Introduction

Population structure has important consequences for the evolution of a species and for implications for conservation, because drift, selection, migration and mutation act to differentiate allelic diversity among semi-isolated populations (Whitlock, 2004). Phenotypic diversity can evolve over short genetic and geographic distances (McKay and Latta, 2002), and semi-isolated populations can diversify and become locally adapted (Taylor, 1991). Furthermore, the accumulation of deleterious mutations is dependent on the global effective population size, as well as on the levels of gene flow between populations (Waples, 2002; Whitlock, 2004). Semi-isolated populations represent the range of diversity within a species or conservation unit over a variety of environmental conditions, and afford resilience to perturbations (Allendorf and Luikart, 2007). To be most effective, conservation actions should consider the existing population structure and the processes that lead to structuring among populations.

The recovery of a species may require human intervention, but care is needed to maintain population structure. Activities such as supportive breeding and translocations are often used to increase population abundance, provide surplus fish for harvest and expand a species’ range. However, these activities present risks after the escape or release of domesticated or nonindigenous fish (Ellstrand et al., 1999; Barilani et al., 2005; Halbert et al., 2005; Naish et al., 2007) that may culminate in introgression or a partial or complete displacement of native populations (Hansen, 2002; Haygood et al., 2003; Vasemagi et al., 2005). Ongoing supportive breeding could result in a loss of genetic diversity and reductions in effective population size (Ryman and Laikre, 1991; Gaffney et al., 1996) and could alter the sub-population structure (Utter, 2001; Birnbaum et al., 2003). Despite these risks, supportive breeding and translocations may provide an effective avenue for the recovery of a population's fitness or for the prevention of a loss of genetic diversity (Tallmon et al., 2004; Edmands, 2007). It thus becomes important to establish the relationship between evolutionary divergences and impacts on population structure that might be expected after human-mediated hybridization (McClelland and Naish, 2007).

When donor and recipient populations differ substantially at neutral genetic loci, assignment tests can identify descendants of introduced fish (Cornuet et al., 1999); however, this task is more difficult among closely related populations that experience high levels of gene flow wherein assignment tests are less reliable (Cornuet et al., 1999), and when baseline data are not available for all source populations. In the absence of complete baseline data, Bayesian assignment approaches have been used to identify fish with native, introduced or introgressed ancestry (Hansen et al., 2001). Alternatively, transfers may alter the fine-scale population structure. Archived tissues can provide an invaluable historical baseline for exploring changes to the allelic diversity within and among populations resulting from extensive human intervention (Nielsen et al., 1999).

Coho salmon populations (Oncorhynchus kisutch) on the West Coast of North America provide a framework for studying the effects of introgression due to supportive breeding and translocations. These populations have been classified within conservation units (here, evolutionarily significant units, ESUs; Waples, 1991) that represent divergent evolutionary components of species diversity. Several studies have shown that barriers to introgression tend be greater in anadromous salmonids than in freshwater or marine fish species, but introgression is frequent after transfers among closely related populations (reviewed in Utter, 2001). Translocations of coho salmon between two ESUs have been shown to be largely unsuccessful (Withler, 1982; Ford et al., 2004), but little is known about the effects of culture on population structure within an ESU. Coho salmon in the Puget Sound ESU of Washington State have been propagated and released into the wild for over 100 years to enhance harvest opportunities (Stickney, 1996), and the composite populations in this ESU have been designated a Species of Concern due to possible changes to genetic diversity caused by culture practices (NMFS, 2004). In an earlier study, we showed that the genetic distance among contemporary populations was the smallest between rivers that experienced the greatest numbers of transfers, and sub-population genetic divergence was reduced in rivers that had the highest number of hatchery releases (Eldridge and Naish, 2007). Population structure could not be explained by geographical distance. However, without a historical knowledge of gene diversity and gene flow before large-scale transfer and culture activities, it is impossible to eliminate the possibility that the observed patterns were due to natural causes.

Here, we examined the stability of the population structure and genetic diversity of coho salmon within an ESU that has experienced ongoing releases into the wild of cultured fish. We used microsatellite loci to compare genetic diversity, and the magnitude and patterns of gene flow among populations occupying six rivers around Puget Sound, Washington in 1938 and in the present, using archival samples, and we compared our results with those published earlier (Eldridge and Naish, 2007).

