Enhancing genomic selection in rubber tree (Hevea brasiliensis): Exploring the impact of genetic relatedness and QTL integration

Rubber tree breeding faces significant challenges, mainly due to the low female fertility of the plant, the long breeding cycles and the complex trait architecture. The advent of genomic selection offers a significant opportunity to explore more effective breeding strategies and accelerate genetic gains. This study investigates genomic prediction strategies for Hevea brasiliensis breeding, focusing on two biparental families connected by a common parent and evaluated among four different sites. The objective was to assess the impact of using full-sib and half-sib populations on prediction accuracy for key traits, rubber production and sucrose content. Results confirmed that prediction accuracies are higher when the training and validation populations consist of full-sibs (0.54 for rubber production and 0.36 for sucrose), as compared to half-sibs (0.17 for rubber production and 0.21 for sucrose). Combining full-sibs and half-sibs in the training population yielded prediction performance comparable to intra-family models (0.52 for rubber production and 0.37 for sucrose), providing a more practical option for breeding programs. Additionally, the integration of QTL information into prediction models for rubber production did not improved accuracy in full-sib (0.53) or half-sib (0.16) validation approaches and reduced accuracy in cross-validation, likely due to the polygenic nature of the trait and genotype-by-environment interactions. Rubber tree breeding programs could benefit from constructing training populations composed of multiple related families, simplifying logistics while maintaining prediction accuracy across more diverse populations than single biparental families. This approach offers a promising pathway to enhance the efficiency and genetic gains in rubber tree genomic selection. • Genomic prediction accuracy for rubber production reached 0.54 within single families. • Sucrose content prediction achieved 0.36 accuracy within single families. • Combining full- and half-sibs preserves prediction accuracy. • A single training population of related families is recommended.