Genome architecture enables local adaptation of Atlantic cod despite high connectivity

Adaptation to local conditions is a fundamental process in evolution; however, mechanisms maintaining local adaptation despite high gene flow are still poorly understood. Marine ecosystems provide a wide array of diverse habitats that frequently promote ecological adaptation even in species characterized by strong levels of gene flow. As one example, populations of the marine fish Atlantic cod (Gadus morhua) are highly connected due to immense dispersal capabilities but nevertheless show local adaptation in several key traits. By combining population genomic analyses based on 12K single nucleotide polymorphisms with larval dispersal patterns inferred using a biophysical ocean model, we show that Atlantic cod individuals residing in sheltered estuarine habitats of Scandinavian fjords mainly belong to offshore oceanic populations with considerable connectivity between these diverse ecosystems. Nevertheless, we also find evidence for discrete fjord populations that are genetically differentiated from offshore populations, indicative of local adaptation, the degree of which appears to be influenced by connectivity. Analyses of the genomic architecture reveal a significant overrepresentation of a large ~5 Mb chromosomal rearrangement in fjord cod, previously proposed to comprise genes critical for the survival at low salinities. This suggests that despite considerable connectivity with offshore populations, local adaptation to fjord environments may be enabled by suppression of recombination in the rearranged region. Our study provides new insights into the potential of local adaptation in high gene flow species within fine geographical scales and highlights the importance of genome architecture in analyses of ecological adaptation.


| INTRODUCTION
Local adaptation characterizes populations that experience higher inherited fitness in their native habitat compared to members of other populations transferred to the same environment (Kawecki & Ebert, 2004). The degree of such ecological adaptation depends on the directional selection of advantageous traits and is counteracted by high connectivity and resulting homogenizing gene flow, implicating a limited potential for local adaptation in populations experiencing high gene flow (Dobzhansky, 1937;Mayr, 1942;Wright, 1931).
Although environmental adaptation can also involve gene expression-induced plastic responses such as morphological, physiological or behavioural changes, these occur without genotypic changes (Reusch, 2014;Via et al., 1995).
Most marine fish populations have traditionally been regarded as large panmictic entities with high connectivity due to the apparent lack of geographical barriers, high dispersal capabilities and slow genetic drift as a result of large effective population sizes (Allendorf, Hohenlohe, & Luikart, 2010;DeWoody & Avise, 2000;Waples & Gaggiotti, 2006). However, this assumption is challenged by an increasing number of genetic studies reporting high levels of local adaptation in marine fish populations despite substantial gene flow (Clarke, Munch, Thorrold, & Conover, 2010;Limborg et al., 2012;Milano et al., 2014;Nielsen et al., 2009;Therkildsen et al., 2013).
Simulation studies have demonstrated that local adaptation can arise in these situations through selection on tightly linked divergent alleles rather than on many single loci (Yeaman & Whitlock, 2011). In line with these expectations, the occurrence of linked alleles (e.g., in the form of chromosomal rearrangements) in locally adapted populations has been reported in studies addressing the genome architecture of fish species such as stickleback (Jones et al., 2012;Roesti, Kueng, Moser, & Berner, 2015), Atlantic herring (Lamichhaney et al., 2017;Martinez-Barrio et al., 2016) and Atlantic cod (Barney, Munkholm, Walt, & Palumbi, 2017;Berg et al., 2015Berg et al., , 2016Bradbury et al., 2013Bradbury et al., , 2014Hemmer-Hansen et al., 2013;Kirubakaran et al., 2016;Sodeland et al., 2016). Chromosomal rearrangements that physically combine genes residing within "supergene clusters" and promote adaptation in connected populations are well known in plants (Lowry & Willis, 2010), and insects (Cheng et al., 2012;Joron et al., 2011) and are widely discussed to play a role in speciation and evolution (Hoffmann & Rieseberg, 2008;Schwander, Libbrecht, & Keller, 2014). However, the relative importance of this mechanism in highly connected marine populations on small geographical scales remains poorly understood.
