ENHANCING GENOMIC PREDICTION MODELS IN <i>MISCANTHUS</i> POPULATIONS BY INCORPORATING THE GENOTYPE-BY-ENVIRONMENT INTERACTION

ABSTRACT Giant Miscanthus giganteus ( Mxg ) is one of the most promising perennial crops to generate biomass feedstock for bioenergy and biobased products. It is derived from the natural inter-species hybridization of Miscanthus sacchariflorus ( Msa ) and Miscanthus sinensis ( Msi ) species, thus population improvement within these species is crucial. Genomic selection (GS) is an attractive option to accelerate breeding of perennial grasses, such as Miscanthus, which requires up to three years of evaluation to produce reliable phenotypic data. Hence, genotypes are observed in multiple years and locations causing inconsistent response patterns from one year to the next, between location, and/or location-by-year combinations. These inconsistencies are known as the genotype-by-environment interaction effect (G×E). Although GS has been successfully implemented in multiple annual crops where straightforward cross-validation schemes exist to assess the levels of predictive ability that can be reached, for perennial crops new cross-validation schemes will help avoid data contamination. Here, we propose a series of cross-validation schemes to evaluate model performance for perennial crops. We perform a case study by analyzing one panel of each species (516 genotypes of Msa , 280 genotypes of Msi ) scored for biomass yield at different locations around the world over several years. The results of the different cross-validation schemes provide insights about the usefulness of GS to accelerate the breeding process of Miscanthus species. In addition, leveraging the G×E effects of different types significantly increases predictive ability (up to 10% in Msa and 30% for Msi ) compared to the conventional approaches based on main effects only.