Atlantic cod (Gadus morhua) populations in the Gulf of Maine (GoM) are at a fraction of their historical abundance, creating economic hardships for fishermen and putting at risk the genetic diversity of the remaining populations. An understanding of the biocomplexity among GoM populations will allow for adaptive genetic diversity to be conserved to maximize the evolutionary potential and resilience of the fishery in a rapidly changing environment. We used restriction-site-associated DNA sequencing (RADseq) to characterize the population structure and adaptive genetic diversity of five spawning aggregations from the western GoM and Georges Bank. We also analyzed cod caught in the eastern GoM, an under-sampled area where spawning aggregations have been extirpated. Using 3,128 single nucleotide polymorphisms (SNPs), we confirmed the existence of three genetically separable spawning groups: (1) winter spawning cod from the western GoM, (2) spring spawning cod, also from the western GoM, and (3) Georges Bank cod. Non-spawning cod from the eastern GoM could not be decisively linked to either of the three spawning groups and may represent a unique component of the resource, a mixed sample, or cod from other unsampled source populations. The genetic differentiation among the three major spawning groups was primarily driven by loci putatively under selection, particularly loci in regions known to contain genomic inversions on linkage groups (LG) 7 and 12. These LGs have been found to be linked to thermal regime in cod across the Atlantic, and so it is possible that variation in timing of spawning in western GoM cod has resulted in temperature-driven adaptive divergence. This complex population structure and adaptive genetic differentiation could be crucial to ensuring the long-term productivity and resilience of the cod fishery, and so it should be considered in future management plans.
Citation: Clucas GV, Kerr LA, Cadrin SX, Zemeckis DR, Sherwood GD, Goethel D, et al. (2019) Adaptive genetic variation underlies biocomplexity of Atlantic Cod in the Gulf of Maine and on Georges Bank. PLoS ONE 14(5): e0216992. https://doi.org/10.1371/journal.pone.0216992
Editor: Arnar Palsson, University of Iceland, ICELAND
Received: August 18, 2018; Accepted: May 2, 2019; Published: May 24, 2019
Copyright: © 2019 Clucas et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Raw read data can be found on the sequence read archive (SRA BioProject accession number: PRJNA521889). Our final SNP datasets used for analyses and details of the outlier loci identified by BayeScan can be downloaded from FigShare (https://doi.org/10.6084/m9.figshare.7683938.v1).
Funding: This work was funded by a National Oceanic and Atmospheric Administration Saltonstall-Kennedy grant (NA14NMF4270014; LK, GS, SC, AK, DG; https://www.fisheries.noaa.gov/national/funding-and-financial-services/saltonstall-kennedy-grant-program-application-process-and-deadlines) and the National Oceanic and Atmospheric Administration College Sea Grant program through the New Hampshire Sea Grant project R/SFA-3 (AK, SC, LK, #NA14OAR4170083, https://seagrant.unh.edu/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Identifying stock boundaries of marine populations remains a challenge despite rapid advances in genetic stock identification methods due to the considerable complexity in the population structure of many marine species . Most marine fishery resources are characterized by large population sizes and high dispersal ability, resulting in subtle patterns of population differentiation [2,3], the characterization and interpretation of which is complex . Given advances in sequencing technologies, thousands to millions of genetic markers can now be used to study population structure, enabling high-resolution interpretations.
Gene flow is often used to infer demographic connectivity, so that groups connected by high gene flow are often considered as one continuous stock. However, populations can be demographically independent despite high gene flow . Further, populations connected by high gene flow may lack genetic differentiation at neutral genomic markers yet exhibit differentiation at adaptive loci as a result of divergent selection pressures. Patterns of gene flow and adaptive genetic diversity contribute to biocomplexity among stocks in life history, morphology, and local adaptation that may be highly relevant to fisheries management [5–7]. Accordingly, there may be unique ecological and functional diversity among stocks [8–11] the maintenance of which could be key to ensuring adaptive capacity or evolutionary potential [12,13]. Biocomplexity may also confer resilience; loss of intra-specific genetic diversity has been linked to reduced population stability and reduced resilience to exploitation and changing environmental conditions [5,11,14–16]. The U.S. Endangered Species Act mandates that populations divergent at putatively adaptive loci warrant designation as separate management units even if they appear panmictic at neutral loci. Therefore, stock identification for fisheries management should aim to use thousands of genetic loci to aid in assessing biocomplexity, including both neutral and adaptive genetic diversity, among populations, especially in regions experiencing rapid environmental change.
Population structure of Atlantic cod (Gadus morhua) is complex [1,17] and many investigations of the various populations and ecotypes that co-exist across the North Atlantic have made cod a canonical example of biocomplexity in a marine species. In U.S. waters, there is considerable biocomplexity in cod stocks [18–22] that is not reflected in the current stock assessment and management structure in place since 1972. The two management units consist of: 1) cod from Georges Bank, offshore of Cape Cod, and in the waters of southern New England; and 2) a group consisting of cod from the western, central, and eastern GoM (Fig 1). However, genetic data [23–28], differences in growth rates [29,30], timing of spawning [19,31], larval dispersal patterns [32,33], movement patterns [18,34–36], spawning site fidelity [37–39], and differences in life history strategy [40,41] all confirm that there is biocomplexity in US cod that conflicts with the stock designations (for a review see REF 20). The misalignment of stock designations with this biocomplexity is implicated as a contributing factor in the failure of these stocks to rebuild despite decades of intensive management .
Samples from Ipswich and Massachusetts Bays and Georges Bank were from known spawning grounds (dark gray circles), whereas the sentinel fishery targeted non-spawning cod in the eastern GoM (circled region). We do not have exact catch locations for all of the eastern GoM samples, as fishermen were not required to declare them, and so we show the overall area in which they were caught. The management unit boundaries of the GoM, Georges Bank, and Scotian Shelf are indicated by the thin black lines.
Populations of Atlantic cod in the Gulf of Maine (GoM) are at a fraction of their historical abundance [22,43]. Currently, cod biomass is estimated at just 5–8% of its target level  and the fishery has been declared an economic disaster. Overexploitation has been implicated as one factor in the extirpation of historical spawning components in the GoM  and the low biomass of the remaining components puts the fishery at risk, especially as the GoM is warming faster than 99% of the global ocean [44–46]. Cod productivity appears to be closely linked to temperature [44,47,48] and adaptive diversity is likely going to be key for the resilience of the fishery in the face of this rapid ecosystem change.
In their investigation of genetic stock structure in the GoM using microsatellites, Kovach et al.,  found genetic differentiation among spawning aggregations in the GoM, primarily stemming from outlier loci. This structure included differentiation among subpopulations that overlap spatially but differ in spawning season (i.e., genetically distinct winter- and spring-spawning populations). Barney et al.,  also found that adaptive variation played a key role differentiating two spawning populations in the western GoM and one on Georges Bank, with differentiation in regions of the genome that were in high linkage disequilibrium (LD) on linkage groups (LG) 2, 7 and 12. These LD blocks are chromosomal inversions [49–51] and have been previously found to segregate between populations of cod in other regions spatially [52–55], temporally [6,56], between stationary and migratory ecotypes [50,51,54,55,57,58], and have been linked to spatial clines in temperature across the North Atlantic [52,53]. Therefore, despite potential gene flow among subpopulations within the GoM, there is growing evidence for adaptive differentiation among spawning aggregations [23,24].
To date, no genetic studies have focused on cod in the eastern GoM (from US waters off midcoast and eastern Maine). Ames’  analysis of historical fishing grounds suggested that cod previously found in these areas comprised separate spawning components that were likely discrete from cod in the western GoM. Spawning components within the eastern GoM have been functionally extirpated  and since 2010, the Maine Center for Coastal Fisheries’ Sentinel Hook Survey Fishery has monitored cod abundance in the region to determine whether stocks are rebuilding [59,60]. Although spawning aggregations have not been observed recently, cod caught by the sentinel survey allow an opportunity to determine the level of connectivity between the eastern GoM and other regions. Characterizing the connectivity of these eastern GoM cod with the more robust western populations is vital for understanding the metapopulation structure of cod in the GoM and their potential for recovery .
In this study, we used genome-wide genetic data to analyze both the neutral and putatively adaptive population structure of cod in the GoM. We sampled seasonally divergent spawning aggregations in the western GoM to characterize gene flow and adaptive divergence between populations that spawn in the same locations but in different seasons. We also investigated connectivity among these western GoM spawning aggregations, a Georges Bank spawning aggregation, and cod caught by the sentinel fishery in the eastern GoM that were not in spawning condition. If there were distinct subpopulations in the eastern GoM, as hypothesized [19,20], and the cod caught in these areas today are remnants or resurgent of these subpopulations, then they could still harbor unique genetic signatures even if they were not spawning at the time of sampling. Alternatively, a lack of differentiation of eastern GoM fish may suggest that there has been contemporary movement of individuals into the eastern GoM, or that these fish were never in fact distinct from other regions.