Atlantic cod (Gadus morhua Linnaeus, 1758) is a benthopelagic, high-fecundity, predatory fish of great commercial and ecological value occurring in a variety of habitats in the North Atlantic and hence constitutes an ideal model for the investigation of local adaptation. Molecular studies inferring the potential for local adaptation in Atlantic cod have a long history, which began with the discovery of adaptive allelic variation in the oxygen-binding protein haemoglobin (Sick, 1961) and the observation of a latitudinal gradient in the distribution of its isoforms (Sick, 1965; for recent reviews see Andersen (2012) and Ross, Behrens, Brander, Methling, and Mork (2013)). Since then, extensive research has contributed to the description of several genetically, phenotypically and behaviourally distinct populations occurring in a wide range of different ecosystems (Lilly et al., 2008). One of the best-investigated examples for apparent local adaptation despite high connectivity is the cooccurrence of two ecotypes of Atlantic cod, the migratory Northeast Arctic cod (NEAC) and the stationary Norwegian coastal cod (NCC), at the same spawning areas along the northern Norwegian coast (Neuenfeldt et al., 2013). While genetic differences between NEAC and NCC were already described in the 1960s (Moller, 1966), the mechanism maintaining differentiation despite ongoing gene flow is still a controversial subject Karlsen et al., 2013). The releases of two successive Atlantic cod genome assemblies (Star et al., 2011;Tørresen et al., 2017) facilitated the investigation of such mechanisms, revealing the presence of large chromosomal rearrangements likely permitting differentiation of these ecotypes despite ongoing gene flow (Berg et al., 2016;Kirubakaran et al., 2016).
On a much smaller spatial scale within the Skagerrak and Kattegat, two confined seas connecting the brackish Baltic Sea with the saline North Sea (Figure 1) Rogers, Olsen, Knutsen, & Stenseth, 2014;Sodeland et al., 2016). These coexisting fish are characterized by distinct lifestyles, with mobile oceanic (offshore) individuals foraging along the coast but possibly returning to North Sea or offshore Skagerrak spawning sites, and sedentary coastal individuals that remain close to the coast and local spawning sites at all times (Espeland et al., 2008;Knutsen et al., 2007;Neuenfeldt et al., 2013;Rogers et al., 2014). In line with this observation, differentiated Atlantic cod has been described between estuarine western Skagerrak fjords and offshore areas, as well as between individual fjords (Jorde, Knutsen, Espeland, & Stenseth, 2007;Knutsen, Jorde, Andr e, & Stenseth, 2003;Knutsen et al., 2011;Olsen et al., 2004). In these cases, the maintenance of differentiation has been associated with seascapes, coastal topography and hydrographic features such as salinity gradients (Ciannelli et al., 2010;Howe et al., 2010;Knutsen et al., 2011;Rogers et al., 2014). Limited migration of coastal cod (Espeland et al., , 2008, spawning site fidelity Skjaeraasen, Meager, Karlsen, Hutchings, & Fern€ o, 2011)

| Genotyping and filtering
DNA was extracted from muscle tissue using standard DNA extraction kits and normalized to 100 ng/ll as described elsewhere (Berg et al., 2015(Berg et al., , 2016 (Purcell et al., 2007) leading to a high-quality SNP set of 7,783 SNPs (for details see Appendix S1 and Table S2).
Variants were further filtered based on linkage to conform with the expectations of models employed in our genetic analyses: the correlation of allele frequencies (r 2 ) was calculated based on genotypic allele counts and 1,125 SNPs with an r 2 > 0.1 were excluded, resulting in a final data set of 6,658 unlinked SNPs.
A second data set including SNPs with detected linkage was generated to investigate the importance of large chromosomal rearrangements containing tightly linked SNPs that may play important roles in the divergence and adaptation of Atlantic cod (Bradbury et al., 2013;Hemmer-Hansen et al., 2013;Bradbury et al., 2014;Berg et al., 2015Berg et al., , 2016Sodeland et al., 2016;Kirubakaran et al., 2016;Barney et al., 2017; see section 2.5 below). All format conversions were either accomplished with in-house scripts, or using the software PGDSPIDER v2.0.8.0 (Lischer & Excoffier, 2012).

