Genomics of end-Pleistocene population replacement in a small mammal

Abstract

Current species distributions at high latitudes are the product of expansion from glacial refugia into previously uninhabitable areas at the end of the last glaciation. The traditional view of postglacial colonization is that southern populations expanded their ranges into unoccupied northern territories. Recent findings on mitochondrial DNA (mtDNA) of British small mammals have challenged this simple colonization scenario by demonstrating a more complex genetic turnover in Britain during the Pleistocene–Holocene transition where one mtDNA clade of each species was replaced by another mtDNA clade of the same species. Here, we provide evidence from one of those small mammals, the bank vole (Clethrionomys glareolus), that the replacement was genome-wide. Using more than 10 000 autosomal SNPs we found that similar to mtDNA, bank vole genomes in Britain form two (north and south) clusters which admix. Therefore, the genome of the original postglacial colonists (the northern cluster) was probably replaced by another wave of migration from a different continental European population (the southern cluster), and we gained support for this by modelling with approximate Bayesian computation. This finding emphasizes the importance of analysis of genome-wide diversity within species under changing climate in creating opportunities for sophisticated testing of population history scenarios.

1. Introduction

Much of high-latitude Europe was uninhabitable for temperate species in the late Pleistocene and the current populations in those regions were thus established at the Pleistocene–Holocene transition [1,2]. It is becoming clear, however, that the original view of postglacial colonization as southern populations merely expanding their ranges into unoccupied northern territories [3] is overly simplistic, as several reported examples of local genetic turnovers suggest that colonization of some high-latitude areas involved input from different source populations at different times, possibly coupled with competition and selection [4,5].

During the Last Glacial Maximum (LGM; 26–19 kya), temperate species were not able to occupy large areas of European higher latitudes because of the presence of ice sheets and adverse climatic conditions, and they only survived in sheltered climatic refugia, typically at lower latitudes [3]. Warming during the Bøling–Allerød period (beginning at 14.7 kya) provided a first opportunity for northwards range expansion in many species [6]. However, this process was largely reversed during the Younger Dryas (12.9–11.7 kya), the last cold event of the Pleistocene preceding the final warming at the beginning of the Holocene. Populations established during the Bøling–Allerød that survived through the Younger Dryas thus probably faced a massive wave of immigration from the south in the early Holocene. It is usually assumed that colonization of an unoccupied area by a particular species prevented subsequent expansions of other populations of the same species into the same area [1]. However, recent findings on mitochondrial DNA (mtDNA) of British small mammals [4,5,7] revealed a genetic turnover in southern Britain during the Pleistocene–Holocene transition where one mtDNA clade of each species was replaced with another mtDNA clade of the same species. Although mtDNA turnovers have been inferred for other species elsewhere and at various times in the past [8–10], the colonization of Britain by small mammals is now one of the best-studied models of mtDNA replacement in association with recolonization at the Pleistocene–Holocene transition [4,5,11].

At the end of the Pleistocene, Britain and continental Europe were connected by a land bridge and small mammals colonized Britain over the ‘Doggerland’ landmass [4] that later became submerged beneath the North Sea [6]. The opening of the English Channel during 9.0–7.7 kya [6] resulted in Britain becoming isolated from continental Europe and prevented further overland colonization [4]. Searle et al. [4] used mtDNA (and in one case chromosomal variants) as the genetic marker, revealing two distinct clades (1–2% divergent on average; figure 1a) of each of five small mammals in Britain, two shrews and three voles [4]. In all of these species, the clades show the ‘Celtic fringe’ pattern, with one clade having a more northern distribution relative to the second one (figure 1a; for further details, see [4]). Both clades in each species also occurred in continental Europe (figure 1a) and the explanation proposed was a two-phase colonization from continental Europe, where the initial colonization during Bøling–Allerød was by the northern clade, and the southern (replacing) clade came from continental Europe across the land bridge after the Younger Dryas [4]. Direct evidence for the timing of the replacement comes from ancient DNA for the water vole (Arvicola terrestris) where specimens from England dated to before the end of the Younger Dryas carry mtDNA of the northern clade presently restricted to Scotland, while mtDNA of post-Younger Dryas specimens is of the southern clade that is found in England today [5]. The most obvious driver that forced the replacement of the northern clade over much of England in each of the five small mammals is the climate change at that time [4]. Up until now it was not however possible to determine whether the replacement in the different species only involved mtDNA or if other parts of their genome also show signatures of replacement [4]. This could range from a selective sweep involving one or a few loci, to complete population replacement, involving a genome of one population at the cost of another [4]. A north–south pattern similar to the two mtDNA clades was observed for two different types of haemoglobin (Hb) of the bank vole (Clethrionomys glareolus) (figure 1b) [11,12], showing that other loci than mtDNA are involved in the replacement in this species. The present study therefore focuses on the bank vole and uses a large number of nuclear single-nucleotide polymorphism (SNP) loci to provide the first genome-wide perspective of population divergence and admixture in a small mammal species showing the ‘Celtic fringe’.

