Evolution of stickleback in 50 years on earthquake-uplifted islands

Significance

On several Alaskan islands, phenotypically variable threespine stickleback fish now live in ponds that were formed during uplift caused by the 1964 Great Alaska Earthquake. We analyzed phenotypic and genome-wide genetic divergence of resident freshwater and oceanic threespine stickleback populations from three islands. These data support the hypothesis that the freshwater populations evolved repeatedly from their oceanic ancestors in the past half-century, and have differentiated to nearly the same extent as populations that were founded thousands of years ago. This work raises the possibility that much of the evolution that occurs when oceanic stickleback invade fresh water takes place in fewer than 50 generations after colonization, rather than gradually over thousands of years.

Abstract

How rapidly can animal populations in the wild evolve when faced with sudden environmental shifts? Uplift during the 1964 Great Alaska Earthquake abruptly created freshwater ponds on multiple islands in Prince William Sound and the Gulf of Alaska. In the short time since the earthquake, the phenotypes of resident freshwater threespine stickleback fish on at least three of these islands have changed dramatically from their oceanic ancestors. To test the hypothesis that these freshwater populations were derived from oceanic ancestors only 50 y ago, we generated over 130,000 single-nucleotide polymorphism genotypes from more than 1,000 individuals using restriction site-associated DNA sequencing (RAD-seq). Population genomic analyses of these data support the hypothesis of recent and repeated, independent colonization of freshwater habitats by oceanic ancestors. We find evidence of recurrent gene flow between oceanic and freshwater ecotypes where they co-occur. Our data implicate natural selection in phenotypic diversification and support the hypothesis that the metapopulation organization of this species helps maintain a large pool of genetic variation that can be redeployed rapidly when oceanic stickleback colonize freshwater environments. We find that the freshwater populations, despite population genetic analyses clearly supporting their young age, have diverged phenotypically from oceanic ancestors to nearly the same extent as populations that were likely founded thousands of years ago. Our results support the intriguing hypothesis that most stickleback evolution in fresh water occurs within the first few decades after invasion of a novel environment.

On March 27, 1964, the largest earthquake ever recorded in North America struck the south coast of Alaska (1, 2). This catastrophic event uplifted islands in Prince William Sound and the Gulf of Alaska in just a few minutes, creating ponds from formerly marine habitat and setting the stage for the diversification of threespine stickleback fish (Gasterosteus aculeatus) in these new freshwater sites. This seismic disturbance provides an excellent opportunity to address long-standing evolutionary questions regarding how often dramatic phenotypic shifts can happen over contemporary timescales (3–7).

Despite examples of rapid divergence in wild populations, evolutionary rates may often be constrained by a suite of factors (8). For example, evolution in new habitats may be limited by waiting times for new beneficial mutations (9–11). Even when adaptation occurs from standing genetic variation, evolution via selection of numerous independent loci of small effect may be time consuming (12–16). We know, however, that evolution can occur rapidly, particularly under artificial selection or in human-altered landscapes (17–21). In addition, empirical studies in the wild—particularly in response to significant environmental changes—have demonstrated that strong selection and rapid evolution over decades may be more common than once thought (22–24).

A rapid evolutionary response is predicted when the intensity of directional selection is strong (11, 25), a scenario likely to occur immediately after a habitat shift or environmental disturbance (26, 27). However, because of previous technological limitations, few studies of rapid differentiation in the wild have included genetic data to fully disentangle evolution from induced phenotypic plasticity. The small numbers of markers previously available for most population genetic studies have not provided the necessary precision with which to analyze very recently diverged populations (but see refs. 28 and 29). As a consequence, the frequency of contemporary evolution in the wild is still poorly defined, and its genetic and genomic basis remains unclear (30).

Advances in sequencing technology now allow the precise inference from genomic data of colonization history and evolutionary patterns that have occurred over just a few generations (31, 32). The threespine stickleback system is ideal for testing hypotheses about contemporary evolution. Postglacial adaptive radiations over the last 12,000–20,000 y in newly available freshwater habitats have spawned divergent phenotypes that demonstrate parallel phenotypic evolution (33, 34), with underlying parallel genetic (35–39) and genomic (40–43) bases. An open question, however, is whether this parallel divergence in stickleback actually requires thousands of years, or whether it can occur in nature over decadal timescales, as is implied by studies of a small number of recently formed artificial and wild stickleback populations (44–50). Also unknown is how often the countless populations of stickleback in geographically close ponds represent invasion followed by local dispersal or independent founding from the sea.

