Breeding for drought tolerance is a challenging task that requires costly, extensive, and precise phenotyping. Genomic selection (GS) can be used to maximize selection efficiency and the genetic gains in maize (Zea mays L.) breeding programs for drought tolerance. Here, we evaluated the accuracy of genomic selection (GS) using additive (A) and additive + dominance (AD) models to predict the performance of untested maize single-cross hybrids for drought tolerance in multi-environment trials. Phenotypic data of five drought tolerance traits were measured in 308 hybrids along eight trials under water-stressed (WS) and well-watered (WW) conditions over two years and two locations in Brazil. Hybrids' genotypes were inferred based on their parents' genotypes (inbred lines) using single-nucleotide polymorphism markers obtained via genotyping-by-sequencing. GS analyses were performed using genomic best linear unbiased prediction by fitting a factor analytic (FA) multiplicative mixed model. Two cross-validation (CV) schemes were tested: CV1 and CV2. The FA framework allowed for investigating the stability of additive and dominance effects across environments, as well as the additive-by-environment and the dominance-by-environment interactions, with interesting applications for parental and hybrid selection. Results showed differences in the predictive accuracy between A and AD models, using both CV1 and CV2, for the five traits in both water conditions. For grain yield (GY) under WS and using CV1, the AD model doubled the predictive accuracy in comparison to the A model. Through CV2, GS models benefit from borrowing information of correlated trials, resulting in an increase of 40% and 9% in the predictive accuracy of GY under WS for A and AD models, respectively. These results highlight the importance of multi-environment trial analyses using GS models that incorporate additive and dominance effects for genomic predictions of GY under drought in maize single-cross hybrids.

-

-

Water deficit is one of the most common causes of severe crop‐production losses worldwide in maize (Zea mays L.). The main goal of this study was to infer about genotype × environment interaction (G × E) and to estimate genetic correlations between drought tolerance traits in maize using factor analytic (FA) multiplicative mixed models in the context of multi‐environment trial (MET) and multi‐trait multi‐environment trial (MTMET) analyses. The traits measured were: grain yield (GY), ears per plot (EPP), anthesis‐silking interval (ASI), female flowering time (FFT), and male flowering time (MFT). Three‐hundred and eight hybrids were evaluated in a total of eight trials conducted under water‐stressed (WS) and well‐watered (WW) conditions across 2 yr and two locations in Brazil. For most of the traits (GY, ASI, and FFT), the magnitude of the genetic variances differed across WS and WW conditions. Genetic correlations between water conditions for FFT and MFT were 0.81 and 0.82, respectively, indicating that it might be unnecessary to measure these traits in both water conditions. Grain yield and EPP showed moderate to high G × E, with genetic correlations of 0.57 and 0.39 between WS and WW conditions, respectively, which suggested that gene expression was not consistent across different water regimes. Therefore, it is necessary to evaluate these traits under both water conditions. Genetic correlations between pairs of traits, in general, were higher under WS conditions compared with WW conditions. Grain yield exhibited moderate correlations with EPP (r = 0.62) and FFT (r = −0.42) under WS conditions. The FA models can be a useful tool for MET and MTMET analyses in maize breeding programs for drought tolerance.

ABSTRACT. We evaluated the phenotypic and genotypic stability and adaptability of hybrids using the additive main effect and multiplicative interaction (AMMI) and genotype x genotype-environment interaction (GGE) biplot models. Starting with 10 singlecross hybrids, a complete diallel was done, resulting in 45 doublecross hybrids that were appraised in 15 locations in Southeast, Center-West and Northeast Brazil. In most cases, when the effects were considered as random (only G effects or G and GE simultaneously) in AMMI and GGE analysis, the distances between predicted values and observed values were smaller than for AMMI and GGE biplot phenotypic means; the best linear unbiased predictors of G and GE generally showed more accurate predictions in AMMI and GGE analysis. We found the GGE biplot method to be superior to the AMMI 1 graph, due to more retention of GE and G + GE in the graph analysis. However, based on cross-validation results, the GGE biplot was less accurate than the AMMI 1 graph, inferring that the quantity of GE or G + GE retained in the graph analysis alone is not a good parameter for choice of stabilities and adaptabilities when comparing AMMI and GGE analyses.