The ECPGR European Evaluation Network (EVA) for Maize involves genebanks, research institutions, and private breeding companies from nine countries focusing on the valorization of maize genetic resources across Europe. This study describes a diverse collection of 626 local landraces and traditional varieties of maize (Zea mays L.) from nine European genebanks, including criteria for selection of the collection and its genetic and phenotypic diversity. High-throughput pool genotyping grouped the landraces into nine genetic groups with a threshold of 0.6 admixture, while 277 accessions were designated admixed and likely to have resulted from previous breeding activities. The grouping correlated well with the geographic origins of the collection, also reflecting the various pathways of introduction of maize to Europe. Phenotypic evaluations of 588 accessions for flowering time and plant architecture in multilocation trials over three years confirmed the great diversity within the collection, although phenotypic clusters only partially correlated with the genetic grouping. The EVA approach promotes conservation of genetic resources and opens an opportunity to increase genetic variability for developing improved varieties and populations for farmers, with better adaptation to specific environments and greater tolerance to various stresses. As such, the EVA maize collection provides valuable sources of diversity for facing climate change due to the varieties’ local adaptation.

Forecasting vulnerability of cultivated and wild species to climate changes is highly challenging. Evolutionary genomic models enable the prediction of mal-adaptation (genomic offset - GO) across environments and future climates under the assumption that populations are currently locally adapted but do not predict but the resulting phenotypic changes. To do so, we developed a new genomic prediction model (GP) integrating both genomic offset (GO) and within-population gene diversity (Hs) to capture genotype by environment interaction and inbreeding effects, respectively (GP-HO-Hs). As proof of concept, we applied this GP-GO-Hs model to a collection of 397 maize populations (landraces) evaluated across 25 environments in Europe using high-throughput DNA pool genotyping. GP-GO-Hs model accurately predicted yield, plant height and flowering time. It increased by 13% the predictive abilities of GP model for predicting yield of new landraces in new environments. GP-GO-Hs model also predicted that the more diverse the landrace, the more stable its agronomic performance across environments. GP-GO-Hs model generated phenotypic adaptive landscapes for each landrace in future climatic scenarios, enabling the identification of landraces with enhanced potential to adapt to future or emerging cultivation conditions. This GP-GO-Hs model could be easily applied to other wild and cultivated species. Teaser: Identify promising landrace adapted to new and future environments by combining genomic selection and offset