Most coho salmon Oncorhynchus kisutch in Washington state spawn at 3 y of age, creating the potential for three temporal populations or “broodlines” at each spawning site. This is generally prevented by a portion of males in each site that mature and reproduce at 2 y of age, resulting in population structure in which the geographic component is stronger than the temporal component. The Quilcene National Fish Hatchery, located on Big Quilcene River in the Hood Canal region of Washington state, selected against late returning coho salmon by excluding all but the earliest returning fish from its broodstock for an unknown number of generations, and restricted gene flow among broodlines by excluding 2-y-old males for 27 generations. The resulting hatchery population exhibited three distinct broodlines that returned in alternating years: an “early” broodline that arrived 1 mo before the wild fish, a “late” broodline that arrived at the same time as the wild fish, and a “middle” broodline that arrived in between these two broodlines. We evaluated temporal and geographic components of population genetic structure in coho salmon from the Quilcene National Fish Hatchery and nine other sites from Puget Sound and the Strait of Juan de Fuca using 10 microsatellite loci. Genetic diversity at the Quilcene National Fish Hatchery was lowest in the early broodline and highest in the late broodline. Divergence among broodlines at the Quilcene National Fish Hatchery was greater than that observed at any other site, and was also greater than that observed between any of the sites. This apparent reversal of the relative magnitudes of temporal and geographic components for this species emphasizes the importance of variable age-at-maturity in shaping population genetic structure.
Coho salmon Oncorhynchus kisutch are anadromous, with spawning populations found in rivers flowing into the North Pacific Basin from approximately 40°N to 65°N (Sandercock 1991). The cultural and economic value of this species has led to its propagation in fish hatcheries throughout its native range and beyond. The typical life history of coho salmon includes a year or more of freshwater residency, followed by smoltification and migration to the marine environment for 18 mo or more before homing to their natal site to spawn (Sandercock 1991). Coho salmon are semelparous, meaning that they spawn once and die shortly thereafter. Typical freshwater habitat includes small coastal streams as well as higher-order headwater streams in larger watersheds. Genetic analyses have revealed weak, but significant, patterns of isolation by distance and emphasized the importance of genetic drift (random changes in allele frequencies over generations due to sampling in finite populations) in shaping population structure in this species (e.g., Olsen et al. 2003).
Most coho salmon in Washington state spawn at 3 y of age (1 y in freshwater and 2 y in the marine environment), creating the potential for three temporally (and genetically) separated spawning groups or “broodlines” at each spawning site. This is generally prevented in the wild by a variable proportion of males at each site that mature and spawn at 2 y (1 y in freshwater and 1 y in the marine environment), known commonly as “jacks.” These precocious males facilitate gene flow among otherwise isolated broodlines. In an evaluation of temporal gene flow in coho salmon populations, Van Doornik (2002) found that the effective proportion of 2-y-old breeders was 0.35 in two wild populations (Ennis Creek and Grizzly Creek), but only 0.02 in a hatchery population (Soos Creek Hatchery). Although 4-y-old spawners represent additional opportunities for gene flow, such individuals comprise less than 5% of spawners returning to streams in British Columbia, and they are virtually absent in Washington state (Sandercock 1991). A genetic survey of coho salmon collections taken from throughout the species' range revealed that population structure included both geographic and temporal components but that the temporal component was generally much weaker (i.e., gene flow across years at a location was much greater than gene flow among locations; Beacham et al. 2011). This result supported the concept of coho salmon populations as geographic (rather than temporal) entities. Population structure in which the geographic component is much stronger than the temporal is common among Pacific salmon species, with pink salmon Oncorhynchus gorbuscha representing a notable exception (e.g., Beacham et al. 1988).
Quilcene National Fish Hatchery (hereafter Quilcene Hatchery), located on Big Quilcene River in the Hood Canal region of Washington state (Figure 1), has been propagating coho salmon for the purpose of enhancing fisheries since 1911. The hatchery population was founded using adult fish returning to Big Quilcene River and has been propagated using returning hatchery-origin fish ever since. Although inadvertently trapped wild-origin fish have been spawned in some cases, no active effort has been made to incorporate them. Sex ratios and spawning protocols have varied over time, but they have remained fixed at one male to one female since 1989 (L. Telles, U.S. Fish and Wildlife Service [USFWS], personal communication). Sporadic transfers of fertilized eggs from other hatcheries in Washington and Oregon occurred before 1974; however, contemporary genetic samples from Quilcene Hatchery still cluster with other populations at Puget Sound and the Strait of Juan de Fuca (Van Doornik et al. 2007), and the population at Quilcene Hatchery is thought to be primarily derived from the ancestral Big Quilcene River stock (USFWS 2009).