For the western GoM, fin clip samples from spawning adult cod were collected on known spawning grounds in Ipswich and Massachusetts Bays at two distinct spawning seasons–late spring (May/June) and winter (December/January; Table 1). The National Marine Fisheries Service granted the GMRI permission to collect samples during fisheries closures for the sole purpose of scientific research via a Scientific Research Letter of Acknowledgment. The fishing vessel F/V Ellen Diane was not subject to the Magnuson-Stevens Act or fishery regulations at 50 CFR part 648 while collecting samples for scientific research. For Georges Bank, we used archived DNA samples from the study of Kovach et al., . These samples were collected by a survey conducted by Canada’s Department of Fisheries and Oceans from spawning adults on the Northeast Peak in February 2006 and, although there is a temporal mismatch between these and the other samples used in this study (Table 1), previous work  and this study shows temporal stability in the population structure in the region. For the eastern GoM, fin clip samples were collected from non-spawning cod caught in the sentinel fishery survey from stations located throughout mid-coast and Downeast Maine, with a majority of the samples from the inshore and nearshore areas and few collected from offshore (Fig 1). Maturity stages were defined based on .
Library preparation and sequencing
DNA was extracted from fin clips with a Qiagen DNeasy Tissue Kit (Qiagen, Valencia, CA, USA), following the manufacturer’s protocols. RAD libraries were prepared and sequenced by Floragenex Inc. (Eugene, OR, USA), following standard protocols of . Libraries were prepared from 50 μL of DNA per sample with concentrations of 10–20 ng/μl. Restriction digestion was performed with the SbfI enzyme and 300–500 base pair fragments were size selected. Individual DNA samples were barcoded with uniquely indexed nucleotides, to allow for the pooling of 95 multiplexed individuals per library. The resulting RAD libraries were sequenced on two lanes of an Illumina HiSeq using Illumina 1 x 100 bp chemistries.
Raw read quality was assessed with FastQC  and reads were demultiplexed using the process_radtags function from the Stacks v1.35 pipeline [64,65]. Adapter trimming and quality filtering were performed with skewer  using the following settings: trim adapters found anywhere in a read (-m any), maximum allowed error rate = 0.1 (-r 0.1), trim 3’ end until minimum quality of 20 is reached (-q 20), trim reads with mean quality < 20 (-Q 20), minimum read length allowed after trimming = 40 (-l 40), filter highly degenerative reads (-n). After this filtering, 15 individuals with fewer than 800,000 reads remaining were discarded from further analysis. These individuals had significantly fewer reads than other individuals and initial single nucleotide polymorphism (SNP) calling trials resulted in more than 30% missing data for these individuals. Therefore, to retain the maximum number of loci, we discarded these individuals before performing our final SNP calling and SNP filtering.
SNP calling was performed using the ipyrad v0.7.21 pipeline  mapping to the cod reference genome (gadMor2; ). This pipeline uses bwa-mem  with default settings for alignment to the reference genome and discards reads that do not map uniquely. We set the minimum depth per locus to five reads and the maximum to 50,000 reads within individuals. We set this high maximum depth filter due to the high depth of sequencing that some of our samples received (Figure A in S1 Appendix). Additional maximum depth filters were applied downstream (see below) to remove potential paralogs. We specified that a locus had to be shared by at least 70% of individuals in each group to be included in the final dataset, while all other ipyrad settings were left as defaults, including the maximum heterozygosity filter of 0.5.
After SNP calling, SNPs were further filtered using vcftools v0.1.15  so that the final dataset included only: biallelic SNPs; one SNP per RADtag; SNPs with a minor allele frequency ≥ 5%; SNPs with less than 30% missing data per sampling location; SNPs with mean depth less than two standard deviations above the mean to avoid paralogs; and SNPs that were in Hardy-Weinberg equilibrium in at least half of the groups at the p < 0.01 significance level. PGDSpider  was used to convert the data between formats for downstream analyses.
Putatively neutral SNP selection
To identify a putatively neutral SNP dataset for determining gene flow among groups, we first thinned the data using vcftools such that SNPs were at least 10 kb apart to avoid tightly linked SNPs. We then used the FST outlier method implemented in BayeScan  to identify outlier loci. We set the prior-odds of neutrality to 10 to favor the detection of outliers and used a q-value cut-off of 0.05 (this means 5% of identified loci are expected to be false positives). Furthermore, we removed SNPs located within regions of the genome with putative inversions. These inverted regions have elevated linkage disequilibrium, including inter-chromosomal linkage disequilibrium , and in some cases show elevated FST among spawning aggregations within the GoM . Therefore, we discarded SNPs within these regions as well as the FST outliers, to create our putatively neutral SNP dataset, hereafter “neutral dataset”. Further analyses were conducted with and without these outlier loci and are referred to as the full dataset and neutral dataset, respectively.
Pairwise FST values between groups were calculated for the neutral and full SNP datasets with the Hudson estimator as it has been shown to be insensitive to variation in sample sizes and was found to be unbiased when tested on SNP datasets [74,75]. To calculate the Hudson estimator, we used the R script of Di Gaetano et al., , correcting for an error on line 14, and modified the script to estimate bootstrapped confidence intervals with 1000 replicates, using the ‘boot’ package in R and the percentile method for the confidence interval.
Population structure was visualized with both datasets using a principal components analysis (PCA), performed using the adegenet v2.1.1 packages in R [77–79]. Before performing the PCA, allele frequencies were scaled and centered using the scaleGen function, and missing values were replaced by the mean allele frequency. PCA was performed with the dudi.pca function.
Discriminant Analysis of Principal Components (DAPC, REF 78) was also performed to assess genetic differentiation. The find.clusters algorithm suggested the most likely number of clusters within the data was equal to one (K = 1) for both datasets, suggesting population differentiation was subtle, and so we used the a priori clustering of individuals by group instead. The number of principal components (PCs) to retain was estimated by cross-validation using the xvalDAPC function with 1000 replicates .
To verify that variation in sequencing depth and missing data did not produce erroneous signals of population structure, we created a more stringent SNP dataset, subsetting the final SNP dataset for SNPs with less than 5% missing data across all sampling locations and repeated the DAPC analysis, as above. With this stringent SNP dataset, we also recoded missing data as 1 and genotypes as 0 and performed a PCA using the prcomp() function in R, scaling and centering the data, to check that patterns of missing data were not driving population structure.
Structure v2.3.4  was also used to find clusters in the full and neutral SNP datasets and to estimate the proportion of ancestry individuals inherited from each cluster, effectively assigning individuals to clusters. The admixture model with correlated allele frequencies was used in all cases, as we expected to find a high level of admixture and gene flow among groups. An initial run was performed with K = 1, allowing lambda to vary, to estimate the value of lambda that was then set in all subsequent runs. Structure was run both with and without sampling locations as priors to differentiate between subtle and strong population differentiation, and values of K from one to six were tested. Each analysis was run for 150,000 generations, discarding the first 50,000 as burn-in, and repeated ten times from a random seed for each value of K. Structure Harvester web v0.6.94  was used to compare replicates, calculate the most likely value of K according to the Evanno method , and prepare files for CLUMPP . CLUMPP aligns the results from replicate runs of Structure to check for multimodality and calculates the average membership coefficients of individuals to each cluster, ready for visualization with DISTRUCT v1.1 . To check for hierarchical structure, we removed groups that appeared differentiated and then repeated the Structure analysis on the remaining groups.
To investigate patterns of genetic differentiation throughout the genome, we created Manhattan plots using the R package qqman for locus-specific FST values from the full dataset for all pairwise comparisons between groups. Locus-specific FST was calculated using the Weir and Cockerham  estimator in vcftools. Spring and winter spawning individuals from Ipswich Bay and Massachusetts Bay were grouped together into “spring spawners” and “winter spawners” for these comparisons due to the genetic similarity between bays in the same season.
Sequencing, SNP calling, and filtering
After discarding individuals with low read counts, we called SNPs in a total of 128 individuals across our six sampling locations (Table 1). The average number of reads per individual was 3.34 million (range: 0.87–12.18 million), with on average 84% of reads retained after quality control, adapter trimming, and mapping. Despite uneven sequencing depth among individuals and groups (Figure A in S1 Appendix), patterns of population structure did not correspond to these differences in sequencing depth or the lane in which individuals were sequenced, indicating that the patterns of population structure were robust. For example, spring spawners from Ipswich Bay, which received high coverage and were sequenced in lane 1, were genetically most closely related to spring spawners in Massachusetts Bay (see Population structure section), which had received much lower coverage and were sequenced in lane 2.
