Living on the edge: Exploring the role of coastal refugia in the Alexander Archipelago of Alaska

Abstract Although islands are of long‐standing interest to biologists, only a handful of studies have investigated the role of climatic history in shaping evolutionary diversification in high‐latitude archipelagos. In this study of the Alexander Archipelago (AA) of Southeast Alaska, we address the impact of glacial cycles on geographic genetic structure for three mammals co‐distributed along the North Pacific Coast. We examined variation in mitochondrial and nuclear loci for long‐tailed voles (Microtus longicaudus), northwestern deermice (Peromyscus keeni), and dusky shrews (Sorex monticola), and then tested hypotheses derived from Species Distribution Models, reconstructions of paleoshorelines, and island area and isolation. In all three species, we identified paleoendemic clades that likely originated in coastal refugia, a finding consistent with other paleoendemic lineages identified in the region such as ermine. Although there is spatial concordance at the regional level for endemism, finer scale spatial and temporal patterns are less clearly defined. Demographic expansion across the region for these distinctive clades is also evident and highlights the dynamic history of Late Quaternary contraction and expansion that characterizes high‐latitude species.

coastal archipelagos, where dynamic glacial advances potentially restructured entire communities repeatedly.
Areas previously glaciated were presumably colonized from multiple ice-free (Beringian, southern continental) regions during the late Pleistocene-early Holocene (14 to 10 kya), with independent recolonization from disparate sources hypothesized to have shaped the contemporary genetic structure of coastal biota. Due to eustatic and isostatic fluxes at the LGM, the Alexander Archipelago (AA) of Alaska and Haida Gwaii of British Columbia experienced sea levels up to 165 m lower (Baichtal, Carlson, & Crockford, 2008;Hetherington et al., 2003;Shugar et al, 2014); however, there remains substantial uncertainty regarding the extent of glaciation in this region (Buma et al., 2013;Carrara, Ager, & Baichtal, 2007;Carrara, Ager, Baichtal, & VanSistine, 2003;Elias, 2013;Fladmark, 1979). The earliest initiation of deglaciation in the region is now estimated at about 17,000 ybp (Lesnek, Briner, Lindqvist, Baichtal, & Heaton, 2018). Although many of the islands were buried under 1,000 m of ice, coastal refugia may have persisted along the exposed western continental shelf (e.g., Baichtal & Carlson, 2010;Fladmark, 1979). This Coastal Refugia Hypothesis remains poorly documented, however, and hinges on whether species persisted and diverged in isolation through the LGM, thereby becoming significant source F I G U R E 1 Sampling scheme, range maps, and North American LGM glacial cover. Sampling localities are shown by major cytb lineage. The thick black lines are the current range for each species, with the addition of P. maniculatus (white line) on the Peromyscus map. The light blue in the bottom right image is LGM glacial ice cover. COP: Colorado Plateau; LGM: Last Glacial Maximum; NPC: North Pacific Coast. Color schemes for species and lineages are held constant across figures populations for recolonization of deglaciated island and continental areas in northwestern North America (Byun, Koop, & Reimchen, 1999Demboski, Stone, & Cook, 1999).
As glaciers receded and sea levels rose, the islands of the AA became increasingly fragmented, although the sequence of fragmentation is complex due to isostatic rebound (Carrara et al., 2007). Subsequent in situ diversification across the AA hypothetically produced endemic populations due to either long-term occupation of the region (paleoendemics that originated from coastal refugia) or recent colonization from sources outside the region (neoendemic; Cook, Dawson, & MacDonald, 2006;Cook & MacDonald, 2013).
Long-term field studies MacDonald & Cook, 2007) have now produced the density of sampling across the archipelago necessary to begin to explore this complexity for mammals.
The AA is one of the planet's most extensive archipelagos with >1,100 named islands including 7 of the 15 largest United States islands. Together with Haida Gwaii to the south, these archipelagos support a significant portion of the remaining coastal temperate rainforest worldwide (DellaSala et al., 2011;Ecotrust & Conservation International, 1992). Most of the islands within this archipelago are managed by the Tongass National Forest (6.9 million ha; United States Geological Survey, 2010) and have been heavily modified by industrial timber harvests and associated fragmentation (e.g., roads) over the past 60 years (Albert & Schoen, 2013;List, 2000;Schoen & Dovichin, 2007). The rugged and ice-laden Coast and Wrangell-St. Elias mountain ranges that border the adjacent mainland to the east and north have acted as barriers to dispersal that filtered the movement of species into and out of the region (Cook & MacDonald, 2013).
Small mammals are optimal for exploration of comparative phylogeographic signatures because they are relatively abundant, widespread, and have limited vagility. We examine the role of historical climate variability in structuring contemporary genetic variation of two rodents, Microtus longicaudus and Peromyscus keeni, and a shrew, Sorex monticola; all are widely co-distributed throughout the AA and adjacent mainland. Previous analyses of mitochondrial DNA variation of a reduced set of localities in the region uncovered significant inter-population variation in these mammals (Conroy & Cook, 2000;Demboski & Cook, 2001;Lucid & Cook, 2004). Although these species are frequently sympatric, M. longicaudus (an herbivore) tends to prefer more open herbaceous habitats, P. keeni (an omnivore) typically occurs in forest and scrub habitats, and S. monticola (an insectivore) usually occupies forested and non-forested habitats with dense ground cover (Van Horne, 1981, 1982Smith & Belk, 1996;Smolen & Keller, 1987;Zheng, Arbogast, & Kenagy, 2003). The dietary isotopic niches of these species largely do not overlap as well (O'Brien, Cook, & Newsome, 2017). If all three species expanded from shared refugia, genetic signatures tracking their expansion histories may be congruent due to the common influence of climatic events (i.e., top-down environmental control), regardless of specific ecological differences or niche requirements.