Historically, broodstock at Quilcene Hatchery included only the first fish to return to the hatchery. Fish trapping and egg collection were started with the earliest fish to return to the hatchery, and they were terminated once the egg-take goals were reached. Return timing is a heritable trait in salmon (Smoker et al. 1998; Tipping and Busack 2004) that can respond quickly to selection (Hendry and Day 2005; Kovach et al. 2013), and coho salmon returning to Quilcene Hatchery were thus subject to directional selection against late return timing. Beginning in 2000, an effort was made to collect broodstock from throughout the return time, and from 2002 to 2007, fish captured later in the run (October) were preferentially incorporated into the broodstock (L. Telles, personal communication). The impact of the original selective regime was still apparent during these years, as the mean return time for Quilcene Hatchery coho salmon was between 2 and 4 wk earlier than that for adjacent wild populations (USFWS 2009).
Of particular interest, the Quilcene coho salmon appeared to return on a 3-y rotating schedule. In the first year, they would return at the same time as the wild fish (i.e., first half of October). In the second year, they would return a month earlier (i.e., first half of September). In the third year, they would return somewhere in between these times. This cycle would repeat, and fish returning in the first year became known as “late broodline” (e.g., 1988, 1991, 1994, …), fish returning in the second year became known as the “early” broodline (e.g., 1989, 1992, 1995, …), and fish returning in the third year became known as the “middle” broodline (e.g., 1990, 1993, 1996, …). It thus appeared that artificial selection against late return timing had been most successful in the early broodline and least successful in the late broodline. Over the past two generations (since 2009), however, hatchery managers have reported that return timing differences are becoming less pronounced (Dan Magneson, USFWS, personal communication; Figure 2c).
Another standard practice at fish hatcheries (including Quilcene Hatchery) for much of the 20th century was exclusion of jacks from the broodstock. Large fish were desirable to the fisheries that these hatcheries were intended to enhance, so exclusion of smaller fish (including jacks) from the broodstock was common practice. The first record of jacks being integrated in the Quilcene Hatchery broodstock was in 1992, when 44 of 531 (∼8%) males used were 2-y-olds (Table 1). Thus, for 81 y (27 generations), little, if any, gene flow among broodlines was facilitated by jacks. In 2003, a formal target of using 10% jacks for the male broodstock was established, and in 2007 the target was increased to 20%. However, the actual proportion of jacks integrated per year has varied (2–26% since 2003) based on available broodstock returning to the hatchery (Table 1).
Between 2005 and 2009, the USFWS conducted a series of reviews on its Pacific Northwest hatcheries to ensure that programs in this region were part of a scientifically sound strategy for conserving wild stocks and managing fisheries. During the review of the Quilcene Hatchery coho salmon program, questions arose regarding how the propagation history at this facility might have influenced population structure (USFWS 2009). The review identified potential genetic risks posed by the Quilcene Hatchery coho program to that hatchery population as well as adjacent wild populations. Here, we present an analysis of temporal and geographic components of population structure based on microsatellite data in coho salmon collected from Quilcene Hatchery and several other sites in Puget Sound and the Strait of Juan de Fuca (Figure 1; Table 2). Our objective was to investigate whether the Quilcene Hatchery population bears genetic signals of selection against late return timing, the elimination of variation in age at spawning, or both.
The data analyzed here included previously published genotypes for collections of coho salmon in the standardized microsatellite baseline presented by Van Doornik et al. (2007). For this purpose, we selected all collections representing Puget Sound and the Strait of Juan de Fuca for which multiple brood years were represented by 20 or more individuals (Figure 1; Table 2). Our analysis also included novel data for collections from Quilcene Hatchery and the adjacent Little Quilcene River. Methods for sample collection, laboratory analysis, and data quality assessment were described by Van Doornik et al. (2007) for the previously published data, and they are described below for the novel data. We use the term “site” to refer to a stream, tributary, or hatchery where coho salmon spawn (e.g., Little Quilcene River is considered a single site, even though fish were collected from several locations within this river). The term “collection” is used to describe a group of fish sampled from a given site at a given time. Two collections of coho salmon are referred to as “in-cycle” if they are part of the same broodline (i.e., the number of years between hatching of individuals in the two collections is a multiple of 3), and “off-cycle” if they are not. For example, collections of fish which hatched in 2006 and 2009 are in-cycle, but both are off-cycle compared with fish that hatched in 2007 and 2008. Collections from sites other than Quilcene Hatchery were included primarily to provide context; we refer to these sites as “reference sites,” and collections from these sites are referred to as “reference collections.”
