endemic to the Iberian, Balkan or Anatolian peninsulas [24]. Although the definition of clades is somewhat arbitrary, the lack of resolution of relationships between monophyletic haplogroups, their high divergence, their geographic coherence, and results from the SAMOVA analysis (Figure 1A,B; Additional file 4) indicate that a grouping into fewer lineages would not be practical and does not describe accurately the mitochondrial DNA diversity exhibited by P. muralis.

The observed phylogeographic structure of P. muralis does not match the current subspecific division of this species [22, 23], as already reported by [27] and [28] for the 11 subspecies occurring in Italy. Regarding the subspecies found outside Italy, many divergent clades fall into P. m. brogniardii (clades 1–3, 5, and 6) and P. m. albanica (clades 12–14, and 18). Though a taxonomic revision is beyond the scope of this study, our results indicate that the current intraspecific taxonomy of P. muralis reflects local phenotypic varieties rather than distinct evolutionary units.

While mitochondrial data showed high variability and a clear geographic structure of genetic diversity, nuclear data showed both relatively low genetic variability and shallow divergence among populations. This finding is in agreement with allozyme data [49]; see also [26] from which low levels of polymorphism and genetic differentiation were identified (number of alleles per locus, A = 1.07; proportion of polymorphic loci, P = 0.07; observed heterozygosity, Ho = 0.028; genetic distance DNEI = 0.036; average values from [49]) among 23 populations of P. muralis attributable to eight mitochondrial lineages described in our study (clades 1, 2, 3, 5, 9, 14, 22, and 23). Despite low differentiation in nuclear markers, FST values between mitochondrial clades are significant for all three nuclear markers and the mean between-clade pairwise distance is significantly different from that observed within clades, indicating that they are significant evolutionary lineages sensu [46]. Furthermore, even though overall nuclear data showed low genetic variation, a weak geographic association of nuclear haplotypes is apparent in sequence data. At the acm4 and mc1r loci, besides a common widespread haplotype, some closely related derived haplotypes, relatively frequent, were found in the Balkans, Turkey, Slovenia, and Eastern France. This geographic association was more evident at the pdc locus, where we observed two main haplotype groups: one including all haplotypes from Italy, and the other comprising haplotypes from the Balkans, Turkey, Slovenia, and Eastern France. Both haplogroups also occurred in Spain and Western France. Such shallow phylogeographic patterns at nuclear loci suggest that the divergence between mitochondrial clades has not been long enough for them to reach reciprocal monophyly at the nuclear level [50–52]. According to mitochondrial DNA TMRCA estimates, divergence among lineages is estimated to have occurred since the Early Pleistocene. Taking into account the slower evolutionary rates of nuclear genes, the genetic pattern observed at the mitochondrial and nuclear level is compatible with a scenario of Pleistocene divergence among lineages. Yet, the use of more variable markers such as microsatellites is needed for a full assessment of the nuclear pattern of variation within this species.

The finding of conspicuous genetic divergence among geographic clusters of populations has provided evidence for the existence of cryptic species within other Podarcis taxa including the P. erhardii[53], P. tiliguerta[31], and P. hispanica[54] species complexes. Given the extensive genetic structure observed in P. muralis and the lack of evidence for syntopy between mitochondrial lineages (except in four populations in Italy, see Figure 1), it might be suggested that some form of barrier to gene exchange has existed between distinct lineages. Nevertheless, we have no evidence that P. muralis constitutes a species complex. Pairwise uncorrected p-distances between clades ranges from 1.0 to 7.4% in cytb and from 0.9% to 4.7% in nd4, which is a level of divergence typically found within (and not between) recognised species of the genus (e.g. P. vaucheri, [44]). Further, there is no evidence for high differentiation in nuclear markers which would be indicative of long-term independent evolutionary trajectories. While a low divergence level among genetic lineages is not necessarily a synonym of absence of reproductive isolation, the available evidence suggest that, contrary to other Podarcis forms, P. muralis is a single species and not a species complex.