To address these questions, we identified populations from three islands (Middleton, Montague, and Danger) in Prince William Sound and the Gulf of Alaska that could have been founded only after the 1964 earthquake (Fig. 1 and SI Appendix, Table S1). Middleton Island was uplifted 3.4 m, creating a new terrace with ponds from a previously submarine platform (1). Similarly, Danger and Montague Islands experienced uplift and creation of new ponds (51). Stickleback now can be found in many of the habitats produced by the earthquake (52). We first analyzed a subset of populations from Middleton Island to describe the pattern of multivariate phenotypic divergence. We then produced and analyzed restriction site-associated DNA sequencing (RAD-seq) data (53, 54) from 25,000 RAD loci in 1,057 individuals collected from a total of 20 populations from all three islands and one mainland population. Deep sequencing yielded a set of 130,000 single-nucleotide polymorphisms (SNPs) and a total of 146 million genotypes. This large genomic dataset allowed us to ask whether phenotypic and genetic divergence in stickleback, thought to require thousands of years, can occur in a fraction of that time. Unlike previous studies that have made inroads into this question (47–50), the high level of biological replication of individually genotyped samples, within and across populations, in the present study avails a battery of population genomic analyses such as analysis of molecular variance (AMOVA), principal component analysis (PCA), and STRUCTURE. These approaches are most appropriate for defining (and assigning individuals to) genetic groupings across recently formed populations potentially still experiencing gene flow, such as those that are the focus of our study. We use this robust dataset to test the parallel origin of several populations against a precisely dated geological event—the Great Alaskan Earthquake of 1964—to ask whether replicated colonization of a large number of newly formed freshwater habitats by oceanic stickleback ancestors occurred independently on different islands and even amid close geographic locales within individual islands.

Fig. 1.

Sampling locations. (A) Prince William Sound and the Gulf of Alaska, with Danger (B), Montague (C), and Middleton (D) Islands boxed. (Inset) Alaska with box representing sampling area. Sites are coded by whether they are freshwater or oceanic habitat and by the dominant ecotype found in the population. Dark gray shading within each island cartoon delineates the approximate pre-1964 shoreline.

Results

Stickleback Are Phenotypically Divergent Between Habitats and Among Populations.

We gathered morphometric data from six populations on Middleton Island (SI Appendix, Fig. S1 and Table S1). We analyzed individuals from one marine habitat [Middleton Population 23 (Mi23)] containing fish only with oceanic phenotypes, and one freshwater site (Mi06) containing fish only with freshwater phenotypes. The four remaining freshwater sites (Mi08, Mi09, Mi13, and Mi19) contained both phenotypically freshwater and oceanic individuals (SI Appendix, Table S2). Freshwater and oceanic ecotypes differed in standard length [F_(1,569) = 943.80; P < 0.001] and the number of lateral plates that form part of the dermal armor [F_(1,569) = 1,734.00; P < 0.001; SI Appendix, Fig. S2]. In a PCA using a set of morphological metrics known to diverge between ecotypes (55) (SI Appendix, Fig. S1 and Table S2), the first principal component (PC1) clearly separated freshwater and oceanic individuals and explained 93% of the total variation (Fig. 2 and SI Appendix, Table S3). Oceanic and freshwater ecotypes differed significantly in scores from the first three PCs [F_(1,553) = 786.52, P < 0.001; F_(1,553) = 16.11, P < 0.001; F_(1,553) = 14.33, P < 0.001; SI Appendix, Fig. S3]. When considering populations separately, freshwater individuals from Mi09 were also differentiated from Mi08 and Mi19 along PC1 (SI Appendix, Fig. S3 and Table S4).

Fig. 2.

PCAs describe the overall distribution of phenotypic variation in six sites from Middleton Island. Each point represents the population mean ± 2 SE. Points for Mi08, Mi09, Mi13, and Mi19 represent means for only phenotypically freshwater individuals as determined before analysis by visual inspection (Materials and Methods). Mi23 is a phenotypically oceanic population originating from a marine habitat, and SO are oceanic individuals pooled from all four sympatric sites included in the phenotypic analysis (Mi08, Mi09, Mi13, and Mi19). High/partially plated groups are in green, and low plated are in blue.

Lateral plates had the largest loadings among all of the traits on PC1 (SI Appendix, Table S5), which was anticipated due to the extensive documentation of differences in lateral plate number between oceanic and freshwater individuals (35–40). To confirm that other traits also contributed to phenotypic divergence, we reran the PCA with all variables except lateral plates. The first two PCs still accounted for a significant amount of the variation (28% and 16%, respectively; Fig. 2 and SI Appendix, Fig. S3 and Table S3), and the ecotypes again differed significantly in scores for the first three PCs [F_(1,546) = 95.30, P < 0.001; F_(1,546) = 12.09, P < 0.001; F_(1,546) = 4.23, P = 0.040; SI Appendix, Figs. S3 and S4 and Table S4]. Lengths of the left pelvic spine, second dorsal spine, pectoral muscle, and pectoral fin insertion had the largest loadings on PC1, whereas snout, eye, and operculum lengths, as well as the distance between dorsal spines, had the largest loadings on PC2 (SI Appendix, Table S5). Similar to the analysis that included lateral plates, scores from the first three PCs differed significantly among populations [F_(6,550) = 27.88, P < 0.001; F_(6,550) = 9.93, P ≤ 0.001; F_(6,550) = 7.38, P < 0.001; SI Appendix, Figs. S3 and S4 and Table S4]. The degree of phenotypic divergence among these very young populations mirrors what has been documented for much older stickleback populations in Alaska and other regions of the world (33).

Genetic Variation Is Partitioned Primarily Between Oceanic and Freshwater Ecotypes.

