2.1 Sample collection

Fin clips and scales were taken from Striped Bass from multiple collections (Table 1). Age of sampled individuals differed by location. YOY juveniles were individuals less than 1 year old (<15 cm long). Saint John juveniles were 1–4 years old and largely spawned in the year 2013. Ages for Saint John River juveniles were obtained from scales. Adults were sexually mature individuals aged 4 years and older. All adults collected were in spawning condition at time of sampling, except for Bras d’Or Lake, Mira River, and Shubenacadie River. Shubenacadie origin Striped Bass migrate to the Stewiacke–Shubenacadie systems from overwintering sites during the sampling period (DFO, 2014; Keyser, Broome, Bradford, Sanderson, & Redden, 2016). Adult bass caught during this period are assumed to be of Shubenacadie River origin for the purpose of population surveys (DFO, 2014). Putative Miramichi River origin Striped Bass were included using fin clips taken from Striped Bass caught in the Bras d’Or Lake, Cape Breton, that have previously been examined using microsatellites and found to match the Miramichi River population (Bentzen, Mcbride, & Paterson, 2014). These samples will hereafter be referred to as Bras d’Or–Miramichi individuals (Box 1).

2.2 Laboratory

DNA was isolated using either NucleoMag^® 96 Tissue (Macherey-Nagel) kit on an epMotion 5075t (Cat. 5075000302), or the E.Z.N.A. Tissue DNA Kit (Omega Bio-Tek). Libraries containing 96 individuals each were prepared using a double-digest restriction-site-associated DNA sequencing (ddRAD-seq or ddRAD) protocol developed by Poland et al. (2012) and modified as described in LeBlanc et al. (2018). Samples were randomized so that each lane contained individuals from multiple locations and sequenced using Illumina^® HiSeq™ 2,500 or Illumina^® HiSeq™ 4000 (San Diego) at Génome Québec Innovation Centre.

2.3 Quality control and analysis

SNPs were demultiplexed and filtered using modified versions of Eric Normandeau's Stacks workflow scripts, available on github (https://github.com/enorman deau/stacks_workflow, downloaded August 2016). Cutadapt v. 1.13 (Martin, 2011) was used to trim adapters from the raw sequences using a maximum error rate (e) of 0.2 and a minimum read length (m) of 50. FastQC v. 0.11.5 (Babraham Bioinformatics) was used to assess sequence quality before and after. Sequences were then trimmed to a uniform length of 85 bp and demultiplexed using the process radtags module of Stacks v. 1.46 (Catchen, Hohenlohe, Bassham, Amores, & Cresko, 2013) using the paired-end option –P. BWA version 0.7.15 (Li & Durbin, 2010) was used to align sequences to the Striped Bass genome (BioProject accession number PRJNA266827) using a minimum seed length (k) of 19, a maximum seed occurrence of 55, and no filtering on output alignment score, and otherwise default parameters. The stacks module pstacks identified reference aligned loci with a minimum depth (m) of 4 using the “snp” model type and an alpha of .1. Loci were assembled into a catalogue using cstacks, sstacks, and rxstacks with default settings, and unclear or unlikely haplotypes, as well as SNPs with a log likelihood <45, were pruned from the dataset. Using the populations module, SNPs were further filtered to remove all loci with a stack depth <5, with >20% missing data in any given location, and any loci not amplified in all locations. We examined the output of populations and removed loci with an Fis < −0.3 to eliminate possible paralogs, and used VCFTools 0.1.13 (Danecek et al., 2011) to remove any loci with a minor allele frequency <0.01, and plink v. 1.90 (Chang et al., 2015) was used to remove loci in linkage disequilibrium with each other.

Structure files created by Stacks were converted to the appropriate input files for downstream analyses using PGDSpider v. 2.1.1.0 (Lischer & Excoffier, 2012). Sibship analyses were carried out in Colony2 v. 2.0.6.5 (Jones & Wang, 2010) on each population separately to ensure individuals were not closely related. Full sibling pairs identified with a probability of >.5 were removed from subsequent analyses. Percent polymorphism of loci in each population was reported by the Stacks populations module, and expected and observed heterozygosity were calculated using the R package adegenet v. 2.1.1 (Jombart, 2008).

