Quaternary climate cycles played an important role in promoting diversification across the Northern Hemisphere, although details of the mechanisms driving evolutionary change are still poorly resolved. In a comparative phylogeographical framework, we investigate temporal, spatial and ecological components of evolution within a suite of Holarctic small mammals. We test a hypothesis of simultaneous divergence among multiple taxon pairs, investigating time to coalescence and demographic change for each taxon in response to a combination of climate and geography.
Beringia, the nexus of the northern continents.
We used approximate Bayesian computation methods to test for simultaneous divergence among eight pairs of taxa, using cytochrome b gene sequences. We calculated coalescence times for eastern and western components of each pair and for the combined pairs, and relate dates to Quaternary climatic periodicity and combinations of environmental events and physical barriers. Population growth and expansion statistics were used to test evolutionary responses among taxa, including range shifts, persistence or periodic extirpation. Species distribution models (SDMs) for each taxon were used to predict their geographical ranges during the present interglacial, Last Glacial Maximum and previous interglacial.
Multiple divergence events across Beringia were primarily coincident with extreme glacial cycles of the late Quaternary. Structure within Beringia is spatially consistent with at least three environmental barriers arising at different times: the Kolyma Uplands, Bering Strait and portions of the Bering Isthmus. Levels of divergence varied substantially, indicating evolutionary processes spanning deep and shallow time-scales. The different demographics among taxa reflect their distinct ecological responses. SDMs predicted regional distributional changes through time and different spatial responses among taxa.
Beringia predominantly constituted a dispersal corridor during the early Quaternary and a major centre of endemism in the late Quaternary. Coincident with severe glacial cycles, small mammal species were ‘caught’ in Beringia and diversified over multiple climatic phases. Relative genetic differentiation across Beringia appears to be related to ecological differences reflecting a gradual adaptation to Beringian environments through time. Some methodological constraints associated with resolving recent (late Quaternary) isolation events or drawing inferences from a single locus are discussed.
Beringia, the vast region spanning eastern Siberia and north-western North America, was first described by Hultén (1937) as a biotic refugium and distinct biogeographical region. Subsequently, Beringia was recognized as an important corridor for transcontinental movement between Asia and North America, a potential centre of endemism, a source for postglacial recolonization, and a region of phylogeographical suture zones that delineate distinct clades within and among species (Abbott & Brochmann, 2003; Cook et al., 2005; Waltari et al., 2007a; DeChaine, 2008). The Beringian Isthmus enabled species to move in pulses and subsequently diverge across the Holarctic (Hoberg & Brooks, 2008, 2010), and yet the spatial and temporal components of these dynamics within Beringia remain poorly understood.
From the early recognition and consideration of Beringia as a large and relatively homogeneous single refugium (Hopkins, 1967; Hopkins et al., 1982) during glacial episodes of the Quaternary (2.6 Ma–present), a more refined view is emerging that Beringia was a shifting mosaic of regional palaeoenvironments through time (Elias et al., 1997; Edwards et al., 2000; Guthrie, 2001; Hoberg & Brooks, 2008). This heterogeneous and dynamic area produced fine-scale population structure (e.g. Hoffmann, 1981; Elias, 1992) and complex community dynamics reflecting responses to high climatic variability even within glacial cycles (Stewart & Lister, 2001; Provan & Bennett, 2008; Hofreiter & Stewart, 2009). Instead of absolute ice-bound isolation through a glacial phase, populations within Beringia may have experienced intermittent local habitat fragmentation and reconnection accompanied by periodic instances of gene flow in response to shifting local climate conditions (DeChaine, 2008; Elias & Crocker, 2008; Shafer et al., 2010). In addition to ephemeral determinants of population structure, the heterogeneous geography in and surrounding Beringia included more persistent barriers to dispersal during both glacial maxima and minima (Cook et al., 2005). The Cordilleran and Laurentide ice sheets delineated the eastern boundary during maxima and extensive mountain ranges and river systems dissected the area. Finally, non-volant terrestrial species were intermittently isolated across Beringia by the Bering Sea during interglacial phases (Sher, 1999). Multiple cycles of isolation and reconnection across major barriers provided repeated opportunities for diversification, but was the periodicity of climatic phases sufficiently long to both initiate and maintain significant (species-level) evolutionary divergence across Beringia during the Quaternary? Was one geographical barrier, climatic event or glacial cycle more important than others in driving divergence? Finally, did a shifting ecological mosaic through time result in one or multiple responses of species to common environmental processes in the region?
Repenning (2001) used empirical evidence from fossil rodent faunas to summarize many of the major evolutionary processes in response to climatic and geophysical changes that have shaped Holarctic diversification through Beringia. The Quaternary is exemplified by a cool climate with repeated cycles of warmer and colder phases brought about by orbital processes (reviewed in Elkibbi & Rial, 2001). Beginning in the middle Quaternary (c. 1 Ma) glacial cycles transitioned from weaker amplitude and shorter periodicity (c. 41 kyr) to stronger amplitude and longer periodicity (c. 100 kyr). Movement through Beringia in the early to middle Quaternary (2.6–1.2 Ma) proceeded via multiple waves of dispersal coincident with the initial onset of the cyclic glacial climate regime and subsequent weak glaciations (41 kyr periodicity). Dispersal of predominantly temperate species during this period was further facilitated by warm climatic conditions on the exposed Bering Isthmus due to a Pacific Ocean heat sink (Repenning, 2001). Few dispersal events are attributed to the stronger glacial cycles of the late Quaternary (100-kyr periodicity) due to the formidable barriers imposed by continental ice sheets or the possibly harsher Beringian climate, limiting dispersal at these times to early glacial onset (Repenning, 2001). This important assessment provides a number of predictions from which we form and test the following hypotheses:
Beringia has constituted a significant barrier resulting in the divergence of multiple pairs of taxa on (H1a) a single occasion (i.e. simultaneous), or (H1b) numerous occasions over the Quaternary. Although multiple dispersal events through Beringia are inferred from fossil evidence (Repenning, 2001), this does not specifically suggest that divergences occurred within Beringia on multiple occasions. Numerous extant pairs of taxa spanning the Holarctic allow us to test this hypothesis.
