We provide the first comparative multispecies analysis of spatial genetic structure and diversity in the circumpolar Arctic using a common strategy for sampling and genetic analyses. We aimed to identify and explain potential general patterns of genetic discontinuity/connectivity and diversity, and to compare our findings with previously published hypotheses.
We collected and analyzed 7707 samples of 17 widespread arctic–alpine plant species for amplified fragment length polymorphisms (AFLPs). Genetic structure, diversity and distinctiveness were analyzed for each species, and extrapolated to cover the geographic range of each species. The resulting maps were overlaid to produce metamaps.
The Arctic and Atlantic Oceans, the Greenlandic ice cap, the Urals, and lowland areas between southern mountain ranges and the Arctic were the strongest barriers against gene flow. Diversity was highest in Beringia and gradually decreased into formerly glaciated areas. The highest degrees of distinctiveness were observed in Siberia.
We conclude that large-scale general patterns exist in the Arctic, shaped by the Pleistocene glaciations combined with long-standing physical barriers against gene flow. Beringia served as both refugium and source for interglacial (re)colonization, whereas areas further west in Siberia served as refugia, but less as sources for (re)colonization.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
To understand how species have been moulded through time and space has fascinated researchers for centuries, and with the development of molecular techniques, new discoveries and connections could be revealed (reviewed in Hewitt, 2001). Hampton Carson (Carson, 1970, 1983) was one of the pioneers in placing genetic phylogenies into a spatial and temporal framework, and in 1987, John C. Avise called this discipline ‘phylogeography’ (Avise et al., 1987). The heart of phylogeographical inference is to understand the underlying temporal and spatial factors that shape the genetic diversity and thus the evolution of a species (Avise et al., 1987; Hewitt, 1996, 2001; Avise, 1998, 2009). Although the possible factors involved are many, such as extent of secondary contact among divergent lineages and dispersal capacity (Avise et al., 1987; Nichols & Hewitt, 1994; Avise, 1998; Hewitt, 2001), the most influential ones are usually associated with barriers against gene flow (Slatkin, 1987). Such barriers may vary through time and space: they can be long-term physical ones, such as oceans and mountain ranges, more short-term physical/environmental ones, such as glaciers, or ecological/environmental ones, such as changing climatic conditions (Slatkin, 1987; Tivy, 1993). As the influence of barriers can be substantial, we expect that species subjected to the same barriers to gene flow show some congruence amongst their phylogeographies (Avise et al., 1987). This expectation has led to the study of comparative phylogeography, where central goals are the testing of hypotheses beyond individual species, addressing how abiotic and/or ecological processes drive evolution within whole communities (Bermingham & Moritz, 1998; Avise et al., 2000; Arbogast & Kenagy, 2001; Byrne et al., 2008; Qiu et al., 2011). Such studies have successfully identified historical events that concordantly changed the pattern of gene flow among populations of different species (Walker & Avise, 1998; Riddle et al., 2000; Zink, 2002; Tribsch & Schönswetter, 2003; Hewitt, 2004b; Leaché et al., 2007; Morgan et al., 2011; Smith et al., 2011; Lorenzen et al., 2012).
On the other hand, it has been stated that phylogeographical patterns are species-specific rather than congruent (Taberlet et al., 1998; Soltis et al., 2006; Stewart, 2008; Edwards et al., 2012). Stochasticity related to plant dispersal, establishment and reproduction does obviously influence distribution ranges, and fast recolonization of plants relies on stochastic long-distance dispersal events (Nathan, 2006). Thus, the incongruence found in some of the larger compilations (e.g. Taberlet et al., 1998; Carstens et al., 2005; Soltis et al., 2006) is not surprising. However, some of this incongruence may be a result of different sampling design rather than different species histories. Most larger comparative studies have been, and still are, typically based on compilations from the literature, where different sampling strategies and genetic marker systems have been used (e.g. Taberlet et al., 1998; Carstens et al., 2005; Soltis et al., 2006). Comparative phylogeography is an expanding field of research with a large underexplored potential, especially through more standardized sampling (Gugerli et al., 2008; Gutierrez-Garcia & Vazquez-Dominguez, 2011), more formal statistical tests (e.g. Hickerson et al., 2006; Carnaval et al., 2009), utilization of methodological tools like geographical information systems (GIS, e.g. Swenson, 2008; Chan et al., 2011), and through a stronger interdisciplinary approach (Barnosky, 2008; Beheregaray, 2008; Avise, 2009; Johnson et al., 2009; Hickerson et al., 2010).
