Simulations, genomic prediction, and environmental modeling to support Eucalyptus breeding decisions

Abstract

Eucalyptus urophylla S.T. Blake is the most widely planted tree species in Brazil. However, long breeding cycles and strong genotype × environment (G×E) interactions may limit genetic gains in this species. This study integrated empirical approaches and in silico simulations to evaluate strategies aimed at increasing selection efficiency, reducing operational costs, and optimizing genotype recommendation. Through simulations, in the first Chapter, we investigated the impact of genomic selection (GS) model recalibration, considering different training set sizes, proportions of phenotyped individuals, numbers of previous generations included, and genetic architectures varying in heritability, number of QTLs, and the presence of additive and dominance effects. We also compared recurrent phenotypic selection strategies, such as traditional progeny testing (TPT) and clonal progeny testing (CPT), applied either alone or in combination with GS and Optimal Cross Selection (OCS). The results indicated that the number of evaluated individuals and phenotype quality are decisive factors for the long-term sustainability of GS, with the CPT + OCS combination providing the best balance between genetic gain and cost. In the empirical studies (Chapters 2 and 3), open-pollinated families of E. urophylla were evaluated across 18 trials distributed in eight Brazilian states. The incorporation of environmental covariates derived from NASA POWER and SoilGrids enabled the construction of environmental similarity matrices and the explicit modeling of G×E interactions through genotype × environmental covariate interaction (G×W). The environmental covariates were further summarized into more informative environmental factors and selected to maximize their representation of the G×E interaction matrix. In Chapter 3, genomic information was incorporated into this population through bulk genotyping. Models including G×W interactions achieved higher predictive ability, particularly for prediction in new environments (CV0). Bulk genotyping used for Genome-Wide Family Prediction (GWFP) showed high correlation with individual allele frequencies while reducing genotyping costs by 24-fold; however, predictive gains were modest. In contrast, environmental modeling showed consistent performance and played a central role in spatial extrapolation, enabling recommendation maps and more informative deployment strategies. Overall, the results demonstrate that the combined use of genomic tools, environmental modeling, and optimized crossing strategies can accelerate genetic gains, enhance predictive robustness across multiple environments, and rationalize operational costs in forest breeding programs.