Genomic Prediction in a Breeding Program of Perennial Ryegrass

We present a genomic selection study performed on 1918 rye grass families (Lolium perenne L.), which were derived from a commercial breeding program at DLF-Trifolium, Denmark. Phenotypes were recorded on standard plots, across 13 years and in 6 different countries. Variants were identified by Genotype-By-Sequencing and the allele frequency of the variant allele was determined at each data point.

GBLUP analyses on heading date, crown rust resistance, and water soluble carbohydrates were performed on a training set of 945 F₂ families (originated by pair crosses), with cross validation within this set. Estimated Breeding Value and prediction accuracies were calculated trough two different cross-validation schemes: (i) k-fold (k=10); (ii) leaving out one parent combination at the time, in order to test for accuracy of predicting new families. Accuracies ranged between 0.56 and 0.97 for scheme (i) and between 0.46 and 0.91 for scheme (ii), depending on the trait. Incorporation of information on parent populations (PPs), performed by using a genomic relationship matrix based on PPs estimated frequencies, led to an increase in accuracy up to 0.07 when leaving out a parent combination at the time.

A larger set of 1791 F₂s were used as training set to predict EBVs of 127 synthetic families (originated from poly-crosses between 5-11 single plants) for heading date and crown rust resistance. Prediction accuracies were 0.93 and 0.57 respectively. Results clearly demonstrate considerable potential in utilizing genomic selection in rye grass.