Mixed‐stock analysis using Rapture genotyping to evaluate stock‐specific exploitation of a walleye population despite weak genetic structure

Abstract Mixed‐stock analyses using genetic markers have informed fisheries management in cases where strong genetic differentiation occurs among local spawning populations, yet many fisheries are supported by multiple, weakly differentiated stocks. Freshwater fisheries exemplify this problem, with many populations supported by multiple stocks of young evolutionary age and isolated across small spatial scales. Consequently, attempts to conduct genetic mixed‐stock analyses of inland fisheries have often been unsuccessful. Advances in genomic sequencing offer the ability to discriminate among populations with weak population structure, providing the necessary resolution to conduct mixed‐stock assignment among previously indistinguishable stocks. We used genomic data to conduct a mixed‐stock analysis of eastern Lake Erie's commercial and recreational walleye (Sander vitreus) fisheries and estimate the relative harvest of weakly differentiated stocks (pairwise F ST < 0.01). Using RAD‐capture (Rapture), we sequenced and genotyped individuals from western and eastern basin local spawning stocks at 12,081 loci with 95% reassignment accuracy, which was not possible in the past using microsatellite markers. A baseline assessment of 395 walleye from 11 spawning stocks identified three reporting groups and refined previous assessments of gene flow among walleye stocks. Genetic assignment of 1,075 walleye harvested in eastern Lake Erie's recreational and commercial fisheries indicated that western basin stocks constituted the majority of harvest during the peak walleye fishing season (July–September), whereas eastern basin individuals comprised much of the early season harvest (May–June). Clear spatial structure in harvest composition existed; catches in more easterly sites contained more individuals of eastern basin origin than did more westerly sites. Our study provides important stock contribution estimates for Lake Erie fishery management and demonstrates the utility of genomic data to facilitate mixed‐stock analysis in exploited fish populations having weak population structure or limited existing genetic resources.


| INTRODUC TI ON
The sustainability of many populations depends on multiple, but often cryptic, breeding groups that are connected by shared habitat and (or) reproductive behaviors (Alves et al., 2010;Cowen et al., 2007;Hilborn et al., 2003). Such complexity contributes to population stability and resilience (Hilborn et al., 2003;Schindler et al., 2010Schindler et al., , 2015 but also complicates conservation and management by increasing the number of regulatory units . In some cases, populations have been sustainably managed by treating discrete local spawning populations (i.e., stocks) as parts of an overall portfolio of population diversity (Schindler et al., 2015;Waples et al., 2020). Although this idea has primarily been applied to marine species, it has also been suggested for freshwater species . However, fisheries management often presumes that assessment information represents a single stock as opposed to multiple stocks, which can lead to unintended overexploitation of local spawning stocks and inappropriate management (Hutchinson, 2008;Li et al., 2015;Stephenson, 1999). The ability to accurately discriminate and identify unique population components (e.g., local spawning stocks) is integral to the conservation and management of populations that fit the portfolio theory model of ecology and evolution, because it provides a means of accounting for variance in stock-specific productivity (Figge, 2004;Sethi, 2010). However, achieving accurate discrimination among population components can be difficult for species that experience high gene flow at small spatial scales (Martinez et al., 2018) or are of young evolutionary age. Thus, the need exists for methods that can deal with the resultant weak population structure.
For decades, molecular markers have been used to untangle complex migratory behavior (Ruzzante et al., 2006;Seeb & Crane, 1999) and resolve stock-specific contributions to mixed-stock fisheries in marine ecosystems (Bernatchez et al., 2017;Milner et al., 2008;Waples et al., 2020). Mixed-stock assessments have become central to the management of Pacific salmonines (Shaklee et al., 1999) such that the contributions of hundreds of salmon spawning populations are evaluated annually using molecular approaches (Beacham et al., 2020;Dann et al., 2013). Similar practices have become essential to Atlantic Cod Gadus morhua management, helping to improve our understanding of seasonal spawning dynamics (Dean et al., 2019) and limit overharvest of less productive stocks (Dahle et al., 2018;Ruzzante et al., 2000).
The fish populations and the fisheries of large freshwater ecosystems, such as the world's Great Lakes, are also often under high fishing pressure Embke et al., 2020;Fluet-Chouinard et al., 2018) and typically experience more environmental stochasticity than marine ecosystems (Strayer & Dudgeon, 2010) and therefore could benefit from portfolio-based management.
Application of mixed-stock assessments that use similar techniques as large marine fisheries has been beneficial in large lacustrine fisheries (Andvik et al., 2016;Bott et al., 2009;Potvin & Bernatchez, 2001;Tibihika et al., 2020), especially for species with similar lifehistory attributes as marine populations (Ludsin et al., 2014). Such assessments in freshwater ecosystems have been hampered by limited genetic differentiation among spawning stocks (e.g., the North American Great Lakes; Chen, Euclide, et al., 2020;Isermann et al., 2020), highlighting to need for approaches that can detect weak genetic structure.
Fish populations supported by multiple stocks are common in the Laurentian Great Lakes, which have a long history of commercial, recreational, and subsistence fishing (Lynch et al., 2016;Regier & Hartman, 1973). Walleye (Sander vitreus) is one of the most ecologically and economically important species in the Great Lakes (Hatch et al., 1987;Ludsin et al., 2014) and has been the focus of many previous stock discrimination research efforts (Chen, Euclide, et al., 2020;Chen et al., 2017;Johnson et al., 2004;Stepien et al., 2015). Lake Erie's walleye population is the largest of the five Laurentian Great Lakes and is supported by multiple spatially and biologically discrete stocks Stepien et al., 2012;Stepien & Faber, 1998;Zhao et al., 2011). Most of the lake's walleye production occurs in the western basin where tributary and open-lake spawning stocks of varying productivity exist Fraker et al., 2015). Individuals from these stocks move throughout Lake Erie during nonspawning periods in search of preferred habitat fisheries indicated that western basin stocks constituted the majority of harvest during the peak walleye fishing season (July-September), whereas eastern basin individuals comprised much of the early season harvest (May-June). Clear spatial structure in harvest composition existed; catches in more easterly sites contained more individuals of eastern basin origin than did more westerly sites. Our study provides important stock contribution estimates for Lake Erie fishery management and demonstrates the utility of genomic data to facilitate mixed-stock analysis in exploited fish populations having weak population structure or limited existing genetic resources.