Collections from Quilcene Hatchery consisted of fin clips taken from adult (3-y-old) coho salmon broodstock during spawning. Collections from Little Quilcene River consisted of fin clips taken from juvenile (≤1-y-old) coho salmon captured during electro-fishing surveys. Collections from Quilcene Hatchery were taken from throughout the spawning season at the hatchery. Each fin clip was preserved in 200 proof nondenatured ethanol and stored at ambient temperature before genetic analysis.
We extracted DNA from a small (∼2-mm2) piece of each fin clip using a DNeasy-96 Tissue kit (QIAGEN). The PCR was used to amplify 11 microsatellite loci: Ocl8 (Condrey and Bentzen 1998), Oki1, Oki10 (Smith et al. 1998), Oki23 (Spidle et al. 2000), One13 (Scribner et al. 1996), Ots3 (Banks et al. 1999), Ots103 (Small et al. 1998), Ots213 (Greig et al. 2003), Ots503 NWFSC (Naish and Park 2002), OtsG422 (Williamson et al. 2002), and P53 (de Fromentel et al. 1992). Loci were amplified in three multiplex reactions in 10-µL volumes consisting of 5.0 µL of 2× Qiagen Multiplex PCR Master Mix (final concentration of 3 mM MgCl2), 2.0 µL of extracted DNA, labeled oligonucleotide primers corresponding to the loci in each multiplex, and deionized water (Table 3). Liquid handling was performed using a Janus Automated Workstation (PerkinElmer). Thermal cycling began with an initial activation step of 95°C for 15 min; followed by 29 cycles of 95°C for 30 s, annealing temperature for 90 s, and 72°C for 60 s; and concluded with an extension step of 60°C for 20 min. Amplification products were size fractionated using an AB3130xl Genetic Analyzer (Applied Biosystems), and raw microsatellite data (electropherograms) were analyzed using genemapper 4.0. Electropherograms were scored by two researchers, and conflicting genotypes that could not be easily reconciled to the satisfaction of both were deleted. To allow estimation of genotyping error rates, 10% of the samples were extracted and amplified a second time.
Collections of juvenile salmonids may contain groups of siblings and may thus yield inaccurate estimates of structure of the populations from which they originate (Allendorf and Phelps 1981). Two measures were taken to reduce the probability that our Little Quilcene River collections included overrepresentation of sibling groups. First, we sampled fish from throughout the drainage, with no more than a few individuals taken from any single location. Second, we used multiple statistical analyses of the genetic data from these fish to allow identification of sibling groups. We used the likelihood method described by Wang (2004) and implemented in the program colony to identify groups of full siblings. This group-likelihood approach to sibship reconstruction uses a simulated annealing technique to search for the best (maximum likelihood) configuration of siblings within a data set. It also allows incorporation of genotyping error rates to prevent false exclusions of sib groups. This analysis was repeated six times on juveniles collected in each year (two runs with three replicates each). We also used the graph based and triplet enumeration algorithms for maximal sibling group ([MSG]; Almudevar and Field 1999) estimation implemented in the program prt (Almudevar and Anderson 2012) to identify full-sibling groups. Groups identified by colony and by both MSG algorithms were labeled as “siblings” and all members but one of each such group were removed. Although it is possible that groups of siblings also exist in the previously published baseline, we chose to retain the data in their published state (see Van Doornik et al. 2007 for a description of quality assessment performed on those data).
Testing for genotypic ratios that departed from Hardy–Weinberg equilibrium was conducted using Fisher's exact tests in genepop version 4.0 (Rousset 2008). The log likelihood ratio statistic (G test) was used to test for linkage disequilibrium (LD) between each pair of loci in each collection. Critical values for both tests were corrected for multiple comparison using standard Bonferroni corrections (Rice 1989). For Hardy–Weinberg equilibrium, α = 0.05/11 loci = 4.5 × 10−3. For LD, α = 0.05/55 pairwise comparisons per collection = 9.1 × 10−4.
To measure genetic diversity within collections, we calculated mean (across loci) heterozygosity (Hs) and allelic richness (number of alleles observed per collection, corrected for unequal numbers of individuals per collection [AR]) for each collection using fstat 188.8.131.52 (Goudet 2001). To assess the significance of differences in Hs and AR between Quilcene Hatchery and Little Quilcene River, a permutation scheme in which individuals were allocated at random among sites (keeping sample sizes the same) with 104 iterations was used. The P value associated with each difference was assessed as the proportion of randomized data sets yielding larger differences than the observed values (2-tailed test with α = 0.05). The inbreeding coefficient ([f]; Weir and Cockerham 1984) was also estimated for each collection using fstat.