This does not mean, however, that the genetic landscape of the species is uniform. On the contrary, genetic variation in P. muralis is structured in many distinct lineages across its range. The phylogeographic pattern observed, with many reciprocally monophyletic allo- parapatric lineages having a Pleistocene divergence, suggests a scenario of long-term isolation in multiple ice-age refugia [1, 55]. Therefore, the geographic distribution of these genetic lineages is likely to reflect the refugial structure of this species and indicate putative areas which allowed the long-term maintenance of its genetic diversity. Assessments of many widespread European species have recovered a classic pattern of higher genetic diversity and multiple genetic lineages within southern European peninsulas [2]. On one level the pattern of P. muralis is not an exception: more than three quarters of all lineages are endemic to the Iberian (2 and 3), the Italian (7–10, 19–23) and the Balkan peninsulas (12–14, 18). The simplest explanation for this phylogeographic pattern is one of a large number of refugia occurring not only across southern peninsulas, as pioneer works in phylogeography have demonstrated at a continental scale (e.g. [1]), but also within each peninsula, conforming to the “refugia within refugia” paradigm [56]. Such compartmentalization is particularly evident in Italy. Multiple, related lineages occur in South Italy, whereas other, unrelated lineages occur in central and northern regions. These latter lineages from central and northern regions were considered by [27] as belonging to a main lineage (clade “1A”) leading to the inference of a main refugial area north of Campania and Apulia. However, though this clade was not supported in any of their phylogenetic analyses (BP ≤ 56, BPP = 0.54), in our analyses its constituent lineages are statistically well supported, unrelated, well differentiated one from the other, and parapatrically distributed. Thus, a scenario of multiple independent refugia in Central and Northern Italy appears to fit the data better. A similar reasoning suggests the existence of multiple independent refugia in the southern Balkan Peninsula. On the other hand, the southern Italian Peninsula would have acted as an independent biogeographic compartment with four related lineages arising from distinct refugia within this main refugial area. The same situation is observed in the westernmost and easternmost areas of the species range - in the Iberian Peninsula and in Turkey. Here, two or three endemic related lineages, respectively, occupy geographically different regions that likely acted as distinct refugia within a main refugium. It is noteworthy that in Iberia, where the species currently has a discontinuous distribution, genetic discontinuities do not fully correspond to geographic isolates, suggesting recent fragmentation dynamics.

A scenario of survival in multiple refugia within a southern peninsula has been inferred for many species of the continental Europe fauna and detailed studies on amphibians and reptiles provide the best evidence (see [57] for a recent review). For example, there is evidence for multiple Iberian refugia in Salamandra salamandra[58], multiple Italian refugia in Triturus carnifex[59] and various Balkan refugia in Lacerta viridis[60]. To our knowledge, however, P. muralis is the first reported case of differentiation into multiple refugia within all Mediterranean peninsulas. In fact, within P. muralis, diversity and divergence levels are not strikingly different among these regions (Table 2), suggesting that all Mediterranean peninsulas significantly and simultaneously contributed to the long-term persistence of this species throughout the Pleistocene glaciations. Therefore, the case of P. muralis not only conforms to the scenario of “refugia within refugia” [56], but it also sets a new phylogeographic pattern of “refugia within all refugia”.

According to the high number of lineages found in southern European peninsulas, the phylogeographic pattern of P. muralis fits the classical southern refugia model to some extent. However, while for many species of continental Europe these Mediterranean peninsulas were both glacial refugia and source areas for northward postglacial colonization [2, 4], in the case of P. muralis peninsular lineages are not found outside of these areas. This means that southern peninsulas acted as glacial refugia but not as sources for postglacial expansion. A possible explanation for such an unusual pattern is that northern areas outside Mediterranean peninsulas were already occupied by the species. In fact, various distinct lineages of P. muralis were found outside southern peninsulas, in the Pyrenees and France (clades 1 and 6), Northern Italy (clade 5), Slovenia and Austria (clade 11) and the Central Balkans (clade 4), reaching the highest latitudes of the species’range. Phylogenetic analyses and permution tests showed that these lineages are well differentiated and evolutionarily independent from southern lineages. The occurrence of two allopatric lineages in Western and Eastern France (clades 1 and 6), which are unrelated and considerably divergent from each other and relative to lineages occurring in Iberia (clade 2 and 3), and on the other side of the Alpine arch in North Italy (clade 5), provides evidence for a long-term persistence of P. muralis in these areas during Pleistocene glaciations, within separate refugia. Considering that the highest genetic diversity of these lineages is found along the southern portion of their range in SW and SE France respectively (data not shown), we can hypothesize that the refugia where these lineages differentiated were located in proximity of the Mediterranean coast and acted as source areas for post-glacial recolonization of Northern France and Germany. This hypothesis has been tested and validated by a recent study focused on P. muralis populations from France and Luxembourg [61], and it is further supported by palaeoclimatic and phylogeographic studies which indicate southern France as an area of prolonged ecological stability where climatic oscillations were attenuated [62] and where genetic signatures of refugia have been found among multiple species, including plants and animals (e.g. [63, 64]). A similar reasoning applies to the allopatric mitochondrial lineages found in Northern Italy, Northern Adriatic, Eastern Slovenia and Austria, and Central Balkans (clades 5, 21, 11, and 4), which are allopatric and reciprocally monophyletic, thus suggesting long-term isolation in multiple ice-age refugia. The long-term persistence of temperate species in Northern Italy (including the Po Plain and the southern foothills of the Alps) and in the Northern Adriatic has been inferred for many amphibians and reptiles in recent phylogeographic studies based on the occurrence of divergent lineages endemic to these areas (e.g., Hyla intermedia, Pelobates fuscus, Pelophylax lessonae, Triturus carnifex, Podarcis sicula; see [59, 65]). Likewise, in the area spanning from the Eastern Alps to the South-Western Pannonian region, palaeobotanical, paleoclimatic, and genetic data indicate that temperate species such as the tree Fagus sylvatica and the butterfly Erebia medusa survived in isolated refugia during glacial periods [20, 21, 66, 67]. Finally, a refugial area in the central Balkans and south-western Carpathians has already been suggested for many species showing high genetic diversity and distinct lineages in this area where also broad-leaved trees survived glacial periods (e.g. [16, 66–68]).

