Phylogeography of sugar kelp: Northern ice‐age refugia in the Gulf of Alaska

Abstract Many Northeast (NE) Pacific fishes and invertebrates survived Pleistocene glaciations in northern refugia, but the extent that kelps survived in northern areas is uncertain. Here, we test the hypothesis that populations of sugar kelp (Saccharina latissima) persisted in the Gulf of Alaska during ice‐age maxima when the western margin of the Cordilleran ice sheet covered coastal areas around the NE Pacific Ocean. We estimated genetic diversities within and phylogeographical relationships among 14 populations along 2,800 km in the NE Pacific and Bering Sea with partial sequences of mitochondrial DNA 5′‐cytochrome oxidase subunit I (COI, bp = 624, n = 543), chloroplast DNA ribulose‐1,5‐bisphosphate carboxylase large subunit‐3′ (rbcL, bp = 735, n = 514), and 11 microsatellite loci. Concatenated sequences of rbcL and COI showed moderate levels of within‐population genetic diversity (mean h = 0.200) but substantial differences among populations (ΦST = 0.834, p < .0001). Microsatellites showed moderate levels of heterozygosity within populations (mean H E = 0.391). Kelps in the same organellar lineage tended to cluster together, regardless of geographic origins, as indicated in a principal coordinate analysis (PCoA) of microsatellite genotypes. The PCoA also showed evidence of nuclear hybridizations between co‐occurring organellar lineages. Individual admixture plots with population clusters of K = 2, 6, and 9 showed increasing complexity with considerable historical admixture between some clusters. A time‐calibrated phylogeny placed divergences between rbcL‐COI lineages at 1.4 million years at most. The time frames of mutation in the rbcL‐COI lineages and microsatellite population clusters differed among locations. The existence of ancient lineages in the Gulf of Alaska, moderate levels of genetic diversity, and the absence of departures from neutrality are consistent with northern refugia during multiple Croll‐Milankovitch climate cycles in the Pleistocene Epoch.