The Arctic provides an excellent system to investigate the existence and strength of large-scale general barrier/gene flow patterns through comparative phylogeography. This vast region harbors a young biome, which emerged in response to the global cooling in the Pliocene (Murray, 1995; Zachos et al., 2001). It provides less habitat complexity and holds fewer, often very widespread species with wider ecological niches compared with temperate areas (Callaghan et al., 2004 and references therein). The region presents long-standing physical barriers such as the Atlantic Ocean and the Ural Mountains, but it has also been strongly affected by more ephemeral, climate-induced barriers. Throughout the Pleistocene, the Arctic experienced large climatic oscillations and recurrent glaciations. Massive continental ice sheets covered vast areas, including most of North America and Europe (Dyke, 2004; Svendsen et al., 2004). The glaciations, the last one ending c. 10 000 yr ago, physically expelled most former vegetation, and made vast areas inaccessible for any species to establish.
A major exception was the area around the Bering Strait, named ‘Beringia’ (Hultén, 1937), which was never glaciated (Dyke, 2004). During the glaciations, the global sea level was also much lower than today, turning the shallow Bering Strait into the Bering Land Bridge connecting Asia and North America (Hopkins, 1967; Elias et al., 1996). Beringia was shown to be a hotspot of species diversity and endemism, and Hultén (1937) proposed that this region served as a major glacial refugium for the arctic flora. A compilation of fossil, phytogeographical, and early molecular evidence supported Hultén's idea of Beringia as a major arctic refugium (Abbott & Brochmann, 2003). Molecular studies have also supported Beringia as an important source for recurrent postglacial plant colonization (Eidesen et al., 2007b; Westergaard et al., 2010). However, the size and overall importance of a major Beringian refugium compared with other possible refugial areas have been questioned (Shafer et al., 2010). A species compilation focusing on flora of the Eurasian Arctic showed that other regions, such as northwestern Russia, have a similarly high number of species to that of unglaciated regions in Beringia (Tkach et al., 2008). Fossil evidence and pollen records have also shown that extensive tundra south and east of the Eurasian ice sheets provided suitable refugial habitats for arctic biota during the last glacial maximum (Huntley & Birks, 1983; Tarasov et al., 2000; Birks, 2008). Possible refugial areas also existed south of the North American ice sheets, although the extension of tundra was much more restricted, as the boreal forest was much closer to the ice sheet margin (Jackson et al., 2000; Prentice et al., 2000). Further, molecular evidence supporting other refugia outside Beringia is currently increasing (Provan & Bennett, 2008; Shafer et al., 2010). For arctic plants, molecular evidence has supported refugial areas located, for example, south and east of the Eurasian ice sheets (Skrede et al., 2006; Winkler et al., 2012), in mountain regions in Japan (Ikeda et al., 2006, 2009), within the Cordilleran ice sheet (Marr et al., 2008), in ice-free regions in the Canadian High Arctic, and south of the Laurentide ice sheet (Tremblay & Schoen, 1999).
The glacial cycles drove major species range shifts in the Arctic, reflected in the phylogeographical pattern of arctic animals (reviewed in Hewitt, 2004a, 2011) and plants (reviewed in Abbott & Brochmann, 2003; Brochmann et al., 2003; Brochmann & Brysting, 2008). The recurrent retreats to isolated glacial refugia resulted in reduced gene flow, leading to divergence, further enhanced through successive founder events during postglacial range expansions (Hewitt, 1996, 1999, 2004a; Ibrahim et al., 1996). The standard genetic models of glacial refugia and colonization predict lower genetic diversity in formerly glaciated areas and higher genetic diversity (including unique or rare alleles, genotypes or markers, i.e. genetic distinctiveness) in glacial refugia (Hewitt, 1996; Widmer & Lexer, 2001; Keppel et al., 2012). The first prediction is often violated in areas of secondary contact, where admixture of formerly separated lineages increases genetic diversity (Sharbel et al., 2000; Petit et al., 2003), and this should be most prominent in species with high dispersal capacity. Contrary to traditional beliefs (e.g. Hultén, 1958; Dahl, 1963; Nordhagen, 1963), many arctic plants have now been shown to colonize over long distances (Brochmann et al., 2003; Alsos et al., 2007; Popp et al., 2011; Hoffmann, 2012), even across the Atlantic Ocean (Hagen et al., 2001; Skrede et al., 2006; Schönswetter et al., 2008), which had often been considered an impenetrable barrier to seed dispersal. The prediction of higher genetic diversity within refugia can be fulfilled if the refugia were located within the present distribution range and the population size remained relatively large. This could be the case for the postulated Beringian refugium, but as shown by the fossil record, other larger refugia for arctic organisms were situated south of the ice and are today replaced by boreal or temperate biota (Huntley & Birks, 1983; Tarasov et al., 2000). Small glacial refugia, such as nunataks protruding above ice caps, are likely to have experienced a drastic reduction in diversity as a result of genetic drift. However, such areas may still contain rare or unique alleles, making them genetically distinct (Hewitt, 1996, 1999, 2004a; Schaal et al., 1998; Stehlik et al., 2001). The overall importance of glacial survival in nunataks or other small in situ refugia is debated (Brochmann et al., 2003; Birks et al., 2012), but for some plant species there is recent molecular evidence supporting such refugia within former heavily glaciated regions, for example, in the Alps (Stehlik et al., 2001; Schönswetter et al., 2005; Ronikier et al., 2008), British Columbia, Canada (Marr et al., 2008, 2013), northern Norway (Parducci et al., 2012), Greenland/Svalbard and southern Scandinavia (Fedorov & Stenseth, 2001; Westergaard et al., 2011).
