Genomic prediction strategies for grain yield stability in second season maize hybrids

Abstract

The genotype by environment interaction (GxE) is a major factor in maize breeding. Therefore, it is essential to select genotypes that are stable across locations and over the years. Genotype stability is even more important for second season maize breeding programs. The use of genomic prediction tools in maize breeding has been frequent. Genomic selection not only reduces the time required per breeding cycle but also allows to study of a higher number of genotypes without significantly increasing the phenotyping costs. Several reports highlight the advantages of including the GxE in the prediction models. However, to predict maize stability has been scarcely reported. Given the above, the present work was carried out aiming to verify the feasibility of predicting second season maize stability and to define simple and efficient strategies to deal with real scenarios of multi-environment trials. Two research works were performed. For this study, a private maize breeding dataset was used. Over 1300 maize hybrids were assessed across 12 environments during the 2012/13 and 2013/14 second seasons. The dataset was split into three: 1) 128 hybrids that were common across the six 2012/13’s environments; 2) all 309 hybrids assessed in the six 2012/13’s environments; 3) all 710 hybrids assessed in the six 2013/14’s environments. Dataset 1 was used in the first research work, while datasets 2 and 3 in the second one. In the first work, the predictions of nine adaptability and stability indices were compared to a multi-environmental approach under a genetically balanced (across the environments) scenario. A BRR (Bayesian Ridge Regression) model was used, and the predictive abilities were measured via cross-validation (10-fold). The Euclidean Distance and MHPRVG (harmonic mean of the relative performance of the breeding values) indices outperformed the multi-environmental approach. In the second work, the BRR model was maintained, however, four prediction scenarios were considered: i) single-environment; ii) stability indices; iii) Multi-environmental (ME) without including GxE, and iv) ME including GxE effects. In addition, it was considered two cross-validation schemes: CV1 (10-fold), and CV2 (whole environment predictions, ME only). The Euclidian distance index did not prove feasible, on the other hand, the MHPRVG results were consistent for both datasets. Considering CV1, the inclusion of GxE was not advantageous, increasing the time required for predictions either without (2012/13) or with marginal gains (2013/14) in predictive ability. Considering CV2, iii e iv presented very poor predictability. In general, considering the second work, the ME approach (iii) outperformed the stability indices (ii), however, in both works, the use of MHPRVG index has proved feasible as a strategy to predict the stability of second season maize hybrids.