| INTRODUC TI ON
A continuing challenge to evolutionary biologists is to identify environmental variables that shape the genetic structures of natural populations through the actions of gene flow, random genetic drift, and natural selection, all of which operate against a backdrop of deep genetic structure created by historical events (Marko & Hart, 2011). A major source of deep structure in temperate and boreal species has been isolations in glacial refugia and vicariant separations forced by massive sheets of ice that spread over North America and Eurasia during the Pleistocene Epoch (Raymo, 1994;Stewart et al., 2010). A widely held view, based on phylogeographic patterns in terrestrial species, posits that northern populations contracted into southern ice-free refugia as glaciers advanced and then expanded northwards as glaciers receded (Hewitt, 1996(Hewitt, , 2000. A consequence of these contractions and expansions was the loss of genetic diversity in northern populations as a result of serial, postglacial colonizations (Hewitt, 1996). In contrast, some terrestrial species appear to have survived in northern refugia without losing genetic diversity (Birks et al., 2005;Stewart & Lister, 2001). Populations surviving in northern refugia may also show phylogeographic breaks with southern populations, as well as heterogeneity among populations resulting from isolations in local refugia.
The shores of the Northeastern (NE) Pacific were periodically covered by the margins of the Cordilleran ice sheet in the Pleistocene Epoch from 2.6 to 0.012 million years (Ma) during cold depressions in 40-100 thousand year climate cycles (Li & Born, 2019;Rasmussen et al., 2014). Since ice margins were irregular, a patchwork of coastal habitats between lobes of glacial ice potentially served as refugia (Carrara et al., 2007;Kaufman & Manley, 2004). Concurrently, global sea levels during glacial maxima dropped as much as 120 m below present-day levels (Rabineau et al., 2006;Rohling et al., 1998), re-sculpting shorelines. Unlike terrestrial refugia, the sizes, number, and locations of coastal refugia, and postglacial dispersal pathways were limited by the more or less linear nature of shorelines. Survival in northern marine glacial refugia has been suggested for several intertidal and shallowwater species of algae (Lindstrom et al., 1997) and a variety of invertebrates (Marko et al., 2010;Marko & Zaslavskaya, 2019).
While numerous genetic studies have been made of kelps along the coasts of Washington, Oregon, andCalifornia (e.g., Alberto et al., 2010, 2011), few population studies have been made at higher latitudes.
Here, we focus on populations of sugar kelp (Saccharina latissima Lane, Mayes, Druehl & Saunders) (Figure 1), a species distributed from Central California through the Arctic and into the North Atlantic (Bringloe et al., 2017;Neiva et al., 2018). Several biological features of sugar kelp bear on understanding its phylogeographic structure in the Gulf of Alaska. Sugar kelp occupy a narrow ecological niche, inhabiting low intertidal and shallow subtidal areas in wave-protected coves or bays and growing well only at 5-17°C in temperate regions (Druehl, 1967;Machalek et al., 1996). Nevertheless, some varieties of this kelp tolerate colder temperatures in Arctic waters (Bringloe et al., 2017;Neiva et al., 2018). Abundance and reproductive output vary greatly among sites and years, so that populations can be ephemeral on decadal time scales (Bekkby & Moy, 2011;A. Raymond, pers. comm.). Individual kelps tend to be perennial in the North Atlantic, but largely annual in Alaska (Bartsch et al., 2008;A. Raymond, pers. comm.).
The reproductive biology of sugar kelp influences connectivity between populations. This kelp alternates between large-bladed sporophytic kelps (Figure 1), anchored to small rocks and pebbles with branched haptera, and a microscopic, filamentous gametophyte phase (Lindeberg & Lindstrom, 2010). Sporophytes (2n chromosome complement) produce large numbers of meiospores, which settle after a brief planktonic phase and sprout into haploid, filamentous gametophytes with an n chromosomal complement. Male gametophytes produce spermatozoa that fertilize oogonia on female gametophytes, producing zygotes that grow in place into large-bladed sporophytic kelps. Even though spores can potentially be transported in coastal currents for several days (van den Hoek, 1987), F I G U R E 1 Photographs of sugar kelp Saccharina latissima. Left: Recruit of the year about 1.5 m in length. Right: Blades of sugar kelp at low tide in a wave-protect area behind the breakwater at Homer Spit, Alaska realized spore dispersals of kelps are generally limited to only a few meters (Anderson & North, 1966;Dayton, 1985;Santelices, 1990;Stein et al., 1995). Nevertheless, the drifting of reproductively mature sporophytes along a shore may occasionally contribute to longdistance dispersals (Saunders, 2014).
The goal of this study was to search for genetic imprints in contemporary populations that would shed light on historical events in the Gulf of Alaska in the Pleistocene Epoch. Partial sequences of genes encoded in mitochondrial (mt) and chloroplast (cp) DNAs were used to reconstruct gene genealogies and test for genetic population structure. We also used microsatellite DNA markers to estimate recent processes influencing genetic variability (Paulino et al., 2016). The expected smaller mutation rates of the two organellar genes potentially resolve deep population events, whereas the apparent larger mutation rates at microsatellite loci potentially resolve contemporary population dynamics.
These data together were used to test whether sugar kelp populations survived in southern refugia, or in local refugia in the Gulf of Alaska.

