Ghd7‑triggered transcriptional regulation for synergistically improving rice eating quality and yield

Abstract

Improving both eating quality and yield remains a major challenge in rice breeding. Here, we identify Ghd7 as the causal gene underlying the grain amylose content QTL qAC7 and reveal Ghd7–triggered transcriptional regulation that balances starch and protein to coordinate yield and quality. Ghd7 transcriptionally activates PCF3 and exhibits dominant epistasis over it. A natural TA/T variant in PCF3 determines its DNA-binding affinity: the PCF3‑T allele strongly represses OsNF-YC4, increasing starch and reducing protein in grain, whereas PCF3‑TA fails to exert this repressive effect. Haplotype and evolutionary analyses show that the favorable allele PCF3‑T has not been selected. Combining Ghd7‑1/4 with PCF3‑T further enhances both yield and eating quality. Our findings uncover a possible transcriptional cascade and highlight Ghd7‑1/4 and PCF3‑T as promising targets for simultaneously improving rice yield and quality.

Rice breeders struggle to boost yield and eating quality simultaneously as the two traits often work against each other. Here, the authors identify a CCT domain transcription regulator encoding gene Ghd7 in determining rice grain amylose content and functioning in balancing starch and protein to coordinate yield and quality.

Introduction

Unlike the significant advancements in increasing rice yield over the past half century through semi-dwarfing, widespread nitrogen fertilizer use, and heterosis, research and breeding efforts focused on improving rice quality have not kept pace^1,2. This disparity can be attributed to historical neglect of rice quality and, more significantly, the observed negative correlation between quality and yield traits^3,4. The current mainstream strategy for balancing rice yield and quality is to combine natural elite alleles or to perform gene editing as conducted by Liu et al.⁵ and Wu et al.⁶. Nevertheless, there may be natural genetic resources that can be exploited to enhance rice yield and quality simultaneously. Previous studies have extensively documented genes or quantitative trait loci (QTL) associated with enhancing rice yield^7,8, as well as those for improving rice quality, albeit to a lesser extent^9,10. Nevertheless, the identification of genes or QTL that can simultaneously enhance both rice yield and quality remains an area yet to be explored.

Rice quality usually includes appearance quality, nutritional quality, cooking and eating quality (CEQ), milling quality, and recently emerging hygienic quality¹¹. Among these, nutritional quality and CEQ are the primary concerns for consumers. The major components of the endosperm are starch (70–80%), protein (7–10%), and lipids (about 1%)¹². Therefore, the main factor determining rice yield and quality for human consumption lies in the composition and quantity of starch and protein. An increasing number of recent studies have indicated that the balance of carbon and nitrogen could be crucial for coordinated regulation of yield and quality^13–16. Since starch and protein are the primary carbon-containing and nitrogen-containing compounds in endosperm, respectively, they are likely genetically coordinated. A main feature of the balance is the trade-off observed between endosperm starch content and protein content^3,17–20, although it is not always absolutely the case²¹. Therefore, by discovering novel regulators that simultaneously regulate endosperm starch and protein content, it may be possible to further elucidate the key factors that coordinately regulate quality and yield. In recent decades, significant progress has been achieved in understanding starch synthesis pathways and the genetic variations^10,22–26. Additionally, some advances have been made in studying the genetic factors influencing cereal protein content^9,10,27. However, these studies have mainly focused on a few QTL with substantial genetic effects or genes identified in mutants. The availability of genetic resources associated with starch or protein content in natural populations remains quite limited.

Ghd7, the first pleiotropic gene identified to control yield in rice, was originally map-based cloned using a segregating population derived from the cross between Zhenshan 97 (ZS97) and Minghui 63 (MH63)²⁸. Moreover, Ghd7 was also found to regulate grain shape²⁹, nitrogen use efficiency³⁰, chlorophyll content³¹, vascular bundle variation³², grain zinc content³³, and stress response³⁴. Previous studies on Ghd7 have mostly focused on its roles in regulating rice heading date and yield formation, while our previous work has demonstrated that Ghd7 maintains the coordinated balance between high grain quality and high yield in rice and that Ghd7's effect on grain quality does not simply depend on the extension of heading date³⁵. Despite these findings, systematic exploration of genetic resources directly regulated by Ghd7 that synergistically improve both rice quality and yield remains insufficient, and further in-depth investigations in this area are urgently needed. OsNF-YC4, an NF-Y family transcription factor, positively modulates protein accumulation and negatively modulates starch synthesis in rice endosperm. Nevertheless, the regulatory relationship between OsNF-YC4 and Ghd7 remains unclear.

In this study, we find that PCF3 (encoding a plant-specific TCP family transcription factor) is a downstream target of Ghd7 and uncover transcriptional regulation by Ghd7, which modulates starch and protein contents in rice grains, thereby influencing rice eating quality and yield. Specifically, in the context of rice quality regulation, Ghd7 exerts an epistatic effect on PCF3. Haplotype analysis reveals that PCF3-T is a favorable allele of the PCF3 gene; overexpression of this allele coordinately enhances both grain yield and eating quality in rice. Furthermore, the combination of favorable alleles of Ghd7 and PCF3 provides an effective strategy for optimizing rice yield while maintaining high eating quality.

Results

Map-based cloning and confirmation of qAC7 for grain amylose content

A doubled-haploid population from a cross between japonica rice varieties Kongyu131 (KY131) and Daohuaxiang 2 (DHX2) was analyzed to identify genes underlying natural variation in grain amylose content. A QTL for grain amylose content, qAC7, was identified in the interval between molecular markers 6144 and 1949 on chromosome 7 (Supplementary Fig. 1). The QTL was repeatedly detected in different environments (Supplementary Table 1) with the allele from DHX2 increasing grain amylose content (Supplementary Table 1).

