Validation and association of candidate markers for adult migration timing and fitness in Chinook Salmon

Abstract Recent studies have begun to elucidate the genetic basis for phenotypic traits in salmonid species, but many questions remain before these candidate genes can be directly incorporated into conservation management. In Chinook Salmon (Oncorhynchus tshawytscha), a region of major effect for migration timing has been discovered that harbors two adjacent candidate genes (greb1L, rock1), but there has been limited work to examine the association between these genes and migratory phenotypes at the individual, compared to the population, level. To provide a more thorough test of individual phenotypic association within lineages of Chinook Salmon, 33 candidate markers were developed across a 220 Kb region on chromosome 28 previously associated with migration timing. Candidate and neutral markers were genotyped in individuals from representative collections that exhibit phenotypic variation in timing of arrival to spawning grounds from each of three lineages of Chinook Salmon. Association tests confirmed the majority of markers on chromosome 28 were significantly associated with arrival timing and the strongest association was consistently observed for markers within the rock1 gene and the intergenic region between greb1L and rock1. Candidate markers alone explained a wide range of phenotypic variation for Lower Columbia and Interior ocean‐type lineages (29% and 78%, respectively), but less for the Interior stream‐type lineage (5%). Individuals that were heterozygous at markers within or upstream of rock1 had phenotypes that suggested a pattern of dominant inheritance for early arrival across populations. Finally, previously published fitness estimates from the Interior stream‐type lineage enabled tests of association with arrival timing and two candidate markers, which revealed that fish with homozygous mature genotypes had slightly higher fitness than fish with premature genotypes, while heterozygous fish were intermediate. Overall, these results provide additional information for individual‐level genetic variation associated with arrival timing that may assist with conservation management of this species.

are sexually premature, undergoing maturation while holding in freshwater. Late-returning populations mature in the ocean, prior to entry into freshwater (Quinn et al., 2016). Variation in migration timing can be divergent both within and across populations Narum, Hess, & Matala, 2010;Waples, Teel, Myers, & Marshall, 2004). More specifically, three distinct phylogenetic lineages have been identified which exhibit both neutral and adaptive divergence, including the coastal, Interior ocean-type, and Interior Columbia River stream-type lineages (Hecht, Matala, Hess, & Narum, 2015). Both the coastal and Interior ocean-type lineages exhibit variation in arrival timing for spawning as early (spring or summer) or late (fall) phenotypes that correspond to sexually premature and mature individuals entering freshwater, respectively.
In contrast, Chinook Salmon from the Interior stream-type lineage exclusively enter freshwater as sexually premature, but exhibit variation as early versus late in their final ascent to spawning grounds (Hess, Whiteaker, Fryer, & Narum, 2014;Narum et al., 2018;Quinn et al., 2016;Waples et al., 2004). However, it is uncertain whether early and late arrival to spawning grounds is a phenotypic trait that is conserved across lineages of Chinook Salmon despite differences in freshwater entry . Despite these differences in freshwater migration timing and arrival to spawning grounds, previous studies indicate that gene flow occurs between early and late phenotypes within populations O'Malley, Jacobson, Kurth, Dill, & Banks, 2013;Prince et al., 2017). However, recent environmental pressures select against the early phenotype and, overall, fish that migrate early are at a greater risk of extirpation (Kareiva, Marvier, & McClure, 2000;Quinn et al., 2016;Quinones, Holyoak, Johnson, & Moyle, 2014;Thompson et al., 2019). In spite of selection favoring the late phenotype under various conditions, the ecological, cultural, and historical importance of the early arriving phenotype remains vital for this species (Quinn et al., 2016;Swezey & Heizer, 1977). As such, maintenance of the two migration phenotypes is an extremely important component of salmonid conservation but application to management decisions based on the genetic architecture of the trait remains a continuing debate (Oke & Hendry, 2019;Pearse, 2016;Quinn et al., 2016;Schindler et al., 2010;Waples & Lindley, 2018).
In light of the recent insight into the genomic basis of salmonid migration timing (Hess et al., 2016;Micheletti et al., 2018;Narum et al., 2018;Prince et al., 2017;Thompson et al., 2019), competing theories on the evolution of the migration phenotype provide contrasting strategies to management plans (Kardos & Shafer, 2018;Oke & Hendry, 2019;Pearse, 2016;Waples & Lindley, 2018). For example, while the current framework for Chinook Salmon conservation is structured around evidence that migration timing arose through a process of parallel evolution and is therefore likely to arise again, recent studies suggest that migration timing is highly associated with a genomic region of major effect that arose from a rare mutational event (Prince et al., 2017). Further, the strong association of candidate genes greb1L/rock1 with migration timing occurs in the face of apparent gene flow of migratory types within populations , which raises uncertainty regarding individuals that are heterozygous for candidate markers and their respective phenotypic expression of migratory traits, frequency of occurrence, and fitness relative to homozygous individuals. Waples and Lindley (2018) point out that prior to making conservation management decisions for salmonids with respect to migration timing, it is crucial to determine the distribution of migration-related alleles across populations and to gain a broader understanding of fitness differences underlying the migration phenotype.
In this study, we developed several markers that span 220 kb on chromosome 28 within and between candidate genes greb1L and rock1 for adult migration timing in Chinook Salmon to investigate their utility for conservation applications following questions outlined by Waples and Lindley (2018). Previous studies provided only allele frequency differences for SNP variants  or very few SNP markers from this candidate genomic region for individuals (Prince et al., 2017;Thompson et al., 2019). This new panel of 33 markers spanning the genomic region of major effect on chromosome 28 allowed for more thorough testing of individual phenotypic association within and among lineages of Chinook Salmon.
Using individual-level genotypes from these candidate markers, we tested for an association with migration phenotypes across three lineages that demonstrate distinct freshwater migration phenotypes but each exhibits variation for early and late arrival timing for spawning. We then tested whether the percent of phenotypic variation explained by candidate markers and the patterns of inheritance on chromosome 28 differs between each of the three lineages.
Pedigree data from one of the populations enabled association tests between the candidate markers and fitness which was based on previous estimates of reproductive success (Janowitz-Koch et al., 2019).
Hereafter, we refer to alleles at candidate loci as premature or mature following previous studies in this species Prince et al., 2017), but individual phenotypes as early or late arrival at spawning grounds to reflect phenotypic variation in arrival timing within each of the three lineages. It is important to point out that although a strong association has been documented between chromosome 28 and migration time, the precise point within the migration cycle that exhibits the strongest association with the genotype has not been documented. Thus, the phenotypes evaluated do not represent states of sexual maturity at freshwater entry (premature vs. mature) or timing of freshwater entry, but rather early versus late arrival timing for spawning of each lineage of Chinook Salmon.

