Genotyping-by-sequencing for genomic selection of perennial ryegrass (Lolium perenne)

Perennial ryegrass (Lolium perenne L.) is the principal source of nutrition for ruminant animals grazed in pastoral agricultural systems in temperate regions of the world, including New Zealand. Ryegrass is an obligately outbreeding perennial plant species, adapted and bred as heterogeneous populations of heterozygous individuals. These inherent features present challenges for plant breeders seeking to increase the rate of genetic gain for complex traits, such as herbage yield, nutritive quality and long-term pasture persistence. Genomic selection (GS), a breeding approach wherein the effects of high-density single nucleotide polymorphism (SNP) markers are used to predict breeding values, may be used to lift genetic gain. Genotyping-by-sequencing (GBS) is a method used to develop rapid and cost-effective high-density SNP marker data, and was chosen as the enabling SNP marker platform for GS in ryegrass. GBS libraries were generated using the enzyme ApeKI for five populations constituting a GS training set. The populations were developed with the aim of training and assessing genomic prediction models in forages. Following sequencing of the GBS libraries, a reference genome-based informatics pipeline based on TASSEL 5.0 was used to process the data. After filtering for missing data (<50% per SNP site), minor allele frequency (MAF>0.05) and, using the KGD (kinship using GBS with depth adjustment) statistical package, Hardy-Weinberg disequilibrium (>-0.05), 1,023,011 bi-allelic SNPs were called within the GS training set. The statistical method KGD, designed specifically for GBS data, which commonly have low read depth, was used to develop an unbiased genomic relationship matrix and GS predictive models for test traits measured in the ryegrass training set. Comparison with other statistical methods demonstrated that KGD was faster and less compute-intensive than other GBLUP methods, 'ridge regression', and random forest, without any loss of prediction accuracy. Components of the data analysis pipeline are available in a public Github repository (https://github.com/AgResearch).