To infer the genetic relatedness of individuals and evolutionary histories of populations on all three islands, we generated RAD libraries for 1,057 individuals from 21 total populations (SI Appendix, Table S1), ranging from 18 to 96 fish per population (mean of 50), and including one locale on mainland Alaska. These libraries were single-end sequenced in 12 lanes of an Illumina HiSeq 2500 (SI Appendix, Tables S6 and S7), resulting in a total of more than 1.4 billion 101-bp sequences that passed several stringent quality filters. Each individual was represented by an average of 1.25 million sequences of which ∼1 million (84%) were aligned to the reference genome, and nearly all (99%) were used by the population genomics analysis pipeline Stacks for downstream analyses. Approximately 130,000 SNPs were identified, most of which were called in every individual, thus providing a powerful dataset for studies of recent and rapid evolution (SI Appendix, Tables S8 and S9).

We performed PCA using the SNP data from all 21 populations. Similar to our phenotypic results, we found that the major axis of genetic variation (PC1) represents a continuum of oceanic to freshwater genotypes, and also separates freshwater populations from the three islands (Fig. 3 and SI Appendix, Fig. S5). The optimum number of clusters (K) in the STRUCTURE analysis was 2, representing oceanic (dark gray) and freshwater (light gray) genotypes. Posterior probability of assignment to the oceanic genotype decreases, and probability of assignment to the freshwater genotype correspondingly increases, as populations increase in PC1 score.

Fig. 3.

The major axis of genetic variation is a continuum of oceanic to freshwater genotypes. (A) STRUCTURE analysis of the entire dataset reveals an optimum K of 2, representing oceanic (dark gray) and freshwater (light gray) genotypes. (B) Distribution of mean principal component (PC) 1 scores for each of 20 populations ± 1 SE. PC1 accounts for 36% of the overall genetic variation. As populations increase in PC1 score, their posterior probabilities of assignment to the freshwater genotypic cluster in the STRUCTURE analysis also increase. (C) Populations plotted by PC1 scores as in B, but with those populations that house different plate morphs split by that criterion; high/partially plated groups are in green and low plated in blue.

Post-1964 Populations Were Not Derived from Preexisting Freshwater Populations.

Historical maps and aerial photos show that no freshwater habitat existed before the 1964 earthquake on Danger Island. However, Middleton Island and the study area of Montague Island had preexisting freshwater sites. On Montague, extensive sampling of the only nearby site (Stump Lake), which is in a separate watershed and approximately 1 overland kilometer from the post-1964 sites on the eastern side of the island, yielded no fish of any species. Additional pre-1964 freshwater sites in more distant watersheds on Montague were at least 2 km from the uplift ponds. On Middleton, stickleback were found in only one of the two preexisting ponds, Mi12 (Fig. 1), which rests on an upper terrace that formed ∼2,400 y ago (2). Stickleback from this site were well separated from the remaining sites in the study system along PC2 (SI Appendix, Figs. S5 and S6), formed a clearly unique genotypic cluster in STRUCTURE (Fig. 4), and were the most differentiated freshwater population on all three islands (Fig. 5 and SI Appendix, Fig. S7 and Table S10). Site 12 is therefore unlikely to have founded the other freshwater populations on Middleton.

Fig. 4.

Uplift island populations form multiple distinct genotypic groupings. Plots of posterior probabilities of assignment of each individual into clusters based upon the results of a STRUCTURE analysis and corresponding PCAs color-coded by K-means clustering assignment; reuse of colors in these three separate STRUCTURE analyses does not signify the same genotypic groupings. Percentages represent the amount of genetic variation explained by each PC. (A) Post-1964 freshwater populations from Middleton Island separate into clusters representing sites 7, 14, and 15, and sites 6, 8, 11, 13, 14, 16, 22, and 28. From sites 8, 13, 14, and 22, only individuals with freshwater phenotypes (low lateral plate counts) were included. Site 12, the pre-1964 population, forms a unique genotypic cluster. (B) Montague sites separate into two major clusters that distinguish sites from watersheds in the southeast (35–37) and southwest (30, 33). Site 33 contains two genotypic clusters. (C) The freshwater (Da04) and oceanic (Da02) sites from Danger also form separate clusters.

Fig. 5.

A neighbor-joining tree illustrates the relationships of the populations. The tree was assembled in POPTREE (95) using 1,000 random loci. Bootstrap values were calculated based on 500 replicates. Circles and squares distinguish freshwater and saltwater environments, with polymorphic (Mi12) and sympatric populations divided according to phenotype: high/partially plated (green) and low plated (blue). Shown are nodes with support greater than 0.75.

One hypothesis for the origin of freshwater stickleback in Mi12 is that they were unintentionally introduced during stocking of trout into this lake in the mid-20th century. The stocked trout originated from a hatchery on mainland Upper Fire Lake north of Anchorage, Alaska. STRUCTURE analyses show that Mi12 and Upper Fire Lake (UFL) represent distinct genotypic clusters with different allele frequencies (SI Appendix, Fig. S8), and the two sites are also highly divergent (F_ST = 0.164; SI Appendix, Tables S10 and S11). Taken together, these results offer no support for the hypothesis of introductions from UFL and instead argue for an older colonization event for Mi12 with a subsequent evolutionary history independent of other sites on Middleton.