| Tissue sample collection
Tissue samples were included from three distinct populations representing the three major phylogenetic lineages of Chinook Salmon in North America (coastal, interior ocean-type, and interior streamtype; Figure S1; Hecht et al., 20155;Narum et al., 2010;Waples et al., 2004). Adult migration phenotypes vary across these three lineages that demonstrate distinct patterns of freshwater entry but each exhibits variation for early and late arrival timing for spawning . Data on the precise timing of freshwater entry were not available. Therefore, samples in this study were classified by early versus late arrival for spawning within each of the three populations (i.e., lineages) and represented independent samples from those included in the original association tests  in order to follow best practices for validating association of candidate markers (Wray et al., 2013). Sample sizes for each collection location and phenotype category are provided in Table 1. Samples from the coastal/Lower Columbia (Cowlitz R.) and Interior ocean-type (upper Columbia/Methow R.) lineages were classified by categorical phenotypes of early and late timing of arrival for further analyses.
Both spring-run (early) and fall-run (late) samples were collected in 2015 from the Cowlitz River hatchery broodstock (Washington) to represent the coastal/Lower Columbia lineage (Narum et al., 2010;Waples et al., 2004). Samples chosen to represent the Interior ocean-type lineage were collected in 2015 from fish with early arrival at Wells Hatchery (Washington; summer-run broodstock) and fish with late arrival at Prosser Hatchery (Washington; fall-run broodstock) that are part of a broad ranging population in the upper Columbia River demonstrating high gene flow (Moran et al., 2012;Narum et al., 2010).
Samples representing the Interior Columbia River stream-type phylogenetic lineage were collected across several years (2010-2011 and 2013-2016) from a naturally spawning population at a weir on Johnson Creek, Idaho that exclusively exhibits spring/summer entry into freshwater . Samples collected from the Interior stream-type lineage provided an estimate of timing of arrival that could be treated as either a continuous trait by weir arrival date, or a categorical trait by implementing a cutoff for early versus late arrival at the weir. Finally, previous studies have determined the reproductive success of each individual passing the weir through pedigree analyses (Hess et al., 2012;Janowitz-Koch et al., 2019), which was used for fitness components of the study ( Figure S2).

