Multi-environment evaluation and genomic prediction of agronomic traits in the southern US rice genepool

The southern US is responsible for 80% of the country’s production of rice, approximately half of which is exported to other countries. Understanding genotypic and environmental factors impacting the historical performance of rice (Oryza sativa L.) is important for directing research efforts to optimize production of this globally important crop. A set of 429 rice genotypes including globally diverse historical parents and advanced japonica breeding lines from southern US breeding programs were phenotyped in 2008 for 8 agronomic traits in Arkansas, Louisiana, and Mississippi. These were also genotyped using a single-nucleotide polymorphism set optimized for genomic prediction/selection. Genotypic and phenotypic data were analyzed via clustering techniques, principal component analysis, and Finlay-Wilkinson regression. Single trait Genomic Best Linear Unbiased Prediction, multi-trait genomic prediction (via mega-scale linear mixed models; MegaLMM), and crop growth modeling (CERES-Rice in the Decision Support System for Agrotechnology Transfer) were used to predict/simulate traits on a per-plant basis. We found that contemporary germplasm from the southern state breeding programs were highly interrelated and distinct from progenitor indica and temperate japonica genotypes. Genomic predictive abilities were high and largely consistent across environments for seed number per panicle, tiller number, and plant height. Although predictive abilities were lower for seed weight, that trait was correlated with seed number per panicle (r = 0.919), and predictive ability was higher for both traits in a multi-trait prediction framework. Furthermore, including data from the two major genotypic clusters identified herein had no penalty on predictive ability. The data and analyses presented herein could inform future genomic and phenotypic investigations and applied breeding in the southern US rice germplasm pool.