In this study, we test three related questions. First, we assess the effects of geological history and climatic conditions on phylogeographic signatures of M. longicaudus, P. keeni, and S. monticola, with a focus on testing the Coastal Refugium Hypothesis (Fladmark, 1979).
We use species distribution models (SDMs; Figure 2) and historical bathymetric reconstructions ( Figure 3) across species to identify areas of endemism and their spatiotemporal relationship to potential refugia, including coastal refugia.
Next, we test whether phylogeographic signatures are concordant across the three species. An expectation of shared history should result in signatures reflecting similar responses to climatic events and geographic barriers and corridors. Conversely, if taxa have idiosyncratic histories, we expect to observe highly distinctive phylogeographic structure.
Lastly, we integrate the niche models, bathymetry, and demographic analyses to test for concordant signatures of demographic and spatial expansion to assess whether the three species responded similarly to the warming trends of the early Holocene. Signatures of historic expansion may be expected in island populations as a result of post-glacial colonization following increased island connectivity that occurred when sea levels were lower during glacial periods.  (Hall, 1981) found in or near Southeast Alaska for each species were represented. Closely related outgroup species (n = 3 Microtus, 40 Peromyscus, and 9 Sorex) were also sequenced. Additionally, we F I G U R E 2 Species distribution models (climate suitability at each time period) for Microtus longicaudus, Peromyscus keeni and Sorex monticola from the Last Inter-Glacial (LIG), Last Glacial Maximum (LGM; solid blue = glacial ice coverage), Mid-Holocene (Mid-Hol.), Current, and Future (approximately the year 2080), including the change in habitat suitability between current and future models used GenBank sequences representing 41 M. longicaudus, and 18 P. maniculatus (Supporting Information Appendix S1).
Posterior probabilities (PP) for nucleotides ≥0.85 were chosen; otherwise ambiguous sites were coded as N. All analyses used phased sequence data. Sequences were edited in Sequencher v4.2 (GeneCodes Corporation), aligned in mega v5.2 (Tamura et al., 2011) using the muscle algorithm and confirmed by eye.
enmTools (Warren, Glor, & Turelli, 2008, 2010 was used to determine highly correlated variables (Pearson correlation coefficient ≥0.75), which we then reduced based on those most biologically relevant to small mammals (i.e., temperature related), to avoid over-parameterized models. Final runs were performed using bioclimatic variables 1 (annual mean temperature), 6 (minimum temperature of coldest month), 7 (temperature annual range), and 11 (mean temperature of coldest quarter), and run using a test percentage of 25%. We obtained species localities from museum databases (e.g., ARCTOS [http://arctos.data-base.uaf.edu] and MaNIS [http://manisnet.org/], Stein & Wieczorek, 2004) in October 2013. Sites with large spatial errors were removed and localities within <12 km of each other were eliminated (Hope et al., 2011) to reduce potential spatial autocorrelation (Reddy & Davalos, 2003), resulting in 127 M. longicaudus, 150 P. keeni, and 145 S. monticola sample localities. SDMs for each species were constructed at each time period using maxenT v3.3.3k (Elith et al., 2006;Phillips, Anderson, & Schapire, 2006;Phillips & Dudik, 2008). Final runs were performed using cross-validation across 10 replicates, with a regularization parameter of 5 (Hope et al., 2011; Warren & Seifert, 2011) and 1,000 iterations. All other values F I G U R E 3 Islands and nearby coastal mainland locations in Alaska, including paleoshorelines (kya = thousands of years ago) and hypothesized island groups (also see Table 2). Red arrows indicate potential colonization across the Alexander Archipelago as a result of change in sea level and glacial cover   We estimated potential island refugia, connectivity within and among hypothesized island groups, and potential colonization pathways at different points since the LGM. To re-create paleoshorelines for three temporal periods: 20 kya (with LGM glacial cover), 14, and 10 kya (Ehlers & Gibbard, 2004), we used information available from Carrara et al., 2003 and Baichtal (pers. com.)  For this study, AA populations were hypothetically designated as refugial or nonrefugial based on both paleoshoreline reconstructions ( Figure 3; Table 1) and climate suitability as determined by the SDMs (Figure 2). Refugia were subaerially exposed, which reflects areas not covered by glacial ice, and not under water (i.e., regions of newly exposed continental shelf
To identify signals of population fluctuation, we estimated historical demography for the Island clades (clade includes both island populations and nearby mainland, and are distinct from Northern or Southern continental mitochondrial phylogroups) within cyt b, we generated Bayesian Skyline Plots implemented in BeasT. Three runs per analysis used a MCMC chain of two billion steps, sampled every two million, with strict molecular clocks and models of evolution (Supporting Information Appendix S4) calculated via modelTesT. Tracer and aWTy were used to assess convergence.