| Genetic differentiation
The population structure was investigated to delineate genetic differentiation and admixture of fjord samples and diverged populations, as well as to test for an isolation-by-distance (IBD) pattern as described earlier in the western North Atlantic cod (Beacham, Brat- 2001). Individual ancestry and the number of genetic clusters (K) were assessed using a hierarchical framework in STRUCTURE V2.3.2 (Pritchard, Stephens, & Donnelly, 2000) under the admixture model with correlated allele frequencies for closely related populations or highly migratory species (Falush, Stephens, & Pritchard, 2003). Five replicates of 100,000 (Monte Carlo Markov chain (MCMC) iterations (discarding the first 10,000 iterations as burn-in) were performed per model, each testing for K = 1 to K = 5. Convergence was confirmed by consistent results in all five replicates (see Table S3). In addition, principal component analyses were performed to display the largest variances in the genotype data (PCA, Appendix S2, Table S4).
Based on the probability that an individual has inherited a genetic marker from one of the two source populations North Sea and Kattegat, H was estimated using two cline parameters that describe the bias (a) and rate (b) of locus-specific introgression into an admixed genomic background (Gompert & Buerkle, 2012

| Biophysical connectivity modelling
Physical transport and connectivity of Atlantic cod eggs and larvae was quantified using a biophysical model to explore geneflow poten- poral and spatial differences in mortality rates is available. Larval drift simulations were repeated for 6 years (1995, 1996, 1998, 2000, 2001 and 2002), which represent negative, neutral and positive periods of the North Atlantic oscillation winter index (National Center for Atmospheric Research, 2015), as winter NAO is known to correlate well with variations in the circulation pattern (Marshall et al., 2001). To include as much variation as possible, results were based on the average of all spawning times, drift depths, drift durations and years with a total of~100M individual drift trajectories.
| 4455 Because of model domain limitations, the North Sea spawning areas did not include the Viking Bank east of Shetland. Connectivity between the spawning areas and the larval settlement areas (western and eastern Skagerrak, and Kattegat) was calculated as the proportion of eggs spawned in area i and settling as larvae in area j.
Furthermore, dispersal patterns from the spawning areas to western Skagerrak fjords were also assessed. As the spatial resolution of the biophysical model is not sufficient to represent the full geomorphology of the inner fjords, only the coastal waters close to the fjord mouths were considered (Soppekilen was not included as the connectivity model cannot resolve this site from the closely situated Hellefjord). The measure of connectivity of the biophysical model only predicts the probability per egg to be transported from i to j.
To obtain a relative estimate of the abundance of eggs reaching a settlement area, we also scaled the inferred connectivity with recent estimates of the spawning stock biomass (SSB, for calculations see Jonsson et al., 2016).

| Chromosomal rearrangements
The genomic architecture was examined to study the impact of large chromosomal rearrangements on population divergence and adaptation. The physical locations of SNPs within chromosomes (here: linkage groups; LGs) were inferred by mapping the flanking regions of all SNPs to the gadMor2 genome assembly (Tørresen et al., 2017) using BLAST v2.2.26+ (Camacho et al., 2009). Querying 10,913 flanking region pairs resulted in 10,804 blast hits, which were subsequently filtered according to the following quality thresholds: identity between query and hit >90%, E-value <1.0 9 10 À42 , and minimum length >100 bp. SNPs not meeting these criteria (n = 182) and SNPs on unplaced contigs (n = 526) were removed. Of the remaining SNPs, the exact positions were retrieved only for highquality SNPs included in this study (7,783, including linked SNPs, see above  Berg et al. (2016)) was inferred using PCA as implemented in the R package ADEGENET v1.4-1 (Jombart, 2008), similar to the approach described by Ma and Amos (2012). Bootstrapping (Efron, 1979; sample size 1,000,000) of individual genotypes was used to calculate the probability of an over-or underrepresentation of the presumably rearranged allele within sampling sites and within western (Tvedestrand, Soppekilen, Hellefjord, Grenland) and eastern (Iddefjord, Gullmarsfjord, Havstensfjord) fjords under the null hypothesis that the frequency of rearranged alleles within a population corresponds to its overall frequency across all populations. Sequential Bonferroni correction was applied to correct for multiple tests (Rice, 1988).  Table S3). According to Evanno's ΔK statistic, an ad hoc quantity  (Table S8). In general, all fjords possess admixed individuals, albeit at lower proportions in Tvedestrand (TVE 34%), Soppekilen (SOP 32.1%), Iddefjord (IDD 34.8%), Gullmarsfjord (GUL 48.9%) and Havstensfjord (HAV 41.7%). In these fjords, mechanical mixing of individuals with different ancestries seems to dominate the population structure.