Here we provide the first comparative multispecies analysis of spatial genetic structure and diversity in the circumpolar Arctic based on a single stringent strategy for sampling and genetic marker analysis. We use GIS-based meta-analyses of 17 widespread north boreal to arctic–alpine plant species sampled in the field over most of their range. Eleven of the species have a full circumpolar distribution. We aim to identify potential general patterns of genetic discontinuity/connectivity and diversity/distinctiveness in the circumpolar Arctic; to address the importance of Beringia as a refugium (Hultén, 1937), the existence of common refugial areas and contact zones (Hewitt, 2004a), and the overall extent of dispersal across barriers, such as the Atlantic Ocean (Brochmann et al., 2003); and to assess the causes leading to the patterns identified. In particular, we assess whether the strongest barriers against gene flow are represented by oceans and mountain ranges or by repeated glacier formation and climate changes.
Materials and Methods
Seventeen plant species (seven herbs, seven dwarf-shrubs, two shrubs and one tree) were collected in the field throughout most of their arctic–subarctic distribution range as well as from more southern mountain regions. Eleven of the species have a circumpolar distribution, while six are restricted to the amphi-Atlantic region. In total, 911 local populations and 7707 plant individuals were successfully genotyped (Fig. 1; Table 1; Supporting Information, Table S1). The species possess different life-history traits but they are all widespread in formerly glaciated areas. Thus, all of them have been able to expand their distribution range postglacially and must have been influenced by the Pleistocene glaciations.
Table 1. Overview of the 17-species data set (for details, see the given references and Supporting Information Table S1)
Polymorphic AFLP markers
Distribution ranges (Dist): AA, amphi-Atlantic; CP, circumpolar; CB, circumboreal; Hultén & Fries (1986); Pop/ind, number of populations and individuals analyzed for AFLPs; Extr, extraction method (CTAB, after Doyle & Doyle (1987) with modifications specified in Schönswetter et al. (2002); DPK and DPMK, DNeasy Plant 96 Kit and Plant mini Kit, respectively); S, number of genetic groups inferred from Bayesian clustering using Structure 2.1 Pritchard et al. (2000). AFLP primers are presented with labeling color and selective nucleotide extension of primers EcoRI/MseI.
Scored in two steps to cover the variation in genetically depauperated regions.
For most populations, leaves from 11 plants were collected 25 m apart along a 250 m transect and dried in silica gel. The sampling design deviated for some small populations and early collections (Table S1). Silica samples and DNA extracts are deposited in the DNAbank at the National Centre for Biosystematics, Natural History Museum, University of Oslo, and one voucher from most populations is deposited at the Natural History Museum in Oslo (O). DNA was extracted following the CTAB protocol after (Doyle & Doyle, 1987) with modifications specified in Schönswetter et al. (2002) or with the DNeasy Plant 96 Kit and/or Plant mini Kit (Qiagen). Genetic diversity was analyzed using amplified fragment length polymorphisms (AFLPs) following Gaudeul et al. (2000) with modifications given in Alsos et al. (2007), except that Dryas octopetala was analyzed using the Applied Biosystems AFLP kit following the manufacturer's protocol (see Skrede et al., 2006). All samples were run on an ABI Prism 3100 Genetic Analyzer (for more details regarding AFLP analyses and detailed locality data and collectors, see Tables 1, S1; Skrede et al., 2006; Alsos et al., 2007, 2009; Ehrich et al., 2007, 2008; Eidesen, 2007; Eidesen et al., 2007a,b; Westergaard et al., 2010). Reproducibility was assessed according to Bonin et al. (2004) using 23–63 replicates per species.
