Divergent evolutionary processes associated with colonization of offshore islands

Abstract Oceanic islands have been a test ground for evolutionary theory, but here, we focus on the possibilities for evolutionary study created by offshore islands. These can be colonized through various means and by a wide range of species, including those with low dispersal capabilities. We use morphology, modern and ancient sequences of cytochrome b (cytb) and microsatellite genotypes to examine colonization history and evolutionary change associated with occupation of the Orkney archipelago by the common vole (Microtus arvalis), a species found in continental Europe but not in Britain. Among possible colonization scenarios, our results are most consistent with human introduction at least 5100 bp (confirmed by radiocarbon dating). We used approximate Bayesian computation of population history to infer the coast of Belgium as the possible source and estimated the evolutionary timescale using a Bayesian coalescent approach. We showed substantial morphological divergence of the island populations, including a size increase presumably driven by selection and reduced microsatellite variation likely reflecting founder events and genetic drift. More surprisingly, our results suggest that a recent and widespread cytb replacement event in the continental source area purged cytb variation there, whereas the ancestral diversity is largely retained in the colonized islands as a genetic ‘ark’. The replacement event in the continental M. arvalis was probably triggered by anthropogenic causes (land‐use change). Our studies illustrate that small offshore islands can act as field laboratories for studying various evolutionary processes over relatively short timescales, informing about the mainland source area as well as the island.


Introduction
Marine islands were pivotal settings for the development of evolutionary theory (Wallace 1880) and continue to inspire evolutionary biologists today (Grant 2008). They tend to be ecologically simple and distinctive compared with mainland settings, with fewer competitors and predators, a different food supply and environmental conditions and a smaller living space, all factors that promote rapid and perhaps substantial evolutionary change relative to mainland source populations. Studies on islands have therefore provided insights into evolutionary processes such as species radiations (Emerson 2002) and the generation of evolutionary novelty (Lachaise et al. 2000).
Marine islands are also interesting for the varied ways in which they can be colonized. First, for islands separating from a mainland, some species may already be present at island formation. Second, islands may be colonized by natural 'sweepstake' dispersal (Simpson 1940). Finally, species may be introduced into islands by humans.
Many of the classic systems for the study of island evolution, particularly species radiations, have involved oceanic islands or archipelagos colonized by sweepstake dispersal, for example Darwin's finches on the Galapagos and Hawaiian drosophila. However, here we consider offshore islands, which can be colonized by all three mechanisms, providing rich opportunities for evolutionary studies, notably of poor dispersers such as small mammals (Berry 1996).
Offshore islands around the Atlantic coast of Europe were formed by end-glacial sea level rise, and thus, small mammals currently found on them may have already been present prior to island separation, and perhaps even before the last glacial maximum (LGM; c. 20 000 BP: Ehlers & Gibbard 2004), if local conditions were compatible with survival (Beirne 1952;Mart ınkov a et al. 2007). Alternatively, the small mammals could have been introduced by humans, for example, within bedding/fodder for livestock (Corbet 1961). Sweepstake dispersal, by rafting on floating masses of vegetation (Morrison et al. 1992) or walking over ice (White & Searle 2008), is also a possibility.
In this study, we aim to distinguish between these different modes of colonization of a European offshore archipelago (Orkney) by a small mammal (the common vole Microtus arvalis) and to document evolutionary change in that species postcolonization. Microtus arvalis is a ground-living, grass-eating small rodent that is common in agricultural meadows (Niethammer & Krapp 1982). The species is widespread in western continental Europe, although it is absent from Britain and Fennoscandia (Fig. 1a). Therefore, the whole British landmass separates the M. arvalis populations in continental Europe and on Orkney. The only species of Microtus in mainland Britain and other islands off mainland Britain is Microtus agrestis, which, like M. arvalis, has a wide continental European range (Shenbrot & Krasnov 2005). The two species belong to unrelated lineages within Microtus (Fink et al. 2010;Mart ınkov a & Moravec 2012).
The colonization history of a population can be inferred through comparisons with potential source populations. If the presence of the Orkney vole is explained by glacial survival or sweepstake dispersal, then surviving representatives of the source populations are no longer available, because there are no extant populations suitably close to Orkney. However, human introduction can come from a distance, and the most likely source area is represented by the closest coastal part of the continental European range of the same mitochondrial lineage as found on Orkney, that is, northern France and Belgium (Haynes et al. 2003;Tougard et al. 2008). For this reason, we focus on this potential source area in our analyses and consider whether such a human introduction is feasible or whether one of the other modes of colonization is more likely.
To explore the colonization history of the Orkney vole and the evolutionary change since colonization, we compare populations from Orkney and the potential source area using mitochondrial DNA sequences and microsatellites. We also use archaeological deposits of the species from Orkney for ancient DNA (aDNA) sequencing and radiocarbon dating (two Orkney vole bones have previously been radiocarbon-dated [Hedges et al. 1987] using technology now outdated), to obtain a more detailed temporal dimension to the colonization history and to support a demographic analysis. Finally, we use geometric morphometrics of a molar tooth as a marker for morphological evolution. This work provides a more sophisticated assessment of colonization and evolution of the Orkney vole than has previously been possible (Berry & Rose 1975;Corbet 1986;Haynes et al. 2003;Thaw et al. 2004;Gorman & Reynolds 2008 evolutionary changes in offshore islands, but also the mainland areas with which they are being compared.