Divergence among each pair of collections was estimated as FST (θ; Weir and Cockerham 1984), and the statistical significance of each estimate was evaluated by comparison to a null distribution based on 104 replicate data sets in which individuals were permuted among collections. To compare observed FST estimates among the three Quilcene Hatchery broodlines to values that might be expected based on genetic drift over 27 generations, we calculated expectations based on the approximation (Wright 1943), where t was the number of generations and Ne was the effective population size (the size of an ideal population that experiences genetic drift at the same rate as the observed population). This approximation assumes there is no gene flow among populations (m = 0.0). Divergence among collections was further assessed using tests of allele frequency heterogeneity as implemented in fstat and represented graphically by performing a correspondence analysis on the table of multilocus genotypes using genetix (Belkhir et al. 2004). To evaluate statistical significance of changes in FST among Quilcene Hatchery broodlines in the most recent three generations, 95% confidence intervals (CIs) were generated for the 2000–2002 generation and the 2009–2011 generation using the bootstrap over loci procedure in FSTAT.
We estimated effective number of breeders per generation (Nb), an analog of effective population size (Ne) that refers to number per year rather than per generation (Waples and Teel 1990), using the linkage disequilibrium method (Waples 2006) in the program ldne (Waples and Do 2008). Alleles with frequencies less than 0.01 were excluded, and 95% CIs for each estimate were assessed using a jackknife on locus procedure (Waples 2006). To assess possible reduction in Nb due to broodstock sex ratios, we also estimated Nb for Quilcene Hatchery collections based on numbers of females (Nf) and males (Nm) crossed in the hatchery each year following Kimura and Crow (1963) as Nb = (4NmNf)/(Nm + Nf). Numbers of males and females used at the hatchery were read from hatchery spawning records for the years for which they were available (1984–2011).
Bayesian clustering of individuals as described by Pritchard et al. (2000) and Falush et al. (2003) and implemented in the program structure was used to evaluate the number of populations (K) sampled at Quilcene Hatchery. The Markov Chain Monte Carlo was run for 4 × 105 steps following 105 burn-in steps. Twenty replicate analyses were performed for each value of K between 1 and 9. The ad hoc statistic ΔK, based on the rate of change in the log probability of data for successive values of K, was used to estimate the number of populations following Evanno et al. (2005) and implemented in the program structure harvester (Earl and Vonholdt 2012).
The 11 microsatellites examined exhibited 321 alleles across 30 collections and 1,864 samples, with individual loci exhibiting between 14 (Ots3) and 56 (Ots103 and OtsG422) alleles (Table 3). Comparison of initial genotypes to those generated for error rate assessment revealed no conflicting genotypes, suggesting a low genotyping error rate. Nevertheless, an error rate of 0.001 was incorporated into the sibship analysis. Genotype data are available in Appendix A1.
Tests for departures from Hardy–Weinberg equilibrium yielded significant results at Ots103 in 21 of 30 collections with an uncorrected α and in 16 of 30 collections after correction for multiple comparisons. Most (20/21 uncorrected α and 16/16 corrected α) departures were associated with positive FIS values, suggesting that the null allele(s) previously documented for this locus (Small et al. 1998) were likely a contributing factor. We thus excluded Ots103 from further analyses. The remaining 10 loci exhibited departures in five or fewer collections with an uncorrected α. After correction for multiple comparisons, only two results were significant: Ots213 in Soos Creek Hatchery 1997 and Ocl8 in Snow Creek 2003.
Graph-based and triplet enumeration algorithms for msg both identified a single trio of full siblings in the 2006 sample from Little Quilcene River and no full-sibling groups in the 2005 sample. The maximal full sibling configuration from colony included only a single pair of siblings. The pair included two of the individuals identified as a trio using msg (the third individual was identified as a half-sibling by both colony runs). A large number of half siblings were also identified, both within and among collections. Half-sibling pairs identified among collections seemed biologically unlikely, and we interpreted the large number of such pairs identified as evidence of our lack of power to exclude false half-sibling pairs. Based on these results, we removed two of the individuals from the trio identified using msg. We did not have confidence in our ability to correctly identify half-sibling groups, so no other individuals were removed.
Significant (corrected α = 0.05/45 tests = 0.001) LD was observed in several collections; however, it was noted that Quilcene Hatchery and Snow Creek were the only two sites from which multiple collections exhibited LD (Table 2). In the first generation of Quilcene Hatchery collections examined (2000–2002), the late broodline (2000) exhibited the greatest proportion of pairs of loci in LD (17.8%). In subsequent generations (2007–2008 and 2009–2011), collections from all three broodlines exhibited high proportions of locus pairs in LD (e.g., 2007 early with 11.1%, 2009 late with 8.9%, and 2011 middle with 8.9%). Little Quilcene River exhibited 2.2% of locus pairs in LD in the 2005 collection, and no pairs in the 2006 collection.