An initial pairwise F_ST analysis was conducted in Arlequin v. 3.5.2.2 (Excoffier & Lischer, 2010), with significance assessed using 10,000 random permutation tests. Individuals caught in the Hudson River and Delaware River in 2012 and 2014 were grouped by location and year in order to assess whether the genetic profile of each location differed from year to year. After confirming no significant differences between years, the two sampling years were pooled together for outlier analyses.

2.4 Constructing a neutral SNP panel and assessing adaptive selection

Outlier loci were removed prior to subsequent population genetic analyses, and a subset of outliers were examined separately. Existing outlier analyses are known to detect high numbers of false positives alongside true outlier loci (De Villemereuil, Frichot, Bazin, François, & Gaggiotti, 2014; Lotterhos & Whitlock, 2014), and a common method of controlling for this is to examine which loci are flagged as having non-neutral divergence patterns by more than one analysis software (De Villemereuil et al., 2014). In the absence of out-group genotypes or known neutral loci, we assessed whether any of our SNPs were under balancing or divergent selection using two methods. BAYESCAN v. 2.1 (Foll & Gaggiotti, 2008) was run with 100,000 iterations, using a burn-in of 50,000, a thinning interval of 10, and a sample size of 5 K. Prior odds were set to 1,000 to minimize false positives while retaining power to detect outliers (Lotterhos & Whitlock, 2014). We also used the recently developed R package OutFLANK (Whitlock & Lotterhos, 2015) with Hmin >0.1 to identify an additional set of outliers. Unlike previous outlier tests like BayeScan, outFLANK uses distribution of allele frequencies across all loci to account for differences in genetic structure among populations (Whitlock & Lotterhos, 2015). Loci identified as outliers at a q-value ≤0.05 by either method were removed to create a dataset of putatively neutral loci for genetic structure analyses. Loci identified as outliers by both methods were mapped to one of 35,010 scaffolds contained in the published Striped Bass genome using the JBrowse genome browser (Skinner, Uzilov, Stein, Mungall, & Holmes, 2009) to identify associated genes showing signatures of selection, and allele frequencies were calculated in Arlequin to investigate divergence patterns across populations.

2.5 Connectivity of Striped Bass locations through population genetic structure

Population structure was assessed using both traditional F_ST and clustering algorithms. Overall hierarchal F_ST and pairwise F_ST were calculated in Arlequin, and significance was assessed using 10,000 random permutation tests. Hierarchal population groupings for overall F_ST were made based on patterns of differentiation seen in clustering analyses and previous studies. Pairwise F_ST values were also calculated on all locations, and pairwise significance was assessed using a chi-square test implemented in the R package strataG v. 2.1 (Archer, Adams, & Schneiders, 2017), and corrected to account for multiple tests using the false discovery rate method detailed in Benjamini and Hochberg (1995). Chi-square tests have high power and low false-positive rates when used on large numbers of biallelic loci, as found in SNP datasets (Ryman et al., 2006).

Isolation by distance (IBD) was assessed using mantel tests implemented in Arlequin v. 3.5.2.2 (Excoffier & Lischer, 2010). Isolation by distance was assessed on all locations, on only Canadian locations, only US locations, and on locations within Chesapeake Bay and Delaware River, using approximate distance between rivers. When calculating distances between rivers, we assumed that Striped Bass make use of the Cape Cod Canal, and that Striped Bass move between the Chesapeake Bay and the Delaware River via the Chesapeake and Delaware Canal based upon tagging study results (Gahagan et al., 2015; Kneebone et al., 2014).

The R package LEA v. 2.0 (Frichot & François, 2015) estimates ancestry coefficients for all individuals using sparse non-negative matrix factorization (sNMF), an algorithm that has been optimized for use with large numbers of genetic markers. Scenarios were run using 1–20 theorized number of distinct genetic populations (K), with 10 repetitions per K value, on all individuals as well as on only US individuals. To ensure results were not biased by differences in sampling size, sNMF was run a second time with a maximum of 30 individuals from any given genetic cluster found in the initial run. The probability of a K being valid was calculated using the cross-entropy criterion. The K values with the lowest minimal cross-entropy value were considered most probable as the true number of ancestral populations (Frichot, Mathieu, Trouillon, Bouchard, & François, 2014). Where the lowest entropy was unclear, clustering results for the lowest K values were manually inspected for informative grouping and consistency across repetitions. Population structure was also assessed using Discriminant Analysis of Principal Components (DAPC; Jombart, Devillard, & Balloux, 2010), implemented in the R package adegenet, using 1–20 assumed clusters (K). The number of putative clusters with the lowest Bayesian information criterion value was chosen to evaluate population groupings. Another DAPC analysis was conducted with samples taken from Canadian rivers excluded, using the same methods described above.