Materials and methods

Study area

This study was conducted in six rivers in the Washington State portion of the Puget Sound coho salmon ESU (Figure 1). Within each river basin, hatcheries have released into the wild millions of coho salmon from a variety of intra- and interbasin sources. Coho salmon releases in and around Puget Sound between 1952 and 2004 have been described previously (Eldridge and Naish, 2007). The goals of many coho salmon hatchery programs in Puget Sound were undefined until recently (Mobrand et al., 2005), and likely comprised a mixture of fishery enhancement and conservation activities. However, in all cases, broodstock was collected from adults during their spawning migrations in freshwater, and offspring were released into the wild as juveniles, often into multiple creeks in the same year.

Figure 1
figure 1

Map of coho salmon sampling locations around Puget Sound, Washington. Numbers correspond to sample labels in Table 1.

Full size image

Tissue samples and microsatellite analyses

Coho salmon tissues from the six Puget Sound rivers were available from two time periods (Table 1). Historical scales or skin patches were collected from adult coho salmon sampled from spawning grounds and from hatcheries in these rivers during the spawning cycle that began in the fall of 1938 (N=367). More recently, scales, fin clips, operculum punches or clips from pectoral girdle were collected from adult and juvenile coho salmon from hatcheries and from the wild between 2001 and 2005 (N=643); scales were stored dry on cards at an ambient temperature and other tissues were stored in ethanol. Additional scales were available for an intermediate period (1968) from adult coho salmon returning to the Voights Creek Hatchery on the Puyallup River (N=73). A sample consisted of fish from a single creek from the same year where possible, except samples labeled as either North Fork, Middle Fork or South Fork in Table 1, which refer to small collections combined from nearby creeks, or samples labeled as Trap, which were collected from the mainstem of a river.

Table 1 Sample information and summary of genetic variation within samples for coho salmon from six rivers around Puget Sound
Full size table

DNA was extracted using the DNeasy kit (Qiagen, Valencia, CA, USA) following the manufacturer's directions, but the historical samples collected in 1938 were eluted in two centrifugation spins of 50 μl each. A total of 12 modern samples were previously screened (Eldridge and Naish, 2007). All historical, intermediate and four modern samples (NOnf05, NOmf01, NOsf01 and SGjk05) were unique to this study. We determined genotypes for all fish at eight microsatellite loci (Olc8, Ogo1a, Ogo2aI, Ogo2aII, Omy1011, One13, Ots3 and P53). A total of 11 loci were originally screened (Eldridge and Naish, 2007), but genotypes in the historical samples could be reliably determined at only eight loci. Alleles at the locus with the shortest products (Ots503, 73–95 base pairs), and at two loci with large products (Ots505 and Oki23, >230 base pairs), could not be distinguished from PCR artifacts and these loci were dropped from the analyses.

Amplification of microsatellite loci in the modern samples followed procedures described in Eldridge and Naish (2007). The PCR reaction reagents for the historical samples are described in Supplementary Table 1 and the PCR thermocycling conditions in Supplementary Table 2. PCR products were visualized on the MegaBACE automated sequencer (Amersham Biosciences, Piscataway, NJ, USA) using a ROX-labeled ET 550 internal size standard. Microsatellite allele size was determined with the Genetic Profiler v.2 software package (Amersham Biosciences). PCR thermocycling and allele scoring were repeated for 118 genotypes from 1938 to determine the rates at which alleles were scored consistently among historical samples. Approximately, 94% of alleles (221 of 236) were scored consistently at all eight loci. Most discrepancies involved a fish being scored once as a heterozygote and a second time as a homozygote so that only one allele was consistent between repeat amplifications.