Several authors have noted misidentifications between split kelp
Hedophyllum nigripes (as Laminaria groenlandica, Saccharina groenlandica, Saccharina nigripes) and sugar kelp Saccharina latissima (Laminaria latissima, Laminaria saccharina) (Bartsch et al., 2008;Longtin & Saunders, 2015). In our study, misidentifications of S. latissima were discovered through molecular analysis of mtDNA 5′-cytochrome oxidase (COI). Young individuals of H. nigripes and S. latissima generally have bullated blades and similar morphologies. However, the blades of older H. nigripes are leathery and slippery to the touch, whereas those of S. latissima are thinner and lack mucilaginous glands (Longtin & Saunders, 2016). In the Gulf of Alaska, H. nigripes occurs on only wave-exposed or current-swept rocky shores consisting of bedrock or large boulders, whereas S. latissima occurs largely in wave-protected inlets and coves and is usually attached to small rocks and pebbles on a sedimentary bottom. While the two kelps can occur along the same stretch of beach, they are segregated into exposed (H. nigripes) and protected (S. latissima) microhabitats. A similar association between wave exposure and the occurrences of H. nigripes (as Saccharina nigripes) and S. latissima was found in the Bay of Fundy, in the NW Atlantic Ocean (Longtin & Saunders, 2016).
A 4-cm 2 piece of frond near the basal meristem was excised from sporophytes, damp-dried, and immediately desiccated with silica beads. Kelps at least 1 m apart were collected to avoid sampling siblings, or closely related individuals. DNA was extracted from 10 to 20 mg of dried tissue with a NucleoSpin ® 96 Plant II Kit (Macherey-Nagel Inc., Düren, Germany). Standard extraction kit protocols were followed, except dried tissues were homogenized at room temperature by crushing or chopping on weighing paper with a scalpel.
Amplicons were electrophoretically separated by size in an Applied  (Tajima, 1989). Sequence divergences between populations were estimated with F ST (Weir & Cockerham, 1984) and Φ ST with appropriate mutation models as determined with MEGA 7 (Kumar et al., 2016). IBD 1.52 (Bohonak, 2002)  A Bayesian tree of evolutionary relationships was produced from unique rbcL-COI haplotypes with BEAST 1.8.4 (Drummond & Rambaut, 2007), with a strict clock, the Yule model of speciation, and the HKY+G mutation model. Nodes in the tree were timecalibrated with a divergence of 6.9 Ma between S. latissima and S. japonica, as indicated in Figure  We also considered using such programs as BEAST to produce the Bayesian skyline plots (Drummond et al., 2005), or Ima2 (Hey & Nielsen, 2004), to reconstruct demographic history. However, the data at hand for sugar kelp are not appropriate for these analyses, as sample sizes for discrete populations of randomly mating individuals are too small. Pooling of genetically heterogeneous populations to achieve larger sample sizes would yield misleading results (Grant, 2015). Additionally, kelps show reproductive skew and hence do not follow the Wright-Fisher model of coalescence used in these programs for simulations to estimate medians and credibility intervals (Eldon & Wakeley, 2006;Grant et al., 2016). Beyond these technicalities, the effects of population growth cannot clearly be distinguished from the effects of reproductive skew on the nucleotide site frequency spectrum, which forms the basis of these analyses (Niwa et al., 2016).

| Microsatellite DNA
We used GENEPOP 4.6 (Rousset, 2008) to test for fit to Hardy-Weinberg genotypic proportions, using Markov chain Monte Carlo chains of 10,000 steps in 100 batches and a Bonferroni correction (Rice, 1989) of p = .05/11 = .0045 to control type I error at α = 0.05.
We used gene diversity analysis (Lewis & Zaykin, 2002) to estimate observed (H O ) and expected (H E ) heterozygosity averaged over loci, to count the number of alleles at each locus, and to estimate the inbreeding coefficient, F IS . HP-RARE (Kalinowski, 2005) was used to estimate allelic richness based on the smallest sample size. We used ML-Null (Kalinowski & Taper, 2006) and GENEPOP to estimate nullallele frequencies.
We examined the geographic and genetic components of population structure with four approaches. GENALEX 6.503 (Peakall & Smouse, 2012) was used to define principal coordinate analysis (PCoA) of allele frequency variability among samples with standardized covariance and with option to estimate missing genotypes. GENALEX was also used to reassign individuals to populations with significance of likelihoods set to 0.01 and to compute an AMOVA of allelic frequencies to estimate the overall levels of diversity within and among populations with 50,000 permutations.
STRUCTURE (Pritchard et al., 2000) was used to estimate population clusters and admixtures of individuals from hybridizations between clusters. We searched for the best fit of the data with population groups ranging from K = 1 to 10 with 10 replicates of 5,000 burn-in and 50,000 MCMC steps for each value of K. We used a uniform prior of admixture, assumed a correlation of allele frequencies among populations, and estimated the probability (maximum likelihood) of the data under the model. Web-based STRUCTURE HARVESTER (Earl & vonHoldt, 2012) was used to summarize the likelihoods of K in the various runs with the approach of Evanno et al. (2005) and to produce estimates of ΔK. Meirmans (2015) has cautioned that since the K statistics are "dubious at best" and "ad hoc," more than one value of K may be of evolutionary relevance.
The model used in STRUCTURE assumes that (1) loci are unlinked, (2) loci are at linkage equilibrium within population clusters, and (3) genotypes in population clusters are in Hardy-Weinberg proportions. To begin, we tested for linkage between loci overall and in the sampled populations with GENEPOP. We found that 2 locus pairs were significantly linked over all (Table S3 in Appendix S1) and 17 of 715 pairs tested showed significant disequilibrium before applying a correction for false positives (Table S4 in Appendix S1).
Hence, there was no indication of physical linkage between pairs of loci: The locus pairs showing disequilibria within populations were scattered among populations and were likely due to demographic history, or to chance. The inbreeding coefficient, F IS , ranged from −0.052 to 0.096, except for one population for which it was 0.217, and averaged 0.038. Hence, the populations did not substantially depart from the Hardy-Weinberg expectations.