2.6 Assessing the power of SNPs and reference pool for population assignment

We tested whether our SNP panel could accurately assign individuals to populations of origin using a leave-one-out protocol implemented in GeneClass2 v. 2.0 (Piry et al., 2004), using the Rannala & Mountain, 1997 Bayesian method (Rannala & Mountain, 1997). Assignment success was compared to results from another genetic assignment algorithm implemented in the R package rubias, again using a leave-one-out protocol. Using this protocol, each individual is assigned to a region using a reference panel composed of all individuals except the one being tested.

We tested assignment success of all sample locations separately, as well as assignment to pooled groups according to previous population groupings used in Gauthier et al. (2013). In both cases, we considered an individual assigned to a population if the confidence score for assignment to that population was 80% or above.

3.1 Filtering

The initial SNP catalogue contained 756,713 loci. After filtering for Ln Likelihood less than −40, the catalogue contained 670,167 loci. After filtering out loci with stack depths of less than five, more than 20% missing data, more than two alleles, and loci present in fewer than 17 populations, the SNP catalogue contained 7,884 loci. After filtering by F_IS values, removing loci with minor allele frequencies <0.01 and loci in linkage disequilibrium with at least one other locus, and removing non-polymorphic loci, we had 1,291 loci. Average read depth across loci for each individual was 55 (range = 9–171), and average read depth for loci across all individuals was 55 (range = 17–131). Sibship analysis found two possible full sibship pairs in the Kennebec River and three in the Chesapeake Bay; one individual from each pair was removed. In addition, the Cape Fear dataset contained three possible full sibship pairs, one trio, and one group of five individuals. One individual from each group was retained, and the rest were excluded from downstream analyses.

Expected and observed heterozygosity levels ranged from 0.25 to 0.38 and observed heterozygosity did not deviate more than 0.02 from expected heterozygosity in any location. Canadian rivers had slightly lower observed heterozygosity (H_O = 0.26–0.33) compared to rivers south of the Bay of Fundy (H_O = 0.35–0.38; Table 2). Similarly, individuals in the Bras d’Or–Miramichi and Shubenacadie River populations had the lowest proportion of polymorphic loci among all locations; almost one quarter of all loci genotyped were fixed in Bras d’Or–Miramichi individuals (Table 2). All other sampled locations had > 95% polymorphic loci.

3.2 Outliers

Outlier analyses identified 35 total outlier loci: BayeScan identified 13 loci as possible outliers, compared to 25 loci found by outFLANK, and 3 loci were identified by both analyses. All 35 potential outliers were excluded from downstream genetic structure analyses, while the three loci identified by both approaches were examined further as putative adaptive loci. These three loci were located on three different scaffolds and were given names according to their scaffold number and base pair position on the scaffold (scaffold_bp). Locus 4437_41108 is located 41,108 base pairs into a large (77,288 bp) scaffold, Msax_4437, inside an intron of insulin-like growth factor 2b (igf2b). The remaining two outliers, 25891_222 and 27535_2519, were located on short (2,825 and 5,316 bp, respectively) scaffolds with no known genes.

Examination of allele frequencies of the three putative outliers revealed that all three loci possessed one allele that was at or near fixation in individuals within Gulf of St. Lawrence and Shubenacadie River and at very low frequencies in US locations (Figure 2), with maximum allele frequency differences of 0.85–0.98. In 25891_222 and 27535_2519, allele frequencies in Saint John River fish were slightly lower than other Canadian locations, while the major Canadian allele of 4437_41108 was present at about a frequency of 0.50 in the Saint John River, Kennebec River, and Hudson River (Figure 2).