Independent Evolution Occurred Quickly Among Islands and Among Regions Within Middleton and Montague Islands.

Using an AMOVA (56) framework, we found that, across all populations, most of the genetic variation (∼74%) was partitioned among individuals, whereas 15% of the genetic variation was attributed to differences among sites within islands (SI Appendix, Table S9). Despite their having distinct phenotypic differences, we found that only 5% of the genetic variation was partitioned between oceanic and freshwater habitats, but 18% of the variation was partitioned among populations within habitat types (freshwater or oceanic). Because the PCA and STRUCTURE results (Fig. 3 and SI Appendix, Tables S10–S12) support low levels of genetic divergence among oceanic populations, this partitioning of genetic variation must be due to differences among freshwater sites and is consistent with independent draws of genetic variation from a common oceanic ancestor during separate originations of these freshwater populations. Furthermore, STRUCTURE, K-means clustering, PCA, pairwise F_ST, and phylogenetic analysis of populations (Figs. 3–5 and SI Appendix, Figs. S5 and S7 and Table S13) all consistently support the presence of three distinct clusters of freshwater populations on Middleton Island. One group (designated FW1) contains sites Mi07 and Mi15. A second (FW2) consists of sites Mi06, Mi11, Mi16, and Mi28, as well as freshwater phenotype individuals from sites Mi08, Mi13, and Mi22. Mi14 harbors a mix of freshwater individuals from FW1 and FW2. A third freshwater grouping is the pre-1964 population, Mi12 (SI Appendix, Table S14). Within Middleton Island, 9% of the variation among sites is attributed to these different regions, whereas individuals within sites, and sites within regions, each account for 5–6% of the variation (SI Appendix, Table S9). F_ST comparisons among populations within FW1 and FW2 range from 0.002 to 0.047, whereas those between pairs of populations among FW1, FW2, and Mi12 range from 0.015 to 0.198 (SI Appendix, Table S10).

Similar to the results on Middleton, STRUCTURE, K-means clustering, phylogenetic clustering, F_ST comparisons, and PCA on Montague Island (Mo) support differentiation between the southeast (Mo35, Mo36, and Mo37) and the southwest (Mo30 and Mo33) (Figs. 3–5 and SI Appendix, Fig. S4 and Table S15). These two derived groupings are designated MoSE and MoSW, respectively. The two populations that comprise MoSW are further separated from each other in STRUCTURE and K-means analyses (Fig. 4). Within MoSW, Mo33 consists of two genotypic clusters that roughly correspond with lateral plate morphs, but there are also individuals with intermediate numbers of lateral plates. STRUCTURE and comparisons of mean genetic PC scores suggest that this is a more admixed population than in sympatric sites on Middleton Island (Mi08, Mi13, Mi14, and Mi22; Fig. 3). That Mo33 is a population in which oceanic and freshwater fish have hybridized is also supported by the distinction of its members from the oceanic genotype (SI Appendix, Fig. S8) and by the comparison of lateral plate number and genetic PC 1 score (SI Appendix, Fig. S10); we see more individuals with intermediate phenotypes and genotypes here than among Middleton sympatric populations. Assignment to either MoSE or MoSW accounts for 17% of the variation among populations on Montague Island, whereas individuals within sites, and sites within MoSE and MoSW, account for ∼3% and 5% of the variation, respectively. The remaining 75% of the variation is at the individual level (SI Appendix, Table S9). Pairwise F_ST between populations within MoSE and MoSW ranges from 0.003 to 0.030, whereas F_ST between pairs of populations chosen from MoSE and MoSW ranges from 0.037 to 0.125 (SI Appendix, Table S10).

The fine-scale population structure within Montague and Middleton Islands suggests that there were at least five independent colonizations (Mi12, MiFW1, MiFW2, MoSE, and MoSW) by oceanic ancestors on these two islands. The presence in a freshwater habitat [Danger Island population 4 (Da04)] of oceanic-looking fish that form a genotypic cluster distinct from nearby oceanic stickleback at Danger Island (Fig. 4 and SI Appendix, Table S16) argues that independent colonization of newly opened habitats continues to occur.

Methods that assign individuals by probability to genetic groupings (e.g., STRUCTURE) are most appropriate for studies of very recently derived populations, or populations that are continuing to experience gene flow. Although phylogenetic approaches that assume a bifurcating relationship among lineages are not strictly appropriate for analyses of populations with these characteristics (32, 57), using allele frequency variation to reconstruct well-supported nodes of a population-level phylogenetic tree can be informative (Fig. 5 and SI Appendix, Fig. S7). Broad topology of the inferred population phylogeny clearly supports the conclusions (presented above) of repeated evolution among and within islands. As expected under recent, independent derivation of freshwater populations from oceanic ancestors, the base of the tree is an unresolved polytomy of clades that includes oceanic and freshwater lineages (Fig. 5). Importantly, the tree provides no more evidence for the pre-1964 population (Mi12) as the sister group to the other Middleton freshwater populations than it does for any of the other clades at the tree’s base, including those found on distant Montague Island. The observed well-supported clade of low-plated fish on Middleton (excluding Mi12) might arise in a scenario of independent colonization; reuse of standing genetic variation from local marine populations could create a signal of cohesion among the young freshwater adapted lineages even when some populations arose independently from the sea.

