dewlap-associated SNPs
To determine whether SNPs associated with dewlap traits are also involved in local adaptation of invasive A. sagrei populations, we combined F ST outlier analyses and genotype-environment association analyses. TheF ST outlier approach aimed to determine whether dewlap-associated SNPs make a large contribution to population genetic differentiation in invasive A. sagrei. We followed the approach described in Bock et al. (2021) for the same set of populations as those used for dewlap measurements here. Briefly, we used VCFtools (v. 0.1.16; Danecek et al., 2011) to calculate SNP-based F STamong 15 independent population pairs found in our dataset. For this, we relied on the full LD-filtered SNP set. We then comparedF ST values for all dewlap-associated SNPs withF ST values observed for an equal number of randomly selected non-dewlap SNPs. For this, we used a linear model in R, which had population pair and locus category (i.e., dewlap or non-dewlap SNP) as predictor variables, and F STvalues as the response variable. As well, we designated asF ST outliers those values that were in the top 5% of the empirical distribution, for each of the 15 population pairs. We then asked whether SNPs that are strongly associated with dewlap traits (i.e., those with the lowest inferred GWAS P value) are also repeatedly classified as F ST outliers in multiple independent population pairs.
The genotype-environment association (GEA) analyses aimed to determine whether dewlap-associated SNPs are also associated with environmental variables that are important from the perspective of dewlap signal effectiveness. We followed the approach described in DeVos et al. (2023) and used a latent factor mixed model (LFMM), as implemented in thelfmm R package (v. 1.1; Frichot et al., 2013). To correct for the confounding effect of population structure, we set K = 2, which corresponds to the main genetic subdivision in our dataset (see ‘Population structure’ results below). Similar to the GWAS analyses described above, we adjusted the GEA P values based on the genomic inflation factor. We then relied on the qvalue R package (v. 2.30.0; Storey et al., 2015) to convert P values to qvalues and to identify genome-wide significant SNPs based on a false discovery rate (FDR) of 5%. We used canopy openness as the environmental variable, given evidence of correlation between this metric and several of the dewlap traits (see ‘Associations between the dewlap, genetic ancestry, and the environment’ results below), and because this metric was considerably more variable than temperature or precipitation across our study populations (Figure S3). Finally, we compared the canopy openness GEA results with the standard GWAS results for dewlap total brightness. We focused on this trait for two reasons. First, dewlap brightness was correlated with canopy openness, as might be expected under local adaptation. Second, the GWAS for this trait revealed several ancestry-independent loci. Thus, we could exclude the confounding effects of genomic background, which may occur for loci identified using the ancestry-specific GWAS.