Genetic response to human‐induced habitat changes in the marine environment: A century of evolution of European sprat in Landvikvannet, Norway

Abstract Habitat changes represent one of the five most pervasive threats to biodiversity. However, anthropogenic activities also have the capacity to create novel niche spaces to which species respond differently. In 1880, one such habitat alterations occurred in Landvikvannet, a freshwater lake on the Norwegian coast of Skagerrak, which became brackish after being artificially connected to the sea. This lake is now home to the European sprat, a pelagic marine fish that managed to develop a self‐recruiting population in barely few decades. Landvikvannet sprat proved to be genetically isolated from the three main populations described for this species; that is, Norwegian fjords, Baltic Sea, and the combination of North Sea, Kattegat, and Skagerrak. This distinctness was depicted by an accuracy self‐assignment of 89% and a highly significant F ST between the lake sprat and each of the remaining samples (average of ≈0.105). The correlation between genetic and environmental variation indicated that salinity could be an important environmental driver of selection (3.3% of the 91 SNPs showed strong associations). Likewise, Isolation by Environment was detected for salinity, although not for temperature, in samples not adhering to an Isolation by Distance pattern. Neighbor‐joining tree analysis suggested that the source of the lake sprat is in the Norwegian fjords, rather than in the Baltic Sea despite a similar salinity profile. Strongly drifted allele frequencies and lower genetic diversity in Landvikvannet compared with the Norwegian fjords concur with a founder effect potentially associated with local adaptation to low salinity. Genetic differentiation (F ST) between marine and brackish sprat is larger in the comparison Norway‐Landvikvannet than in Norway‐Baltic, which suggests that the observed divergence was achieved in Landvikvannet in some 65 generations, that is, 132 years, rather than gradually over thousands of years (the age of the Baltic Sea), thus highlighting the pace at which human‐driven evolution can happen.


