Envirotyping can increase genomic prediction accuracy of new environments in grain sorghum trials depending on mega‐environment

Abstract

Grain sorghum [Sorghum bicolor (L.) Moench] is an important crop native to Africa and grown in many subtropical and temperate regions worldwide. The variability in production environments underscores the plasticity of sorghum genotypes and provides an opportunity to predict sorghum hybrid performance in novel environments. Reaction norms informed by envirotype data can aid in modeling the differential responses of genotypes across multi‐environment trials and ultimately increase the prediction accuracies of hybrid performance. In this study, a combination of genomic and enviromic information was applied to predict grain sorghum hybrid performance across United States sorghum production environments. Seven models based on additive, dominance, environment, and envirotype effects in two forms, and their interactions were tested under three cross‐validation schemes that simulate genomic prediction scenarios encountered by breeding programs. Relationship matrices for hybrids and environments were created from molecular and envirotypic data, respectively, to predict hybrid performance under a hierarchal Bayesian framework. When predicting unobserved environments, the envirotypic reaction norm model based on typologies produced the highest prediction accuracies for every temperate environment, but results vary in the subtropical environments, which has not been demonstrated in prior enviromic research. This study also demonstrated for the first time that envirotype data can effectively define established United States sorghum production regions and may assist breeders in resource allocation. Results support using sparse hybrid trials when breeding programs apply genomic predictions. Therefore, envirotype‐assisted genomic prediction models can expand the information available to a sorghum breeding program and aid breeders in efficiently designing hybrid trials.