Of glaciers and refugia: a decade of study sheds new light on the phylogeography of northwestern North America


Aaron B.A. Shafer, Fax: (780) 492-9234;
E-mail shafer@ualberta.ca


Glacial cycles have played a dominant role in shaping the genetic structure and distribution of biota in northwestern North America. The two major ice age refugia of Beringia and the Pacific Northwest were connected by major mountain chains and bordered by the Pacific Ocean. As a result, numerous refugial options were available for the regions taxa during glacial advances. We reviewed the importance of glaciations and refugia in shaping northwestern North America’s phylogeographic history. We also tested whether ecological variables were associated with refugial history. The recurrent phylogeographic patterns that emerged were the following: (i) additional complexity, i.e. refugia within refugia, in both Beringia and the Pacific Northwest; and (ii) strong evidence for cryptic refugia in the Alexander Archipelago and Haida Gwaii, the Canadian Arctic and within the ice-sheets. Species with contemporary ranges that covered multiple refugia, or those with high dispersal ability, were significantly more likely to have resided in multiple refugia. Most of the shared phylogeographic patterns can be attributed to multiple refugial locales during the last glacial maximum or major physiographic barriers like rivers and glaciers. However, some of the observed patterns are much older and appear connected to the orogeny of the Cascade-Sierra chain or allopatric differentiation during historic glacial advances. The emergent patterns from this review suggest we should refine the classic Beringian-southern refugial paradigm for northwestern North American biota and highlight the ecological and evolutionary consequences of colonization from multiple refugia.


Throughout the Quaternary, the distribution of the world’s biota has been shaped by climatic fluctuations. During this time, the ebb and flow of glaciers forced populations into several major ice age refugia. These glacial refuges were the location where species survived the ice ages, regardless of geography or size, and contained some degree of suitable habitat (Holderegger & Thiel-Egenter 2009). Within these refugia, populations were restricted and often isolated, becoming genetically differentiated over time. When the ice-sheets receded, species colonized newly available terrain in a leading edge expansion, resulting in the reduction of genetic variation in a clinal fashion (Hewitt 2004). However, increased genetic diversity in colonized populations did occur when lineages from separate refugia mixed (Petit et al. 2003). Historic subdivision and population isolation within refugia, termed refugia within refugia, also helped maintain genetic diversity (Gómez & Lunt 2007). In addition, some species persisted in cryptic glacial refugia (Stewart & Lister 2001; Provan & Bennett 2008) that contained quasi-hospitable climates capable of sustaining at least some temperate flora and fauna. The locations and internal complexity of these major and cryptic ice age refugia, and the subsequent colonization of deglaciated terrain, are partially responsible for shaping the genetic structure and distribution of contemporary biota (Hewitt 1996, 2000, 2004; Widmer & Lexer 2001).

Determining the location of refugia typically requires the knowledge of species distributions prior to glaciation (Waltari et al. 2007). Historically, glacial refugia have been inferred from paleoecological data. Fossil and pollen records provide historical snap-shots that are used to reconstruct refugial locations and communities; unfortunately, these methods have considerable shortcomings. Fossil evidence is often sparse or incomplete and pollen analyses have taxonomic and spatial limitations (see Ritchie 1995; McLachlan et al. 2005), especially in the Arctic and alpine (Birks & Birks 2000). The lack of sufficient data means that smaller refugia on islands and nunataks may never be discovered because they either are overlooked or leave no discernable evidence (Pielou 1991). For example, Haida Gwaii has long been hypothesized as a North American coastal refugium (Heusser 1989) and is widely applied in biogeographic models. However, paleoecological findings from the last glacial maximum are extremely limited for this region (i.e. Lacourse & Mathewes 2005; Wigen 2005). This paucity of paleoecological data makes it exceedingly difficult to infer refugial communities and colonization routes from postulated refugia.

Phylogeography provides a valid alternative for inferring refugia and colonization patterns. This approach does not require historical samples, as it is based on the geographic distribution of genealogical lineages that have evolved since the initial separation (Avise et al. 1987; Avise 2000). Under this framework, hypotheses regarding how populations responded to geologic and climatic fluctuations can be tested (Knowles 2001). Early phylogeographic models were very simple; the ‘southerly refugia model’ (Bennett et al. 1991) predicted a leading edge expansion from southern refugia to the north following glacial retreat, resulting in a latitudinal gradient of decreased genetic diversity (Hewitt 1996, 2000, 2004). These models have become obsolete as they do not take into account lineage mixing (Petit et al. 2003), additional complexity (e.g. refugia within refugia) (Gómez & Lunt 2007) or deeper historical associations (Lovette & Bermingham 1999). To account for such multifaceted phylogeographic histories, analytical methods have expanded to encompass coalescent and ecological niche modelling (Carstens & Richards 2007; Waltari et al. 2007) and phylogeography now pulls from a wide array of disciplines (Knowles & Maddison 2002; Knowles 2009). These advancements coincide with unprecedented growth in the field (Knowles 2009).

In the past decade, the expanding interest in phylogeography has led to the discovery of novel patterns and refugial sites. To better understand these complex patterns, researchers strive to find congruence among codistributed taxa and connect the observed patterns to underlying processes (Bermingham & Moritz 1998; Avise 2000). This approach is required to identify refugia and colonization routes in the absence of, and in addition to, paleoecological data. Given the plethora of studies and continued evolution of the field, it has become critical to periodically review the broader phylogeographic patterns among taxa within regions (e.g. Europe—Taberlet et al. 1998; Petit et al. 2002; North America—Avise 2000; Soltis et al. 2006; Jaramillo-Correa et al. 2009). There is now sufficient published data to systematically review the complex biogeographic history of northwestern North America and thereby identify shared patterns of colonization and identify cryptic refugia.

Reviewing the major patterns of northwestern North America

Ice-sheets covered much of Canada during the Pleistocene, reaching their largest extent some 20 000 years ago (Clark et al. 2009). The glacial advances in North America were among the most extensive worldwide (Velichko et al. 1997) and followed regular climatic intervals known as the Croll-Milankovich cycle (Imbrie 1985; Muller & MacDonald 1997). This cycle dictated glacial movement throughout the Pleistocene, causing biota to respond by shifting their range or going extinct. The fossil record shows large changes in species distributions that corroborate these cycles (Bennett 1997; Williams et al. 1998). For most northwestern biota, two large refugia were available; Beringia and the Pacific Northwest (Fig. 1; Hultén 1937; Pielou 1991). Species could have either retreated north, south or in both directions during glacial advances. This has resulted in the broad classification of colonizing taxa as either Beringian or southern in origin (Youngman 1975; Hoffman 1981).

Figure 1.

 Extent of ice during the last glacial maxima in northwestern North America. Ice layer is from Dyke et al. (2003) and is partially transparent to show the underlying mountain network. Major refugia and ice-sheets are labelled. Inset map contains the states and provinces of northwestern North America: AK—Alaska, YT—Yukon Territory, NWT—Northwest Territories, BC—British Columbia, AB—Alberta, WA—Washington, OR—Oregon, ID—Idaho, MT—Montana, WY—Wyoming.

While the glaciations are widely recognized as having significant impacts on species distributions across multiple continents (Taberlet et al. 1998; Soltis et al. 2006), the northwest of North America has a unique complexity because of the physiography of the region. There are essentially two contiguous north–south mountain ranges: the Rocky Mountains, which define the eastern extent of the northwest, and the Coast Mountains, which occur along the Pacific coast. Moreover, the Cascade/Sierra orogeny was recent enough to have detectable impacts on current species distributions (Graham 1999; Brunsfeld et al. 2001). Based on this orogeny, Brunsfeld et al. (2001) proposed a historic vicariant pattern between the coast and the northern Rockies.

Adding to the intricacies of this region is the potential existence of cryptic refugia. Soltis et al. (1997) suggested that coastal refugia (i.e. Haida Gwaii and the Alexander Archipelago) were important for surviving biota. In areas where glaciation was not complete, species could also have persisted in small within-ice refugia (e.g. nunataks; Pielou 1991). Moreover, the influence of Beringia was largely overlooked in earlier reviews of the northwest. In the decade since Brunsfeld et al. (2001), new patterns and processes have been uncovered in this region with many including a role for Beringia and cryptic refugia. Based on this, studies have called for the re-examination of the phylogeographic patterns and a thorough assessment of cryptic refugia in northwestern North America (e.g. Cook et al. 2001; Demboski & Cook 2001; Janzen et al. 2002; Fedorov et al. 2003; Spellman et al. 2007).

The aim of this review is to characterize the broad-scale phylogeographic patterns in northwestern North America and to therefore identify refugial locations and colonization routes. We have three major predictions: (i) the phylogeographic history of northwestern North America will be complex, but repeated patterns and substructure (i.e. refugia within refugia) will be evident in both plants and animals; (ii) major and cryptic refugia will be detected and supported by at least some paleoecological evidence; and (iii) certain ecological traits (e.g. habitat specialization) may be associated with refugial history (Table 1). Given the complex glacial and physiographic history of this region, we attempted to distinguish between older historical events (e.g. mountain orogeny) and more recent substructuring because of multiple glacial refugia. We also discuss the evolutionary and ecological implications of the inferred broad-scale phylogeographic patterns in northwestern North America.