We initially identified 73,495 SNPs across all RADtags using Ipyrad. After SNP filtering, we retained 3,128 SNPs which had a mean sequencing depth of 77.5X when averaging across all individuals (range 20.8–113 X, Figure B in S1 Appendix). This comprised our full SNP dataset which included potentially adaptive loci (Table A in S1 Appendix) and loci with up to 30% missing data per sampling location. To further verify that variation in sequencing depth and missing data did not produce erroneous signals of population structure, we further subsetted this dataset to remove SNPs with more than 5% missing data, creating a stringent SNP dataset of 1,660 SNPs. To obtain the neutral dataset, we thinned our full SNP dataset to remove any SNPs that were < 10kb apart, removed an additional 47 SNPs that were identified as outliers (the details of these are available on figshare: https://doi.org/10.6084/m9.figshare.7683938.v1) and another 106 SNPs located in regions in high linkage disequilibrium on LG 2, 7, and 12. This resulted in a neutral dataset of 2,689 SNPs.
When we used the full dataset, which included the neutral SNPs plus the 106 SNPs in inverted regions on LG 2, 7 and 12, and 47 outlier SNPs identified by BayeScan, all pairwise FST values between sampling locations were small, but confidence intervals did not overlap zero (Table 2). Comparisons between the bays in the western GoM showed stronger differentiation by spawning season than geographic location. Georges Bank was more differentiated from the western GoM than the eastern GoM. Lastly, the eastern GoM showed similar levels of differentiation with the western GoM winter spawners as with Georges Bank, but higher differentiation with the western GoM spring spawners.
The weighted mean value is given above and the confidence interval is given below.
Results from PCA and DAPC showed patterns consistent with those described by the pair-wise FST comparisons and differentiation of the spring spawners. Individual scores formed three clusters on the PCA along PC1 (Fig 2A), which explained 1.6% of the variance. This three-cluster pattern is typical when there are two alternate forms of a large, high linkage disequilibrium genomic block driving genetic differentiation, because individuals can effectively be genotyped as having either of the homozygous states (clusters at the extremes of PC1) or the heterozygous state (the cluster in the middle of PC1). Spring spawners from Ipswich and Massachusetts Bays were found in the middle and right-hand cluster and shifted towards the upper right-hand corner of the plot; winter spawners were only found in the middle and left-hand cluster and were shifted towards the lower left-hand corner of the plot. Similar to the spring spawners, winter spawners from Ipswich and Massachusetts Bays overlapped with one another. Georges Bank and eastern GoM cod were found in all three clusters, but towards the lower half of the plot where the winter spawners were also found. There was no additional clustering in the PCA when other PCs were plotted. SNP loadings (Figure C in S1 Appendix) revealed that SNPs in the inverted region on LG 7 were largely responsible for the patterns along PC1, while SNPs in the inversion on LG 12 differentiated individuals along PC2. A single SNP on LG18 also seemed to have a large effect in differentiating populations on PC1. The clusters on the far left and right of PC1 correspond to individuals homozygous for the inverted and non-inverted regions of LG 7, while individuals in the middle cluster are heterozygous for the inversion. The differentiation along PC2 is the result of similar patterns stemming from the inversion on LG12.
a) PCA including all individuals, b) DAPC including all individuals, c) PCA after the Massachusetts Bay outlier individuals were removed, d) DAPC after the same Massachusetts Bay outlier individuals were removed. The percentage of the variation explained by each principle component is shown on the axis labels for the PCAs, and the cumulative percentage of variation explained by the discriminant functions is shown on the inset graphs for the DAPCs. The number of principle components retained in b) was 20 and in d) was 26, as assessed using cross-validation.
DAPC provided an informative way to visualize these patterns of differentiation among the groups (Fig 2B). Spring spawners tended to group together into one cluster, winter spawners into a second cluster, and Georges Bank and eastern GoM into a third cluster when individuals were assigned to their a priori sampling group. The same clustering was also observed when the DAPC was repeated with the stringent SNP dataset of 1,660 loci (i.e. SNPs with only 5% missing data; Figure D in S1 Appendix) and there was no structure in the missing data (Figure E in S1 Appendix), confirming that the population structure we observed had a biological basis. The PCA plot using the full SNP dataset, and to a lesser degree the DAPC plot, revealed two individuals in the Massachusetts Bay winter spawning aggregation that appeared to be more genetically divergent than all others. We therefore removed these outliers and repeated both analyses (Fig 2C and 2D). The patterns we observed were stable after removal of these outlier individuals.
The differentiation of the spring spawners was also evident in a Structure analysis using the full SNP dataset. When K (number of populations) was varied from two to four, the spring spawners appeared differentiated from other groups (Fig 3). This pattern was evident in analyses both with location priors (Fig 3) and without location priors (Figure F in S1 Appendix). The Evanno method suggested K = 2 was the most likely value of K in both cases, although the estimated posterior probability of the data increased up to K = 4 (Figures F and G in S1 Appendix).
Neutral population structure
When we analyzed the data after removing the 47 outlier SNPs and the 106 SNPs on LG 2,7, and 12 within the inversions, pairwise FST values among all groups were much smaller than for the full dataset. FST values were close to zero with the neutral dataset, although the confidence intervals again did not overlap zero, suggesting the possibility of some fine-scale differentiation among groups (Table 3). Pairwise FST was almost the same for two comparisons: Georges Bank versus eastern GoM and Ipswich Bay winter versus Massachusetts Bay winter (Table 3). The results could suggest that cod in these pairs of populations were under similar selection pressures to one another, since the inclusion of outlier SNPs did not change the levels of genetic differentiation between these pairs. Georges Bank and Massachusetts Bay winter spawners had on average slightly higher pairwise FST values compared to most other comparisons, suggesting subtle neutral differentiation of these two populations.
The weighted mean value is given above and the confidence interval is given below.
PCA and DAPC provided much less evidence for genetic differentiation among groups at putatively neutral SNPs in comparison to the analyses with the full dataset (Fig 4). PCA showed some limited patterns of neutral differentiation, with the Massachusetts Bay winter individuals clustering to the left and to a lesser extent the Ipswich Bay winter individuals clustering to the lower right portion of the plot (Fig 4A). DAPC suggested weak separation of Georges Bank, Massachusetts Bay spring, and Massachusetts Bay winter, with individuals from Ipswich Bay spring and winter and eastern GoM grouping together in the central portion of the plot (Fig 4B). Removal of the two outlier individuals from the Massachusetts Bay dataset caused most of the differentiation of Massachusetts Bay winter to disappear (Fig 4C and 4D) and caused the pattern in the DAPC to largely collapse, with a weak separation of Georges Bank and the two spring spawning populations remaining and broad overlap of the other three groups.
a) PCA with all individuals; b) DAPC with all individuals. Removal of two outlying individuals from Massachusetts Bay winter largely collapses all groups in both c) PCA and d) DAPC. The percentage of the variation explained by each principle component is shown on the axis labels for the PCAs, and the cumulative percentage of variation explained by the discriminant functions is shown on the inset graphs for the DAPCs. The number of principle components retained in b) was 24 and in d) was 13, as assessed using cross-validation.
The lack of clear differentiation at neutral loci was also apparent with our Structure analyses (Figure H in S1 Appendix). The Evanno method for determining the ‘true’ value of K suggested K = 2 was most likely for analyses both with and without location priors. At values of K from two to four, the outlier individuals in Massachusetts Bay winter were assigned to a separate cluster, while other populations did not show clear patterns of differentiation (Figure H in S1 Appendix). This is not surprising since the neutral genetic differentiation among these populations is subtle and Structure is known to perform poorly when FST is less than 0.05 .
Genome-wide patterns of differentiation
To investigate patterns of differentiation at the genome-wide scale, we plotted the locus-specific FST values for all pairwise comparisons between groups (with winter and spring spawners from Ipswich Bay and Massachusetts Bay grouped together by spawning season given the genetic similarity of these groups). Our full dataset included 62 SNPs from the inversion on LG7 and 61 from the inversion on LG12, making it possible to see the influence that these inversions were having on the inferred population structure (Fig 5). The inversion on LG 7 appeared to segregate between winter and spring spawners, between spring spawners and eastern GoM cod, and to a lesser extent between winter spawners and Georges Bank spawners. The inversion on LG 12 appeared to segregate between spring spawners and Georges Bank spawners, between spring spawners and eastern GoM cod, and to a lesser extent between winter and spring spawners. Since our dataset only included 28 SNPs from the inverted region on LG2, we were not able to make inference about the contributions of this inversion. We also found evidence for adaptive variation outside of the known chromosomal inversions: one SNP on LG 18 showed very high (> 0.8) FST between spring spawners and all other groups, and a small number of SNPs on LG 11 also showed elevated FST in all comparisons except between Georges Bank versus eastern GoM cod. Overall, comparisons between spring spawners and other groups tended to show the most elevated FST values, while winter spawners seemed somewhat similar to Georges Bank and eastern GoM cod, the comparison between the latter two showing the overall lowest levels of differentiation.