K E Y W O R D S
genetic stock identification, fisheries management, Great Lakes, Portfolio theory, RADcapture, RAD-seq, stock discrimination (Kershner et al., 1999;Raby et al., 2018), where they intermix with walleye from smaller spawning stocks in the central and eastern basins (Matley et al., 2020;Vandergoot & Brenden, 2014;Zhao et al., 2011). Migration of individuals from western basin stocks into Lake Erie's eastern basin is predicted to have a disproportionate influence on local commercial and recreational fisheries because of presumed differences in population productivity and abundance (Zhao et al., 2011). However, the exact degree to which individuals from western versus eastern spawning stocks are harvested by recreational and commercial fisheries in a year or season is unclear. Further, efforts to discriminate among western and eastern basin stocks to facilitate mixed-stock assessments using biological markers have been largely unsuccessful (Chen et al., 2017;Hedges, 2002;Johnson et al., 2004;Riley & Carline, 1982). These needs, in turn, limit management options in Lake Erie (MDF, JMR, TMM).
A recent study using thousands of single nucleotide polymorphisms (SNPs) genotyped with restriction sited associated DNA sequencing (RAD-seq) showed high-accuracy reassignment of walleye to the basin of origin, suggesting potential use high-throughput sequencing for mixed-stock assessments (Chen, Euclide, et al., 2020).
Building on this study, we developed a Rapture panel containing thousands of genetic markers to conduct a mixed-stock analysis of walleye harvest in Lake Erie's eastern basin. Specifically, our objectives were to 1) quantify the relative contributions of western and eastern basin walleye spawning populations to commercial and recreational harvest within the eastern basin, and 2) determine whether these contributions varied spatiotemporally. We hypothesized that individuals spawned in the western basin would comprise the majority of harvest for both types of fisheries, but that harvest composition would vary spatially and temporally owing to annual seasonal migrations (Kershner et al., 1999;Raby et al., 2018).
In testing this hypothesis, we ultimately demonstrate the ability of genomic data to resolve genetic stock structure and facilitate mixedstock analysis in ecosystems with weak population structure, to the benefit of fisheries management.