| SNP genotyping and quality control
Genotyping was completed following protocols in Janowitz-Koch (2019) using the genotyping-in-thousands by sequencing method (GT-seq; Campbell, Harmon, & Narum, 2015). DNA was extracted from fin tissue using a Chelex 100 method (Sigma-Alrich). In addition to putatively neutral SNP markers from Janowitz-Koch et al., (2019) chromosome 28 (Table S1) that are associated with early and late arrival to spawning grounds, further referred to as premature and mature alleles, respectively . These 33 markers were chosen based on highly significant association results for SNPs that spanned a 220 Kb genomic region of significance on chromosome 28 that included candidate genes of greb1L, rock1, and the intergenic region between them that presumably includes the promoter (Table S1; Narum et al., 2018). To ensure adequate quality control, all samples and loci with ≥10% genotyping failure were removed from further analyses.

| Partitioning haplotype blocks
Since candidate markers were developed in physical proximity, we first determined whether multiple SNPs on chromosome 28 were in strong linkage disequilibrium (LD

| Development of a neutral SNP marker set
To generate an unlinked (i.e., not in LD) and selectively neutral set of SNPs that could be used as a covariate for population structure and relatedness in further analyses, we used a previously published panel of 298 SNP markers and a single sex marker from the Chinook sdY region (Janowitz-Koch et al., 2019). We removed both samples and loci with ≥10% genotyping failure. To ensure that markers from this panel were unlinked and selectively neutral, we first removed SNPs with high physical linkage using a sliding-window approach in PLINK v1.90 (Chang et al., 2015;Purcell et al., 2007). More specifically, one SNP from a pair with an R 2 value greater than 0.9 was removed from SNP windows of 50 shifted by five SNPs per iteration. Three candidate SNPs from chromosome 28 were also used as a positive control for the outlier analysis, since we presumed that those SNPs were not selectively neutral. We then used BayeScan V2.1 (Foll & Gaggiotti, 2008) and OutFLANK (Whitlock & Lotterhos, 2015) to determine and remove outlier SNPs. Each population was analyzed independently using program defaults, and we identified loci putatively under divergent selection with FDR <0.05. To generate a consensus list of a neutral set of markers, SNP markers identified as outliers in any of the three populations were removed from further analysis. The R package adegenet (Jombart, 2008) was used to create a principal component analysis (PCA) for candidate chromosome 28 markers and putatively neutral markers to contrast patterns of adaptive versus neutral structure.

