Prospects of Genomic Selection for Wheat Improvement

Genomic selection (GS) is a novel approach which has been suggested as a complementary tool to phenotypic selection to improve genetic gain of quantitative traits. However, there is still limited empirical evidence on the practical application of GS for wheat breeding. The objectives of this study were i) to evaluate the potential of single and multiple-trait GS prediction models for wheat breeding ii) to examine improvements in prediction accuracy when modelling genotype by environment (GxE) interactions. This study was based on 231 hexaploid wheat lines that were genotyped using the wheat 90K iSelect assay and evaluated for a number of agronomic and end-use quality traits in ten environments in western Canada. Several single-trait (RR-BLUP, G-BLUP, BayesA, BayesB, BayesCπ, BRR, BL, and RKHS) and multiple-trait (MT-BayesA, MT-BayesA-matrix, and MT-BayesA-scalar) models were generated to predict yield and other traits. The effect of modelling GxE interactions on GS model prediction accuracy was assessed using a marker by environment interaction model and a reaction norm model that incorporates environmental data. The average prediction accuracies ranged from 0.5 to 0.8 for the various traits and models. Prediction accuracy for single trait prediction models was similar to that when multiple trait prediction models were used. For some traits, multi-trait prediction accuracy was lower to single trait prediction, but this was dependent on the inter-trait correlation. Modeling GxE interactions improved prediction accuracy dependent on the prediction problem. Results obtained from this study are promising but further research is still required to fully implement GS for wheat breeding.