| Study system and sample collection
Lake Erie has a surface area of ~26,000 km 2 and is composed of three basins. The western basin is shallowest with most areas 3-7 m in depth. The central basin is deeper (15-18 m deep), and the eastern basin is the deepest (15-25 m) with the deepest point in the lake being 65 m (Holcombe et al., 2003). Four large walleye stocks exist in the western basin. The largest stock spawns on open-lake reefs, whereas the others spawn in three large tributaries (Maumee, Sandusky, and Detroit rivers;DuFour et al., 2015;Fraker et al., 2015; Figure 1).
Spawning occurs in the central basin on nearshore reefs and in the F I G U R E 1 Map of collection locations of walleye (N = 397) within Lake Erie that were used to develop the baseline genotypes for Lake Erie spawning stocks. The background color of each spawning location indicates the final reporting group into which that local spawning stock was grouped based on assignments generated from discriminant analysis of principal components (DAPC). This analysis used Bayesian Information Criterion to group walleye taken from these known spawning locations during 2012-2018 spawning seasons (Table 1) into the four clusters. Pie charts display the posterior probability of each DAPC cluster proportion averaged across all individuals collected at that spawning location (color scheme is identical to that used for spawning locations). Lake basins (western, central, and eastern) are delineated by dashed black lines Grand River, Ohio (Stepien et al., 2018). Although the contributions of their resulting offspring to the lake-wide fishery remains unknown, we suspect them to be less than the western basin stocks because their spawning populations are smaller. More spawning stocks exist in the eastern basin, where spawning aggregations have been reported in small New York tributaries and nearshore reefs along the southeastern shore, as well as in Ontario's Grand River on the northern shore of the eastern basin (Zhao et al., 2011;  Given that spawning-site fidelity for Lake Erie walleye is high (70%-98%; Hayden et al., 2018) and all individuals used for initial panel development (N = 96) and development of our baseline classification functions (N = 397) were (1) collected at known spawning locations that are spatially segregated from nonspawning and (2) determined to be mature and in spawning condition (i.e., males were ripe and females were gravid but hard or "spent"), we are confident that all individuals of known origin were spawning at or had recently spawned at that their collection locations (Table 1; Figure 1). Individuals of unknown origin (N = 1274) were sampled during the spring, summer, and fall of the 2016-2018 fishing seasons ( Table 2).
The total length (nearest 1 mm) of each individual, as well as its harvest location (nearest creel grid location), was collected by agency personnel (e.g., creel agents) at boat ramps and docks (Table S1). To maximize our ability to quantify seasonal variation in stock-specific harvest in eastern Lake Erie's fisheries, the majority (N = 1021, 80%) of the unknown origin walleye used in our mixed-stock analyses were from the 2017 harvest, with samples chosen in proportion to their proportional harvest during May through December (Table 2). We also selected 130 samples from the July 2016 commercial harvest and 122 from the July 2018 recreational harvest to investigate interannual variation during the season of peak harvest (Table 2).

| Rapture panel development
We developed a Rapture bait panel (Ali et al., 2016) to reduce cost and increase the power and consistency of genetic stock TA B L E 1 Walleye samples collected and genotyped for Rapture panel development and identification of reporting groups in mixed-stock analysis of the recreational and commercial harvests in the eastern Lake Erie during 2016-2018

Spawning stock
Year ( Note: Columns show total sample size (N) and genetic diversity estimates including observed (H O ) and expected (H E ) heterozygosity, allelic richness (AR), and inbreeding coefficient (F IS ) for final Rapture panel single nucleotide polymorphism (SNP) genotypes. These SNPs were identified during initial RAD-sequencing for panel development. Afterward, microhaplotypes containing these SNPs were genotyped using a Rapture panel developed from the 395 baseline individuals and used to identify reporting groups (note the higher AR of microhaplotype loci than SNP loci in Baseline individuals). Panel Development samples generated the initial RAD-sequencing data that were subsequently used to develop the Rapture panel. This panel, in turn, was used to describe the Baseline population structure in Lake Erie and assign individuals of unknown origin.
identification. Loci containing SNPs and microhaplotypes (short DNA fragments containing multiple polymorphic SNPs) were identified by conducting RAD-sequencing on 96 walleye collected at spawning sites across the lake (Table 1) Sequence data for the remaining 14,418 loci that met quality standards were sent to Arbor Biosciences for bait development (custom oligonucleotides that help to isolate desired genomic regions).
Arbor Biosciences conducted additional quality filters, including a blast alignment to the yellow perch (Perca flavescens) genome (Feron et al., 2019) that removed an additional 2337 loci, and then synthesized two baits per-locus to create a final panel consisting of TA B L E 2 Number of walleye of unknown origin collected in eastern Lake Erie's commercial and recreational fisheries that were genotyped (N total), successfully assigned (classified) to one of three natal areas (i.e., spawning groups: western/central basin, eastern basin, or Ontario Grand River; N assigned), and assigned to a noneastern basin spawning group (N assigned to west/central basin)   (Jombart, 2008). To avoid model overfitting, the optimal number of principal components necessary to explain among-group variance was identified in ADEGENET using the optim.a.score function (Jombart, 2008). Finally, we used AMOVA and permutation tests of significance implemented in the poppr v.2.8.3 R package (Kamvar et al., 2014) to determine how much variance was explained when sites were grouped by putative reporting groups.

