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. Mixed linear models that identify allele associations to environment, while controlling for genome-wide variation at other loci, have emerged as the method of choice when studying local adaptation.Despite their importance, it is unclear whether this approach performs better than identifying environmentally-associated SNPs without accounting for population structure. To examine this, we first use the mixed linear model GEMMA, and simple Spearman correlations, to identify SNPs showing significant associations to climate with and without accounting for population structure. Subsequently, using Italy and Sweden populations, we compare evidence of allele frequency differentiation (FST), linkage disequilibrium (LD ≈ 2 ��� ), fitness variation, and functional constraint, underlying these SNPs. Using a lenient cut-off for significance, we find that SNPs identified by both approaches, and SNPs uniquely identified by Spearman correlations, were enriched at sites showing genomic evidence of local adaptation and function but were limited across Quantitative Trait Loci (QTL) explaining fitness variation. SNPs uniquely identified by GEMMA, showed no direct or indirect evidence of local adaptation, and no enrichment along putative functional sites. Finally, SNPs that showed significantly high FST and LD, were enriched along fitness QTL peaks and cis-regulatory/nonsynonymous sites showing significant functional constraint. Using these SNPs, we identify genes underlying fitness QTL, and genes linking flowering time to local adaptation. These include a negative regulator of abscisic-acid (FLDH) and flowering time genes PIF3, FIO1, and COL5.

Linking genomic signatures of selection to expression variation and direct evidence of local adaptation

“…LD was measured using the package 'PLINK' (Purcell, et al 2007) and it was estimated as the mean square coefficient of correlation ( 2 ��� ). AFD and LD were estimated in a previous study (Price, et al 2020).…”

“…To examine whether the proportion of (E/GxE) genes with cis-regulatory/nonsynonymous SNPs showing evidence of local adaptation (estimated using AFD and/or LD) is significantly higher than expected by chance we used a circular permutation test. This test has been previously explained in detail (Price, et al 2020); but in brief, AFD's and/or LD's are shifted across the genome (not randomly shuffled) and according to certain criteria (e.g., AFD>0.60) we estimate the proportion of genes with high AFD and/or LD cis-regulatory/nonsynonymous SNPs. This was repeated a thousand times and the resulting permutation distribution is compared to the observed proportion/number of genes with high AFD.…”

“…Simulation studies comparing various methods used to identify genetic variation underlying local adaptation while accounting for the effects of population structure have shown that the power of each method can significantly change depending on the underlying evolutionary scenario, in addition to other factors (De Mita, et al 2013;de Villemereuil, et al 2014;Lotterhos and Whitlock 2015;Yoder and Tiffin 2017). Furthermore, in examining the link between GWA/population-genomic methods and empirical evidence of local adaptation, the strength of this link changed depending on the method(s) used in studies (Fournier-Level, et al 2011;Lasky, et al 2014;Yoder, et al 2014;Exposito-Alonso, et al 2018;Price, et al 2018;Price, et al 2020).…”

“…To re-examine the link between genetic, expression, and fitness variation of Arabidopsis populations in different climates, and the role of the CBF-pathway in local adaptation to temperature, the current study examines the following data: (a) re-sequenced genomes of locally adapted (Ågren and Schemske 2012) South Italy and North Sweden (Price, et al 2020); (b) expression of Italy and Sweden genotypes under control and cold-acclimation conditions (Gehan, et al 2015); and (c) quantitative trait loci (QTL) explaining fitness variation of Italy and Sweden recombinant inbred lines grown in a series of reciprocal transplant experiments (Ågren, et al 2013). More specifically, we examine the link between allele frequency differentiation (AFD) and linkage disequilibrium (LD) at cis-regulatory (sites found 1kb upstream from the transcriptional start site) and nonsynonymous sites, to patterns of expression (E and GxE) and genetic-tradeoff QTL, while taking into account the effects of selective constraint (or purifying selection) at nonsynonymous sites.…”

“…More specifically, we examine the link between allele frequency differentiation (AFD) and linkage disequilibrium (LD) at cis-regulatory (sites found 1kb upstream from the transcriptional start site) and nonsynonymous sites, to patterns of expression (E and GxE) and genetic-tradeoff QTL, while taking into account the effects of selective constraint (or purifying selection) at nonsynonymous sites. Among population genomic signatures of local adaptation we chose AFD and LD, since these were previously found to be enriched along fitness QTL (Price, et al 2020).…”