Modeling genotype by environment interactions using genomic and environmental data in apple

Abstract

Genotype by environment interactions (G × E) affect the response of plant genotypes to environments, leading to changes in the extent of phenotypic differences between genotypes grown in contrasting environments. Understanding of G × E and its prediction are important for breeding and deployment of new apple varieties. Multi-environment trials, such as the apple reference population (apple REFPOP), allow to evaluate the effect of G × E for a diverse set of genotypes. Large-scale genomic, phenotypic, and environmental data can be incorporated in genomic prediction models estimating G × E, which may lead to increased genomic predictive ability for genotypes in different environments. However, the large datasets from multi-environment trials increase complexity of statistical models accounting for G × E and lead to long computational times, which hinder extensive model exploration. Here, new modeling techniques designed to increase computational efficiency were used to estimate genomic predictive ability for the genotypes of the apple REFPOP, which were phenotyped for eleven traits in up to 25 environments (i.e., 25 combinations of location and year). Reported proportions of the variance components related to genetic, environmental, and residual sources of phenotypic variation contribute to understanding of the effect of G × E on the studied traits. The estimated computation times and predictive abilities show that increased efficiency and accuracy of multi-environment genomic predictions can be reached using the tested new modeling techniques. These improved genomic prediction models will prove valuable for future breeding decisions in apple.