Joint analysis of phenotype-effect-generation identifies loci associated with grain quality traits in rice hybrids

Rice (*Oryza sativa* L.) is a staple food for over half the world's population. Increasingly, consumers prioritize rice cooking quality alongside yield. Many quality traits, like protein content and chalkiness, are correlated with yield. Hybrid rice offers higher yields than inbred varieties, but often requires quality improvement. To achieve high-yield, high-quality hybrids, a thorough genetic dissection of hybrid quality traits is crucial to identify superior crosses. Rice grain quality is categorized into milling, appearance, cooking and eating, and nutritional qualities. Milling quality involves brown rice ratio (BRR), milled rice ratio (MRR), and head rice ratio (HRR), with HRR being a key indicator. Appearance quality, linked to grain length (GL), grain width (GW), GL/GW ratio (GLWR), and endosperm translucency/chalkiness, favors translucent rice. Cooking and eating quality depend on starch properties, including amylose content (AC) and alkali spreading value (ASV), where intermediate AC and ASV are preferred. Protein content (PC) is a significant nutritional aspect, with intermediate levels being optimal. While many grain quality traits in inbred rice are well-studied, with numerous genes cloned and characterized, dissecting these traits in hybrids is more complex due to non-additive effects like dominance. Analyzing these effects requires integrating both inbred and hybrid populations. Existing methods often estimate heterosis as the difference between hybrid and mid-parent values, but these methods are less powerful than those using joint additive and dominant effect models, which have been shown to outperform models focusing solely on separate effects. This study aims to address this gap by developing a statistical pipeline to analyze both additive and dominant genetic effects simultaneously using both inbred and hybrid populations to improve the identification of superior rice hybrids with improved grain quality.

Extensive research exists on rice grain quality traits in inbred varieties. Dozens of genes influencing grain size have been identified, including *GS3*, *GL3.1*, *GW7/GL7* (grain length), *GW2*, *GW5/qSW5*, *GSS* (grain width), and *GS6*, *GS9*, *TGW6*, *GW8/SPL16* (grain size). Genes like *OsRab5a* regulate endomembrane organization and protein trafficking, impacting amyloplast formation and chalkiness. *Chalk5* encodes a vacuolar H+-translocating pyrophosphatase influencing chalkiness. Amylose content (AC), significantly impacting cooking quality, is controlled by genes like *Waxy* and *Wx*, with various alleles (*Wx^a*, *Wx^b*, *wx*, *Ws^x*, *Wx^P*, *Wxⁱ*, *Wx^x*) influencing starch synthesis. *SSIIIa*, a defective soluble starch synthase gene, also plays a role in resistant starch production, which in turn is heavily dependent on *Wx* allele expression. *OsSSIIa* (starch synthase II) is a major determinant of gel temperature. Breeding superior hybrids requires careful selection of parental inbred lines, a process complicated by non-additive genetic effects. Previous studies on other species have highlighted the importance of joint additive and dominant effect models for identifying loci associated with traits. For example, a study on cattle identified dominant loci with larger impacts than those found with additive-only models. In maize, a study identified QTLs categorized into additive, dominant, and epistatic effects. Ideally, a powerful model should simultaneously analyze additive and dominant effects using both parental inbred and hybrid populations.

This study utilized a diallel mating design, crossing 113 indica rice accessions (male parents) with five male-sterile varieties (female parents) to create 565 hybrid test crosses. Phenotypes for 12 quality traits were collected from both parental inbreds and hybrids. Whole-genome sequencing was performed on the parental inbreds to identify single nucleotide polymorphisms (SNPs). Genotypes of hybrids were inferred from parental genotypes. General combining ability (GCA) and heterosis (H) were calculated. A statistical pipeline called Joint analysis of Phenotypes, Effects, and Generations (JPEG) was developed to analyze additive and dominant genetic effects using both original phenotypes (V and T) and derived phenotypes (G and H). The pipeline employs the BLINK multiple loci model within GAPIT to iteratively add associated additive and dominant effects as covariates. This controls for population structure and cryptic associations. Initial covariates included the first three principal components and dummy variables for female parents. Iterations stopped when no additional additive or dominant effects could be added. The study employed various statistical analyses, including assessment of phenotype distributions, correlation analysis of traits, population structure analysis using neighbor-joining trees and principal component analysis (PCA), and a genome-wide association study (GWAS) using the JPEG pipeline. Heritability was estimated for each trait in both inbred and hybrid populations using a mixed linear model. Finally, the accuracy of the prediction of hybrid performance using identified loci was assessed using two-fold cross-validation (CV). Genomic best linear unbiased prediction (gBLUP) was used as a benchmark method for comparison. The JPEG pipeline source code is publicly available on GitHub.

