Accuracy of genomic selection for alfalfa biomass yield in two full-sib populations

Abstract

Alfalfa (Medicago sativa) is one of the most important leguminous forages, widely planted in temperate and subtropical regions. As a homozygous tetraploid, its complex genetic background limits genetic improvement of biomass yield attributes through conventional breeding methods. Genomic selection (GS) could improve breeding efficiency by using high-density molecular markers that cover the whole genome to assess genomic breeding values. In this study, two full-sib F1 populations, consisting of 149 and 392 individual plants (P149 and P392), were constructed using parents with differences in yield traits, and the yield traits of the F1 populations were measured for several years in multiple environments. Comparisons of individual yields were greatly affected by environments, and the best linear unbiased prediction (BLUP) could accurately represent the original yield data. The two hybrid F1 populations were genotyped using GBS and RAD-seq techniques, respectively, and 47,367 and 161,170 SNP markers were identified. To develop yield prediction models for a single location and across locations, genotypic and phenotypic data from alfalfa yields in multiple environments were combined with various prediction models. The prediction accuracies of the F1 population, including 149 individuals, were 0.11 to 0.70, and those of the F1 population, consisting of 392 individuals, were 0.14 to 0.67. The BayesC and RF models had the highest average prediction accuracy of 0.60 for two F1 populations. The accuracy of the prediction models for P392 was higher than that of P149. By analyzing multiple prediction models, moderate prediction accuracies are obtained, although accuracies will likely decline across multiple locations. Our study provided evidence that GS can accelerate the improvement of alfalfa yield traits.