To finely map the qAC7 locus, we backcrossed a double haploid line containing the chromosome segment 6144-1949 from DHX2 to KY131 by four times. Two near-isogenic lines (NILs) in the KY131 background, NIL(KY131) (genetic background recovery rate: 100%) and NIL(DHX2) (genetic background recovery rate: 97.29%), were selected following self-pollination of heterozygous BC₄F₁ plants. From 2400 BC₄F₂ individuals, we identified 430 recombinants in the 6144 - 1949 interval (Fig. 1a). The progeny of the recombinants was genotyped for the qAC7 locus, which was narrowed within an approximate 245.8Kb interval between markers 9042 and 1150 (Fig. 1a–c). In this process, we found that one of the co-isolated single-nucleotide polymorphism (SNP) markers, TYH-1, was located in the second exon of Ghd7, and was one of the functional variation sites of Ghd7 (labeled as S_1629³⁶) (Fig. 1b, d). The full-length Ghd7 gene, including the 2000 bp promoter and 500 bp 3’ untranslated region, was sequenced and compared between the parents, KY131 and DHX2 which confirmed that there was a single SNP base substitution, namely TYH-1 or S_1629, in the gene (Fig. 1d and Supplementary Fig. 2). Based on the genotyping results²⁸, the Ghd7 derived from KY131 belongs to the Ghd7−0a type, while the Ghd7 derived from DHX2 belongs to the Ghd7−2 type. The Ghd7−0a allele had a stop codon created by the SNP at the S_1629 site, resulting in premature termination of translation^9,28. Consequently, Ghd7 derived from KY131 was considered non-functional. A single BC₄F₂ plant heterozygous only for the 9042-1150 interval was used to develop a small segregating population of 96 individuals. Phenotyping revealed a linkage between plant height and grain amylose content in this population. Specifically, taller plants exhibited higher grain amylose content, whereas shorter plants had lower grain amylose content (Fig. 1e). The frequency distribution histogram showed that the ratio of plants with low plant height to high plant height was approximately 1:3 (n = 96, χ² = 0.108 < χ² _{0.05, 1} = 3.84, P > 0.05), indicating the presence of a single dominant allele for plant height from DHX2 (Supplementary Fig. 3). Taken together, we considered Ghd7 to be a high-confidence candidate gene for qAC7.

Fig. 1. Map-based cloning of qAC7.

a,b Fine mapping of the qAC7 region using a mapping population of 2400 plants. No. of Recs. indicates the number of recombinants between molecular markers flanking qAC7. c Genotypes and amylose contents phenotypes of the recombinants established by progeny testing. K and D represent the genotypes of KY131 and DHX2, respectively. d The differences in the Ghd7 gene between parents, KY131and DHX2; red arrow indicates the functional variant SNP. e Mean plant heights and grain amylose contents of homozygous and heterozygous lines in a BC₄F₃ population. f Grain amylose content of Ghd7 complemented plants (CO- and CO+ represent negative and positive plants, respectively), Ghd7 knockout plants (Ghd7-KO), and Ghd7 overexpression plants (Ghd7-OX). Data are means ± s.e.m (n = 16 biological replicates). Error bars, s.e.m. P values were generated by two-tailed t tests. Source data are provided as a Source Data file.

For validation, we generated complementation (CO) transgenic lines in the japonica variety KY131 background by introducing the Ghd7 allele from DHX2 (Supplementary Fig. 4a, c). Compared with transgene-negative plants, T₁ transgene-positive plants (CO + ) showed a significant increase in grain amylose content, plant height and heading date (Fig. 1f and Supplementary Fig. 4c, 5). A CRISPR-Cas9 construct targeting the second exon of Ghd7 was developed and transform into Nipponbare (NIP, a japonica variety that carries the same Ghd7 allele as DHX2²⁸) (Supplementary Fig. 4b, c). After several generations of genotypic and phenotypic screening, we obtained a homozygous Ghd7 knockout (KO) mutant line. As expected, grains produced from homozygous Ghd7-KO plants showed a lower amylose content as well as lower plant height and earlier heading period (Fig. 1f and Supplementary Fig. 4c, 5). We also obtained a line overexpressing Ghd7 in Zhonghua11³⁴ (ZH11, a japonica variety carrying the same Ghd7 allele as DHX2²⁸). Reverse transcription quantitative PCR (RT-qPCR) analysis confirmed that the transcript level of Ghd7 was significantly upregulated in the overexpression lines compared with wild-type ZH11 plants (Supplementary Fig. 6). Compared ZH11 as the control, the Ghd7-OX plants had higher grain amylose content, higher plant height and later heading date (Fig. 1f and Supplementary Fig. 5). These results suggested that Ghd7 was the causal gene underlying the qAC7 QTL.

Ghd7 synergistically affects rice quality and yield

To assess the effects of Ghd7 on grain quality, we first analyzed its expression in various tissues. Eleven tissues from the NILs were assayed for temporal and spatial expression patterns of Ghd7 by quantitative PCR (qPCR) (Supplementary Fig. 7). Ghd7 expression levels were much higher in NIL(DHX2) endosperm at 10 and 15 days after fertilization (DAF) than in other tissues. This expression pattern corresponded well to the grain filling stage critical for grain quality formation. In subcellular localization analysis, the protein encoded by the allele from DHX2 colocalized with the nucleus-specific marker (WHITE-CORE RATE 1 protein)¹⁹, demonstrating nuclear localization of Ghd7 protein (Supplementary Fig. 8), consistent with previously reported results²⁸. Next, we measured the quality traits of NIL(KY131) and NIL(DHX2) using data collected from a field trial (Table 1). Compared with NIL(KY131), NIL(DHX2) had 1.1% higher grain amylose content and 4.9% higher grain starch content, leading to an increase of 4 mm in gel consistency and a 4.1 increase in taste value, a decrease of 6 mg/g in glutelin content and a decrease of 13.1 mg/g in total protein content. Similar results were observed in complementation (CO) materials. We also observed a change in the same quality traits in the knockout (KO) and overexpression lines (Table 1). These results strongly suggest that the known pleiotropic gene Ghd7 also significantly affects grain quality traits, and that enhancing its expression can improve the cooking and eating quality (CEQ) of rice. Analysis of key agronomic traits of these genetic materials (Supplementary Fig. 5) confirmed that Ghd7 can greatly increase yield in rice by increasing the number of grains per panicle, consistent with a previous report²⁸. Therefore, Ghd7 is an excellent gene that can be targeted to simultaneously improve CEQ and increase yield.

Table 1.

