Structure and function of rice hybrid genomes reveal genetic basis and optimal performance of heterosis

Heterosis, or hybrid vigor, is a crucial phenomenon in agriculture where the F1 generation of heterozygous plants exhibits superior performance compared to its homozygous parental lines. The exploitation of heterosis has led to significant yield increases in crops like maize and rice since the 1930s and 1970s, respectively. This is particularly important given the increasing global population and climate change pressures on food production. Three main hypotheses – dominance, overdominance, and epistasis – attempt to explain the genetic basis of heterosis. While studies have provided evidence supporting these hypotheses in various plants (e.g., tomato, sorghum, *Arabidopsis*, maize), a comprehensive genome-wide analysis of rice hybrids is lacking. This study aims to fill this gap by performing a large-scale genomic analysis of a diverse collection of rice hybrids and their progeny, offering a dynamic perspective on heterotic loci and breeding history. The study investigates the genetic basis of phenotypic changes in yield-related traits, flowering time, and grain quality, focusing on both *indica-indica* and *indica-japonica* hybrids. Finally, a genomic selection model is developed to aid in predicting and optimizing the performance of future hybrid combinations, furthering the understanding and application of heterosis in rice breeding.

Previous research has identified several quantitative trait loci (QTLs) contributing to rice heterosis, but a comprehensive analysis scanning genome-wide breeding footprints in a dynamic context was missing. Studies have shown the role of overdominance or pseudo-overdominance in driving single-locus heterosis in some plants, and genetic complementation (dominance) in others. Large-scale genomic analyses have provided evidence for the partial dominance of heterozygous loci in yield traits, suggesting that multiple heterozygous genomic loci contribute non-linearly to heterosis in rice. However, a comprehensive analysis combining extensive phenotypic and genomic data to systematically study heterosis across the history of rice hybrid breeding has been lacking until now.

This study utilized 2,839 rice hybrid cultivars released between 1976 and 2020, representing the majority of commercially used hybrids in China. These hybrids were resequenced at an average depth of 35-fold, yielding high-quality SNP and InDel data. To further investigate heterotic loci, 18 representative hybrids were selected to generate 9,839 F2 progeny, which were also sequenced and genotyped. Phenotypic data were collected for both hybrids and F2 individuals, encompassing heading date, morphological traits, yield-related traits, and grain quality. The data were analyzed through several methods including: calculation of nucleotide diversity, phylogenetic analysis, principal component analysis (PCA), admixture analysis, genome-wide association studies (GWAS) using a mixed linear model in TASSEL, identification of japonica introgression, QTL mapping (linkage mapping and association analysis), calculation of dominance-effect/additive-effect (d/a) index, estimation of phenotypic variance explained (PVE), and construction of a genomic selection model using the GBLUP method. A web-based tool was also created to provide easy access to the data.

The study divided the rice hybrid breeding history into three periods (1976–2000, 2001–2010, and 2011–2020) based on policy changes, technological advances, and the increased use of two-line hybrids. Analysis revealed that *indica-indica* hybrid improvement involved broadening genetic resources, pyramiding favorable alleles, and eliminating inferior alleles at negative dominant loci. In *indica-japonica* hybrids, widespread genetic complementarity contributed to heterosis in yield traits, with dominance effect loci having a greater impact than overdominance. GWAS identified several known and novel loci associated with various traits. Improvement breeding showed a trend toward accumulating more favorable alleles for most traits, resulting in larger source and sink organs, slightly shorter heading dates, and improved grain quality. Analysis of japonica introgression in *indica-indica* hybrids revealed that the introduction of favorable japonica alleles played a role in the improvement process. The construction of a genomic selection model using GBLUP achieved high prediction accuracy for several traits and showed its applicability across different rice hybrid subgroups. The model also enabled the identification of potential hybrid combinations that exceed the performance of existing cultivars.

The findings highlight the complexity of rice hybrid breeding, showing it as a multi-stage process involving genetic diversity expansion, allele pyramiding, and parental line improvement. The increase in genetic diversity contrasts with the usual reduction seen in inbred lines. The shift towards more two-line hybrids broadened the genetic base by introducing japonica introgression, providing favorable alleles. The genomic selection model offers a powerful tool for breeders to efficiently optimize hybrid combinations, potentially accelerating breeding cycles and reducing costs. Although the model used an additive plus dominance model, the inclusion of dominance did not improve prediction accuracy, suggesting limitations in current models for handling non-additive effects. Future research should focus on more sophisticated models and a deeper understanding of the genetic architecture to fully exploit heterosis in rice.

This study provides a comprehensive genomic analysis of rice hybrid breeding, revealing the genetic basis of heterosis and offering a powerful genomic selection model for future breeding efforts. The findings demonstrate the success of strategies like broadening genetic resources, pyramiding favorable alleles, and improving parental lines. The developed genomic model significantly enhances the efficiency of hybrid rice breeding, accelerating improvement and potentially leading to greater food security. Future research should address the limitations of current models in capturing non-additive effects and further explore the complex genetic interactions underlying heterosis.

The study's main limitation is that the model's accuracy in predicting heterosis relies heavily on the quality and quantity of the training dataset. The model may not be perfectly generalizable to other populations or environments. The study mainly focused on Chinese rice hybrids, and its findings may not fully apply to hybrids developed in other regions. Additionally, the analysis primarily focused on yield and quality; additional traits should be incorporated for a more comprehensive understanding. The relatively small number of indica-japonica hybrids compared to indica-indica hybrids could affect the strength of conclusions drawn about intersubspecific heterosis.