A Reference-Free K-mer Framework Enhances Genomic Prediction in Highly Heterozygous Woody Species: A Case Study in Litsea cubeba

Abstract

Abstract Litsea cubeba, an economically important woody species in the Lauraceae, is widely cultivated for spice and essential oil production. Core agronomic traits, particularly fruit morphology and yield per plant, directly determine its commercial value. However, genetic improvement of complex traits in this perennial species is hindered by intrinsic biological constraints, including a prolonged juvenile phase and an extended generation interval. Moreover, the genetic architecture and regulatory mechanisms underlying key agronomic traits remain poorly resolved. Conventional single nucleotide polymorphism (SNP)-based approaches, which depend on a single reference genome, often fail to capture large structural variants and non-reference sequences, thereby limiting the predictive performance of genomic selection (GS). To address these limitations, we performed SNP- and K-mer-based genome-wide association analyses to dissect the genetic basis of coordinated fruit morphological development and biomass accumulation. The results indicated that the K-mer strategy not only recapitulated most SNP-associated signals but also uniquely captured a novel locus associated with the fruit shape index. Additionally, we implemented a reference-free K-mer-based genomic prediction framework to overcome reference bias and incorporate additional genetic variation. Compared with SNP-based baseline models, the K-mer strategy improved prediction accuracy for key agronomic traits by 4.48%–7.71%. Collectively, this study elucidates the polygenic architecture and pleiotropic regulatory networks governing core agronomic traits in L. cubeba and demonstrates that reference-free K-mer-based strategies can enhance genomic prediction performance. These findings provide a conceptual and methodological framework for GS-assisted molecular breeding in highly heterozygous woody species.