Specimens
Details of all Microtus arvalis specimens used in this study are listed in Tables 1 and 2 and Tables S1-S3 (Supporting information) and mapped in Figs 1 and 2 and Figs S1 and S2 (Supporting information). Geometric morphometrics were applied to 553 modern specimens from 27 localities, while 651 M. arvalis specimens from 70 localities were used for the analysis of modern DNA (125 specimens were used in both cytb and microsatellite studies). Ancient DNA analysis was conducted on 37 archaeological specimens from eight localities. Nineteen of these were among 23 archaeological specimens used for radiocarbon dating, calibrated using Calib Rev. 5.0.1 (Stuiver & Reimer 1993) and IntCal04 (Reimer et al. 2004), with dates before present (BP) standardized to 1950 AD.

Geometric morphometrics
Modern M. arvalis from populations in Orkney and continental Europe (Fig. 1a) were compared by geometric morphometrics of dental phenotype. Digital photographs of the first lower molar (M 1 ) were used to record two-dimensional Cartesian coordinates of 18 anatomical landmarks at the bases and tips of the lingual and labial cusps as well as 12 equidistant semilandmarks on a manually drawn curve along the anterior loop (Fig. 1b).
Digitization of landmarks and semi-landmarks was performed using TPSdig 2 (Rohlf 2010a). Position, orientation and scaling information from the raw coordinates were standardized by a generalized Procrustes analysis (GPA) using TPSrelw 1.49 (Rohlf 2010b). To combine landmarks and semi-landmarks in the GPA,  the semi-landmarks along the anterior loop curve were constrained to slide along an estimated tangent at each sliding point using the bending energy method (Bookstein 1997). The overall size parameter for the M 1 is centroid size (square root of the sum of squared distances of landmarks and semi-landmarks from the centroid) and comparisons between populations use a box plot of logtransformed values produced with 'R' v. 2.13.1 (R Development Core Team 2011). The shape variables are the Procrustes coordinates obtained after the GPA, and variation among populations is displayed by principal component analysis (PCA) using MorphoJ 1.05c (Klingenberg 2011).

Cytochrome b sequencing (modern DNA)
A total of 283 specimens of M. arvalis were sequenced from 18 localities in Orkney and 50 localities in continental Europe (particularly from around the potential area of human introduction: northern France and Belgium and inland from there). Total genomic DNA was isolated using the DNeasy Tissue Kit (Qiagen, Halden, Germany) and PCR-amplified using previously described primers (Table S4, Supporting information; Jaarola et al. 2004). PCR products were purified with QIAquick PCR purification kit (Qiagen) and commercially sequenced using BigDye Terminator Sequencing chemistry (Applied Biosystems, Foster City, CA, USA) with newly designed primers (Table S4, Supporting information) and run on ABI PRISM 3730xl sequencers (Applied Biosystems). Complete cytb sequences (1143 bp) were generated (GU190383-GU190665).

