2.8 | Association of loci with traits
To detect loci associated with each of the eight traits, we examined genome-wide association (GWA) of SNP genotypes (association mapping) and GWA of local ancestry (admixture mapping) at ddRAD loci with the phenotypic values using mixed linear models. Despite the similarities between association mapping and admixture mapping, some differences between them in terms of aims and methods result in both merits and demerits of each (Buerkle and Lexer 2008). Association mapping can deal with complicated background of ancestry and variation within ancestral lineages underlying phenotypes using more loci densely distributed across a genome, while admixture mapping can work with fewer loci sparsely distributed owing to larger blocks of linkage disequilibrium in recombinants of distinct ancestries (Buerkle and Lexer 2008).
Phenotypic values of individual trees with the same Qd ancestry were expected to differ between the inland and coastal sites due to phenotypic plasticity. To exclude such confounding effects with environment and plasticity and to conduct GWA in both sites together, we scaled phenotypic values (mean = 0 and variance = 1) in each site separately and applied a mixed linear model to pooled data of the scaled values in both sites. Environmental heterogeneity within the sites were incorporated as random errors in the models.
To conduct GWA, we used single-locus GWA (Yu et al. 2006) implemented in the package rrBLUP 4.5 (Endelman 2011) in R 3.3.2. To discriminate confounding factors with genetic structure of sampled trees in both sites, we used the Q + K model with a population structure matrix Q calculated from probabilistic principal component analysis using the function ppca in the package pcaMethods (Stacklies et al. 2007) and a kinship matrix K calculated from realized additive relationship using the function A.mat in rrBLUP. The first four principal components were used for the Q matrix. All parameters in the GWA were set at the default values. Significant associations were regarded as < 0.05 false discovery rate (FDR) of –log10p calculated using the function p.adjust in R 3.3.2. Q-Q plots of expected and observed –log10 p values were examined for associations of SNP genotypes and local ancestry.
To search for genes involved in the adaptation to coastal environment around the trait-associated loci, we selected genes (protein ID) nearest to the loci detected by association mapping and within regions including the loci detected by admixture mapping from PM1N v2.3 Q. roburannotation database (https://urgi.versailles.inra.fr/OakMine_PM1N) (Plomion et al. 2018). We obtained gene ontology (GO) terms and protein descriptions of the genes.