Statistical analyses

Each locus within each sample was tested for the presence of null alleles, and for scoring errors due to stuttering or allelic dropout (Micro-Checker v.2.2.1, Van Oosterhout et al., 2004). The genetic diversity within each sample was described by determining the number of unique allele classes at each locus, the observed and expected heterozygosities and the sample inbreeding coefficient (FIS) (GenAlEx v.6.1, Peakall and Smouse, 2006). To test for panmixia within samples, the significance of FIS over all loci in each sample was tested using the Hardy–Weinberg probability exact test (Genepop v.3.4, Raymond and Rousset, 1995) and tests for genotypic equilibrium were conducted between all pairs of loci (FSTAT v2.9.3.2, Goudet, 1995). ‘Outlier’ loci or samples were defined as those that consistently contributed to significant tests at the P<0.05 level. Outlier loci may arise because of null alleles, inconsistencies in amplification or linkage between loci, and outlier samples may suggest deviations from random mating. To determine whether any locus, locus pair or sample was involved in a disproportionately large number of significant deviations from Hardy–Weinberg or genotypic equilibrium, χ2-tests were conducted comparing the observed number of significant tests with the expected number, if significant results were distributed randomly across loci, locus pairs or samples.

Changes in genetic diversity among and within rivers were assessed by testing whether the number of unique allelomorphs changed over time. We compared the allelic richness between samples (Na, number of allelomorphs in samples of a common size) to test for a change in the number of allelomorphs. For each river and period, the allelic richness for a locus was determined by counting the number of allelomorphs in a draw of 2N alleles (as opposed to genotypes), taken without replacement, from a pool of all samples from a single river during the historical (1938), intermediate (Puyallup River in 1968) or modern (2001–2005) period. The draws were repeated 1000 times and the average was recorded. The process was repeated after combining all the samples from a given time period to compare the overall historical and modern periods. Wilcoxon paired sample tests (Zar, 1999) were conducted to determine if the allelic richness among all eight loci was significantly different between historical samples (1938) and modern samples (2001–2005) from the same river, and between all rivers combined. The Samish River was excluded from the allelic richness tests because of small historical sample size.

The temporal and spatial scales at which allelic diversity was distributed within and between historical (1938) and modern samples (2001–2005) were evaluated by examining the genotype diversity within and among samples grouped by period or by river, using a series of hierarchical analysis of molecular variance (AMOVA) tests, interpolating for missing data with 999 random permutations to test for significance (GenAlEx v.6.1, Peakall and Smouse, 2006). We used genotype diversity (Φ-statistics) rather than allelic diversity (F-statistics) because of deviations from the Hardy–Weinberg equilibrium in some samples (Supplementary Table 3).

Isolation by distance among rivers historically (1938) and recently (2001–2005) was assessed by Mantel tests (Mantel and Valand, 1970), with 999 random permutations to determine significance (GenAlEx v.6.1). We compared rivers rather than individual samples because we were interested in the effects of transfers on the gene flow between rivers. In this analysis, the genetic distance between each pair of rivers (here, ΦRT) was determined using an AMOVA that interpolated for missing data (GenAlEx v.6.1). When only one site was sampled in each river, the ΦST value estimated by AMOVA was used as the ΦRT value. Geographic distances between rivers were determined to be the shortest path in fresh or saltwater between the respective river mouths using the distance calculator in the US Geological Survey National Map Viewer (http://nmviewogc.cr.usgs.gov/viewer.htm).

Pairwise tests of population differentiation based on randomizing multi-loci genotypes between samples were conducted (FSTAT v2.9.3.2, Goudet, 1995) to estimate the reproductive isolation between pairs of samples within and between rivers and between different years, and ΦST was calculated for each pair of sample (GenAlEx v.6.1). Population structure was visualized using a multidimensional scaling plot of ΦST (SPSS, version 10.0.5, SPSS, Chicago, IL, USA). To evaluate the changes in genetic distances over time, a matrix of the historical ΦST values was compared with a matrix of the modern values between the same pairs of sampling locations using a Mantel test, with 999 random permutations to determine significance (GenAlEx v.6.1). Modern samples collected over 2 years were available for some locations, in which case only the earlier sample was used.