Figure 1.

Genetic replacement in British bank voles. Previously described spatial patterns of (a) mtDNA [4] and (b) haemoglobin [11,12] are shown. The phylogenetic tree of representative complete mtDNA genome sequences in (a) shows the placement of the British northern (blue circles) and southern (red circles) mtDNA in two divergent (1% on average) clades, which both also contain samples from continental Europe (empty circles). (c) Genome-wide ancestry proportions from the two genomic clusters at each of the six sampling sites (see electronic supplementary material, table S1) estimated from SNP genotype data of 39 voles with ADMIXTURE [13]. The thin dashed line shows the England–Scotland border.

2. Material and methods

(a). Samples

A total of 39 bank voles from six localities on an approximately north–south transect through Britain [11] were analysed (figure 1c; see electronic supplementary material, table S1). Ten bank voles from southern Sweden and 10 from the Netherlands (figure 2a) were included to represent continental European populations of the same two mtDNA clades as found in Britain [16].

Figure 2.

Genetic relationships of the British and continental bank vole populations. (a) Principal components analysis of SNP genotype data. The first two components are plotted for 39 voles from the six sites in Britain and 20 from two sites in continental Europe (see electronic supplementary material, table S1). The map of Europe is shown for reference, displaying the Doggerland land bridge (brown). (b) Tree relating the non-admixed British populations (CON, DEV and ABD) to the two continental populations (NL and SE) inferred with TreeMix [14,15], taking into account the effect of admixture in NL (arrow). The admixture rate of 0.5 was used as estimated with Admixture [13] (see electronic supplementary material, figure S4). Clustering of the non-SE ancestry of NL with CON and DEV, to the exclusion of SE and ABD, is highly supported by bootstrap analysis. The basal trifurcation reflects the lack of information about the ancestral branching order.

(b). RNA sequencing, transcriptome assembly and single-nucleotide polymorphism genotyping

A poly(A) RNA library was prepared from spleen of each vole [17] and sequenced to generate 100 bp paired-end reads by using the Illumina HiSeq 2000 sequencing platform (NCBI BioProject PRJNA429463). A reference transcriptome was constructed by assembling trimmed and filtered [18,19] reads with Trinity [20] and merging putative isoforms into gene models [21]. The reads for each vole were then mapped to the reference transcriptome with Bowtie2 [22] and SNPs were identified with mpileup of SAMtools [23] and filtered based on quality and biological criteria [24–26] using VCFtools [27], PLINK [28] and custom scripts, as detailed in the electronic supplementary material text (data and script files available through Dryad: http://dx.doi.org/10.5061/dryad.db470 [29]).

(c). Population structure and phylogeny analysis

The mean observed and expected heterozygosity and F_st were estimated with Arlequin v. 3.5 [30]. As a first step to look at the population structure, we performed principal component analysis (PCA) of the individual genotypes with Eigensoft [31]. We then used Admixture program [13] to estimate the genetic ancestry for each vole. The program estimates for each individual the proportion of its genome (Q) descending from each contributing population assuming a population admixture model in which the sampled genotypes are derived from one of K ancestral populations. To estimate the value of the K parameter, Admixture's cross-validation procedure [32] was ran for values from 1 through 8, with 10 replications for each K. We also ran Admixture for random subsets of 1000 SNPs each created by sampling SNPs from the filtered dataset with replacement. The same population model was fitted with Structure, which adopts a Bayesian approach [33] rather than maximum-likelihood estimation employed by Admixture [13]. We ran Structure for K values ranging from 1 through 5 with five replications for each value, and determined the best K with the method described in [34].

To infer the evolutionary relationships among populations we used the maximum-likelihood method implemented in TreeMix, which estimates the population tree based on SNP allele frequency data and allows for the admixture to be incorporated into the tree-building process [14,15]. To obtain a measure of confidence in the tree, we performed a bootstrap analysis where we randomly sampled sets of 500 SNPs to generate each of 100 bootstrap replicates. For more details, see the electronic supplementary material text.