To further investigate genetic differentiation in each putatively inverted region on LG 2, 7, and 12, we extracted the SNPs within each LD block from our full SNP dataset using the boundaries determined by Barney et al., . We then repeated our PCA using those subsets of SNPs.
The patterns from the inverted regions on LG 7 and 12 showed three well separated clusters in the PCAs, with PC1 explaining 36.4% and 25.6% of the variation, respectively (Fig 6A and 6B). These clusters likely corresponded to the homozygous collinear ‘genotype’, the heterozygous genotype (where individuals have one copy of the inversion and one copy of the collinear region), and the homozygous inverted genotype. ‘Genotyping’ individuals for the inversion based on which cluster they fell into showed that winter and spring spawning cod have different frequencies of the inversions. At LG 7, spring spawners were mostly homozygotes for one form of the inverted region (left-most cluster) and heterozygotes (middle cluster), while winter spawners had higher frequencies of the alternative homozygous state (right-most cluster) and many heterozygotes. Eastern GoM individuals showed a similar pattern to the winter spawners, while Georges Bank were found across all three clusters. At LG 12, spring spawners were mainly homozygous for one form of the inversion (right-most cluster) and heterozygotes (middle cluster), while winter spawners were mainly homozygous for the alternate allele or heterozygotes. Georges Bank and eastern GoM individuals showed similar haplotype frequencies as the winter spawners.
PCAs for each of the LD blocks on a) LG 7, b) LG 12, and c) LG 2. The percentage of the variance explained by each axis is shown on the axes labels.
The PCA based on the 28 SNPs from the inversion on LG 2 was not as informative (Fig 6C), which is not surprising given the small number of SNPs and the overall lack of elevated FST values in this region (Fig 5). However, three weakly defined clusters were visible on the PCA that related to each of the homozygous states and the heterozygous state for the inversion. Groups were not well differentiated with these 28 SNPs on LG 2, with members of each group found in all three clusters, but some separation of spring spawners primarily in the left and central cluster and winter spawners primarily within the right and central clusters could be seen. Georges Bank individuals grouped primarily in the right and central clusters, similar to the winter spawners, with fewer individuals in the left cluster. No additional clustering was apparent when additional PCs were plotted for any of the LD blocks.
From our genome-wide analysis of population structure, we found evidence for differentiation among three major spawning groups: cod that spawn in the spring in Ipswich and Massachusetts Bays, cod that spawn in the winter in the same bays, and cod spawning on the northeast peak of Georges Bank. The spring spawning cod in the western GoM appeared to be the most genetically differentiated from the other spawning populations. Levels of genetic differentiation were low overall (small magnitude FST) and primarily driven by loci that are putatively under selection, the majority of which were located in the genomic regions on LG 7 and 12 containing chromosomal inversions known to be divergent among other cod populations in both the Northeast and Northwest Atlantic [51–54,58,73]. Little genetic structure was apparent from neutral markers alone among these populations, but Georges Bank and the spring spawning western GoM population displayed some weak neutral differentiation. Divergence at outlier loci despite gene flow has been observed among Atlantic cod populations in other parts of its range . Our findings suggest that selection is likely maintaining differentiation at adaptive loci, particularly among spring spawning cod that may experience divergent selection pressures due to their different spawning season.
The low neutral genetic differentiation suggests that there is some level of ongoing gene flow among these groups of cod, especially between the western GoM bays within the same spawning season. Small but significant levels of genetic differentiation are common in marine fishes as a result of their high dispersal potential and large effective population sizes [1,88], which can make interpretations challenging. Even low levels of neutral genetic differentiation can be biologically meaningful , but translating these low levels of genetic differentiation into useful estimates of demographic connectivity is not easy . Evidence from individually tagged cod in the GoM supports a hypothesis of there being connectivity among the western GoM spawning locations within the same season, as individuals tagged on the primary Massachusetts Bay spring-spawning site have been recaptured or detected via acoustic telemetry in the vicinity of other spawning sites during the spawning season . Spawning site fidelity is high [90–92], but cod may spawn in multiple nearby locations within a season .
Slightly greater neutral differentiation between winter and spring spawning groups and between the western GoM spawners and Georges Bank spawners suggest that gene flow is less prevalent among these groups. This finding is consistent with results from tagging studies, which reveal very little movement of GoM cod to areas outside of this region [36,38,90,93], supporting the view of demographic independence of cod in the GoM and Georges Bank. Movement patterns of spring and winter spawning cod in the western GoM have been studied extensively and suggest dispersal from the spawning grounds after spawning and return to the spawning grounds during distinct spawning seasons [35,90,93], consistent with genetic differentiation of these seasonal spawning groups.
Our findings agree with previous work that found significant genetic differentiation primarily driven by loci putatively under selection among spawning groups in the Gulf of Maine [23,28]. Barney et al.,  localized this genetic differentiation to the inversions on LG 2, 7 and 12, although they focused their investigation on these genomic regions and did not investigate differentiation elsewhere in the genome. Contrary to the findings of Barney et al., , but consistent with Wirgin et al.,  and Kovach et al., , we found that spring spawning cod are the most genetically differentiated of the spawning populations in the western Gulf of Maine and Georges Bank. This inconsistency between the studies is likely a result of small sample sizes, because Barney et al.,  sampled ten individuals from each spawning group, so their estimates of the inversion frequencies in each population may not have been precise.
In addition to differences in the frequency of inversions on LG 7 and 12 among the groups studied, we found several outlier SNPs with elevated FST on LG 11 and 18, with the former segregating among all three primary groups (winter spawners, spring spawners, and Georges Bank), and the latter segregating primarily between the spring spawning groups and the others. Our small sample of genomic SNPs was insufficient to perform a genome-wide scan for selection because a much higher density of SNPs and an estimate of genome-wide linkage disequilibrium would be necessary . Therefore, further investigations into these outlier loci should be conducted with a denser SNP dataset, which may uncover additional targets of selection elsewhere in the genome. In a similar fashion, our ability to detect differences in the frequency of the inversion on LG 2 was limited by the small number of SNPs we recovered in this region. This inversion was previously found to segregate between winter and spring spawning cod in Massachusetts Bay and to contain a number of genes that may have temperature-mediated functions . Our dataset uncovered a relatively large number of SNPs within the inversions on LG 7 and 12, which showed frequency deviations at these sites. Differences in the frequencies of these inversions among populations are likely the result of divergent selection because cod are thought to have large effective population sizes [95–97] resulting in weak genetic drift.
There is a growing body of evidence from a variety of taxa that chromosomal inversions can capture multiple adaptive alleles governing complex traits and therefore act as ‘supergenes’ (e.g. Anopheles mosquitoes , three-spine sticklebacks , Heliconious butterflies ). In Atlantic cod, the inversions on LG 1, 2, 7, and 12 that segregate as biallelic loci between cod populations and ecotypes [49,50,54,55,57,58], may also be acting as supergenes housing suites of genes with adaptive phenotypic effects. The inversion on LG 7 has previously been linked to variation in temperature in both the Northwest and Northeast Atlantic , variation in salinity and oxygen concentrations in the Baltic Sea , and with migratory and resident ecotypes in the Northeast Atlantic that also experience different temperature regimes [57,102]. The inversion on LG 12 has also been linked to temperature [53,55,56] and with inshore and offshore populations on the Skagerrak coast of Norway that show variation in their tendency to migrate . Barney et al.,  identified 306 and 407 genes within the inverted regions on LG 7 and 12, respectively, but they did not find significant enrichment for any specific genomic pathways, and so how these regions could be adaptive under different temperature regimes is currently unknown.
