Genome-wide prediction for maize single-cross hybrids using the GBLUP model and validation in different crop seasons

Abstract

The aim of this study was to perform genome-wide selection using a set of Dart-seq markers associated to the additive-dominant genomic best linear unbiased prediction (GBLUP) model to predict maize grain yield in different crop seasons and locations. Genotyping was performed with Dart-seq markers from 447 lines coming from a germplasm bank of a private maize breeding company. Crossing these lines provided 838 single-cross hybrids evaluated in six locations in the winter crop season of 2013 and 797 single-cross hybrids evaluated in four locations in the summer crop season of 2013/2014. Four k-fold levels were applied on the full panel of 23,153 Dart genotyping-by-sequencing markers and samples of 50% of the available markers. The different crop seasons were used as training and validation populations to estimate the predictive accuracy. The magnitude of the correlations between predicted and observed hybrids ranged from 0.82 to 0.89 in the winter crop season and from 0.56 to 0.76 in the summer crop season. The correlations between combinations tested in different crop seasons and locations were encouraging (0.53). Predictive ability was highly influenced by the genetic background and also by the interaction between crop seasons. The coincidences between the genomic values of the summer crop and winter crop, in terms of discard, were 89 and 90%. This result shows the possibility of using genomic prediction in breeding programs for initial discard of low-yielding genotypes. The GBLUP method was able to generate high correlations between predicted and observed hybrids, even at high levels of missing in k-fold and in different locations and crop seasons.