Estimates of Nb based on genetic data were generally highest in wild populations (e.g., Ennis Creek, Big Beef Creek, Grizzly Creek, and Dewatto River) and lower in hatchery populations (Quilcene Hatchery and Skagit Hatchery); however, there were several exceptions (e.g., high Nb in Soos Creek Hatchery) and CIs were often very large (Table 2). The lowest Nb values were observed in Snow Creek, a naturally spawning population that was nearly extirpated in the late 1990s when it became the target for an integrated recovery hatchery program (Washington Department of Fish and Wildlife 2003). Average Nb for all temporal collections from Quilcene Hatchery was 149, and jackknife CIs for every collection included this value. Of the three broodlines at Quilcene Hatchery, the middle broodline (2002, 2008, 2011) had the largest 95% CIs (Table 2). The CIs for Little Quilcene River were higher than 149 in 2005, but included this value in 2006. Estimates of Nb at Quilcene Hatchery based on sex ratios for each brood year were between 6 and 11 times larger than estimates based on genetic data (Table 1).
The AR was lower at Quilcene Hatchery than at any other location examined. Of the three broodlines, the late (2000, 2009) had the highest AR (Table 2; Figure 3). Heterozygosity varied less than AR, but followed a similar pattern: Quilcene Hatchery samples generally had lower Hs than samples from elsewhere, and collections from the late broodline exhibited the highest Hs values for Quilcene Hatchery (Table 2). In comparing measures of diversity between Quilcene Hatchery and Little Quilcene River, the 2-tailed test results indicated that Hs was not significantly different (P = 0.063), but that AR was (P = 0.026).
The global estimate of FST was 0.032, with pairwise estimates among collections ranging from 0.000 to 0.078 (Table A1). Within locations, divergence among in-cycle years (FST = 0.003) was generally lower than among off-cycle years (FST = 0.015; Table 4). Divergence among off-cycle years was highest at Quilcene Hatchery, but appeared to be diminishing in recent years. For the 2000–2002 generation, FST = 0.061 (95% CI 0.046–0.078), whereas in the 2009–2011 generation, FST = 0.036 (95% CI 0.032–0.040). Among wild populations in Hood Canal (Little Quilcene River, Dewatto River, and Big Beef Creek), average FST among off-cycle collections was 0.006 (Table 4). Assuming 27 generations of genetic drift, starting FST = 0.006, m = 0, and Ne = 150 (based on average Nb estimates described above), we estimated an expected value of FST = 0.092 among broodlines at Quilcene Hatchery. If Ne was increased to 250 individuals, then expected FST = 0.059. Nearly all collections examined were statistically different from each other, as indicated by pairwise FST tests and allele frequency heterogeneity tests (α = 0.05). Nonsignificant FST estimates were observed among some pairs of in-cycle collections from Quilcene Hatchery (2001 vs. 2007, 2002 vs. 2008) and Soos Creek Hatchery (1995 vs. 1997, 1996 vs. 1998), as well one pair from Grizzly Creek (1996 vs. 1999; Table A1). The same pairs, except one (Quilcene Hatchery 2001 vs. 2007; P = 0.025), exhibited nonsignificant allele frequency differences. Both tests yielded significant results between all other pairs of collections.
The correspondence analysis revealed that Quilcene Hatchery and Little Quilcene River collections were distinct from the reference collections and that divergence among three broodlines at Quilcene Hatchery was a dominant feature in the context of all collections examined here (Figure 4). The primary axis separated Quilcene Hatchery and Little Quilcene River from the reference collections and accounted for 19.4% of the variance. The secondary axis separated the early broodline (2001, 2007, 2010) from the middle broodline (2002, 2008, 2011) and accounted for 13.4% of the variance. The third and final axis examined separated the Little Quilcene River samples from Quilcene Hatchery samples and accounted for 7.7% of the variance.
Results of the STRUCTURE analysis revealed that the mean log probability of the data (X) showed the greatest increase between K = 2 and K = 3. For values of K greater than 3, the log P(X) did not appear to increase; however, the standard deviation of the mean estimates did increase (Figure A1). Evaluation of ΔK with successive values of K from 2 to 8 revealed the most likely number of populations at Quilcene Hatchery was three (plots of the log probability of the data (X) and ΔK for K = 1–8 are presented in Figure A1). At K = 3, proportions of inferred ancestry indicated less mixing among the three broodlines for individuals collected in 2000–2002 than those collected in 2009–2011 (Figure 5).