Cytochrome b sequencing (ancient DNA)
All aDNA extractions were performed in a laboratory in Durham where no modern molecular biology or post-PCR work is undertaken and where Microtus were analysed for the first time. Before use, all materials and work surfaces were wiped with 10% bleach, and the workspace was UV-irradiated overnight. Samples of Microtus bone were excised using a scalpel blade and crushed within aluminium foil. The bone powder was incubated overnight on a rotator at 55°C in 500 lL of extraction buffer (0.5M pH 8.0 EDTA, 0.1M pH 8.0 Tris, 0.05% w/v SDS) with 8 lL of proteinase K (0.3 mg/ mL). Digested samples were then extracted using the Qiagen QIAquick PCR purification method, as described in Nichols et al. (2007). Final elutions of aDNA were collected in 50 lL of TE buffer following the QIAquick protocol and stored at À20°C. Negative extraction controls (lacking bone powder) were also performed in parallel in a ratio of approximately 1:10.
Ancient DNA was successfully obtained from 37 specimens (of 190 attempted) from Neolithic to Viking  (Table S4, Supporting information), with the missing cytb sequence at the 5′-end. Each PCR used 2 lL of aDNA extract in a 25 lL volume with Hi-Fidelity Platinum Taq (Invitrogen, Carlsbad, CA, USA). Purification and sequencing of PCR products followed the protocols used for modern DNA. As further checks on the aDNA authenticity additional to the negative extraction and PCR controls, two samples were internally replicated from bone powder (R59 and R60), producing identical sequences. All sequences with unique mutations not found in modern Orkney voles were amplified at least twice for confirmation. The regions of overlap between the PCR products were checked and found to be consistent. The assembled cytb coding regions for all ancient voles produced a readable protein sequence with no frameshift or nonsense mutations. M. arvalis is known to possess a nuclear copy of cytb (DeWoody et al. 1999) that is easily identifiable due to unique fixed mutations and insertions and was not amplified from any of the aDNA extracts.
Six samples (R44, R58, R125, R131, R189 and R191) were sent to Egham where the entire procedure, including extraction and amplification, was replicated. Two samples (R125 and R131) were sequenced in their entirety and produced identical sequences to those in Durham. Radiocarbon dating and high collagen levels found in 19 samples from Orkney that had been successful for aDNA extraction give corroborative evidence for good biochemical preservation of the samples.  (Shenbrot & Krasnov 2005) where three obscurus localities are off the map to the east. Black arrow shows the most likely source locality of the Orkney voles, based on microsatellite data (see Table 1). sequencer (Applied Biosystems), and their lengths determined using GENEMAPPER 3.7 (Applied Biosystems).

Microsatellites
For each locus and each population, tests for departure from Hardy-Weinberg equilibrium were performed with ARLEQUIN 3.12 , and significance levels were corrected for multiple testing using the sequential Bonferroni-Holm procedure (Rice 1989). There were no significant departures from equilibrium. Mean allelic richness over loci (A R ) was calculated with FSTAT version 2.9.3.2 (Goudet 1995) and the mean heterozygosity (H) over loci and population differentiation (F ST ) with Arlequin.
A Bayesian phylogenetic tree of all modern sequences with the complete (1143 bp) cytb was estimated in MrBayes 3.1.2 (Ronquist & Huelsenbeck 2003). To avoid long-tree artefact (Marshall 2010), the Γ-distribution shape parameter a was fixed to 1.0736. The value was obtained from the GTR+Γ substitution model selected by Akaike information criterion in MrModeltest 2.3 (Posada & Crandall 1998;Nylander 2004). The two separate Markov chain Monte Carlo (MCMC) analyses each comprised one cold and eleven heated chains. Samples were drawn from the posterior every 1000 steps over 10 million steps. The chain temperature was 0.06, and two chain swaps were attempted every step to optimize mixing. The first 3000 sampled trees were discarded as burn-in, with the resulting average standard deviation of split frequencies equal to 0.007. The 95% credibility interval of the tree length included a maximum-likelihood estimate of the tree length obtained from RAXML 7.2 (Stamatakis 2006).
Median-joining networks were constructed in Network 4.2 (Bandelt et al. 1999) with an equal transition/transversion ratio. The analysis was carried out using the 1130-bp alignment because aDNA sequences were included.
Radiocarbon-dated and modern M. arvalis cytb sequences were analysed using the program BEAST 1.5.1 , utilizing the 1130-bp alignment because of inclusion of aDNA sequences. For the Orkney data set, the GTR+Γ model of nucleotide substitution was selected using the Akaike information criterion. All BEAST analyses used this model, along with a strict molecular clock. A comparison using Bayes factors selected the constant-size coalescent prior as the most appropriate demographic model. With ancient sequences in the data set, the potential for undetected sequence errors to influence the analyses was also modelled in BEAST (Rambaut et al. 2009). All other parameters were co-estimated with the phylogeny, with samples drawn from the posterior every 1000 steps over a total of 10 million steps. The first 1 000 000 steps were discarded as burn-in. Acceptable mixing and convergence to stationarity were checked using the program TRACER 1.4.1 .
For modern DNA sequences from continental European populations, we used a demographic model simulating population expansion and used Bayes factors to compare a strict molecular clock, uncorrelated lognormal relaxed clock and uncorrelated exponential relaxed clock (Drummond et al. 2006). A model of exponential growth was identified as the best-fitting demographic model using Bayes factors. Samples were drawn from the Markov chain every 10 000 steps over a total of 100 million steps, with the first 30% of sampled trees discarded as burn-in. Estimated effective sample sizes were >200. Mean mutation rate was fixed to that obtained from the tip-dated sequence analysis described above, and each lineage that was expected to exhibit unique demographic history in the target time frame was analysed separately.
For the specific question of time of colonization of Orkney, we used the program IMa (Hey & Nielsen 2007) to estimate the splitting time between the modern Orkney sample (N = 57 for mtDNA, N = 114 for microsatellite loci) and European mainland samples from the potential area of human introduction: northern France, Belgium and nearby areas of Germany (N = 46 for mtDNA, N = 92 for microsatellite loci). Initial runs were performed to estimate parameters, followed by three replicate runs with different random number seeds to check for consistency of results (results from the final run reported). Input data were 1143-bp cytb sequences (assuming the HKY mutation model and the mean mutation rate estimated in BEAST; priors on the range of mutation rate scalars were set one order of magnitude above and below) and 14 microsatellite loci. The inheritance scalar was set at 0.25. Metropolis coupling was implemented using 20 chains and a geometric heating model (term 1: 0.99; term 2: 0.95). The burn-in period was 7 h. Upper bounds were set for prior distributions for theta, migration rate and the splitting time. The final run saved 499 469 trees per locus and ran for 705 h (50 million steps following burn-in). Convergence was tested by ensuring that effective sample sizes exceeded 50 and parameter trend lines were flat.
We analysed the microsatellite data using approximate Bayesian computation (ABC) to estimate demographic parameters of the colonization history of Orkney by M. arvalis. Comparisons were made between the continental European and Mainland Orkney populations listed in Table 1 (and mapped in Fig. S1, Supporting information), except that the four continental populations with small sample sizes (15 individuals or fewer) were excluded. Again, we focused on populations from the region of continental Europe that most likely represented a source for the introduction. We specified an ABC model with 11 demographic parameters in which the Orkney population diverged from a continental population after a bottleneck. Moreover, both the Orkney and the continental populations were allowed to pass through independent bottlenecks. The bottleneck of the continental population could occur either before or after the colonization of Orkney. For simplicity, the model was simulated with instantaneous growth after all three bottleneck events. Further details are given in Data S1, Table S5 and Fig. S3 (Supporting information).