| Rapture sequencing
We tested reassignment accuracy of putative reporting groups in the assignPOP v.1.1.8 R package (Chen et al., 2018). Assignment tests were performed using Monte-Carlo cross-validation in which individuals were randomly resampled as a training set, using remaining individuals as a test (holdout) set to avoid upwardly biased test results (Anderson, 2010;Waples, 2010). Specifically, we chose three proportions (0.5, 0.7, and 0.9) of individuals from each reporting group and used either half (chosen randomly) or all loci as training data (N = 6 total combinations) to perform the assignment test. Each combination of training data and the test dataset were iterated 30 times for a total of 180 assignment tests. This procedure allowed us to evaluate variation in assignment accuracy and how different proportions of training individuals influenced the assignment results.
The assignPOP predictive models were built using a support vector machine (with linear kernel and parameter cost = 1), which has been shown to generate higher assignment accuracies than other models (i.e., LDA, naiveBayes, decision tree, and random forest; Chen et al., 2018) and had the highest accuracy in our preliminary runs. We ran the reassignment test twice: once with every spawn-

| Mixed-stock assignment of harvested walleye
Once all harvested individuals were genotyped, each was assigned to a reporting group (i.e., natal sources identified during baseline analysis) using the microhaplotype genotypes and the SVM parameter in the assign.x function of assignPOP (Chen et al., 2018).
As an additional estimate of mixed-stock proportion in 2017, the total stock mixture was estimated in rubias with identical reporting groups and reference as assignPOP and bootstrapped proportions (100 bootstraps). As confirmation of individual assignments, we also used rubias to estimate the stock mixture of all individuals assigned to each reporting group. Individual assignments were presumed to be correct as long as the genetic mixture of individuals in each reporting group exceeded 90%.

| Stock-specific harvest dynamics
To determine how spawning stock contributions to the eastern basin fisheries varied through space and time, all walleye that could be successfully assigned to one of our reporting groups were summa- latitude of capture, longitude of capture, and month of harvest. All explanatory variables were first run independently against the response variable (proportion of walleye assigned to the west/central basin reporting group), with any variable not found to be significant (ANOVA; α = 0.05) being removed from future models. To test the relationship between fishery type and harvest proportion type was converted to a numeric dummy variable (0 or 1). Once a subset of variables that were all independently significant was identified, models containing all possible combinations of these variables were built, and the best model was chosen based on the approach outlined in Symonds and Moussalli (2011). We did not include interaction terms in these models because we did not have any a priori predictions to justify them. In short, each model was fit in R using the glm function using default settings (Gaussian error structure) and then ranked based on AIC c using the AICcmodavg v.2.3 R package whereby the model with the lowest AIC c was considered the most parsimonious.
Finally, the Evidence Ratio (ER): was calculated in the qpcR v.1.4-1 R package where Δ best equals 0 as a measure of how much more likely the best model was than model n whereby a low evidence ratio indicates that a model is more similar to the best model (Mazerolle, 2020;Spiess, 2018).