Table 1.   Predicted phylogeographic patterns in northwestern North America
ISpecies will have persisted in refugia in either Beringia and/or the continental United States during glacial advances and share recolonization routes
IIAdditional structure, refugia within refugia, is likely present in the major refugia
IIISome species have persisted in cryptic refugia
IVEcological variables will be associated with refugial history
VGeological events and barriers may correspond to phylogeographic breaks


Study area

For the purpose of the review, we are considering northwestern North America to include the following states and province/territories: Alaska, Yukon, Northwest Territories, Alberta, British Columbia, Washington, Oregon, Idaho, Wyoming, and Montana (Fig. 1). The ecoregions within this area include the Pacific coast temperate rain forest, the northwestern forested mountains, western desert, taiga and tundra. The physiography of the region is varied, but notably two major mountain ranges run longitudinally: the Coast Mountains along the Pacific Ocean and the Rocky Mountains in the interior (Fig. 1).

Literature review

We first searched Web of Science using the keywords ‘phylogeography’ and ‘cryptic refugia.’ Selected studies had to encompass a portion of northwestern North America. Because biogeographic patterns of the northwest are intertwined with northerly refugia, we included ‘Beringia’ in our searches. A citation search for all papers referencing the early reviews of Soltis et al. (1997) and Brunsfeld et al. (2001) was also conducted. Our focus was on literature post-2000 since the reviews by Soltis et al. (1997) and Brunsfeld et al. (2001) summarized much of the earlier work that was carried out in the region. We included some earlier phylogeographic studies as supportive evidence where applicable. Retrieved papers were qualitatively assessed on: (i) sampling, (ii) observed pattern and (iii) robustness of pattern. Differing from Brunsfeld et al. (2001), we incorporated nonmesic taxa. We sought out additional paleoecological and ecological evidence in the literature to support the purported patterns.

Characteristics of species and refugial type

Following the methodological approach applied by Hickerson & Cunningham (2006), we tested whether certain ecological traits were correlated with the refugial history of vertebrates and plants. We used a dichotomous scale similar to that of Bhagwat & Willis (2008) with each species evaluated for (i) dispersal ability (0—low, vs. 1—high), (ii) habitat specialization (0—generalist, vs. 1—specialist) and (iii) size of contemporary range (0—encompassing a single refugium, or 1—multiple refugia). The dependant variable was survival during the Pleistocene in single or multiple refugia. The scoring schemata are presented in Appendix S1. In an effort to standardize how each ecological variable was scored across species, we utilized three major sources for data collection: the International Union for Conservation of Nature (http://www.iucn.org), NatureServe (http://www.natureserve.org/) and the United States Department of Agriculture (http://plants.usda.gov/). The majority of species had range maps and ecological/biological descriptions at these sites. Because we were examining a wide array of taxa, scoring was conservative and standardized when possible. If multiple studies were used to draw inferences on refugial history, species were only entered once in the data set. Because one of the covariates (i.e. range) appeared highly predictive, Firth’s penalized-likelihood logistic regression (Firth 1993; Heinze & Schemper 2002; Heinze 2006) was used to assess the strength and direction of the relationship between the ecological variables and refugial history. This method avoids issues of separation arising from highly predictive covariates and has been shown to provide good estimates of logistic regression coefficients (Heinze & Schemper 2002). Univariate models were conducted on the entire data set as well as on individual groups (mammals, birds, herpetofauna and plants). The models were evaluated by converting the logistic b coefficient to an odds ratio (OR) using the formula OR = eb, with the 95% confidence interval calculated using the formula: inline image.

Results and discussion

Literature review

Since 2000, over 4800 refereed papers were catalogued under the keyword ‘phylogeography.’ Of these, ‘Beringia’ and ‘western North America’ were listed as keywords 41 and 183 times, respectively. An additional 100 studies were listed under the keyword ‘cryptic refugia.’ All keywords showed a general increase in the number of publications over the past decade. Focusing on species that had a distribution encompassing a part of northwestern North America, 126 relevant phylogeographic studies were retrieved in our literature review (Table 2). Our qualitative assessment of northwestern North America’s phylogeographic patterns is included in Table 2. Overall, plants and mammals made up the bulk of the phylogeographic studies (60%), but insects, birds, fish, herpetofauna and parasites contributed to the review. We found that including nonmesic taxa revealed fewer plant studies (= 39) compared to vertebrate studies (n = 87). This contrasts with the review of Brunsfeld et al. (2001), but is similar to the review of eastern North America (Soltis et al. 2006). Unlike Soltis et al. (2006), mammals are not underrepresented in our review (n = 36) and we found no relevant turtle studies. The differences between the composition of our literature review and that of Brunsfeld et al. (2001) and Soltis et al. (1997) can largely be attributed to our inclusion of nonmesic taxa. The distinctive biotic composition in both northwestern and eastern North America accounts for some of the differences with Soltis et al. (2006).

Table 2.   Summary of major phylogeographic studies for northwest North America. Specific refugial sites are noted in the Refugia column. Substructure is in reference to phylogeographic breaks found within a particular refuge. The list of abbreviations is included below the table
TaxonCommon nameMarkersObserved patternAnalysis and support*ReferenceRefugia
  1. *Studies showing >80% bootstrap support or posterior probability were considered to have high support, with those below 80% considered moderate.

  2. Abbreviations for each column listed alphabetically, ‘Markers’: AFLP—amplified fragment length polymorphism, cp—chloroplast DNA, ISSR—inter-simple sequence repeat, min sat—mini-satellite, mtDNA—mitochondrial DNA, nuDNA—nuclear DNA, res sites—restriction sites, seq—sequence, μsat—microsatellite. ‘Observed Pattern’: AA—Alexander Archipelago, BC—British Columbia, E—East, HG—Haida Gwaii, LGM—last glacial maxima, Mts—Mountains, N—north, NRM—Northern Rocky Mountains, PNW—Pacific Northwest, RM—Rocky Mountains, S—south, W—west. ‘Analysis and Support’: AMOVA—analysis of molecular variance, ENM—ecological niche modeling, IBD—isolation by distance, ME—minimum evolution, ML—maximum likelihood, MP—maximum parsimony, NCA—nested clade analysis, NJ—neighbour joining, PCA—principal component analysis, SAMOVA—spatial analysis of molecular variance, UPGMA—unweighted pair-group method with arithmetic mean.