| RE SULTS
We present the results for each of the organellar markers separately to facilitate comparisons with other studies, but use the results for concatenated rbcL-COI sequences to make our major inferences about population structure.

| Cytochrome oxidase-5′ (COI)
Sample sizes ranged from 6 to 90 and averaged 33 kelps (Tables S5   and S6 in Appendix S1). Ten nucleotide polymorphisms in a 624 bp fragment of COI defined 11 haplotypes, but only three haplotypes were abundant. A central abundant haplotype (MT040306 in 73.1% of kelps) was connected to 9 peripheral haplotypes by one mutation (Figure 2a). This lineage corresponds to lineage "A" of Neiva et al. (2018). All or nearly all of the kelps in 12 of the 14 samples carried the same common haplotype ("fixed") ( Figure 2b).
However, kelps from Port Moller (sample 2) carried a unique haplotype (MT040307) that was one mutation removed from the central haplotype, and kelps from Auke Bay (11) carried another haplotype (MT040308) also one mutation removed from the central haplotype ( Figure 2a,b). Six of the samples had one or two private haplotypes that were one mutation removed from the central haplotype.
Haplotype diversity (h) ranged from 0.0 in 8 samples to 0.145 in sample 7 and was 0.429 ± 0.024 overall (Table S6 in Appendix S1).
Nucleotide diversity (θ π ) ranged from 0.0 in several samples to 0.029% in sample 9 and was 0.079% ± 0.076% in the pooled sample. Tajima's D was marginally significant at three locations and in the pooled sample (D = −1.452, p = 0.041). Overall, the number of observed haplotypes (N H ) was 11, when only 4.39 were expected under neutrality (N EH ). A total of 10 private haplotypes appeared at six locations.
Genetic divergence (Φ ST ) between populations ranged from 0.0 between pairs fixed with the same haplotype to 0.968 between populations fixed, or nearly fixed, with different haplotypes (Table S7 in Appendix S1). 90.9% of the overall diversity was due to differences among populations, and 9.1% was contained within populations as different haplotypes among plants (Table S8 in Appendix S1).
Haplotype MT040323 appeared disjunctively at Sand Point (3) but also several hundred km away at Cordova (9).
The overall number of rbcL haplotypes was 7, but the expected number under neutrality was 10.5. Haplotype diversity (h) ranged from 0.0 in 6 samples to 0.515 ± 0.027 in sample 3 and was 0.702 ± 0.013 in the pooled sample (Table S10 in Appendix S1).
Tajima's D was not significant in any of the samples, nor in the pooled sample (D = 0.453, p = 0.716).
Divergences (Φ ST ) between populations ranged from 0.0 between populations with the same haplotype to 1.0 between populations with different haplotypes (Table S11 in Appendix S1). Overall, Φ ST = 0.788 (p < 0.00005) among samples. AMOVA indicated that 78.8% of the total variation was due to differences among populations and 21.2% was due to differences among kelps within populations (Table S12 in Appendix S1).
TA B L E 1 Sample information and estimates of genetic parameters based on concatenated fragments of mitochondrial DNA cytochrome oxidase I-5′ (COI) and ribulose-1,5-bisphosphate carboxylase/oxygenase large subunit-3′ (rbcL) (1,359 base pairs combined) in samples from the Gulf of Alaska and southeastern Bering Sea (locations 1-14) Note: Haplotype designations consist of the last two digits of the GenBank Accession Numbers for rbcL-COI and correspond to haplotype designations in Figure 2e. Sample numbers as in Table 1. Table 1 based on concatenated fragments of mitochondrial DNA 5′-cytochrome oxidase (COI) and ribulose-1,5-bisphosphate carboxylase/oxygenase large subunit-3′ (rbcL) (1,359 base pairs) with the Tamura and Nei (1993) model of mutation

