Genomic selection of forage quality traits in winter wheat

Abstract

Phenotyping forage quality traits is time-consuming in forage wheat breeding. In this study, prediction accuracies of three genomic selection (GS) models (ridge regression best linear unbiased prediction [RRBLUP], Gaussian kernel [GAUSS], and Bayesian LASSO [BL, where LASSO stands for least absolute shrinkage and selection operator]) for forage quality traits of winter wheat (Triticum aestivum L.) were compared using two genotype sampling methods. In addition, the impact of training population (TP) size and marker density on prediction accuracy was explored. The study was done using a diversity panel (n = 298) that was genotyped using 90K single nucleotide polymorphisms (SNPs) and pheno-typed for forage quality traits including crude protein, acid detergent fiber, neutral detergent fiber, sugars, lignin content, and in vitro true dry matter digestibility. Generally, the three models produced similar prediction accuracies, which ranged from 0.34 to 0.61, for all traits. The sampling method had little effect on accuracy. Crude protein was one of the traits with the highest prediction accuracy, and it required only 1000 markers to attain its highest prediction accuracy value. Increasing TP size and marker density increased accuracies of all traits, and increasing the TP size was more effective than increasing marker density. For this panel, the optimal TP size (nTP) was 150, at which point prediction accuracies of all traits, except for sugars, reached over 90% of the highest value at nTP = 250. However, the sampling method for marker density had no effect on accuracy. The results suggest that GS can be an alternative approach to facilitate selection of forage quality traits during forage wheat breeding.