For each species, genetic structure, diversity and distinctiveness were calculated based on the AFLP markers. Several methods were combined to subdivide the samples of each species into geographically consistent genetic groups. Model-based clustering was performed with the program Structure 2.1 (Pritchard et al., 2000) at the bioportal of the University of Oslo. The AFLP multilocus phenotypes were treated as diploid multilocus genotypes, with the unknown alleles as missing values, and run under a no-admixture model. The burn-in period was set to 100 000 iterations followed by 1 000 000 iterations. The number of clusters, K, was varied from two to 20 in 10 independent runs. The most appropriate number of genetic groups was determined using the procedure described in Evanno et al. (2005). When there was doubt about the number of groups, the most conservative number was retained. In addition, the groupings were evaluated on the basis of principal coordinate analyses and tree construction algorithms (neighbor-joining and maximum parsimony (Alsos et al., 2007; Eidesen et al., 2007b). Some sampling locations (populations) contained individuals from two, or several, genetic groups. For the mapping procedure, these sampling locations were assigned to the most frequent genetic group.
Two subspecies of Saxifraga rivularis are described, but we only included the amphi-Atlantic subspecies rivularis in this study (see Westergaard et al., 2010). Two data sets were available for this taxon: a smaller data set with 22 populations, 207 individuals and 78 usable markers; and an extended data set with 28 populations, 245 individuals and 45 usable markers. Both data sets were analysed with Structure. In our analyses, two main groups were detected in both data sets (a Svalbard group and an amphi-Atlantic group; Fig. 1), and the Structure results were congruent among the 22 populations included in both data sets. Thus, for the Structure analyses, we used the results from the largest data set containing additional populations, to cover a larger geographical range. We used the smaller data set to estimate genetic diversity and distinctiveness, as the smaller data set with more markers and more evenly sampled populations would give estimates that were more accurate.
Using AFLPdat (Ehrich, 2006), mean intrapopulation genetic diversity was estimated as the average number of pairwise differences for all populations of each species (D; Nei, 1987) and distinctiveness based on rare AFLP markers was calculated as frequency-down-weighted marker values (DW; Schönswetter & Tribsch, 2005). For S. rivularis, the data set with the highest number of markers, but a lower number of populations, was used for DW and diversity calculations. In total, genetic diversity was calculated for 868 populations, as populations represented by one individual were excluded (Table S1). Genetic diversity was particularly low in the northern populations of Arabis alpina and Micranthes stellaris. In order to take into account all genetic variation present in the northern samples, without excluding markers that could be difficult to distinguish with certainty from other markers present in other geographic regions, the data sets for these species were scored in two steps. First, all the northern individuals were scored together with a subset of the southern samples. Secondly, a subset of the northern individuals, including all the different multilocus genotypes identified in the first scoring, were scored together with all southern individuals. In the data analyses, the second scoring including more markers was used, except for the estimation of genetic diversity in the northern populations. As the A. alpina and M. stellaris data sets contained different numbers of genetic markers, a range-wide estimation of DW was not possible. We therefore excluded these two species from the DW analyses, and in total 778 populations were included in the calculations of DW (Table S1).
To reveal general patterns of genetic discontinuity/connectivity and diversity/distinctiveness, we assembled the patterns from all species into three metamaps, representing genetic structure, diversity and distinctiveness, respectively, using GIS software ArcView 3.3 and ArcGIS 9.3 (Esri Inc., Redlands, CA, USA). A grid with a spatial resolution of 50 × 50 km was superimposed upon the northern hemisphere world map (projection: Lambert Equal-Area Azimuthal North Pole). Sampling positions (representing populations) of the 17 species, including information about their genetic group affiliation based on Structure analyses, diversity and distinctiveness estimates, were overlaid on this map (Table S1). As these input points were not randomly or systematically sampled and thus potentially biased, robust GIS methods and principles were preferred. A grid cell size of 50 km was chosen because this scale is small enough to separate the genetic groups into different areas, but still large enough to obtain large-scale patterns like global genetic barriers and diversity patterns. We chose to work in a raster format that made GIS overlay analyses and interpolations possible. Making grids out of vector data (here sampling points and minimum convex polygons) implies some loss of accuracy, but we regarded this issue to be of minor importance because we were focusing on large-scale patterns globally in the northern hemisphere.
A border-strength map was obtained from the Structure group affiliations by the following procedure (steps 1–8):
For each species, estimates of the geographical distribution of each genetic group were obtained as the nonoverlapping minimum convex polygons based upon all sampling positions of populations belonging to the genetic group in question.