Divergence events between pairs of taxa are coincident with (H2a) weaker and shorter glacial cycles of the early to middle Quaternary, or (H2b) stronger and longer cycles of the late Quaternary. Repenning (2001) concluded that weaker glacial cycles resulted in a temperate Bering Isthmus without impenetrable North American continental ice sheets, allowing species to move through Beringia and spread over the northern continents. Major glaciations of the late Quaternary provided not only an ice barrier on the eastern edge of Beringia but also significant regional barriers (Galbreath & Cook, 2004) and climatic gradients (Guthrie, 2001; Elias & Crocker, 2008) within Beringia. Taxa ‘caught’ in Beringia during these more extreme, and longer, glacial cycles may have experienced population fragmentation and the associated evolutionary divergence across this region, possibly accompanied by adaptation to regional conditions.
Within the larger Beringian realm, significant divergence occurred across (H3a) the Bering Strait during shorter interglacial phases, or (H3b) some other physical or environmental feature during extended glacial maxima. On an evolutionary time-scale, even the glacial periodicity during the late Quaternary of c. 100 kyr is relatively brief. The Bering Strait was an absolute barrier to terrestrial mammals on multiple occasions, but only for short periods (c. 20 kyr).
H4: Ecological response
All small mammal taxa have (H4a) the same response, or (H4b) different responses to common environmental processes. Species with different ecological habitat associations and climatic tolerances should exhibit different demographic responses to a given climate, resulting in shifting communities through time and alternately favouring temperate species at different times or within different regions of Beringia than temperate–subarctic or arctic species (for detailed definitions of these ecological associations, see Repenning, 2001).
We test these hypotheses using genetic evidence and species distribution models (SDMs) for 11 small mammal taxa (eight taxon pairs) currently distributed across the northern continents. Multiple individual phylogeographical assessments provide evidence of variable evolutionary responses within this suite of Holarctic taxa. By standardizing population-demographic, coalescent-simulation and modelling approaches, we provide a comparative assessment of temporal, spatial and ecological components of evolution within the resident small mammal community of Beringia.
Materials and methods
For full details of all methods see Appendix S1 in Supporting Information.
Study system, taxa and sampling
We follow Hultén (1937) in delimiting Quaternary Beringia to between 125° E and 130° W and from 45°50′ N north to the Arctic Ocean. This area lies roughly between the Lena River in Siberia and the Mackenzie River in western Canada and extends southward to the Kurile Islands, Aleutian Islands, northernmost Southeast Alaska and north-western British Columbia.
We investigated amphi-Beringian rodents (six species) and shrews (five species; Table 1). Populations represent intraspecific or interspecific taxon pairs according to current taxonomy (Wilson & Reeder, 2005) and are split into two groups based on evolutionary associations:
Table 1. Summary of taxon pairs including west and east populations across Beringia, sample sizes (n), length of gene sequences for the mitochondrial cytochrome b gene, and previous literature
Fredga et al., 1999; Fedorov & Goropashnaya, 1999; Ehrich et al., 2000; Wickström et al., 2001; Fedorov & Stenseth, 2002; Fedorov et al., 2003; Smirnov & Fedorov, 2003
Chernyavsky et al., 1993; Fredga et al., 1999; Chernyavsky & Kartavtseva, 1999; Fedorov et al., 2003
Lance & Cook, 1998; Brunhoff et al., 2003; Frisman et al., 2003; Galbreath & Cook, 2004; Iwasa et al., 2009
These taxon pairs – two interspecific: Lemmus sibiricus and Lemmus trimucronatus, and Dicrostonyx torquatus and Dicrostonyx groenlandicus; and two intraspecific: Microtus oeconomus and Myodes rutilus – exhibit high levels of genetic differentiation across Beringia. All except Dicrostonyx have a phylogeographical break in the region of the Kolyma and Omolon rivers, separating a western geographical clade that stretches into central Siberia from an eastern geographical clade that spans the Bering Strait into North America (Fig. 1a).
These taxon pairs – two interspecific: Sorex portenkoi and Sorex ugyunak, and Sorex camtschatica and S. ugyunak; and two intraspecific: Sorex minutissimus and Sorex tundrensis – show isolation across the Bering Strait. Sorex ugyunak is recognized as the sister species of either S. portenkoi or S. camtschatica (van Zyll de Jong, 1982), and although these species are minimally diverged from each other (Hope et al., 2012), we retain two separate taxon pairs (Hutterer, 2005).
Mitochondrial cytochrome b (cyt b) sequences were retrieved from GenBank (see Appendix S2). Data sets for Myodes rutilus, S. camtschatica, S. ugyunak and S. portenkoi were supplemented by sequencing additional specimens. Tests of neutral evolution (McDonald & Kreitman, 1991) and likelihood ratio tests (LRT) for adherence to a molecular clock (Felsenstein, 1981) were performed for each data set.
