Genomic prediction of biomass yield in two selection cycles of a tetraploid alfalfa breeding population

Alfalfa (Medicago sativa L.) is a widely planted perennial forage legume grown throughout temperate and dry subtropical regions in the world. Long breeding cycles limit genetic improvement of alfalfa, particularly for complex traits such as biomass yield. Genomic selection (GS), based on predicted breeding values obtained using genome-wide molecular markers, could enhance breeding efficiency in terms of gain per unit time and cost. In this study, we genotyped tetraploid alfalfa plants that had previously been evaluated for yield during two cycles of phenotypic selection using genotyping-by-sequencing (GBS). We then developed prediction equations using yield data from three locations. Approximately 10,000 single nucleotide polymorphism (SNP) markers were used for GS modeling. The genomic prediction accuracy of total biomass yield ranged from 0.34 to 0.51 for the Cycle 0 population and from 0.21 to 0.66 for the Cycle 1 population, depending on the location. The GS model developed using Cycle 0 as the training population in predicting total biomass yield in Cycle 1 resulted in accuracies up to 0.40. Both genotype × environment interaction and the number of harvests and years used to generate yield phenotypes had effects on prediction accuracy across generations and locations, Based on our results, the selection efficiency per unit time for GS is higher than phenotypic selection, although accuracies will likely decline across multiple selection cycles. This study provided evidence that GS can accelerate genetic gain in alfalfa for biomass yield.