| Genome-wide association analyses
We used GWAS to confirm an association between markers on chromosome 28 and early versus late timing of arrival within each population. First, we ran a GWAS using the full panel of 33 SNP markers on chromosome 28. There were zero SNPs with MAF < 0.01, and therefore, all candidate markers were included in association analyses. The GWAS was conducted using the mixed linear model (MLM) function implemented in the GAPIT R package, which allows for the inclusion of fixed and random effects to account for population structure and relatedness, respectively (Lipka et al., 2012;Zhang et al., 2010). For the Lower Columbia and Interior ocean-type populations, arrival date was binary and represented numerically as one or two for early versus late arrival timing, respectively. For the Interior stream-type population, ordinal date of arrival was used as a continuous variable in the models. However, the distance from the ordinal date of arrival to the median arrival day of 216 was used to bin data to allow for comparisons across lineages. To account for variation in arrival timing across years in the Interior stream-type population, the year of return (i.e., arrival) was included as a fixed effect in the model. Genetic sex was also included as a fixed effect for all models. If genetic sex was not available from the sex marker, phenotypic sex was used instead (<5% of individuals). In the program GAPIT, the first three PCs to account for neutral population structure were included as a fixed effect for each association test. The neutral markers were then used to generate a kinship matrix using the VanRaden method (VanRaden, 2008), which was included as a random effect in the models. Parameter estimates from univariate associations between each SNP and arrival timing in R version 3.5.3 were used to calculate the percentage of premature-versus matureassociated alleles for each individual (RCore, 2016).
Due to the fact that our markers were in high LD, we used the program BLINK (Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway) as implemented in the GAPIT R package to test the effect of multiple markers simultaneously (Huang, Liu, Zhou, Summers, & Zhang, 2018). BLINK uses a stepwise mixedmodel regression with forward inclusion and backward elimination to perform model optimization while accounting for SNPs in high LD (Pearson correlation coefficient = 0.999). By using this approach, the proportion of phenotypic variance explained by the most causal SNPs was maximized. If multiple SNPs were in high LD with redundant explanatory power, the first SNP from a block was chosen. All fixed and random effects, including those accounting for population structure and relatedness, remained consistent between MLM and BLINK analyses.
To determine whether there was an association between fitness and candidate markers on chromosome 28, we used a subset of data that were available from Johnson Creek that included reproductive success (i.e., number of returning adult offspring for each spawning adult; n = 1,346; see Janowitz-Koch et al., 2019) as a continuous variable in GAPIT. We first ran an MLM using the full panel of 33 SNP markers on chromosome 28 to test for an association with fitness. Year of return, sex, age, and population structure (based on the neutral marker list) were included as fixed effects in the model, and kinship from neutral markers was included as a random effect. Analyses were then run separately for males and females, and sex was removed as a fixed effect from the models. Finally, we ran a stepwise analysis using the BLINK model described above to account for SNPs in high LD. Because the number of offspring demonstrated a negative binomial distribution (see Figure S2) and GWAS in GAPIT assumes a normal distribution, we ran an independent association for males and females separately to estimate the effects of year, age, arrival time to spawning grounds (analyzed using linear and quadratic second-order orthogonal polynomial contrasts), and any SNPs that significantly predicted arrival timing in BLINK models. By using SNPs that significantly predicted arrival timing, we were able to determine the relationship between chromosome 28 markers, fitness, and arrival timing. Using a negative binomial distribution model and a log link function, we used the stepwise AIC regression procedure to determine the best fit model. To ensure that all models were fit to the same number of observations, individuals with missing genotypes were removed from the analysis (n = 75 out of 1,346; 6%).
The stepwise negative binomial GLMs were run using the glm.
nb and stepAIC function as part of the MASS package in R (Venables & Ripley, 2013). We report both raw and adjusted P-values for all models using the Benjamini-Hochberg FDR-controlling procedure (Benjamini & Hochberg, 1995 (Visscher et al., 2014). Therefore, results from the GREML analysis should be interpreted with caution.

| Data trimming and development of a neutral marker set
After removing samples that genotyped at ≤90% of markers, a total of 2,790 individuals were used for further analyses (  (Table S3).

| Haplotype block estimation
We then used Haploview to visualize potential haplotype blocks.
For both the Lower Columbia and the Interior ocean-type populations, one haplotype block with 32 SNPs was estimated (31 out of 32 SNPs were the same between the two populations; Figure 2a, b;

| Association analyses
Initial association tests in GAPIT using MLM with all 33 candidate markers validated that the majority of markers were significantly associated with arrival timing for all three lineages (FDR-corrected p < .05; Figure 3; Table S5). The strongest association was consistently observed for markers in the intergenic region upstream of both greb1L and rock1, or within the rock1 gene. Markers located in 3' UTR regions or downstream of each gene displayed lower signals of association ( Figure 3; Table S5).
We then used a stepwise approach as implemented in BLINK to determine the most significant SNPs after accounting for SNPs in high LD. For the Lower Columbia population, the marker Ots28_11071377 was significantly associated with arrival timing (FDR-corrected p < .05; Table 2) and for the Interior ocean-type population, the marker Ots28_11143508 was significantly associated with arrival timing. Two markers (Ots28_11078636 and Ots28_11202190) were significantly associated with arrival timing for the Interior stream-type population, one from each haplotype block (FDR-corrected p < .05; Table 2). Across all populations, R 2 values provided evidence that a large amount of phenotypic variation was explained by these significant SNPs in each lineage, with 28.7% for the Lower Columbia population, 77.9% for the Interior oceantype population, and 4.7% for the Interior stream-type population (Table 2)  For the MLM association test in GAPIT using all 33 candidate SNPs combined across sexes, 13 SNPs were significantly associated with fitness (FDR-corrected p < .05; Table S6; Figure 7). When sexes were analyzed separately, there were no SNPs significantly associated with fitness for females (FDR-corrected p > .05; Table S6), but 13 SNPs were significantly associated with fitness for males (FDRcorrected p < .05; Table S6). Overall, the only significant markers were within or upstream of the rock1 gene or upstream of greb1L, but none were in greb1L (Table S6; Figure 7). When using the BLINK stepwise approach to determine the most significant SNPs after accounting for SNPs in high LD, zero SNPs were significant for males, females, or sexes combined (Table S6).
To expand upon the potential effect of various factors on individual fitness in the Interior stream-type population, we used F I G U R E 3 Association of each candidate SNP marker with arrival timing within three lineages of Chinook Salmon from MLM as implemented in GAPIT. The genome position of markers on chromosome 28 (NCBI accession GCA_002831465.1) is depicted in the gene diagrams above the x-axis. Y-axis represents − log 10 (FDR-corrected p-value)  (Table 3). When sexes were combined, we found that arrival time, sex, year, age, Figure 3) and relationship to the average arrival timing phenotypes in Chinook Salmon. Results are presented separately for each locus for each population. Black and white bars represent homozygous premature and mature genotypes, respectively, and gray bars represent the heterozygous genotypes. Y-axis represents early (−1) and late (1) binary phenotypes for the Lower Columbia (a) and Interior ocean-type (b) populations. For the Interior stream-type population (c), y-axis represents the distance from ordinal day 216 (August 4th), the cutoff used for early versus late arrival to spawning grounds. Significant SNPs from BLINK results are represented in bold boxes. Black SNPs represent those within the intergenic region, while red SNPs represent those located on rock1