Genetic divergence among off-cycle return years of coho salmon at Quilcene Hatchery revealed the existence of three distinct populations corresponding to the early, middle, and late broodlines identified earlier based on run timing. The magnitude of divergence between these three broodlines was greater than that among many wild populations of coho salmon, even approaching the degree of divergence observed among previously described evolutionarily significant units, the units of coho salmon that are distinct enough to merit treatment as individual species under the US Endangered Species Act (ESA 1973, as amended; Waples 1991). The broodlines exhibiting return timing most divergent from wild populations (early and middle broodlines) also exhibited microsatellite allele frequencies most divergent from wild populations and exhibited the lowest genetic diversity. We observed a recent decrease in the divergence between broodlines at Quilcene Hatchery coinciding with the increased incorporation of jacks into the hatchery broodstock. By systematically excluding jacks for 81 y, Quilcene Hatchery provided a rare opportunity to study the impacts that variable age at maturity has on coho salmon population structure.
Divergent return time exhibited by Quilcene Hatchery broodlines
An initial question that arose when interpreting our results was why the three broodlines returned at three different times. The observed pattern suggests that culture practices impacted the early broodline to a greater extent than they did the late broodline and that gene flow among broodlines was subsequently restricted. Different impacts on the three broodlines might have been caused by differences in selective pressure or by differences in response to selection and possibly genetic drift. A hypothesis involving differences in selective pressure might speculate that the founders of the early broodline were captured earlier than founders of the other two broodlines or that some other environmental perturbation was greater during the year the early broodline was founded. Of course, for this to explain the observed pattern, we would need to assume that the return timing shift was accomplished in a single generation, or else that the selective pressure was on a 3-y cycle (e.g., every third year collection was done early). Alternatively, the magnitude of selection may have been relatively constant, and divergence for the trait (return timing) might reflect stochastic fixation at different loci influencing this trait in the three broodlines. Genetic variation would have likely been lost due to selection against all but the earliest returning fish, and the population might have also been subject to increased genetic drift often associated with hatchery populations (e.g., Waples and Teel 1990). We are not aware of collection practices or environmental conditions that were repeating on a 3-y cycle, but we cannot rule out such conditions either. Regardless of the initial cause of the divergent return timing exhibited by the three broodlines, we expect that the pattern persisted because of a lack of gene flow among broodlines (i.e., exclusion of jacks from the hatchery broodstock).
Impacts of elimination of variable age at maturity
During spawning in the wild, 3-y-old males typically use a strategy of fighting for and defending access to females, whereas jacks typically use a strategy that involves sneaking in adjacent to a courting pair during spawning and attempting to fertilize a portion of the eggs (Gross 1985). As the frequency of a given strategy increases in a population, increased competition among males adopting that strategy will reduce the mean fitness associated with it (Gross 1985; see also Hutchings 2004). This frequency-dependent disruptive selection is expected to maintain jacks at some frequency within each population, which will facilitate gene flow among broodlines (e.g., Van Doornik 2002) and thus increase Nb in the population (Johnstone et al. 2013). Generally low point estimates of Nb in Quilcene Hatchery, Skagit Hatchery, and Snow Creek (Table 2) may reflect reductions due to exclusion of jacks, but the large and overlapping confidence intervals suggest caution in interpreting this result. Reductions in Nb for hatchery populations relative to adjacent wild populations may be based on a number of culture practices (e.g., use of small numbers of fish, skewed sex ratios, large variance in reproductive success; see review by Naish et al. 2008). For example, high variance in offspring produced per female in hatchery populations can greatly reduce Nb (e.g., Hedrick et al. 1994). A recent comparison of hatchery-origin and natural-origin collections of steelhead Oncorhynchus mykiss from the Hood River, Oregon, revealed reduced Nb and increased LD in the hatchery-origin collections (Christie et al. 2012). Increased LD in the Quilcene Hatchery and Snow Creek collections might reflect high variance in reproductive success due to culture practices, but it might also reflect recent mixing of divergent lineages. This seems especially likely in the case of Quilcene Hatchery, where incorporation of jacks in recent generations essentially amounted to crossing divergent (FST ∼ 0.06) broodlines. Comparison of the STRUCTURE plots for the earliest (2000–2002) and latest (2009–2011) generations examined here suggested that this reincorporation was successful in facilitating gene flow in a stepping-stone pattern (i.e., from the early broodline to the middle broodline, from the middle broodline to the late broodline, and from the late broodline to the early broodline).