ReviewN/S split at Cascade/Sierra. HG refugeCo-distributed patternsSoltis et al. 1997    
ReviewNRM and Coastal cladesCodistributed patternsBrunsfeld et al. 2001    
Tamiasciurus spp.Tree squirrelsmtDNA seq, allozymesRefugia in NRM and Blue MtsHigh MP & moderate ML supportArbogast et al. 2001  
Tamias ruficaudusRed-tail chipmunkmtDNA seqE/W divide along Bitterroot Mts, multiple NRM cladesModerate to high Bayesian, MP, ML & NCA supportGood & Sullivan 2001  
Tamias amoenusYellow-pine chipmunkmtDNA seqCoastal and continental clades, two refugia in PNWModerate to high ML, MP & NCA supportDemboski & Sullivan 2003  
Tamias spp.Red-tail & yellow-pine chipmunksmtDNA seq, morphometricClearwater refuge, introgression zoneModerate MP, Bayesian & NCA support, discordant with morphologyGood et al. 2003  
Glaucomys sabrinusFlying squirrelmtDNA seqRecent colonization into AANo ML structure, reduced diversity on islandBidlack & Cook 2001   
Spermophilus paryiiArctic ground squirrelmtDNA seqMultiple clades in BeringiaHigh ML support, molecular clockEddingsaas et al. 2004  
Microtus oeconomusTundra volemtDNA seqBeringian cladeHigh NJ & MP support, molecular clock, nucleotide diversityBrunhoff et al. 2003   
Microtus oeconomusTundra volemtDNA seq, nucDNA seqMultiple clades in BeringiaHigh Bayesian support, molecular clock, nucleotide diversityGalbreath & Cook 2004  
Microtus longicaudusLong-tailed volemtDNA seqDistinct clades along coast, AA & RMHigh NJ & MP support, molecular clockConroy & Cook 2000;
Spaeth et al. 2009
Microtus richardsoniWater volemtDNA seqCoastal and NRM clades with coastal colonizationHigh ML support but not ENMCarstens et al. 2005; Carstens & Richards 2007  
Clethrionomys gapperiRed-backed volemtDNA seqCoastal and NRM cladesHigh MP, NJ & Bayesian support, mismatch & diversity analysesRunck & Cook 2005  
Peromyscus spp.Deer and Keen’s mousemtDNA seqN & S cladesHigh NJ support, coalescence based MIGRATEZheng et al. 2003  
Peromyscus keeniKeen’s mousemtDNA seqDistinct AA cladeModerate NJ support, NCA & molecular clockLucid & Cook 2004   
Peromyscus keeniKeen’s mousemtDNA seqP. keeni in Haines Junction distinct, potential Beringian originHigh NJ & MP support, high divergence among conspecificsLucid & Cook 2007    
Peromyscus maniculatusDeer micemtDNA seq, μsatsDistinct PNW & California clades mixing in Oregon, recent gene flowHigh ML & Bayesian supportYang & Kenagy 2009  
Phenacomys longicaudusRed tree volemtDNA seqN/S split in Oregon. E/W division across Willamette ValleyNetwork analysis, diversity measuresMiller et al. 2006b  
Dicrostonyx groenlandicusCollared lemmingmtDNA seqCanadian Arctic & Beringian cladesModerate to high NJ, ML & MP support, molecular clockFedorov & Stenseth 2002  
Lemmus spp.LemmingsmtDNA seqCanadian Arctic & Beringian cladesHigh NJ & ML support, population expansion modelsFedorov et al. 2003  
Sorex spp.ShrewsmtDNA seqCoastal (Oregon to Alaska) & continental cladesHigh MP & ML support, high divergenceDemboski & Cook 2001  
Ochotona princepsAmerican pikamtDNA seqE/W split & N/S split in RMHigh ML support, molecular clock, population expansion modelsGalbreath et al. 2009  
Lepus arcticusArctic haremtDNA seqCanadian Arctic & Beringian cladesHigh Bayesian & low NJ support, haplotype distributionWaltari & Cook 2005  
Mustela ermineaErminemtDNA seqBeringian, S, continental, AA & HG cladesModerate to high MP, ML & NCA supportFleming & Cook 2002 
Martes americanaAmerican martenmtDNA seq, res sitesCoastal/RM & continental cladesHigh MP support & network analysis, high divergenceStone et al. 2002   
Ursus americanusBlack bearmtDNA seqCoastal (HG) & continental clade, Hecate refuge potential HG sourceHigh MP supportByun et al. 1997  
Ursus americanusBlack bearmtDNA seq, res sitesCoastal clade to Oregon/CaliforniaHigh MP supportStone & Cook 2000   
Canis lupisGray wolfmtDNA seqS refugeLow NJ support, haplotype & diversity distributionsLeonard et al. 2005   
Canis lupisGray wolfmtDNA seqBC inland & coastal differentiationHaplotype distribution, divergenceMunoz-Fuentes et al. 2009   
Gulo guloWolverinemtDNA seqSingle refuge & rapid colonizationModerate NJ support, diversity measuresTomasik & Cook 2005   
Gulo guloWolverinemtDNA seq, μsatsSingle Beringian refugeStar NJ tree, NCA & PCA supportCegelski et al. 2006   
Vulpes vulpesRed foxmtDNA seqBeringian refuge, distinct S refuge in PNWHigh ML & Bayesian support, haplotype distributionAubry et al. 2009  
Odocoileus hemionusMule deermtDNA seqBlack-tail deer refuge in Oregon/WashingtonHigh MP & NCA support, diversity measuresLatch et al. 2009  
Ovis spp.Bighorn and thinhorn sheepmtDNA seqCryptic refugia in McKenzie Mts & N BCAMOVA, network analysis, molecular clockLoehr et al. 2006
Oreamnos americanusMountain goatmtDNA seq, μsatsN and S clades. Potential refugium in N BCHigh ML & Bayesian support, distribution of diversityShafer et al. 2010 
Rangifer tarandusCariboumtDNA seqBeringian & S refugiaModerate/high Bayesian support, distribution of haplotypesFlagstad & Roed 2003  
Ambystoma macrodactylumLong-toed salamandermtDNA seqHG, coastal, Salmon River, Blue Mts, Clearwater & Montana cladesLow ML, ME & MP support, NCA, shallow divergenceThompson & Russel 2005  
Anaxyrus boreasWestern toadmtDNA seq, res sitesNW clade & Klamath–Siskiyou Mts refuge, Columbia River refugeHigh MP & Bayesian support, shallow divergenceGoebel et al. 2009  
Ascaphus trueiTailed frogmtDNA seq, allozymesN & S clades along coast, Klamath–Siskiyou Mts refugeHigh MP & ML support, network analysis, molecular clockNielson et al. 2001, 2006  
Ascaphus montanusTailed frogmtDNA seq, allozymesClearwater & Salmon River Mts refugiaHigh MP & ML support, network analysis, molecular clockNielson et al. 2006  
Ascaphus spp.Tailed frogmtDNA seqCoastal & NRM cladesML support, ENMCarstens et al. 2005; Carstens & Richards 2007  
Batrachoseps wrightiOregon salamandermtDNA seq, RAPDN/S split in Oregon & N colonization, Columbia River refugeHigh NJ & MP support, distribution of diversityMiller et al. 2005  
Plethodon vandykei & P. idahoensisVan Dyke’s salamandermtDNA seqNRM & coastal clades, Clearwater drainage refugeHigh ML & MP support, NCA, molecular clockCarstens et al. 2004  
Plethodon elongatus & P. stormiDel Norte salamandermtDNA seq4 clades, Cascades/Sierra & Klamath–Siskiyou refugiaHigh MP, ML & ME support, high divergenceMahoney 2004  
Plethodon larselliLarch Mountain salamandermtDNA seq, RAPDColumbia River split & N colonizationModerate NJ & MP support, NCA, distribution of diversityWagner et al. 2005  
Plethodon spp.Plethodontid salamandersmtDNA seqCoastal & NRM clades, ancient vicarianceHigh ML & ENM supportCarstens et al. 2005; Carstens & Richards 2007  
Dicamptodon spp.Giant salamandersmtDNA seqCoastal & NRM cladesHigh ML supportCarstens et al. 2005  
Dicamptodon spp.Giant salamandersmtDNA seqCoastal refuge for D. aterrimusHigh MP & ML support, Bayesian hypothesis testingSteele et al. 2005   
Dicamptodon tenebrosusPacific giant salamandermtDNA seqN/S clades, Columbia River Valley & Klamath–Siskyou Mts refugiaHigh MP & ML support, Bayesian hypothesis testingSteele & Storfer 2006  
Dicamptodon copeiCope’s salamandermtDNA seqMultiple western clades, Columbia River refuge, N colonizationModerate MP, ML & Bayesian support, NCASteele & Storfer 2007  
Rhyacotriton variegatusSouthern torrent salamandermtDNA seqCentral Oregon split, Yaquina River barrier, N colonizationNetwork analysis, AMOVA, spatial autocorrelationMiller et al. 2006a  
Taricha granulosaRough-skinned newtmtDNA seq, allozymesOregon split, N colonization from Klamath–Siskiyou refugeModerate NJ, ML & Bayesian support, distribution of diversityKuchta & Tan 2005  
Charina bottaeRubber boamtDNA seqPNW Clade & N/S split at Sierra NevadaModerate MP & ML supportRodriguez-Robles et al. 2001   
Crotalis viridisWestern rattlesnakemtDNA seqE/W split at RM, effect of NRM & coast orogenyModerate MP & ML support, Templeton testsPook et al. 2000  
Crotalis viridisWestern rattlesnakemtDNA seqE/W split at RM, PNW basal for C. v. oreganusModerate MP & ML supportAshton & de Queiroz 2001  
Thamnophis sirtalisGarter snakemtDNA seqHG & S clades, N and S colonizationHigh ML support, hypothesis testingJanzen et al. 2002  
Pseudacris regillaPacific tree frogmtDNA seqDistinct coast & NRM cladesHigh MP, ML & ME support, network analysisRipplinger & Wagner 2004  
Pseudacris regillaPacific tree frogmtDNA seqDistrict Coast & NRM clades & a split in OregonModerate MP & Bayesian support, NCARecuero et al. 2006  
Rana luteiventrus, R. pretiosiaColumbia spotted frogmtDNA seqSeparate RM clade, distinct Wyoming and Coast groupsHigh MP & ML support, NCABos & Sites 2001  
Rana luteiventrus & R. pretiosaColumbia spotted frogmtDNA seqDistinct Coast & NRM clades, unique Blue Mts clade.High MP, ML & Bayesian supportFunk et al. 2008  
Oporornis tolmieiMacGillivray’s warblermtDNA seqHighest diversity in Oregon, likely N colonizationNJ & MP support, MST, distribution of diversityMila et al. 2000   
Dendroica petechiaYellow warblermtDNA seqE/W split corresponding to RMModerate NJ support, distribution of diversityMilot et al. 2000  