| Baseline population structure and identification of reporting groups
Sequencing of 397 walleye collected from known Lake Erie spawning sites during the spawning season produced a total of 1,953,890,992 reads and an average effective per-sample coverage of 124.2 (standard deviation = 44.2). Of the 12,091 baited SNP loci identified, 8482 loci passed our genotyping criterion (present in ≥80% of individuals) and allele count filter (minor allele count >3). Microhaplotypes at each locus had an average of 3.8 alleles, ranging 1 to 21 (Table S2).
Two individuals contained genotypes at fewer than 70% of loci and thus were removed. As a result, the baseline population structure analysis was conducted using 395 individuals genotyped at 92% of retained loci.
Three to four genetically similar clusters of spawning stocks were identified with DAPC ( Figure 2) and Pairwise F ST (Table 2) Ontario's Grand River, individuals respectively while the remaining 16 individuals from Ontario's Grand River were genetically similar to west/central basin individuals and assigned to cluster 1. A similar pattern was apparent with pairwise F ST (Table 3)  Reassignments of individuals to their spawning stock were low with assignPOP (<75%; Figure S1). However, when multiple stocks were combined into reporting groups that reflected the three primary clusters identified by DAPC and pairwise F ST (western basin stocks, eastern basin stocks minus the Ontario Grand River, and the Ontario Grand River) reassignment accuracy was much higher (93%; Figure 3). The mean reassignment accuracy was near perfect for both the eastern basin and west/central basin reporting groups (96% and 99%, respectively) and slightly lower (85%) for the Ontario's Grand River reporting group. Reassignment accuracy was similar for all sets of training individuals, regardless of whether 50% or 100% of the loci were used (Figure 3). Furthermore, assessment of our baseline samples with rubias found similar reporting group accuracy as assignPOP ( Figure S2), and a significant amount of variance was explained when spawning stocks were classified into these three reporting groups (AMOVA p = 0.01; variance between reporting groups = 1.9%, variance between populations within reporting group = 0.2%). These results gave us confidence that the west/ central basin, east basin, and Ontario Grand River reporting groups could be used for mixed-stock analyses.

| Assignment of mixed-stock individuals of unknown origin
Rapture sequencing of the walleye of unknown origin harvested in eastern Lake Erie's recreational and commercial fisheries produced a  (3) the Grand River, Ontario. Reassignment accuracy was determined using either 0.5 or 1 proportion of training loci (gray and white bars, respectively) and a support-vector machine algorithm Chen et al., 2018), with training samples for each grouping consisting of 0.5, 0.7, or 0.9 proportion of the collected individuals (chosen randomly). The remainder of individuals (0.5, 0.3, or 0.1) was used as the test (holdout) dataset to determine reassignment accuracy. Box plots portray medians (thick black line), interquartile ranges (ends of boxes), and outliers (black dots) temporally and spatially ( Figure 5). During the spring, the proportion of walleye originating in the west/central basin was low (<0.10) with nearly all fish originating from eastern basin (non-Grand River, ON) spawning stocks (Figures 4b and 5).  (Figure 4c).

F I G U R E 4
Proportion of walleye harvested in eastern Lake Erie's recreational and commercial fisheries during May through December 2017 that were assigned to the west/central basin reporting group. Proportions for both fishery types are presented for all harvested fish across all months (a), as well as for each month (b) and each harvest location (longitude; c). Each colored point represents the proportion of fish assigned to the west/central basin from a particular sampling event (harvest grid × date combination; N = 48). The blue line and gray background in panel C represent the least-squares regression line from a generalized linear model and its 95% confidence interval. Note that higher sample sizes existed for core months of harvest, June-September, and thus only sampling events that contained >6 assigned individuals were included (see Table 2) Based on examination of harvested walleye collected during July 2016-2018, contributions from the west/central basin also varied among years ( Table 2). The percentage of west/central basin individuals harvested in the July 2016 recreational fishery was 20%, which was smaller than the percentage of west/central basin origin fish harvested in the recreational (54%) or commercial (38%) fisheries during 2017. However, during July 2018 the percentage of walleye of west/central basin origin in the commercial harvest was 90%, which was higher than either the commercial or recreational harvest during July in the previous two years (Table 2).