| Concatenated COI and rbcL
Sample sizes of the concatenated sequences were slightly smaller than those of the two genes individually, because both genes were not successfully sequenced in some kelps. in populations 2, 7, and 9, but was not significant overall (p = 0.203).
Φ ST ranged from 0.0 between populations fixed with the same haplotype to 0.971 between populations 1 and 13 (Table 3). A majority of population pairs (66 of 78) showed significant sequence divergences. AMOVA indicated that 83.4% of the variability was due to differences among populations on average, and 16.6% was contained within populations (Table S13 in Appendix S1). A weak, but

| Microsatellites
Estimates of null-allele frequencies with GENEPOP and ML-null differed considerably, as ML-null estimated null-allele frequencies for invariant loci (Tables S14 and S15 in Appendix S1). Estimates of null-allele frequencies with GENEPOP were systematically smaller than those with ML-null. For the GENEPOP results, no locus nor population stood out with consistently large frequencies of null alleles. Hence, we used the entire microsatellite data set for population analyses. This will have the effect of underestimating some measures of diversity and blunting the precision of population analyses to a small extent.

| D ISCUSS I ON
The cornerstone of phylogeographic inference is the analysis of geographic variability in organellar genes that do not undergo recombination during gamete formation and that are inherited uniparentally (Avise, 2000). In our study, we included both mitochondrial and plastid DNA markers, which fulfill these two requirements. However, the use of these markers alone limits the kinds of insights that can be made about population dynamics, such as the extent of hybridization between lineages. Hence, we also include 11 microsatellite loci in our study. Mitochondrial and chloroplast genes and microsatellite markers are inherited independently of one another, so these markers together provide a multifaceted view of phylogeographic history.
Another strength of our study is the wide geographical coverage (2,800 km) and the analyses of larger sample sizes than are customary in studies of seaweeds, which have largely targeted taxonomic hypotheses (e.g., Lane et al., 2007). We follow with discussions of the molecular markers themselves and then with the pattern of population structure these markers depict. The results reveal a complex population history arising from repeated isolations in northern glacial refugia and postglacial expansions.
Contrary to the results for other seaweeds (Grant, 2016), the plastid gene rbcL in sugar kelp is more polymorphic than mitochondrial F I G U R E 4 Bayes tree of concatenated rbcL and COI haplotypes (1,359 bp) of Saccharina latissima in the Gulf of Alaska, Bering Sea, and Russia. Numbers at nodes represent estimated age of the node in millions of years (Ma) and the posterior probability of support for the node based on 10 million MCMC trees. Bars represent 95% highest probability densities for the age of a node. The NW Pacific kelp Saccharina japonica serves as an outgroup taxon to date the nodes in the tree using a divergence of 6.9 Ma ( One concern in the choice of microsatellite markers is the presence of a sufficient amount of diversity to detect population structure. 12 microsatellite loci were initially described by Paulino et al. (2016). Of these, 11 loci were used in our study and in studies by Neiva et al. (2018) and Naess (2019). 10 of these loci were used by Breton et al. (2018). Average heterozygosity in Alaskan popula-   (Chybicki & Burczyk, 2009), as restricted spore dispersal in the subdivided kelp populations limits outbreeding.
The high level of correct individual assignments back to originating populations also gives further confidence in the microsatellite data (Table 5).