3.3 Population structure

Population structure analyses were conducted using 1,256 SNPs deemed to be neutrally evolving after outlier analyses. Overall, F_ST was 0.086 and highly significant (p < .001), while pairwise values varied from 0 to 0.20. Correction for multiple testing did not change significance of any pairwise F_ST values. Canadian populations tended to be highly genetically distinct, while populations in the US migratory range were less genetically differentiated. The highest pairwise F_ST values occurred between Canadian rivers and all other locations (save for F_ST between the Bras d’Or–Miramichi individuals and Mira River), with F_ST values ranging from 0.09 to 0.20. Pairwise F_ST between Mira River and Bras d’Or–Miramichi River was 0.007 and nonsignificant (p-value = .08; Table 3). Among the three US regions identified with genetic clustering, F_ST values ranged from 0.012 to 0.035, while within-region F_ST values were lower (0 to 0.011). Within the Delaware River–Chesapeake Bay region, 20 of 28 comparisons were significant (Table 3). The majority (18) of significant comparisons were between James River individuals and all other locations in this region, as well as between populations in the Nanticoke and Choptank Rivers, the only two rivers sampled along the eastern side of the Chesapeake Bay, and all other locations within the region. The Delaware River, upper Chesapeake Bay, the Potomac River, and the Rappahannock River all had very low and mostly nonsignificant F_ST values with each other. The Kennebec River had very low (0.004 to 0.008) but significant (p < .01) pairwise F_ST compared with Hudson River individuals. Similarly, the Roanoke River and Cape Fear River had small (F_ST = 0.004) but significant differences (p < .001; Table 3).

Isolation-by-distance (IBD) and differentiation patterns differed among Canadian and US locations. Significant isolation by distance was found when all locations were considered (r = .84, p < .001). Within Chesapeake Bay, there was no significant isolation-by-distance pattern (r < .001, p = .50); however, when US locations in North Carolina, Hudson River, and Kennebec River were included, IBD became significant (r = .61, p = .03). When only Canadian populations were considered, there was no significant IBD (r = .49, p = .21). When all samples were run using DAPC, the most likely number of clusters was four, as estimated using the Bayesian information criterion (Figure S1). Canadian Striped Bass formed three groups, and all US Striped Bass were assigned to a fourth group (Figure 3; Table 4). This general pattern was seen when DAPC was run assuming five and six genetic groups (Figure S2). Using LEA, the number of genetic clusters (K) with the lowest entropy across 10 runs was 6 (Figure S1). We visualized clustering patterns for K values 4 through 7 to identify hierarchal clustering patterns as K increases (Figure 4). In all simulations, Canadian Striped Bass clustered into the same three groups as in DAPC. North Carolina rivers separated into their own cluster at K = 5, while Kennebec River and Hudson River separated at K = 6, and at K = 7, the two rivers on the eastern coast of Chesapeake Bay (Nanticoke River and Choptank River) primarily belong to the seventh cluster (Figure 4). The same clustering pattern was seen when US samples were analyzed separately from Canadian samples (Figures S3 and S4). When LEA was run with balanced sampling numbers, the lowest entropy was K = 4 as seen in DAPC analyses. Canadian locations clustered into three regions, while all US Striped Bass were clustered together. Mean assignment per location remained high when K was increased to 6, with the same clustering pattern seen in the full dataset (Figure S3).

Due to the genetic distinctness of Canadian Striped Bass, it was possible to identify both putative migrants and admixed individuals within these populations. Admixture was seen in a single individual in the Mira River and eight individuals in the Saint John River. Admixed individuals within the Saint John River had approximately equal assignment to the Saint John River cluster and either the Chesapeake Bay or Shubenacadie clusters. The single admixed individual in the Mira River had equal assignment to the Bras d’Or–Miramichi cluster and to both the Hudson River and Chesapeake Bay. Additionally, three individuals caught in the Mira River appear to be migrants from Hudson River or Chesapeake Bay. When migrants and admixed individuals were removed from Mira River, both observed and expected heterozygosity for this population became 0.25, and the proportion of polymorphic loci dropped to 75% from 96%. Genetic variation within the Saint John River population also lowered slightly with the removal of admixed individuals: Observed and expected heterozygosity became 0.31, and percent polymorphism became 94% from 98%. Significance of pairwise F_ST values remained the same with and without migrants and admixed individuals. F_ST between Mira River and Bras d’Or–Miramichi Striped Bass became 0.001.

3.4 Assignment

Analyses were performed at two spatial resolutions to determine the geographic scale to which reliably natal assignments could be made. When individuals were compared to all 15 collection locations in GeneClass 2, 53% were assigned back to their collection location (Table S1). Of the remaining individuals, 71% were assigned to a different river including 91% of Mira River Striped Bass assigned to Bras d’Or–Miramichi, 59% of Cape Fear River Striped Bass assigned to the Roanoke River, and 48% of Striped Bass from the Delaware River and the Chesapeake Bay assigned to a different river in that area. Assignment patterns seen in North Carolina and Delaware River–Chesapeake Bay correspond to the geographic regions used in previous studies, which grouped North Carolina rivers together, and the Delaware River with all Chesapeake Bay rivers (Gauthier et al., 2013). Assignment success and accuracy were similar when analyzed with rubias, with 51% of individuals assigned back to their collection location and 76% of the remaining individuals assigned with high likelihood to a different river.