| INTRODUC TI ON
Humans have dramatically impacted the Earth's surface and promoted striking ecosystem and biodiversity alterations over the course of the last two centuries, hence becoming an evolutionary force of extraordinary influence (Albuquerque et al., 2018;Ceballos et al., 2015;Hooper et al., 2012). Human activities generate major pressures on habitats and organisms and are associated with evolutionary changes that can occur within tens of years, a phenomenon known as "contemporary evolution" (Besnier et al., 2014;Otto, 2018;Pelletier & Coltman, 2018;Stockwell et al., 2003). Human-driven evolution can happen at a pace and extent that is significantly higher than that of natural causes (Bull & Maron, 2016;Hendry et al., 2008;Palumbi, 2001;Therkildsen et al., 2019). Anthropogenic activities have altered and created novel niche spaces and species' responses to ecosystem alterations vary from avoidance to adaptation, including exploitation (Bull & Maron, 2016).
Humans are fundamentally changing connections within and among ecosystems over a wide range of spatial scales and habitat types, hence modifying the levels of connectivity (Crook et al., 2015).
Such changes can pose direct threats to communities, but may also create novel environments that influence the evolutionary trajectories of populations and species (Allendorf et al., 2012), and can alter the phenotypic landscapes of species by decreasing or increasing genetic diversity (Figure 1) . Many examples of contemporary evolution are associated with colonization events, species introductions, or invasions (Colautti & Lau, 2015;Johnston & Selander, 1964;Reznick & Ghalambor, 2001). Populations colonizing new environmental conditions can be exposed to novel selective forces that lead to adaptive divergence and differentiation from the original population (Björklund & Gustafsson, 2015;Hendry et al., 2002).
The construction of navigation canals is an example of human-facilitated connectivity between two previously isolated ecosystems (Galil et al., 2007). Canals can link marine and freshwater bodies allowing aquatic organisms to disperse to new areas and eventually colonize novel environments (Crook et al., 2015).
One such connectivity change took place in 1880, when the lake Landvikvannet (henceforth denoted as Landvik for abbreviation), on the southern Norwegian Skagerrak coast, was artificially connected to the adjacent marine fjord (Strandfjorden, Grimstad, Norway) by a 3 km long narrow canal (Reddal Canal). The construction of the canal, built to transport logs down to the dockyards by the sea as well as to drain the lake to increase the surface of arable land, lowered the water level in the lake by 3 m, turning the lake brackish as saltwater inflows over the tidal cycle while there is a continuous flux of freshwater from streams into the lake (Kanalkontoret, 1883). This human alteration drove changes in species assemblages, facilitating the colonization of marine species like the Atlantic herring (Clupea harengus) (Linnaeus, 1758) and European sprat, Sprattus sprattus (Linnaeus, 1758). Although it is unsure when these marine species colonized Landvik, the first sprat sample taken by the Institute of Marine Research dates back to 1999.
The European sprat is a small pelagic fish that is widely distributed from northern Norway to Morocco, the Baltic Sea, the northern Mediterranean basins, and the Black Sea (Debes et al., 2008).
Three geographically distinct genetic groups have been described with nuclear markers: (a) Norwegian fjords, (b) Baltic Sea, and (c) a wide-ranging component spanning the North Sea, Kattegat-Skagerrak in north to the Celtic Sea, and Bay of Biscay in south (Glover et al., 2011;Limborg, Hanel, et al., 2012;Limborg et al., 2009;Quintela et al., 2020). Furthermore, mitochondrial control region revealed two additional demes in the Mediterranean Sea, Gulf of Lyon, and Adriatic Sea (Debes et al., 2008). Differences found in candidate loci for divergent selection between the fresh-to brackish water Baltic Sea and fully marine populations suggest that local adaptation to low salinity is likely (Quintela et al., 2020), as has been shown in other Clupeid species such as the Atlantic herring in the Baltic Sea (Guo et al., 2016;Limborg, Helyar, et al., 2012), and the European anchovy (Engraulis encrasicolus Linnaeus, 1758) in the Adriatic (Ruggeri et al., 2016) and Tyrrhenian Seas (Catanese et al., 2017). The colonization of Landvik's brackish waters might have been possible due to the sprat's standing genetic variation allowing adaptation to a range of salinities, as conditions in Landvik partly resemble those in the Baltic Sea, the largest brackish water body in the world (Florian Berg, 2018).
The relatively recent colonization of Landvik by sprat provides an opportunity to study a contemporary evolution process, testing whether the creation of this new environment has promoted genetic differentiation from standing variation through ecological adaptation. This happens when barriers to gene flow evolve between populations due to divergent selection, with niche adaptation and competition as driving mechanisms (Bolnick, 2004;Schluter, 2000).
Landvik's salinity is similar to that of parts of the Baltic Sea, which thus allows the use of it as a replicate model to study parallel evolution and the role of the environment in ecologically driven speciation (Bailey et al., 2017;Bolnick et al., 2018).
To test for local adaptation and parallel evolution, we first characterized Landvik sprat with a suite of recently developed SNP markers and investigated the origin and connectivity of the lake population using a set of 42 geographically explicit samples, most of which were described in Quintela et al. (2020). Secondly, we investigated whether loci putatively under selection could be identified across these samples. Correlation between outlier loci and two environmental variables, salinity and temperature, was examined to test the potential role of selection in population divergence, and the possibility to identify genetic signals of parallel evolutionary change between Landvik and the Baltic Sea populations with respect to the marine populations. Spawners and embryos have been identified as the most temperature-sensitive stages in the life cycle of fish (Dahlke et al., 2020).

| Sampling and environmental data
Data about temperature and salinity corresponding to the average summer values for the period 2005-2012 were retrieved from NOAA database (National Oceanic and Atmospheric Administration). The depth at which measurements were chosen was 10 m for being relevant both for spawners and embryos (Table 1).

| DNA isolation and genotyping
DNA was extracted from fin clips stored in ethanol using the Qiagen DNeasy 96 Blood & Tissue Kit in 96-well plates, each of which contained two or more negative controls. All 45 samples were genotyped with the 91 SNPs for which protocols are described in their entirety in Quintela et al. (2020). In addition, a subset of 15 of the 45 samples was genotyped with eight microsatellite markers (see Table A1), as described in Glover et al. (2011