Incongruent Phenotypic and Genetic Variation Indicates Recurring Introgressive Hybridization.

Five of the populations genotyped on Middleton Island contained fish with oceanic characteristics as well as fish with typical freshwater phenotypes (Fig. 1). Such ponds could house populations segregating phenotypes as polymorphisms or could be regions of secondary contact between two genetically differentiated populations. In one of these ponds (pre-1964 Mi12), fish segregate a polymorphism in lateral plate number but form a united genetic group. In the four other sites, however, STRUCTURE analyses reveal both oceanic and freshwater genotypes (Mi08, Mi13, Mi14, and Mi22; Figs. 3 and 4), supporting the hypothesis of secondary contact of distinct populations. As expected, in these populations, the genetic divergence between oceanic and freshwater ecotypes is highly correlated with plate variation (SI Appendix, Figs. S9 and S10), and a significant relationship exists between the genetic PC1 scores and plate number (r² = 0.54, t = −33.61, P < 0.001). Two main clusters of points comprise individuals with concordant freshwater or oceanic phenotypes and genotypes (SI Appendix, Figs. S9 and S10). Individuals in freshwater habitats with discordant phenotypes and genotypes may be products of introgressive hybridization. Further advocating for introgression is that, out of the 132 individuals surveyed from oceanic habitats on Danger (Da02) and Middleton Islands (Mi17 and Mi23), one individual had a freshwater genotype and another had a freshwater phenotype (SI Appendix, Fig. S11). These data lend further credence to gene flow from derived freshwater populations back into oceanic populations.

Discussion

Patterns of Genetic Differentiation Support Independent Evolution After the Earthquake.

Stickleback on three seismically uplifted islands harbor extensive genetic variation that is partitioned primarily among individuals. A significant amount of the remaining genetic variation, however, is also partitioned among fish with divergent lateral plate phenotypes in oceanic and freshwater habitats, as well as among those with freshwater phenotypes from freshwater sites (SI Appendix, Table S9). Genetic divergence between oceanic and freshwater ecotypes has previously been documented across the stickleback holarctic distribution (41) and among populations in close proximity within Alaska (40). The overall divergence (as measured by F_ST) that we find here, however, is lower for several freshwater–oceanic pairs than has been documented previously and is consistent with much more recent freshwater colonization from an oceanic ancestor (Fig. 6). In addition, genetic diversity is not much lower in freshwater habitats than in marine, indicating that either the initial colonizing populations were large or (more likely) that recurrent gene flow has transpired between oceanic and freshwater stickleback since the time the populations were founded (SI Appendix, Table S8). Overall, the pattern of partitioning of genetic variation among populations supports independent draws of genetic variation from a common oceanic ancestor during the origin of freshwater populations, with subsequent gene flow from nearby oceanic stickleback.

Fig. 6.

Post-1964 populations have diverged phenotypically from marine ancestors nearly as much as have older, postglacial freshwater populations. Comparison of pairwise F_ST between freshwater and oceanic populations from Cook Inlet (mainland) and uplifted islands demonstrates that within just a few decades freshwater populations can show freshwater–marine divergence comparable to mainland populations that were likely founded thousands of years ago. Cook Inlet data from Hohenlohe et al. (40). Each of the 24 sites is color-coded by the dominant lateral plate phenotype (high/partially plated in green; low plated in blue).

Our deep population genomic data clearly support a recent oceanic origin and subsequent adaptive differentiation of resident freshwater stickleback in post-1964 ponds on all three islands. Intraisland genetic variation also partitions freshwater sites from Middleton and Montague into three and two groups, respectively (Mi12, MiFW1, MiFW2, and MoSE, MoSW) that are geographically distinct but, in the case of Middleton Island, closely neighboring. A possibility existed that the pre-1964 population on Middleton Island (Mi12) was the progenitor of the younger freshwater populations on the island. Our data do not support this hypothesis. Rather, the high genetic divergence between oceanic populations and Mi12 (F_ST ∼ 0.16; SI Appendix, Table S10) is consistent with its founding at the time the terrace was formed by an older uplift event, ∼2,400 y ago (2). Similarly, the two genetic groupings on Montague are found on opposite sides of the island (Fig. 1) and are separated by a mountain range, which likely prevents gene flow between them. However, gene flow may occur within watersheds, particularly the southeast watershed (Fig. 4 and SI Appendix, Table S15). STRUCTURE analysis also revealed two genotypes in Mo33 (Fig. 4 and SI Appendix, Table S15); the continuous range of posterior probabilities of genetic assignment among individuals there argues for secondary contact occurring soon after colonization, subsequent generations of hybridization, and no recent influx of oceanic fish.