(d). Testing historical scenarios with approximate Bayesian computation

To determine a population historical scenario that could best explain the observed SNP variation we compared six alternative scenarios using approximate Bayesian computation (ABC) [35–37]. For the description of the scenarios, see the electronic supplementary material (text and electronic supplementary material, figure S1). We simulated 10 000 datasets under each of the six scenarios using DIYABC v. 2.1.0 [38]. We then applied a tree-based classification method ‘random forest’ (RF) [39,40], which generates a set of binary decision trees trained by bootstrap samples of the simulated datasets [40,41]. The RF classifier is then applied to the observed dataset and the scenario receiving the most votes over all trees is selected [40,42]. An estimate of the posterior probability of the selected scenario is then obtained as the probability complement of the model selection error estimated by a regression RF [40,43]. For more details, see the electronic supplementary material text.

3. Results

The final dataset consisted of 10 988 SNPs genotyped in 59 voles (for more details, see the electronic supplementary material text). While levels of heterozygosity are similar between samples, F_ST (s.d.) of 0.277 (0.191) shows that a substantial proportion of overall diversity is partitioned among the geographical locations (see electronic supplementary material, tables S1 and S2, and figure S2).

In the PCA plot (figure 2a), bank voles from the same site are clustered together. PC1 explains 13% of the total genetic variation and mainly separates the British populations from those in continental Europe. ABD is also separated from the other British sites by PC1, but on the opposite side to the continental populations. PC2 explains 10% of the genetic variation and largely accounts for a north-south genetic gradient within Britain (figure 2a).

The clustering analyses with Admixture and Structure (results not shown) provided very similar results. Within Britain, the most probable number of genomic clusters is K = 2 (figure 1c). The admixture proportions between the two clusters show a clear north–south cline where the sites at the extreme north (ABD) and extreme south (CON and DEV) are without admixture and the proportions of southern ancestry at the geographically intermediate sites vary between 60% (MLN) and 98% (GLS) (figure 1c). These results hold true in analyses of random subsets of SNPs (see electronic supplementary material, figure S3), showing the pattern is not driven by a small number of highly differentiated loci. An analysis including the two reference continental populations (SE for the northern mtDNA clade, NL for the southern mtDNA clade) identified K = 3 as the most probable number of clusters and showed both continental populations as having ancestry from a separate cluster, although NL also had nearly equal contribution from the cluster already identified in southern Britain (see electronic supplementary material, figure S4).

In order to assess the historical relationships of the British bank voles to the two continental populations we selected sites within Britain without admixture to represent the British northern (ABD) and southern (CON and DEV) genomic clusters (figure 1c). Because NL has admixed ancestry, any attempt to build a tree relating it to the other populations is complicated by that admixture [15]. Using the TreeMix algorithm that allows for an estimate of admixture to be incorporated in the tree [14] we show that the non-SE ancestry in NL forms a well-supported clade with CON and DEV, but the branching order between the southern ancestor and the ancestors of SE and ABD remains unresolved (figure 2b).

By comparing the competing historical scenarios with ABC-RF (see the electronic supplementary material, figure S1) we found that in the best-supported scenario (scenario 6; p = 0.461) the ancestor of SE diverged first followed by divergence of the ancestors of ABD and CONDEV at about the same period of time. This was followed about three times more recently by nearly equal admixture between the ancestors of CONDEV and SE that gave rise to a population ancestral to NL (see electronic supplementary material, table S3).

4. Discussion

We have been able to apply a set of over 10 thousand SNP loci to the specific setting of recolonization of Britain at the end of the last glaciation to test the replacement hypothesis with greater rigour than ever before. This is important because the colonization of Britain by small mammals is one of the best studied models of lineage replacement in association with recolonization at the Pleistocene–Holocene transition [4,5].

(a). The admixture origin of British bank voles

Although genetic clines may also arise from other evolutionary processes, such as isolation by distance [44], the observed PCA configuration where the British sites are arranged along a major axis of variation (PC2 in figure 2a) is consistent with a history of admixture between ancestral populations closely related to ABD on one hand and CON and DEV on the other [31,45]. The model-based clustering (figure 1c) also infers the British bank voles as a mixture of two ancestral genomes whose proportions increase to the north and south, respectively, similar to the patterns of mtDNA and Hb (figure 1a,b). However, while bank voles in southern Britain share some ancestry with NL, as expected if they stem from the same source population [16], bank voles in northern Britain do not show detectable ancestry from SE. A possible explanation is that, in contrast to mtDNA [16], the nuclear genome of the first colonists of Britain did not derive from the same continental source population as that of SE. This hypothesis is corroborated by the two-colonization scenario picked by ABC-RF. Taken at face value, the estimated time of divergence of the ABD ancestor from the SE ancestor (38.4 kya; assuming complete population turnover twice a year [46]) is consistent with the first colonization stemming from a source population that diverged from the population ancestral to SE prior to the LGM. The divergence time between the ancestors of ABD and CONDEV (35.7 kya) suggests that the second colonization originated from yet another LGM refugium, while the time of origin of NL (12.8 kya) is consistent with this colonization being related to the climate shift around the Younger Dryas. Taken together, while other evolutionary processes can produce a similar geographical pattern to that observed in bank vole populations in Britain, given the results of the phylogenetic analyses, the ABC-RF modelling (including dating of divergence) as well as the evidence from previous mtDNA studies [4,16], the double colonization hypothesis seems by far the most likely explanation.

