General combining ability model for genomewide selection: accuracy, marker imputation, and genetic diversity within maize biparental populations

Abstract

A general combining ability (GCA) model enables genomewide selection in the cross between parents A and B before the A/B cross itself is phenotyped. Prior A/* and */B populations, where * indicates any other parent, are used as the training population in the GCA model. I conducted three studies that utilized phenotypic data (grain yield, moisture, and test weight) and single nucleotide polymorphism data for 969 A/B crosses in the Monsanto maize (Zea mays L.) breeding program. The first study aimed to determine if the GCA model is useful for genomewide selection in an A/B cross, and to assess the influence of training population size, number of crosses in the training population, linkage disequilibrium, and heritability on the prediction accuracy (rMP) with the GCA model. Increases in each of these factors improved the prediction accuracy. The GCA model led to selection responses (R) that were 68 to 76% of those eventually achieved with phenotypic selection. The second study aimed to determine: (i) if marker imputation increases R and rMP within biparental crosses; (ii) the number of markers needed to reach a plateau in rMP; and (iii) the lowest number of assayed SNP markers that can be used for imputation without a significant decrease in rMP. Marker imputation made the GCA model as good as or better than the A/B model (which used the A/B cross itself) in terms of R and rMP. The rMP values did not increase significantly beyond 500 imputed markers for grain yield, and 1000 imputed markers for moisture and test weight. The third study aimed to determine if genomewide selection and phenotypic selection lead to comparable losses in genetic diversity within a biparental population. Phenotypic selection for grain yield, moisture, and an index of these two traits did not cause a significant loss in genetic diversity among the selected lines. Genomewide selection of the best 5% of lines led to a small but statistically significant loss in genetic diversity. Overall, my results suggest that the GCA model is effective for genomewide selection within an A/B cross, prior to phenotyping the progeny in the cross itself.