| Phylogenetic and demographic analyses
All loci across all species had varying levels of polymorphism and genetic diversity (  Cook, 1996).
Overall, measures of genetic diversity for the Island clades were low for all three species (Supporting Information Appendix

| Testing phylogenetic models under the Coastal Refugia Hypothesis
Differences in genetic divergence within and between areas identi-

| D ISCUSS I ON
Phylogeographic studies help us understand how past environmental history has influenced genetic structure, but historical context (Grant & Grant, 2003) also provides a crucial foundation for forecasting how anthropogenic impacts, such as habitat conversion (e.g., old-growth logging) or climate-warming, will structure insular populations (Christensen et al., 2007;Fahrig, 2003;Olson, 1989).
We found that genetic structure in three sympatric small mammals of the Alexander Archipelago was influenced by a complex history of deep isolation and subsequent colonization. Genetic footprints, combined with assessment of paleoecology, helps identify both past refugial locations and the contemporary geographic barriers that now structure populations. In the case of the AA, genetic structure in three co-distributed species was influenced by ice cover and lower sea levels-factors that left paleoendemic signatures reflecting their longer-term in situ divergence. Those isolates subsequently served as source populations for recolonization throughout the archipelago.
A dominant feature in the data is an overall signal of island endemism and mainland demographic expansion, with idiosyncratic spatial (island) and temporal patterns among the three study species. Signals of overall demographic expansion among all three species across the entire AA and along the adjoining mainland are consistent with the Coastal Refugia Hypothesis, although details of histories differed across species.