Characteristics of the Orkney voles
Among extensive samples of Microtus arvalis collected from across their European range, it is clear that the island populations on Orkney (seven samples) and Guernsey (one sample) are highly distinctive compared with their continental counterparts in first lower molar morphology (Fig. 1). This relates to both size (the island voles generally had larger teeth: Fig. 1e) and shape (the Orkney voles were divergent from continental and insular European M. arvalis in the principal components scatter plot, with the M 1 displaying a relatively broader anterior loop: Fig. 1c, d).
At the 14 microsatellite loci screened, the mean number of alleles per locus and heterozygosities were systematically lower in the Orkney populations than in northern France, Belgium and nearby areas of Germany (Table 1), the areas of continental Europe from where introduced voles most likely originated. There was no overlap in heterozygosity values and the small overlap in mean number of alleles per locus related to Mainland Orkney populations, which had distinctly higher diversity values than other Orkney Island populations.
Overall F ST between population samples was very high (0.382, P < 0.0001) in agreement with previous studies on the species (e.g. Heckel et al. 2005). Pairwise comparisons between Orkney populations ranged from F ST = 0.094 to 0.682 (mean: 0.412) and between 0.007 and 0.346 for the continental populations (mean: 0.177; Table S6, Supporting information).

Special features of the mitochondrial DNA variation
Together with previously published results, our new data confirm that M. arvalis of the same (Western-North) cytb lineage found today in France, Belgium, the Netherlands, Germany and Switzerland also occurs on Orkney ( Fig. 2 and Table S2, Supporting information). Fig. 2 shows very clearly that the coastline of France and Belgium provides possible maritime access of the Western-North cytb lineage to Orkney, consistent with our contention that this is the most likely source region for an introduction of M. arvalis there.
In the Bayesian phylogenetic tree (Fig. 2) and the median-joining network (Fig. 3), the Orkney cytb sequences form an unsupported monophyletic clade within the Western-North lineage. However, within that context, the sequences are very distinct from those from coastal Belgium/France and other regions in continental Europe, separated by at least four mutations. Thus, despite detailed sampling of the Western-North lineage in general and along the coast of France and Belgium in particular (Fig. 2), we found no cytb sequences there that are clearly ancestral to those of the Orkney voles.
Instead, over this coastal region, an area of more than 10 000 km 2 (Table 3), a single mtDNA haplotype (and sequences separated by one mutation from it) is dominant. None of these haplotypes relate closely to the Orkney M. arvalis haplotypes (Fig. 3). This 'starburst' of the coastal French and Belgian sequences suggests their recent expansion and dispersal. A significant deviation from neutral expectation for Tajima's D and Fu's Fs, a small Rozas's R 2 and a significant signal in pairwise DNA sequence comparisons in the mismatch distribution are consistent with this (Table 3).
Considering now the Orkney sequences, the same cytb lineage and two of the same haplotypes as those in modern specimens were also found in M. arvalis dating back to the Neolithic period ( Fig. 3; Table S3, Supporting information). Interestingly, the central Orkney haplotype found in archaeological specimens from Mainland Orkney and Westray was detected in modern specimens from Westray and Burray not Mainland Orkney. A second haplotype was found in both archaeological and modern sequences from Mainland Orkney.
Both archaeological and modern sequences from Mainland Orkney are represented by many haplotypes in the network (9 and 12, respectively), and there are five archaeological and two modern haplotypes from Westray (Fig. 3).   Fig. 2. Node size is proportional to haplotype frequency and edge length to number of substitutions separating the haplotypes. Guernsey is an offshore island also analysed in the geometric morphometric analysis (see Fig. 1). For neutrality test statistics and the mismatch distribution, significant values (P < 0.01) consistent with population expansion are given in bold. *This included all localities within 100 km of the coast. The area sampled was estimated conservatively as a rectangle with one side given by the distance between the end localities along the coast and the other by the mean distance to the coast of all localities.

