A complex scenario of glacial survival in Mediterranean and continental refugia of a temperate continental vole species (Microtus arvalis) in Europe

2.1 Population sampling and laboratory procedures

We sampled 261 common vole specimens from 10 localities throughout its range in the Iberian Peninsula (see details in Appendix 1 and Figure 1), using Sherman traps between 2012 and 2014. From each individual, we took 2-mm tissue biopsies from the ear and preserved them in 96% ethanol. Afterward, all individuals were released at their respective collection site, and tissue and DNA samples were stored at the Instituto de Investigación en Recursos Cinegéticos DNA data bank. All handling procedures were approved by the UCLM Ethics Committee (reference number CEEA: PR20170201) and in accordance with the Spanish and European policy for animal protection and experimentation. We completed our sampling with a compilation of 151 published cytochrome b sequences, collected in 68 localities across 22 countries covering the entire European range of M. arvalis (references and GenBank accession numbers are indicated in Appendix 1).

The phylogeographic architecture of common vole has traditionally been studied mainly using the mtDNA cytochrome b gene (Fink et al., 2004; Haynes et al., 2003; Heckel et al., 2005; Martínková et al., 2013; Stojak et al., 2015; Stojak, Wójcik, et al., 2016; Tougard et al., 2008) or in combination with microsatellite and Y chromosome data (Beysard & Heckel, 2014; Stojak, Borowik, Górny, McDevitt, & Wójcik, 2019; Stojak, McDevitt, et al., 2016). Although some discordances are expected among gene tree topologies at different markers and the species tree (Kubatko, Carstens, & Knowles, 2009), previous studies on the phylogeography of this and related species suggest that mtDNA is an excellent marker for identification of glacial refugia and study of range expansion (Herman et al., 2014; Martínková et al., 2013; Stojak, McDevitt, et al., 2016). We are fully aware of the benefits of multilocus versus approaches single-marker approaches, but multilocus analyses would have constrained us to use a reduced number of populations. Moreover, former studies that have used autosomal and sex-chromosome markers in M. arvalis have provided congruent phylogeographical topologies and very similar patterns of genetic structure and dating events to those found with mtDNA (Beysard & Heckel, 2014; Braaker & Heckel, 2009; Heckel et al., 2005; Martínková et al., 2013; Stojak, McDevitt, et al., 2016), albeit much finer phylogenetic resolution was obtained with cytochrome b than with nuclear markers at the continental and regional scale (Heckel et al., 2005). Consequently, we used here this mitochondrial gene, given the need for sufficient resolution and direct comparison of our findings with earlier range-wide research on this species.

We digested samples overnight in 250 μl lysis buffer (0.1 M Tris-HCl, 0.005 M EDTA, 2% SDS, 0.2 M NaCl, pH 8.5) and Proteinase K (10 ng/μl) and extracted total genomic DNA from live specimens using a standard AcNH4 protocol. We adjusted DNA extractions to a working dilution of 25 ng/μl and PCR-amplified the entire cytochrome b mitochondrial gene (1,143 bp) using primers L7 (5′–ACCAATGACATGAAAAATCATCGTT–3′) and H6 (5′–TCTCCATTTCTGGTTTACAAGAC–3′) (Tougard et al., 2008). We cleaned up the PCR products using Exonuclease I and Shrimp Alkaline Phosphatase (Fermentas), and direct sequencing was performed with the same PCR primers and the BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) on an ABI PRISM 3130xl Genetic Analyzer (Applied Biosystems). We manually checked chromatograms for stop codons and obvious sequencing errors, and aligned them using the software BioEdit 7.0.5.3 (Hall, 1999). All the sequences obtained are deposited in GenBank (MG874847–MG874883).

2.2 Bayesian phylogeographic inference

We reconstructed the phylogeographic history of M. arvalis through continuous space and time, by modeling the spatial dispersion of its lineages using Bayesian phylogeographic inference in BEAST2 (Bouckaert et al., 2014; Lemey, Rambaut, Drummond, & Suchard, 2009; Lemey et al., 2010). This model combines geographic and genetic data, using the geographic location of each specimen and accommodating branch-specific variation in dispersal rates. For this analysis, we selected one sample per each haplotype found at each locality, totaling 197 sequences and 141 haplotypes, from 78 localities across 23 countries (Appendix 1). Our final alignment was restricted to a common length of 953 bp (Dataset S1). We created a xml input file in BEAUTI v.2 (Bouckaert et al., 2014), and used a jitter option of 0.50 to create unique coordinates for individuals collected at identical sites.