We also assigned fish to theoretical populations using a Bayesian algorithm that can account for an admixture of fish and partial baselines. Accounting for partial baselines was necessary because our historical samples did not include all Puget Sound sources of transferred fish (Eldridge and Naish, 2007). The Bayesian Analysis of Population Structure (BAPS v.4.14) computer program uses two steps to determine the mixture proportions in unknown samples (Corander and Marttinen, 2006). The algorithm first determines the optimum number of theoretical populations (up to a user-defined KMAX that we set to 20 for 20 runs), which is used to determine the admixture proportions in all samples. Historical baselines were used to ‘train’ the algorithm, and during the admixture run, the minimum sample size was 5, the number of iterations was 100, the number of reference fish from each population was 200 and the number of iterations for reference fish was 20. Output is the proportion of a fish's genome derived from different sources, which we averaged across fish within the same sample to determine the assignment percentage. We used this approach rather than considering admixture proportions (Hansen et al., 2001), because most fish were admixed.

Results

Sample quality and tests for random mating

The success of PCR varied between markers and samples. More than 72% of genotypes in the historical and 92% in the modern samples were determined (estimates based on the average number of genotypes determined per fish in Table 1). No sample or locus was involved in more deviations from Hardy–Weinberg equilibrium than expected (historical samples: χ2=9.3, df=12, P=0.67; modern samples: χ2=17.68, df=16, P=0.34; all loci: χ2=12.00, df=7, P=0.10). The historical samples did not deviate from genotypic equilibrium more than expected (χ2=4.36, df=12, P=0.98), nor did any pair of the eight loci (χ2=14.44, df=27, P=0.98). Deviations from genotypic equilibrium were distributed nonrandomly among the modern samples (χ2=50.71, df=16, P<0.001). Five or more significant tests were observed in each of the Child Creek (SGcc05), Sauk River (SGsa05) and Samish River (SMen05) samples. Disequilibrium in the two Skagit River samples may have resulted from a nonrandom mating in these samples (as suggested by significant FIS, Table 1, Supplementary Table 3).

Results generated using Micro-Checker indicated a significant homozygote excess in some of the historical samples at the same loci wherein deviations from Hardy–Weinberg equilibrium were observed. Failure to amplify large alleles or skipping base pairs during PCR replication (that is, stuttering) is unlikely to be the source of homozygote excess because the excess was distributed across allele classes. Furthermore, null alleles are probably not responsible, as no single locus was systematically affected across populations. To affirm that the problem was not caused by a tendency to incorrectly score a heterozygous fish as a homozygote, we compared the historical and modern samples for the number of loci at which fish were homozygous, and found no significant difference (χ2=4.2, df=6, P=0.65). Alternatively, homozygote excess could be explained by a Wahlund effect. Among the historical samples, fish from nearby creeks or from a trap were pooled and may have represented mixed populations; therefore, we separated the fish from different creeks in the older samples and repeated the Micro-Checker analyses. Only one creek in the 1938 NF Nooksack River continued to show homozygote excess after analyzing creeks separately, although the power to detect deviations among individual creeks was reduced because of small sample sizes (2–17 fish). The larger groupings were maintained for further analyses because they provided a greater sample representation, despite the possibility of the Wahlund effect. Inclusion or exclusion of the older NF Nooksack River sample (NOnf38) did not affect the results; therefore, only results including this sample are presented.

Changes in genetic diversity and population structure

Allelic diversity declined between historical and modern samples. Of the 120 unique allelomorphs observed at the eight loci, 16 allelomorphs appeared only in the historical samples, whereas four appeared only in the modern samples and another one was observed only in the intermediate sample. These allelomorphs were rare, occurring at frequencies below 0.10. Similarly, allelic richness declined over time for seven loci, and was unchanged for the eighth when all rivers were pooled, a statistically significant change (Table 2). However, in comparisons involving single rivers, allelic richness was significantly lower only in the Skagit River, possibly reflecting the smaller sample sizes when rivers were considered on their own.

