Unraveling the impact of genotype by environment interaction complexity and a new proposal to understand the contribution of additive and non-additive effects on genomic prediction in tropical maize single-crosses

Abstract

Unraveling the impact of genotype by environment interaction complexity and a new proposal to understand the contribution of additive and non-additive effects on genomic prediction in tropical maize single-crosses The use of molecular markers to predict non-tested materials in field trials has been extensively employed in breeding programs. The genomic prediction of single crosses is a promising approach in maize breeding programs as it reduces selection cycle and permits the selection of promising crosses. Accounting for nonadditive effects on genomic prediction can increase prediction accuracy of models depending on the traits genetic architecture. Genomic prediction was first developed for single environments and recently extended to exploit the genotype by environment interactions for prediction of non-evaluated individuals. The employment of multi-environment genomic models is advantageous in several aspects and has enabled significant higher prediction accuracies than single environment models. However, only a small number of studies regarding the inclusion of non-additive effects in these models are reported. Moreover, the genotype by environment interaction complexity can largely impact the prediction accuracy of these models. Thus, the objectives were to i) evaluate the contribution of additive and non-additive (dominance and epistasis) effects for the prediction of agronomical traits with different genetic architecture in tropical maize single-crosses grown under two nitrogen regimes (ideal and stressing), and ii) verify the impact of the genotype by environment interaction complexity, and the inclusion of dominance deviations, on the prediction accuracy of hybrids grain yield using a multienvironment prediction model. For this, we used phenotypic and genotypic data of 906 single-crosses evaluated during two years, at two locations, under two nitrogen regimes, totaling eight contrasting environments (combination of year x locations x nitrogen regimes). The traits considered in the study were grain yield, ear, and plant height. The results regarding the inclusion of additive and non-additive effects (dominance and epistasis) in genomic prediction models suggest that non-additive effects play an important role in stressing conditions, having a high, medium and low contribution for phenotypic expression of grain yield, plant height, and ear height, respectively. The inclusion of dominance deviations in multi-environment prediction model increases the prediction accuracy. Furthermore, a linear relationship between genotype by environment complexity and prediction accuracy was found.