Style Length in Wild Rhododendron: From Cryptic Genetic Variation to Efficient Genome Prediction Strategy

<title>Abstract</title> <p> Background Style length (SL) is a pivotal reproductive trait influencing mating systems and speciation in flowering plants. However, the genetic basis of continuous SL variation in wild plant species remains largely unexplored. Here, we dissected the genetic architecture of SL using a natural hybrid complex comprising <italic>Rhododendron delavayi</italic> , <italic>R. irroratum</italic> , and their hybrid derivative <italic>R. agastum</italic> (n = 260). Results Phenotypic analysis revealed continuous variation and high narrow-sense heritability ( <italic>h</italic> ^{<italic>2</italic>} = 0.34–0.39) in hybrid-involved populations, contrasting sharply with the low heritability ( <italic>h²</italic> = 0.04) estimated when the two parental species were pooled. This contrast suggests that hybrid populations harbor greater additive genetic variance for style length than the parental species pool. Genome-wide association study (GWAS) identified only a limited number of weak-effect loci (min <italic>P</italic>  = 9.17 × 10⁻⁸) without any major quantitative trait loci (QTLs), while population differentiation (FST) confirmed widespread genomic divergence between the parental species. Genomic selection (GS) analysis further supported a highly polygenic genetic architecture: prediction accuracy increased monotonically with the inclusion of GWAS-relaxed markers and reached a plateau at an optimal threshold of <italic>P</italic>  &lt; 10 ^− 3.5 (2,615 SNPs). This marker set achieved prediction accuracy comparable to that obtained using the whole-genome marker set (5.3 million SNPs) while substantially reducing genotyping costs. Comparative evaluation of seven GS models further demonstrated the robustness of GBLUP for predicting this polygenic trait. Conclusion Our findings indicate that SL in <italic>Rhododendron</italic> is controlled by a typical polygenic architecture dominated by numerous minor-effect loci. Furthermore, a GWAS-informed marker prioritization strategy provides an effective balance between prediction accuracy and genotyping cost, offering a practical framework for genomic prediction and breeding in long-generation woody ornamental plants. </p>