Table 2 Temporal comparisons of allelic richness at eight microsatellite loci for coho salmon from five rivers around Puget Sound, Washington
Full size table

Genotype distances among the historical and modern samples tended to be small (ΦST<0.04 historically and <0.10 currently, Figure 4 and Supplementary Table 4), and a significant genotype differentiation occurred primarily between pairs of modern samples (P<0.05; Supplementary Table 4). The historical samples often consisted of few fish, which limited the power to detect significant genotype differentiation between samples. Only the modern sample from the 2001 NF Nooksack River (NOnf01) was highly diverged from most other samples (ΦST>0.06), a result observed previously for this same sample location and year (Small et al., 2004; Eldridge and Naish, 2007). However, a second sample from the same location collected in 2005 (NOnf05) was less diverged from other modern Nooksack and Skagit River samples (ΦST<0.06), although this sample was smaller (N=17). Bias in the analyses of modern samples might have been caused by the outlier modern NF Nooksack River sample (NOnf01); therefore, all further statistical analyses were conducted with and without this sample.

The Mantel tests indicated that there was a significant positive relationship between genetic distance and geographic distance between rivers historically (Figure 2, filled diamonds and solid line; r=0.595, Mantel test P=0.007). Recently, the isolation-by-distance pattern was still found to be positive, but the relationship was no longer significant (Figure 2, empty circles and dashed line; r=0.322, Mantel test P=0.127) and on an average, distances were larger than they were historically. Removal of the modern NF Nooksack River sample (NOnf01) did not qualitatively affect these results.

Figure 2
figure 2

Geographic distance vs genetic divergence among pairs of rivers (ΦRT) historically (1938, filled diamonds) or recently (2001–2005, empty circles) for coho salmon in six rivers around Puget Sound, Washington. The lines are regression historically (solid line) or recently (dashed line).

Full size image

In the hierarchical AMOVA, over 95% of genotype variation was between fish within samples, but the variance in genotype frequencies increased within and among rivers over time (Table 3). The structure was not significant between the historical and modern groups (ΦRT=0.001), but the structure among rivers was significant both historically (ΦRT=0.008) and during contemporary periods (ΦRT=0.014). Similarly, the differentiation between samples within rivers was significant in the historical (ΦSR=0.015) and modern samples (ΦSR=0.039). The increased divergence within and among rivers recently was largely due to a single population that has become highly diverged over time. The 2001 NF Nooksack River sample was previously identified as an outlier, and here, it accounted for a large proportion of the within-river variation among modern samples; removing this single sample from the combined samples across both time periods reduced the variation among sites within rivers (ΦSR=0.040 vs ΦSR=0.031), which was also observed among the modern samples (ΦSR=0.039 vs ΦSR=0.028).

Table 3 Hierarchical analysis of genotype frequency variation among groups (ΦRT), among sample sites within groups (ΦSR), and among all sample sites (ΦST) for coho salmon around Puget Sound, Washington
Full size table

A multidimensional scaling plot of the genotype distances revealed that the historical and modern samples overlapped in their distributions, except for the previously mentioned 2001 NF Nooksack River (Figure 3). The pairwise genotype distances between locations changed over time, such that there was no relationship between historical and modern comparisons between the same locations (Figure 4; r=−0.099, Mantel test P=0.331). One notable change was that the NF Nooksack River sample was consistently more diverged from other locations recently than it was historically (white diamonds, Figure 4).

Figure 3
figure 3

The first two dimensions of a multidimensional scaling plot based on pairwise ΦST, an analog of FST based on genotype diversity, of coho salmon around Puget Sound, Washington sampled in 1938 (gray triangles) or from 1968–2005 (black dots) and calculated from allele frequencies at eight microsatellite loci. Not included are the GRdm38, NOmf38 and NOmf01 samples, which were of small size (fewer than 15 fish).

Full size image
Figure 4
figure 4

Historical (1938) vs modern (2001–2005) genotype divergence among pairs of locations (ΦST, an analog of FST based on genotype diversity) for coho salmon from six rivers around Puget Sound, Washington (filled symbols). Empty symbols indicate pairwise divergences with the outlier NF Nooksack population. The line is the regression among all samples.