A possible source area for the introduction of M. arvalis to Orkney
We used ABC on the microsatellite data to compare 12 current populations (both coastal and inland) from northern France, Belgium and nearby areas of Germany as possible source populations for Orkney voles (Table 1 and Fig. S1, Supporting information). Our results confirmed that, if M. arvalis were introduced to Orkney from this source region, they most likely originated from coastal Belgium/France (three of the four best-supported populations: Table 1) rather than inland continental populations. The best-supported modern source population is from Stalhille on the Belgian coast ( Fig. 2 and Table 1) and geographically the closest point to Orkney within the distribution of the Western-North lineage.

Time of colonization of Orkney
We obtained new 14 C dates for 23 specimens from archaeological sites dating from the Late Neolithic period to the Viking Age ( Table 2). The estimated dates are consistent with those of the archaeological contexts from which their bones were recovered. They show definitively that M. arvalis was present on Orkney at least 5100 BP. Two molecular analyses suggest that they did not arrive much earlier than this. An ABC analysis of microsatellites shows modal point estimates of time of colonization of Orkney that vary between 3000 and 5700 BP (between Stalhille, Belgian coast and three localities on Mainland Orkney). An IMa analysis (Hey & Nielsen 2007) of ancient and modern mtDNA sequences and microsatellites yields an estimated colonization time of 4515 years (90% CI: 2568-9019). Both analyses assume a single generation per year. Multiple generations (for which there is no evidence: Gorman & Reynolds 2008) would imply a more recent post-Neolithic colonization, which does not fit well with our new 14 C dates (voles on Orkney at latest 5100 BP) and aDNA sequencing of archaeological specimens (same genetic lineage as the modern specimens). While the radiocarbon and molecular dating indicate that voles arrived on Orkney about 5000 BP, the time to the most recent common ancestor (tMRCA) for the Orkney cytb sequences is much earlier, c. 15 400 years (95% CI of 8100-25 400 years), based on Bayesian coalescent analysis incorporating modern and ancient DNA data. Using the same mutation rate (l = 3.27 9 10 À7 mutations/site/year) as for Orkney sequences, the tMRCA for the sequences from coastal France/Belgium (the putative source area for the Orkney voles) is only 2356 years (95% CI of 723-4941 years). Mismatch analysis (Rogers & Harpending 1992) provides further evidence of the recentness of the derivation of the cytb sequences in this potential source area for the Orkney voles, showing an expansion beginning 1860 BP (95% CI: 1201-2665).

Cytb mutation rate
Estimates of mutation rates for phylogeographical studies have traditionally been based on fossil calibrations, often founded on very incomplete information. For Microtus, Triant and DeWoody (2006) used a fossil calibration point of 1.5 Ma to provide a mutation rate for cytb of approximately 8 9 10 À8 mutations/site/year. Although this is probably the best-substantiated fossilbased estimate of the cytb mutation rate for Microtus in the literature, we considered it inadequate for our needs. This is because the fossil calibration point likely has errors (it is not based on precisely dated geological events or fossil transitions) and is geologically very much earlier than the time frame for our study. Instead, we were able to calibrate our rate estimate using the aDNA sequences from radiocarbon-dated specimens as part of a Bayesian coalescent analysis. This method has the advantage of producing an estimate of the cytb mutation rate relevant to precisely the population that we were studying and the temporal scale of interest.
The mutation rate we obtained and used here (3.27 9 10 À7 mutations/site/year) was about four times higher than that of Triant and DeWoody (2006), but comparable to other aDNA-based estimates of mutation rate for a variety of species (Ho et al. 2011). Navascu es & Emerson (2009) caution against possible upward bias of such aDNA-based estimates under extreme demographic scenarios, but there is a growing consensus in the literature that aDNA-based estimates are better than traditional fossil-based mutation rates, which are too low for dating divergences and demographic events that have occurred over the last few thousand years (Ho & Larson 2006;de Bruyn et al. 2011). With regard to our aDNA-based cytb mutation rate for M. arvalis, it is extremely close to that obtained for another Microtus, which used a very recent geological event as a calibration point (M. agrestis; 3.89 9 10 À7 mutations/site/year; Herman & Searle 2011). This provides independent support for our aDNA-based estimate, although, as with all molecular dating, interpretations have to be viewed with caution.