Timing of divergence
We used the program beast 1.6 (Drummond & Rambaut, 2007) to estimate the timing of divergence, based on coalescence time to most recent common ancestor (TMRCA) per designated group, applying an assumed mutation rate for each group of taxa. Closely related taxa generally have similar rates of mutation (Kumar & Subramanian, 2002) so we assumed a common rate for rodents and another for shrews. We used an average mutation rate for rodents of 4% Myr−1 based on other studies (discussed in Brunhoff et al., 2003) and for shrews we used a mutation rate estimated for Sorex minutissimus of 5.5% Myr−1 (Hope et al., 2010).
We grouped individuals as ‘western’ or ‘eastern’ for each taxon pair, set substitution models as determined by MrModeltest 2.3 (Nylander, 2004), and applied a coalescent tree prior that assumed exponential growth and the appropriate mutation rate, using a relaxed uncorrelated lognormal clock model for all analyses. All analyses included 100 million generations (sampling every 1000). We calculated TMRCA with 95% posterior probability distributions using Tracer 1.5 (Rambaut & Drummond, 2007).
To assess changes in population size for each taxon, we generated Bayesian skyline plots (BSP) implemented in independent beast analyses using the Bayesian skyline coalescent tree prior (Drummond et al., 2005). For tests of demographic expansion, we used DnaSP 5 (Librado & Rozas, 2009) to calculate Tajima's D (Tajima, 1989) and Fu's FS (Fu, 1997). Nucleotide and haplotype diversity were calculated for each group using DnaSP to assess genetic variability.
Tests of simultaneous divergence
We tested a null hypothesis of simultaneous divergence under an approximate Bayesian computation (ABC) analytical framework (Beaumont et al., 2002; Hickerson & Meyer, 2008) using msBayes 20120510 (Hickerson et al., 2007; Huang et al., 2011). Although the precise physical features within our study area that led to evolutionary divergence between pairs of taxa are unknown, we know that Beringia encompasses suture zones for multiple taxa (Cook et al., 2005). We therefore designed divergence tests to evaluate the number of temporally distinct divergence events across Beringia rather than focusing on specific putative barriers.
We tested three separate groups to control for differences in estimated mutation rate between rodents and shrews. These included two separate analyses for rodent and shrew taxon pairs using respective mutation rates of 4% Myr−1 and 5.5% Myr−1, and an analysis of all eight pairs of taxa combined. The latter analysis was repeated once using the ratio of mutation rates for rodents and shrews as a relative standard rate (1.0:1.375) and once using a common mutation rate of 4% Myr−1 for all taxon pairs.
We summarized posterior distributions using estimates of Ω [var(τ)/E(τ); Bayesian posterior credible interval] and Ψ (number of divergence events) across taxon pairs. We assessed significance of estimates for number of divergences where Ω = 0 corresponds to simultaneous (= one) divergence and we could not reject this scenario if the 95% confidence interval (CI) of Ω encompassed 0. In addition, although we recognize that estimates of Ψmode (the number of divergence events observed most frequently) can be inaccurate, we retained Ψ for use in a Bayes factor (BF) test of Ψ = 1 against Ψ > 1, where BF < 0.33 is considered substantial evidence against a simultaneous divergence (Jeffreys' scale of interpretation; Jeffreys, 1961). For all analyses using rodent taxa we set maximum τ (average divergence) = 2 coalescent units (number of 2Ne generations, where Ne is the female effective population size) due to their relatively higher genetic diversity, and for shrews maximum τ = 1 coalescent unit. Individual analyses for rodent and shrew taxon pairs used π (nucleotide diversity), θw (Watterson's theta), πnet (Nei and Li's net nucleotide divergence), and the denominator of Tajima's D as summary statistics. The prior of the ancestral population size (Nanc) before the divergence was assumed to be uniformly distributed, and the maximum value of this distribution was determined empirically: we ran the simulations under three different maximum values (0.1, 0.25 or 0.5 of the current Ne), and selected the value which fitted the observation best using BF as a criterion. We ran 8 × 106 simulations, retaining in each analysis 500 simulations with best fit to the empirical data.
Species distribution models
To assess differences in the spatial ecology of our focal species and corresponding consequences for species response to historical climate change, we produced SDMs for each taxon based on 19 environmental variables from the WorldClim data set (Hijmans et al., 2005). We first generated SDMs for current climatic conditions (based on a 30-year climate record from 1971 to 2000) and then projected these models onto simulated climate data representing the Last Glacial Maximum (LGM; 21 ka) and last interglacial (LIG; 130 ka). We used two LGM data sets: NCAR-CCSM, version 3.0 (Collins et al., 2006) and MIROC, version 3.2 (Hasumi & Emori, 2004). The LIG data are based only on the CCSM model. We used current and past monthly climate data at 2.5′ (4-km) spatial resolution. Waltari et al. (2007b) and Peterson & Nyári (2008) discuss the process of layer development for the historical data more fully.
