In the presence of population structure: From genomics to candidate genes underlying local adaptation

Abstract

Understanding the genomic signatures, genes, and traits underlying local adaptation of organisms to heterogeneous environments is of central importance to the field evolutionary biology. To identify loci underlying local adaptation, models that combine allelic and environmental variation while controlling for the effects of population structure have emerged as the method of choice. Despite being evaluated in simulation studies, there has not been a thorough investigation of empirical evidence supporting local adaptation across these alleles. To evaluate these methods, we use 875 Arabidopsis thaliana Eurasian accessions and two mixed models (GEMMA and LFMM) to identify candidate SNPs underlying local adaptation to climate. Subsequently, to assess evidence of local adaptation and function among significant SNPs, we examine allele frequency differentiation and recent selection across Eurasian populations, in addition to their distribution along quantitative trait loci (QTL) explaining fitness variation between Italy and Sweden populations and cis‐regulatory/nonsynonymous sites showing significant selective constraint. Our results indicate that significant LFMM/GEMMA SNPs show low allele frequency differentiation and linkage disequilibrium across locally adapted Italy and Sweden populations, in addition to a poor association with fitness QTL peaks (highest logarithm of odds score). Furthermore, when examining derived allele frequencies across the Eurasian range, we find that these SNPs are enriched in low‐frequency variants that show very large climatic differentiation but low levels of linkage disequilibrium. These results suggest that their enrichment along putative functional sites most likely represents deleterious variation that is independent of local adaptation. Among all the genomic signatures examined, only SNPs showing high absolute allele frequency differentiation (AFD) and linkage disequilibrium (LD) between Italy and Sweden populations showed a strong association with fitness QTL peaks and were enriched along selectively constrained cis‐regulatory/nonsynonymous sites. Using these SNPs, we find strong evidence linking flowering time, freezing tolerance, and the abscisic‐acid pathway to local adaptation.

Polygenic adaptation of rosette growth in Arabidopsis thaliana

“…The analysis of polygenic scores and patterns of differentiation suggests that much of the variation observed within Europe has been shaped by natural selection, rather than by the burden imposed by deleterious mutations. Leveraging polygenic associations in local adaptation studies remains challenging [78]. Methodological developments that improve the use polygenic associations for the study of local adaptation are needed to consolidate these conclusions.…”

“…Traits with polygenic architecture are controlled by variation in many loci of low frequency and/or low effect sizes and dissecting their evolution is arguably a major challenge today in evolutionary biology [77][78][79][80]. Specifically, random SNPs with outlier frequency are not always sufficiently corrected for with the kinship matrix and these may give rise to spurious associations.…”

Putative chromosomal rearrangements are associated primarily with ecotype divergence rather than geographic separation in an intertidal, poorly dispersing snail

“…Overcorrection and loss of association signals may occur when methods that account for genome-wide demographic history capture signals of selection correlated with population structure (Atwell et al, 2010), such as the inclusion of latent factors of genetic variation associated with locally adapted populations. This loss of significance signal after inclusion of population structure covariates has been observed in studies of major loci associated with adaptation in Arabadopsis thaliana (Price et al, 2020) and sea-age variation in Salmo salar Atlantic salmon (Barson et al, 2015;Sinclair-Waters et al, 2020). Correction for population structure by modelling allelic covariance can also be impeded by the presence of structural variants with highly correlated allele frequencies (Lotterhos., 2019), and correction for population structure using a genomic kinship matrix resulted in the loss of association signal in large, putatively adaptive F I G U R E 5 Hierarchically clustered heat map of correlations between all compared traits, and the first three PCs from principal component axis on all 11,504 SNPs, and the first canonical axis from separate redundancy analyses on shape, size-corrected shape, ornamentation and size haploblocks in Helianthus sunflower ecotypes (Todesco et al, 2020).…”

“…Consequently, genomic regions associated with ecological isolation may exhibit allelic covariance (Lotterhos & Whitlock, 2015;Wang & Bradburd, 2014). These studies indicate the need for statistical tools better suited to detecting the genomic basis of trait variation in wild populations (Price et al, 2020).…”

Polygenic adaptation of rosette growth inArabidopsis thaliana

“…The analysis of polygenic scores and patterns of differentiation suggests that much of the variation observed within Europe has been shaped by natural selection, rather than by the burden imposed by deleterious mutations. Leveraging polygenic associations in local adaptation studies remains challenging [76]. Methodological developments that improve the use polygenic associations for the study of local adaptation are needed to consolidate these conclusions.…”