Wilsonia pusillaWilson’s warblermtDNA seq, res sitesCoastal and RM divergence, highest diversity in Alberta & AlaskaNetwork analysis, distribution of diversityKimura et al. 2002  
Dendroica coronataYellow-rumped warblermtDNA seqPartial differentiation between E & W, Oregon basalHigh ML & NJ support, network analysisMila et al. 2007   
Geothylpis trichasCommon yellowthroatmtDNA seqE, W & Nevada groupsNetwork analysis, distribution of diversityLovette et al. 2004   
Poecile rufescenChestnut-backed chickadeeμsatsDistinct lineages in HG & AlaskaBayesian assignment, distribution of diversity, private allelesBurg et al. 2006 
Poecile gambeliMountain chickadeemtDNA seqCoastal & RM cladeHigh ML support, network analysis, MDIVSpellman et al. 2007  
Melospiza melodiaNorthwestern song sparrowμsatsHG refuge, N colonizationHigh Bayesian & NJ support, distribution of diversity & allelesPruett & Winker 2005  
Troglodytes troglodytesWinter wrenmtDNA seqE/W split potentially from PleistoceneHigh ML support, DIVA, distribution of diversityDrovetski et al. 2004   
Troglodytes troglodytesWinter wrenmtDNA seq, AFLPE/W split at RM, pre-PleistoceneBayesian assignment, PCA, molecular clockToews & Irwin 2008   
Catharus ustulatusSwainson’s thrushmtDNA seqCoast & RM split from PleistoceneSequence divergence, distribution of diversity, expansion modelsRuegg & Smith 2002  
Cyanocitta stelleriStellar’s JayμsatsHG & S clusters, likely refugiaBayesian assignment, distribution of diversity & alleles, high FSTBurg et al. 2005  
Branta canadensisCanada goosemtDNA seq, μsatsMultiple clades in Beringia, likely multiple refugiaHigh ML, ME & MP supportScribner et al. 2003 
Dendragapus obscurusBlue grousemtDNA seqCoast & NRM clades, N/S split within NRMMP support, distribution of diversity, estimates of gene flowBarrowclough et al. 2004  
Lagopus mutusRock ptarmiganmtDNA seq, nuDNA seqArctic, Beringia, & Aleutian Island cladesModerate NJ support, network analysis, distribution of diversityHolder et al. 1999, 2000 
Grus canadensisSandhill cranemtDNA seqArctic cladesHigh ML & NJ support, coalescent molecular clockRhymer et al. 2001  
Grus canadensisSandhill cranemtDNA seq, μsatsBeringian clade & secondary contactBayesian assignment, PCA, distribution of diversityJones et al. 2005   
Aix sponsaWood duckmtDNA seqCoast & RM split, Pleistocene divergenceNetwork analysis, distribution of diversity, molecular clockPeters et al. 2005  
Strix occidentalisSpotted owlmtDNA seqCascade & Sierra Nevada split, distinct Washington cladeMP support, network analysis, distribution of diversityBarrowclough et al. 1999  
Salvelinus alpinusArctic charrmtDNA seqS, Beringian, & Arctic cladesModerate, MP, NJ & ML support, haplotype distributionBrunner et al. 2001 
Gobbiesox maeandricusNorthern clingfishmtDNA seqN & S coastal cladesModerate ML support, network analysisHickerson & Ross 2001  
Salvelinus malma & S. confluentisDolly varden and bull troutmtDNA seq, nuDNA seqBeringian & S clades for Dolly varden. S clade for bull troutModerate NJ & ML support, distribution of haplotypesRedenbach & Taylor 2002 
Thymallus arcticusArctic graylingμsats, mtDNA seq, res sitesN/S split in Beringia & likely refuge in Nahanni RiverModerate NJ & ML support, distribution of haplotypesStamford & Taylor 2004 
Salvelinus namaycushLake troutmtDNA res sitesN & S clades & likely a Nahanni River refugeModerate ML support, distribution of haplotypesWilson & Hebert 1998 
Pinus flexilisLimber pinemtDNA res sitesRefuge in N Wyoming (E foothills)Distribution of alleles, degree of differentiationMitton et al. 2000  
Pinus flexilisLimber pineallozymesE & W refugia for NRM populationHigh among population GST, UPGMA supportJorgensen et al. 2002  
Pinus albicaulisWhitebark pinemtDNA seq, cpSSRYellowstone, Columbia basin, & Oregon clades, N colonizationAMOVA, spatial distribution of haplotypesRichardson et al. 2002  
Pinus lambertianaSugar pinecp seqN/S split at Cascade-Sierra interface, N refuge in WashingtonModerate MP support, distribution of diversityListon et al. 2007  
Pinus balfourianaFoxtail pinecp, mtDNA, nuDNA seqRefugia in Klamath Mts & S Sierra Nevada MtsBayesian clusters, coalescent-based isolation with migrationEckert et al. 2008  
Pinus contortaLodgepole pinemtDNA, min satMontana, Columbia River basin, Cascades, HG/AA & Beringia cladesBayesian assignment, UPGMA, distribution of diversityGodbout et al. 2008
Picea sitchensisSitka sprucecp STSS refuge & possibly HGStrong spatial structure, no bottleneck signatureGapare & Aitken 2005; Gapare et al. 2005  
Picea glaucaWhite sprucecp seqS clade & refuge in AlaskaSAMOVA, KST, spatial distribution of haplotypesAnderson et al. 2006  
Salix melanopsisDusky willowcp seqClearwater & Salmon River clades, W colonization to the coastModerate ML & Bayesian support, divergence valuesBrunsfeld et al. 2007  
Salix melanopsisDusky willowcp seqCoastal & NRM cladesML support but not ENMCarstens et al. 2005; Carstens & Richards 2007  
Larix lyallii & l. occidentalisSubalpine and western larchμsatsE/W split at NRM, likely two distinct refugia for both speciesNJ & UPGMA support, distribution of diversityKhasa et al. 2006  
Boechera spp.Rockcresscp seqMontana/Idaho refuge, potential Arctic or Beringian refugiaModerate MP support, network analysis, distribution of diversityDobes et al. 2004b  
Boechera spp.Rockcresscp, nucDNA seq, μsatsN/S split in WNA, Cascades basal to NRMNJ support, distribution of alleles & diversityDobes et al. 2004a   
Boechera spp.RockcressμsatsN/S split in NRM, suggestive of multiple refugiaBayesian assignment, distribution of diversitySong et al. 2006  
Boechera spp.Rockcresscp seqPossible Klamath–Siskiyou & Blue Mts refugiaNetwork analysis, distribution of haplotypes & diversityKiefer et al. 2009  
Packera paucifloraAlpine groundselcp res sitesN colonization & refuge in BeringiaNetwork analysis, AMOVA, distribution of diversityBain & Golden 2005  
Sedum lanceolatumYellow stonecropcp seqRM split at Wyoming basin, possible refuge in N. MontanaDistribution of diversity & alleles, high differentiationDeChaine & Martin 2005  
Townsendia hookeriEaster daisiescp seq, cp res sitesBeringia & S refugia, refuge within LGM in SW AlbertaDistribution of diversity, ploidyThompson & Whitton 2006 
Packera spp.Ragwort/groundselcp res sitesRefugia in SW Alberta on nunataks in P. pseudaurea and P. conterminaDistribution of haplotypesGolden & Bain 2000  
Chrysosplenium iowenseIowa golden saxifrageISSRS refuge, but Alberta population from refuge within icePCoA, distribution of diversity, high differentiationLevsen & Mort 2008  
Cardamine constanceiConstance’s bittercresscp seq4 NRM refugia (Clearwater)Moderate ML & Bayesian support, NCA, high divergenceBrunsfeld & Sullivan 2005  
Oxyria digynaMountain sorrelcp res sitesRefugia in NBC & western USA, N colonizationNCA, spatial distribution of haplotypesMarr et al. 2008  
Saxifraga oppositifoliaPurple saxifragecp res sitesBeringia refuge, suggested Arctic refugeModerate MP & NJ support, spatial distribution of haplotypesAbbott et al. 2000  
Saxifraga oppositifoliaPurple saxifragecp res sitesRefuge in Beringia, two distinct clusters in AlaskaDistribution of diversity, high differentiationAbbott & Comes 2003  
Oxytropis spp.LocoweednuDNA seq, RAPDRefugia in NE Alaska and S AlaskaModerate ML & UPGMA support, AMOVAJorgensen et al. 2003  
Dryas integrifoliaMountain avenscp res sitesRefugia in Beringia and high ArcticModerate NJ support, diversity & differentiation of haplotypesTremblay & Schoen 1999  
Daphnia pulexDaphnidmtDNA res sitesPotential refugia in Canadian ArcticNJ support, distribution of haplotypes & diversityWeider et al. 1999   
Melanoplus spp.GrasshoppersmtDNA seqMultiple NRM refugia in Montana and IdahoML, MP & NJ support, coalescent modelsKnowles 2001  
Prophysaon coeruleumArionid slugmtDNA seqKlamath Mts refugia, effect of orogenyModerate Bayesian & ML supportWilke & Duncan 2004   
Dendroctonus rufipennisSpruce beetlesmtDNA seq, μsatsBeringian, S & PNW refugia with deep divergencesModerate MP & Bayesian support, Bayesian assignmentMaroja et al. 2007 
Adelges cooleyiGallmtDNA seq, AFLPCoastal and RM splitModerate NJ support, Bayesian assignment, NCAAhern et al. 2009  
Greya politellaSeed parasitemtDNA seq, AFLPNRM refugia in Salmon, Bitterroot Rivers & southern Oregon distinctModerate MP & Bayesian support, network analysisRich et al. 2008  
Soboliphyme baturiniNematodemtDNA seqCoastal & interior division, AA refugeModerate Bayesian support, distribution of diversityKoehler et al. 2009 
Lobaria pulmonariaLichenμsatsCoastal & interior divergence, possible Vancouver Island refugeUPGMA, AMOVA, distribution of diversityWalser et al. 2005 
Cavernularia hulteniiLichennuDNA seqBeringia & S split in WNA. Possible Pacific Island refugeNetwork analysis, diversity distributions (mismatch)Printzen et al. 2003 
Armillaria ostoyaeFungusnuDNA seqCoastal & RM splitHigh NJ, MP & Bayesian supportHanna et al. 2007  
Tricholoma matsutakeFungusnuDNA seq, AFLPRM basal, coastal split corresponds to Sierra/Nevada orogenyModerate MP, NJ & ML support, IBDChapela & Garbelotto 2004  