Consistent with findings from other Atlantic cod populations, temperature and/or offshore and vertical (depth) movement patterns may also be selective pressures that maintain the polymorphisms that we observed. Spring spawning cod in the western GoM differed from the winter GoM and the Georges Bank spawners in the frequencies of the inversions on both LG 7 and 12 (and LG 2 according to REF 24), suggesting they experienced divergent selective pressures from those groups. Winter spawning cod in the western GoM appeared to be under similar selective pressures as cod spawning on Georges Bank, based on more similar frequencies of the inversions (although they are not fully connected by gene flow at neutral loci, supporting the divergence of these spawning groups). Divergent temperature-associated selection pressures could act at different life stages (e.g., egg development, larval growth, juvenile settlement, adult migration). For example, eggs that are spawned in the spring in the western GoM will experience warmer waters (based on surface temperature) than those spawned in the winter at the same location and on Georges Bank (Georges Bank cod also spawn in the winter). However, depth differences during juvenile settlement may also be a factor; spring-spawned juveniles settle at greater depths (up to 80 m) and a narrower range of temperatures (<10°C) compared with winter-spawned juveniles, which settle in shallower inshore waters (< 30 m) at a greater range of temperatures (5–15°C; ref ; M. Dean Massachusetts Division of Marine Fisheries, pers. comm.). Lastly, temperature differences may also be associated with adult movement patterns if populations differ in their offshore migration patterns and their affinity for deeper, cooler waters outside of the spawning season. Spring spawning cod in Massachusetts Bay move to offshore feeding grounds in the summer and fall, and then overwinter in deep (>150 m) offshore basins . Further work identifying the inversion haplotypes of the different spawning aggregations and associating them with life history data is needed to fully understand the effects of temperature, depth, or migration-driven selection on these spawning populations.
The eastern GoM cod caught by the sentinel fishery showed some of the lowest levels of neutral genetic differentiation from spawning cod from Ipswich Bay, but this differentiation increased when outlier loci were included. If we assumed that the cod caught in the eastern GoM were residents to that area, then this would suggest that there may be a high level of connectivity between Ipswich Bay and the eastern GoM, and that selection may result in the putatively adaptive allelic frequency differences that we observed, favoring the genotypes of the winter-spawned Ipswich cod. One possibility is that cod from Ipswich Bay could have moved in to fill the vacant niche after spawning aggregations largely disappeared from the eastern GoM . Alternatively, connectivity may have pre-dated the loss of spawning aggregations in the eastern GoM and may represent a long-term phenomenon. However, evidence to support high connectivity between eastern and western GoM is lacking. Cod tagged in the western GoM have rarely been recaptured outside of the western GoM [35,90], supporting the idea of high spawning site fidelity to the western GoM. An alternative scenario is that the eastern GoM cod we sampled were from small, remnant spawning aggregations that have persisted in the region but have gone undetected. Differences at outlier loci on LG11 between both the winter and spring spawning western GoM cod and the eastern GoM cod point toward a separate origin of these cod, perhaps as a remnant of the historically larger aggregations.
Compared to Ipswich Bay, the eastern GoM cod showed a higher level of neutral differentiation with Georges Bank cod, suggesting there could be demographic separation of these two groups. However, putatively adaptive differentiation was low, which may suggest that similar selection pressures act on cod from eastern GoM and Georges Bank. Perhaps cod in these areas experience a similar temperature regime resulting from similar oceanographic conditions, or they may exhibit similar habitat preferences. However, we cannot confirm these hypotheses given the little information we have about these eastern GoM fish and the lack of known spawning in this region. Tagging studies conducted over the last 40 years have shown some migratory connectivity between Georges Bank, the Scotian Shelf, and the Bay of Fundy [34,35]. Few cod with conventional tags have been recaptured in eastern GoM , but the few recaptures likely reflect the relatively low intensity of fishing in the eastern GoM. Fishery-independent inferences of cod movement from geolocation of archival tags suggest relatively high connectivity between western and eastern Gulf of Maine is possible during certain seasons . Therefore, cod caught by the sentinel fishery in the eastern GoM could have been recent migrants from adjacent areas.
Our results do not provide a clear conclusion to the source of the cod caught by the sentinel fishery in the eastern GoM, because signals of neutral and adaptive genetic variation do not align to consistently identify them with either the western GoM or Georges Bank spawning populations. It is also possible that the cod caught in this area today are a mixed stock, resulting from foraging migrations of cod that spawn in other areas. Indeed, the samples we analyzed in this study were collected during a 3-month period (June–August), during which cod could have moved into this area from different locations (including possibly a mix of Canadian and U.S. waters). Further studies using genetic data from historical samples (e.g., otolith collections) pre-dating the collapse of the eastern GoM stocks are needed to definitively test the hypothesis about demographic separation of this population in the past. These studies could indicate whether there has been long-term connectivity between eastern GoM and other areas, which would be informative for investigating potential mechanisms for rebuilding of cod in the eastern GoM.
Implications for stock structure
The patterns of fine-scale population structure that we found have been confirmed by several studies [23,24,28] and are in contrast to the current two stock strategy for assessment and management. This strategy groups all eastern and western GoM cod into one stock, and cod from southern New England to Georges Bank into a second, southern stock (Fig 1). After conducting an interdisciplinary review, Zemeckis et al.,  suggested that there may be five separate subpopulations of cod in US waters. These subpopulations correspond to: 1) an eastern Georges Bank population that spawns in the winter and early spring; 2) a northern spring coastal complex that spawns in the spring in Ipswich Bay, Massachusetts Bay, and Bigelow Bight; 3) a southern complex that includes winter spawners from Ipswich Bay, Massachusetts Bay, and Jeffreys Ledge, early-spring spawners from Stellwagan Bank, and winter spawners from Nantucket Shoals, the Great South Channel, southern New England and the Middle Atlantic; 4) a mid-coast GoM population; and 5) an eastern GoM population. The latter two subpopulations were defined by Ames  and are now severely depleted, with no spawning observed in these areas for the last two decades. Our results provide further evidence consistent with the metapopulation model proposed by Zemeckis et al.,  and show that the current GoM stock encompasses multiple, genetically distinct sub-populations that likely harbor different local adaptations. The lack of recognition in the current management system of the two seasonally distinct western GoM subpopulations and a third possibly distinct subpopulation in eastern GoM may put this biocomplexity at risk.
Despite high dispersal potential in marine species at either larval or adult stages, local adaptation is common (e.g., purple sea urchin, Strongylocentrotus purpuratus , grey reef sharks, Carcharhinus amblyrhynchos , hamlet fish, Hypoplectrus spp., ) and local adaptation has often been linked to temperature (e.g. American lobster, Homarus americanus, , sea scallops, Placopecten magellanicus, , northern shrimp, Pandalus borealis, ). Conserving locally adapted populations should be a high priority for fishery managers, because the functional aspect of adaptive genetic diversity is arguably more important to preserve than neutral genetic diversity . Adaptive genetic diversity is also one of the components that determines the adaptive capacity of species and populations , because species with higher genetic diversity (both neutral and adaptive) have a greater probability of surviving rapid environmental change through adaptation. Atlantic cod are likely to be severely affected by the rapid warming of the Gulf of Maine, which has warmed faster than 99% of the global ocean . Therefore, preserving adaptive genetic diversity, especially if it is linked to thermal regime, is likely going to be critical for this species into the future.
Preserving population diversity or biocomplexity is also critical for fishery stability and resilience by providing diverse responses to the environment [5,15,16]. Portfolio effects, whereby a high diversity of discrete populations reduces the variability in a stock [5,11], can provide stability and resilience in fisheries . The loss of Atlantic cod spawning components  has likely impacted the productivity and stability of the population and the fishery targeting this resource. Furthermore, the adaptive differences among spawning populations suggest that the potential for severely depleted populations to recover through migration from other spawning populations could be limited, because immigrants may have low fitness due to lack of suitable local adaptations . Accordingly, preservation of the remaining biocomplexity of Atlantic cod within the Gulf of Maine and Georges Bank should be considered in future fishery management efforts.
Funding was provided by National Oceanic and Atmospheric Administration Saltonstall-Kennedy Grant number NA14NMF4270014 and the National Oceanic and Atmospheric Administration College Sea Grant program through the New Hampshire Sea Grant project R/SFA-3 (#NA14OAR4170083). We are grateful to Yong Chen, Anna Henry, Mattie Rodrigue, John Carlucci, Robin Alden, and Ted Ames for cod samples from the sentinel fishery, and the Massachusetts Department of Marine Fisheries sample collection in the Gulf of Maine. We also thank Ted Ames and Karen Alexander for their inspiration and insightful conversations about cod in the eastern Gulf of Maine and David Berlinsky for introducing us to the spring and winter spawners in the Gulf of Maine. Kris Wojtusik and Katie Shink provided assistance with sample preparation in the laboratory.
- 1. Hauser L, Carvalho GR. Paradigm shifts in marine fisheries genetics: ugly hypotheses slain by beautiful facts. Fish Fish. 2008;9: 333–362. https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1467-2979.2008.00299.x
- 2. Ward RD, Woodwark M, Skibinski DOF. A comparison of genetic diversity levels in marine, freshwater, and anadromous fishes. J Fish Biol. Wiley/Blackwell (10.1111); 1994;44: 213–232.
- 3. DeWoody JA, Avise JC. Microsatellite variation in marine, freshwater and anadromous fishes compared with other animals. J Fish Biol. Wiley/Blackwell (10.1111); 2000;56: 461–473.