Grain quality and yield traits in NILs and transgenic plants

Line	n	AC (%)	Starch content (%)	GC (mm)	Glu (mg/g)	Pro (mg/g)	Glo (mg/g)	Alb (mg/g)	Protein content (mg/g)	Taste value
NIL(KY131)	20	16.0 ± 0.2	82.5 ± 1.9	126.0 ± 0.8	45.0 ± 3.2	2.1 ± 0.2	3.4 ± 0.3	3.4 ± 0.3	75.9 ± 3.7	75.9 ± 1.5
NIL(DHX2)	20	17.1 ± 0.3	87.4 ± 1.8	130.0 ± 1.6	39.0 ± 3.0	2.4 ± 0.3	3.4 ± 0.2	3.2 ± 0.2	62.8 ± 3.0	80.0 ± 1.2
P value		7.6 × 10-4	2.8 × 10-2	3.6 × 10-2	2.1 × 10-5	2.7 × 10-1	8.9 × 10-1	4.7 × 10-2	5.4 × 10-5	1.7 × 10-8
CO-	20	17.8 ± 0.6	82.0 ± 1.1	97.4 ± 7.4	58.6 ± 5.3	5.5 ± 0.5	2.6 ± 0.4	2.9 ± 0.6	70.6 ± 2.3	77.0 ± 1.0
CO+	20	19.5 ± 0.4	86.5 ± 1.7	103.2 ± 8.9	39.7 ± 7.5	5.4 ± 0.4	2.1 ± 0.3	3.0 ± 0.5	61.7 ± 3.5	83.0 ± 1.3
P value		1.0 × 10-4	9.1 × 10-7	2.5 × 10-5	1.4 × 10-8	8.3 × 10-1	2.8 × 10-4	3.8 × 10-1	1.4 × 10-8	3.0 × 10-13
NIP	20	20.3 ± 0.4	87.1 ± 1.1	112.3 ± 4.5	48.2 ± 3.7	3.1 ± 0.4	5.5 ± 0.4	2.7 ± 0.3	98.6 ± 3.7	87.0 ± 0.1
Ghd7-KO	20	15.94 ± 0.6	84.0 ± 0.8	95.0 ± 0.8	61.4 ± 7.9	3.2 ± 0.2	4.6 ± 0.6	2.7 ± 0.2	115.9 ± 7.7	84.7 ± 0.5
P value		3.2 × 10-13	2.1 × 10-5	5.8 × 10-3	2.6 × 10-4	6.3 ×; 10-1	1.9 × 10-3	7.8 × 10-1	4.4 × 10-4	2.2 × 10-3
ZH11	20	19.7 ± 0.6	72.6 ± 1.5	84.5 ± 2.4	52.9 ± 7.8	2.7 ± 0.6	5.3 ± 0.2	4.3 ± 0.4	107.8 ± 3.0	82.6 ± 1.1
Ghd7-OX	20	22.4 ± 0.6	76.7 ± 2.1	92.4 ± 2.2	42.9 ± 6.5	2.1 ± 0.6	4.9 ± 0.6	2.9 ± 0.4	103.9 ± 2.7	85.9 ± 0.8
P value		7.9 × 10-5	3.8 × 10-4	9.3 × 10-7	8.1 × 10-5	4.7 × 10-1	6.3 × 10-1	2.2 × 10-2	5.3 × 10-7	1.8;×10-7

NIL, near-isogenic line; n, number of plants with the same genotype; each replicate corresponds to an individual rice plant The data are based on field trials using randomized block designs with three replications. All data are means ± s.e.m. “+” and “-” represent transgene-positive and negative plants, respectively. CO, complemented plants. KO, knockout plants. OX, overexpression plants. AC, amylose content; GC, gel consistency; Glu, grain glutelin content; Pro, grain prolamin content; Glo, grain globulin content; Alb, grain albumin content; Taste value, a comprehensive assessment of the viscosity, hardness, and appearance of cooked rice. P values were generated by two-tailed t tests. Source data are provided as a Source Data file.

Since 1990, KY131 has been a predominant rice variety in the planting areas of Northeast China, attributed to its exceptional traits including lodging resistance, high yield, early maturity, and cold resistance³⁷. According to the China Rice Data Center (https://www.ricedata.cn/), KY131 ranks among the most widely planted conventional japonica rice varieties in China. However, its grain quality remains relatively average, limiting its overall potential. Conversely, DHX2 is a well-known japonica rice variety with superior tasting quality, meeting the highest rice quality standards set by the Ministry of Agriculture and Rural Affairs of China. However, DHX2 is susceptible to lodging and a lower yield. Combining the strength of KY131 and DHX2 would be highly significant, enabling the expansion of KY131’s planting area and the increase of its commercial value. For this, the Ghd7 allele from DHX2, which can synergistically improve rice quality and yield, could be exploited. For this reason, we grew the NILs containing the Ghd7 allele from DHX2 in the Wuchang region of Northeast China and in Wuhan region of central China, and conducted a comprehensive study of quality and yield at the two locations. The grain amylose content, total starch content, gel consistency, and taste value of the NIL(DHX2) line in Wuchang increased by 2.2%, 1%, 8.7 cm, and 3.3, respectively, compared with NIL(KY131), and the grain protein content, grain chalkiness rate and grain chalkiness area were decreased by 8.4 mg/g, 89.2% and 95.5%, respectively (Supplementary Fig. 9). The NIL(DHX2) line exhibited a 13 cm increase in plant height, delayed heading date by 11.5 days, increased number of grains per panicle of 33.7, a 2 cm extension in panicle length, and a 17 g increase in yield per plant (Supplementary Fig. 10, 11). There were no significant changes in number of panicles, effective panicles per plant, seed setting rate, and 1000-grain weight per plant. In Wuhan, the grain amylose content, total starch content, gel consistency and taste value of NIL(DHX2) plants increased by 0.6%, 4.9%, 5 cm, and 5.3, respectively, compared with NIL(KY131), and grain protein content, grain chalkiness rate and grain chalkiness area were decreased by 9.1 mg/g, 49.8% and 87.9% respectively (Supplementary Fig. 12). NIL (DHX2) exhibited a 20.3 cm gain in plant height, a 20.7-day later heading date, a 58.8 more grains per panicle, a 3.8 cm longer panicle, and a 13.4 g higher yield per plant compared to NIL(KY131) (Supplementary Fig. 10, 13). The number of panicles per plant, effective panicles per plant, seed setting rate, and 1000-grain weight remained unchanged. These phenotypic discrepancies between the two locations are most likely attributed to distinct photoperiod and temperature conditions, which may modulate the regulatory effects of the Ghd7-mediated transcriptional regulation, as well as genotype-by-environment interactions. These results show that the Ghd7 allele from DHX2 synergistically improved the eating quality and yield of KY131 across different environments.