Each minimum convex polygon was converted to raster data (presence or absence of the genetic group) by scoring presence in all 50 km × 50 km grid cells that were included, fully or partly, in the convex polygon.
For each species, second estimates of the distributions of genetic groups were obtained by assigning each of the formerly unassigned grid cells to the genetic group represented by a presence grid cell that was situated closest in space to the target cell. The second estimate provided sharp borders in the middle between genetic groups for all species. The maximum possible expansion range for each species was restricted to the circumpolar region for circumpolar/circumboreal species, and to hemicircumpolar for the amphi-Atlantic species (Table 1). Thus, a border appeared where expanding genetic groups met. In this way, the distribution limits towards the south, or towards the east or west for the noncircumpolar species, did not generate a border.
For each species and each of the sharp genetic-group borders resulting from step 3, the genetic dissimilarity of the two genetic groups was calculated as the average number of pairwise differences (Table S2) between individuals belonging to the different groups using Arlequin v.2.0 (Schneider et al., 2000). The genetic distance was ranged to a scale from one to two to allow direct comparisons, using the formula: ranged value = (value – MIN value)/(MAX value – MIN value) + 1. Adding one was done to avoid zero values, which were used as ‘no border’ value. The genetic dissimilarity was treated as a characteristic of each genetic border.
The width of each genetic-group border was adjusted to proportionality with the genetic distance between groups by the formula n = 4 + 4x, where n = breadth of border (number of grid cells), and x is the scaled genetic dissimilarity. Thus, for instance, the border was treated as four grid cells (200 km) wide when the ranged value of genetic difference between two groups was one and as eight grid cells (400 km) wide when the difference was two units. We chose to adjust the width of each border according to the genetic distance between groups rather than increasing the value in the sharp border grid cell. This procedure would allow the strongest borders to contribute more in the border overlay analysis, despite uncertainties of exact placement of borders as a result of uneven sampling intensity and the assumption from step 3 that borders were in the middle between assigned groups. For each species, all grid cells were classified as border cells (1) or nonborder cells (0).
For all 17 species, maps of the grid cells' status as border cells or nonborder cells were superimposed.
For each grid cell, the 17 binary values for border status were summed to give an overall border affiliation index value for each grid cell.
The final gene flow and border map was obtained by calculating a moving average of the overall border affiliation index in a quadratic neighbourhood of 15 × 15 cells.
To create diversity and distinctiveness maps, we standardized the estimates of diversity (D) and distinctiveness (DW) to enable the comparison between species following the formula: standardized value = (value – MIN value)/(MAX value – MIN value). For each species, the standardized genetic diversity and distinctiveness measures were assigned to the grid cells where samples were available and outlines of the species' current distribution based on Hultén & Fries (1986; Fig. 1) were added to the maps. Degrees of diversity/distinctiveness were then spatially interpolated to fill the unsampled grid cells within the outlined distribution range of each species using block kriging interpolation version 3.2 for ArcView 3.3. Adding the respective maps for the single species together produced a metamap for genetic diversity and distinctiveness.
In total, 78–325 AFLP markers were scored per species (Table 1), and the AFLP reproducibility was high for all species (estimated to 95–99.4%, mean ± SD, 98.3% ± 1.07 ).
The number of genetic groups identified in the Structure analyses varied from two to seven (Fig. 1; Table 1). In all species with fully circumpolar ranges, one of the genetic groups was found in Beringia, often expanding across previously glaciated North America (red dots; Fig. 1) or expanding into Siberia (blue dots; Fig. 1). In all species, at least one genetic group was identified in Europe, and sometimes this group also occurred eastwards into Asia and/or westwards into Iceland and Greenland (green dots in Fig. 1). Several of the circumpolar dwarf shrubs showed similar patterns. For example, five main groups were identified in Cassiope tetragona ssp. tetragona: a Siberian group, a Beringian group, a Canadian group, an east Canadian/west Greenlandic group, and an east Greenlandic/Scandinavian group (Fig. 1). Similar geographical patterns with five groups were revealed in Betula nana s. lat. and Vaccinium vitis-idaea, although the ranges of their groups were somewhat different. Vaccinium uliginosum also showed a similar pattern with five main groups, but in this case the Beringian group extended across Canada, and the populations from the southern European mountains formed a group of their own. Separate groups in the European mountains were also found in several other species, such as Arctous alpina, Salix herbacea, Arabis alpina and Micranthes stellaris.