Overall, our findings support the hypothesis of at least six independent colonization events by oceanic ancestors across the three islands. Because we examined a subset of freshwater habitats on only three of the many islands and coastal regions throughout Prince William Sound and the Gulf of Alaska impacted by uplift, this is likely to be a significant underestimate of the number of times that freshwater stickleback have evolved independently in the last 50 y throughout this region. Furthermore, our results demonstrate a range of populations along the oceanic to freshwater phenotypic and genomic continuum, from Da04, which closely resembles oceanic populations, to Mi07 and Mi15, which appear to have the most extreme freshwater phenotypes and genotypes (Fig. 3 and SI Appendix, Figs. S5, S9, and S10). This spectrum likely reflects colonization and hybridization history. Da04 could be a more recently founded population that experiences regular gene flow with oceanic stickleback, whereas Mi07 and Mi15 could have been founded earlier and remained isolated.

Phenotypic Differentiation Supports a Role for Strong Divergent Selection.

Despite being founded less than 50 y ago, freshwater stickleback populations on Middleton Island are differentiated from oceanic fish across multiple morphological features. As expected, variation in lateral plate count is a major driver of the phenotypic divergence we measured. However, even when lateral plate number is removed from the PCA, traits involved in foraging, defense, and swimming also distinguish the two ecotypes (SI Appendix, Table S5). Many of these traits have a known genetic basis (35–37, 58, 59), with quantitative trait loci mapping to many different linkage groups (35, 36, 39, 58–62), strongly supporting that the phenotypic divergence we observe on Middleton Island is due primarily to evolution rather than phenotypic plasticity.

Even in freshwater habitats where oceanic and freshwater ecotypes are sympatric, they are phenotypically divergent to nearly the same extent as allopatric ecotypes (Fig. 2 and SI Appendix, Figs. S3 and S4). Given the topography of Middleton Island and the locations of the newly formed freshwater habitats, secondary contact between oceanic and freshwater fish is the most probable cause of this sympatric co-occurrence. This scenario has likely been repeated over many years and creates the potential for gene flow that would inhibit phenotypic divergence simply by drift (30, 63–66). The clearly discontinuous trait distributions between ecotypes are therefore most likely due to strong divergent selection on the phenotypes in the alternative oceanic and freshwater habitats, in agreement with other studies of selection in artificially seeded stickleback populations (38, 50, 67–69). The strength of selection (s) on the region containing Eda—the locus associated with lateral plate loss (35, 36, 70)—in oceanic stickleback transplanted to artificial freshwater ponds was found to be 0.52 (47, 69, 71). Even selection coefficients an order of magnitude smaller on other traits would maintain adaptive phenotypic differentiation in the face of all but the highest levels of gene flow.

Most Stickleback Evolution May Occur in the First Decades After Colonization.

The postglacial adaptive radiation of threespine stickleback in Cook Inlet is a well-described model of rapid evolution for which there is ample evidence of phenotypic and genetic divergence between ecotypes likely occurring over thousands of years (35, 40, 60, 72). However, the independent colonization of newly available freshwater habitat by oceanic ancestors on Middleton, Montague, and Danger Islands has occurred on a timescale that is orders of magnitude shorter. Although life history traits vary among stickleback populations, 50 y likely represents only 25 to a maximum of 50 generations (73). Remarkably, phenotypic divergence of freshwater fish on Middleton Island is nearly the same amount as freshwater stickleback from mainland Alaska populations that were probably founded about 13,000 y ago (35) (Fig. 5), despite the overall genetic divergence being lower in the much younger Middleton populations. Our results demonstrate that evolution can occur in this species on contemporary timescales, and present the tantalizing hypothesis that much of the previously documented postglacial divergence of freshwater stickleback populations that has been inferred to have occurred over thousands of years (40, 74) actually occurred during the first few generations following colonization.

Although phenotypic differentiation in decades is a very rapid rate that is not often observed in nature (reviewed by refs. 6 and 75), it has previously been documented in threespine stickleback (e.g., refs. 49, 50, 52, 76, and 77) as well as other fish species. In addition, ancestral and derived populations of mosquitofish (Gambusia affinis) diverged in several life history traits in less than 60 y (78). Similarly, Pacific salmon (Oncorhynchus spp.) adapted to different breeding environments and developed partial reproductive isolation over a period of 13–26 generations (79). Parallel life history evolution has also been observed in the guppy (Poecilia reticulata) in only about a decade as a result of changes in predation pressure (3). Here, for the first time (to our knowledge), we document the fine-scale population genetic patterns associated with similarly rapid phenotypic evolution of stickleback after colonization of new habitats of known age in the wild.

Introgressive Hybridization Maintains the Pool of Standing Genetic Variation.

The possibility for recurrent gene flow between fish in different habitats persists. Our data document that oceanic fish are still entering freshwater ponds to breed or are becoming trapped after high tides or storm surges, as has been observed in other regions of southcentral Alaska (e.g., ref. 76). Although most fish selected for genotyping from freshwater sites containing both ecotypes were morphologically freshwater or oceanic, a few were intermediate in phenotype, and STRUCTURE results indicated a degree of hybridization in these ponds (Figs. 3 and 4). These results suggest that at least some individuals are likely early-generation hybrids between the differentiated oceanic and freshwater ecotypes. Thus, despite the phenotypic diversification in the two habitats, gene flow in sympatry may decrease divergence in neutral genomic regions. Our data also clearly support the “transporter hypothesis” (80) that standing genetic variation in stickleback is maintained by low levels of recurrent hybridization between freshwater and oceanic stickleback.