TA B L E 1 (Continued)
second set was to estimate genetic diversity through allelic richness, and hence, results derived of the microsatellite data will be mainly presented in Appendix 1.

| Statistical analysis
All statistical analyses were performed separately for SNPs and microsatellites. The observed (H o ) and unbiased expected heterozygosity (uH e ) as well as the inbreeding coefficient (F IS ) were computed for each sample with GenAlEx v6.1 (Peakall & Smouse, 2006). The genotype frequency of each locus and its direction (heterozygote deficit or excess) was compared with Hardy-Weinberg expectations (HWE) using the program GENEPOP 7 (Rousset, 2008), as was linkage disequilibrium (LD) between pairwise loci.
Landvik sprat were compared with the remaining collections using pairwise F ST (Weir & Cockerham, 1984) computed with ARLEQUIN v.3.5.1.2 (Excoffier et al., 2005). The Bayesian clustering approach implemented in STRUCTURE v. 2.3.4 (Pritchard et al., 2000), and conducted using the software ParallelStructure (Besnier & Glover, 2013), was used to identify genetic groups under a model assuming admixture and correlated allele frequencies without using population information as a prior. Ten runs with a burn-in period consisting of 100,000 replications and a run length of 1,000,000 MCMC iterations were performed for K = 1 to K = 7 clusters. To determine the number of genetic groups, STRUCTURE output was analyzed using two approaches: (a) the ad hoc summary statistic ΔK of Evanno et al. (2005), and (b) the four statistics (MedMed, MedMean, MaxMed, and MaxMean) both implemented in StructureSelector (Li & Liu, 2018). The ten runs for the selected Ks were then averaged with CLUMPP v.1.1.1 (Jakobsson & Rosenberg, 2007) using the FullSearch algorithm and the G′ pairwise matrix similarity statistic, and graphically displayed using barplots. Furthermore, the relationships between Landvik and the reference samples were examined using discriminant analysis of principal components, DAPC (Jombart et al., 2010) implemented in adegenet (Jombart, 2008), as well as with the principal coordinates analysis (PCoA) built using Nei (1978)'s genetic distance between pairs of populations with F I G U R E 2 Map of the sampling sites as well as detailed view of Landvikvannet. Codes and associated full names of sampling locations can be found in Two analytic approaches, BayeScan (Foll & Gaggiotti, 2008) and LOSITAN (Antao et al., 2008), were combined to detect loci deviating from neutral expectations and therefore reflecting either eventual selective responses or linkage disequilibrium with genes under divergent selection (Lewontin & Krakauer, 1973). In BayeScan, sample size was set to 10,000 and the thinning interval to 50. Loci with a posterior probability over 0.99, corresponding to a Bayes Factor > 2 (i.e., "decisive selection" (Foll & Gaggiotti, 2006)), were retained as outliers.
In LOSITAN, a neutral distribution of F ST with 1,000,000 iterations was simulated, with forced mean F ST at a significance level of 0.05 under an infinite allele model for SNPs and under a stepwise model for microsatellites. To avoid pseudo replication, outlier analyses were conducted using a random sample of 300 individuals from each of the four genetic clusters identified with STRUCTURE (after excluding southern distant outgroups). Analyses were performed either using jointly the four sets of samples or using subsets, as appropriate.
Adaptation to local environments often occurs through natural selection acting on a large number of loci, each having a weak phenotypic effect. LFMM, "latent factor mixed model" (Frichot et al., 2013), was used to assess whether salinity or water temperature could be a potential selective pressure driving local adaptation by identifying loci