| Contemporary population structure
An understanding of the genetic background of kelps is important for the development of commercial cultivars (Goecke et al., 2020).
The extent that populations are connected through gene flow is an important consideration for the management of a developing seaweed industry in Alaskan waters. While no signal of IBD was detected among populations, the sharing of low-frequency rbcL-COI haplotypes between some neighboring locations reflects connectivity between locations. For example,locations 3,4,5,6,7,9,11,and 14 had low-frequency haplotypes differing from the local common lineage, most likely indicating migration from other populations.  Table 1 Levels of microsatellite divergence between populations varied considerably, but showed only a weak correlation between genetic and geographical distance (Figure 3b,c). The lack of divergence between adjacent populations can readily be explained by gene flow between them, as for example between populations 5 and 6 in Kachemak Bay, which appear to consist of a single population in the bay. The association between genetic and geographical distance, however, weakens between populations separated by geographical distances greater than about 300 km, indicating that population history has played a greater role in sculpting population structure on larger geographic scales in the Gulf of Alaska than has contemporary levels of gene flow between populations.
The microsatellite allele-frequency similarity between these populations, which are separated by several hundred kilometers in some cases (Figure 5), is unlikely due to ongoing gene flow, given the poor ability of kelp meiospores and gametes to disperse long distances (Anderson & North, 1966;Gaylord et al., ,2002Gaylord et al., , , 2004Gaylord et al., , , 2006

| Temporal patterns of population clustering
The Location numbers correspond to number in Table 1 two historical scenarios. If sugar kelp abundances had been greatly reduced in the past, the two groups may represent an ancient geographical partition into two major population groups. If, on the other hand, population abundances have been more or less stable over time, the two genetic groups reflect the survival of only two genetic clusters among many. The two geographically disjunctive population clusters are reminiscent of the phylogeographic patterns of COI variability ( Figure 2b). COI lineage C is scattered in patches across the Gulf of Alaska, loosely reflecting the disjunct microsatellite cluster split between the eastern Aleutians and Southeast Alaska 2,000 km away.
The them. This contrast between patterns of divergence in microsatellite and in rbcL-COI markers cannot entirely be due to patterns of gene flow, but may also reflect differences in mutation rates.
The organellar DNA and microsatellite data together in these population pairs show different rates of evolution in the marker classes. In some cases, the rbcL-COI lineages were the same in the population pairs but in other cases they were different. Common origins of two populations would be expected to start the populations in the same clusters with the same rbcL-COI haplotypes and microsatellite allele frequencies. Unexpectedly, the organellar genes have evolved more rapidly than the microsatellite loci in the population pairs 7-9, 10-11, and 13-14 (Figures 2f, 5, and 7c). The haplotype frequency shift in clusters 10-11 occurred in the COI sequences, but shifts between the pairs 1-9 and 13-14 occurred in the highly variable rbcL nucleotide site mentioned above. In other cases, microsatellite loci have evolved faster than organellar genes at locations 4, 7, 10, 12, and 13, which differ from one another at microsatellite loci but still bear the same COI-rbcL haplotype in lineage A (Figures 2f, 5, and 7c). Also, kelps at locations 3 and 9 carry lineage D haplotypes, but have diverged in microsatellite frequencies. These results support the contention that mutation rates in organellar genes are not generally larger than in nuclear genes, or vice versa (Karl et al., 2012).
Despite coastal glaciers, environmental reconstructions indicate that oceanic conditions remained conducive to the growth of kelps around the Gulf of Alaska and along the coast of British Columbia.
Ocean temperatures in the Gulf of Alaska were relatively warm, dropping 5-6°C from present temperatures during glacial maxima (COHMAP-Members, 1988;Kutzbach et al., 1993). Sea surface temperature (SST) reconstructions indicate that the NE Pacific was not covered in perennial sea ice during glacial maxima (COHMAP-Members, 1988), although icebergs were common (Keigwin & Gorbarenko, 1992). Surface salinities of coastal waters were lower, retarding convection and lessening the mixing of surface waters with nutrient-rich deep waters (Gong et al., 2019;Worne et al., 2019;Zahn et al., 1991). Even so, depressed levels of nutrients were unlikely to interrupt completion of the sugar kelp life cycle.
Farther to the west, glacier-free shorelines around a smaller Bering Sea may also have served as glacial refugia. Lower sea levels during glacial maxima exposed the Bering Land Bridge, which remained unglaciated with ice-free southern shores. Nevertheless, seasonal sea ice covered the Bering Sea (Sancetta, 1983;Sancetta et al., 1984), leading to environmental conditions similar to those along the seasonally ice-covered shores of the Arctic Beaufort Sea, where sugar kelp presently occur (Bringloe et al., 2017. The unique haplotypes at Port Moller (2) in the Southeastern Bering ( Figure 2) support the concept of a southern Beringian shoreline refugium.  (Clague & James, 2002;Ryder et al., 1989;Thorson, 1980). If surviving populations had been limited to a southern refuge, postglacial invasions of northern shores would have been unimpeded by northern coastal glaciers and rapid colonizations would have led to genetically homogeneous northern populations (Hewitt, 1996). This is not supported by the strong genetic heterogeneity among populations in the Gulf of Alaska.