Transcriptional regulation and epistatic interaction between Ghd7 and PCF3

A potential downstream target gene of Ghd7, PCF3, identified by ChIP-seq, was also among the candidate genes identified by genome-wide association study of grain glutelin content (Supplementary Fig. 14a, 15 and Supplementary Table 2). Yeast one-hybrid (Y1H) mutagenesis assays confirmed that the CCACC motif is absolutely required for physical association between Ghd7 and the PCF3 genomic fragment (Supplementary Fig. 14b). Consistent with this result, EMSA and ChIP-qPCR separately validated the direct binding of Ghd7 to the PCF3 locus at the in vitro and in vivo levels (Fig. 2a, b). The luciferase reporter assay in rice protoplasts performed using a dual-luciferase reporter assay system (Fig. 2c, d) showed that Ghd7 protein could enhance the transcriptional activity of PCF3. RT-qPCR used to examine the relative expression levels of PCF3 in two sets of NILs and Ghd7-associated transgenic materials (Fig. 2e-h), showed that PCF3 gene was significantly less expressed in lines lacking Ghd7 (NIL(KY131), NIL(ZS97) and CO-) compared to those carrying the Ghd7 gene (NIL(DHX2), NIL(NYZ) and CO + ). Similarly, compared to the control line ZH11, PCF3 expression was increased in Ghd7 overexpression lines (OX-Ghd7) and decreased in Ghd7-silenced lines (Ami-Ghd7) (Fig. 2h). These results indicate that Ghd7 positively regulates the expression of the PCF3 gene.

Fig. 2. Ghd7 promotes PCF3 gene expression.

a EMSA assay. Non-labeled probes were used as competitors. DNA-protein complexes and free probes are indicated by arrows, respectively. Unlabeled competitive DNA was present at 50-fold and 100-fold molar excess over the labeled probe. The experiments were replicated three times with similar results. b Ghd7 mediated ChIP–qPCR enrichment (relative to input) of CCACC³⁵ contained fragments (asterisk) from PCF3. Data are means ± s.e.m (n = 3 biological replicates). Error bars, s.e.m. P values were generated by a two-tailed t test. c,d Transient expression assays of the transcriptional activity of Ghd7 on the promoter of PCF3 gene. The 713 bp PCF3 exon fragment was inserted in the reporter vector containing the mini35S promoter and the LUC gene. The effectors consisted of the negative vector None and the positive vector None-Ghd7 containing the full-length Ghd7. For relative LUC activity (d) the regulatory effect of None-Ghd7 on 190luc-ARE1 was used as a positive control³⁰. Data are means ± s.e.m (n = 10 biological replicates). Error bars, s.e.m. Significant differences were determined by Duncan’s tests. e–h Relative expression of the PCF3 gene in developing endosperm of Ghd7-related genetic materials. Total RNA was extracted from endosperm at 10 days post-flowering. Data are means ± s.e.m (n = 3 biological replicates). Error bars, s.e.m. P values were generated by two-tailed t tests. Source data are provided as a Source Data file.

To further elucidate the genetic relationship between Ghd7 and PCF3, we generated both knockout and overexpression plants of PCF3 in the near-isogenic lines NIL (KY131) and NIL (DHX2), respectively (Fig. 3a and Supplementary Fig. 16a, b). Grain morphological analysis showed that in the Ghd7-loss-of-function NIL(KY131) background, knockout of PCF3 significantly reduced grain thickness and 1000-grain weight, whereas PCF3 overexpression caused no obvious phenotypic changes. In the NIL(DHX2) background with functional Ghd7, PCF3 knockout had no significant effect on grain thickness and 1000-grain weight, while PCF3 overexpression markedly decreased these two traits. For grain length and grain width, neither knockout nor overexpression of PCF3 induced any significant alterations (Supplementary Fig. 17, 18h). Quality phenotypic analysis revealed that in the NIL(KY131) genetic background (where Ghd7 function is lost), knockout of the PCF3 gene did not induce any alterations in grain quality-related phenotypes (Fig. 3b, c). In contrast, in the NIL(DHX2) genetic background (with normal Ghd7 function), PCF3 knockout led to an increase in chalkiness rate and protein content, accompanied by a decrease in gel consistency and taste value (Fig. 3b, c). Regardless of the genetic background, overexpression of PCF3 consistently resulted in a simultaneous reduction in amylose content and protein content, along with a corresponding increase in taste value (Fig. 3c). PCF3 knockout mutant lines in both NIL(KY131) and NIL(DHX2) genetic backgrounds had reduced phenotypic values for all assessed agronomic traits except heading date and 1000-grain weight (Supplementary Fig. 18). Analysis of grain quality and agronomic traits revealed that some phenotypic values of NIL(KY131)-PCF3-OX plants reached to the levels comparable to those of NIL(DHX2), including total starch content in grains, total protein content in grains, taste value, number of panicles per plant, number of effective panicles per plant, seed setting rate and grain size (Fig. 3c and Supplementary Fig. 17, 18), indicating that the overexpression of PCF3 can partially restore the phenotypic defects caused by the loss of Ghd7 function. Further two-factor interaction assays revealed that the regulatory effects of PCF3 on yield and eating quality rely heavily on the functional status of Ghd7 (Supplementary Fig. 19). The distinct non-parallel phenotypic trends between Ghd7 and ghd7 backgrounds verify a significant non-additive epistatic interaction between the two genes.