Several of our results indicated greater divergence among the three broodlines at Quilcene Hatchery than among the reference sites and thus challenged our concept of coho salmon populations as geographic entities. The reference collections (especially those representing wild populations) exhibited geographic components of population structure that were much greater than temporal components (Table A1) as expected based on past surveys (Van Doornik et al. 2007; Johnson and Banks 2008; Beacham et al. 2011). In such cases, where jack-mediated temporal gene flow is relatively strong, the term broodline has little relevance. In contrast to this pattern, we observed greater divergence among off-cycle years at Quilcene Hatchery (mean pairwise FST = 0.051) than among pairs of reference sites (FST = 0.003–0.043, mean = 0.018). Moreover, our estimate of temporal divergence among broodlines at Quilcene Hatchery in 2000–2002 (FST = 0.061) was comparable to past estimates of divergence among evolutionarily significant units (e.g., Hood Canal vs. Columbia River, FST = 0.050; Hood Canal vs. Oregon, FST = 0.048; Hood Canal vs. California, FST = 0.067; Beacham et al. 2011). This observed value fell closer to that predicted by a model of 27 generations of genetic drift with Ne = 250 (FST = 0.059) than with Ne = 150 (FST = 0.092), suggesting that some gene flow might have occurred during this time. Incorporation of wild fish in the broodstock, especially in the late broodline, seems a likely explanation. The few other coho salmon populations where temporal structure approaches or is greater than geographic structure are also associated with hatcheries (e.g., Quinault Hatchery; USFWS 2011), presumably reflecting a lack of temporal gene flow in these populations as well. This reversal of the relative magnitudes of temporal and geographic population structure in populations such as Quilcene Hatchery illustrates an important role that jacks may have in shaping coho salmon population structure.
Genetic risks associated with propagation of Quilcene Hatchery coho salmon
The USFWS recently completed a broad-scale evaluation of the risks and benefits associated with each of its National Fish Hatcheries. In the review of Quilcene Hatchery, two types of genetic risks were identified (USFWS 2009). The first was the genetic risk faced by the hatchery population. It was reasoned that if the population existed as three divergent broodlines, then the loss of returns in any one year due to stochastic events could result in the loss of unique genetic resources. Our results suggest that divergence among the broodlines has decreased in recent years and the amount of genetic diversity in what were the relatively depauperate broodlines (early and middle) appears to have increased in a short number of generations since jacks were reintroduced. The STRUCTURE results indicated greater mixing among broodlines in 2009–2011 than in 2000–2002 (Figure 5), suggesting that returning adult fish (from which the samples were taken) are carrying an increasing fraction of genetic resources from multiple broodlines. Finally, reduced divergence among the broodlines in the 2009–2011 generation (FST = 0.036) compared to the 2000–2002 generation (FST = 0.061) indicates increased gene flow (Table A1). These results suggest the incorporation of 2-y-old males at Quilcene Hatchery has been at least partially successful in addressing genetic risks to the hatchery population.
The second genetic risk of the Quilcene Hatchery program involved potential introgression with adjacent wild populations. The USFWS (2009) found that the native population of natural-origin coho salmon that once inhabited the Big Quilcene River had been heavily influenced by the hatchery, and probably no longer existed as a self-sustaining entity. Several nearby rivers, however, are home to naturally spawning populations that likely represent what remains of the ancestral coho salmon from this region (e.g., Little Quilcene River, Big Beef Creek, and Dewatto Creek). Introgression of hatchery fish into such populations may have several outcomes, including changes in Nb via the Ryman–Laikre effect (reduced effective population size accompanying the demographic boost provided by hatchery supplementation programs; Ryman and Laikre 1991), hybridization and concomitant reduction in fitness in the wild (Araki et al. 2008; Theriault et al. 2011), and ultimately homogenization or replacement of the wild populations (e.g., Williamson and May 2005). Eldridge et al. (2009) compared historical (1938) and contemporary (2005) collections of Puget Sound coho salmon and found a slight decrease in AR and weakening of isolation-by-distance, which they attributed to hatchery propagation in this region. Alternative strategies for reducing this risk include a “similar strategy” in which hatchery fish are raised to be as similar to wild fish as possible, and a “divergent strategy” in which hatchery fish are raised to be as divergent from wild fish as possible. The goal under the similar strategy is to create hatchery fish so similar to wild fish that there is no loss of fitness when the two cross (crosses common, but fitness cost is low). The goal under the divergent strategy is to make hatchery fish so different from wild fish that introgression will not occur (crosses rare, but fitness cost is high). Baskett and Waples (2013) found the greatest reduction in fitness of wild populations occurred in the transition between the extremes of the similar and divergent paradigms. Several of the collections examined here represent small populations which are geographically near Quilcene Hatchery, yet have retained distinct allele frequencies, lower LD, and higher AR than Quilcene Hatchery. Tests of FST significance and allele frequency heterogeneity suggest that populations such as those in Little Quilcene River, Dewatto River, and Big Beef Creek have not been genetically homogenized by Quilcene Hatchery. One result that may be indicative of gene flow from Quilcene Hatchery to a wild population is the proximity of Little Quilcene River to Quilcene Hatchery in the correspondence analysis. These results suggest that Quilcene Hatchery has generally succeeded in implementing a divergent strategy but that there may be some gene flow especially in the case of the Little Quilcene River. If increasing gene flow among broodlines at Quilcene Hatchery results in later return timing for the middle and early broodlines, it seems possible that this could lead to increased potential for introgression between hatchery and wild populations. Because we do not know the extent to which differences in return timing have limited gene flow between these populations, however, the impact of increased gene flow among broodlines at Quilcene Hatchery on wild populations is difficult to predict.