| Predicting contributions from the west/ central basins
To better predict proportional contributions of western/central basin spawning stocks to eastern Lake Erie's recreational and commercial fisheries, we sought to use AICc and an all-subsets approach to evaluate a set of general linear models. Two variables (fishery type and mean total length) were not significantly related to the proportion of west/central basin fish (ANOVA: both r ≤ 0.1, both p ≥ 0.05), we removed them prior to AIC c analysis. All other variables F I G U R E 5 Monthly assignment of walleye harvested in eastern Lake Erie's commercial and recreational fisheries during May through December 2017 to either the west/central basin or eastern basin reporting group. The size of each pie chart in each harvest zone corresponds to log 10 normalized mean number of fish harvested in that zone, with the pie chart being located in the harvest centroid of each rectangular harvest zone. Only two colors are shown because no harvested individuals were assigned to the Grand River (Ontario) reporting group. See Table 2

| D ISCUSS I ON
Our study used a sequencing-based genetic Rapture panel to reveal weak, yet sufficient, population structure and to reliably quantify the relative contributions of distinct spawning stocks to the harvest in a mixed-stock fishery. Specifically, using microhaplotypes taken from mature walleye collected at 11 Lake Erie spawning locations during the spawning season, we developed classification functions that could assign walleye of unknown origin to a putative spawning region with a high degree of accuracy. Our findings supported our expectation that walleye from the western basin contribute substantially to the eastern basin fisheries for much of the year. Even so, we found seasonal and spatial variation in the proportion of walleye originating from the west/central basin reporting group. This result indicates that the smaller eastern basin stocks (except for the Grand River, Ontario) comprise a large portion of harvest during certain times (e.g., during the spring) and in certain areas of the eastern basin (e.g., locations east of 79.5°W). In addition to providing critical information on stock-specific harvest of walleye in the eastern basin of Lake Erie that can benefit fishery management, our study represents one of the first uses of Rapture data for mixed-stock assignment in fisheries (but see Carrier et al., 2020) and highlights the value of genomic approaches for management and conservation applications involving species with weak population structure.

| Population structure and reassignment accuracy
Although our Rapture panel could not reliably discern individual spawning stocks within each basin, we were able to use it to discriminate among three reporting groups (i.e., west/central basin, east basin, and Ontario Grand River) with high reassignment accuracy (93%). Previous attempts at differentiating Lake Erie stocks of walleye using microsatellite loci Stepien et al., 2012) or mitochondrial DNA (mtDNA) haplotypes (Gatt et al., 2004;Haponski et al., 2014;Stepien & Faber, 1998) similarly identified genetic differences between basins, but also could not consistently or reliably assign individuals to independent stocks (20%-87% assignment accuracy).
Our success in developing reliable assignment functions emanated from three sources. First, in contrast to earlier studies (Stepien et al., 2012;Strange & Stepien, 2007), we aggregated spawning stocks with low genetic differentiation and included more stocks into our baseline Chen, Euclide, et al., 2020).
Doing so helped us to identify a robust set of reporting groups, albeit at a coarser spatial scale. Second, we used high-throughput sequencing to genotype thousands of microhaplotype loci, instead of using a microsatellite panel or mtDNA, which increased our statistical power and identified useful genetic variation among stocks that could be used to assign harvested individuals of unknown origin to a natal source (i.e., area of origin). Third, extensive preliminary tests of assignment accuracy in both assignPOP and rubias provided us with high confidence in individual assignments. These individual assignments of walleye facilitated the detection of fine-scale patterns Note: The number of estimated parameters for each model (K), Akaike's information criterion for small sample sizes (AIC c ), change in information criterion between sequential models (∆AIC c ), AIC c weight (AIC c Wt), log likelihood (LL), and evidence ratio (ER) is provided. The most parsimonious model is in bold-face text.
TA B L E 4 Statistics from comparisons of general linear models used to describe the proportion of western basin walleye (PropWB) harvested in eastern Lake Erie's recreational and commercial fisheries during May through December 2017 as a function of longitude of each harvest grid (lon), latitude of each harvest grid (lat), and the month of harvest (month) in stock-specific harvest, which will have direct implications for the sustainable management of this ecologically and economically important population (Hatch et al., 1987;Kayle et al., 2015).