Fig. 3. Ghd7 is a positive regulator of PCF3 and exhibits dominant epistasis over it.

a Plant morphology of near-isogenic lines containing Ghd7 and PCF3 mutant plants. Scale bar, 10 cm. b Grain appearance of the near-isogenic lines and the PCF3 mutants. Note: PCF3 knockout affects chalkiness only in NIL(DHX2) (Ghd7 functional), confirming dominant epistasis. (c) Grain quality traits of the near-isogenic lines and the PCF3 mutants. PCF3 mutation impacts traits only when Ghd7 is functional, consistent with dominant epistasis. Data are means ± s.e.m (n = 8 biological replicates). Error bars, s.e.m. Significant differences were determined by Duncan’s tests. Source data are provided as a Source Data file.

PCF3 negatively regulates the expression of OsNF-YC4

PCF3 encodes a plant-specific transcription factor belonging to the TEOSINTE BRANCHED 1, CYCLOIDEA AND PROLIFERATING CELL FACTOR (TCP) family^38,39 (Supplementary Fig. 20a). PCF3 was predominantly localized to the nucleus and highly expressed in the stem and panicle (Supplementary Fig. 20b, c). To identify its downstream targets, we first employed DNA Affinity Purification Sequencing (DAP-seq) technology to screen for potential targets of PCF3 protein at the genome-wide level. Sequencing results revealed that the top 10 potential binding motifs of PCF3 all contained the 5-base sequence CCCAC (Supplementary Data 1), suggesting that this sequence is probably the core binding motif of PCF3. Further investigation using the ChIP-Hub regulomics comprehensive analysis platform⁴⁰ showed that OsNF-YC4, a gene recently reported to positively regulate protein content while negatively regulating starch content in rice grains⁴¹, is highly likely a direct downstream target gene of PCF3 (Supplementary Data 2). To verify this hypothesis, we first analyzed the DNA sequence of OsNF-YC4, and the results indicated that its gene region contains a total of 9 CCCAC binding motifs, including 7 in the promoter region and 2 in the 3’UTR region (Fig. 4a). Subsequent EMSA demonstrated that the PCF3 protein could specifically bind to the CCCAC motifs at all sites except for those at the P4, P5, and P8 positions (Fig. 4b). To explore the regulatory effect of PCF3 on OsNF-YC4, a dual-luciferase reporter assay was performed in the rice protoplast system. The results showed that PCF3 significantly inhibited the transcriptional activity of OsNF-YC4 (Fig. 4c, d). DAP-qPCR results demonstrated that the Halotag-PCF3 fusion protein interacts in planta with six DNA fragments (designated as P1, P2, P3, P6, P7, and P9) containing the CCCAC motif (Fig. 4e). Consistently, when DAP-seq data were visualized using the IGV software, two distinct PCF3 binding enrichment peaks were observed in the promoter region of OsNF-YC4, and these peaks were located at positions corresponding to the aforementioned five CCCAC motifs in the promoter region (Fig. 4f). Furthermore, detection of OsNF-YC4 expression levels in the endosperm and seedlings of PCF3 knockout plants further confirmed that PCF3 exerts a negative regulatory effect on OsNF-YC4 expression (Fig. 4g). To clarify the genetic relationship between PCF3 and OsNF-YC4, we generated the single mutants of PCF3 (P3-KO) and OsNF-YC4 (Y4-KO), as well as their double mutant (P3Y4-KO) (Supplementary Fig. 21). Phenotypic analysis showed that the double mutants (P3Y4-KO1 and P3Y4-KO2) exhibited a phenotype highly comparable to that of the OsNF-YC4 single mutants (Y4-KO1 and Y4-KO2) (Fig. 4h, i and Supplementary Fig. 21). Collectively, these experimental results confirm that OsNF-YC4 is a direct downstream target gene of PCF3.

Fig. 4. PCF3 is a negative regulator of OsNF-YC4.

a The gene structure diagram of OsNF-YC4. Black-filled rectangles represent exons, while white-filled rectangles represent the promoter or untranslated regions (UTRs). Small black inverted triangles indicate the positions of probes used in the electrophoretic mobility shift assay (EMSA). b EMSA assays. Non-labeled probes were used as competitors. DNA-protein complexes and free probes are indicated by arrows, respectively. Unlabeled competitive DNA was present at a 50-fold or 200-fold molar excess over the labeled probe. The experiments were replicated three times with similar results. c,d Transient expression assays of the transcriptional activity of Ghd7 on the promoter of PCF3 gene. The reporter vector containing the mini35S promoter and the LUC gene was inserted in the 2 kb PCF3 promoter fragment (c). The effectors consisted of the negative vector None and the positive vector None-PCF3 containing the full-length PCF3. For relative LUC activity (d) the regulatory effect of None-Ghd7 on 190luc-ARE1 was used as a positive control. Data are means ± s.e.m (n = 10 biological replicates). Error bars, s.e.m. Significant differences were determined by Duncan’s tests. e Halotag–PCF3 mediated DAP–qPCR enrichment (relative to input) of CCCAC-containing DNA fragments from OsNF-YC4. Different letters denote significant differences (P < 0.05) from a Duncan’s multiple range test. f Visualization of DAP-seq results using IGV software. The black dashed lines indicate the positions of the motifs bound by the PCF3 protein, corresponding sequentially to P1, P2, P3, P0, P4, P5, P6, P7, P8 and P9 in Figure a from left to right. g Relative expression of the OsNF-YC4 gene in developing endosperm and seedlings of KY131 and P3-KO. Total RNA was extracted from endosperm at 10 days post-flowering and from two-week-old seedlings. OsActin (LOC_Os03g50885) was used as the control. Data are means ± s.e.m (n = 3 biological replicates). Error bars, s.e.m. P values were generated by two-tailed t tests. h Appearance of grains from the single mutants of PCF3 and OsNF-YC4, and their double mutant. Scale bar, 5 mm. i Main grain quality traits of PCF3 and OsNF-YC4 single and double mutants, including chalkiness rate, amylose content, total starch content, protein content, gel consistency, and taste value. Data are means ± s.e.m (n = 6 biological replicates). Error bars, s.e.m. Significant differences were determined by Duncan’s tests. Source data are provided as a Source Data file.

