Performance of Genomic Prediction for a Sugarcane Commercial Breeding Program

Abstract

Sugarcane is a clonally propagated crop of economic importance in tropical areas and is mostly used for production of sugar, ethanol, energy and animal feed. Cultivars are hybrids between two autopolyploid species, the domesticated "noble cane" Saccharum officinarum L. (2n=80) and the wild Saccharum spontaneum L. (2n=40-128). In this study genomic selection was evaluated as a tool to increase efficiency in the breeding program. A population of 1882 clones from two breeding cycles was genotyped by sequencing resulting in a filtered set of 55k SNPs, providing extensive genome coverage. This population was phenotyped for plot weight, Brix, fiber and sucrose content, with replicated measurements taken on first season crop and ratoon crop harvests. Broad-sense heritabilities ranged from 0.69 to 0.90. Genomic prediction accuracy was assessed with genomic best linear unbiased prediction models in two ways: for clonal prediction of the genotyped clones and for parental prediction of their respective progenitors. In clonal prediction accuracies ranged from 0.07 to 0.39 in cross validation within a breeding cycle, and 0.01 to 0.32 in predictions across cycles. In parental prediction accuracies varied from 0.14 to 0.17 for Brix, and from 0.20 to 0.26 for plot weight. We observed a strong genotype by year interaction effect leading to reduced accuracies when predicting across breeding cycles. The genomic predicted breeding value using progeny data, achieved similar accuracies as clonal prediction. These results could be taken into account in the deployment of genomic selection for a sugarcane breeding program. We also investigated the use of high dosage information in the representation of SNP data from sugarcane. Association analysis and genomic prediction were performed using four fiber traits, for a countinuous marker representation that can represent high dosage of alleles, and for a discrete representation, that is limited in distinguishing heterozygous from homozygous states. We observed an increase in the number of significant hits in association tests when using dosage coding. In genomic prediction, differences were small between continuous and discrete coding, but in most of the cases there was an advantage when using continuous coding.