We collated georeferenced occurrence points for each species from natural history collection databases (e.g. Arctos, http://arctos.database.museum; Appendix S2) to determine current distributions, and used Maxent 3.3.3e (Phillips et al., 2006; Phillips & Dudík, 2008) to construct SDMs. For each analysis, we generated five replicates under the ‘cross-validate’ option to ensure that results were robust to perturbation of the species' locality data. Other settings were left at defaults. In initial rounds of analyses, we implemented a jackknife manipulation to assess the relative contributions of all 19 bioclimatic variables in predicting current distributions. To avoid over-fitting models to occurrence points (Peterson & Nakazawa, 2008), we then generated summary maps based on the eight most influential variables for each taxon.
Study system, taxa and sampling
We investigated the comparative phylogeography of eight taxon pairs within Beringia, with samples ranging from 6 to 89 individuals, and using 626–1143 bp of cyt b (Table 1). We could not reject the assumption of neutral evolution for any data set based on McDonald–Kreitman tests.
Genetic distance and timing of divergence
Divergence estimates from beast analyses were plotted along a Quaternary timeline of the global oxygen isotope ratio record (Fig. 2) to illustrate relative coalescence times (TMRCA) for each taxon pair (both combined and as separate populations) in relation to glacial periodicity. Coalescence times for combined rodent taxon pairs are significantly greater than for either western or eastern separately whereas TMRCA for combined and separate taxon pairs are similar among most shrews. Mean TMRCA for rodent taxon pairs range through the Quaternary from 2.47 to 0.17 Ma (Table 2, Fig. 2) although confidence limits are broad. Individual western and eastern TMRCA for rodents are coincident with a late Quaternary timeframe and range from 0.30 to 0.12 Ma. TMRCA for combined shrew taxon pairs range from 0.07 to 0.02 Ma, spanning only the most recent Wisconsinan glacial (0.13–0.01 Ma).
Table 2. Genetic diversity indices and divergence estimates for all small mammal taxa in this study. n = sample size; Hd = haplotype diversity; π = nucleotide diversity; D = Tajima's D (with associated P-value); FS = Fu's FS (with associated P-value); TMRCA = time to most recent common ancestor (coalescence time with associated 95% confidence interval)
95% CI (Ma)
Dicrostonyx torquatus/D. groenlandicus
Lemmus sibiricus/L. trimucronatus
M. oeconomus (western)
M. oeconomus (eastern)
M. rutilus (western)
M. rutilus (eastern)
Sorex camtschatica/S. ugyunak
S. portenkoi/S. ugyunak
S. minutissimus (western)
S. minutissimus (eastern)
S. tundrensis (western)
S. tundrensis (eastern)
The population size change through time inferred through BSPs among several rodent lineages showed generally increasing effective population sizes through time, although Dicrostonyx (both species) and L. sibiricus showed essentially no change considering the broad confidence intervals. Population growth for both Microtus (Fig. 3) and Myodes (Appendix S3) began earlier in eastern Beringia than in western Beringia, but growth was coincident with the Wisconsinan glacial period and slowed by the LGM in all instances. In Microtus oeconomus, a decline is evident most recently in the demographic history within eastern Beringia (Fig. 3). As was the case for lemmings (Dicrostonyx and Lemmus), there were no significant demographic trends for shrews in western Beringia except for S. minutissimus, in which growth in western Beringia appears to have occurred earlier (pre-LGM) than in eastern Beringia (post-LGM; Appendix S3). In eastern Beringia, S. tundrensis experienced growth coincident with the transitional climate from the late Wisconsinan glacial (c. 40 ka) to the Holocene (11 ka – present), whereas the growth of S. ugyunak populations followed the LGM. Population growth for both Microtus and Myodes is also reflected by significant expansion statistics (Table 2). Expansion statistics are not significant for western Beringian shrew lineages, except for S. minutissimus, but are significant for eastern Beringian lineages (Table 2). The lowest values of both nucleotide and haplotype diversity were found among shrew taxon pairs (Table 2).
Tests of simultaneous divergence
For the combined analyses of rodents and shrews, and also the analysis of the four pairs of rodent taxa, significantly high values of Ω indicate multiple divergences (Table 3, Appendix S3). For analysis of the four pairs of shrew taxa, the 95% CI of Ω encompasses zero, so we could not reject simultaneous divergence, although a BF test of Ψ indicates substantial evidence against simultaneous divergence (Table 3, Appendix S3). Joint posterior probability density plots should exhibit either a peak posterior probability situated centrally in relation to the x and y axes [E(τ) and Ω, respectively] to illustrate a high degree of discordance between divergence times among taxon pairs (= multiple divergences) or a peak posterior probability flat along the y-axis (Ω) to illustrate zero discordance and therefore simultaneous divergence (Hickerson et al., 2006a). Our plots exhibit a central peak for both analyses combining eight taxon pairs but for the shrew analysis and to a lesser extent the rodent analysis, peak posterior probabilities were intermediate (Appendix S3). Prior and posterior curves of Ω for shrew taxa were also closely overlapping further suggesting weak statistical power (Appendix S3).
Table 3. Results of msBayes analyses. μ = mutation rate; Max. τ = maximum average divergence, in coalescent units, depending on relative genetic diversity; Ω = [var(τ)/E(τ)]; Ψ = number of divergence events; BF = Bayes factor for Ψ = 1 against Ψ > 1. Values in bold suggest substantial support for multiple divergence events
μ (% Myr−1)
Ω 95% CI
PP Ψ = 1
PP, posterior probability.