Haplotype analysis of the PCF3 gene

Based on genomic variation data and functional annotations from the RiceVarMap v2.0 database, we identified a potential functional variant of PCF3, vg1103758534, which is located in the third exon of the gene. This variant is predicted to cause a frameshift mutation, thereby impairing the PCF3 protein function (Fig. 5a). Based on this variation, the 533 accessions within the Mini Core Collection of rice germplasm resources⁴² could be classified into two haplotypes: haplotype T (also referred to as PCF3^T) and haplotype TA (also referred to as PCF3^TA) (Fig. 5b and Supplementary Data 3). Among the 600 accessions (533 core germplasm accessions plus 446 previously sequenced wild rice accessions⁴³, excluding those genotypes with PCF3 missing), approximately 43% of accessions carried the haplotype T, while haplotype TA had a higher frequency in the Japonica (Jap) subspecies (Supplementary Table 3 and Supplementary Data 3). By leveraging the whole-genome variation data of 533 rice accessions from the RiceVarMapv2.0 database and our phenotypic data, we found that rice varieties with the PCF3^T allele showed distinct advantages over those with the PCF3^TA allele regarding yield, taste value, plant height, tiller number per plant, effective tiller number per plant, spikelet number per plant and grain amylose content traits (Fig. 5c, d and Supplementary Fig. 22a–i). Analysis of nitrogen utilization data from 55 mainstream cultivated varieties (including 24 officially recognized green super rice varieties and 31 candidate green super rice varieties)⁴⁴, combined with genotyping of their PCF3 allelic variants via Sanger sequencing (Supplementary Data 4), revealed that varieties carrying the PCF3^T allele exhibited significantly higher nitrogen partial factor productivity from applied N (PFPN) compared to those carrying the PCF3^TA allele (Supplementary Fig. 22j). Therefore, PCF3^T represents a superior allele.

Fig. 5. The natural variations of PCF3.

a Visualization of PCF3 variation. Potential functional variant locus vg1103758534 is indicated by the orange arrow. b Schematic of PCF3 gene structure and the location of the potential functional variant. c,d Comparison of yield and taste value between two PCF3 haplotypes in 533 rice accessions. P values were generated by a two-tailed t test. “n” represents the number of samples for the two distinct haplotypes. For relative LUC activity e the regulatory effect of None-Ghd7 on 190luc-ARE1 was used as a positive control. Data are means ± s.e.m (n = 10 biological replicates). Error bars, s.e.m. Significant differences were determined by Duncan’s tests. f EMSA assays. DNA-protein complexes and free probes are indicated by arrows, respectively. The experiments were replicated three times with similar results. Source data are provided as a Source Data file.

To further dissect the molecular mechanism underlying the superior performance of the PCF3^T allele over the PCF3^T allele, we compared the transcriptional activation activity of proteins encoded by these two alleles on the OsNF-YC4 gene using a luciferase reporter assay in a rice protoplast system (Fig. 5e and Supplementary Fig. 23). The results showed that the T-type protein significantly repressed the transcriptional activity of OsNF-YC4, whereas the TA-type protein exerted no such repressive effect. This finding was further supported by an EMSA assay: the T-type transcription factor could bind normally to the five CCCAC binding motifs (P1, P2, P3, P6, and P9) in the promoter region of OsNF-YC4, whereas the TA-type transcription factor could only bind to motif P9 (Fig. 5f). Collectively, these observations strongly suggest that this functional variant in PCF3 is likely the fundamental cause of the differences in yield and quality traits among rice plants.

Haplotype distribution, evolutionary characteristics, and domestication selection analysis of the PCF3 gene

The geographic distribution map of haplotypes illustrates that the T haplotype of PCF3, marked by red dots, is distributed across multiple continents, including the Americas, Europe, Africa, and Asia. In contrast, the TA haplotype, represented by dark green dots, is relatively concentrated in Asia, with only a few scattered occurrences in regions such as Africa and Europe (Fig. 6a). This distinct spatial distribution pattern suggests that there may be differences between the two haplotypes in terms of geographic diffusion and evolutionary trajectory. Phylogenetic analysis showed that PCF3 is quite diverse across the range of wild rice, Ind, and Jap accessions (Fig. 6b). Using the nucleotide polymorphism of the PCF3 gene in 533 core germplasm accessions and 106 wild rice accessions, we classified these materials into five major haplotypes (Fig. 6b, c). Among them, Hap1-3 and Hap 4-5 belong to the types TA and T, respectively (Fig. 6b, c). Our results demonstrated that the PCF3 gene has undergone differentiation in the common wild rice population, with 63% of common wild rice accessions carrying the TA genotype and 37% carrying the T genotype (Fig. 6b–d, Supplementary Table 3 and Supplementary Data 3). Phylogenetic inference based on the haplotype network revealed that Hap5, located at the central hub of the network, was postulated as the ancestral haplotype. This inference is supported by its pivotal position connecting all other derived haplotypes via the fewest mutational steps. Two major derived haplotypes, Hap2 and Hap4 (each with a frequency of 188), diverged directly from Hap5. Hap2 is predominantly found in the Jap subspecies, while Hap4 is primarily associated with Ind, suggesting potential sub-species-specific adaptive evolution. Another derived haplotype, Hap3 (frequency: 43), also originated directly from Hap5. Furthermore, Hap1 (frequency: 90) was identified as a terminal derivative, arising from a subsequent mutation on the Hap2 background. The network topology, combined with haplotype frequency distribution, clearly delineates an evolutionary pathway: the ancestral haplotype Hap5 gave rise to Hap2, Hap3, and Hap4 via independent single mutations, whereas Hap1 originated from a subsequent mutation on the derived Hap2 background (Fig. 6c and Supplementary Fig. 24). Furthermore, three-dimensional principal component analysis revealed that samples of the T group exhibited substantial spatial overlap and closer distribution proximity to the wild group, whereas samples of the TA group were relatively distanced from the wild group (Fig. 6e). These observations suggest a closer phylogenetic relationship between the T lineage and the wild group.