Colonization history of Orkney voles
It is striking that there is substantial cytb variation in M. arvalis in Mainland Orkney and over the whole archipelago (Fig. 3, Table 3). Island populations often show low genetic diversity (Frankham 1997). This can relate to small population sizes and/or population bottlenecks associated with colonization of islands, particularly by sweepstake dispersal or human introduction. The high cytb diversity could indicate that the Orkney population of M. arvalis represents an island relict of a previously continuous mainland population, perhaps dating back to before the LGM (Beirne 1952). This would fit with the long tMRCA for the molecular variation on Orkney, potentially dating back to 25 400 BP (within the 95% CI).
However, there are strong arguments against glacial survival of the Orkney vole population. First, the IMa analysis based on microsatellites and cytb and the ABC analysis based on microsatellites provide a date of arrival around 5000 BP, considerably more recent than the LGM. Second, all the other species of small mammals on Orkney are most reasonably viewed as human introductions (Yalden 1982), necessitating a special case for M. arvalis as a glacial survivor. Third, it is very difficult to make this special case given that M. arvalis is not a species currently associated with arctic or even moderately high latitude conditions. Its range extends eastward beyond Lake Baikal and yet barely traverses north of the 60th parallel ( Fig. 2; Shenbrot & Krasnov 2005). Orkney was under or near a glacial ice sheet at the LGM (Bowen et al. 2002) and M. arvalis is not part of the fossil fauna known from Britain from the last glacial period (Yalden 1999;Currant & Jacobi 2001). Fourth, M. arvalis is not currently found in Britain (Fig. 2). It is therefore contrary to think that M. arvalis should be a glacial relict on Orkney rather than M. agrestis, when the latter occurs further north in Eurasia (beyond the 70th parallel) and is distributed throughout Britain, including on many offshore islands, while M. arvalis only occurs on Orkney. There have been no land connections between mainland Britain and Orkney after conditions ameliorated following the LGM (Yalden 1982), and hence why M. agrestis (and other wide-ranging small mammals in Britain, such as common shrews Sorex araneus and bank voles Myodes glareolus) failed to colonize Orkney. If M. arvalis did not survive on Orkney itself during the last glacial period, the absence of the species in Britain means there are no grounds to suggest sweepstake colonization from there. It is conceivable that there could have been sweepstake colonization of Orkney from Doggerland, the landmass connecting Britain, the Low Countries and Denmark until about 8000 BP (Weninger et al. 2008)but this would require the survival of small mammals on floating mats of vegetation over a substantial marine gap between Doggerland and Orkney.
Thus, human introduction is by far the most likely explanation for the occurrence of M. arvalis on Orkney.
From the IMa and ABC dates, this introduction at about 5000 BP fits well with the earliest radiocarbon dates for archaeological M. arvalis from Neolithic contexts (5100 years old: Table 2) and the beginnings of the Neolithic culture on Orkney (5600 BP: Ritchie 2001;Schulting et al. 2010). Voles could have been brought to Orkney by Mesolithic hunter-gatherers, as early as c. 9000 BP, but no vole remains have been found in the one excavated Mesolithic site on Orkney (Lee & Woodward 2009), in contrast to their abundance at Neolithic and later sites (Yalden 1999;Thaw et al. 2004).
If, as appears most likely, the voles were introduced by Neolithic settlers about 5000 BP, various other implications flow from our molecular data, which are of considerable archaeological interest.
First, the introduction implies long-distance maritime travel by Neolithic people between continental Europe and Orkney, extending on findings from elsewhere (e.g. Broodbank 2006). Our study highlights the Belgian coastline as the most reasonable source of the Orkney voles on the basis of available genetic data. This suggests Neolithic cultural linkages between Belgium and Orkney, of worthwhile focus for future archaeological investigation. Microtus arvalis were not introduced successfully into mainland Britain, which is consistent with relatively direct transport to Orkney from the continental source area.
Second, if the introduction occurred about 5000 BP, then, because the tMRCA for the Orkney voles is so long (15 400 years), substantial numbers of female voles must have been introduced to explain the cytb variation observed in modern and archaeological Orkney voles. High genetic diversity is already evident in the 16 aDNA sequences dating to 4200 BP or earlier, separated by up to 10 mutations ( Fig. 3 and Table S3, Supporting information) and which produce an estimate for the tMRCA (14 780 years; 95% CI of 4681-36 379 years) similar to that of the full ancient and modern data set.
To explain a substantial number of voles arriving accidentally on Orkney implies transport of plentiful grass livestock bedding/fodder in which the vole stowaways could have survived. This in turn may suggest the direct movement of livestock as part of the proposed Neolithic linkage between Belgium and Orkney.
Alternatively, deliberate transport of voles onto Orkney could explain the large numbers introduced (Thaw et al. 2004). It is conceivable that voles were taken as food items, pets or for cultural/religious purposes -M. arvalis is docile in captivity (Berry 2000), so could theoretically have been 'tamed'. This suggestion of deliberate transportation of small rodents has a precedent: it has been argued that Pacific rats (Rattus exulans), now present on islands throughout Oceania, were intentionally conveyed by Polynesians as a food source (Matisoo-Smith & Robins 2004).

