Genetic architecture and predictive modeling of heading date in japonica rice cultivars

Directional selection within elite breeding pools has narrowed the genetic base, yet phenotypic diversity persists among modern cultivars. Here, we analyzed 257 Korean japonica rice cultivars to dissect the genetic and environmental determinants of heading date. A genome-wide association study (GWAS) identified Hd1, Ghd7, OsPRR37, and Hd16 as major-effect genes in the population. Given the central role of Hd1 in variation of heading date, subpopulation-specific GWAS based on Hd1 functionality was performed to uncover minor-effect variants. Regression models using alleles identified in GWAS explained a substantial portion of phenotypic variance and achieved higher predictive accuracy than genome-wide SNP-based prediction within this population. Allele-specific genomic prediction with the additive linear model LASSO closely matched the nonparametric machine learning model XGBoost, indicating that additive effects largely accounted for variation in heading date. A redefined accumulated temperature index (ATI) model further enabled estimation of cultivar-specific thermal requirements within an optimized developmental window. We also discuss how the Hd1-Type14 allele was preferentially utilized for developing early maturing cultivars in Korean breeding programs. These findings demonstrate that integrating allele-based and environment-based prediction provides an effective framework for improving breeding precision for regional adaptation.