- 4. Waples RS, Gaggiotti O. What is a population? An empirical evaluation of some genetic methods for identifying the number of gene pools and their degree of connectivity. Mol Ecol. 2006;15: 1419–1439. pmid:16629801
- 5. Hilborn R, Quinn TP, Schindler DE, Rogers DE. Biocomplexity and fisheries sustainability. Proc Natl Acad Sci. 2003;100: 6564–6568. pmid:12743372
- 6. Therkildsen NO, Hemmer-Hansen J, Hedeholm RB, Wisz MS, Pampoulie C, Meldrup D, et al. Spatiotemporal SNP analysis reveals pronounced biocomplexity at the northern range margin of Atlantic cod Gadus morhua. Evol Appl. 2013;6: 690–705. pmid:23789034
- 7. Ruzzante DE, Mariani S, Bekkevold D, André C, Mosegaard H, Clausen LAW, et al. Biocomplexity in a highly migratory pelagic marine fish, Atlantic herring. Proc R Soc B Biol Sci. 2006;273: 1459–1464. pmid:16777738
- 8. Funk WC, McKay JK, Hohenlohe PA, Allendorf FW. Harnessing genomics for delineating conservation units. Trends Ecol Evol. 2012;27: 489–496. pmid:22727017
- 9. Crandall KA, Bininda-Emonds ORP, Mace GM, Wayne RK. Considering evolutionary processes in conservation biology. Trends Ecol Evol. Elsevier Current Trends; 2000;15: 290–295. pmid:10856956
- 10. de Guia APO, Saitoh T. The gap between the concept and definitions in the Evolutionarily Significant Unit: the need to integrate neutral genetic variation and adaptive variation. Ecol Res. Springer-Verlag; 2007;22: 604–612.
- 11. Schindler DE, Hilborn R, Chasco B, Boatright CP, Quinn TP, Rogers LA, et al. Population diversity and the portfolio effect in an exploited species. Nature. 2010;465. pmid:20520713
- 12. Nicotra AB, Beever EA, Robertson AL, Hofmann GE, O’Leary J. Assessing the components of adaptive capacity to improve conservation and management efforts under global change. Conserv Biol. 2015;29: 1268–1278. pmid:25926277
- 13. Harrisson KA, Pavlova A, Telonis-Scott M, Sunnucks P. Using genomics to characterize evolutionary potential for conservation of wild populations. Evol Appl. Wiley/Blackwell (10.1111); 2014;7: 1008–1025. pmid:25553064
- 14. Stachowicz JJ, Whitlatch R, Osman R. Species Diversity and Invasion Resistance in a Marine Ecosystem. Science (80-). American Association for the Advancement of Science; 1999;286: 1577–1579. pmid:10567267
- 15. Kerr LA, Cadrin SX, Secor DH. Simulation modelling as a tool for examining the consequences of spatial structure and connectivity on local and regional population dynamics. ICES J Mar Sci. Oxford University Press; 2010;67: 1631–1639.
- 16. Kerr LA, Cadrin SX, Secor DH. The role of spatial dynamics in the stability, resilience, and productivity of an estuarine fish population. Ecol Appl. Wiley-Blackwell; 2010;20: 497–507. pmid:20405802
- 17. Robichaud D, Rose GA. Migratory behaviour and range in Atlantic cod: inference from a century of tagging. Fish Fish. Wiley/Blackwell (10.1111); 2004;5: 185–214.
- 18. Wise JP. Cod groups in the New England area. Fish Bull. 1963;63: 189–203. https://www.st.nmfs.noaa.gov/spo/FishBull/63-1/wise.pdf
- 19. Ames EP. Atlantic cod stock structure in the Gulf of Maine. Fish Res. 2004;29: 10–28.
- 20. Zemeckis DR, Martins D, Kerr LA, Cadrin SX. Stock identification of Atlantic cod (Gadus morhua) in US waters: an interdisciplinary approach. ICES J Mar Sci. 2014;71: 1490–1506.
- 21. Goode GB. The fisheries and fishery industries of the United States: 1851–1896. Washington, DC: Government Printing Office; 1887. https://archive.org/details/fisheriesfishery52goodrich
- 22. Alexander KE, Leavenworth WB, Cournane J, Cooper AB, Claesson S, Brennan S, et al. Gulf of Maine cod in 1861: historical analysis of fishery logbooks, with ecosystem implications. Fish Fish. Wiley/Blackwell (10.1111); 2009;10: 428–449.
- 23. Kovach AI, Breton TS, Berlinsky DL, Maceda L, Wirgin I. Fine-scale spatial and temporal genetic structure of Atlantic cod off the Atlantic coast of the USA. Mar Ecol Prog Ser. 2010;410: 177–195.
- 24. Barney BT, Munkholm C, Walt DR, Palumbi SR. Highly localized divergence within supergenes in Atlantic cod (Gadus morhua) within the Gulf of Maine. BMC Genomics. 2017;18: 1–14.
- 25. Ruzzante DE, Taggart CT, Cook D. A nuclear DNA basis for shelf- and bank-scale population structure in northwest Atlantic cod (Gadus morhua): Labrador to Georges Bank. Mol Ecol. Wiley/Blackwell (10.1111); 1998;7: 1663–1680.
- 26. Lage C, Kuhn K, Kornfield I. Genetic differentiation among Atlantic cod (Gadus morhua) from Browns Bank, Georges Bank, and Nantucket Shoals. Fish Bull. National Marine Fisheries Service; 2004;102: 289–298.
- 27. Weiss E, Sanderson M, Gerlach G. Population structure and essential fish habitat mapping of western Georges Bank cod. Final Rep Coop Res Partners’ Program, Northeast Reg Off Natl Mar Fish Serv. 2005.
- 28. Wirgin I, Kovach AI, Maceda L, Roy NK, Waldman J, Berlinsky DL. Stock Identification of Atlantic Cod in U.S. Waters Using Microsatellite and Single Nucleotide Polymorphism DNA Analyses. Trans Am Fish Soc. 2007;136: 375–391.
- 29. Begg GA, Hare JA, Sheehan DD. The role of life history parameters as indicators of stock structure. Fish Res. Elsevier; 1999;43: 141–163.
- 30. Tallack S. Regional growth estimates of Atlantic cod, Gadus morhua: Applications of the maximum likelihood GROTAG model to tagging data in the Gulf of Maine (USA/Canada) region. Fish Res. Elsevier; 2009;99: 137–150.
- 31. Berien P, Sibunka J. Distribution patterns of fish eggs in the U.S. NE Continental Shelf Ecosystem, 1977–1987. NOAA Tech Rep NMFS. 1999;145: 310.
- 32. Churchill JH, Jeffrey R, Chen C. Processes controlling retention of spring-spawned Atlantic cod (Gadus morhua) in the western Gulf of Maine and their relationship to an index of recruitment success. Fish Oceanogr. Wiley/Blackwell (10.1111); 2011;20: 32–46.
- 33. Lough RG, Hannah CG, Berrien P, Brickman D, Loder JW, Quinlan JA. Spawning pattern variability and its effect on retention, larval growth and recruitment in Georges Bank cod and haddock. Mar Ecol Prog Ser. Inter-Research Science Center; 2006;310: 193–212.
- 34. Hunt J, Stobo W, Almeida F. Movement of Atlantic cod, Gadus morhua, tagged in the Gulf of Maine area. Fish Bull. 1999;97.
- 35. Tallack S. Proceedings from a workshop to identify future research priorities for cod tagging in the Gulf of Maine, 12 February, 2009, Gulf of Maine Research Institute. Northeast Fish Sci Cent Ref Doc. 2009; https://repository.library.noaa.gov/view/noaa/3697
- 36. Tallack S. Stock identification applications of conventional tagging data for Atlantic cod in the Gulf of Maine. In (pp. 1–15). Proc from 2nd Int Symp Adv Fish Tagging Marking Tech Auckland, NZ Am Fish Soc. 2011; 1–15.
- 37. Perkins H, Chenoweth S, Langton R. The Gulf of Maine Atlantic cod complex, patterns of distribution and movement of the Sheepscot Bay substock. Bull Natl Res Inst Aquac. 1997;3: 101–107.
- 38. Howell WH, Morin M, Rennels N, Goethel D. Residency of adult Atlantic cod (Gadus morhua) in the western Gulf of Maine. Fish Res. Elsevier; 2008;91: 123–132.
- 39. Zemeckis DR, Hoffman WS, Dean MJ, Armstrong MP, Cadrin SX. Spawning site fidelity by Atlantic cod (Gadus morhua) in the Gulf of Maine: Implications for population structure and rebuilding. ICES Journal of Marine Science. 2014. pp. 1356–1365.