The metamap of genetic structure (Fig. 2) revealed that there is no genetic exchange across the North Pole, and that the overall strongest barriers to genetic exchange in the Arctic occur across the Arctic and Atlantic Oceans, across the Greenlandic ice cap, along the Ural Mountains, and between southern alpine areas and the Arctic. Weaker borders were indicated across the Bering Strait, southwards from Hudson Bay separating eastern North America, and around the largest rivers in Siberia, Lena and Kolyma (Fig. 2). Larger areas with strong connectivity, indicating high amounts of gene flow, were found within central Canada, between Fennoscandia, Iceland and the British Isles, in Siberia between the Lena River area and the Ural Mountains, and within northern European Russia.
The highest degrees of genetic diversity in the Arctic were found around the Bering Strait, especially on the Siberian side (Fig. 3a). The degree of diversity decreased pronouncedly both eastwards and westwards from Beringia towards the Atlantic Ocean. The decrease was gradual throughout North America except for a slight increase in southeastern Canada, while in Eurasia, there was a distinct drop in diversity from the eastern to the western side of the Ural Mountains. In nonarctic Eurasia, high diversity was observed in the European Alps, the Caucasus, and the Altai mountains (Fig. 3a).
The pattern of genetic distinctiveness was only partly congruent with the pattern of diversity. Much higher degrees of distinctiveness were observed in Eurasia, including the southern mountain regions, than in North America (Fig. 3b). In the Eurasian Arctic, the highest genetic distinctiveness was found in eastern and western Siberia. High distinctiveness was also recorded in parts of Scandinavia, central Europe and the European Alps. In North America, the pattern observed for genetic distinctiveness was rather similar to that for genetic diversity. The greatest distinctiveness was found in American Beringia, from where it gradually decreased eastwards towards the Atlantic coast and West Greenland, except for a somewhat higher level in southeastern Canada. In contrast to the pattern in diversity, genetic distinctiveness was somewhat higher in East Greenland than in West Greenland.
We found distinct overall trends that can be reasonably explained by a combined effect of glacial oscillations and more long-standing geographical barriers. Our data indicate several barriers to gene flow. As shown by Hoffmann (2012) based on species distributions, our results confirm that the North Pole area is an efficient barrier against gene flow. The three major genetic borders we identified within the Arctic are also concordant with long-standing physical barriers to gene flow: the Arctic and Atlantic Oceans, the Greenlandic ice cap, and the Ural Mountains (Fig. 2). The drop in diversity westwards of the Ural Mountains (Fig. 3a) lends further support to the hypothesis that this mountain range has been, and for low-arctic species still is, an efficient barrier to gene flow. However, it is possible that the Ural Mountains do not act as a barrier per se, but that the shift towards less continentality moving from east (Arkhipov et al., 2005) to west (Velichko et al., 2005) may filter establishment. One should also keep in mind that this climatic asymmetry has persisted throughout the Pleistocene, leading to recurrently more extensive glaciations in the west (Astakhov, 2008), which may generate a genetic boundary and shift in genetic diversity that coincide with the Ural region.
The fourth strong boundary identified, situated outside the Arctic and running through central Europe, can be explained by two factors. First, during interglacials such as the current one, the environmental conditions in lowland central Europe are not suitable for arctic–alpine plants, restricting gene flow between northern and southern populations. Secondly, during glacial times, many species had refugia in and adjacent to the central and southern European mountains that functioned as sources for interglacial colonization in the south (Tribsch & Schönswetter, 2003), while the northern region often was colonized from source populations situated further north, in unglaciated areas south and east of the ice (Skrede et al., 2006; Ehrich et al., 2007). Refugia in and adjacent to the European mountains are also supported by increased genetic distinctiveness (Fig. 3b; discussed further later).
The weaker borders identified in our meta-analysis (Fig. 2) may either indicate areas with less extensive congruence (i.e. barriers acting more efficiently against certain species than others), or represent less efficient barriers to gene flow for all species. Some barriers may also have changed through time, such as in unglaciated Beringia, where the eastern part was repeatedly connected to the western part via the Beringian land bridge during the long-lasting glaciations, whereas gene flow was interrupted by the Bering Strait during the short interglacials (Hopkins, 1967; Elias et al., 1996). That this boundary is depicted as weak in our analysis supports the idea that, taken over long time spans, the Bering Strait has functioned more as a bridge than as a barrier (DeChaine, 2008).
