Genomewide selection utilizing historic datasets improves early stage selection accuracy and selection stability

Abstract

AbstractOver the past decade, commercial maize (Zea mays L.) breeding programs have generated a large quantity of complimentary phenotypic and genotypic datasets on their recurrent breeding populations. As many genetic subgroups and alleles are recycled over the years, these data can be a valuable resource for predicting the performance of future breeding cycles. Our objective was to test the efficacy of using historic breeding data within a genomewide selection (GWS) framework to improve plant breeding selections in early stage testing of maize lines. Across a multiyear maize breeding dataset propriety of Syngenta Seeds, LLC., we compared the selection accuracy of GWS and phenotypic selection (PS). Genomewide selection improved selection accuracy relative to PS by 15% for grain yield, 21% for grain moisture, 3% for test weight, and 58% for ear height. Additionally, we demonstrate the stability of GWS for plant selections through a stratified sampling procedure in which loss of testing environments was simulated. Plant selections with GWS were robust and stable with the simulated loss of up to four testing environments (out of seven). Specifically, under this scenario, where four early stage testing environments were lost, the average GWS selection accuracies only decreased relatively by 10, 8 and 5% for grain yield, grain moisture, and test weight, respectively, compared with an average relative reduction for PS by 35, 13, and 12% for grain yield, grain moisture, and test weight, respectively. These results indicate that an abundance of historic genotypic and phenotypic data can compensate for a lack of preliminary yield trials.