F I G U R E 3 (Continued)
showing unusual associations with these environmental factors compared to the genetic background. Thus, the environmental information used corresponded to the season of the year where fish are at its most temperature-sensitive stages (Dahlke et al., 2020). This method, which has formerly proved to be efficient for a suite of scenarios of demographic history (Lotterhos & Whitlock, 2015;de Villemereuil et al., 2014), uses a linear mixed model to test for associations between genetic variation and environmental factors, while controlling for neutral genetic structure with (random) latent factors. Ten runs of LFMM were conducted using 1,000 sweeps for burn-in and 10,000 additional sweeps. The number of latent factors was set at K = 4 according to STRUCTURE outcome as suggested by Frichot et al. (2013). The corresponding z-scores of the ten replicates were combined following the recommendations described in Frichot and François (2015). First, the genomic inflation factor (λ) was obtained after computing the median of the squared (combined) z-scores for each K, divided by the median of the chi-square distribution with one degree of freedom. Finally, p-values were adjusted using the genomic inflation factor (λ), and false discovery rates were set using the Benjamini and Hochberg (1995) algorithm.
In addition, the relationship between genetic distance (F ST ) and each environmental factor was examined using Mantel (1967) tests F I G U R E 4 Origin of Landvik sprat: Neighbor-joining tree placing Landvik in context with the reference samples genotyped at 91 SNP loci (NJ tree using after removing the loci under positive selection can be found in Figure A3 in Appendix 1). To increase the resolution, analyses were performed after excluding the distant southern outgroups to investigate whether the correlations conformed the expectations of "Isolation by Environment" (IBE); that is, pattern in which genetic differentiation increases with environmental differences irrespective of geographic distance (Wang & Bradburd, 2014), as opposed to "Isolation by Distance" (IBD), which refers to the increase of genetic differentiation with geographic distance as a result of restricted gene flow and drift (Rousset, 1997;Slatkin, 1993;Wright, 1943 Allele frequency shifts at outlier loci are expected to be driven by selective responses toward strong ecological gradients leading to local adaptation, either due to directly associated genes or through hitchhiking (linkage) with associated genes (Gagnaire et al., 2015). Lowfrequency alleles can also reach high frequencies through allele surfing during population range expansion (Excoffier & Ray, 2008). Major allele frequencies (MAF) per sample were displayed through heatmaps and graphs as appropriate.
TA B L E 2 LFMM analysis for salinity and temperature (measured both in summer at 10 m depth) Note: Outlier analyses were performed using 300 randomly sampled individuals per genetic cluster instead of geographically explicit samples. Cells shaded in dark grey depict significant associations at LFMM after genomic inflation correction as well as candidates for positive selection according to LOSITAN after FDR correction (P(Simul F ST <sample F ST )) and BayeScan (log 10 (PO)). Cells shaded in light grey depict candidates to balancing selection. BayeScan did not detect deviations from neutrality in the pairwise comparisons. Flanking sequences of SNP loci were blasted against the GenBank and annotated genes in the vicinity of SNP markers were indicated as appropriate (empty cells depict no hit). All the annotated genes are Predicted for Clupea harengus.

| Genetic differentiation
All the approaches used to compare Landvik with the reference samples highlighted the distinctness of the lake sprat, putting also in evidence the low gene flow occurring between the brackish lake and   Table A3 in Appendix 1) whereas   showed that across all samples 86% of the individuals genotyped at SNPs were correctly assigned to their respective clusters (Table A4 in Appendix 1). The correct self-assignment per cluster ranged from 84% for Norwegian fjords to 100% in the Mediterranean Sea outgroups. In Landvik, 89% of the individuals were correctly assigned to the Landvik cluster, albeit with temporal differences: In 2012 and 2015, the percentage of correct assignment to cluster was of 96%-98%, respectively, whereas in 2019, it dropped to 60% as 21 individuals (i.e., 30% of the total in LAND19) were assigned to the Norwegian fjord cluster, 11 of them to the neighboring coastal samples (i.e., LYS, SORF, and TVE). As seven out of the 21 individuals showed an ancestry of q > 0.8 to the Norwegian fjord cluster, the hypothesis of them being migrants is plausible (see Figure A1 in

| Genetic relationships of Landvik sprat
The determination of the origin of Landvik sprat is hampered by the high levels of differentiation between this population and the reference samples. Pairwise F ST between Landvik and Norwegian fjord sprat were lower than any of the remaining comparisons hence revealing higher genetic relatedness than to brackish Baltic sprat (Table S1 in Supplementary File). Likewise, the Norwegian samples from LYS, TVE, and SORF, which are the geographically nearest to Landvik, were also the genetically closest ( Figure 3c). Furthermore, the NJ tree built with all the SNPs not only highlighted the distinctness of Landvik, but also showed that the lake sprat could stem from the sprat of the Norwegian fjords as Landvik shared a node in the phylogenetic tree with LYS ( Figure 4).