The constituent genetic loci of traits under differential selection between oceanic and freshwater habitats map to different linkage groups (62), suggesting that numerous regions need to be reassembled. However, linkage disequilibrium created and maintained by strong selection in each habitat (81), perhaps abetted by structural variation in the stickleback genome (43, 81, 82), may significantly reduce the number of independent regions that need to be “transported” and rapidly reintegrated. Supporting this hypothesis is our finding that parallel evolution in this species happens not only on the scale of different continents or dispersed geographic regions as has been amply documented (41, 55, 81, 83), but also on smaller spatial scales as nearby islands or even geographically proximate ponds on the same island. These data argue that rapid parallel evolution over decades in stickleback may occur frequently because it is underlain by a pliant genomic architecture that is itself the product of millions of years of evolution [sensu (84)]. If the findings from stickleback are generalizable to other systems, then rapid evolution in the wild may be more common than previously documented.

Materials and Methods

Site Selection.

By comparing pre- and post-1964 maps and aerial imagery (Aerometrics, US Geological Survey, Bureau of Land Management) of islands in Prince William Sound and the Gulf of Alaska, we identified ponds that now lie on terrain that had been submarine before the 1964 Great Alaska Earthquake. Multiple water bodies fitting this criterion were found on Montague and Danger Islands, in Prince William Sound, and on Middleton Island, in the Gulf of Alaska (Fig. 1 and SI Appendix, Table S1). Middleton Island has only a single pre-1964 pond (Mi12) that rests on an upper terrace formed ∼2,400 y ago (2). Mi12 was stocked with rainbow trout from the Upper Fire Lake (UFL) hatchery (Eagle River, AK) in the 1960s. Therefore, stickleback samples from UFL are also included in our analysis.

Field Collections.

Collections were made in the summers of 2005 (Montague and Middleton), 2010 (Danger and Middleton), and 2011 (Middleton, Montague, and UFL). Threespine stickleback were collected using 0.32- and 0.64-cm mesh minnow traps set near shore and left overnight, killed with an overdose of MS-222 anesthetic, and preserved in 95% ethanol. Salinity was measured at each site by YSI-80 meter and ranged from 0.1 to 1.4 ppt for freshwater sites and 21.4–26.4 ppt for oceanic sites. We used samples from 1 oceanic and 1 post-1964 freshwater site on Danger Island, 5 post-1964 freshwater sites on Montague Island, and 2 oceanic sites, 12 post-1964 freshwater sites, and 1 pre-1964 freshwater site on Middleton Island (Fig. 1 and SI Appendix, Table S1).

Sample Preparation.

Caudal and pectoral fins were clipped for DNA extraction using the Qiagen DNeasy kit. DNA and soma were assigned a unique identification number for association of genotype with phenotype. Bodies were fixed in 10 mg paraformaldehyde in 100 mL 1× PBS for at least 48 h, bleached in a 0.05% hydrogen peroxide solution, stained in a 0.1% Alizarin red S solution, destained in 1% KOH, and preserved in 700 mL/L (70%) undenatured ethanol.

Phenotypic Analysis.

Left sides of stained fish were photographed using a tripod-mounted Canon digital SLR camera and Canon EOS ViewerUtility software. For consistency, one observer counted lateral plates from photographs, and individuals were grouped into one of the following categories: high plated (complete row of anterior, supporting, and posterior plates), partially plated (gap in posterior plates), or low plated (only anterior and/or supporting plates present).

For individuals from six sites on Middleton Island (SI Appendix, Table S2), fish longer than 32-mm standard length (SL) (anterior tip of upper jaw to posterior end of hypural plate) were analyzed for morphology such that only adult phenotypes were included (70). The lengths of the left pelvic spine and second dorsal spine were measured using digital calipers and the number of gill rakers on the first right arch were counted under a dissecting microscope from ≤30 fish per site. The following traits were measured digitally to the nearest 0.01 mm from photographs using tpsDig2, version 2.04, software: SL, snout length (anterior tip of upper jaw to closest margin of orbit), vertical orbit diameter, horizontal opercle width, length of pectoral fin insertion, length of the first lateral plate behind the second dorsal spine, distance between the insertion of the second and third dorsal spines, and length of the base of the dorsal fin (SI Appendix, Fig. S1 and Table S2). The numbers of principal anal, dorsal, and caudal fin rays were counted. Also recorded was the diagonal length of the smooth area anterior to the left pectoral fin, as a proxy for the size of the fin abductor muscle.

Phenotypic Statistical Analyses.

