Implementing Genomic Selection in a temperate Rice Breeding Program

Improving selection efficiency and rate of genetic gain in plant breeding is essential to meet the growing global demand for food, fiber and fuel. Rice is a major crop in Uruguay, and the adoption of optimized agronomic practices and production technologies over the last 40 years has fueled sustained increases in planted area, production and yield. However, the impressive production yields achieved in the past two decades have begun to plateau, indicating that Uruguayan varieties may have reached their maximum yield potential, opening the door for intensified crop improvement efforts. A population of 329 tropical-japonica elite rice lines from the local Rice Breeding Program of the National Institute of Agricultural Research (INIA-Uruguay) was genotyped by Genotyping by Sequencing (GBS) and phenotyped for grain yield and two grain quality parameters: yield after milling and chalkiness percentage. After imputation and filtering, a total of 44,604 SNP markers were used for population structure and GS cross-validation analyses. Five statistical models were used to calculate prediction accuracies: RR-BLUP, Bayesian Lasso, Kinship BLUP, Random Forest and Reproducing Kernel Hilbert Spaces (RHKS). Prediction accuracies ranged from 0.1 to 0.337 for grain yield, 0.037 to 0.26 for milling yield, and 0.2 to 0.397 for chalkiness. Among the models tested, RHKS outperformed for grain yield while Random Forest outperformed for both grain quality parameters, suggesting that epistatic interactions must be taken into account.