Multiple, Single Trait GWAS and Supervised Machine Learning Reveal the Genetic Architecture of Fraxinus excelsior Tolerance to Ash Dieback in Europe.

Abstract

Common ash (Fraxinus excelsior) is under intensive attack from the invasive alien pathogenic fungus Hymenoscyphus fraxineus, causing ash dieback at epidemic levels throughout Europe. Previous studies have found significant genetic variation among genotypes in ash dieback susceptibility and that host phenology, such as autumn yellowing, is correlated with susceptibility of ash trees to H. fraxineus; however, the genomic basis of ash dieback tolerance in F. excelsior requires further investigation. Here, we integrate quantitative genetics based on multiple replicates and genome-wide association analyses with machine learning to reveal the genetic architecture of ash dieback tolerance and of phenological traits in F. excelsior populations in six European countries (Austria, Denmark, Germany, Ireland, Lithuania, Sweden). Based on phenotypic data of 486 F. excelsior replicated genotypes we observed negative genotypic correlations between crown damage caused by ash dieback and intensity of autumn leaf yellowing within multiple sampling sites. Our results suggest that the examined traits are polygenic and using genomic prediction models, with ranked single nucleotide polymorphisms (SNPs) based on GWAS associations as input, a large proportion of the variation was predicted by unlinked SNPs. Based on 100 unlinked SNPs, we can predict 55% of the variation in disease tolerance among genotypes (as phenotyped in genetic trials), increasing to a maximum of 63% when predicted from 9155 SNPs. In autumn leaf yellowing, 52% of variation is predicted by 100 unlinked SNPs, reaching a peak of 72% using 3740 SNPs. Based on feature permutations within genomic prediction models, a total of eight nonsynonymous SNPs linked to ash dieback crown damage and autumn leaf yellowing (three and five SNPs, respectively) were identified, these were located within genes related to plant defence (pattern triggered immunity, pathogen detection) and phenology (regulation of flowering and seed maturation, auxin transport). We did not find an overlap between genes associated with crown damage level and autumn leaf yellowing. Hence, our results shed light on the difference in the genomic basis of ADB tolerance and autumn leaf yellowing despite these two traits being correlated in quantitative genetic analysis. Overall, our methods show the applicability of genomic prediction models when combined with GWAS to reveal the genomic architecture of polygenic disease tolerance enabling the identification of ash dieback tolerant trees for breeding or conservation purposes.