Ecological variables associated with refugia l history

We found suitable data on 103 vertebrate and plant species (Appendix S2) to assess the association between contemporary range, habitat specificity and dispersal ability with refugial history. Scoring of multiple refugia was based on phylogeographic data and refugial scenarios suggested by the original authors (Appendix S3). All studies had patterns consistent with Pleistocene refugia. Phylogeographic breaks that could not be distinguished from contemporary or pre-Pleistocene influences were not included. We used penalized regression to offset any bias in sample size and because some explanatory variables almost perfectly predicted the dichotomous refugial history (Heinze 2006). Habitat specialists appeared less likely to have persisted in multiple refugia (Table 3). On the other hand, we found a positive and significant relationship between multiple refugia and dispersal ability (= 1.82) and contemporary range (= 2.86). The OR of dispersal and range was significantly above one, indicating that species with high dispersal ability and large contemporary ranges were more likely to have resided in multiple refugia during glacial advances (Table 3). Analyses of individual groups showed the same general pattern but with lower levels of significance (Table 3).

Table 3.   Univariate Firth’s penalized-likelihood logistic regression results. Data on ecological variables and phylogeographic history were collected from 103 species in northwestern North America (Appendixes S1 and S2)
Phylogeographic historyEcological variableOdds ratio95% confidenceP value
Multiple refugiaRange overlap
Habitat specificity
Dispersal ability

After applying a penalty to the log-likelihood, the odds ratio of contemporary range with respect to multiple refugia was the highest across the entire data set and within groups (Table 3). Similar positive associations between northern refugia and range, habitat specialization and dispersal ability were observed for European plants and vertebrates (Bhagwat & Willis 2008). Svenning & Skov (2004) found that northern European trees filled most of their potential range (relative to southern species) and attributed the difference to dispersal ability. Efficient dispersal of northern populations would have facilitated efficient range shifts during glacial advances and possibly permitted gene flow between disjunct populations. Moreover, there appears to be a direct link between phylogeography and dispersal ability, such that historical substructure within species can create conditions that promote efficient dispersal (discussed further in the Ecological and Evolutionary Implications section). Overall, these data suggest species with large contemporary ranges and high dispersal ability are significantly more likely to have resided in multiple refugia.

Biotic responses to the changing Pleistocene climate

During glacial advances, most species in western North America retreated to major refugia in either Beringia or the Pacific Northwest (Fig. 1: Hultén 1937; Pielou 1991). Some widespread plant species like rockcress, Boechera spp. (Dobes et al. 2004a) and white spruce, Picea glauca (Anderson et al. 2006), along with mammals such as the ermine, Mustela erminea (Fleming & Cook 2002), caribou, Rangifer tarandus, (Flagstad & Roed 2003) and red fox, Vulpes vulpes, (Aubry et al. 2009) occupied both major refugia. However, it quickly becomes apparent that there is a considerable amount of variation and complexity within species that challenge this simplistic view of single northern and southern origins for western North America’s biota. In terms of the biotic response, we focused on three major areas: (i) south in the Pacific Northwest; (ii) north in Beringia and iii) within areas overlooked or previously undiscovered (i.e. cryptic refugia).

South of the ice-sheets

Refugia in the Pacific Northwest were originally split into two mountainous regions: the Cascade/Coast Range and the northern Rockies (Fig 3; Brunsfeld et al. 2001). This phylogeographic split is still identified as a predominant pattern among comparative studies (Carstens et al. 2005; Jaramillo-Correa et al. 2009). Within vertebrates, we see this split in herpetofauna like salamanders Plethodon vandykei and Plethodon idahoensis (Carstens et al. 2004), Pacific tree frog, Pseudacris regilla (Ripplinger & Wagner 2004; Recuero et al. 2006), western toad, Anaxyrus boreas (Goebel et al. 2009) and the spotted frogs, Rana luteiventrus and Rana pretosia (Bos & Sites 2001; Funk et al. 2008). We also see a similar pattern in the American pika, Ochotona princeps (Galbreath et al. 2009), long-tailed vole, Microtus longicaudus (Conroy & Cook 2000; Spaeth et al. 2009) and yellow-pine chipmunk, Tamias amoenus (Demboski & Sullivan 2003). Blue grouse, Dengragapus obscurus, show this same split (Barrowclough et al. 2004) which resulted in recent recognition as distinct species (Banks et al. 2006).

Figure 3.

 Postulated colonization routes in northwestern North America. Large blue arrows indicate major routes, while smaller yellow indicate smaller or postulated routes.

In mammals, a larger break of the Coast Range from the continent is commonly observed for species endemic to the northern coniferous forests (Stone et al. 2002; Arbogast 2007; Yang & Kenagy 2009) or distributed across North America (Runck & Cook 2005; Latch et al. 2009). This overarching phylogeographic split between the Coast Range and northern Rockies/continent can largely be attributed to the orogeny of the Cascade/Sierra chain that occurred 5-2 million years ago (Graham 1999; Brunsfeld et al. 2001). The rise of this mountain range created a rain shadow, leading to the xerification of the Columbia River basin and thus a disjunct mesic forest (Daubenmire 1975). Where utilized, molecular dating of clade divergences often corresponds to this event (e.g. Wilke & Duncan 2004; Toews & Irwin 2008). Some other shared phylogeographic splits involving the mountain chains are not specifically from orogeny, but allopatric diversification during historic glacial advances (Johnson & Cicero 2004; Weir & Schluter 2004).

In addition to the east–west split, there is also a north (British Columbia to Oregon) south (Oregon to California) separation within taxa in the Coast/Cascade region that is commonly observed (Soltis et al. 1997; Jaramillo-Correa et al. 2009). This split has become affectionately known as the ‘Soltis line’ (Fig 3; Brunsfeld et al. 2007). Examples of this split range from the blue-grey tail-dropper slug, Prophysaon coeruleum (Wilke & Duncan 2004), torrent salamander, Rhyacotriton variegatus (Miller et al. 2006a), to sugar pine, Pinus lambertiana (Liston et al. 2007). Explanations for the ‘Soltis line’ are less obvious as the break is dynamic (angled dashed line in Fig. 2) and occurs in the absence of any obvious physiographic break. Soltis et al. (1997) attempted to reconcile this pattern by proposing different refugial-colonization scenarios resulting from Pleistocene glaciations (Soltis et al. 1997). Our accumulated data (Table 2) suggest both multiple Pleistocene refugia in the Pacific Northwest (Fig. 2) and northern colonization (which were Soltis et al.’s (1997) original hypotheses) could explain the ‘Soltis line.’

Figure 2.

 Important biogeographic sites within identified refugia. Putative refugial locations are labelled, while important rivers and phylogeographic breaks (dotted line) are shown. In some cases, rivers acted as refugial sites. Bordered-dashed areas indicate potential cryptic refugia. Gray lines bordered by black denote separate ice caps inferred from Dyke et al. (2003) that may have acted as barriers.

While the large-scale patterns are east–west and north–south separations in the southern refugium, the literature suggests more complex patterns and structure within both the coastal and northern Rocky regions (Fig. 2). In plants, Godbout et al. (2008) found evidence for three separate refugia of lodgepole pine, Pinus contorta, in the Cascades, Columbia River basin, and east of the Rockies in Montana; these refugia are supported by the fossil record (Baker 1976; Mehringer et al. 1977; Mack et al. 1978; Carrara et al. 1986). Similarly, phylogeography of the whitebark pine, Pinus albicaulis suggested the coastal mountains, Clearwater basin and Yellowstone were distinct refugia (Richardson et al. 2002), some of which are supported by fossil pollen (Baker 1990). In the Wyoming basin, a phylogeographic split for yellow stonecrop, Sedum lanceolatum occurs, suggesting two distinct refugia in this region (DeChaine & Martin 2005). Brunsfeld & Sullivan (2005) and Brunsfeld et al. (2007) also found four distinct groups of Constance’s bittercress, Cardamine constancei in the Clearwater River drainage, as well as a north–south split between the Clearwater and Salmon rivers in dusky willow, Salix melanopsis. These latter two refugia are also observed in herpetofuana (Nielson et al. 2001, 2006; Carstens & Richards 2007). In the long-tailed vole, individuals from Wyoming and Colorado are basal in their phylogeny (Conroy & Cook 2000), indicative of Pleistocene refugium on the eastern side of the Rockies. In red-tailed chipmunks, Tamias ruficaudus, an east–west split occurs in the northern Rockies as well as a subdivision within the Clearwater drainage (Good & Sullivan 2001): all of these examples indicate historical substructure and multiple refugia within the northern Rockies.

We also observed multiple refugial sites in the Cascade/Coast range. For Stellar’s jay, Cyanocitta stelleri (Burg et al. 2005), Canada goose, Branta canadensis (Scribner et al. 2003), and the spotted owl, Strix occidentalis (Barrowclough et al. 1999), the Coast Mountains are implicated as a refugium, but the specific locations could not be determined. More precise refugial locations are inferred from herpetofauna. The Columbia River area was suggested as a refuge for the Oregon salamander, Batrachoseps wrighti (Miller et al. 2005), and Larch Mountain salamander, Plethodon larselli (Wagner et al. 2005). The Klamath–Siskiyou Mountains appeared to be a refugium for the tailed frog, Ascaphus truei (Nielson et al. 2001, 2006), Del Norte salamander, Plethodon elongatus (Mahoney 2004), and rough-skinned newt, Taricha granulosa (Kuchta & Tan 2005). Some species like the western toad (Goebel et al. 2009) and salamanders, Dicamptodon spp. (Steele & Storfer 2006, 2007), likely resided in both of these refugia. Additional support for these refugia comes from the phylogeographic studies of both foxtail pine, Pinus balfouriana (Eckert et al. 2008), and rockcress (Kiefer et al. 2009), which suggested the Klamath–Siskiyou Mountains was their refugial location. The Blue Mountains in northeast Oregon have also been identified as a regional refuge (Arbogast et al. 2001; Thompson & Russel 2005; Nielson et al. 2006; Carstens & Richards 2007; Funk et al. 2008; Kiefer et al. 2009). Phylogeographic patterns from Dolly Varden, Salvelinus malma and bull trout, Salvelinus confluentus suggested the Chehalis River in Washington was a coastal refuge (Redenbach & Taylor 2002).