Pairwise fixation indices (F ST ) were calculated to characterize the population structure between the different sampling sites and to assess the connectivity through isolation-by-distance (IBD) estimates. Isolation by distance was assessed using a Mantel test among fjord sampling sites only, or including the reference populations, and considering either direct geographical distances between sampling coordinates or least-cost paths restricted to marine and shelf areas.
However, no significant correlation was detected for any of the comparisons (Fig. S4). In summary, these results describe the presence of differentiated western Skagerrak fjord cod, and a mixed occurrence of North Sea and Kattegat cod within eastern Skagerrak fjords.

| Chromosomal rearrangements
Large genomic regions exhibiting strong linkage disequilibrium (LD) on several Atlantic cod chromosomes (linkage groups; LG) have recently been reported (Berg et al., 2015(Berg et al., , 2016Kirubakaran et al., 2016;Sodeland et al., 2016). Likely all of these regions represent large chromosomal inversions as suggested in previous studies (Berg et al.,2016;Sodeland et al.,2016), and empirically demonstrated for the linked region on LG1 (Kirubakaran et al., 2016). As our data set was filtered for LD using a strict filtering cut-off (r 2 > 0.1), most SNPs within the rearranged regions were removed due to strong signals of LD, with the remaining ones not influencing the genetic structure (Fig. S5). However, as these genomic regions have been suggested to carry genes responsible for local adaptation to low salinity, temperature and oxygen levels (Berg et al., 2015;Bradbury et al., 2010), these linked SNPs were used in separate analyses to investigate the occurrence and segregation of the chromosomal rearrangements between sampling sites. Our data revealed three of the four putative inversions previously described by Berg et al. (2015): LG2 (position 18,609,660,985;~5.05 Mbp),LG7 (position 13,622,181,520;~9.56 Mbp) and LG12 (position 426,445,150;~13.02 Mbp). The inversion on LG1 has so far exclusively been found in comparison with the Northeast Arctic cod (Berg et al., 2016;Kirubakaran et al., 2016), and was not detected in our data using the R package INVERSION. However, a comparison of SNPs within the linked region on LG1 in our data with the previously pub- LG2 within western fjords (p < .001), but not within eastern fjords. As the Oslofjord clustered with the North Sea group, it was excluded from this comparison; however, the rearranged allele on LG2 was also significantly overrepresented (p < .01) when the Oslofjord was included within the western fjords. In summary, these findings suggest that the particular genomic architecture of Atlantic cod may contribute to the potential for local adaptation to a low salinity environment.

| DISCUSSION
How local adaptation can be maintained in the face of gene flow is a long-standing question in evolutionary biology, which we are now beginning to understand owing to the profound advances in sequencing technology and genomic analysis tools (Tigano & Friesen, 2016 F I G U R E 4 Biophysical model of larval connectivity. (a) Modelled connectivity from four spawning areas to the eastern and western Skagerrak coasts, and to the Kattegat, expressed as the proportional larval supply. Larval supply was calculated as the probability of larval dispersal from the spawning areas scaled with the respective spawning stock biomass (SSB). (b) Modelled connectivity from the same four spawning areas to western Skagerrak fjords expressed as larval supply by scaling the dispersal probability with the respective SSB, and normalized to the target area. For TVE, HEL and GRE, only the fjord mouths were included in the model. Error bars show the standard deviation of simulations for six years (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002). For abbreviations of fjords see legend Figure 1 of chromosomal rearrangements in high geneflow species. Marine organisms provide ideal models to study this question, owing to their varied habitats and the lack of physical barriers. By combining genomic analyses of ecologically distinct Atlantic cod populations with biophysical modelling of dispersal, we were not only able to unravel cryptic population structure and detect ecologically differentiated populations, but also identified chromosomal rearrangements as a potential mechanism enabling local adaptation despite high connectivity.