Full size image

The Bayesian analysis of individual admixture indicated that there were most likely 11 theoretical populations among the historical and current Puget Sound coho salmon samples (Figure 5). The theoretical populations were not equivalent to any real population; therefore, we interpreted assignments to the same theoretical population as indicating a shared common ancestry, that is, alleles identical by descent. Each historical and modern sample partially assigned to all 11 theoretical populations, and rarely did the assignment percentages exceed 50%; therefore, only the theoretical population to which the largest percentage of the genome assigned and the percent assignment are shown for each sample (Figure 5). The predominant assignment for the historic samples was different for each sample, but the modern samples were predominantly one of five different origins, and most current populations cluster on the basis of a history of mutual exchange. The Skagit River was the most common source for fish introduced into the Nooksack and Stillaguamish Rivers in the north (Eldridge and Naish, 2007), and the 2001 SF Nooksack sample, both Stillaguamish samples and three of the Skagit samples were predominantly one of two groups observed in two historic Skagit samples (either SGjk38 or SGdi38). The Green River was the most common source for fish introduced into the Puyallup River in the south (Eldridge and Naish, 2007), and the four Green and Puyallup River samples were predominantly of the group that dominated in the Green River historically. The two NF Nooksack samples may be of native ancestry because they were predominantly of the same group that was dominant in the historic SF Nooksack. A grouping that is not easily explained by exchange history included the MF Nooksack sample, one SF Nooksack sample, the Samish sample, and the Skagit Sauk sample and the historical NF Stillaguamish sample, This anomalous grouping may reflect introgression after transfers from a river, in addition to the six rivers considered (Eldridge and Naish, 2007).

Figure 5
figure 5

Theoretical population to which the largest percentage of the genome was assigned (indicated by pattern or shading of bar) and the percent assignment for historical and modern samples of Puget Sound coho salmon. A Bayesian assignment method (BAPS v.4.14) ‘trained’ using the samples from 1938 was used to create the 11 theoretical populations (each indicated by a unique pattern or shading of bar) and to perform assignments. Sample labels are as in Table 1. Each of the samples from 1938 predominantly assigned to a different theoretical population, but the 2001–2005 samples predominantly assigned to one of five theoretical populations.

Full size image

Discussion

Our results revealed changes in genotype diversity and fine-scale population structure among coho salmon within and between neighboring rivers around Puget Sound, Washington since 1938. The heirachical AMOVA, multidimensional scaling plot of genetic distances (Figure 3), assignment tests (Figure 5) as well as an earlier study (Eldridge and Naish, 2007), all indicated that the current populations of coho salmon in the Puget Sound ESU are descended from historical populations within the ESU. Historically, there was a pattern of reproductive isolation by geographic distance that remains but is no longer statistically significant (Figure 2), and we found no relationship of genetic distances in the historical and modern comparisons between populations sampled at the same locations (Figure 4). Rather, there currently exists a significant relationship between genetic distance and transfer history (Eldridge and Naish, 2007). In addition, populations in different rivers share more of their ancestry recently than they did historically, and generally cluster on the basis of a history of mutual exchange (Figures 3 and 5). Coho salmon lost allelic diversity as well, suggesting that genetic drift increased due to a severe bottleneck in many populations. The hierarchical AMOVA also indicated that populations in different rivers are more diverged now than they were historically, as are different populations in the same river, which could result from an increased genetic drift within populations. There are two hatchery practices that may account for these changes in genetic diversity: transfers between hatcheries, which could alter patterns of gene flow and pooling the milt of multiple parents, which could reduce genetic diversity (that is, the Ryman–Laikre effect Ryman and Laikre, 1991). Similar changes in population structure have been observed in anadromous fish species subject to migration from genetically diverged populations after hatchery transfers (Nielsen et al., 1997; Hansen, 2002). At the same time, finding what seems to be a remnant population in the NF Nooksack River is additional evidence that habitat partitioning may allow some native populations to resist swamping by large releases of nonnative fish and heavy exploitation (Nielsen et al., 2001; Hansen, 2002).

Null alleles or allelic dropout due to poor DNA quantity and quality is a concern with historical samples (Nielsen et al., 1999). In addition, excess homozygote classes due to pooling diverged populations (a Wahlund effect) might have affected the interpretation of the results. However, we used tests that do not rely on an assumption of the Hardy–Weinberg equilibrium, namely, AMOVA based on genotype frequencies, allelic richness tests and genotype frequencies (Φ-statistics) as opposed to allele frequencies (F-statistics). Repeating the analyses with and without the poor quality 1938 NF Nooksack River sample did not detectably alter the results of the AMOVA and isolation by distance tests, suggesting that excess homozygote classes in the archived samples did not substantially bias the results. In any case, it is unlikely that null alleles or allelic dropout caused systematic errors in allele or genotype frequencies, leading to the results reported here.

