Genomic prediction using phenotypes from pedigreed lines with no marker data

Abstract

Until now, genomic prediction (GP) in plant breeding has only used information from individuals that have been genotyped. Information from nongenotyped relatives of genotyped individuals can also be used. Single-step GP combines marker and pedigree information into a single relationship matrix to perform GP. The objective of this study was to evaluate single-step GP in a wheat breeding program. We compared the performance of pedigree-based, marker-based, and single-step models (ABLUP, GBLUP, and HBLUP, respectively). Data consisted of 1176 genotyped (via genotyping-by-sequencing) and 11,131 nongenotyped wheat lines replicated in five management environments at the CIMMYT experiment station in Obregon, Mexico. Analyses involved three scenarios: (i) all lines had pedigree information but only some were genotyped, with phenotypes from one or two environments in the 2011–2012 season, (ii) all lines had genotype and pedigree information and phenotypes from four or five environments in the 2012–2013 season, and (iii) the combination of Scenarios 1 and 2. Prediction accuracies were calculated by fivefold cross validation on plant height, maturity, heading date, and grain yield. The single-step HBLUP outperformed GBLUP and pedigreebased ABLUP in all cases. We conclude that the single-step procedure combining pedigree and genomic marker data should be favored where appropriate data is available for GP in wheat breeding programs.