Evolutionary processes affecting voles on Orkney and the continental source area
Compared with continental European M. arvalis, those on Orkney and another offshore island (Guernsey) are divergent in terms of tooth morphology, including increased tooth size. For Orkney, this divergence may have occurred over c. 5000 years, if introduced during the Neolithic. In addition to having larger teeth, the Orkney and Guernsey M. arvalis have a larger body size than continental voles (Gorman & Reynolds 2008). Quick-evolving rodent gigantism has been described previously on islands of the northeast Atlantic (Corbet 1961;Angerbjorn 1986), but not within such a precisely defined time frame. A range of selective factors have been proposed to explain this gigantism, including an absence of small mammalian predators (Lomolino 1985), and a genetic basis for gigantism has been identified in island house mice (Chan et al. 2012). Elsewhere, we further explore the dynamics of morphological evolution for the Orkney M. arvalis using archaeological specimens (T. Cucchi, R. Barnett, N. Mart ınkov a, et al. submitted) extending substantially on previous studies (Berry & Rose 1975;Corbet 1986).
Despite their similarity in large tooth and body size, Guernsey and Orkney voles exhibit distinctive mtDNA haplotypes (Fig. 3). It is therefore most reasonable to consider that the Guernsey and Orkney voles attained their large body size independently. It is not clear whether Guernsey was colonized naturally before it became an island, as part of the continental European late glacial/ postglacial species expansion (Haynes et al. 2003;Heckel et al. 2005;Tougard et al. 2008), or whether the voles were introduced by people after it became an island (Gorman & Reynolds 2008). However, given that Guernsey voles are likely to come from the same general (northern France/Belgium) source area as the Orkney voles and that they are also different in mtDNA from current northern France/Belgium populations, there would be much interest in further detailed comparison of Orkney, Guernsey and northern France/Belgium voles.
In addition to the operation of selection in the evolution of Orkney voles suggested by morphology, stochastic processes appear to have been important based on microsatellites. The population on the largest island, Mainland Orkney, has retained much of the microsatellite variation found in continental Europe, while all the other Orkney Islands (which are considerably smaller: Fig. 1) show very low levels of microsatellite variation, consistent with founder events and genetic drift. Similar stochastic processes can also explain microsatellite vari-ation among Scottish Island populations of common shrew (White & Searle 2007a).
Our findings with regard to morphology and microsatellites in M. arvalis are unsurprising in comparison with previous studies on island small mammals, but the results from our mtDNA analyses are more unexpected. Although the cytb sequences from Orkney and the proposed source area for the Orkney colonization both belong to the Western-North lineage of M. arvalis, the sequences are remarkably divergent given the time frame for colonization. Also, it might have been expected that (as for the microsatellites) variability would have been lower on Orkney than in continental Europe. In fact, the opposite is the case. Taking either the principal island (Mainland Orkney) or the whole archipelago, mtDNA diversity is higher in Orkney than in coastal France/Belgium (Table 3). Our dating analysis also shows that the mtDNA sequences in coastal France/Belgium have a much more recent derivation than the Orkney sequences.
So, here we are seeing another facet of evolution in association with the colonization of offshore islands, in this case occurring in the mainland population. The presence of derived sequences in coastal France/Belgium suggests a replacement event in M. arvalis, with one mtDNA type (the current type) replacing another (the Orkney type), similar to aDNA findings in other species (Barnes et al. 2002;Pergams et al. 2003;Hofreiter et al. 2007). The fact that there is an affiliation between coastal Belgium and Orkney on the basis of microsatellite genotypes argues against a complete population replacement (e.g. by extinction-recolonization) as an explanation for the mtDNA result. Instead, within-population processes of selective sweeps or genetic drift are implicated, and more likely expressed in the mtDNA data, as a single locus with small effective population size than in the microsatellite data. We cannot be sure what environmental factors promoted the replacement. There could, for instance, have been a local, unrecorded disease outbreak. However, it is notable that the replacement occurred over a period when M. arvalis populations would have changed dramatically due to human land-use change, and this appears the most likely driver of the replacement. Over several thousand years, sustained forest clearance in continental Europe (Rackham 1998;Cyprien et al. 2004) created new agricultural habitats and associated selection pressures that essentially expanded the opportunities for M. arvalis as a species that particularly exploits managed grassland (Niethammer & Krapp 1982). In such a habitat, M. arvalis populations can undergo massive population expansions and crashes (Delattre et al. 1992) that reduce longterm effective population size, promoting genetic change through drift. On Orkney (which saw the rapid decline of low shrubs and tree species with the arrival of Neolithic farmers: Bunting 1996), M. arvalis utilizes a range of open habitats and does not show the same dramatic population fluctuations as seen in parts of continental Europe (Gorman & Reynolds 2008).

