Abstract
Adaptive genetic divergence occurs when selection imposed by the environment causes the genomic component of the phenotype to differentiate. However, genomic signatures of natural selection are usually identified without information on which trait is responding to selection by which selective agent(s). Here we integrate whole-genome-sequencing with phenomics and measures of putative selective agents to assess the extent of adaptive divergence in threespine stickleback occupying the highly heterogeneous lake Mývatn, NE Iceland. We find negligible genome wide divergence, yet multiple traits (body size, gill raker structure and defence traits) were divergent along known ecological gradients (temperature, predatory bird densities and water depth). SNP based heritability of all measured traits was high (h2 = 0.42 – 0.65), indicating adaptive potential for all traits. Whilst environment-association analyses identified thousands of loci putatively involved in selection, related to genes linked to neuron development and protein phosphorylation, only allelic variation linked to pelvic spine length was concurrently linked to environmental variation (water depth) - supporting the conclusion that divergence in pelvic spine length occurred in face of gene flow. Our results suggest that whilst there is substantial genetic variation in the traits measured, phenotypic divergence of Mývatn stickleback is mostly weakly associated with environmental gradients, potentially as a result of substantial gene flow. Our study illustrates the value of integrative studies that combine genomic assays of multivariate trait variation with landscape genomics.
Keywords: adaptive divergence, gene flow, environmental gradients, genome scans, landscape genomics, Gasterosteus aculeatus

Introduction

Elucidating the genetic basis of adaptive divergence in natural populations is an enduring goal of evolutionary biology . Doing so can provide insight into evolutionary processes occurring in the wild, including the mechanisms associated with adaptive divergence, and the extent to which divergence takes place in the face of gene flow . Genetically, adaptive divergence is expected to manifest as blocks of differentiation across the genome, at regions containing genes that contribute to adaptation to divergent local environments . Genome scan studies that test these expectations have identified genomic regions associated with adaptation to divergent ecological niches in numerous species (e.g., ). This has been termed a “reverse ecology” approach, whereby loci associated with adaptation may be identified without measuring the traits themselves . However, genome scan studies on wild populations are seldom able to provide precise information on which aspects of the phenotype selection is acting on, or which environmental factors are imposing selection .
A comprehensive view on the genomic mechanisms associated with adaptive divergence requires studies that combine phenotypic, environmental and genomic data. Accordingly, integrative approaches that combine association mapping with landscape genomics or selection scans to map gene-phenotype-environment associations could be a powerful means to infer the genomic basis of adaptation . Association mapping studies (e.g., genome-wide-associations, GWA ) identify specific loci that underlie divergent traits, whereas landscape genomic studies can aid in determining loci associated with adaptive divergence, under the assumption that loci should be correlated with environmental variation that is directly or indirectly causing selection . Combining association mapping with landscape genomics can strengthen the identification of genomic signatures of selection by allowing inference on whether causal variants of phenotypic variation are concurrently associated with environmental variation. This would be especially true in cases where correlations between phenotype and environment are mirrored in genetic polymorphisms, where at some quantitative trait loci, allele frequencies differ between groups that inhabit different environments.
In the absence of dispersal barriers, many populations remain connected by gene flow during the process of adaptive divergence, often along environmental clines . Gene flow is expected to constrain divergence, swamping locally adapted alleles and breaking up favourable allele combinations through recombination . Whilst in cases of substantial gene flow there may be little genome-wide divergence, responses to natural selection may be present at specific genomic regions (islands of divergence; ). Identifying genomic divergence in the presence of gene flow is a major challenge because most genome scan approaches require grouping individuals, which is not usually possible when individuals remain connected . Our perspective on adaptive divergence may therefore be biased towards studies where physical barriers to gene flow have facilitated divergence. Although such studies have provided great insight into evolutionary processes, studying processes of divergence in populations connected by gene flow can greatly improve our understanding of the relative roles of natural selection and gene flow in adaptive divergence .
Here, we employ GWA and landscape genomic approaches to map gene-phenotype-environment associations in threespine stickleback that inhabit Mývatn, a highly environmentally heterogeneous lake in NE Iceland. Threespine stickleback is a well-established model system in evolutionary biology . Within freshwater systems, there is evidence for repeated adaptive divergence at both phenotypic and genomic levels , most commonly across the benthic-limnetic axis (e.g. ) but also across a range of other selective agents (e.g., predation ). However, most of the studies focus on simple environmental contrasts (e.g., benthic vs limnetic or lake vs stream), and only few studies have aimed to test intralacustrine divergence across environmental gradients.
Mývatn is a large (37 km2) lake, where temperature, water depth, invertebrate, and vertebrate (including stickleback) densities vary over space and time . Stickleback habitats in this lake can crudely be divided to five main types, across which stickleback vary phenotypically . Previous work found that male stickleback had relatively larger brains in a ´lava´ (warm) than a ´mud´ (colder) habitat , relatively longer spines in the north basin than the south basin , and divergence in gill raker morphology and diet among some of the habitats . Evidence for population genetic divergence of stickleback across the lake is mixed. Using samples collected between 1999 and 2002, found evidence for genetic divergence using a suite of nuclear and mitochondrial markers between stickleback inhabiting the ´lava´ and ´mud´ habitats (microsatellites: FST = 0.08; mtDNA: FST = 0.223), suggesting the presence of two contrasting morphs. In contrast, using samples collected in 2009 and 12 nuclear microsatellite loci (seven of which were the same as in Ólafsdóttir et al. 2007), found little evidence for neutral genetic divergence of stickleback across five habitat types (average pairwise FST = 0.004), suggesting extensive gene flow.
Given the known phenotypic divergence in traits typically under selection in stickleback, coupled with spatial variation in possible selective agents, our main goal here was to identify genomic signatures of selection in Mývatn stickleback occupying different environments. Genomic signatures of selection are typically defined as genomic regions which are disproportionately divergent between groups compared to the rest of the genome . We extended this definition to strengthen our identification of signatures of selection: we expected that genomic regions that bear a signature of selection should be both divergent across ecological axes, and contain loci associated with variation in divergent traits. We further measured SNP-based additive genetic variation of divergent traits to gain insight into the evolutionary potential of traits that are spatially divergent.