Rodents + shrews
Rodents + shrews
Species distribution models
The modelled distributions show general patterns of predicted occurrence and absence. These indicate shared barriers and refugial areas, but also show different predicted responses to historical climate change, reflecting variable ecological associations (Appendix S3). Xeric tundra-associated species such as D. torquatus, D. groenlandicus, L. sibiricus, and S. ugyunak are predicted during all timeframes to occur only at the highest latitudes. The two shrew species with limited distributions in eastern Siberia (S. portenkoi and S. camtschatica) have closely matching historical predictions despite allopatric contemporary ranges. The remaining taxa constitute widespread temperate–subarctic species and are collectively predicted to have broad distributions that avoid the highest latitudes (Appendix S3). The main areas of lowest prediction for all taxa include the Kolyma Uplands, from the Chukchi Sea in the north to the Sea of Okhotsk in the south (Fig. 1), and much of central Beringia coincident with inundated continental shelf south and east of Chukotka (Fig. 1, Appendix S3). All taxa are predicted to have increased their ranges during the LGM relative to present predictions. Currently widespread temperate–subarctic species including Microtus oeconomus, Myodes rutilus, S. minutissimus and S. tundrensis are also predicted to have had larger ranges than present during the LIG (a period of slightly warmer climate than at present). The remaining taxa are predicted to have had smaller ranges at the LIG than at present.
Beringia is the centre of the terrestrial Holarctic although, paradoxically, it also represents the periphery of both northern continents. A complex biogeographical history, combined with dramatic geophysical and climatic change through time, provides an ideal opportunity to investigate spatial and ecological responses to dynamic environmental processes over a relatively brief evolutionary timeframe (Cook et al., 2005). Our comparative study of 11 small mammal species involved multiple tests of community dynamics not possible through previous investigations of individual taxa or pairs of taxa distributed through Beringia. We tested four sets of hypotheses related to the temporal, spatial and ecological components of Beringian diversification.
Tests for one simultaneous divergence versus numerous divergences across Beringia strongly suggest that the species differentiated on multiple occasions. beast analyses corroborate the test results from msBayes and indicate that the divergence times for different taxa varied considerably, particularly among rodent taxa and between rodents and shrews (Fig. 2, Table 2). These results stress that diversification across Beringia is an ongoing process.
Coalescent estimates of TMRCA place the divergences between all pairs of taxa within the Quaternary (Fig. 2, Table 2). We further tested whether Quaternary diversification among extant Beringian taxa was a consequence of weak, short glacial cycles or strong, long cycles. As reported in other studies (e.g. Fedorov & Goropashnaya, 1999; Fedorov et al., 2003), the lemmings (including Dicrostonyx and Lemmus) have a long history of occupation within Beringia. The western and eastern lineages of lemmings represent clearly differentiated species across Beringia with mean TMRCA coincident with the onset of the Quaternary glacial climate regime for Lemmus and coincident with the mid-Quaternary transition to more extreme 100-kyr glacial cycles for Dicrostonyx (Fig. 2). Although we recognize that confidence intervals around these means are broad, our coalescent estimates of divergence dates for lemmings are consistent with fossil evidence (Repenning, 2001) and we conclude that a Beringian barrier, coupled with the Quaternary climate, drove the speciation events. Among the other rodent taxa, Microtus oeconomus and Myodes rutilus are both considered single species across Beringia but exhibit significant divergence between western and eastern lineages, with a mean TMRCA dated to 405 and 174 ka, respectively (Fig. 2, Table 2). These dates are coincident with major glacial cycles of the late Quaternary. Four shrew taxon pairs share a more recent history of divergence across Beringia. All western and eastern lineages were isolated by the Bering Strait at least c. 11 ka, but only S. tundrensis exhibits substantial genetic divergence between west and east suggesting a separation deeper than the LGM (Fig. 2, Table 2). Most of the current diversity over all the mammals we examined within Beringia is therefore a consequence of the stronger glacial cycles of the late Quaternary, although at least one pair of lemming taxa started diverging much earlier. Also, the finding that half of the detected divergence events appear to have occurred within the most recent Wisconsinan glacial (shrews; Fig. 2) raises the possibility that every glacial cycle of the late Quaternary resulted in multiple invasions of Beringia followed by allopatry, but only a small proportion eventually led to speciation.
We tested a hypothesis of divergence across the Bering Strait versus some other physical or environmental barrier as the major feature promoting allopatry and Beringian endemism. By perceiving Beringia as a general barrier across which multiple populations have differentiated, we were able to assess the relative evolutionary importance of multiple more discrete features of the region as identified in previous phylogeographical and palaeoenvironmental studies. In particular, SDMs provide predictions for persistent barriers over multiple timeframes. Easternmost Siberia has been postulated as a significant barrier for multiple taxa (Galbreath & Cook, 2004; Fedorov et al., 2003; Fig. 1, Appendix S3). This region, extending from the vicinity of the Kolyma Uplands eastwards to the Bering Strait is considered to have been particularly inhospitable to all but the most xeric-adapted species through much of the Quaternary, comprising steppe-tundra, arid steppe or even polar desert (Elias & Crocker, 2008). During glacial phases, the exposed northern continental shelf is likely to have constituted a harsh climatic barrier that prevented many species from circumnavigating the Kolyma Uplands to the north (Galbreath & Cook, 2004) although arctic-adapted species may have benefited from such xeric conditions. Lemming SDMs predict expansion to occupy exposed northern continental shelf and at least D. torquatus spread eastwards to the Bering Strait. Although the initial divergence between Dicrostonyx species may have been coincident with a barrier in the Kolyma region, this is not evident from their current distributions (Fig. 1) and cannot be inferred considering the deep genetic divergence between the two species (Fedorov & Goropashnaya, 1999). The occurrence of D. groenlandicus on Wrangel Island (Fig. 1) indicates that a potential contact zone may have existed previously in the Kolyma region and may indicate that D. groenlandicus was subdivided within Beringia during the LGM (Fedorov & Goropashnaya, 1999), although this is minimally supported by SDMs (Appendix S3). To the south of the Kolyma Uplands, a narrow corridor is predicted by the SDMs to be hospitable to all but the most arctic-adapted species (Fig. 3, Appendix S3), and may have allowed movement between central and westernmost Beringia. Sorex camtschatica currently occurs both west and east of the Kolyma region through this southern area (Fig. 1). Although the area is geographically restricted, it may represent the predominant corridor for movement of temperate species through Beringia.