| Genetic signatures of northern refugia
Third, rapid population growth is expected when new habitats are colonized, producing an excess of low-frequency mutations over that expected for stable populations (Avise, 2000). However, the rbcL and COI haplotype frequency distributions failed to show these telltale departures from neutrality (Table 1). The absence of genetic imprints of geographic expansion and "recent" population growth is underscored by deep divergences between lineages dating to one million years or more ( Figure 4).  (Marko et al., 2010). The northern clingfish (Gobbiesox meandricus) shows high diversities in northern populations, also indicating persistence during glacial maxima in northern refugia (Hickerson & Ross, 2001). Survival in northern marine refugia has also been postulated for Pacific cod (Bigg, 2014;Canino et al., 2010).
Genetic signatures of northern refugia have also been found for several species of algae. In addition to sugar kelp, split kelp (Hedophyllum nigripes)  and wing kelp (Alaria "marginata" complex) ) have a mosaic population structure and high levels of genetic diversity in the Gulf of Alaska that reflect persistence in several northern refugia. Some species in the kelp genus Agarum also have geographical distributions that are consistent with persistence in northern glacial refugia (Boo et al., 2011).
Finally, an study of Pacific dulse (Palmaria mollis) using random amplified polymorphic DNA revealed a phylogeographic break between populations between SE Alaska and south-central and western Alaska that likely indicates secondary contact between populations isolated in northern glacial refugia (Lindstrom et al., 1997). In fact, a break in the geographic distributions of several algae at Southeast Alaska indicates contact between northern species that had to have survived glaciations in Alaskan waters and southern species from southern refugia (Lindstrom, 2009). The genetic signatures of many intertidal and shallow subtidal marine algae indicate that local northern ice-age refugia were common.

| CON CLUS ION
The pattern of divergence among populations estimated with organellar DNA differed somewhat from the pattern estimated with microsatellite markers, a contrast that appears to be due to an interaction between gene flow and mutation in the different markers.
Unquestionably, restricted dispersal between populations has led to a fragmented genetic population structure that failed to produce a signal of isolation by distance in the rbcL-COI markers and only a weak signal in the microsatellite markers. Superimposed on innate patterns of gene flow are chance mutations in the molecular markers and population upheavals from numerous ice ages in the Pleistocene producing genetic imprints of local extinctions and colonizations.
Our argument that sugar kelp survived episodes of glaciation in northern refugia rests on three findings. First, while we were not able to make a robust test of the diversity gradient hypothesis because of the lack of data for populations in areas below the southern boundary of the last ice sheet, we found a considerable amount of genetic diversity in the Gulf of Alaska that was not predicted by the diversity-gradient hypothesis. Second, the source of genetic di-

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