Altogether these phylogeographic, palaeobotanical and palaeoclimatic studies provide evidence for the long-term persistence of temperate organisms outside the traditional refugia in southern peninsulas, and underline the prominent contribution of these northern refugia to the present-day genetic pools of many temperate species.

Conclusions

In recent decades, the phylogeography of the Western Palaearctic has been mainly centered around the southern refugia model, which sets a sharp dichotomy between Mediterranean peninsulas and Northern Europe, the former being areas of long-term persistence of temperate species and sources for the postglacial assembly of northern biotas. Our data do not strictly conform to this model: a pattern of northern purity was not detected, whereas several evolutionary units endemic to northern regions were identifed. Overall, the occurrence of many reciprocally monophyletic lineages within P. muralis suggests a history of long-term persistence in multiple glacial refugia across its distribution, providing a paradigm of ice-age survival of temperate species in Mediterranean and extra-Mediterranean refugia [12]. Contrary to previous hypotheses [25, 27, 28] this species did not occupy most of its present range only recently, either from the Italian Peninsula or from any other southern peninsular refugium.

In the last two decades mounting evidence for glacial survival in northern refugia through the Pleistocene has been found among various organisms representing different biogeographical groups ([12] for a review). These studies suggest that the assembly of northern European biota was not simply the outcome of long-distance postglacial re-colonization from southern peninsulas [6, 7]. Rather, extra-Mediterranean refugia allowed the long-term persistence and differentiation of distinct genetic lineages of many organisms in the Western Palaearctic and acted as source areas for post-glacial colonization of Central and Northern Europe [12]. These biogeographic insights in the Western Palaearctic are paralleled by phylogeographic and palaeobotanical evidence for survival during ice ages of populations of temperate organisms in northern refugia located in the Eastern Palaearctic [69], as well as in the Nearctic region [70–72] providing directions for future research in the Holarctic ecozone concerning the role of northern regions as areas of long-term persistence of temperate biotas.

Understanding the relative contribution of Mediterranean and extra-Mediterranean refugia to the current patterns of genetic diversity may allow us to significantly improve our current knowledge regarding glacial refugia dynamics, the extent and the routes of dispersal between subregions and their consequences in terms of evolutionary and demographic processes [9–12, 73]. Moreover, our understanding of northern refugial biodiversity has deep consequences for conservation. The survival of temperate biota in northern refugia across a matrix of unsuitable glacial landscapes challenges the view that organisms must have gone extinct in northern regions during glaciations, and suggests that, in many cases, postglacial colonisation routes – and thus migration capacity - tracking the expansion of suitable habitats have been much smaller than previously thought [9–12, 68, 74]. Our appraisal of past species resilience, and of their ability to track suitable habitats in response to past climatic shifts, is crucial to better forecast their future performance under global climate change scenarios [10, 73, 75–77], and thus essential for more informed decisions aimed at long-term conservation of biodiversity.