Fig. 6. Haplotype diversity, geographic distribution, population differentiation, and selection signal analysis of the PCF3 gene.

a Geographic distributions among 533 accessions. Colored circles represent different PCF3 alleles. The map layers used in the figures are derived from the Natural Earth dataset (https://www.naturalearthdata.com/). b Phylogenetic tree of PCF3. c Haplotype network of PCF3. Circle size is proportional to the number of samples for a given haplotype. Red spots represent unobserved, but inferred haplotypes. Lines between haplotypes represent mutational steps between alleles. d Frequency of PCF3 alleles identified in common wild rice. e Three - dimensional principal component analysis (PCA) score plot of samples from different groups (T, TA, and Wild). The percentages in parentheses next to PC1, PC2, and PC3 represent the proportions of variance explained by each principal component, which are 6.66%, 3.36%, and 2.39%, respectively. Different colors denote different sample groups: red for T, orange for TA, and green for Wild. This plot illustrates the distribution and separation of samples across the three principal components, providing insights into the differences in the overall characteristics among the groups. f Selective sweep analysis of the PCF3 genomic region. The upper panel shows the density distribution of log₂ θ_π ratio values, with blue, gray, and red colors representing log₂ θ_π ratio < −0.888, −0.888 <log₂ θ_π ratio <0.878, and log₂ θ_π ratio > 0.878, respectively, along with the cumulative percentage curve. The lower panel is a scatter plot where the x-axis denotes log₂ θ_π ratio and the y-axis represents Fst. Blue dots indicate selected regions identified in the 533Admix population, red dots represent selected regions in the Wild population, and gray dots are non-selected regions. The dashed lines mark the thresholds for significant selection (log₂ θ_π ratio < −0.888 or > 0.878, and Fst ≥ 0.202). Source data are provided as a Source Data file.

To investigate whether artificial selection has contributed to the domestication of PCF3, genomic data from 533 core accessions of cultivated rice and 106 accessions of O. rufipogon were analyzed to assess the nucleotide diversity of the PCF3 locus and its flanking regions. In the Aus and Ind subgroups, the frequencies of the T allele all were 0.54, showing only a slight increase compared to that in the wild rice population (0.37) (Supplementary Table 3 and Supplementary Data 3). The frequency of this allele in the Jap subgroup was significantly lower than that in the wild rice population (Fig. 5d). Such a distribution pattern of allele frequencies across different subgroups is a typical manifestation of subgroup differentiation under neutral evolution. To further investigate whether the PCF3 gene has undergone domestication selection, selective sweep analysis was performed, and the results revealed that PCF3 is localized in the gray region and does not lie within genomic segments subjected to strong positive selection (Fig. 6f). Selective sweep analysis demonstrated that in the PCF3 gene region, the θπ_Wild/θπ_group ratios showed no significant reduction relative to genome-wide background levels—evidenced by the absence of conspicuous troughs in θπ_Wild/θπ_group distribution across all cultivated subgroups (Admix: ~5; Aus: ~40; Ind: ~12; Jap: ~15; VI: ~50), where a deep trough would typically indicate diversity loss from artificial selection. Concurrently, population differentiation analysis revealed no elevated F_st peaks at this locus compared to genomic averages, as a prominent peak would have suggested directional selection-driven divergence between subgroups. This multi-evidence confirms that PCF3 escaped strong artificial selection during rice domestication (Supplementary Figs. 25, 26). Furthermore, the results of the Tajima’s D value line plot indicated that there was no consistent negative trend in Tajima’s D values among different subgroups of cultivated rice (a negative Tajima’s D value is often associated with recent positive selection), and there was no significant differentiation compared with the wild rice population (Supplementary Fig. 27). This result is inconsistent with the core logic of domestication selection, which requires directional fixation of favorable alleles across different cultivated rice subgroups to form consistent genetic differentiation characteristics among subgroups. Based on the comprehensive analysis of the above multiple dimensions, it can be concluded that the PCF3 gene is highly likely not subjected to artificial selection during the domestication of cultivated rice. The significant phenotypic differences between its T allele and TA allele are essentially attributed to the functional genetic variation that already exists in wild rice being naturally retained in the cultivated rice population, and this variation has no correlation with artificial selection during the domestication stage.

The combination of elite alleles of Ghd7 and PCF3 is conducive to the coordinated improvement of rice yield and quality

Given that previous studies have clearly identified Ghd7-1 and Ghd7-4 as two elite haplotypes of the Ghd7 gene^28,35, a question worthy of in-depth investigation arises: whether the combination of elite haplotypes of Ghd7 with those of PCF3 will yield more positive effects?

The geographical distribution of the genotype combinations of Ghd7 and PCF3 reveals that the two most superior genotype combinations, T-Ghd7-4 (with Ghd7 genotype Ghd7-4 and PCF3 genotype T, denoted by red dots) and T-Ghd7-1 (with Ghd7 genotype Ghd7-1 and PCF3 genotype T, denoted by pink dots), are relatively concentrated in Southeast Asia and East Asia, respectively (Fig. 7a). In contrast, their distributions in other regions are relatively sparse and scattered. Subsequently, a comparative analysis was conducted on the agronomic traits and quality traits of these 8 genotype combinations (Fig. 7b–d and Supplementary Fig. 28), with the results presented as follows: for grain amylose content, the TA-Ghd7-2, T-Ghd7-1, and T-Ghd7-2 genotype combinations exhibited a wider distribution range, while the remaining genotype combinations showed relatively concentrated distributions (Fig. 7b). For the yield trait, all TA-containing genotype combinations shared a highly similar numerical distribution range with T-Ghd7-0. In contrast, the other three T-containing genotype combinations (T-Ghd7-4, T-Ghd7-1, and T-Ghd7-2) displayed significantly higher numerical distributions, with their values concentrated in a higher interval (Fig. 7c). For taste value, TA-Ghd7-1, TA-Ghd7-2, T-Ghd7-4, and T-Ghd7-1 performed significantly better than the other four genotype combinations, reaching a higher level (Fig. 7d). Furthermore, we observed that the two elite allele combinations, T-Ghd7-4 and T-Ghd7-1, not only had a greater number of panicles per plant and effective panicles per plant but also exhibited lower grain protein content (Supplementary Fig. 28). In conclusion, the T-Ghd7-4 and T-Ghd7-1 genotype combinations showed the optimal performance in terms of yield and eating quality.

