Genomic prediction of rice mesocotyl length indicative of directing seeding suitability using a half-sib hybrid population

Abstract

Direct seeding has been widely adopted as an economical and labor-saving technique in rice production, though problems such as low seedling emergence rate, emergence irregularity and poor lodging resistance are existing. These problems are currently partially overcome by increasing seeding rate, however it is not acceptable for hybrid rice due to the high seed cost. Improving direct seeding by breeding is seen as the ultimate solution to these problems. For hybrid breeding, identifying superior hybrids among a massive number of hybrids from crossings between male and female parental populations by phenotypic evaluation is tedious and costly. Contrastingly, genomic selection/prediction (GS/GP) could efficiently detect the superior hybrids capitalizing on genomic data, which holds a great potential in plant hybrids breeding. In this study, we utilized 402 rice inbred varieties and 401 hybrids to investigate the effectiveness of GS on rice mesocotyl length, a representative indicative trait of direct seeding suitability. Several GP methods and training set designs were studied to seek the optimal scenario of hybrid prediction. It was shown that using halfsib hybrids as training set with the phenotypes of all parental lines being fitted as a covariate could optimally predict mesocotyl length. Partitioning the molecular markers into trait-associated and -unassociated groups based on genome-wide association study using all parental lines and hybrids could further improve the prediction accuracy. This study indicates that GS could be an effective and efficient method for hybrid breeding for rice direct seeding.