| Western Skagerrak fjords possess locally differentiated Atlantic cod despite high connectivity and a mix of North Sea and Kattegat cod
The ecological peculiarity of the low-saline Baltic Sea and the transition zone connecting it with the saline North Sea have led to the evolution of unique linages (Johannesson & Andr e, 2006). Nevertheless, based on unlinked SNPs, the overall population differentiation of Atlantic cod within this area was weak, as also shown in earlier studies and explained by large effective population sizes and high gene flow (Knutsen et al., 2011;Nielsen, Grønkjaer, Meldrup, & Paulsen, 2005). Comparatively strong differentiation was detected between North Sea/English Channel/Skagerrak and Kattegat/western Baltic samples, reflecting the geographical separation (Figure 1) as well as a separation resulting from adaptation to low salinity as shown previously for Atlantic cod, but also many other species of the eastern Baltic Sea (Berg et al., 2015;Johannesson & Andr e, 2006;Lamichhaney et al., 2012;Sj€ oqvist, Godhe, Jonsson, Sundqvist, & Kremp, 2015). However, no genetic differentiation was detected within these strongly separated North Sea-like and western Balticlike groups (Appendix S3).
Contrary to these well defined populations, the eastern Skagerrak fjords appeared to be composed of a mix between North Sealike and western Baltic-like individuals, indicating that these fjords are part of the distributional area of the two major evolutionary units detected in this study. These fjords may experience larval recruitment through a strong influx of central North Sea water into the Skagerrak, as well as less-saline Kattegat water entering along the coast Danielssen et al., 1997;Jonsson et al., 2016;Knutsen et al., 2004;Stenseth et al., 2006). In agreement with these predominant ocean currents, a large fraction of individuals from the eastern Skagerrak fjords appeared to be of North Sea origin (Figure 2), while our biophysical model suggested greater larval connectivity with the Kattegat and western Baltic ( Figure 5). However, the model did not include the North Sea Viking bank spawning ground which has significantly increased its contribution during the last decades , suggesting that the influence of the North Sea spawning areas to the eastern Skagerrak is larger than shown in our modelling. We did not detect genetically differentiated individuals that would be indicative for a distinct fjord population in eastern Skagerrak fjords, although differentiation between Atlantic cod larvae inside and outside Gullmarsfjord was previously found (Øresland & Andr e, 2008). It is unknown if recent reductions in abundance along the eastern Skagerrak coast (Sved€ ang & Bardon, 2003;Sved€ ang & Svenson, 2006) indicate the loss or severe decimation of a genetically differentiated population in this region.