| Stock contributions to eastern basin fisheries
The contribution of west/central basin stocks to eastern basin fisheries was predicted to be large based on previous estimates suggesting that greater than 90% of the harvest in the eastern basin comes from western basin stocks (Walleye Task Group The magnitude and extent of annual migrations have been hypothesized to be associated with western basin population size (Zhao et al., 2011) water temperature (Raby et al., 2018), and prey availability (Kershner et al., 1999). Therefore, interannual variation in these factors likely drives walleye distribution and harvest dynamics throughout Lake Erie (Dippold, Adams, et al., 2020). In years when eastward migration is either delayed or limited, eastern basin stocks may be exploited to a higher degree (Dippold, Aloysius, et al., 2020). Over longer periods of time, exploitation of individual stocks may also vary with their relative abundance owing to differences in productivity and recruitment success between eastern and western basin stocks. Behavioral differences among stocks might also influence contributions to the mixed fishery. For example, there is evidence that walleye from Ontario's Grand River do not mix evenly with walleye from the west/central and east basin reporting groups (Jackson et al., 2003;Matley et al., 2020), and no fish in our mixedstock analysis were assigned to the Ontario Grand River despite previous evidence of Grand River walleye contributing to the eastern basin harvest (Jackson et al., 2003). Further understanding of walleye migration patterns and their environmental predictors may inform harvest quotas in Lake Erie, as they have in other fisheries (Bradbury et al., 2016;Shaklee et al., 1999;Vähä et al., 2011).
One of the most surprising findings of this study was that eastern basin walleye stocks have a larger influence on eastern basin fisheries than previously thought and likely contribute substantially to harvest from year-to-year. In particular, fisheries in the eastern-most portion of the eastern basin appear to be sup- mixed-stock assessments of walleye were largely seen as unfeasible or too imprecise to merit the investment. Now that a methodology capable of conducting high-accuracy mixed-stock assignment has been established, the likelihood that re-assessments will take place is greatly increased.
Continued climate warming is predicted to interrupt natural patterns of recruitment for many species, including walleye in western Lake Erie (Brander, 2007;Dippold, Aloysius, et al., 2020). The ability to manage populations like investment portfolios, whereby agencies invest in the protection of multiple spawning stocks could promote stability within these ecosystems and the fisheries that they support (e.g., Schindler et al., 2015;Waples et al., 2020).
However, this strategy depends on the ability to distinguish among population components. The development and use of new molecular tools provide opportunities to incorporate portfolio theory into the management in new fish populations that could benefit from this strategy.
Without a doubt, our ability to understand stock-specific production and harvest in Lake Erie will depend on the continued use of genomic methods. The increased diagnostic power of our Rapture panel made it possible to make precise estimates stock structure.
Indeed, when genetic differentiation is low, as is the case in Lake Erie walleye  this study) having a high number of genetic markers becomes more important for high assignment accuracy (Benestan et al., 2015;Larson et al., 2014;Waples & Gaggiotti, 2006). The high assignment accuracy allowed us to go beyond mixture analysis, which is generally considered to be more accurate than individual assignments when genetic structure is low (Manel et al., 2005) and investigate fine-scale variation in harvest proportions. Genotyping-by-sequencing methods, like Rapture, offer a feasible way to obtain high diagnostic power in a nonmodel species and can provide reliable mixed-stock estimates, even when population components (e.g., local spawning populations) are weakly differentiated. Such approaches could be used in other ecosystems to offer conservation and management agencies the ability to quantify the relative contributions of local spawning populations to larger population they support.
With the promise that genomic tools hold, we are optimistic that molecular studies of exploited freshwater populations can begin to be used more consistently to monitor contemporary changes in population structure and microevolution in response to anthropogenic change. Collecting the data necessary for mixed-stock analysis can be problematic for data-poor fisheries that lack established molecular resources such as SNP or microsatellite marker panels.
Our approach does not require prior-knowledge about the genetic background of a population and therefore could provide a means to conduct mixed-stock assessments in most ecosystems (Allan et al., 2005;Irvine et al., 2019). In this way, we are confident that the continued use of genomic approaches, like the one demonstrated herein, can aid efforts to unravel the complexities associated with threatened or exploited populations that are supported by multiple local breeding populations such that they can remain sustainable both now and in the face of future ecosystem change.

CO N FLI C T O F I NTE R E S T
There is no conflict of interest declared in this article.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in Geome at https://geome -db.org/workb ench/proje ctoverv iew?proje ctId=190 and Dryad at https://doi.org/10.5061/ dryad.4b8gt htb2. All raw sequence files are available via the NCBI short read archive under the BioProject ID's: PRJNA699858,