In terms of the physiographic associations, water refuges are clearly important for aquatic taxa, and some of the above phylogeographic substructure can be attributed to geographic barriers like rivers and canyons, elevation and volcanic events (Wakabayashi & Sawyer 2001; Monsen & Blouin 2003; Miller et al. 2006a; Takacs-Vesbach et al. 2008; van Tuinen et al. 2008; Galbreath et al. 2009). Much of the codistributed taxa show the major divisions between the northern Rockies and coastal region, and the ‘Soltis line’ along the Pacific coast (Soltis et al. 1997; Brunsfeld et al. 2001). These large breaks are explained by mountain orogeny and Pleistocene glaciations. But the vast majority of phylogeographic structure appears to be related to multiple refugial locales in the Pacific Northwest. There is considerable historic substructure and support for the refugia within refugia model in the Pacific Northwest. Similar to refugia on the Iberian Peninsula (Gómez & Lunt 2007), this historic substructure suggests genetically differentiated populations persisted throughout Pleistocene glaciations in the Pacific Northwest and have been an important source of genetic variability in colonizing taxa.


Phylogeographic studies have shown that Beringia was a refuge for numerous species during the Pleistocene glaciations. Molecular evidence from plants such as mountain avens, Dryas intergrifolia (Tremblay & Schoen 1999), purple saxifrage, Saxifraga oppositifolia (Abbott et al. 2000; Abbott & Comes 2003), locoweeds, Oxytropis spp. (Jorgensen et al. 2003), white spruce (Anderson et al. 2006) and Townsend’s daisy, Townsendia hookeri (Thompson & Whitton 2006) support Beringia as a refuge. Mammals that show this pattern include lemmings, Lemmus and Dicrostonyx (Fedorov & Stenseth 2002; Fedorov et al. 2003), tundra voles, Microtus oeconomus (Brunhoff et al. 2003; Galbreath & Cook 2004), thinhorn sheep, Ovis dalli (Loehr et al. 2006), collared pikas, Ochotona collaris (Galbreath et al. 2009), Alaska marmot, Marmota broweri (Steppan et al. 1999), Arctic shrew, Sorex arcticus (Fumagalli et al. 1999) and brown bears, Ursus arctos (Leonard et al. 2000; Barnes et al. 2002). Many of these mammals have fossil evidence supporting this pattern (e.g. Weber et al. 1981; Storer 2004).

Like the Pacific Northwest, additional substructure is observed within Beringia. Abbott & Comes (2003) and Jorgensen et al. (2003) found divergent Alaskan clades of saxifrage and locoweeds. In mammals, tundra voles show Beringian splits that appear to have been induced by glaciers (Galbreath & Cook 2004), and thinhorn sheep show evidence for a previously undetected refugium in southeast Beringia (Loehr et al. 2006). A distinct haplotype in Keen’s mouse, Peromyscus keeni from Haines Junction, Yukon (Lucid & Cook 2007) suggests a similar pattern. Ground squirrels (Spermophilus parryii) show at least four distinct clades in Alaska where their bifurcations date to glacial events (Eddingsaas et al. 2004). In addition, compelling evidence for an inland freshwater refugium southeast of Beringia that centred on the Nahanni river exists for both lake trout, Salvelinus namaycush (Wilson & Hebert 1998) and Arctic grayling, Thymalus arcticus (Stamford & Taylor 2004). This refugium was situated at the juncture of the Cordilleran and Laurentide ice-sheets that appeared to contain periglacial freshwater lakes throughout the Pleistocence (Dyke & Prest 1987). The overarching cause of the Beringian splits is still unclear, but Jorgensen et al. (2003) attributed the breaks to the coastal ice shield (Fig. 2). The Yukon River Delta is also situated between phylogeographic splits (Eddingsaas et al. 2004) and likely helps to maintain vicariant signals. Given the ubiquitous nature of glaciations and both the Yukon River and ice shield as potential barriers, they should prove to be a shared phylogeographic break among more of the region’s taxa (Riddle 1996). Moreover, these extrinsic factors have served to create and maintain genetic diversity in Beringia throughout multiple Pleistocene glacial cycles.

Beringia also reveals the utility of phylogeography for reconstructing refugial communities when paleoecological data produced different scenarios. Fossil evidence for wolverines, Gulo gulo pointed to the possibility of both southern and Beringian refugia (Bryant 1987). However, extensive sampling and phylogeographic analyses found no evidence for a southern refugium (Tomasik & Cook 2005; Cegelski et al. 2006). This suggests the southern lineage went extinct (or has had limited success), and Beringia has been the major source for North American wolverines. A slightly different pattern is observed in mountain goats, Oreamnos americanus, where fossils only support a southern refugium (Cowan & McCrory 1970; Rideout & Hoffman 1975). The distribution of genetic diversity and haplotypes points to a second northern refugium for mountain goats (Shafer et al. 2010). Thus, instead of the predicted northern decrease in diversity (Hewitt 2004), mountain goats showed discrete hotspots of diversity in both the north and south (Shafer et al. 2010). In both examples, by utilizing both phylogeographic and paleoecological data, a clearer picture of the refugial origin and distribution of diversity was produced.

Cryptic refugia: the Alexander Archipelago and Haida Gwaii

There is considerable evidence in the literature for multiple refugia off the coast of British Columbia and Alaska. Vancouver Island has been suggested as a coastal refugium (Heusser 1960; Pojar 1980) and some genetic (Walser et al. 2005; Godbout et al. 2008) and paleoecological (Ward et al. 2003) evidence support this. But the vast majority of studies suggest that Haida Gwaii and the Alexander Archipelago were the major coastal refugia. The chestnut-backed chickadee, Poecile rufescens (Burg et al. 2006), Stellar’s jay (Burg et al. 2005) and northwestern song sparrow, Melospiza melodia (Pruett & Winker 2005) all showed increased genetic diversity on Haida Gwaii—a pattern consistent with refugia (Hewitt 1996). Haida Gwaii and/or the Alexander Archipelago have been suggested as a refuge for garter snakes, Thamnophis sirtalis (Janzen et al. 2002), the water flea, Daphnia pulex complex (Weider et al. 1999), and the lichen, Cuvernularia hultenii (Printzen et al. 2003).

For plant species, phylogeographic evidence of refugia along coastal Alaska and northern British Columbia is accumulating (Gapare & Aitken 2005; Gapare et al. 2005; Godbout et al. 2008). Fossil plant and pollen records also support refugia in this area (Warner et al.1982; Peteet 1991; Hansen & Engstrom 1996). Most contentious is whether mammals utilized these islands as a refuge during the last glacial maximum. Byun et al. (1997) postulated that the coastal clade of black bears, Ursus americanus, near Haida Gwaii were from an island refuge (but see Demboski et al. 1999; Byun et al. 1999). Expanded sampling of black bears has failed to fully resolve the debate (Stone & Cook 2000; Peacock et al. 2007). However, phylogeographic support is observed in the long-tailed vole (Conroy & Cook 2000) and Keen’s mouse, with the latter having fossil evidence pre-dating the Holocene (Lucid & Cook 2004). Mountain goats too may have survived on the Alexander Archipelago (Shafer et al. 2010), and fossil evidence does support the presence of an ungulate species around the last glacial maximum (Heaton & Grady 2003). Phylogeography of the ermine and a codistributed nematode, Soboliphyme baturini, also support an island refugium (Fleming & Cook 2002; Koehler et al. 2009).

Vertebrate fossils are limited on Haida Gwaii (Wigen 2005); but considerable fossil evidence from the Alexander Archipelago supports the persistence of available terrestrial habitat (Heaton et al. 1996; Dixon et al. 1997; Heaton & Grady 2003). Geological evidence shows portions of this area were ice-free (Scudder & Gessler 1989; Josenhans et al. 1995; Carrara et al. 2007). Collectively, the accumulation of phylogeographic, fossil and geological data provide near definitive evidence for a Pleistocene refugium that included mammals. One scenario is an extended coastal refugium including Haida Gwaii, the Alexander Archipelago and an exposed Queen Charlotte Sound and Hecate Strait. Haida Gwaii was connected to the ice-free coast during the Pleistocene by the Hecate Strait, and both the Hecate Strait and Queen Charlotte Sound were at various times exposed, containing freshwater lakes and terrestrial flora (Barrie et al. 1993; Josenhans et al. 1995; Hetherington et al. 2003, 2004; Lacourse et al. 2003, 2005): this appears to be what prompted Byun et al. (1997) to suggest Haida Gwaii as a refugium for bears. An exposed Hecate Strait could have facilitated gene flow to the coast and southward, but also potentially north to the Alexander Archipelago. Although a glacier is believed to have separated Haida Gwaii from Alaska (Bornhold & Barrie 1991; Barrie & Conway 1999) open water stretches were considerably shorter (Barrie & Conway 1999). The potential ice or water barrier would therefore have been limited and potentially navigable by terrestrial mammals, or simply bypassed by aerial dispersers. This extended coastal refugium is supported by shared fossil assemblages (Heaton et al. 1996; Heaton & Grady 2003; Wigen 2005) and genetic ancestry (Fleming & Cook 2002; Cook et al. 2006) of the islands. Such a scenario could be used to explain the widespread coastal haplotype in bears (Byun et al. 1997; Demboski et al. 1999; Stone & Cook 2000) and essentially doubles the available area making it more plausible that terrestrial vertebrates could have survived the glacial maxima in this region.