(b). The end-Pleistocene colonization of Britain

Therefore, the separation in the PCA (figure 2a) and the absence of admixture at ABD (figure 1c) probably means that this northernmost site represents the early colonization of Britain during the Bøling–Allerød. However, the bank vole population in Britain would likely have contracted and/or become sparse during the Younger Dryas cold spell, which would have made it susceptible to replacement by the second population arriving during the final warming after the Younger Dryas [4]. The SNP data suggest that the replacement in Britain involved invasion by whole genome of the second arriving population. The replacement therefore would presumably have been in the form of a moving contact zone rather that selective sweeps at particular loci [47]. A more detailed analysis would require further focused sampling in the contact zone, but according to the available SNP data the genomic cline now appears to be centred near Edinburgh (MLN) in southern Scotland (figure 1c). However, not all genes have boundaries positioned in the same place. While for the mtDNA the northern boundary between the two clades is at a similar location as for the SNPs (figure 1a), the contact between the two Hb types is positioned further south, in northern England (figure 1b). In addition, there is a peripheral distribution of the ‘northern’ mtDNA clade (that of the first colonist) also in the extreme south of England delimited by a boundary roughly coincident with the River Thames (figure 1a). Such differences in spatial patterns between markers, including relictual disjunctive distributions of the replaced type (evident in mtDNA here), are not unexpected for moving contact zones, which can often leave trails of markers [48]. It is likely that this applies to the case of bank voles in Britain and the present phylogeographic pattern is the result of westward and northward contact zone movement in favour of the second colonizing population arriving from the east over the Doggerland land bridge (figure 2a). Indeed, the discrepancy in the nuclear and mtDNA genome histories in the continental range can possibly be explained by similar replacement events taking place in continental Europe.

A role for natural selection has been suggested as a driver of the movement of contact zones [47]. No outlier loci were revealed that would pinpoint specific genomic regions involved in adaptive divergence between the two colonists (see electronic supplementary material text). However, when adaptation occurs in traits that are encoded by many genes, as expected for ecological traits, it would be difficult to detect with outlier loci because no individual region of the genome is likely to show strong signatures of selection [49]. Interestingly, physiological differentiation related to resistance to cellular oxidative stress, on which selection can potentially act, has been described between the Hb of the first and second bank vole colonists [11]. It is thus possible that selection has had a role in driving the replacement in Britain and that the differences between markers are due to different patterns of selection.

The application of genome-wide SNP data to the specific setting of recolonization of Britain allowed us to test the population replacement hypothesis with much greater rigour than has previously been attempted. Although mtDNA turnovers have been shown to have occurred in other species elsewhere and at various times in the past, the colonization of Britain by small mammals is one of the best-studied models of mtDNA replacement in association with recolonization at the Pleistocene–Holocene transition (with signatures of replacement in at least five separate species). Thus, although in a confined geographical area (Britain), it is a well-studied and important area to gain proof that true population replacements did occur during the climatic change at the Pleistocene–Holocene transition.

Our finding is of particular relevance to population replacements in response to current climate change. The population shifts that are currently occurring are often not into unoccupied territories, but into areas already occupied by other populations, which may or may not lead to replacement [50]. Disentangling the processes behind replacements that occurred in the past, therefore, is of general interest as it may help understand and ultimately predict the response of species to current and future climate change.

Data accessibility

Sequence data: NCBI BioProject PRJNA429463. Transcriptome assembly, SNP genotyping data and script files: http://dx.doi.org/10.5061/dryad.db470 [29].

Authors' contributions

P.K. and J.B.S. conceived the study. P.K. designed and coordinated the study, carried out the data analysis and wrote a first draft of the manuscript. S.M. participated in the design of the study and carried out the molecular laboratory work. M.K. carried out the assembly and SNP calling. W.B. performed the random forest analysis. All authors contributed to writing and gave final approval for publication.

Competing interests

We declare we have no competing interests.

Funding

The study was carried out with the financial support from the Czech Science Foundation (grant nos P506-11-1872 and 16-032485) and the Ministry of Education, Youth and Sports of the Czech Republic (projects KONTAKT II LH15255 and EXCELLENCE CZ.02.1.01/0.0/0.0/15_003/0000460 OP RDE).