Résumé
Due to their local adaptation to contrasted environments, maize landraces are a valuable source of genetic diversity for facing climate change and low input agriculture challenges, however they are underutilized and undercharacterized. High-throughput pool genotyping (HPG) has proven to be efficient to characterize genetic structure, gene diversity (Hs) and identify selective sweeps in landraces. Here, we investigated the interest of genomic prediction (GP) and genomic offset (GO) approaches using HPG to predict adaptation of landraces to various environments. To do so, 397 European landraces were evaluated for yield, plant height, and flowering time in the ECPGR multi-environment trials of the European Evaluation Network (EVA) network including 25 environments. The objectives were to : i) evaluate the effect of Hs and GO on yield, height and flowering, ii) predict these traits using GP under different cross-validation (CV) scenarios, and iii) evaluate the effect of including Hs and GO in the GP model. Gene-diversity of landraces (Hs) was positively correlated with yield (R2 = 0.29) and plant height (R2 = 0.28) while GO was negatively correlated with these two traits. GP yielded high predictive abilities that varied according to cross-validation scenario, reaching a minimum of 0.52 when predicting new genotypes in new environments. Integrating Hs and GO in the GP model increased predictive abilities by up to 13.2% when predicting new genotypes in a new environment. Our results show that GP using HPG is a promising tool to predict the value of maize landraces in genebanks, notably their adaptation to various environments, and that including gene diversity and genomic offset in the prediction model could improve predictive abilities.