We used a marginal likelihood estimate (MLE) model selection procedure to evaluate the relative fit of our data to (a) models that assume no branch-specific rate variation in dispersal rates (i.e., homogeneous Brownian diffusion); and (b) relaxed random walk (RRW) models that assume different distributions for rate variation among branches (i.e., Cauchy RRW, lognormal RRW, and gamma RRW). We generated posterior distributions of each model by running 7 × 10⁷ Markov Chain Monte Carlo (MCMC) generations that were sampled every 7,000 generations, after discarding the first 10% samples as burn-in. We assessed convergence by visualizing the outputs of posterior distributions and effective sampling sizes (ESS) in TRACER v.1.4.8. (Rambaut & Drummond, 2007).

We then calculated Bayes factors for each model, and used them to select the spatial diffusion model that is best supported by the data. Bayes factors were derived from their MLE and based on path sampling (PS) method (Baele et al., 2012), MCMC chain of 1 × 10⁶ generations and 50 path steps. Our data fitted best the Cauchy-distributed RRW diffusion rate model (MLE = −5,749.378); therefore, we applied this model to all subsequent analyses (gamma-distributed RRW: MLE = −6,468.140; lognormal-distributed RRW: MLE = −5,897.870, and homogeneous Brownian diffusion: MLE = −5,962.448).

To visualize the spatial diffusion of M. arvalis lineages, we inferred a maximum clade credibility (MCC) tree and projected it on a spatial map. We ran four independent MCMC chains for 2.5 × 10⁷ generations—using the BEAGLE library (Ayres et al., 2012) to improve computational performance—and sampled their posterior distributions every 2,500 generations. Following burn-in of the first 10% trees, we combined the four independent analyses using LogCombiner 2.4.1 (whereby all parameters reached ESS >200) and randomly subsampled 10,000 trees that were used to construct a MCC tree in TreeAnnotator 2.4.1. We used the resulting MCC tree as input for the program Spread 1.0.4 (Bielejec, Rambaut, Suchard, & Lemey, 2011) to generate a keyhole markup language file for visualizing spatial diffusion in Google Earth over the complete posterior distribution of trees. We used the Time Slicer function in Spread 1.0.4 to estimate the 80% highest posterior density (HPD) region of the unobserved ancestral locations for each branch that intersects ten specific time points, which roughly coincide with major internal nodes within the MCC tree: 50,700 years before present, ybp; 43,500 ybp; 36,300 ybp; 33,400 ybp; 23,700 ybp; 21,100 ybp (coincident with the last glacial maximum, LGM); 16,300 ybp; 12,600 ybp; 7,500 ybp, and the present.

We also explored temporal changes in the effective population size using a Bayesian skyride tree prior implemented in BEAST v. 1.8.1 (Drummond, Suchard, Xie, & Rambaut, 2012). We used the HKY model of sequence evolution and a strict molecular clock with an average substitution rate across the tree of µ = 3.27 × 10^–7 substitutions/site/year (assuming a generation time of 1 year) (Martínková et al., 2013). We obtained this value from the most recently published—and likely most accurate—rate for cytochrome b of M. arvalis (Martínková et al., 2013; Stojak, McDevitt, et al., 2016; Stojak et al., 2015; Stojak, Wójcik, et al., 2016), estimated using ancient DNA sequences from radiocarbon-dated specimens. We ran four independent MCMC chains for 3 × 10⁷ generations sampling every 3,000 generations. Convergence was assessed in TRACER v.1.4.8. by confirming that the ESS for all parameters was higher than 200 and that independent runs yielded similar posterior distributions. We combined the four runs using LogCombiner v1.8.1 (http://beast.bio.ed.ac.uk/logcombiner) after discarding the first 10% of sampled generations as burn-in, and obtained estimates and credible intervals for each parameter and demographic reconstruction using TRACER v.1.4.8.

With the aim to infer within- and between-lineage evolutionary relationships among haplotypes analyzed, we constructed a median-joining network (Bandelt, Forster, & Röhl, 1999) using the software NETWORK 5.0 (available at http://www.fluxus-engineering.com/sharenet.htm). Genetic distances were calculated through a pairwise distance matrix in MEGA v. 7.0.16 (Kumar, Stecher, & Tamura, 2016).