The analysis of 12 rice grain quality traits revealed that most traits showed normal distributions in inbred lines, with some exceptions such as ASV (skewed towards lower end) and PGWC (skewed towards upper end), reflecting selection for quality improvement. Hybrids generally showed values between those of their parents, suggesting additive genetic effects. However, the order of female parent performance did not always remain constant in hybrids, indicating the complexity of hybrid quality traits. Strong positive correlations were observed between CD and PGWC, while moderate negative correlations existed between GL and GW. A total of 7,734,465 raw SNPs were obtained from whole-genome sequencing of 120 parental varieties, with 1,619,588 passing quality control. The average distance between markers was 196.8 bp, and LD decay was observed within ~200 kb. Population structure analysis revealed genetic diversity among the female parents, with some clustering. Four outliers among the test crosses were identified, having origins distinct from the predominantly indica rice varieties. The JPEG pipeline identified 192 significant SNPs, which clustered into 128 loci associated with at least one of the 12 traits. These included 44 loci with additive effects, 97 with dominant effects, and 13 with both. Forty-two loci were located near 17 known genes, including major quality genes like *Wx* (AC), *GS3* (GL, GLWR), and *GWS* (GW, GLWR). The combined analysis of inbreds and hybrids yielded higher power than analyses of either group alone. Heritability estimates varied among traits, with total heritability (additive + dominant) for hybrids serving as the benchmark for prediction accuracy. Pyramiding the identified loci significantly improved the accuracy of hybrid performance prediction compared to gBLUP, explaining over 45% of the total genetic variation for most traits. Cross-validation showed that the JPEG pipeline can accurately predict the phenotypes in a testing population that was not used in training.

This study successfully developed and validated the JPEG pipeline for identifying loci associated with rice grain quality traits in hybrids, successfully incorporating both additive and dominant genetic effects. The identification of 128 loci, many near known genes, significantly advances our understanding of the genetic architecture underlying these complex traits. The higher power of the combined inbred and hybrid analysis highlights the importance of integrating different data sources for a comprehensive understanding. The substantial improvement in prediction accuracy over gBLUP showcases the pipeline's practical value in breeding programs. However, the study's limitation in fully dissecting overdominance due to the relatively small number of loci with both additive and dominant effects suggests that future research could be focused on investigating more complex genetic interactions such as epistasis and incorporating more sophisticated models for this purpose. Also, the limited number of female parents might influence the analysis, and expanding the dataset in this dimension could enhance the accuracy and generalizability of the findings.

This research presents a novel JPEG pipeline for joint analysis of phenotypes, effects, and generations in hybrid rice, successfully identifying numerous loci associated with 12 grain quality traits. This pipeline significantly improves the accuracy of genomic prediction for hybrid performance. The findings provide valuable insights for rice breeding programs to improve grain quality and yield by pyramiding identified loci. Future research could focus on expanding the dataset, incorporating epistatic effects, and applying the JPEG pipeline to other crops.

The study's primary limitation is the relatively small number of female parents used. This might restrict the generalizability of the findings. Furthermore, the analysis focused primarily on additive and dominant effects, potentially overlooking other types of genetic interactions such as epistasis, which could influence the genetic architecture of grain quality traits. The inference of hybrid genotypes from parental data might introduce some uncertainty. The accuracy of genotype imputation could be improved by including more advanced imputation techniques.