| Selection tests and detection of loci associated with environmental factors
Both outlier detection analyses (LOSITAN and BayeScan) as well as LFMM were conducted after excluding the southern distant groups due to their low sample size. LOSITAN reported four loci (4.4%) under positive selection, whereas BayeScan reported two, one of them in agreement with LOSITAN (Table 2). After genomic inflation correction, LFMM identified three loci associated with salinity and seven with temperature (Table 2), although the strength of the association was larger with salinity ( Figure 5a,b). Locus Ssp263 was associated with temperature as well as flagged as an outlier by both procedures, whereas locus Ssp210 was associated with salinity, flagged as outlier with BayeScan and marginally with LOSITAN. Locus Ssp248, the one F I G U R E 6 Major allele frequency for the four loci showing the largest differentiation in Landvikvannet compared to the Norwegian sprat. Allele frequency per sample was plotted versus the shortest water distance between each site and HOL (the northernmost one). The coloring pattern followed STRUCTURE barplot, that is, green for the Norwegian fjords and purple for Landvikvannet showing the strongest association with salinity (log 10 (P0) = 130.1), was annotated to the vicinity of a predicted protein kinase C epsilon in the herring genome whereas only one of the loci associated with temperature could be annotated, that is, locus Ssp319 to TOG array regulator of axonemal microtubules 1 also in the herring genome (  (Table 3) and revealed a similar pattern in low salinity waters (i.e., Landvik and the Baltic Sea) as opposed to marine waters (Norwegian fjords), which could suggest that Landvik sprat evolved from the Norwegian make-up to adapt to low salinity environments. Conversely, no temperature-related pattern for a similar process was obvious.
Finally, the major allele frequency of seven of the SNPs (Ssp253, Ssp321, Ssp260, Ssp268, Ssp213, Ssp251, and Ssp236) showed a remarkable drop in Landvik compared with the Norwegian fjord samples illustrating a change that could have happened in less than 132 years (see Figure 6 for four of them).

| D ISCUSS I ON
The brackish lake Landvik, created after excavating a 3 km long canal to the sea in 1880, represents a model system in which to investigate the potential for marine organisms to adapt to rapidly emerging new environments in the marine realm. Here, we showed that European sprat, a small pelagic marine fish, were able to colonize and develop a genetically highly distinct population in few decades.  (Quintela et al., 2020). The study also suggests signatures of contemporary adaptation to brackish habitat in Landvik sprat population, which represents a potential model system to study parallel evolution in comparison with the Baltic.