| Shared geologic and climatic history
Historic SDMs are consistent with the paleoendemic signatures and identify that suitable environmental conditions existed in Southeast Alaska for these species during both the LIG and LGM (Figure 2).
Inconsistent estimates within each species partially stem from our inability to calibrate trees with fossils, and thus account for rate decay (Ho, Phillips, Cooper, & Drummond, 2005 Demographic expansion was identified in the Island clades of all three species (Bayesian skyline plots and expansion statistics), consistent with deglaciation of these areas ( Figure 5; Supporting Information Appendix S4). All three Island clades have lower estimates of mitochondrial and nuclear diversity, compared to their continental counterparts, perhaps reflecting the influence of historically small population sizes. When tested as a single population, rather than individual islands, coalescent simulations (i.e. expansion statistics; Supporting Information Appendix S4) identified the source populations as originating from the islands of Southeast Alaska, rather than mainland. The deeper history of Island clades and relatively higher numbers of endemics within this coastal region for each species (Cook & MacDonald, 2001) is most consistent with their extended persistence in the region followed by contemporary isolation across the fragmented archipelago, a finding that corresponds to the history of ermine in the region (Colella et al., 2018).
Overall, there are signals of shared history across M. longicaudus, P. keeni, and S. monticola, but the idiosyncratic influence of mutation rates, selection and drift, combined with independent populationlevel responses to historical climate and variable pathways are potentially reflected in incongruent aspects of the phylogeographic patterns. Nuclear loci for all three species lack consistent signatures of geographic structure across the region (Figure 4 and Supporting Information Appendix S6 and Appendix S7), potentially due to a combination of incomplete lineage sorting and differential rates of male-biased dispersal (Foster, 1965;Helmus, Mahler, & Losos, 2014;McCabe & Cowan, 1945). The possibility of human-mediated transportation seems unlikely given island-specific resolution of mtDNA.
The mtDNA data are not consistent with our original prediction that these widespread species should have relatively high levels of gene flow across the region. Instead, the data suggest that repeated genetic exchange or admixture during periods of lowered sea levels throughout the late Pleistocene and early Holocene have been followed by segregation and divergence in these species.

| Coastal Refugia Hypothesis
Although we are just beginning to explore the complex AA, preliminary studies suggest a significant role for northern coastal refugia in diversifying and structuring contemporary communities (e.g., de Volo et al., 2013;Hannon et al., 2010;Shafer et al., 2011). Reconstruction of paleoshorelines is complex due to non-linear changes as a result in lithospheric rebound (Josenhans, Fedje, Pienitz, & Southon, 1997) and submerged signatures of glacial extent, but our reconstructions of historical island connectivity, and potential colonization pathways suggest the potential for multiple LGM glacial refugia in Southeast Species distribution models (Figure 2) for each species suggest suitable climate supported offshore habitat on the exposed shelf and select western islands (Table 1) (Hetherington et al., 2003). Cowan (1935) suggested both P. keeni and S. monticola survived the Wisconsinan glaciation in coastal refugia in the AA. Although Klein (1965) and others (Heaton & Grady, 2003) conclude all small mammals likely failed to survive the LGM, there are pre-LGM fossils of long-tailed voles from Prince of Wales Island (Heaton & Grady, 2003. Lack of fossils on Prince of Wales during the LGM does not eliminate the possibility, however, that these species persisted further west in coastal refugia on the continental shelf when oceans were >120 m lower during the LGM. Presence of paleoendemic lineages has implications for the application of island biogeographic theory to the AA (e.g., Conroy et al., 1999), early human colonization of the New World (Achilli, Olivieri, Semino, & Torroni, 2018), and for understanding the evolution of continental biota (Riddle, 2016). Island isolation in particular may require special consideration as measurements of isolation generally assume the source population is on the mainland. If source populations for some species were actually from coastal refugia, then diversity measurements would be complicated due to multiple colonization sources and routes (Figure 3), a conclusion consistent with Lucid and Cook (2004) who showed that island area had more influence on contemporary island genetic diversity, than isolation (as measured by distance from the mainland). For example, the source for Prince of Wales populations would traditionally be measured from the mainland immediately to the east, but source populations might instead be from refugia to the west (e.g., Forrester or Coronation complexes).
Diversification of fauna in coastal refugia and then subsequently on the Alexander Archipelago also raises the possibility that continental diversity in northwestern North America has been influenced through recolonization of the mainland from islands (e.g., Filardi & Moyle, 2005 where the distribution of the island lineages now extend far beyond the boundaries SE AK (Dawson et al., 2014;Fleming & Cook, 2002;Peacock et al., 2007;Small, Stone, & Cook, 2003).

