Development of Genomic Selection for Perennial Ryegrass

Abstract

Mixed-species pasture based on perennial ryegrass (Lolium perenne) and white clover (Trifolium repens) is the foundation for profitable production from temperate grasslands. In theory, genomic selection (GS) offers an opportunity to lift the rate of genetic gain in these species. Critical research questions include to what extent theoretical expectations for GS can be realized in practice, and understanding the genetic and economic implications of GS in breeding programs and on farm. We describe a limited experiment to derive and cross-validate genomic estimated breeding values (GEBVs) from 211 perennial ryegrass plants evaluated for plant herbage dry weight (DW) and days-to-heading (DTH), using field phenotypic data and up to 10,885 markers typed via genotyping-by-sequencing (GBS). Using Ridge Regression-BLUP and Random Forest regression, cross validation prediction accuracies ranged from r = 0.16–0.34 (DW) and r = 0.52–0.56 (DTH). Accuracy was not influenced by marker density, but there was an interaction between statistical model and trait. The data indicate that, in these elite breeding populations, low marker densities in a limited training population dataset may be viable for generation of GEBVs in perennial ryegrass. Variance attributable to population structure, rather than linkage disequilibrium, is likely the primary basis of GEBV accuracy in this study. Generation of a training set of increased size, scope and greater relevance for key economic traits is outlined in the context of developing a GS capability with an Australasian focus.