| Origin of the Landvik population
The relationship between genetic differentiation and shortest water distance revealed that samples from Landvik strongly departed from any geographically driven expectation (see Figure 3d), a situation also described Landvik is also inhabited by a taxonomically close species to sprat: the Atlantic herring. Landvik herring are considered as a self-sustaining and somewhat stationary population, characterized by slower growth, smaller length at maturity, lower vertebral count, shorter life span, higher relative fecundity, and divergent genetic profiles compared to the migratory oceanic herring in other parts of the Norwegian waters (Eggers, 2013;Eggers et al., 2014;Silva et al., 2013). Meristic trait vertebral count is often used as a population identifier in herring (e.g., Berg et al., 2017;Mosegaard & Madsen, 1996;Rosenberg & Palmén, 1981), and the observation that vertebral count in Landvik herring is similar to that in herring populations in the brackish Western Baltic Sea has led to the hypothesis that Landvik was colonized by low salinity adapted herring of Western Baltic Sea origin (Berg et al., 2019;Eggers et al., 2014). In addition, factorial crossing experiments performed at a range of salinities ranging from 6 to 35 revealed adaptation of herring populations to their native salinity conditions and also that adaption to salinity is transmitted to the offspring within the following generation (Berg et al., 2019). In contrast to herring, which rely on a benthic spawning habitat for depositing eggs, sprat is a pelagic spawner. As such, salinity may exert an even stronger selection pressure in sprat to avoid neutrally buoyant eggs from sinking into deeper anoxic water layers, as has for example been observed in Atlantic cod, Gadus morhua, adapted to spawning in brackish waters (Berg et al., 2015;Nissling et al., 1994). In both cod and herring, local adaptation is implied to be swift and ongoing, and working on standing genetic variation (e.g., Berg et al., 2015;Lamichhaney et al., 2012).
The origin of Landvik sprat is unknown, but, based on inference from herring, it would be conceivable that the lake could have also been colonized by fish from the Baltic Sea, already adapted to brackish waters, given the parallelism in the environmental conditions.
However, the analysis of Landvik in conjunction with the reference samples available in this study does not appear to support the hypothesis of the Baltic Sea as the source, but points toward founders from Norwegian fjordic sprat. This is particularly endorsed by the lower genetic differentiation between Norwegian sprat and Landvik, as well as by the neighbor-joining tree showing that Landvik sprat stems from the Norwegian cluster. Taking into consideration that NJ analyses are sensitive to outliers, the tree was recalculated after purging the candidate loci to positive selection detected by LOSITAN and BayeScan. The new NJ tree confirmed that the node from which Landvik sprat stem was the Norwegian sample, LYS (see Figure A3 in

| Adaptation as a consequence of brackish water colonization
Transitions from marine to freshwater habitats constitute dramatic shifts between adaptive habitats that have occurred not only on macroevolutionary time scales, but also in the recent past (Lee & Bell, 1999). During the last two centuries, humans have been changing connections between freshwater and marine ecosystems thus facilitating freshwater introductions (Crook et al., 2015). Drastic differences in salinity, parasites, competitors, and predators between marine and freshwater environments exert divergent selective pressures on the corresponding populations. Salinity showed strong associations with 3.3% of the loci analyzed in the present study. The genetic change experienced by the Norwegian sprat colonizing Landvik could be attributed to the strong directional selection driven by the low salinity in the lake. Rapid evolutionary changes are predicted in the face of strong selection following habitat shifts or environmental disturbances (Burke & Long, 2012;Kopp & Matuszewski, 2014;Losos et al., 1997;Turcotte et al., 2011), as happened in Landvik when the lake was artificially connected to the sea circa 150 years ago.
Similar processes have been documented in other species such as the threespine stickleback, which managed to evolve from oceanic ancestors to colonize the freshwater ponds that were formed during uplift caused by the Great Alaska Earthquake in 1964 (Lescak et al., 2015). Adaptation of a newly established resident population to the brackish environment often proceeds very fast, over the course of several decades (Barrett et al., 2008;Lescak et al., 2015;Marques et al., 2018). and North Adriatic Sea. These SNP outliers were also associated with salinity variability or involved in a critical stage of fertilization process.
The Baltic Sea was formed after the latest ice age, approximately 10,000-15,000 years ago, although its "ecological age" is circa 8,000 years (Lass & Matthäus, 2008). The combination of young geological age and contrasting environmental conditions to the surrounding oceans resulted in fast processes of adaptive evolution, which led to species living in the edge of their physiological tolerance (Ojaveer et al., 2010). The degree of differentiation between marine and brackish sprat is higher in the comparisons Norway fjords versus Landvik (mean F ST = 0.080, range 0.029-0.117) than in Norway fjords versus Baltic (mean F ST = 0.037, range 0.026-0.047).
Hypothesizing that the origin of Landvik sprat is the Norwegian fjord sprat invokes the possibility that the genetic changes occurring in the lake took place in a maximum of 65 generations (<132 years) rather than gradually over thousands of years. This hypothesis was also put forward in the threespine stickleback, which achieved in 50 years similar levels of divergence as populations that had diverged thousands of years ago (Lescak et al., 2015). Most likely, such rapid adaptation to a new environment cannot depend on de novo mutations and must rely primarily on standing genetic variation (Matuszewski et al., 2015;Dolph Schluter & Conte, 2009) as it has been demonstrated in the threespine stickleback populations (Terekhanova et al., 2019). The high speed of adaptation of Landvik population to the brackish environment probably has been possible by freshwater tolerance alleles being present in ancestral marine populations.