Ots28_11071377
Ots28_11072994  Table 3; Figure 8). For a model that included males only, we found that arrival time, year, age, and marker Ots28_11202190 significantly predicted fitness (FDR-corrected p < .05; Table 3; Figure S7). However, the significant quadratic term in the model suggests that the relationship with arrival timing and fitness is nonlinear. When testing a female-only model, we found that year was the only significant predictor of fitness (FDR-corrected p < .05;  (Figure 8).  Interior ocean-type lineages. However, the percentage of phenotypic variation for arrival timing that was explained by the top two markers in the Interior stream-type population was much lower, potentially reflecting polygenicity, stronger environmental effects, or differences in trait designations in this population. These results were similar to a previous study that demonstrated a significant but weaker association of migration timing for spawning in the Interior stream-type population relative to the other two lineages .

For Lower Columbia and Interior ocean-type lineages of Chinook
Salmon, early arriving fish demonstrated a much greater proportion of premature alleles across markers than late arriving fish. The proportion of premature alleles rose to nearly 100% in or near the most significant markers, suggesting that the markers that accounted for the most variation in arrival time potentially provide the most accurate representation of allele frequencies for this trait. We observed the same pattern for late arriving fish that had a higher percentage of mature alleles, with almost 100% in or near the top significant markers. For both populations, the percent of premature alleles in F I G U R E 6 Percentage of premature and mature alleles for the top two significant SNPs (a and b) from BLINK results in early and late arriving fish in the Interior stream-type population. Gray and turquoise bars represent premature and mature alleles, respectively, while x-axis represents ordinal date. Density plot of ordinal date is represented by the black line on the secondary y-axis. All return years are combined it is unlikely to prevent extirpation under a case of complete loss of the early migrating phenotype, a finding that has been seen in other studies (Thompson et al., 2019).
For the Interior stream-type population, there were still a relatively high percentage of premature alleles found within late arriving fish and mature alleles found within early arriving fish.
These results suggest that genetic variation for both early and late arrival to spawning grounds may exist within some interior populations. However, it is possible that the cutoff used to define binary early versus late arrival phenotypes in this population does not adequately capture true variation in maturation timing. Unlike the other populations examined that have variable freshwater entry timing, fish from the Interior stream-type lineage only enter freshwater as sexually premature but may exhibit a bimodal pulse in the final ascent to the spawning grounds. Thus, Interior stream-type fish do not experience the same trade-off as the other two populations examined in this study whereby fish maturing in the ocean experience potentially longer access to feeding but a higher predation risk (Quinn et al., 2016). Therefore, the early versus late distinction in the Interior stream-type population is based on arrival behavior (i.e., final ascent to the spawning grounds), which may not be equivalent to premature versus mature freshwater entry patterns. A recent study (Thompson et al., 2019)  Taken together, this suggests that while there is a significant association between candidate markers and the arrival phenotype in the Interior stream-type population, yearly environmental variation could play a role in both phenotypic expression of the trait in the form of plasticity (intra-generational) and in selection for the trait in certain years compared to others (inter-generational). Previous research has shown that ocean and stream conditions directly affect migration timing (Anderson & Beer, 2009;Hodgson, Quinn, Hilborn, Francis, & Rogers, 2006;Jonsson, Jonsson, & Hansen, 2007;Mundy & Evenson, 2011) and can drive selection on the trait (Crozier, Scheuerell, & Zabel, 2011;Kovach, Gharrett, & Tallmon, 2012;Quinn, Hodgson, Flynn, Hilborn, & Rogers, 2007 stream-type population is influenced by annual environmental conditions that could interact with genotypes at greb1L and rock1. Across all three populations, there was evidence that heterozygotes at the most significant candidate markers demonstrated early arrival timing which provides information regarding the mode of inheritance patterns beyond previous studies (Prince et al., 2017;Thompson et al., 2019). Specifically, for SNPs in or near the top significant markers, particularly markers in the rock1 gene, heterozygous individuals were highly skewed toward early arrival for spawning, suggesting dominant inheritance for premature alleles in rock1, and to a lesser extent, the intergenic region between greb1L and rock1. These results suggest that the alleles for early arrival are not masked (recessive) in the heterozygous state and may be dominant in rock1, and thus would be readily lost from populations under natural selection against early arrival (Thompson et al., 2019).