In the second analysis, Striped Bass were pooled into six geographic regions or proposed reporting groups: the Gulf of St. Lawrence, Shubenacadie River, Saint John River, Kennebec–Hudson River, Delaware–Chesapeake Bay, and Roanoke–Cape Fear. The three regions containing US Striped Bass correlate with groupings made by Gauthier et al. (2013), and all six regions correspond to one of the six genetic clusters identified in LEA. Under this scenario, both GeneClass2 and rubias assigned 99% of individuals to a reporting group: 97% to the group from which it was collected, and 2% to a different group (Table 5).

4.1 Genetic diversity

Genetic diversity was highest among sampled US locations and lowest in the relatively isolated populations in the Gulf of St. Lawrence and the Shubenacadie River. All sampled US populations had similar mean observed heterozygosity, including the small and hatchery-supported Cape Fear River, suggesting that genetic variation comparable to the rest of the US range is being maintained in this population. The Kennebec River population similarly does not show signs of reduced genetic diversity, despite the Kennebec River's recent restoration with stocked Hudson River Striped Bass (G. Whippelhauser, pers. comm.). Within Canada, the highest observed heterozygosity was seen in the Saint John River, only slightly lower than values seen in the United States, suggesting this population has retained substantial genetic diversity despite its apparent decline in numbers since the 1970s (Andrews et al., 2017; LeBlanc et al., 2018). Diversity values calculated for all populations were comparable to patterns seen in a recent range-wise microsatellite study (Wirgin et al., 2020). Lower genetic diversity in northern, previously glaciated locations has been seen in other anadromous fish species such as American Shad (Alosa sapidissima; Hasselman, Ricard, & Bentzen, 2013) and is expected in populations on the edge of a range expansion (Bernatchez & Wilson, 1998; Hewitt, 1996). The lowest observed heterozygosity values seen in this study (0.26–0.28) were similar to observed heterozygosity seen in other anadromous fishes examined using SNP markers, such as Blueback Herring (Ho = 0.28–0.30), Alewife (Ho = 0.22–0.27; Baetscher et al., 2017), and Chinook Salmon (Oncorhynchus tshawytscha;Ho = 0.26–0.32; (Clemento, Abadía-Cardoso, Starks, & Garza, 2011).

4.2 Outlier loci represent regions of major effect

Most ecologically relevant traits are thought to be polygenic, involving small allele frequency differences of many genes (Pavey et al., 2015; Yeaman, 2015). All three outliers identified in this study showed high allele frequency changes among populations (0.85–0.97 maximum allele frequency differences), which suggests the presence of single genes or regions of major effect. The two unannotated outlier loci identified in this study (25891_222 and 27535_2519) showed a strong tendency toward fixation for one allele in Canadian populations, and very low frequency of that allele in southern populations, while locus 4437_41108 showed a tendency toward fixation of allele A in Shubenacadie River and Gulf of St. Lawrence Striped Bass and very high allele frequencies of allele B in populations south of the Hudson River. Within Striped Bass in the Saint John River, Kennebec River, and Hudson River, in contrast, both alleles were maintained at relatively equal frequency. Further characterization of the Striped Bass genome and anchoring of existing scaffolds into linkage groups will be invaluable for placing these outlier loci into a wider genomic context, and sequencing of igf2b in northern versus southern individuals will shed light on whether locus 4437_41108 is associated with nonsynonymous mutations within this gene.