| CON CLUS ION
Historical climate and coastal refugia shaped genetic structure of species of the high-latitude Alexander Archipelago. Multiple lines of evidence suggest all three small mammals have paleoendemic lineages in the region, a finding consistent with other recent work on endemics in the region such as the Prince of Wales/Haida Gwaii ermine (Colella et al., 2018). Although this broad spatial pattern is concordant, questions remain regarding whether the timing of divergence coincides across these taxa. Cyclic climatic changes may produce similar spatial patterns that have different temporal signatures. Signals of demographic expansion across the region for these distinctive clades are also evident and roughly concordant. More detailed documentation of Late Quaternary changes in sea level and glacial cover along the North Pacific Coast, in addition to expanded genome-scale sampling of these and other endemic organisms, however, are needed to refine the number, location, and influence of glacial refugia.
Assessments of genetic structure across an array of species in complex landscapes, such as this coastal archipelago which experienced dynamic sea level fluctuations (e.g., Dolby et al., 2018), provide an initial framework for scientifically defensible management decisions (Gutrich et al., 2005;Pritchard, Jones, & Cowley, 2007). Future SDMs for these species forecast serious impacts, especially on the outer islands of the AA (Figure 2). Those outer islands now support a disproportionate number of subspecies (Cook & MacDonald, 2001) and endemic lineages of mammals (Cook et al., 2006) and other taxa (Sikes & Stockbridge, 2013).
Those islands also have experienced extensive anthropogenic habitat conversion (e.g., clear-cut logging of old-growth forests) over the past six decades with only minimal monitoring of impacts on biodiversity (Cook et al., 2006;Orians & Schoen, 2013). More generally, similarities across species are identifiable through the use of integrated analyses (i.e., phylogenetic reconstructions, SDMs); approaches that could be extended to other taxa in the AA or other high-latitude fragmented systems, such as Haida Gwaii of British Columbia (Reimchen & Byun, 2006), the Japanese Archipelago (Millien-Parra & Jaeger, 1999), or British Isles (Vincent, 1990) to help conserve regionally distinctive biota experiencing dynamic environmental change (Avise, 2008;Hendry et al., 2010).

ACK N OWLED G M ENTS
We acknowledge the efforts of personnel associated with the fieldwork and museum curation at the University of New Mexico's (UNM)

CO N FLI C T O F I NTE R E S T
None declared.

DATA ACCE SS I B I LIT Y
Sample locations with museum numbers, latitude and longitude, major cyt b clade, and associated GenBank accession numbers (Supporting Information Appendix S1), as well as between group net genetic divergences of cyt b among refugial and nonrefugial Southeast Alaskan