However, in the Interior stream-type population, the relationship between alleles and arrival timing to spawning grounds was dependent on year, providing additional support that balancing selection While balancing selection may maintain phenotypic variation for both early and late migration phenotypes through disruptive selection related to high stream temperatures in the middle of migration , other scenarios for balancing selection may also include sexual selection and negative frequency-dependent selection. In this study, fitness was significantly associated with one marker in particular in males, but not females, that was located within rock1 (Ots28_11202190). Although not measured in this study, these results could be indicative of intralocus sexual conflict whereby distinct optimums in migration timing differentially impact fitness for each sex, maintaining standing genetic variation at loci association with migration time (Cox & Calsbeek, 2009 Using a third analysis that accounted for non-normality of fitness data, we again found that marker Ots28_11202190 significantly predicted fitness for males and combined sexes, but not for females.
Therefore, the GWAS results for fitness should be interpreted with caution and may only apply to this single population, but future studies should expand on the relationship between arrival time and fitness, along with the possibility of intralocus sexual conflict, in other populations such as the Lower Columbia and Interior ocean-types.
As suggested by Waples and Lindley (2018), gaining a broader representation for genetic variation in migration timing across populations, examining patterns of inheritance in rock1/greb1L, and determining phenotypes of heterozygotes are all areas of research that require careful evaluation in individual populations of Chinook Salmon prior to making conservation decisions. The results of this study provide strong support that timing of arrival is driven by genetic variation at adjacent candidate genes greb1L and rock1 on chromosome 28 across populations representing distinct lineages of Chinook Salmon. After accounting for LD, the most significant SNPs were located within rock1 or upstream of rock1 and greb1L, providing more precise information regarding the specific candidate genes underlying phenotypic variation for alternative arrival time phenotypes in Chinook Salmon. We also show that when examining highly significant SNPs predicting arrival time, premature alleles were present at very low frequency in late arriving populations from the Lower Columbia and Interior ocean-type lineages, which provides limited or no potential for recovery of early fish from populations that are nearly fixed for mature alleles (Waples & Lindley, 2018). Furthermore, alleles for early arrival follow patterns suggestive of dominance inheritance in rock1 and therefore readily lost from populations under natural selection against the early phenotype. However, populations from the Interior stream-type lineage may have standing genetic variation that selection could act upon under scenarios of differing environmental conditions, which would be dependent on the amount of gene flow between populations, inheritance pattern of premature alleles, and strength of selection. By using a systematic approach in this study to examine an important phenological trait in salmonids, we broadened our understanding for genetic variation and adaptive potential of Chinook Salmon populations, providing additional information to inform conservation management decisions.

ACK N OWLED G EM ENTS
We would like to thank biologists at Yakama Nation, Washington Department of Fish and Wildlife, and the Nez Perce Tribe for providing samples. We thank David Graves for providing the map of the Columbia River Basin and Stuart Willis for custom R scripts used to generate variance components of models. Laboratory-based work was completed by Janae Cole, Stephanie Harmon, Travis Jacobson, Lori Maxwell, and Rebecca Sanders. We would also like to thank the anonymous reviewers for providing valuable and insightful feedback on our manuscript. Finally, we would like to acknowledge Bonneville Power Administration for funding of this research.

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

DATA AVA I L A B I L I T Y S TAT E M E N T
Phenotype data, genotype data, and custom R scripts are available on the Dryad Digital Repository: https://doi.org/10.5061/dryad.