We do not consider the Bering Strait to be a significant barrier, despite the firm isolation of terrestrial species during interglacial warm phases. For taxa such as Dicrostonyx, the Bering Strait may have helped maintain isolation between populations that initially diverged as a consequence of inhospitable environments during glacial phases (Fedorov & Goropashnaya, 1999). Evidence from shrews (e.g. S. minutissimus) indicates a recent expansion into eastern Beringia followed by isolation across the Bering Strait with rising sea levels (Appendix S3; Hope et al., 2010). Despite separation through the Holocene (a time span comparable to other interglacial warm phases), little differentiation across this barrier is evident.
Finally, SDMs indicate other potential climatic barriers for many taxa in the central Beringian Isthmus (Appendix S3) that may reflect a xeric region south of present Chukotka that was next to a wetter region further east. The wetter region may be coincident with the mesic ‘buckle’ described by Guthrie (2001). Central Beringia may constitute a moisture barrier for multiple species despite the absence of a significant physical barrier (Guthrie, 2001; DeChaine, 2008; Elias & Crocker, 2008). Beringia seemingly provided climatic gradients across both latitude and longitude leading to regional variability in the evolutionary responses of ecologically distinct taxa (Stewart & Dalén, 2008; Hofreiter & Stewart, 2009).
H4: Ecological response
We tested for consistency among the Beringian small mammals in their ecological responses to common environmental processes and found that distinct evolutionary processes reflect differences in ecology among species. In response to environmental change, species can shift range to follow optimal conditions, persist in a given area, or be locally extirpated. Arctic species (Lemmus, Dicrostonyx) persisted within Beringia through multiple glacial cycles. Tundra-associated shrews (S. ugyunak, S. portenkoi) also persisted in Beringia through the most recent glacial advance (Hope et al., 2012). SDMs for these shrew species and lemmings indicate the possibility of northern Beringian distributions persisting through time. Both coalescent and SDM analyses indicate that S. tundrensis persisted through the last glacial phase in discrete regions coincident with major river corridors that served as buffers to unfavourable conditions (Hope et al., 2011). In contrast, temperate or temperate–subarctic species are predicted to avoid the coldest and most arid areas. The temperate–subarctic vole Microtus oeconomus largely occupies open wet tundra and meadows within Beringia, although it occurs in more temperate habitats elsewhere across its broad Holarctic distribution. Beringian (eastern) M. oeconomus have been isolated for multiple glacial cycles, as indicated by the significant genetic divergence from the central Siberian (western) lineage (Fig. 2; Table 2; Galbreath & Cook, 2004), and the times of demographic expansion during the most recent glacial period were offset between west and east lineages (Fig. 3). Together, the evidence suggests regional persistence in different areas of Beringia accompanied by periodic demographic change in response to local conditions (Galbreath & Cook, 2004). The most temperate taxa, including Myodes rutilus and S. minutissimus, are associated primarily with taiga habitats but can also be found in tundra or open boreal zones (Hope et al., 2010). Myodes rutilus diverged across Beringia over the last two glacial cycles (Fig. 2, Table 2), again indicating regional persistence within Beringia accompanied by local demographic change. Sorex minutissimus exhibits a similar response but has a more recent history in Beringia with a rapid expansion eastwards since the LGM (Hope et al., 2010; Fig. 2). Although SDMs indicate suitable habitat in eastern Beringia as far back as the LIG (Appendix S3), molecular data do not support occurrence within this region before the latest Wisconsinan (Table 2).
In summary, different responses to climate fluctuations include: (1) persistent panmictic populations of tundra-associated (arctic) species occupying single xeric regions within Beringia (e.g. Dicrostonyx, Lemmus, although there is no evidence to suggest that there was only one common region in western or eastern Beringia, respectively), (2) persistent species responding largely independently of the prevailing climate, such as S. tundrensis persisting in separate river systems (Hope et al., 2011), (3) persistent species that respond differently to local climate change in western and eastern Beringia respectively, reflecting significant regional differences in the timing and severity of environmental change (Microtus oeconomus, Myodes rutilus), and (4) species shifting rapidly into and through Beringia in response to favourable climate (e.g. S. minutissimus; Hope et al., 2010). As such, macroevolutionary processes across Beringia are influencing a shifting biotic mosaic, reflecting turnover and differential responses to climate change through time (Hoberg & Brooks, 2008, 2010).
Because diversification across Beringia has been an ongoing process through the Quaternary, it may be that different responses to environmental change among extant Beringian species reflect stages in a single sequential process. This evolutionary process includes not only divergence across Beringia but also local adaptation as a consequence of endemism spanning successive glacial episodes.