In contrast, the western Skagerrak fjord samples included varying levels of genetically differentiated individuals that clustered neither with the North Sea-like nor the western Baltic-like group (Figure 2b), indicative of the existence of a local western Skagerrak coastal or fjord cod population(s). The existence of such local populations is also supported by the biophysical model results, which explained a large fraction of larval supply by local recruitment (Figure 4). Local fjord cod has previously also been assumed to exist at the northern Norwegian coast (Jørstad & Naevdal, 1989;Myksvoll, Jung, Albretsen, & Sundby, 2014), and differentiation between fjord, coastal or oceanic cod has been shown in two closely related gadiids, the Pacific cod (Gadus macrocephalus) and the polar cod (Boreogadus saida) (Cunningham, Canino, Spies, & Hauser, 2009;Madsen, Nelson, Fevolden, Christiansen, & Praebel, 2015). Fjord systems represent semi-enclosed ecosystems where water exchange is restricted by a narrow connection with the outer sea, often further reduced by a tall entrance sill, thus creating an inner estuarine circulation (Howe et al., 2010). Such conditions have been shown to hamper gene flow as a result of stationary behaviour with reduced adult migration and restricted egg and larval dispersal (Bergstad, Jørgensen, Knutsen, & Berge, 2008;Ciannelli et al., 2010;Espeland et al., 2007Espeland et al., , 2008Jung et al., 2012;Knutsen et al., 2007;Rogers et al., 2014). Consequently, the strongest genetic differentiation and the largest fraction of local western Skagerrak fjord individuals was found in the particularly isolated Hellefjord (Molvaer, Green, & Baalsrud, 1978) and Grenland fjord (Danielssen & Føyn, 1973) (  (Lilly et al., 2008), an environmental flexibility that likely required the acquisition of locally adaptive traits. It has recently been described that such adaptations, especially in highly connected organisms like oceanic fish, can arise through the segregation of chromosomal rearrangements, where recombination is suppressed and important functional genes are inherited together (Feder, Egan, & Nosil, 2012;Thompson & Jiggins, 2014;Tigano & Friesen, 2016). While empirical evidence for this theory is still scarce, it is well supported by studies on stickleback (Jones et al., 2012;Roesti et al., 2015). Recently, haplotype blocks associated with ecological adaptation were also detected in the Atlantic herring, but it is unclear if inversions are the causative mechanism (Lamichhaney et al., 2017;Martinez-Barrio et al., 2016). In contrast, a series of recent studies employing genomewide data to dissect Atlantic cod population differentiation, discovered exceptionally large chromosomal rearrangements that are likely to be inversions on several linkage groups (LGs), which were suggested to play a major role for the adaptive abilities of Atlantic cod (Barney et al., 2017;Berg et al., 2015Berg et al., , 2016Bradbury et al., 2013Bradbury et al., , 2014Hemmer-Hansen et al., 2013;Kirubakaran et al., 2016;Sodeland et al., 2016). These recent studies, including this study, therefore contribute remarkable examples in the marine environment to a growing body of literature identifying chromosomal rearrangements and inversions as an important mechanism to maintain contrasting ecotypes in intermixing populations (Cheng et al., 2012;Hoffmann & Rieseberg, 2008;Joron et al., 2011;Lowry & Willis, 2010).
For example, adaptation to low-saline and hypoxic environments as occurring in the Baltic Sea strongly depends on the ability for osmoregulation and effective oxygen management (Andersen et al., 2009;Berg et al., 2015). Berg et al. (2015) (Nielsen, Hansen, Ruzzante, Meldrup, & Grønkjaer, 2003)) than to the North Sea population. In addition, we found a significant overrepresentation of the rearranged LG2 allele in the Hellefjord and Grenland fjord samples ( Figure 5a), an allelic shift that has recently also been described between oceanic and coastal cod groups (Sodeland et al., 2016).
Both fjords have high freshwater influx, causing a low-saline surface layer above oceanic water with 25-30& salinity (Danielssen & Føyn, 1973;Molvaer et al., 1978), comparable to salinity gradients in the Kattegat/western Baltic (Madsen & Højerslev, 2009 Westin, 1997; for a recent review see H€ ussy, 2011), a mechanism that for example prevents lethal sinking of the eggs to the hypoxic deeper layers in the Baltic Sea. In contrast, the eggs of marine Atlantic cod populations are neutrally buoyant at salinities of~33& (Thorsen, Kjesbu, Fyhndr, & Solemdal, 1996). Similar to Baltic cod, eggs of fjord cod are neutrally buoyant in the low-saline water layers of fjords, which not only prevents sinking of the eggs to hypoxic layers, but also retains the eggs inside the sheltered fjord area (Ciannelli et al., 2010;Espeland et al., 2007;Jung et al., 2012;Knutsen et al., 2007). Egg buoyancy can be regulated by the in-and efflux of solutes (Reading et al., 2012), and many SNPs in or close to genes coding for membrane trafficking proteins have been identified within the rearranged region on LG2 (Berg et al., 2015). This accumulation of adaptive variation could be explained by diversifying selection shaping the rearranged region in the likely absence of recombination between the alleles. In ecosystems where regulation of egg buoyancy provides an evolutionary advantage, an increase in the frequency of the rearrangement might be expected.
