Mechanical harvesting could help address the challenges related to the cost and availability of labor experienced by Florida’s fresh market tomato industry. However, tomato fruit are very susceptible to bruising, which is likely to be exacerbated by mechanical harvest. Previous studies have found that fruit firmness plays an important role in cultivar resistance against internal bruising. The UF/IFAS tomato breeding program has worked to develop tomato lines with traits important for mechanical harvest, including a compact growth habit (CGH) and increased fruit firmness. To investigate the inheritance and genetic architecture of fruit firmness among CGH lines in the program, bi-parental populations were developed from firm and soft inbred parents. Genome wide association analysis identified multiple significant fruit firmness quantitative trait loci (QTLs) with minor effects, underscoring the quantitative nature of the fruit firmness trait among the population studied. The dissection of fruit firmness variance components revealed mostly additive variance components. Genomic selection (GS) models were successfully trained to predict fruit firmness, demonstrating the viability of GS integration into the UF Tomato Breeding program. Significant gains in prediction accuracy and computational efficiency were achieved through model parameters fine-tuning such as training population size and marker density optimization, and modeling of significant QTLs as fixed effects. This work demonstrates that significant fruit firmness variability exists in the UF/IFAS tomato breeding program germplasm and can be selected for to develop firm-fruited CGH tomatoes intended for mechanical harvest. Furthermore, the successful training of fruit firmness GS models will aid in the efficient development of CGH fresh market tomato cultivars intended for mechanical harvest. The training of multi-trait and multi-environment genomic selection models to leverage the variance-covariance information between traits and between testing environments may result in increased prediction accuracies and needs to be explored.