The colonization of Beringia by a hypothetical temperate species would initially have been rapid in response to a transitional climate between glacial and interglacial phases (e.g. S. minutissimus during the latest Wisconsinan glacial). Subsequent isolation across a barrier such as the Kolyma Uplands during the early onset of glacials or across the Bering Strait during interglacials would have led to either local extirpation or persistence in small populations coupled with initial local adaptation. This hypothetical species, although still associated with taiga habitats and exhibiting a strong demographic response to transitional climate (e.g. M. rutilus), would have now persisted in Beringia through multiple glacial cycles and would begin to exhibit regional differences that may reflect adaptation along a Beringian environmental gradient (e.g. a moisture gradient from west to east). After additional climate cycles, divergence within the species could increase, with regional differences becoming more pronounced (e.g. Microtus oeconomus). The possibility that the individuals of M. oeconomus found within Beringia are better adapted to temperate–subarctic environments than individuals of M. oeconomus found occupying a broader fundamental niche further west in Eurasia should be further explored. SDMs derived from both western and eastern lineages of M. oeconomus do not predict occurrence of the species further west in Eurasia (where it presently occurs), possibly reflecting local adaptation to more arctic environments in Beringia.
Xeric-adapted steppe-tundra species such as Dicrostonyx or Lemmus may have evolved through a similar process of initial invasion followed by isolation, persistence and adaptation over multiple glacial cycles. For these species, an initial barrier between western and eastern Beringia may well have been a strong environmental gradient (e.g. mesic buckle; Guthrie, 2001) across Beringia instead of a geomorphic feature, particularly considering that the Kolyma Uplands did not extend north onto the exposed continental shelf during glacial maxima. In eastern Beringia, D. groenlandicus, and particularly L. trimucronatus, extend south through much of Alaska and are not strictly limited to the northernmost areas as are their western Beringian counterparts (Fig. 1). This difference may reflect adaptation in response to a moisture gradient across Beringia (DeChaine, 2008; Elias & Crocker, 2008) from xeric in western Beringia to mesic further east. Sorex ugyunak and closely related species west of the Bering Strait could represent newly arrived precursors to the speciation process among xeric-associated Beringian species, as these shrews are among the most arid-adapted soricines (Hope et al., 2012). Sorex camtschatica, a recent arrival to the taiga of the Kamchatka Peninsula and Magadan coast (Fig. 1) may, however, reflect recent adaptation to more temperate habitats in western Beringia. As such, the evolutionary process of divergence and adaptation within Beringia again reflects multiple ecological pathways. Temperate mesic-associated species may adapt to more xeric and arctic conditions whereas species that were initially more xeric-associated and arctic-associated may later become adapted to wetter and warmer conditions. It seems that Beringia may exert strong environmental pressures on many if not most species moving into this area, exemplifying the region's role as a significant biological filter (Hopkins, 1967, 1982).
The concept of Beringia (particularly eastern Beringia) as a holding ground for high-latitude species during the Quaternary may represent the most significant mechanism for producing local endemism. The majority of transcontinental dispersals through Beringia during the Quaternary were eastward (Waltari et al., 2007a). Species currently in eastern Beringia with a Palaearctic origin but only a limited Nearctic distribution are likely to represent recent arrivals (e.g. S. minutissimus; Hope et al., 2010). However, as suggested by Repenning (2001), other species maintained this limited Nearctic distribution through multiple glacial cycles of the late Quaternary, perhaps due to the increased amplitude or severity of glaciation. Even species that shifted dramatically in response to climate change were eventually restricted by continental ice. For instance, genetic evidence indicates that Myodes rutilus previously expanded to occupy areas further south and east in North America, but was extirpated following the onset of a previous glacial phase. Introgressed haplotypes of M. rutilus (not present in current North American populations) were found within a related species (Myodes gapperi) in Canada (Runck et al., 2009). Myodes rutilus is again spreading through Canada and has come into secondary contact with M. gapperi. Competitive interactions coupled with glaciations may have significantly hindered the establishment of Beringian taxa further into North America. Following repeated glacial cycles, prolonged adaptation to the Beringian climate may in turn have maintained a limited distribution, such as is seen with Microtus oeconomus or L. trimucronatus.
The small mammals of Beringia currently comprise pairs of taxa ranging from genetically indistinct to clearly differentiated species and yet taxonomy does not reflect these levels of divergence. Based on coalescent estimates, western and eastern lineages of M. oeconomus could be considered incipient species, being significantly more diverged than the shrew taxon pairs. The current taxonomy, which recognizes S. portenkoi, S. camtschatica and S. ugyunak as different species (Hutterer, 2005), is not supported by our analyses. The results of msBayes analyses for shrew taxa could not reject a simultaneous divergence and estimates for TMRCA are comparable between the two pairs of taxa (Table 2). Further, time to coalescence for each combined pair was not significantly longer than either individual ‘species’, indicating negligible differentiation (Fig. 2, Table 2).