Cryptic refugia: the Arctic

In the far north, the Canadian Arctic has been suggested as an important refuge (Macpherson 1965; Pielou 1991). Fossil evidence is limited (Harington 1990) and those fossils found like caribou (Stewart & England 1986) and muskox, Ovibos moschatus (Maher 1968) lack accompanying phylogeographic support. However, a wide variety of taxa show phylogeographic evidence of an Arctic refugium. Arctic hare, Lepus arcticus (Waltari & Cook 2005), saxifrage (Abbott et al. 2000) and mountain aven (Tremblay & Schoen 1999) populations likely resided in Arctic refugium. Dunlin, Calidris alpina (Wennerberg 2001), Canada goose (Scribner et al. 2003) and rock ptarmigan, Lagopus mutus (Holder et al. 1999) appeared to utilize the Arctic throughout the Pleistocene. Some aquatic biota like Arctic charr, Salvelinus alpinus (Brunner et al. 2001) and water fleas (Weider et al. 1999) also persisted in an Arctic refuge.

With the accumulation of data supporting an Arctic refugium (as well any cryptic refugia for that matter), a refinement of biogeographic paradigms begins to arise. Beringia is considered one of the two major sources of biota in the northwest (Hultén 1937; Pielou 1991); however, this is not the case for all taxa. In lemmings, their phylogeographic pattern suggests they utilized an Arctic refuge throughout the Pleistocene (Fedorov & Stenseth 2002; but see Fedorov et al. (2003) for a possible periglacial refuge). But most interestingly, Fedorov et al. (2003) noticed a reduced geographic distribution of Beringian haplotypes in lemmings. This led the authors to suggest that Beringia was only a minor source for postglacial colonization of lemmings and thus the biogeographic models of Beringia and temperate North America may not be applicable to all northern species (Fedorov et al. 2003). It is premature to propose a full shift in biogeographic paradigms for the northwest, but these studies question the Beringian model as the major Pleistocene refuge and biotic source for the Arctic and northwest. Patterns like that observed by Fedorov et al. (2003) will undoubtedly be species dependent, but future studies should consider and test the influence of Arctic refugia. Such examples provide hypotheses to be tested in the statistical phylogeographic framework and will help refine and rewrite the role of major refugia in shaping contemporary biotic distributions.

Refugia within the ice-sheets

In addition to refugia flanking the Laurentide and Cordilleran ice-sheets, evidence for refugia within the ice-sheets during the last glacial maximum is mounting. In mountain sorrel, Oxyria digyna, a highly cold tolerant species, the haplotype distribution suggests colonization northward to the Arctic from northern British Columbia (Marr et al. 2008). This region was mostly covered by ice-sheets, but scattered nunataks (Ryder & Maynard 1991) may have supported persistent populations. A thinhorn sheep refugial population also appears to have persisted in northern British Columbia (Loehr et al. 2006). In southwestern Alberta, phylogeographic evidence suggests numerous plants survived within the ice-sheets including groundsel, Packera spp. (Golden & Bain 2000) and Easter daisies (Thompson & Whitton 2006) (see also Levsen & Mort 2008). Geological evidence has led to speculation of an ice-free corridor (Jackson 1979; Rutter 1984) or scattered nunataks (Burns 1980; Dyke & Prest 1987; Pielou 1991) that may have sustained disjunct plant populations in this region. Many of the ecological attributes (e.g. cold tolerance) thought to promote survival on nunataks are not restricted to these species (Marr et al. 2008), which suggests other northwest taxa may have survived on such refuges. These findings suggest additional colonization routes (Fig. 3) and the need to refine the Beringian and southern paradigm for the biota of northwestern North America. Similar to the Arctic refuge, future studies in the northwest should test for the presence of nunatak refugia, especially if the ecological characteristics of the species in question would promote survival in cryptic refugia.

Colonization post-Pleistocene

Dispersal ability was associated with a history of multiple refugia (Table 3). However, postulating the colonization routes and dispersal corridors from these refugia proves to be a more challenging task. For highly vagile taxa, colonization appeared to be so efficient that current population structure is a mixture of genetic lineages from different Pleistocene refugia. For example, mule deer, Odocoileus hemionus (Latch et al. 2009), the common raven, Corvus corax (Omland et al. 2000) and Canada goose (Scribner et al. 2003) all show considerable mixing of Pleistocene lineages in the northwest, making colonization routes virtually impossible to deduce. On the other hand, taxa that have limited dispersal abilities exhibit population structure that often strongly reflects their Pleistocene distributions. This is true for many herpetofauna, in which limited dispersal ability and habitat specialization acted together to limit their spread post-Pleistocene (e.g.Nielson et al. 2001; Carstens et al. 2005; Carstens & Richards 2007; Funk et al. 2008).

Despite the difficulty, colonization routes have been traced for many plants based on phylogeographic study. These studies come largely from south of the ice-sheets, but northerly routes have also been inferred (Fig. 3). The inland dispersal hypothesis and ancient vicariance models (Brunsfeld et al. 2001) are still accurate for some mesic taxa (Carstens et al. 2005; Carstens & Richards 2007). In contrast, both the dusky willow, (Carstens et al. 2005; Brunsfeld et al. 2007) and whitebark pine (Richardson et al. 2002) likely colonized the coast from the northern Rockies. In addition, whitebark pine (Richardson et al. 2002) and lodgepole pine (Godbout et al. 2008) show a northern colonization route that diverges north of the Columbia River basin and appears to follow mountain ranges. Species surviving within the ice-sheets (e.g. Marr et al. 2008) provide novel sources from which colonization would have emanated (Fig. 3). In many instances, demographic expansion statistics corroborate colonization from Beringian (Fedorov et al. 2003; Galbreath & Cook 2004), southern (Lessa et al. 2003; Shafer et al. 2010), and cryptic refugia (Fedorov et al. 2003; Lessa et al. 2003). There is also evidence that the physiography of mountains helped direct dispersal. Bull trout show a northern colonization pattern that appears to utilize mountain river systems (Redenbach & Taylor 2002). Southerly dispersal along the Coast Mountains has been suggested for garter snakes (Janzen et al. 2002), and landscape genetic patterns of northwestern mammals such as thinhorn sheep (Worley et al. 2004), wolverines (Schwartz et al. 2009), black bears (Cushman et al. 2009), and mountain goats (Shafer et al. 2010) all support mountain ranges acting as corridors.

Ecological and evolutionary implications

Two important issues arise from discovering novel refugia. The first is accurate reassessment of migration rates and modelling species-specific responses to climate change. Provan & Bennett (2008) noted the importance of understanding past phylogeographic patterns with respect to predicting future range shifts in species affected by climate change. Phylogeography is important because migration rates that are inferred when there are unrecognized cryptic refugia will be overestimated (McLachlan et al. 2005; Svenning & Skov 2007). Thus, predictions of species-specific responses to climate change are likely flawed without correct phylogeographic and refugial inferences. The second issue is identifying units of conservation. Bhagwat & Willis (2008) found evidence that survival in northern refugia was associated with unique biogeographic traits. These localities likely harbour high levels of genetic diversity and should be of highest conservation priority (Bhagwat & Willis 2008). If refugial locations are on the periphery of the range, there is an added urgency (Hampe & Petit 2005). Future studies in northwestern North America should recognize the importance of phylogeography for climate models and identifying conservation units. For example, McLachlan et al. (2005) re-evaluated migration rates of two tree species in eastern North America based on chloroplast DNA reconstructions; this approach could easily be applied in the northwest. In addition, given the peripheral, endemic and refugial origin of biota on the Alexander Archipelago and Haida Gwaii, conservation and management plans should recognize the unique phylogeographic status of these islands, especially with anthropogenic pressures mounting (Cook et al. 2006).

Some of the patterns we describe have identifiable evolutionary and ecological implications. Similar to our results (Table 3), European biota showed an association between dispersal ability and refugial history, such that habitat generalists and efficient dispersers were more likely to have survived in northern refugium (Bhagwat & Willis 2008). Svenning & Skov (2004) also found northerly distributed plants were better dispersers; but it is difficult to tease apart the cause-and-effect relationship between refugial history and dispersal ability. Evidence suggests the refugia within refugia scenario can produce highly efficient dispersers. Admixture between divergent plant populations often produces high-ploidy groups (Abbott & Brochmann 2003). In northern climates, increased ploidy is associated with successful colonization of recently deglaciated terrain (Brochmann et al. 2004). Polyploids may be adaptable to a wider array of ecological conditions (Stebbins 1950) and be more likely to maintain genetic variability during long-distance dispersal and bottleneck events (Brochmann et al. 2004). This pattern has also been observed in Arctic water fleas (Dufresne & Hebert 1997). Thus, a link is formed between refugial history and dispersal, such that (at least in some plants and animals), the refugia within refugia scenario produces efficient dispersers and can partly explain the success of some northerly distributed taxa.

