The identification of grain size genes by RapMap reveals directional selection during rice domestication

The study addresses the long-standing challenge of efficiently identifying causal genes underlying quantitative trait loci (QTL) in crops, focusing on rice grain size (length, width, and shape). Traditional bi-parental mapping and map-based cloning are reliable but time- and labor-intensive and often capture only a fraction of natural phenotypic variation. Alternative multi-parental approaches (GWAS, MAGIC, NAM) improve mapping resolution and diversity but still struggle to pinpoint causal genes and require complex, lengthy population development. Grain size is a key domestication and breeding trait with limited known QTGs and unclear regulatory interactions. The authors propose RapMap, a rapid mapping and cloning strategy using F2 gradient populations (F2GPs) derived from minor-phenotypic-difference parents, guided by a co-segregation standard for single-locus models, to accelerate discovery of multiple QTGs and reveal domestication signatures at population scale.

The paper reviews quantitative genetics methods used in crops: bi-parental mapping (F2, RIL, DH, BC) and map-based cloning; GWAS leveraging historical recombination but limited by population structure and LD; and multi-parental designs (MAGIC, NAM) that enhance recombination and reduce structure but are costly and slow to develop, with few causal genes isolated to date. Prior rice grain-size genes involve pathways such as G protein signaling, BR/IAA/CK signaling, peptide/MAPK signaling, ubiquitin-proteasome, epigenetics, and transcriptional regulation, yet only ~15 QTGs were known, leaving interactions and population-level variation unresolved. Archaeological and evolutionary studies indicate seed enlargement as an early domestication trait, but the molecular basis and selection dynamics for rice grain size at population scale remained unclear.

RapMap strategy and implementation: 1) F2 gradient populations (F2GPs): Assemble a diverse rice mini-core collection (541 accessions). Define a phenotypic difference index (PDI) and select minor-phenotypic-difference parents (PDI typically 8–25%, threshold ≤20%) across gradient phenotype groups. Construct 6–9 F2GPs in the first round by crossing adjacent gradient parents, maximizing the chance of single-locus segregation. Predict minimum crosses via N = PDI(core collection)/PDI(threshold). 2) Bulked segregant analysis (BSA): Measure target trait in ~200 F2 plants per cross. Create high- and low-value DNA bulks (top/bottom ~15%). Genotype bulks using RICE6K SNP arrays or whole-genome sequencing (QTL-seq) to detect SNP/InDel clusters linked to causal loci. 3) Co-segregation standard (quality control): Define 12 single-locus genetic models (complete dominance, semi-dominance, over-dominance) sharing the criterion that homozygous genotypes (AA, aa) co-segregate with distinct phenotypic classes in the segregating population. Use flanking InDel/SNP markers to genotype all individuals and confirm each candidate QTL and simultaneously identify NIL-like lines (NIL-LLs) as heterozygotes at the locus. If F2 fails the standard, derive F3/F4 families with clear segregation. 4) Fine mapping and cloning: From heterozygous NIL-LLs, screen sufficient recombinants (KASP high-throughput genotyping; rapid NaOH-based DNA extraction). Delimit QTL to small intervals and identify candidate ORFs using comparative sequencing (functional variants), expression data, and annotations. Validate causal genes via NIL-LL phenotyping, gene editing (CRISPR/Cas9), and complementation with native promoters. 5) Phenotyping and analyses: Grain size traits measured on air-dried, filled grains (n=10 per plant). Multiple linear regression across 541 accessions to estimate gene effects and predictive ability; 10-fold cross-validation. Population genetics using wild (n=446), landrace (n=2462), and cultivar (n=784) datasets: allele frequencies, nucleotide diversity (π), selective sweep scans in 2 Mb windows (100 kb window, 20 kb step), π ratios within 500 kb around loci, and GWAS (FaST-LMM) for association significance.

