Concerns about the potential for genetically-based loss of fitness in hatchery salmon date back more than 45 years (e.g., Reisenbichler and McIntyre 1977), and have been a focus of hatchery reform efforts for several decades (Hard et al. 1992; Mobrand et al. 2005). Several studies comparing the reproductive success of hatchery- and natural-origin steelhead when spawning in nature have found large and rapid (single generation) fitness losses that have been inferred to be genetic based on comparing the reproductive success of hatchery-origin fish with zero, one, or two natural-origin parents (Araki et al. 2007; Araki et al. 2008). Although evidence for such rapid, heritable fitness loss has not been found in other propagated Pacific salmon species, it has been commonly assumed that inadvertent domestication selection in hatcheries is a serious risk to natural salmon populations of all species (Mobrand et al. 2005; Araki et al. 2008; Fraser 2008; Anderson et al. 2020).
In this study, we used whole genome sequence data to directly examine the degree of genomic divergence between hatchery and natural Chinook salmon in two different hatchery-supplemented populations over two time periods of potential divergence. The shortest time period was a comparison of fish from within the same cohort but identified as having high reproductive success in either the hatchery brood or in the natural stream. If hatchery propagation leads to strong, rapid selection for genotypes that are deleterious in nature but advantageous in the hatchery, then we might expect to see evidence for this in the form of larger than expected allele frequency differences between fish that were successful as brood compared to those that were successful in the stream. We did not see this, however. In both the Wenatchee River and Catherine Creek populations, the genomic distribution of divergence between the successful stream and successful brood fish was nearly the same as expected if the two sets of samples were drawn at random from the same statistical population (Fig. 3B, D). This suggests that, to the degree that these pressures exist, they are not creating greater than random differences in allele frequencies over the course of a single generation. It is important to note, however, that our power to detect such differences is fairly low, illustrated by the quantiles of the null distribution of FST. The power of increased sample size can be clearly seen by comparing the narrower distribution of permuted FST values for the between population comparison (n = 80) relative to the within population comparisons (n = 40; Table 2). Nonetheless, our results indicate that to the degree that divergent selection in the stream and brood exists, it is not creating large differences in allele frequencies over the course of a single generation.
In contrast to the stream/brood comparisons, hatchery and natural origin fish were notably more divergent than would be expected if the two groups were drawn at random from the same cohort (Fig. 3C, E). This was particularly true in the Wenatchee River population. The hatchery/natural comparisons involve at least one generation of separation, because this comparison includes only the progeny of fish that spawned in different environments. Therefore, even in the absence of any differential selection between the hatchery and stream environments, a small amount of differentiation between hatchery and natural fish is expected due to a generation of drift. Spatially non-random spawning of hatchery and natural fish could also create additional opportunities for reduced geneflow (Williamson et al. 2010; Ford et al. 2015; Hughes and Murdoch 2017). It is therefore not surprising that the observed divergence is somewhat greater than the permuted distribution. Our results do provide an empirical measurement of this divergence, however. For example, even though the hatchery/natural comparisons within each population had mean FST 10-20X lower than the comparison between the two populations, the maximum values and the higher quantiles were similar both within and between populations (Table 2). In other words, the maximum divergence between hatchery and natural fish in each population was similar to the maximum divergence between populations that are considered to be different ESUs.
The excess divergence between hatchery and natural fish in the Wenatchee River was associated primarily with a peak on chromosome 8, near the slc7a2 and pdgfrl genes (Fig. 7, Table S3). The function of neither gene has been studied in salmonids, but in mammals the slc7a2 gene encodes a cell membrane protein involved in cationic amino acid transport (Hoshide et al. 1996) and variation in this gene has been associated with various cancers (Sun et al. 2020; Xia et al. 2021). In zebrafish (Danio rerio) the protein produced by this gene is expressed in macrophages involved in central nervous system health (Demy et al. 2020). The pdgfrl gene has also been identified as a tumor suppressor in humans (Guo et al. 2010) and has been associated with a blood vessel inflammation disease (Hou et al. 2013). The finding by Waters et al. (2018) of a peak on divergence between integrated hatchery and natural Chinook salmon near this region of chromosome 8 supports the possibility that variation in this region may be involved in hatchery adaptation more generally. Analysis of additional samples of hatchery and natural Chinook salmon in the Wenatchee River and elsewhere will be needed, however, to test this hypothesis further.
