Isolation by distance and isolation by environment
To better understand the drivers potencially explaining the observed patterns of spatial genetic variation in E. coccineum , we performed distance-based redundancy analysis (db-RDA) and partial db-RDA at the individual level on the complete PG_dataset and on loci potentially influenced by selection. RDA is a multiple linear regression method performed between a matrix of dependent variables and matrices of independent variables, which has been demonstrated as more appropriate than other methods (i.e., Mantel test) when multiple variables are analyzed to identify drivers of population genetic structure (Nadeu et al., 2016; Orsini et al., 2013).
A dependent matrix, calculated with a Bray-Curtis dissimilarity index (Bray and Curtis, 1957), of co-dominant variant call format of each individual was used as a response variable. Predictor matrices were geographic distance (IBD), environmental disparity (IBE) and degree of shared co-ancestry (IBA). As environmental data, we used seven environmental variables with low level of covariation (see Supplementary table 1). For the distance matrix, we used a Principal Coordinates of Neighbourhood Matrix (PCNM), using a truncation threshold of 0.05 for long distances with the function pcnm() within the R package vegan v.2.5.7 (Oksanen et al., 2007). PCNM is commonly used to transform spatial distances into rectangular data matrices suitable for constrained ordination or regression analyses (Borcard and Legendre 2002). Populations Q-values obtained with STRUCTURE K = 2, which separated populations into North and Center-South ancestry linages (see results for more details), were used as co-ancestry variable. Prior to the analysis, all three independent matrices were scaled to a mean of zero and variance of one with the scale() function in R.
Among populations variation in E. coccineum was partitioned into components explained by geographic distance (IBD), ecological gradients (IBE) and co-ancestry (IBA), or their combination (i.e., constrained by the effects of the remaining two independent matrices), using the vartpart() function within R package vegan v.2.5.7 (Oksanen et al., 2007). Significance of each partition was tested with the anova.cca() function through 1,000 permutations on the R package vegan v.2.5.7 (Oksanen et al., 2007).