The coalescent-based tests for simultaneous divergence that we implemented rely on a number of important assumptions that warrant further consideration (Carstens et al., 2005; DeChaine & Martin, 2006; Hickerson et al., 2006a). Although variance around coalescent parameter estimates based on a single mitochondrial locus are broad and increasing the number of loci can improve divergence estimates (Hickerson et al., 2006b; Huang et al., 2011), inferences from msBayes have been shown to be robust to both single-locus data sets and small sample sizes (Hickerson et al., 2006a). However, our data set illustrates that too few taxon pairs and/or too little genetic information from a single locus coupled with recent isolation may limit resolution of the number of divergences. For instance, the separate analysis for shrew taxon pairs resulted in an estimate of Ω that encompassed zero, suggesting a potentially simultaneous divergence, but with substantial support for multiple divergences from BF tests of Ψ (Table 3). Multiple divergences are reasonable considering the coalescent divergence estimates from beast analyses (Table 2, Fig. 2), but there is a lack of consistent statistical support. In general, values of Ψ were not well resolved, and although Ω is a more rigorous estimator of simultaneous versus non-simultaneous divergence (Hickerson et al., 2006a), our data were not sufficient in this test to resolve clear estimates of either statistic.
Probably the most important factor limiting our ability to resolve the relative number and timing of divergences for Beringian mammals is the fact that we used a single locus, because data from multiple loci were not available for all taxa. We recognize the benefits of using multiple independent loci in coalescent analyses (Edwards, 2009), as they can narrow the confidence limits around coalescent estimates of population parameters and new methods allow the incorporation of multilocus data sets within msBayes for multiple species comparisons (Huang et al., 2011). However, the Beringian mammalian system is dominated by recent divergences and population structure that is still largely unresolved. Incorporating data from a feasible number of nuclear loci may not provide clearer resolution considering that neutral nuclear substitution rates can be an order of magnitude lower than for cyt b (Lynch et al., 2006), although at the minimum, multilocus data sets can provide independent tests of inferences derived from mtDNA alone (Galbreath et al., 2011).
This investigation of multiple Holarctic mammals assesses the processes that drive evolution across Beringia. We have tested a suite of eight pairs of small mammal taxa for simultaneous divergence across Beringia and rejected that hypothesis in favour of multiple distinct divergence events through time among these taxa. This explicit analysis of pairs of taxa extends the implications of multiple independent studies, while controlling for the genealogical stochasticity inherent in coalescent processes (Carstens et al., 2005; Hickerson et al., 2006a). Climate-driven changes have resulted in multiple diversification events among small mammals within Beringia, particularly during the late Quaternary. Small mammal fossils reflect abundant movement of temperate species through Beringia during the early Quaternary (Repenning, 2001), but little in situ diversification occurred until the barriers associated with stronger (100-kyr) glacial cycles confined populations for extended periods within Beringia. The degree of specialization to Beringian environments, reflected by ecological and demographic differences between species may be directly related to occupancy times within Beringia. Those species with negligible genetic differentiation across Beringia are the most recent additions to local communities and have responded to climate change by rapid and extensive demographic fluctuations. Species exhibiting deeper divergence across Beringia reflect persistent and more localized demographics that indicate possible adaptation to regional conditions or sufficiently broad fundamental niches to withstand major environmental fluctuations.
These different scenarios highlight a conundrum that warrants further investigation, particularly considering the possibility that future climate warming may have no analogue within a Quaternary timeframe (MacDonald, 2010). Are well-differentiated Beringian endemics highly adapted to a narrow range of climate conditions that may not exist in the near future, or will such species be resilient to dramatic change? In addition, how will persistent species respond ecologically to community change as a new wave of temperate species rapidly advances into Beringia during the current warming phase? SDMs coupled with corroborating genetic evidence provide a useful method for evaluating the dynamics of community change based on climate variables. Predicted distributions at the LIG provide a surrogate for future scenarios considering warmer conditions than at present and illustrate that arctic-adapted species (including many of the oldest components of the current Beringian small mammal fauna) may be the most vulnerable to severe population decline in response to a warming climate and a transition to more mesic conditions within Beringia (Elias et al., 1997; Fedorov & Goropashnaya, 1999).
Resolution of our comparative phylogeographical analyses is limited by recent allopatry across Beringia with only minimal genetic divergence, a small number of pairs of taxa within each group and single-locus data sets. However, we have shown that repeated episodes of allopatry over multiple glacial cycles are likely to play an important role in speciation. The Beringian region presents multiple barriers to dispersal related to regional climatic conditions through time coupled with geographical features. Repenning (2001) concluded that up to the mid-Quaternary, significant diversification in both North America and Eurasia was influenced by a process of transcontinental movement through Beringia and subsequent speciation elsewhere. Although late Quaternary glaciations may have restricted transcontinental passage, evidence from small mammals indicates a continued role of Beringia in promoting speciation through local endemism and adaptation to arctic environments.
Statistical analyses were facilitated by the University of Alaska Fairbanks, Life Science Informatics Portal, accessed online at http://biotech.inbre.alaska.edu/. Funding was provided by the American Society of Mammalogists Grants-in-Aid of Research, National Park Service, Beringian Coevolution Project (NSF0196095 and 0415668), UNM Biology Department Gaudin Scholarship, and the U.S. Geological Survey's Alaska Regional Executive Department of the Interior (DOI) on the Landscape initiative. We thank E.P. Hoberg and D.J. Hafner for insightful comments. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.
This research is part of Andrew G. Hope's doctoral dissertation. He studies how environmental processes govern evolution through time and space, with an emphasis on Holarctic vertebrate community dynamics and responses to climate change.
Author contributions: A.G.H. and J.A.C. conceived the ideas; A.G.H. collected the data; A.G.H. and N.T. analysed the data; S.L.T. and J.A.C. secured financial support and provided laboratory facilities; A.G.H. and K.E.G. led the writing; all authors contributed to writing.