4.3 Canadian populations are highly distinct

Rivers near the Gulf of St. Lawrence (Bras d’Or–Miramichi and Mira River), the Shubenacadie River, and the Saint John River were consistently, highly differentiated from each other and from US populations (F_ST = 0.13–0.20). Phylogeographic theory predicts that populations founded after the last glacial retreat will show less intraspecific divergence than their southern counterparts (Bernatchez & Wilson, 1998). Unexpectedly high divergence in Canadian populations has been seen in other anadromous fishes along the North American Atlantic coast and has been attributed both to the circuitous coastline created by the Nova Scotia peninsula and to a complex hydrography within the Bay of Fundy that drives differentiation of native fish populations (Hasselman et al., 2013; King, Kalinowski, Schill, Spidle, & Lubinski, 2001; McConnell, Ruzzante, O’Reilly, Hamilton, & Wright, 1997). Variation in habitat is known to drive differentiation of anadromous fish species such as Atlantic Salmon (Bradbury et al., 2014) and Dolly Varden Char (Salvelinus malma; Bond, Crane, Larson, & Quinn, 2014). The Shubenacadie River, in particular, is the only tidal bore river wherein Striped Bass are known to successfully spawn (Rulifson & Dadswell, 1995), and the extreme environmental conditions that eggs and larvae must tolerate in this river may contribute to its increased population differentiation (Rulifson & Tull, 1999). Unexpectedly high genetic divergence in Canadian populations could also be the result of small initial colonization sizes driving changes in allele frequencies that persist to the present day (Excoffier & Ray, 2008).

Genetic similarity between the Mira River and Bras d’Or–Miramichi Striped Bass indicates that these two groups have the same origin. It is likely that Striped Bass currently residing in the Mira River migrated from the Miramichi River at some point after the formation of suitable estuarine habitat and nursery areas some 500–800 years ago (Andrews, Dadswell, et al., 2019). While Striped Bass in the Mira River appear behaviorally distinct, demonstrating multiannual residency and spring upstream migration shown in an acoustic telemetry study in 2012–2015 (Andrews, Dadswell, et al., 2019; Buhariwalla, 2018), our data suggest that this potential spawning population is not genetically distinct from Striped Bass found within the Gulf of St. Lawrence.

Also identified from Mira River samples were individuals of putative US origin (3/22 samples). Recent evidence that Striped Bass move between the US Atlantic coast and the northeastern coast of Nova Scotia is scarce. In 1983, a single fish tagged in the Kouchibouguac River in the Gulf of St. Lawrence in 1983 was later recaptured in the Wye River, Maryland (Hogans & Melvin, 1984), indicating that this fish likely passed through the northeastern coast of Nova Scotia. In contrast, none of the hundreds of Striped Bass with internal acoustic tags in the Roanoke River, Hudson River, New England coast, Bay of Fundy, and Miramichi River (Andrews, Wallace, Gautreau, Linnansaari, & Curry, 2018; Broome, 2014; Callihan et al., 2015; Douglas, Bradford, & Chaput, 2003; Gahagan et al., 2015; Pautzke, Mather, Finn, Deegan, & Muth, 2010) have ever been detected passing the Halifax Line of acoustic receivers on the eastern coast of Nova Scotia (Andrews, Dadswell, et al., 2019). Thousands of Striped Bass in the Gulf of St. Lawrence (DFO, 2010; Douglas et al., 2003; Hogans & Melvin, 1984), the Bay of Fundy (Broome, 2014; Rulifson & Dadswell, 1995), and along the US coast (Pautzke et al., 2010; Richards & Rago, 1999; Waldman et al., 1990) have been externally tagged from the 1960s to the present day (Andrews, Dadswell, et al., 2019), only one of which has ever been caught on the far eastern shores of Nova Scotia (Andrews et al., 2018; Douglas et al., 2003). This apparent isolation may be caused by a physical isolation of the Gulf of St. Lawrence before the Canso Strait opened postglacier retreat (Shaw & Courtney, 2002) and after the Canso Causeway was built in 1955 (Vilks, Schafer, & Walker, 1975), or influenced by a sharp temperature change between the two water bodies (Rulifson & Dadswell, 1995). A “genetic breakpoint” has been described in several other species along eastern Nova Scotia at ~45°N (close to the City of Halifax; Stanley et al., 2018). Increasing ocean temperatures are predicted to drive Striped Bass populations north, but this remains a poorly studied region.

