Longitudinal genomic prediction of cacao disease resilience identifies robust witches' broom disease targets.

Introduction: Genomic studies in cacao ( Methods: We analyzed repeated-harvest phenotypes from 102 cacao accessions to derive longitudinal targets describing baseline performance and temporal change for healthy pod rate, frosty pod rot (FPR), and witches' broom disease (WBD). Genomic prediction was evaluated using random and relatedness-blocked cross-validation with single nucleotide polymorphism (SNP) marker sets aligned to the Criollo and Matina reference genomes. Robustness was assessed across trajectory model specifications, cross-validation block structures, and multi-trait selection indices. Results: Mixed-model analyses demonstrated temporal repeatability ranging from 0.78 to 0.89 across four harvests, with particularly high accession-level consistency for WBD traits. Genomic prediction models showed that WBD-derived targets, particularly branch and flower-cushion WBD symptom components, yielded more consistent predictive performance than FPR or healthy pod rate. These signals remained supported after 10,000-label permutation testing, Benjamini-Hochberg false-discovery-rate correction, and multiple robustness analyses. Cross-reference comparisons indicated that WBD predictability was robust to the choice of reference genome, whereas other traits exhibited higher reference sensitivity. Finally, Pareto-based multi-trait prioritization identified a stable shortlist of candidate accessions optimizing both disease resilience and productivity. Discussion: Repeated-harvest disease phenotypes harbored reproducible accession-level temporal structure, with WBD-related longitudinal traits remaining genomically predictable across diverse validation settings. These findings indicate that repeated-harvest trajectory targets provide useful genomic-prediction endpoints for prioritizing resilient germplasm in perennial cacao breeding programs, while emphasizing the need for validation in larger multi-environment trials.