Speciation and local adaptation are also important evolutionary consequences of multiple refugia. North American mountain sheep provide a textbook example of glacial-induced speciation and differentiation during the Pleistocene (Cowan 1940; Pielou 1991; Geist 1999). Thinhorn sheep arose from Beringian refugium while bighorn sheep, Ovis canadensis, evolved south of the ice-sheets. During the most recent glacial advance, isolation and differentiation produced detectable morphological and genetic differences between populations of thinhorn sheep (Worley et al. 2004). Variation in thinhorn sheep coat colour arose during this time because of hybridization with bighorn sheep in an isolated Pleistocene refuge (Worley et al. 2004; Loehr et al. 2006) and now plays an important role in dominance hierarchies (Loehr et al. 2008). Similarly, glacial-induced vicariance also promoted speciation and morphological divergence in northwestern pikas, Ochotona spp. (Guthrie 1973; Galbreath et al. 2009; Hafner & Smith 2010). Locally adapted variation occurring in part from refugial history has been also observed. One dramatic example comes from lake whitefish, Coregonus clupeaformis, where multiple glacial refugia led to divergence among populations in eastern North America (Bernatchez & Dodson 1990, 1991). Secondary contact and sympatric divergence among these lineages produced unique ecotypes (Pigeon et al. 1997) that are morphologically and ecologically distinct. Rogers & Bernatchez (2007) and Renaut et al. (2010) found that differential natural selection on adaptive traits likely maintains these ecotypes. With genome-wide scans becoming increasingly accessible, future studies may reveal the full extent to which adaptive variation in populations may stem in part from their phylogeographic history.

Future directions and review of major findings

The accuracy with which phylogeographic patterns are inferred is influenced by a number of interacting factors. One major limitation has been the lack of extensive sampling. Collection of tissues in phylogeographic studies is often opportunistic and may not adequately cover a species distribution. As a result, additional substructure and confidence in phylogenetic topologies may be limited. In our review, we found numerous instances where a species was analysed multiple times in the literature and when more samples were included, additional (or alterations to) refugial locations were inferred (e.g. tailed frog, Nielson et al. 2001, 2006; Carstens et al. 2005, Nielson et al. 2006; Columbia spotted frog, Bos & Sites 2001; Funk et al. 2008; dusky willow, Carstens et al. 2005; Carstens & Richards 2007; Brunsfeld et al. 2007). We predict that more intensive sampling will yield similar substructure and identify additional refugial sites in many species. Although collecting samples from the complete species range would help alleviate this problem, it is clearly not feasible logistically or financially in many cases. One solution is the development and utilization of integrated field inventories and permanent archives. This allows researchers the ability to access information and material for additional study. For example, Kuhn et al. (2010) utilized the Parks Canada DNA Repository to augment their ancient DNA with modern samples, enabling them to examine temporal changes in caribou population structure. Online databases also allow for comparisons among laboratories around the world (see Wang et al. 2009 for a good example). Such repositories and databases depend upon cooperation and support from parties often with different interests, but are a necessity if we wish to ameliorate the costs and difficulties associated with sample collection.

Another issue is that phylogeographic inference has largely relied on a single locus. The markers of choice, mainly mtDNA and cpDNA, have received tremendous scrutiny in the literature (e.g. Hurst & Jiggins 2005; Zink & Barrowclough 2008) stemming from their sometimes-inaccurate depiction of population and species histories. For example, in the montane frog, Rana cascadae, mtDNA suggested high levels of divergence, almost to the order of separate species, for the Olympic Peninsula population (Monsen & Blouin 2003). When nuclear markers were examined, it became clear that the Olympic Peninsula population was not isolated, and gene flow with other Washington populations had been ongoing. In such instances conservation designations and management practices would be misinformed if based solely on mtDNA. These examples of mitochondrial-nuclear discordance are rampant in the literature, which has led to multiple independent nuclear genes being put forth as a potential solution for phylogeography (Hare 2001). However, as evidenced by Table 2, nuclear genes have not yet been widely utilized in phylogeographic studies, likely because of a limited number of appropriate markers.

One possible solution may come from the development of genome-wide resources for nonmodel organisms made possible by the reduced costs of next generation sequencing and single-nucleotide polymorphism (SNP) screening (Gilad et al. 2009; Pool et al. 2010; Thomson et al. 2010). SNPs have slow mutation rates and limited homoplasy making them useful markers for inferring population history (Brumfield et al. 2003; Brito & Edwards 2009). Genome-wide approaches have recently been used to examine the phylogeography of balsam poplar, Populus balsamifera (Keller et al. 2010) and Lycaeides butterflies (Gompert et al. 2010). More importantly, these resources are ripe for the burgeoning field of statistical phylogeography that has shifted away from single-gene trees and towards accounting for genealogical discordance and genetic stochasticity when inferring population histories (Knowles 2009). However, a few limitations still need to be considered with these new genetic markers. Conceptually, genomic resources do not produce a result that is as visually intuitive as gene trees (Brito & Edwards 2009); thus a shift in thought process is required. Unlike what is performed with mtDNA, recombination (Nachman 2001) and ascertainment bias (Brumfield et al. 2003) should be considered, as they can impair phylogeographic interpretations. But overall, there is clear promise to these markers and they are being embraced by phylogeographers (Holsinger 2010). As the phylogeographic toolbox expands to include genome-wide data sets, the improved power and resolution will make questions that were inconceivable at one-point, become routine in nonmodel organisms.

Improved phylogeographic inferences are also being made with the advancement of analytical and statistical methods (Knowles 2009; Nielsen & Beaumont 2009). Already the field has seen considerable growth with the application of coalescent analysis (Knowles & Maddison 2002; Richards et al. 2007). Coalescent approaches allow for the testing of specific phylogeographic hypotheses (Carstens & Richards 2007; Hickerson et al. 2010) providing a quantitative assessment of tree divergence. Although these models are based on certain assumptions that are not likely met in reality (i.e. panmixia, stable population sizes), which could lead to errors in calculating divergence times (Hickerson et al. 2006, 2010), software programs now exist (see Excoffier & Heckel 2006; Kuhner 2008) that can untangle complex demographic parameters (e.g. population size and growth), along with estimating the time of divergence. The use of Approximate Bayesian Computational framework could further improve phylogeographic inference (Hickerson et al. 2006; Beaumont et al. 2010; Bertorelle et al. 2010). This approach can incorporate demographic histories that will affect coalescent times (Hickerson et al. 2006, 2007) and can be used in a comparative framework, which would allow for testing the congruence of divergence times across taxa (Hickerson et al. 2006). Ecological niche modelling (ENM) is another important method being added to the phylogeography toolbox (Peterson 2001; Carstens & Richards 2007). Researchers can use ENM to predict species’ distributions pre-Pleistocene and empirically evaluate how predicted historical refugia correspond to the phylogeographic structure (Carstens & Richards 2007; Knowles et al. 2007; Waltari et al. 2007). Landscape genetics (see Manel et al. 2003; Storfer et al. 2007) can also be used in conjunction with phylogeography to assess the habitat associations that have helped produce phylogeographic structure. Most importantly, these advances bring phylogeographic analysis towards a statistically valid comparative approach, rather than purely ‘descriptive phylogeography,’ which is an important step forward for the field.

In the decade since the review of Brunsfeld et al. (2001), empirical studies have revealed a wealth of phylogeographic information from the biota of northwestern North America. The recurrent phylogeographic pattern that emerged was additional complexity, i.e. refugia within refugia, in both the Beringia and southern refugia. This substructure is connected to mountain orogeny and common physiographic features like rivers, mountains and the Alaskan ice shield. There was also near conclusive evidence for multiple cryptic refugia in the Alexander Archipelago and Haida Gwaii, the Canadian Arctic, as well within the ice-sheets. These cryptic refugia force us to refine the classic two-refuge paradigm of Beringian and southern sources for the colonization of northwestern North America. In the next decade as the field transitions from being descriptive to statistical in nature, future phylogeographers should view the purported patterns as hypotheses to test under a more rigorous statistical framework. Because phylogeography is such an integrative field we stress that all available information, especially paleoecological data, should be used in conjunction with new methodology to help formulate and test phylogeographic hypotheses (Cruzan & Templeton 2000). More comprehensive sampling schemes (made available at permanent archives), genomic data and new analytical methods should be used in future phylogeographic studies. Less conventional approaches like ancient DNA (Krajick 2002), hotspot clusters (Swenson & Howard 2005) and parasites (Criscione et al. 2005) can also be used to reconstruct biogeographic scenarios and patterns. With the ever-expanding phylogeographic toolbox, novel patterns and ecological traits associated with refugial history will continue to be discovered, along with our understanding of the ecological and evolutionary consequences of Pleistocene glaciations.


This work was supported by Alberta Ingenuity and Natural Sciences and Engineering Research Council of Canada (NSERC) scholarships to ABAS. The Alberta Conservation Association (ACA) also contributed. SDC and DWC are both supported by NSERC and ACA grants. Special thanks to Aaron’s father, William Shafer, for his interest and support of this research. Thanks to Bryan Carstens for conversations on the biogeography of the region. This manuscript benefited from the valuable comments of four anonymous reviewers.

Aaron Shafer is finishing his PhD (at the University of Alberta) on the evolutionary history and population structure of mountain goats. His research interests include molecular ecology and evolution of mammals. Catherine Cullingham is currently a postdoctoral fellow at the University of Alberta where she is examining genomics of lodgepole and jack pine populations in western Canada to contribute towards understanding mountain pine beetle spread. Steeve Côté is a professor in the Biology Department of Université Laval and senior scientist at the Centre for Northern Studies. He has conducted research on ungulates in the Arctic and Rocky Mountains for over two decades. Dave Coltman is a professor in the Department of Biological Sciences of University of Alberta. His lab is focused on molecular ecology and has worked on a wide range of critters in northwestern North America.