Offshore islands as field laboratories
There has been a tendency to view offshore island populations of small mammals (and other organisms with low density and low dispersal) as genetic deviants from the 'norm'. This is because studies of various species have shown results similar to ours for morphology (substantial change) and microsatellites (loss of variation) (Lomolino 1985;Frankham 1997;Boessenkool et al. 2007;Millien 2011). These have included detailed studies on small mammals such as wood mouse Apodemus sylvaticus (Angerbjorn 1986;Michaux et al. 1996), masked shrew Sorex cinereus (Stewart & Baker 1992) and common shrew (White & Searle 2007a. However, as we have demonstrated with our M. arvalis mtDNA studies, island populations can also represent genetic 'arks', retaining the ancestral genetic variation, while evolutionary and other processes on the mainland may lead to a loss of that ancestral variation. Islands may have importance therefore in conservation of genetic variation. A further example involving human introduction of a small mammal onto an offshore island is provided by the Eurasian red squirrel Sciurus vulgaris. Thus, Irish red squirrels have genetic variants that apparently derive by introduction from Britain, but these are now absent in that source population (Finnegan et al. 2008;Searle 2008). For low density and low dispersal organisms such as small mammals, we suggest that genetic surveys of mainland areas should, where available, include populations from neighbouring offshore islands. It is very likely that those island populations will provide a new perspective on the temporal and spatial dynamics of the genetic variation in that region.
The 'ark' concept that we discuss here is of course more general. Populations colonizing new areas will take the genetic and nongenetic characteristics of the source population, and some of those characteristics may subsequently be lost in the source population but retained in the population in the new area. In this way, for instance, the United States is a 'linguistic ark' for various English words that would have been common in the British Isles at the time of settlement of North America by the British, but which have subsequently fallen into disuse in the homeland (e.g. 'fall' meaning 'autumn').
Returning to genetic characteristics of offshore islands, in addition to their potential as genetic 'arks', they also hold potential as field laboratories to study genetic change in the islands themselves. Compared with the classic evolutionary studies on oceanic islands, those based on offshore islands will tend to view events over shorter timescales and thus provide a different perspective on evolutionary processes. Offshore islands are particularly valuable for studying initial stages of diversification, with the opportunity (as in the current study together with T. Cucchi, R. Barnett, N. Mart ınkov a, et al. submitted) to follow island populations from their foundation to the present day using advanced genetic and morphometric tools as applied to modern and ancient populations of different ages. Extremely accurate dating of ancient populations may be possible (e.g. in archaeological settings). With this short time duration and close proximity to the mainland, there is also a greater chance to find the precise source area for the island colonization, which allows interesting comparison of evolutionary processes on the mainland and island. This brings us back to the value of offshore island populations in interpreting mainland processes. Offshore islands are an underutilized resource for evolutionary analysis, with great potential. In some ways, they represent study systems intermediate between those in a continental setting and those on oceanic islands; they have the simplicity of the oceanic island system yet are clearly relevant to continental situations.

Data accessibility
The DNA sequences have been deposited in GenBank (GU190383-GU190665). The TreeBASE entry for the phylogenetic tree of cytb sequences shown in Fig. 2, the microsatellite genotypes, the input file for the IMa analysis, the modern and ancient cytb sequences, and the morphological coordinates collected from 2D images of skulls are all available through DRYAD: doi:10.5061/dryad.9rf5m.

Supporting information
Additional supporting information may be found in the online version of this article.
Data S1 Material and Methods.
Table S1 List of modern M. arvalis samples used for morphometrics, including country and site of origin.

Table S2
List of all modern M. arvalis used for cytb analysis and collected for this study, including country and site of origin; arranged according to cytb haplotype.

Table S3
List of all ancient specimens of M. arvalis that successfully provided a cytb sequence with details of location collected, calibrated age range (where obtained) and GenBank Accession Number for the sequence.

Table S4
List of primers used for the amplification of cytochrome b from M. arvalis.

Table S5
Prior distributions of the ABC model parameters (as illustrated in Fig. S3).

Table S6
Pairwise F ST values between population samples analysed with microsatellites (see Table 1 and Fig. S1).