Leveraging spatio-temporal genomic breeding value estimates of dry matter yield and herbage quality in ryegrass via random regression models

Abstract

AbstractJoint modeling of correlated multi-environment and multi-harvest data of perennial crop species may offer advantages in prediction schemes and a better understanding of the underlying dynamics in space and time. The goal of the present study was to investigate the relevance of incorporating the longitudinal dimension of within-season multiple harvests of biomass yield and nutritive quality traits of forage perennial ryegrass (Lolium perenne L.) in a reaction norm model setup that additionally accounts for genotype-environment interactions. Genetic parameters and accuracy of genomic breeding value predictions were investigated by fitting three random regression (random coefficients) linear mixed models (gRRM) using Legendre polynomial functions to the data. All models accounted for heterogeneous residual variance and moving average-based spatial adjustments within environments. The plant material consisted of 381 bi-parental family pools and four check varieties of diploid perennial ryegrass evaluated in eight environments for biomass yield and nutritive quality traits. The longitudinal dimension of the data arose from multiple harvests performed four times annually. The specified design generated a total of 16,384 phenotypic data points for each trait. Genomic DNA sequencing was performed using DNA nanoball-based technology (DNBseq) and yielded 56,645 single nucleotide polymorphisms (SNPs) which were used to calculate the allele frequency-based genomic relationship matrix used in all genomic random regression models. Biomass yield’s estimated additive genetic variance and heritability values were higher in later harvests. The additive genetic correlations were moderate to low in early measurements and peaked at intermediates, with fairly stable values across the environmental gradient, except for the initial harvest data collection. This led to the conclusion that complex genotype-by-environment interaction (G×E) arises from spatial and temporal dimensions in the early season, with lower re-ranking trends thereafter. In general, modeling the temporal dimension with a second-order orthogonal polynomial in the reaction norm mixed model framework improved the accuracy of genomic estimated breeding value prediction for nutritive quality traits, but no gain in prediction accuracy was detected for dry matter yield. This study leverages the flexibility and usefulness of gRRM models for perennial ryegrass research and breeding and can be readily extended to other multi-harvest crops.