The finding of isolation by distance in the historical but not the contemporary samples was likely not an artifact of outlier comparisons. There were seven ΦRT values equal to zero among the historical samples (Figure 2). In total, four of these pairwise comparisons involved the Stillaguamish River, which is in the geographic middle of the rivers comprising the ESU, and therefore may have naturally exchanged a number of fish with rivers to the north and south (Figure 1). There was also a large genetic distance between the following two geographically close modern samples: the Stillaguamish and Samish rivers (Figure 2). Removing all pairwise comparisons associated with the modern Stillaguamish or Samish rivers did not change the results of the Mantel test for isolation by distance (results not shown), indicating that these samples had little effect on the conclusions. In addition, a simple comparison of the pairwise genetic distances between historical and modern samples (Figure 4) showed that population structure had likely changed over time.

Previously, it was reported that rivers that exchanged a large number of fish were the most similar, and it was hypothesized that the release of millions of coho salmon may have contributed to the homogenization of populations within rivers (Eldridge and Naish, 2007). However, coho salmon populations within rivers are more diverged recently (ΦSR Table 3). There are two possible explanations for this outcome. First, the number of modern populations in the previous study was greater than that in this study and compared hatchery populations directly with wild populations. This study only included samples that were collected at the same locations as those of the historical populations, and may thus have a reduced resolution at this scale. Second, it seems that some populations were resistant to transfers and releases of cultured fish. The NF Nooksack River population became increasingly diverged from other populations, suggesting that cultured fish released into the wild may affect only some populations within a river. Previously, Small et al. (2004) postulated that behavioral and environmental forces might favor the reproductive isolation of certain wild coho salmon populations that do not experience direct hatchery activities, despite large numbers of cultured releases within 12 river kilometers. It is possible that culture practices can increase gene flow between donor and recipient populations, but nearby populations in the same river may remain isolated if they are not targeted by intentional cultured releases.

A fundamental question remains: Without the hatcheries, would historical and current populations of coho salmon from the same location be more closely related and would the current isolation-by-distance pattern be significant? A temporal stability of population structure has been observed in a number of taxa including marine, anadromous and freshwater fishes, (Nielsen et al., 1997; Tessier et al., 1997; Hansen et al., 2002; Bernal-Ramirez et al., 2003) and amphibians (Hoffman et al., 2004), often despite severe population bottlenecks or migration due to transferring genetically distinct populations. But stability of local population structure is by no means universal (Ruzzante et al., 2001), and depends on the spatial and temporal scales considered (Whitlock, 2004). It seems reasonable to assume that the population structure of long-standing populations of coho salmon should be stable over 60 years or 20 generations, in part because it takes many generations of reproductive isolation for the population differences in allele and genotype frequencies to evolve (Allendorf and Luikart, 2007). Hence, it is important to determine the natural factors and human actions that can disrupt population structure. Extinction and recolonization within a metapopulation is one natural factor (Whitlock, 2004), and differences may be observed between nuclear and mitochondrial DNA if there are sexual differences in philopatry (Bernal-Ramirez et al., 2003), but it is not clear whether bottlenecks or high levels of natural gene flow also lead to an instability of population structure (Ruzzante et al., 2001; Hoffman et al., 2004).

We believe that artificial culture practices including transfers and supportive breeding are the most likely mechanisms by which the pattern of gene flow and levels of genotypic diversity among Puget Sound coho salmon might be altered. Clearly, human actions such as hatchery supplementation and transfers can have dramatic effects on the structure of wild populations (Nielsen et al., 2001; Hansen, 2002), but again, not always (Ford et al., 2004; Ruzzante et al., 2004). Studies that have detected adverse effects after transfers usually have been on systems in which many fish were transferred and released (hundreds of thousands to millions), in which transfers occurred for a long time (>10 years) or involved donor populations that were closely related to the recipient population (this study, Nielsen et al., 2001; Hansen, 2002). Systems in which negative effects after transfers were restricted to a small area or were absent usually involved fewer fish (tens of thousands), were over a shorter duration (<10 years) or involved source populations that were highly diverged from native populations (Ford et al., 2004; Ruzzante et al., 2004).