Our analyses do indicate some weaker borders that may indicate reduced gene flow across larger rivers, especially along the Omolon/Kolyma and Lena rivers in Siberia. Genetic borders along one or several of the large Siberian rivers, like the Omolon/Kolyma, Lena, Yenisei and Ob, have been shown for several animal species and species complexes, including lemmings, voles, shrews, hares and dunlins (Wenink et al., 1996; Fedorov et al., 1999, 2003; Fedorov & Stenseth, 2002; Brunhoff et al., 2003; Galbreath & Cook, 2004; Waltari et al., 2004; Hope et al., 2011, 2012). Rivers are, however, regarded as rather weak barriers for plants, so these borders may result from several factors acting in concert. The eastern Siberian montane glaciers suggested to have surrounded these river valleys during glacial periods (Barr & Clark, 2009, 2012) is one factor that could have contributed to reduced gene flow in these areas. Another possibility is that the overall genetic borders we identified may not be directly linked to physical barriers, but rather may reflect contact zones shared among different species. Contact zones are found in areas where differentiated genetic groups, expanding from different glacial refugia, met during the colonization process after glaciations (Hewitt, 1996, 1999, 2001). These contact zones do often correlate with physical barriers that slow down the expansion process. Although our metamap is rather rough, our data fit remarkably well with the compilation of contact zones done by Hewitt (2004a), where borders were identified along the Mackenzie, Kolyma and Lena River, near the Ural Mountains, and along the Greenlandic ice cap (Fig. 2).
Hewitt (2004a) also identified a contact zone along the location of the last remnants of the Scandinavia ice sheet. Recent studies have added further support to the existence of this contact zone (e.g. Skrede et al., 2006; Sonsthagen et al., 2011), but no boundary in Scandinavia emerged in our metamap. The lack of current physical barriers in this region may have diluted the effect of this contact zone.
Our data show high degrees of genetic diversity and distinctiveness in Beringia (Fig. 3a,b). This supports the existence of a large, long-standing refugium in this region, as hypothesized by Hultén (1937) based on phytogeographical evidence (see also Abbott & Brochmann, 2003). However, recent studies have identified several cryptic refugia within the Beringian region, as ‘refugia in the refugium’ (Abbott & Comes, 2004; Pruett & Winker, 2005, 2008; Shafer et al., 2010). The presence of such cryptic refugia would also increase the distinctiveness and diversity in this region. Still, none of the species we analyzed had more than one genetic group restricted to Beringia, leaving the alternative of multiple refugia within Beringia less likely, at least for currently widespread plant species.
Hultén further hypothesized that the majority of arctic plants initially radiated east- and westwards from Beringia, and reached circumpolar distributions before the onset of the Pleistocene glaciations (Hultén, 1937). The overall circumpolar diversity pattern we identified, with the highest diversity in Beringia, and reduced diversity when entering the formerly heavily glaciated areas in North America and Europe, fits nicely with Hultén's hypothesis of initial radiation from Beringia.
This decrease in diversity in formerly glaciated areas also fits the genetic model of leading edge expansion (Hewitt, 1996; Ibrahim et al., 1996); genetic diversity is lost via repeated bottlenecks during rapid expansion into formerly glaciated areas. This pattern has probably been enhanced through several glaciation cycles. The gradual decrease, not only in diversity but also in genetic distinctiveness, from Beringia and eastwards into North America, supports the idea that Beringia served as an important source of postglacial colonization of extra-Beringian areas.
The eastward expansion from Beringia apparently continued into West Greenland, that is, until meeting the Greenlandic ice cap, which is the only strong barrier to gene flow identified in the Nearctic and situated next to the barrier created by the Arctic and Atlantic Oceans. East Greenland is thus sandwiched between two major barriers against gene flow, and is apparently strongly isolated. There is also a weak increase in distinctiveness in this region, supporting isolation. The increased distinctiveness could reflect glacial survival in East Greenland, supporting the previous hypothesis that this area held an in situ glacial refugium (Funder, 1979; Westergaard et al., 2011). However, although our data clearly suggest that the Arctic and Atlantic Oceans are a strong barrier relative to continuously inhabitable continental areas, this barrier is not impermeable, as often stated in earlier phytogeographical literature (e.g. Dahl, 1963). Several previous molecular studies have shown gene flow across the Atlantic Ocean (e.g. Bennike, 1999; Aares et al., 2000; Hagen et al., 2001; Schönswetter et al., 2008; Skrede et al., 2009).