An additional difference between hatchery and natural origin fish in our study was the estimated effective number of breeders (Nb), which was lower for the hatchery-origin samples in both populations (Table 1). This result is similar to previous observations in the Wenatchee River population (Ford and Williamson 2010), and to estimates of hatchery and natural Nb of Chinook salmon in the nearby Yakima River (Waters et al. 2015). For the Wenatchee River and Catherine Creek populations, the lower Nb in the hatchery is reflective of the generally smaller number of breeders used in the hatchery compared to the number spawning in the streams. For these populations, the hatchery environment is also markedly more productive than the stream environment: in the Wenatchee River population between 1989 and 2014, each brood spawner produced an average of 7.4 returning adults, compared to 1.03 returning adults per stream spawner (Hillman et al. 2021), and in Catherine Creek the values are 6.1 adults/spawner for the brood and 0.4 for the stream (EB, unpublished data). Despite this high productivity of the hatcheries, however, Nb is lower due to the smaller number of spawners.
The observed divergence between the Wenatchee River and Catherine Creek samples provides a useful benchmark for the degree of differentiation between distinct, albeit fairly closely related, Evolutionarily Significant Units (Myers et al. 1998). The distribution of FST values was markedly (and unsurprisingly) greater than the permuted null distribution for all quantiles (Fig. 3A). Despite this, the absolute level of divergence was certainly not large, with a mean FST of only 0.01, a 75% quantile of only 0.02, and no fixed differences between the samples. The only obvious peak of divergence occurred in the GREB1L/ROCK region of chromosome 28, which has been previously associated with run timing variation in both Chinook salmon and steelhead (see Waples et al. 2022 for a recent review). Our results continue to support this association in two ways. First, we found that variation within this region was associated with run timing within the Wenatchee River (Fig. 6). This association is similar to what has been found in other Chinook salmon populations, including both coastal (Prince et al. 2017; Thompson et al. 2020) and Interior Columbia populations (Koch and Narum 2020; Willis et al. 2021). Variation at the GREB1L/ROCK1 region therefore appears to influence run timing in the Wenatchee River Chinook population, even though the population as a whole has an early (spring) run timing distribution. This is similar to what has been previously observed in Johnson Creek Chinook salmon (Narum et al. 2018), another early-run Interior Columbia population. Second, the patterns of variation between the Wenatchee River and Catherine Creek samples are also consistent with an association of run timing at this genomic region. Compared to other Interior spring/summer run Columbia River Chinook salmon populations, Catherine Creek has a particularly compressed run timing distribution when measured at the mouth of the Columbia River (see Fig. 3 in Sorel et al. 2021). The Wenatchee River population, in contrast, has a much broader distribution when measured the same way. These patterns are reflected within the peak of divergence between these two populations, where the Wenatchee River sample was markedly more variable compared to the Catherine Creek sample (Fig. 5, Fig. 6).
To our knowledge, this is the first study to evaluate the degree of divergence between hatchery and natural salmon in supplemented populations at the whole-genome level. The study is largely exploratory and will benefit from replication, so our results should be interpreted cautiously. Larger sample sizes, especially, will be helpful for improving the signal to noise ratio for detecting outlying regions of divergence. Additional temporal sampling will also be helpful. For example, a supplemented population might change genetically over time in a different way from an unsupplemented population, even if there was only modest differentiation between the hatchery and natural components of the population at any moment time. Evaluating whether such temporal changes are due to supplementation per se appears conceptually difficult for any single population study, but perhaps by evaluating large numbers of genomes from multiple supplemented and unsupplemented populations, the question could be addressed. Despite these limitations, our results illustrate the promise of using whole-genome population analyses to address an important and long-standing problem in conservation biology.