Trait-dependent optimization of genotyping effort enhances genomic prediction in Japanese plum

Abstract

Japanese plum (Prunus salicina Lindl.) breeding is constrained by long juvenile phases and the logistical limitations of large-scale genotyping. Single-step Genomic Best Linear Unbiased Prediction (ssGBLUP) provides a unified framework to integrate genomic, pedigree, and phenotypic information offering a potential solution under partial genotyping scenarios. In this context, this study evaluated how varying proportions of genotyped individuals influence genomic prediction performance in a multi-family Japanese plum population. Approximately 800 trees were genotyped using a combined low-density SNP panel with imputed high-density markers and phenotyped over two seasons for the flowering date (FD), harvest date (HV), fruit weight (FW), soluble solid content (SSC), and juice acidity (AD). SsGBLUP models were implemented under eight genotyping proportions (10–100%). Prediction accuracy ($\hat{r}$) and predictive ability (PA) increased with genotyping effort but exhibited trait-dependent plateaus. For FD, theoretical accuracy exceeded 0.90 with only 15% genotyping, while for HV and FW, PA obtained with 50% genotyping was statistically equivalent to that achieved under full genotyping. In contrast, SSC and AD required higher genotyping proportions to approach maximal performance. These findings suggest that optimal genotyping effort is trait-dependent and that reasonable genomic prediction can be achieved without full population genotyping. The study provides preliminary quantitative guidance for implementing single-step genomic prediction in perennial fruit breeding programs under realistic genotyping constraints.