• Using one round of RapMap with 15 F2GPs over three years, eight grain-size genes were cloned: grain length genes GS3 (including two new alleles GS3-5 and GS3-6), GL7, GS2, and a novel gene GL1 (LOC_Os01g63930, cytochrome P450); grain width genes GW5, GW8, GW7, and a novel gene GW5.1 (LOC_Os05g25350, receptor-like kinase). - Grain-length QTLs mapped to chromosomes 3, 3, 3, 3, 7, 1, 7, and 2 (Crosses 1–8) with PVEs in their respective F2GPs of 44%, 85%, 90%, 79%, 94%, 50%, 58% and 83%; fine-mapping intervals of 20, 18, 14, 54, 24, 19, 56, and 3.7 kb, respectively. - Grain-width QTLs mapped to chromosomes 8, 8, 8, 7, 5, 5, and 5 (Crosses 1–7) with PVEs of 61%, 51%, 58%, 90%, 60%, 53%, and 93%; fine-mapping intervals of 168, 36, 168, 24, 28, 18, and 18 kb, respectively. - Novel functional variants: GL1 exhibits expression-level differences among haplotypes (WZ1 higher expression correlates with longer grains). NIL(WZ1) grains were 7.4% longer and 8.3% heavier than NIL(9311). Complementation of GL1WZ1 increased grain length and weight; CRISPR knockout in ZH11 showed no change, consistent with a weak/null allele background. - GW5.1: An 11.8 kb LTR-retrotransposon insertion disrupts ORF1 (RLK) in the wide-grain allele (Zhenshan 97/AA), causing loss of function. NIL(gw5.1) had 14.9% wider grains and 12.9% heavier weight versus NIL(GW5.1). CRISPR frameshift mutants increased width; complementation decreased width, confirming ORF1 as a negative regulator of grain width. - Across 541-accession mini-core, multiple regression with functional variants: four length genes (GL1, GS2, GS3, GL7) explained 67.4% of grain-length variation; four width genes (GW5, GW5.1, GW7, GW8) explained 66.8% of grain-width variation; all eight explained 77.2% of grain shape (length-width ratio). Major contributors: GS3 (62.5% of length) and GW5 (44.3% of width); GS3 and GW5 explained 55.0% and 47.5% of shape, respectively. - Prediction performance (10-fold CV): correlation r=0.82 for length (CV R2=0.669), r=0.79 for width (CV R2=0.663), r=0.87 for shape (CV R2=0.762). - Domestication signals: Stepwise accumulation of long- and slender-grain alleles from wild to landrace to cultivar, especially in indica. Average grain length and length-width ratio increased and width decreased from landrace to cultivar. Nucleotide diversity (π) reduced at grain-size loci through domestication. - Selective sweeps in cultivars: pronounced π valleys around GS3 (~263 kb, long-grain gs3), GW5 (~307 kb, wide-grain gw5), and GW7/GL7 (~229 kb). Within 500 kb windows, average π ratios (target/alternative allele) were significantly lower than chromosome-wide: GS3 0.14 vs 0.74; GW5 0.27 vs 0.85; GW7/GL7 0.42 vs 1.73 (P<2.2×10^-16), indicating strong artificial selection. - Correlations: Strong positive correlations among phenotypic effect (PVE), GWAS association significance, selection strength, and DNA variation across the eight genes (e.g., r≈0.94 between PVE and association significance; r≈0.85 between DNA variation and selection). RapMap complemented GWAS by identifying rare/minor-effect genes (GS2, GL1).

RapMap addresses the core challenge of rapidly isolating causal QTGs underlying complex traits by engineering F2 gradient populations and enforcing a co-segregation standard that simultaneously verifies QTL effects and yields NIL-like lines for fine mapping. This three-in-one framework reduces time and labor compared with traditional linkage and multi-parental designs and overcomes GWAS limitations (population structure, LD, rare alleles) by enabling detection of both major and minor-effect genes. Application to rice grain size cloned eight genes in three years, capturing more than two-thirds of natural phenotypic variation and achieving high predictive accuracy. Population-scale analyses revealed directional selection for long, slender grains—particularly in indica—reflected by increased favorable allele frequencies, reduced nucleotide diversity, and clear selective sweeps at major loci (GS3, GW5, GW7/GL7). The observed strong correlations between effect size, association signal, selection strength, and sequence variation link phenotypic impact with domestication pressures, providing mechanistic insight into how human preference and cultivation shaped grain morphology. The results demonstrate RapMap’s utility for comprehensive genetic dissection and its potential to inform genomic breeding strategies by optimizing allele and haplotype combinations for desired grain shapes and adaptation.

The study introduces RapMap, a rapid, scalable QTL mapping and cloning strategy that integrates population design (F2GPs), a co-segregation quality-control standard, and NIL-based fine mapping to accelerate discovery of causal genes. In rice, RapMap identified eight grain-size genes—including two novel genes (GL1, GW5.1) and two new GS3 alleles—that collectively explain most variation in grain size and shape within a diverse panel, and uncovered strong, directional selection favoring slender, long grains, particularly in indica. RapMap’s advantages—speed, resolution, power to detect both common and rare alleles, and direct gene validation—make it broadly applicable to other crops for functional genomics and genomic breeding. Future work should: (1) run additional RapMap rounds with new gradient crosses to capture remaining heritability; (2) incorporate epigenetic and structural variants, gene–gene interactions, and environment into models; (3) extend to other traits and species; and (4) leverage the identified alleles with gene editing and haplotype design for rapid neo-domestication and precise ideotype breeding.

• Approximately 30% of grain size/shape heritability remains unexplained at the population level, potentially due to limited sample size, untested parents/alleles, epigenetic and structural variants not captured by SNP/InDel genotyping, epistasis, environmental effects, and phenotyping noise. - Some QTLs are difficult to narrow to very small intervals due to recombination frequency and marker availability, occasionally necessitating candidate-gene approaches. - Not all F2GPs meet the co-segregation standard initially; replacement with F3/F4 families may be required. - Functional validation faced practical constraints (e.g., transformation in indica backgrounds), leading to some validations in japonica. - Population-level effect estimates depend on allele frequencies; rare alleles (e.g., GS2) can have small PVE despite large bi-parental effects.