Gene flow among brood years in wild populations of coho salmon, as facilitated by jacks, provides a mechanism for the maintenance of genetic diversity in this species and creates a hierarchical structure in which the temporal component is small relative to the geographic component. The results presented here revealed that a nearly century-long experiment in which jacks were excluded from the breeding population accomplished what one might predict: the creation of three populations from one. They further suggest that the recent reincorporation of jacks has predictably resulted in increased gene flow among the broodlines.
Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any archived material. Queries should be directed to the corresponding author for the article.
To cite this archived material, please cite both the journal article (formatting found in the Abstract section of this article) and the following recommended format for the archived material.
Smith CT, Baumsteiger J, Ardren WR, Dettlaff Y, Hawkins DK, Van Doornik DM. 2014. Data from: Eliminating variation in age at spawning leads to genetic divergence within a single salmon population. Journal of Fish and Wildlife Management 6(1):4–18. Archived in Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.4kp00
Table A1. Pairwise FST estimates for collections of coho salmon Oncorhynchus kisutch from Quilcene Hatchery and other sites in Puget Sound and the Strait of Juan de Fuca taken between 1997 and 2011. Values highlighted in gray were not statistically significant (α = 0.05) based on a permutation test.
Found at DOI: http://dx.doi.org/10.5061/dryad.4kp00 (57 KB PDF).
Figure A1. Numbers of inferred populations (K) in samples of Quilcene Hatchery coho salmon Oncorhynchus kisutch based on Bayesian clustering analysis. Collections were taken between 1999 and 2011. Both (a) mean (±SD) log probability of the data (X), and (b) ΔK (rate of change in the log probability of data) are plotted for K = 1–8. These results suggest that our collections from Quilcene Hatchery represent three populations.
Found at DOI: http://dx.doi.org/10.5061/dryad.4kp00 (76 KB PDF).
Appendix A1. Genotype data for ten microsatellite loci in collections of coho salmon Oncorhynchus kisutch from Quilcene Hatchery and other sites in Puget Sound and the Strait of Juan de Fuca taken between 1997 and 2011.
Found at DOI: http://dx.doi.org/10.5061/dryad.4kp00 (157 KB TXT).
Reference A1. [USFWS] U.S. Fish & Wildlife Service. 2009. Quilcene, Quinault, and Makah National Fish Hatcheries: Assessments and Recommendations. Portland, Oregon.
Found at DOI: http://dx.doi.org/10.5061/dryad.4kp00 (2859 KB PDF).
Reference A2. [USFWS] U.S. Fish & Wildlife Service. 2011. Genetic profile for Quinault Hatchery coho salmon. Longview, Washington.
Found at DOI: http://dx.doi.org/10.5061/dryad.4kp00 (54 KB PDF).
We are grateful for information regarding historical and present hatchery practices provided by Ron Wong, Larry Telles, Dan Magneson, Dave Zajac, and Don Campton. Excellent laboratory assistance was provided by Stephanie Walden, Matthew Diggs, Jennifer Von Bargen, and Brice Adams. We are grateful to Patricia Crandell, two anonymous reviewers, and the Associate Editor for providing helpful comments on earlier versions of this manuscript.
This work was funded by the USFWS Pacific Region.
Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Citation: Smith CT, Baumsteiger J, Ardren WR, Dettlaff Y, Hawkins DK, Van Doornik DM. 2014. Eliminating variation in age at spawning leads to genetic divergence within a single salmon population. Journal of Fish and Wildlife Management 6(1):4–18; e1944-687X. doi: 10.3996/122013-JFWM-086
The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service.