- 40. Sherwood GD, Grabowski JH. Exploring the life-history implications of colour variation in offshore Gulf of Maine cod (Gadus morhua). ICES J Mar Sci. Oxford University Press; 2010;67: 1640–1649.
- 41. Conroy CW, Calvert J, Sherwood GD, Grabowski JH. Distinct responses of sympatric migrant and resident Atlantic cod phenotypes to substrate and temperature at a remote Gulf of Maine seamount. ICES J Mar Sci. 2018;75: 122–134.
- 42. Kerr LA, Cadrin SX, Kovach AI. Consequences of a mismatch between biological and management units on our perception of Atlantic cod off New England. ICES J Mar Sci. Oxford University Press; 2014;71: 1366–1381.
- 43. NEFSC (Northeast Fisheries Science Centre). Operational assessment of 19 northeast groundfish stocks, updated through 2016. 2017. https://www.nefsc.noaa.gov/publications/crd/crd1717/
- 44. Pershing J, Alexander MA, Hernandez CM, Kerr LA, Le Bris A, Mills KE, et al. Slow adaptation in the face of rapid warming leads to collapse of the Gulf of Maine cod fishery. Science (80-). 2015;350: 809–812. pmid:26516197
- 45. Saba VS, Griffies SM, Anderson WG, Winton M, Alexander MA, Delworth TL, et al. Enhanced warming of the Northwest Atlantic Ocean under climate change. J Geophys Res Ocean. John Wiley & Sons, Ltd; 2016;121: 118–132.
- 46. Thomas AC, Pershing AJ, Friedland KD, Nye JA, Mills KE, Alexander MA, et al. Seasonal trends and phenology shifts in sea surface temperature on the North American northeastern continental shelf. Elem Sci Anth. 2017;5: 48.
- 47. Klein ES, Smith SL, Kritzer JP. Effects of climate change on four New England groundfish species. Reviews in Fish Biology and Fisheries. 2017. pp. 317–338.
- 48. Hare JA, Morrison WE, Nelson MW, Stachura MM, Teeters EJ, Griffis RB, et al. A vulnerability assessment of fish and invertebrates to climate change on the northeast u.s. continental shelf. PLoS One. 2016;11: e0146756. pmid:26839967
- 49. Sinclair-Waters M, Bradbury IR, Morris CJ, Lien S, Kent MP, Bentzen P. Ancient chromosomal rearrangement associated with local adaptation of a postglacially colonized population of Atlantic Cod in the northwest Atlantic. Mol Ecol. 2017; 1–13.
- 50. Kirubakaran TG, Grove H, Kent MP, Sandve SR, Baranski M, Nome T, et al. Two adjacent inversions maintain genomic differentiation between migratory and stationary ecotypes of Atlantic cod. Mol Ecol. 2016;25: 2130–2143. pmid:26923504
- 51. Barth JMI, Berg PR, Jonsson PR, Bonanomi S, Corell H, Hemmer-Hansen J, et al. Genome architecture enables local adaptation of Atlantic cod despite high connectivity. Mol Ecol. 2017;26: 4452–4466. pmid:28626905
- 52. Bradbury IR, Hubert S, Higgins B, Bowman S, Borza T, Paterson IG, et al. Genomic islands of divergence and their consequences for the resolution of spatial structure in an exploited marine fish. Evol Appl. Wiley/Blackwell (10.1111); 2013;6: 450–461. pmid:23745137
- 53. Bradbury IR, Hubert S, Higgins B, Borza T, Bowman S, Paterson IG, et al. Parallel adaptive evolution of Atlantic cod on both sides of the Atlantic Ocean in response to temperature. Proc R Soc B Biol Sci. 2010;277: 3725–3734. pmid:20591865
- 54. Sodeland M, Jorde PE, Lien S, Jentoft S, Berg PR, Grove H, et al. “Islands of Divergence” in the Atlantic Cod Genome Represent Polymorphic Chromosomal Rearrangements. Genome Biol Evol. Oxford University Press; 2016;8: 1012–1022. pmid:26983822
- 55. Berg PR, Star B, Pampoulie C, Bradbury I, Bentzen P, Hutchings J, et al. Trans-oceanic genomic divergence of Atlantic cod ecotypes is associated with large inversions. Heredity (Edinb). 2017;119: 418–428. pmid:28930288
- 56. Therkildsen NO, Hemmer-Hansen J, Als TD, Swain DP, Morgan MJ, Trippel EA, et al. Microevolution in time and space: SNP analysis of historical DNA reveals dynamic signatures of selection in Atlantic cod. Mol Ecol. Wiley/Blackwell (10.1111); 2013;22: 2424–2440. pmid:23551301
- 57. Hemmer-Hansen J, Nielsen EE, Therkildsen NO, Taylor MI, Ogden R, Geffen AJ, et al. A genomic island linked to ecotype divergence in Atlantic cod. Mol Ecol. 2013;22: 2653–2667. pmid:23611647
- 58. Berg PR, Star B, Pampoulie C, Sodeland M, Barth JMI, Knutsen H, et al. Three chromosomal rearrangements promote genomic divergence between migratory and stationary ecotypes of Atlantic cod. Sci Rep. 2016;6: 23246. pmid:26983361
- 59. Rodrigue M. Developing a collaborative research program to evaluate fine-scale groundfish dynamics in eastern Maine. Electronic Theses and Dissertations. University of Maine; 2017. https://digitalcommons.library.umaine.edu/etd/2810
- 60. Henry AM. Assessing social resilience of Maine fishing communities and developing a longline groundfish survey. Electronic Theses and Dissertations. University of Maine; 2013. p. 130.
- 61. Burnett J, O’Brien L, Mayo RK, Darde JA, Bohan M. Finfish maturity sampling and classification schemes used during Northeast Fisheries Center bottom trawl surveys, 1963–89. NOAA Tech Memo NMFS-F/NEC-76. 1989;76: 14.
- 62. Baird NA, Etter PD, Atwood TS, Currey MC, Shiver AL, Lewis ZA, et al. Rapid SNP discovery and genetic mapping using sequenced RAD markers. PLoS One. 2008;3: e3376. pmid:18852878
- 63. Andrews S. FastQC: a quality control tool for high throughput sequence data. 2010.
- 64. Catchen JM, Amores A, Hohenlohe P, Cresko W, Postlethwait JH. Stacks: building and genotyping Loci de novo from short-read sequences. G3 Genes, Genomes, Genet. 2011;1: 171–82. pmid:22384329
- 65. Catchen J, Hohenlohe PA, Bassham S, Amores A, Cresko WA. Stacks: An analysis tool set for population genomics. Mol Ecol. 2013;22: 3124–3140. pmid:23701397
- 66. Jiang H, Lei R, Ding S-W, Zhu S. Skewer: a fast and accurate adapter trimmer for next-generation sequencing paired-end reads. BMC Bioinformatics. BioMed Central; 2014;15: 182. pmid:24925680
- 67. Eaton D. ipyrad: interactive assembly and analysis of RADseq data sets. 2015.
- 68. Tørresen OK, Star B, Jentoft S, Reinar WB, Grove H, Miller JR, et al. An improved genome assembly uncovers prolific tandem repeats in Atlantic cod. BMC Genomics. BMC Genomics; 2017;18: 1–23.
- 69. Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXiv Prepr arXiv13033997. 2013;
- 70. Danecek P, Auton A, Abecasis G, Albers CA, Banks E, DePristo MA, et al. The variant call format and VCFtools. Bioinformatics. 2011;27: 2156–2158. pmid:21653522
- 71. Lischer HEL, Excoffier L. PGDSpider: an automated data conversion tool for connecting population genetics and genomics programs. Bioinformatics. 2012;28: 298–299. pmid:22110245
- 72. Foll M, Gaggiotti O. A genome-scan method to identify selected loci appropriate for both dominant and codominant markers: a Bayesian perspective. Genetics. 2008;180: 977–993. pmid:18780740
- 73. Bradbury IR, Bowman S, Borza T, Snelgrove PVR, Hutchings JA, Berg PR, et al. Long Distance Linkage Disequilibrium and Limited Hybridization Suggest Cryptic Speciation in Atlantic Cod. Yin T, editor. PLoS One. Public Library of Science; 2014;9: e106380. pmid:25259573
- 74. Bhatia G, Patterson N, Sankararaman S, Price AL. Estimating and interpreting FST: the impact of rare variants. Genome Res. Cold Spring Harbor Laboratory Press; 2013;23: 1514–21. pmid:23861382
- 75. Hudson RR, Slatkin M, Maddison WP. Estimation of levels of gene flow from DNA sequence data. Genetics. 1992;132.
