Using non-addive effects in genome-wide association studies and genomic predictions to improve biotic stress tolerance in peach

International audience ; Accounting for genetic architecture is crucial to breed for sustainable disease resistances and tolerances in plants. Indeed, (i) harnessing together minor and major effects genes allows to design a more durable plant immunity with large spectrum (ii) access to non-additive variance allows for a better exploitation of the total genetic variance when it comes to breeding, which is particularly relevant for clonally propagated crops. Our study system, Prunus persica (peach tree), is a major temperate fruit crop characterized by an overall high susceptibility to several pests and diseases, illustrated by a frequency treatment index around five times higher than in cereals. In this work, we phenotyped symptoms of two pests (leafhopper and twig moth) and four diseases (rust, leaf curl, mildew and shot hole) under low pesticide cover over three years in a peach core-collection replicated at three sites. This population consists in 192 unique accessions representing peach worldwide diversity and has been genotyped with the IRSC 16K SNP array. We used linear mixed models and the natural orthogonal interactions approach (Vitezica et al. 2017) to explicitly decompose genetic variance into additive, dominant and epistatic effects, and genotype x environment interactions. Genome-wide associations studies (GWAS) were performed with single-locus mixed models including kinships accounting for different dominance inheritance patterns. Genomic predictions consisted in a comparison of five GBLUP models incorporating different combinations of non-additive and inbreeding effects. After describing significant non-additive genetic variance and inbreeding effects across traits, we show that in addition to additive quantitative trait loci (QTLs), three to eight additional QTLs have been detected when accounting for dominant architecture. We were also able to improve genomic predictions by up to +0.05 in predictive ability with models incorporating non-additive and inbreeding terms in comparison to the additive ...