| Parallel or convergent evolution under similar selection pressures?
Adaptation to a radically different environment is likely to be genetically complex and to involve many loci, as it has been shown for other species (Terekhanova et al., 2019). Locus Ssp210, showing strong association with salinity, was reported to be a candidate to positive selection in the comparison between marine and brackish samples (in Norway vs. Baltic and marginally in the comparison Norway vs. Landvik) but not in the comparison between brackish environments (Lanvik vs. Baltic), which could suggest parallel or convergent evolution processes in Landvik and the Baltic Sea diverging from marine sprat. Despite the geographic proximity between Landvik and the Norwegian coastal sites, strong genetic divergence is found among those samples, probably due to differences in abiotic parameters (salinity) between habitats.
Parallel evolution under similar selection pressure has been widely observed in populations of the same species, for example, in bacterial experiments (Baym et al., 2016), recurrent adaptations of pathogens to their hosts (Collins & Didelot, 2018), and marine threespine sticklebacks that have independently colonized many freshwater habitats (Stuart et al., 2017). Baltic Sea and Landvik sprat populations could well be the results of parallel or convergent evolution (Arendt & Reznick, 2008), as it has been observed in lake and stream sticklebacks (Colosimo et al., 2005;Stuart et al., 2017). They dwell in discrete and divergent habitats, and are derived from ancestral marine populations, increasing the likelihood of them reusing similar ancestral genetic variants for adaptation. However, unlike the case of some stickleback populations that colonized lakes and streams after the last glaciation from the same ancestral population (Bolnick et al., 2018;Therkildsen et al., 2019), the history of Landvik  (Bassham et al., 2018). This population structure and history would provide many opportunities for parallel evolution when new freshwater populations were established from the marine stickleback population (Stern, 2013). In sticklebacks, that genetic parallelism is seen on finer geographic scales (Jones et al., 2012;Nelson & Cresko, 2018) but not globally, a pattern attributed to founder events and the loss of genetic diversity following colonization of the Atlantic (Fang et al., 2020). Landvik adaptation to brackish waters could have followed a similar pattern, where adaptation independent from Baltic populations has been a consequence of demographic forces of the founder event of the lake from the Norwegian coast populations. However, our study has strong limitations to disentangle whether it is a case of molecular parallelism or independent adaptation. Future genomic studies may help reveal the evolutionary history of the sprat and the molecular mechanisms involved in its different adaptations to brackish environments. Study the genetics of convergence can help shed light on fundamental questions in evolutionary biology, including whether natural selection is constrained and repeatable or instead characterized by many molecular paths to similar phenotypes.
The uniqueness of Landvik sprat suggests that an appropriate management should be considered for this population. A next step using whole-genome sequencing will allow to further explore intraclusters standing genetic variation as well as the origin of Landvik population. Parallel evolution in response to similar environmental pressures strongly suggests evolution by natural selection; however, the underlying genetic basis of this process is unclear. Landvik sprat thus provides an excellent opportunity for testing the genomic aspects of evolutionary repeatability.

ACK N OWLED G M ENTS
We would like to thank the technicians and the commercial fishermen that helped with the data collection. This study was primarily financed through funding from the Norwegian Department of Trade and Fisheries. Funding was also provided by the European Maritime and Fisheries Fund for the project "Fordeling af makrel, brisling og silde-bestande," 33113-B-16-065, and the Research Council of Norway project "CoastRisk" (299554).

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

DATA AVA I L A B I L I T Y S TAT E M E N T
The raw data for SNPs and microsatellites are available in https://doi.

F I G U R E A 3
Origin of Landvikvannet sprat: Neighbor-joining tree placing Landvikvannet in context with the 40 reference samples after removing the SNP loci candidates to directional selection flagged by LOSITAN and BayeScan. Names of sampling sites by numbers can be retrieved in Table 1