4.1 European glacial refugia and postglacial expansion

Our phylogenetic reconstruction was congruent with previous studies and recovered with high support six lineages of M. arvalis (Bužan et al., 2010; Fink et al., 2004; Haynes et al., 2003; Martínková et al., 2013; Stojak et al., 2015; Tougard et al., 2008). The well-delimited geographic distribution of these lineages exposes the importance of allopatric factors in shaping the genetic structure of this species at a continental scale, probably through multiple isolation events during the mid- and late Würm ice age (60,000–11,500 ybp). Our model-based phylogeographic analysis inferred the location of the most recent common ancestor of all haplotypes in a region between the Alps and southern Germany around 51,000 ybp (95% HPD: 62,400 –39,500 ybp). This location is consistent with the oldest fossil remains of M. arvalis found in SW Germany (Kowalski, 2001) and relates closely to what (Tougard et al., 2013) proposed as the common cradle of M. arvalis in “western Central Europe.” Moreover, our date estimates for the root of the tree and of major lineages coincided with recent research (Stojak, McDevitt, et al., 2016). This scenario is clearly not consistent with the fossil evidence (oldest fossil remains of M. arvalis in Central Europe from the Late Cromerian, 465,000 ybp; (Kowalski, 2001)) although accurate species identification in fossil remains appear to be problematic and should always be considered cautiously for this and related species (Navarro et al., 2018; Tougard, 2016). The extreme similarity between sibling vole species (Markova et al., 2010) and the morphological changes probably suffered by the species between the mid-Holocene and the present day are known to be unfavorable for diagnosing M. arvalis in the fossil record (Markova, Beeren, van Kolfschoten, Strukova, & Vrieling, 2012). Here, we focus on extant lineages, and therefore, we lack direct information about extinct lineages, which are often those found in the fossil record but that probably did not survive past periods (Tougard, 2016). This probably explains the large discrepancy in estimated origin and divergence times of major M. arvalis lineages between our study and the fossil evidence. In the future, research involving ancient DNA can contribute to shed light on these complex scenarios.

Even though the Bayesian posterior mean estimate should provide a reasonable basis for inferences, it is unwise to use our data to make strong statements about a single origin and subsequent lineage diversification or that different lineages originated more or less at the same time, given that 95% HPD estimate for the root of the tree overlaps with the origin of the Western, Italian, Balkan, and Central + Eastern lineages. In any case, M. arvalis could have survived in the Alps during the LGM and Younger Dryas periods (YD; 12,900–11,700 ybp, (Rasmussen et al., 2006), maybe in regions of lower elevation between the northern slope of the Alps and the Danube River in southern Germany. Most importantly, this area could represent not only the origin of dispersal for the species in Europe, but also an extra-Mediterranean glacial refugium.

The timing and pattern of cladogenetic events also point to geographic isolation during the last glaciation as the major driver of lineage divergence, as well as to the existence of multiple refugia for M. arvalis. Most of the major splits between lineages occur around the time of the LGM, such as the split between Western-North and Western-South lineages (mean: 34,028 ybp. 95% HPD: 45,447–23,882 ybp), the split between Eastern and Central lineages (mean: 24,072 ybp. 95% HPD: 15,823–32,275 ybp), or the split between the Iberian and Western-South lineages (mean: 21,069 ybp. 95% HPD: 29,743–14,087 ybp). For instance, the most recent common ancestor of the Eastern lineage was located west of the Danube River around 13,639 ybp (95% HPD: 18,809–8,820 ybp), which could imply that a glacial refugium in the eastern Alps was the origin of this lineage. However, based on the molecular data alone, we cannot rule out the existence of a Carpathian refugium, as it has been proposed for numerous species (Kotlik et al., 2006; Pazonyi, 2004; Sommer & Nadachowski, 2006) including M. arvalis (Stojak et al., 2019, 2015).

Similarly, the high differentiation of the Balkan lineage relative to the other Central Eastern lineages could be the result of allopatric divergence, following isolation in a Ponto-Mediterranean refugium in the northwestern Dalmatian coast (Dinaric Alps). Furthermore, and despite the co-occurrence of both Eastern and Balkan haplotypes in that region at the present time (Bužan et al., 2010; Stojak et al., 2015), it appears that the Balkan lineage has historically maintained a very restricted distribution, and has not expanded nor contributed to the gene pool of northern or eastern European populations. The large and recent expansion detected (Figure 4) in populations of the Eastern lineage to the center and northwestern Balkan Peninsula and south of the Danube River would explain their presence there. Based on these, we could first consider the Balkan Peninsula as a region of endemism of M. arvalis, and second, refute previous hypotheses suggesting that the Eastern lineage originated from a refugium in the Balkans (Haynes et al., 2003; Heckel et al., 2005).