Harvest and habitat changes in the form of urbanization and logging have been extensive in the Puget Sound area (Pess et al., 2002), and population bottlenecks can affect genetic diversity (Hauser et al., 2002). Harvest and habitat changes may have contributed to population size decreases and reduced gene flow, and therefore reduced genotypic diversity within populations and increased the genetic distance between populations. However, neither harvest nor habitat changes seem solely sufficient to explain the changed population structure observed among rivers. We expect that fish dislocated from their natal sites by habitat changes would most likely seek out a suitable habitat nearby, and thus would likely blur but not disrupt an isolation by a distance pattern of gene flow (Leider, 1989). Moreover, separating harvest effects from the effects of hatcheries may be difficult, high harvest rates would not have been sustainable without the benefit of hatchery supplementation. For example, starting in the 1960s, escapement goals for some rivers were set to the amount needed to meet the broodstock goals for the local hatchery, about 10% of the total number of adults for some Puget Sound rivers (Nehlsen et al., 1991).

The available demographic data do not suggest a substantial decrease in the number of fish reaching maturity over time (Weitkamp et al., 1995); however, there is a reliance on hatcheries for a substantial portion of the production of adult coho salmon (over 50% of all adults in Puget Sound are born in hatcheries (Weitkamp et al., 1995)). We hypothesize that the dominant spawning population shifted from wild to hatchery and that this shift was coupled with poor hatchery practices such as pooling milt of multiple males, which may have led to a reduction of allelic diversity due to a decrease in the effective size within rivers (Ryman and Laikre, 1991; Eldridge and Killebrew, 2007).

Salmon hatchery operations started in Puget Sound in the 1890s, which may have altered the coho salmon population structure before the time the historical samples were collected in 1938. However, our study showed changes in population structure and genetic diversity between 1938 and the present, a finding that does not depend on the knowledge of the unaffected population structure. In addition, it was common hatchery practice to release coho salmon as presmolts (fry and fingerlings) before the 1940s. These releases were generally not successful, and it is unlikely that in the presence of healthy wild coho salmon populations, the hatchery-origin fish would have had much influence on the genetic variation (Kelez, 1937; Flagg et al., 1995).

The actual ancestries of the modern population could not be determined using the Bayesian assignment tests, although this is not surprising. Even in the absence of transfers, it may be very difficult to assign fish when populations are weakly diverged (Latch et al., 2006). In addition, population structure has changed from a relatively clean isolation-by-distance pattern of gene flow to one in which it is influenced by both natural migration and hatchery-mediated gene flow, which can make ancestry more difficult to resolve. Introductions have occurred from outside the Puget Sound coho salmon ESU, but despite these transfers, all contemporary Puget Sound coho salmon populations examined here are closely related to the historical populations we sampled within the ESU. Thus, there is no strong evidence that a highly differentiated exogenous population has substantially introgressed with, or replaced, the native populations, supporting observations in other systems (Utter, 2001).

Here, we showed that fine-scale population structure and gene flow have changed within a coho salmon ESU that experienced the release of hundreds of millions of fish over a long time period. This finding adds to a previous study, in which we showed that the numbers of fish transferred and released affected the coho salmon population structure (Eldridge and Naish, 2007). Fisheries propagation as a tool in restoration may become increasingly important (Leber et al., 2004), and interactions with wild populations may be inevitable (Naish et al., 2007). The evolutionary importance of metapopulation structure and gene flow has long been recognized for enabling adaptation through trial and error (Wright, 1932), for providing novel genetic diversity to counteract the effects of inbreeding by genetic drift (Mills and Allendorf, 1996) or for maintaining the long-term viability of the species in the face of environmental change (Hilborn et al., 2003). Conservation management plans should recognize the potential effects of culture on gene flow and the fine-scale structure of the target species, and recovery approaches should use fish from the target population as much as possible.