We found overall more genetic borders, more distinct areas of genetic connectivity, and more areas with high genetic distinctiveness in Eurasia than in North America. The Eurasian areas separated by the borders in Fig. 2 correspond to previously postulated refugial areas, such as Beringia, central Siberia, western Siberia, and central Europe (reviewed by, e.g., Weider & Hobæk, 2000; Hewitt, 2004a, 2011). Except for Beringia, there is also higher distinctiveness in these areas. Thus, our results suggest the existence of several smaller, separated refugia in Eurasia, and fewer, but larger, refugia in North America. Similar patterns have been found for more temperate species as well, where stronger genetic differentiation as a result of refugial isolation were found in Europe (Dumolin-Lapegue et al., 1997; Petit et al., 2003) than in North America (Magni et al., 2005). However, one should bear in mind that our sampling in central and southern range limits in North America were sparse for several species, and we may have lost signals of the other suggested refugial areas in North America (Shafer et al., 2010), including, for example, refugia within the boundaries of the Cordilleran ice sheet (Marr et al., 2008; Shafer et al., 2010), ice-free regions in the Canadian high Arctic, and areas south of the Laurentide ice sheet (Tremblay & Schoen, 1999). An exception might be the weak increase in diversity and distinctiveness we identified in southeastern Canada, as well as the border surrounding this area, which lends some support to former hypotheses of a refugium in the eastern North American coastal region (Steig et al., 1998; Tremblay & Schoen, 1999).
Examining the identified areas of connectivity, it is evident that northern Europe was not only colonized from the south, but was also heavily influenced from the east. This pattern has been shown for several species of both plants and animals (Taberlet et al., 1995; Jaarola & Searle, 2002; Palmé et al., 2003; Malm & Prentice, 2005; Knopp & Merilä, 2009; Sonsthagen et al., 2011). Further, there seems to be an overall close connection among the British Isles, Scandinavia, and Iceland, which is also a commonly found association in the literature (Abbott et al., 1995, 2000; Hagen et al., 2001; Westergaard et al., 2008; Thórsson et al., 2010).
The enhanced degrees of both diversity and distinctiveness observed in the southern mountain regions could result from survival and colonization from several smaller adjacent refugia, as shorter migration distances will retain more diversity, and intermixing of formerly separated lineages (hybrid zones) may increase diversity (Hewitt, 1996, 1999, 2004a; Petit et al., 2003). Many smaller refugia have been confirmed or suggested in and around southern mountain ranges in previous studies, for example, in the European Alps (Tribsch & Schönswetter, 2003; Schönswetter et al., 2005; Ehrich et al., 2007), and in the Caucasus and Altai (Skrede et al., 2006). The increased distinctiveness indicated in central Europe north of the Alps (Fig. 3a,b) is more puzzling. Only a few of the species we analyzed currently have occurrences in central Europe north of the Alps. Thus, the increased distinctiveness indicated in this region may reflect the occurrence of relict and/or small, isolated populations.
Our data also showed elevated degrees of distinctiveness within the former heavily glaciated region in Scandinavia. The presence of in situ glacial refugia may represent an explanation of the enhanced distinctiveness is this area as well. Such refugia have been indicated using various genetic methods for, for example, lemmings (Fedorov & Stenseth, 2001), Sagina caespitosa (Westergaard et al., 2011), and conifer trees (Parducci et al., 2012).
Our procedure to identify borders between the genetic groups included spatial GIS methods of assigning no-value grid cells to their nearest neighbor group, that is, drawing a borderline in the middle between assigned groups. A caveat here is that we do not know whether barriers always occur in the middle between different groups (this is also related to the uncertainty in the nonsystematic sampling of points). However, as our data set is large and widely distributed (17 species, each with two to seven genetic groups), we argue that the produced maps, on a northern hemisphere scale, reasonably reflect the genetic variation in our original data.
We conclude that the overall genetic structure in widespread arctic–alpine plants can be explained by long-standing physical barriers to gene flow combined with refugial isolation and expansions in response to glaciation/deglaciation (Fig. 2), and that the overall genetic diversity pattern in the Arctic does support the common perception of reduced genetic diversity in formerly glaciated areas. The formerly unglaciated Beringia served both as a major refugium and as a major source of interglacial colonization of widespread arctic–alpine plants. The high diversity in areas further west in Siberia also indicates long-term persistence, but the higher distinctiveness here suggests that these areas were less important as sources for large-scale (re)colonization of plants.
We thank Gro Hilde Jakobsen, Virginia Mirré, Liv Guro Kvernstuen, Siri Kjølner, Øyvind Stensrud, all collectors, and all others who at some point have been involved in the ARKTØK project. Special thanks to Inger Skrede and Kristine Bakke Westergaard, who were Masters students on the project, for full access to their already published data. The Research Council of Norway provided funding (grants 150322/720 and 146515/420 to C.B.).