The presence of a genetically distinct population of Striped Bass in the Saint John River following its suggested extirpation in the 1970s has been debated for over a decade (Andrews et al., 2017). Two previous studies have found evidence of unique genotypes distinct from US and Shubenacadie River Striped Bass, and present in adults (Bentzen & Paterson, 2008) and juveniles (LeBlanc et al., 2018). A third study examined a mixture of 17 juveniles and 25 adults collected from the Saint John River in 2014 and found that all fish showed admixture between Shubenacadie River and US genotypes with no unique cluster (Wirgin et al., 2020). The 17 juveniles examined by Wirgin et al. (2020) are included in this present study, along with 15 additional juveniles collected in 2015–2017. In contrast to Wirgin 2020s results, most juveniles we examined showed a distinct genetic signature and admixture was only seen between the Saint John River cluster and either Shubenacadie or US genotypes. We detected no Shubenacadie–US hybrids. Adult and juvenile Saint John River Striped Bass examined in both studies are also part of an ongoing (6 + year) acoustic telemetry study. Initial telemetry results clearly demonstrate differing migratory patterns between adults genotyped as Shubenacadie origin (which left the river to spawn), adults genotyped as belonging to the Saint John River cluster (which migrated upstream) and those of US origin Striped Bass (which aggregated around the Hammond River, a tributary of the Saint John River; Andrews, Linnansaari, et al., 2019).

Striped Bass from US populations have been found in Minas Basin (Bay of Fundy; Rulifson, McKenna, & Dadswell, 2008) and are thought to enter the Shubenacadie River as well, and this study found no evidence that migrants successfully spawn in the Shubenacadie River. In contrast, juveniles from the Saint John River were admixed with Shubenacadie River and Chesapeake Bay populations. It is unknown how often this gene flow occurs now or prior to the population's apparent disappearance in the 1970s. All admixed juveniles we examined had approximately equal assignment to the Saint John River and the Shubenacadie River/Chesapeake Bay clusters, suggesting intraspecific F1 hybrids. The first-generation hybrids and the distinctness of the Saint John River–US genotypes suggest that these admixed juveniles may be a recent development. There is little information about the proportion of US migrants in the Saint John River before the population crash and no information about possible admixed individuals (Andrews et al., 2017). Larger Striped Bass are more likely to migrate and to travel far (Andrews, Dadswell, et al., 2019; Callihan et al., 2014; DFO, 2018), and as Striped Bass populations recover, there is an increase in the number of older, larger individuals making migrations (Callihan et al., 2014). We hypothesize that the admixed juveniles result from small numbers of local spawners making admixed offspring more prevalent, increased migration from recovering populations, and a climate-induced northward range shift.

4.4 US Regional Structure

In contrast to Canadian Striped Bass, F_ST values among US locations were much lower (F_ST = 0.000–0.035) and support three regions with low but significant genetic divergence: (a) Hudson River and Kennebec River, (b) Delaware River and Chesapeake Bay, and (c) North Carolina Rivers. Overall, our results suggest individual spawning populations within the Delaware River and Chesapeake Bay make up a large metapopulation connected by extensive gene flow, with lesser amounts of gene flow between this region and populations to the north and south.

Connectivity among populations of Striped Bass along the Atlantic Coast has been investigated in several previous studies (Able, Grothues, Turnure, Byrne, & Clerkin, 2012; Bentzen & Paterson, 2008; Brown et al., 2005; Callihan et al., 2015; Gauthier et al., 2013) and is influenced by gene flow, stocking, and possibly recolonization following local extirpation. Striped Bass between Maine and North Carolina are highly migratory (>1,000 km; Callihan et al., 2015), compared to the more restricted migratory range of populations in Canada and the apparent complete residency of populations south of Roanoke River, North Carolina. Several tagging studies have previously documented movement of Striped Bass among Chesapeake Bay, Kennebec River, Hudson River, and Roanoke River (Callihan et al., 2015; Dorazio, Hattala, McCollough, & Skjeveland, 1994; Gahagan et al., 2015; Kneebone et al., 2014; Waldman et al., 1990). The presence of an isolation-by-distance pattern of differentiation among US locations but not among Canadian locations further supports gene flow among populations in this range.

Our study also investigates the current genetic profile of the recently restored Kennebec River population of Striped Bass. The Kennebec River is one of several rivers in Maine that likely once hosted a native population of Striped Bass (Little, 1995). This population declined and was likely extirpated by the late 1930s, and was subsequently stocked with over 260,000 Striped Bass juveniles from 1982 to 1991 from the Hudson River in an attempt to restore the population (G. Whippelhauser, pers. comm.). Considering stocking, it is unsurprising that the juvenile Striped Bass caught in the Kennebec River in this study were most similar to the Hudson River. F_ST values between the two rivers were low (F_ST = 0.008) but significant, and similar to values seen among some rivers within the Chesapeake Bay, indicating a similar level of relatedness despite the large geographic distance between them (approximately 620 km from mouth to mouth). This similarity was also seen in a recent microsatellite study that examined Kennebec River juveniles (Wirgin et al., 2020), which found no statistically significant difference between the Kennebec River and Hudson River.