Within the Western-South lineage, the split between the Iberian and the French clades around the LGM suggests divergence in two separate refugia. One refugium might have existed near the region of Aquitaine, in southwestern France (i.e., Pyrénées-Atlantiques), from where the species likely expanded to northern Europe, and another one south of Pyrenees—or south of the Ebro River Valley—that further extended to the center and northwest of the Iberian peninsula. According to the temporal-spatial diffusion pattern, the Iberian Peninsula served as a glacial (Mediterranean) refugium for extant populations of common vole during the last cold stage of the Pleistocene, including the LGM. This Iberian clade, however, did not back-colonize the Pyrenees or France during the present interglacial period, as could be deduced by the absence of Iberian haplotypes in populations of these two regions. The Holocene expansion of Iberian voles could have been limited ecologically—maybe by predators or competition with closely related species—or geographically, by physical barriers such as the Ebro River following deglaciation (see below). Consequently, Iberian voles can be considered as a long-term isolated endemism of the Iberian Peninsula (see also Tougard et al., 2008) without admixture with other mitochondrial lineages, and concordant with previous subspecific designation. The most recent common ancestor of tentative M. a. asturianus specimens was estimated around 9,000 ybp in north-central Spain, and the more recent split within this clade was dated 8,500–4,000 years ago. The geographic expansion of the Iberian clade during the late Holocene (i.e., last 5,000 years) seems to be a key factor shaping its current population genetic structure.

In Central Europe, the presence of a glacial refugium has already been suggested for M. arvalis (Fink et al., 2004; Tougard et al., 2008), and for other species (reviewed in Schmitt & Varga, 2012), but for the first time we were able to pinpoint its geographic location using model-based phylogeographic inference. There is good evidence that several continental species had multiple differentiation centers around the European high mountain systems (Schmitt, 2007; Stewart & Lister, 2001), possibly due to the existence of more humid conditions around these mountains in a context of increased dryness westward during the coldest phases of the Pleistocene in Europe. For this group of species, it has been argued that water-limited conditions, rather than temperature, may have limited the species’ glacial distributions along an east–west gradient in response to moisture (Schmitt, 2007), and there is recent evidence of the importance of climate for the genetic structure of this and related species (Stojak et al., 2019).

Our analyses, however, failed to conclude whether only one or multiple refugia occurred scattered along the Alps and adjacent areas. Multiple glacial refugia have been proposed along the northern, southern, southwestern, and eastern slopes of the Alps (Schönswetter, Stehlik, Holderegger, & Tribsch, 2005). Particularly for M. arvalis, the Alpine region is a well-known contact zone between three evolutionary lineages (Western-North, Central, and Italian), and both the northern and southern slopes of the Alps have been proposed as sources of post-LGM recolonization (Braaker & Heckel, 2009). Interestingly, the inferred location and time for the origin of expansion of the current lineages of M. arvalis coincide with the main routes of westward expansion of agriculture in Europe during the Neolithic period ~10,200–2,000 ybp (Larson et al., 2007; Rowley-Conwy, 2011). The treeless tundra or steppe that had previously extended over much of Europe was progressively replaced by woodlands from the beginning of the Holocene, 11,000 years ago (Hejcman, Hejcmanová, Pavlŭ, & Beneš, 2013; Huntley & Birks, 1983; Turner & Hannon, 1988). Thus, it could be argued that the clearance of woodlands to turn them into cultivated lands (Roberts et al., 2018), and the progressive anthropogenic transformations of the landscape (Bouma, Varralyay, & Batjes, 1998; Garcia, Alda, et al., 2011; Garcia, Mañosa, et al., 2011; Novenko, Volkova, Nosova, & Zuganova, 2009; O’Connor & Shrubb, 1986; Ruddiman, 2003), could have facilitated the expansion and connectivity of populations of steppe and grassland species in a postglacial environment (Bouma et al., 1998; Garcia, Alda, et al., 2011; Garcia, Mañosa, et al., 2011; O’Connor & Shrubb, 1986; Pärtel, Bruun, & Sammul, 2005).