Four of the sites we focused on for phenotypic analysis contained sympatric freshwater and oceanic individuals. Each individual was visually classified as being either oceanic or freshwater based on size, defensive armor, and swimming traits (52, 76). To verify the visual classification, K-means clustering was performed using five ecotype specific traits: SL, number of left lateral plates, and size-adjusted lengths of the left pelvic spine, second dorsal spine, and diagonal of the smooth surface anterior to the pectoral fin. Statistical analyses were concordant with prior visual classifications: only 1.2% of the individuals were categorized differently by the cluster analysis than by visual classification. For all subsequent statistical tests, the results of the visual classification were used to assign each fish to its category.

PCAs were used to uncover the major axes of phenotypic variation among 557 individuals from the six sites (SI Appendix, Tables S2–S5). Not all traits were reliably visible on all photos, and gill rakers were not counted on all individuals; to accommodate these missing morphological data, the ppca function in the pcamethods package (85) was used to perform probabilistic PCAs using the R statistical software platform (86). Because morphometric traits exhibit allometric growth patterns, we performed PCAs with size-standardized data. To size-standardize the data, we used a model I regression to test for a significant relationship between each morphometric trait and SL separately for each population. For those traits with significant regressions, the residuals were calculated and included in PCAs along with unstandardized data for meristic traits and morphometric traits that did not have significant size relationships.

Because PCs are by definition orthogonal, we analyzed each separately. Individual ANOVAs were performed on scores from each of the first three PCs, followed by Tukey post hoc tests, to determine whether populations and ecotypes significantly differed in morphology. For population-level analyses, oceanic-looking individuals from sympatric sites (SO; Mi08, Mi09, Mi13, and Mi19) were pooled due to small sample sizes and compared with each of five freshwater populations as well as oceanic individuals from the marine habitat.

RAD Library Preparation and Sequence Analysis.

Genomic DNA from each of the 1,057 individuals from all three islands and UFL was digested with the restriction enzyme SbfI-HF (NEB), and RAD-seq libraries were created as previously reported (40, 53, 54). Uniquely barcoded samples representing 76–96 individuals were run per lane in 12 total lanes of sequencing on an Illumina HiSeq 2500 platform, and on average, each lane resulted in ∼157 million SE sequences, of which about 113 million were retained (72%). The 101-nt-long reads included 6-nt in-line barcodes to identify individual fish. Raw sequence data were demultiplexed by barcode and filtered for quality using the process_radtags program in the Stacks software suite (87, 88). Reads were aligned against the stickleback reference genome (version BROADs1, Ensembl release 64) using GSnap (89), allowing for up to five mismatches and gaps of length 2, disabling terminal alignments, and requiring unique alignments. The alignments were processed and genotypes were called for each locus across all samples using the pstacks, cstacks, and sstacks programs from Stacks. For a locus to be included in further analyses, we required that it be present in all populations and successfully genotyped in at least 75% of individuals from each population. See SI Appendix, Tables S6 and S7, for tallies of raw and retained reads.

Statistical Approach.

Population genetic statistics (major allele frequency, percent polymorphic loci, and Wright’s F statistics F_IS and F_ST) were calculated for every SNP using the populations program in Stacks (87, 88). For biallelic SNP markers, π is a measure of expected heterozygosity, which is an overall indication of a population’s genetic diversity. F_IS measures the reduction in observed heterozygosity compared with that which is expected for a locus in a population and can indicate nonrandom mating or cryptic population structure (90, 91).

To analyze population structure, we used the populations program in Stacks to output filtered SNP data from all RAD loci across all populations into a file formatted for STRUCTURE (92). Because of computational limitations, we randomly chose three subsets of 1,000 SNPs each to complete the analysis. These three subsets yielded comparable results, so data from only one subset are included. STRUCTURE analyses were performed on all sites together, as well as subsets of the data for fine-scale detection of population structure (Figs. 2 and 3 and SI Appendix, Fig. S2). For all analyses, 10,000 burn-in steps and 10,000 replicates were used, with 10 runs for each potential K (number of genotypic groups). The optimal K for each analysis was chosen using the deltaK method (93).

This same set of SNPs was used in GenoDive (94) to conduct K-means cluster analyses using the same ranges of K tested in STRUCTURE. We also tested a higher level of population structure using AMOVAs (56) that nested sites within islands, within habitat type (defined by oceanic versus freshwater), or within mean lateral plate morph (complete, partial, or low). We used PCA to identify the major axes of genetic variation. We analyzed phylogenetic relationships among our study populations by constructing neighbor-joining trees with 500 bootstrap iterations for three sets of 1,000 loci using POPTREE (95). These three datasets produced concordant trees, so we present results from only one. A phylogenetic tree was also created using Treemix (96).

Materials and Data Availability.

Sampling locations are in Fig. 1 and SI Appendix, Table S1. Single-nucleotide polymorphism (SNP) data from the sequences generated for this study for the 3,000 loci used in these analyses have been deposited in Dryad (dx.doi.org/10.5061/dryad.pn85t).

Fish were collected under Alaska Department of Fish and Game Permits SF-2005-020, SF-2010-029, and SF-2011-153, and all research protocols with vertebrates were approved by the University of Alaska Anchorage and University of Oregon Institutional Animal Care and Use Committees.