Within the Chesapeake Bay and Delaware River, F_ST values were very low. The highest values were seen when comparing James River individuals to other rivers in the Bay, as well as Nanticoke and Choptank Rivers, both located on the east coast of the Bay, to rivers on the west coast. A small amount of differentiation between the east and west coast of Chesapeake Bay was also seen in the most recent microsatellite study done on Striped Bass (Wirgin et al., 2020) and may be due to the channel of deeper water that runs through the center of the Bay. The lowest F_ST values within the Chesapeake–Delaware region were seen between individuals from the head of Chesapeake Bay and Delaware River, supporting the hypothesis that these two groups of Striped Bass are not genetically distinct from one another. Previous genetic studies have found conflicting results on whether the growing Striped Bass population in the Delaware was distinct from the Chesapeake Bay. An analysis of mitochondrial length–frequency differences in the recovered Delaware River Striped Bass found significant differences in minor length–frequency alleles from Chesapeake Bay Striped Bass (Waldman & Wirgin, 1994). Minor length–frequency differences were also seen among tributaries within the Chesapeake Bay (Wirgin, Maceda, et al., 1993), and microsatellite studies which found significant F_ST values between the Delaware River and the Chesapeake Bay also found F_ST values of the same magnitude among tributaries within the bay (Gauthier et al., 2013). Decades of observations of adult Striped Bass using the Chesapeake and Delaware Canal to transit between the Chesapeake and Delaware estuaries during spawning season (Kneebone et al., 2014; Koo & Wilson, 1972; Nichols & Miller, 1967) support the likelihood that the Delaware River population receives a high amount of gene flow from Chesapeake Bay Striped Bass, on a similar magnitude as seen among rivers within the Bay. Whether the current Delaware River Striped Bass are descended from a remnant population that was genetically similar to the Chesapeake Bay or whether they are descended from Chesapeake Bay Striped Bass that recolonized the river, it seems clear that Delaware River Striped Bass are part of a complex network of gene flow among the tributaries of the Chesapeake Bay.

4.5 Assignment

Self-assignment tests were performed on the SNP panel generated in this study to assess its utility as a reference dataset for future mixed-stock analyses. Previous attempts to use genetic markers for mixed-stock analysis have met with limited success. Most recently, a study conducted self-assignment tests using GeneClass2 on 14 microsatellites and was able to assign 60% of Striped Bass from the Hudson River, Chesapeake Bay (including the Delaware River), North Carolina, and South Carolina to a region of origin (Gauthier et al., 2013). Our SNP panel showed the highest assignment success when overlapping populations were grouped into the same reporting groups used by Gauthier et al. (2013). We were able to assign 99% of Striped Bass to a region of origin with >80% confidence. When individuals were assigned to river of origin (rather than region of origin), assignment success was much lower and individuals were misassigned to other rivers within the same region, reflecting the low genetic differentiation among these rivers. The assignment success rate seen within regions is likely an indication that rivers within a region are not demographically independent.

Statistical biases when using large panels of SNP loci have been identified in assignment tests that use simulated individuals to predict assignment accuracy of a set of loci (Anderson, Waples, & Kalinowski, 2008); however, the data in this present study indicate that self-assignment tests in the absence of simulations can result in misleadingly high confidence values. In light of emerging techniques allowing high-throughput genotyping of large numbers (>1,000) of loci (Ali et al., 2016), researchers looking to assess stock composition of increasingly closely related populations should interpret confidence scores with these issues in mind when choosing a geographic resolution in which to assign fish. In addition, admixed individuals seen in the Saint John River were assigned to one of their parent populations with high confidence, suggesting that assignment in both GeneClass2 and rubias is insensitive to the presence of admixed individuals. When performing mixed-stock analysis on locations with large numbers of hybrid individuals, assignment may be better conducted using a genetic clustering algorithm such as those found in LEA or STRUCTURE. Overall, our SNP panel constitutes a significant improvement over other genetic markers in assigning Striped Bass to regional areas along the Atlantic coast and will be invaluable to the development of a highly accurate and reliable genetic tool for mixed-stock analysis of the species across the central and northern portion of their range.