The geographic structure of the lineages of M. arvalis also hints at factors limiting postglacial recolonization and/or current dispersal of the species. The major mtDNA lineages of M. arvalis in Central Europe are well delimited by the major river systems (Figure 1). For example, the Loire and Rhone rivers in France divide the Western-North and Western-South lineages, something already indicated by Tougard et al. (2008). Likewise, the Ebro River in Spain separates the tentative M. a. asturianus specimens and M. a. arvalis from the Western-South clade, which includes populations from the Pyrenees and southern France. Overall, the distribution of M. a. asturianus in Iberia is limited to the north and south by the Ebro and Tajo rivers, respectively. The range of the Italian lineage is also limited to the south by the Po, and the Central lineage is roughly delimited by the transitional zone between Oder and Vistula to the west and by the Danube to the south (Stojak, McDevitt, et al., 2016; Stojak, Wójcik, et al., 2016). Furthermore, the part of the Danube basin in the eastern Balkan region seems to represent a geographic boundary between the Balkan and Eastern lineages. Similarly, and in view of the current geographic distribution (Figure 1), the Don-Volga river system marks the eastern limit of the arvalis range (Bulatova et al., 2007; Jaarola et al., 2004; Tougard et al., 2013). As could be expected, these barriers are porous, and most lineages show haplotypes that have leaked to the other side of these rivers, somewhat blurring the boundaries between lineages near the headwaters. This could either be a consequence of recent dispersal during periods of reduced runoff in the Holocene (Bernárdez et al., 2008; Combourieu-Nebout et al., 2013), or by passage through river headwaters, where lineages often come into contact (Fouquet et al., 2012). In fact, most European lineages of M. arvalis meet in a Central European region near de Alps (Figure 1), which coincides also with the headwaters of the major European rivers. Even more recently, river bridges and other human infrastructures also allow voles to cross these barriers. More populations from putative areas of secondary contact and lineage admixture, as well as the use of additional nuclear genetic markers, will be used in future studies to assess patterns of hybridization and introgression between lineages.

Our findings, therefore, provide support for a major role of large European river systems in shaping geographic boundaries of M. arvalis (Stojak, McDevitt, et al., 2016). In contrast to the widely accepted role of mountain systems (e.g., Alps and Pyrenees) precluding postglacial expansion of many temperate species in southern Mediterranean peninsulas (Habel et al., 2005; Hewitt, 1999; Schmitt, 2007), or the role of humans as responsible for the colonization of some islands (Martínková et al., 2013), very little is known about a similar potential role of rivers.

It is expected that during the deglaciation of Europe, the distribution and extent of freshwater systems suffered dramatic changes; these had a significant effect in the postglacial recolonization of terrestrial species. When ice sheets started to retreat from their maximum extent, between 22,000 and 17,000 ybp, there was an increased meltwater production and river water runoff, causing a sudden reactivation of the hydrological cycle (Mangerud et al., 2004; Ménot et al., 2006). Therefore, these water bodies may have played a dual role, either as barriers for dispersal, and/or as corridors, along water banks, consequently affecting gene flow and genetic differentiation, especially in small mammals or species with low vagility such as voles.

In conclusion, M. arvalis represents a remarkable study organism that combines attributes of continental and Mediterranean faunal elements (De Lattin, 1967; Schmitt & Varga, 2012). On one hand, Central Europe acted as a glacial refugium and as a source for other European populations, and on the other hand, additional Mediterranean refugia acted as drivers of endemism. Although there is abundant literature on Mediterranean species that found shelter in extra-Mediterranean glacial retreats (Provan & Bennett, 2008; Schmitt & Varga, 2012), very few studies report continental species in both Mediterranean and extra-Mediterranean refugia. The pattern we describe for the common vole is, however, highly congruent with the phylogeographical pattern in the bank vole M. glareolus, a European forest vole species with Mediterranean and continental refugia, and in which the Mediterranean phylogroups did not contribute to the postglacial recolonization of much of their range (Deffontaine et al., 2005). To the best of our knowledge, so far, the co-occurrence among Mediterranean and extra-Mediterranean glacial refugia for continental species is mainly restricted to eastern Europe, with few examples referred to continental species with refugial areas in the Balkan Peninsula, such as butterflies (Gratton, Konopiński, & Sbordoni, 2008; Junker et al., 2015; Schmitt, Rákosy, Abadjiev, & Müller, 2007), reptiles (Ursenbacher, Carlsson, Helfer, Tegelström, & Fumagalli, 2006), and invertebrates (Pinceel, Jordaens, Pfenninger, & Backeljau, 2005). Whether M. arvalis majorly follows a continental or a Mediterranean specific biogeographical pattern (Schmitt, 2007) is still unclear, in part because of the difficulty to classify ecologically diverse and widely distributed species. More research on continental species is needed to test the generality of the biogeographical pattern of M. arvalis in Europe.