- 76. Di Gaetano C, Fiorito G, Ortu MF, Rosa F, Guarrera S, Pardini B, et al. Sardinians Genetic Background Explained by Runs of Homozygosity and Genomic Regions under Positive Selection. Chaubey G, editor. PLoS One. Public Library of Science; 2014;9: e91237. pmid:24651212
- 77. Jombart T. adegenet: a R package for the multivariate analysis of genetic markers. Bioinformatics. 2008;24: 1403–1405. pmid:18397895
- 78. Jombart T, Ahmed I. adegenet 1.3–1: new tools for the analysis of genome-wide SNP data. Bioinformatics. 2011;27: 3070–3071. pmid:21926124
- 79. Team RC. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2016.
- 80. Jombart T, Devillard S, Balloux F. Discriminant analysis of principal components: a new method for the analysis of genetically structured populations. BMC Genet. 2010;11: 1.
- 81. Pritchard JK, Stephens M, Donnelly P. Inference of population structure using multilocus genotype data. Genetics. 2000;155: 945–959. pmid:10835412
- 82. Earl DA. STRUCTURE HARVESTER: a website and program for visualizing STRUCTURE output and implementing the Evanno method. Conserv Genet Resour. 2012;4: 359–361.
- 83. Evanno G, Regnaut S, Goudet J. Detecting the number of clusters of individuals using the software STRUCTURE: A simulation study. Mol Ecol. 2005;14: 2611–2620. pmid:15969739
- 84. Jakobsson M, Rosenberg NA. CLUMPP: a cluster matching and permutation program for dealing with label switching and multimodality in analysis of population structure. Bioinformatics. 2007;23: 1801–1806. pmid:17485429
- 85. Rosenberg NA. DISTRUCT: a program for the graphical display of population structure. Mol Ecol Notes. 2004;4: 137–138.
- 86. Weir BS, Cockerham CC. Estimating F-statistics for the analysis of population structure. Evolution (N Y). 1984; 1358–1370.
- 87. Latch EK, Dharmarajan G, Glaubitz JC, Rhodes OE. Relative performance of Bayesian clustering software for inferring population substructure and individual assignment at low levels of population differentiation. Conserv Genet. Springer Netherlands; 2006;7: 295–302.
- 88. Waples RS. Separating the wheat from the chaff: Patterns of genetic differentiation in high gene flow species. Journal of Heredity. Oxford University Press; 1998. pp. 438–450.
- 89. Knutsen H, Olsen EM, Jorde PE, Espeland SH, André C, Stenseth NC. Are low but statistically significant levels of genetic differentiation in marine fishes “biologically meaningful”? A case study of coastal Atlantic cod. Mol Ecol. Wiley/Blackwell (10.1111); 2011;20: 768–783. pmid:21199035
- 90. Zemeckis DR, Liu C, Cowles GW, Dean MJ, Hoffman WS, Martins D, et al. Seasonal movements and connectivity of an Atlantic cod (Gadus morhua) spawning component in the western Gulf of Maine. ICES J Mar Sci J du Cons. 2017; fsw190.
- 91. Siceloff L, Howell WH. Fine-scale temporal and spatial distributions of Atlantic cod (Gadus morhua) on a western Gulf of Maine spawning ground. Fish Res. 2013;141: 31–43.
- 92. Dean MJ, Hoffman WS, Zemeckis DR, Armstrong MP. Fine-scale diel and gender-based patterns in behaviour of Atlantic cod (Gadus morhua) on a spawning ground in the Western Gulf of Maine. ICES J Mar Sci. Oxford University Press; 2014;71: 1474–1489.
- 93. Loehrke JL. Movement patterns of Atlantic cod (Gadus morhua) spawning groups off New England. University of Massachusetts, Dartmouth, MA. 2013.
- 94. Lowry DB, Hoban S, Kelley JL, Lotterhos KE, Reed LK, Antolin MF, et al. Responsible RAD: Striving for best practices in population genomic studies of adaptation. Mol Ecol Resour. 2017;17: 366–369. pmid:28382730
- 95. Therkildsen NO, Nielsen EE, Swain DP, Pedersen JS. Large effective population size and temporal genetic stability in Atlantic cod (Gadus morhua) in the southern Gulf of St. Lawrence. Fraser D, editor. Can J Fish Aquat Sci. 2010;67: 1585–1595.
- 96. Poulsen NA, Nielsen EE, Schierup MH, Loeschcke V, Grønkjær P. Long-term stability and effective population size in North Sea and Baltic Sea cod (Gadus morhua). Mol Ecol. Wiley/Blackwell (10.1111); 2006;15: 321–331. pmid:16448403
- 97. Herbinger CM, Doyle RW, Taggart CT, Lochmann SE, Brooker AL, Wright JM, et al. Family relationships and effective population size in a natural cohort of Atlantic cod (Gadus morhua) larvae. Can J Fish Aquat Sci. 1997;54: 11–18.
- 98. Love RR, Steele AM, Coulibaly MB, Traore SF, Emrich SJ, Fontaine MC, et al. Chromosomal inversions and ecotypic differentiation in Anopheles gambiae: the perspective from whole-genome sequencing. Mol Ecol. Wiley/Blackwell (10.1111); 2016;25: 5889–5906. pmid:27759895
- 99. Jones FC, Grabherr MG, Chan YF, Russell P, Mauceli E, Johnson J, et al. The genomic basis of adaptive evolution in threespine sticklebacks. Nature. 2012;484: 55–61. pmid:22481358
- 100. Joron M, Frezal L, Jones RT, Chamberlain NL, Lee SF, Haag CR, et al. Chromosomal rearrangements maintain a polymorphic supergene controlling butterfly mimicry. Nature. Nature Publishing Group; 2011;477: 203–206. pmid:21841803
- 101. Berg PR, Jentoft S, Star B, Ring KH, Knutsen H, Lien S, et al. Adaptation to low salinity promotes genomic divergence in Atlantic Cod (Gadus morhua L.). Genome Biol Evol. 2015;7: 1644–1663. pmid:25994933
- 102. Righton DA, Andersen KH, Neat F, Thorsteinsson V, Steingrund P, Svedäng H, et al. Thermal niche of Atlantic cod Gadus morhua: Limits, tolerance and optima. Mar Ecol Prog Ser. 2010;420: 1–13.
- 103. Howe AB, Correia SJ, Currier TP, King J, Johnston R. Spatial Distribution of Ages 0 and 1 Atlantic Cod (Gadus morhua) off the Eastern Massachusetts Coast: 1978–1999: Relative to" Habitat Area of Special Concern". Massachusetts Div Mar Fish Tech Rep TR-12. 2002;
- 104. Pespeni MH, Palumbi SR. Signals of selection in outlier loci in a widely dispersing species across an environmental mosaic. Mol Ecol. 2013;22: 3580–3597. pmid:23802552
- 105. Momigliano P, Harcourt R, Robbins WD, Jaiteh V, Mahardika GN, Sembiring A, et al. Genetic structure and signatures of selection in grey reef sharks (Carcharhinus amblyrhynchos). Heredity (Edinb). Nature Publishing Group; 2017;119: 142–153. pmid:28422134
- 106. Picq S, McMillan WO, Puebla O. Population genomics of local adaptation versus speciation in coral reef fishes (Hypoplectrus spp, Serranidae). Ecol Evol. Wiley-Blackwell; 2016;6: 2109–2124. pmid:27099711
- 107. Benestan L, Quinn BK, Maaroufi H, Laporte M, Clark FK, Greenwood SJ, et al. Seascape genomics provides evidence for thermal adaptation and current-mediated population structure in American lobster (Homarus americanus). Mol Ecol. Wiley/Blackwell (10.1111); 2016;25: 5073–5092. pmid:27543860
- 108. Van Wyngaarden M, Snelgrove PVR, DiBacco C, Hamilton LC, Rodríguez-Ezpeleta N, Jeffery NW, et al. Identifying patterns of dispersal, connectivity and selection in the sea scallop, Placopecten magellanicus, using RADseq-derived SNPs. Evol Appl. Wiley/Blackwell (10.1111); 2017;10: 102–117. pmid:28035239
- 109. Jorde PE, Søvik G, Westgaard J-I, Albretsen J, André C, Hvingel C, et al. Genetically distinct populations of northern shrimp, Pandalus borealis, in the North Atlantic: adaptation to different temperatures as an isolation factor. Mol Ecol. Wiley/Blackwell (10.1111); 2015;24: 1742–1757. pmid:25782085
- 110. Peterson DA, Hilborn R, Hauser L. Local adaptation limits lifetime reproductive success of dispersers in a wild salmon metapopulation. Nat Commun. Nature Publishing Group; 2014;5: 3696. pmid:24739514