Abstract
Genomic prediction (GP) in mango breeding faces challenges due to the species' complex biology, long cycles, and limited reference populations. To accelerate genetic improvement, this study integrated data from diverse global populations to increase the reference population size. It included three mango collections reserved in Australia (225), USA (161), and China (224), totaling 610 individuals. Fruit weight (FW) and total soluble solids (TSS) were measured in multiple datasets, while several other traits were measured in specific datasets. We evaluated genetic diversity, performed genome-wide association studies (GWAS), and assessed GP accuracy using standard, genotype-by-environment (GxE), and multitrait models, both within and across collections. Findings revealed a highly admixed genetic structure, with faster linkage disequilibrium (LD) decay in the Chinese collection, indicating higher genetic diversity. Data integration significantly enhanced GWAS power, identifying 19 quantitative trait loci (QTLs) for FW and 9 for TSS. GxE models consistently achieved higher or comparable prediction accuracies for FW and TSS compared to the non-GxE models, especially when combining Australian and US collections. This was not the case when predicting into or from the Chinese collection, mostly due to differences in the phenotyping protocol. While single-trait models performed comparably to multitrait models in predicting new individuals (Cross-Validation: CV1), multitrait models significantly improved prediction accuracy in scenarios with incomplete phenotypic records (CV2). This study demonstrates that strategic global data integration significantly enhances GWAS power and GP accuracy in mango. This collaborative approach is crucial for developing more efficient and accelerated breeding programmes for mango and other perennial trees.