In addition to our samples from Hellefjord and Grenland fjord, our € Oresund sample from the western Baltic also shared a significant overrepresentation of the rearranged allele on LG2, which occurs at very high frequency in eastern Baltic cod (Berg et al., 2015). However, our Belt Sea and Kattegat samples did not show an increased occurrence of the rearranged LG2 allele although the genetic structure analyses suggested genetic similarity between the Kattegat and western Baltic samples, indicative for additional adaptive variation outside the large rearrangements. Interestingly, the rearranged LG12 allele was found to be significantly overrepresented in our North Sea and Oslofjord samples, with high occurrences also in the eastern Skagerrak sample (Figure 5c). Concordantly, this allele was recently found to occur at higher frequency in oceanic compared to coastal Atlantic cod populations and was suggested to play a role in ecological adaptation (Sodeland et al., 2016). It has previously also been associated with an adaptation to temperature (Berg et al., 2015;Bradbury et al., 2010), which could thus be relevant with regard to survival and abundance of Atlantic cod in the face of global warming (Drinkwater, 2005). However, similar to the Kattegat/western Baltic samples, which shared most genetic variation but showed a distinct pattern in the occurrence of the rearranged LG2 allele, the North Sea, Oslofjord, Skagerrak and English Channel samples were not distinguishable based on SNPs outside the rearranged regions, but showed a distinct distribution of the rearranged LG12 allele. This contrast between the genomewide profile that rather reflects connectivity and geography, and the chromosomal rearrangements that seem to cluster according to environment, indicates that despite the high gene flow between Atlantic cod populations important genes under adaptive divergent selection likely reside within rearranged regions.

| Significance and summary of the study
Because of their relatively higher fitness in their native habitat compared to introduced populations, locally adapted populations are often irreplaceable once vanished (Kawecki & Ebert, 2004;Reiss, Hoarau, Dickey-Collas, & Wolff, 2009). Human activity has led to the collapse of several fish stocks (Myers, Hutchings, & Barrowman, 1996;Pinsky, Jensen, Ricard, & Palumbi, 2011) and populations of Atlantic cod regionally suffer from overexploitation and population decline (Bartolino et al., 2012;Bonanomi et al., 2015;Sved€ ang & Bardon, 2003;Sved€ ang & Svenson, 2006), causing predator-prey shifts and imbalance of sensible ecosystems (Baden, Emanuelsson, Pihl, Svensson, & Aberg, 2012;€ Ostman et al., 2016). Thus, priorities are high to clarify the potential and occurrence of local adaptation in such high gene flow species, as well as to improve our understanding of the genetic mechanism for adaptation to conserve genetic resources in a globally changing world.
Our study showed that: 1.) the here described North Sea, Kattegat/western Baltic and western Skagerrak fjord cod genotypes most likely correspond to the previously identified oceanic and coastal ecotypes, respectively, thus shedding light on the long-standing question whether local fjord ecotypes exist and 2.) western Skagerrak fjord cod, despite high connectivity with the North Sea, may possess adaptations facilitating a life in a low salinity environment similar to Atlantic cod from the Baltic Sea. The genes encoding these adaptations are suggested to partially reside in large chromosomal rearrangements, regions that due to their reduced recombination are known to promote adaptive population divergence (Feder & Nosil, 2009;Kirkpatrick & Barton, 2006;Thompson & Jiggins, 2014).
In contrast, no locally differentiated fjord cod was detected in the eastern Skagerrak fjords, supporting the absence or suspected loss of local populations along the Swedish coast (Sved€ ang & Bardon, 2003). We thus emphasize the importance of taking genome architecture into account when characterizing ecological adaptation, particularly for species characterized by high gene flow.

ACKNOWLEDG EMENTS
We thank Mariann Arnyasi, Matthew P. Kent and Sigbjørn Lien (Nor- The nomenclature of linkage groups in this study follows Hubert, Higgins, Borza, and Bowman (2010).

AUTHOR CONTRIBU TI ON
The study was conceived and designed by C.