Fig. 7. Synergistic positive effects from the combination of Ghd7 and PCF3 elite alleles.

a Geographical distribution map of the eight genotypic combinations of Ghd7 and PCF3. The map layers used in the figures are derived from the Natural Earth dataset (https://www.naturalearthdata.com/). b–d Grain amylose content, yield, and taste value from the eight different genotypic combinations. n indicates the number of rice accessions in each genotype group. e Plant architecture of the GC2-T line and parental plants. Scale bar, 10 cm. f The stubble with remaining culm structures in the field. Each red circle in the picture indicates a stem that has been cut. g Phenotypic display of all plump grains per plant in GC2-T line and parental plants under field-grown conditions. h Grain appearance of the GC2-T line and parental plants. i Agronomic traits and quality traits of the GC-T line and parental plants. Data are means ± s.e.m. Exact n values (biological replicates) are indicated in the figure. P values were determined by two-tailed Student’s t test. Source data are provided as a Source Data file.

As the Ghd7-1 allele exists extensively in cultivated rice, whereas the Ghd7-4 allele only accounts for a frequency of 13.5% in natural germplasm resources³⁵, we therefore chose Guichao 2 (GC2) (carrying the Ghd7-4 allele but without T (Supplementary Data 3)) as the target to test the performance of the elite genotype combination T-Ghd7-4. To this end, the superior allele T from TB154E-TB-2 was introduced into GC2 plants. Field phenotypic determination results showed that compared with the original GC2, the yield components and quality traits of GC2-derived line (GC2-T) (genetic background recovery rate: 97.2%) were significantly optimized (Fig. 7e–h). For yield-related traits: the number of panicles per plant of GC2-T increased from 12.33 in GC2 to 17.18 (an increase of 4.85), the number of effective panicles per plant increased from 9.25 to 11.82 (an increase of 2.57), and these improvements ultimately led to a significant 14.92% increase in yield per plant (Fig. 7e–g, i and Supplementary Fig. 29). For quality-related traits: the chalkiness rate of GC2-T was drastically reduced from 82.65% in GC2 to 58.69% (a reduction of 28.99%); the grain protein content decreased from 111.38 mg/g to 108.24 mg/g (a reduction of 2.82%); in contrast, the grain amylose content increased from 21.36% to 22.96% (an increase of 7.5%), the gel consistency extended from 24.05 mm to 52 mm (an increase of 27.95 mm), and the taste value also increased from 56.05 to 63.5 (an increase of 7.45) (Fig. 7h, i). Collectively, these findings demonstrate that the elite genotype combination of Ghd7 and PCF3 enables the synergistic improvement of both rice quality and yield.

Discussion

Both KY131 and DHX2 are excellent rice varieties. Although both carry the Wx^b allele, they exhibit significant differences in eating quality. Ghd7 is likely one of the primary factors causing this substantial difference. In another study³⁵, we found that Wx, the major gene for amylose synthesis, is a direct downstream target of Ghd7, and that Ghd7 promotes transcription of the Wx gene. Since Ghd7 acts as a negative regulator of grain protein content, varieties carrying Ghd7 (such as DHX2 and Koshihikari) have more starch and less protein in the grain, resulting in superior eating quality. An independent study⁴⁵ has shown that different photoperiods cause variation in metabolite accumulation in rice grain, with short-day conditions significantly increasing the contents of amino acids, vitamins, and flavonoids, thus enhancing the nutritional value and sensory quality. Varieties carrying Ghd7 have a longer growth period, allowing grain-filling under short-day conditions. Ghd7 can sense changes in light signals through its interaction with phytochrome A and B (OsPHYA and OsPHYB)⁴⁶, suggesting that Ghd7 likely regulates eating quality via photoperiod. In this study, we improved the eating quality and obtained higher yields in KY131 derivatives carrying the Ghd7-2 allele from DHX2. This work indicated that exploitation of Ghd7 will simultaneously improve eating quality and yield. Combining the superior alleles of PCF3 and Ghd7, as well as overexpression of the superior PCF3 allele, should further synergistically improve yield and quality in current materials with Ghd7.

Particularly, our findings further clarify the molecular mechanism underlying the functional variation of PCF3 in regulating rice quality and yield (Fig. 8). The T/TA variant located in the third exon of PCF3 acts as a key switch determining whether rice can achieve synergistic high quality and high yield. Specifically, the PCF3-T allele encodes a functional protein that maintains normal DNA-binding capacity, enabling specific binding to five CCCAC core motifs in the OsNF-YC4 gene. This binding directly inhibits the transcriptional activity of OsNF-YC4, a gene known to positively regulate grain protein content and negatively regulate starch content. The suppressed expression of OsNF-YC4 disrupts the original balance of carbon-nitrogen metabolism in the endosperm, leading to a significant reduction in grain protein content and a corresponding increase in starch content—two key changes that collectively improve rice eating quality (e.g., increased taste value and gel consistency, reduced chalkiness rate). In contrast, the PCF3-TA allele, due to the insertion of the A base, causes a frameshift mutation that impairs the structural integrity of the encoded protein. This structural defect abolishes its ability to recognize and bind to the CCCAC motifs of OsNF-YC4, as confirmed by the absence of specific DNA-protein complexes in EMSA assays. Consequently, OsNF-YC4 remains highly expressed, promoting protein accumulation in grains while inhibiting starch synthesis, ultimately resulting in deteriorated eating quality (e.g., decreased taste value, increased chalkiness) and failing to support yield improvement. This mechanism explains why the T allele of PCF3, but not TA, can act synergistically with Ghd7 to enhance both quality and yield.

Fig. 8. Proposed model for synergistic improvement of rice yield and quality by the Ghd7–triggered transcriptional regulation.

Ghd7 transcriptionally activates PCF3 and exerts a dominant epistatic effect on it. A natural T/TA sequence variant in the third exon of PCF3 acts as a key molecular switch coordinating rice yield and eating quality. The functional PCF3-T protein specifically binds six CCCAC core motifs of OsNF-YC4 (five in the promoter, one in the 3’UTR), significantly repressing its transcription. This leads to reduced grain protein accumulation, promoted starch synthesis, and improved eating quality. In contrast, the PCF3-TA allele undergoes a frameshift mutation due to base insertion, disrupting protein structure and completely losing its ability to bind the OsNF-YC4 promoter. Consequently, OsNF-YC4 remains highly expressed, causing excessive grain protein, insufficient starch, deteriorated eating quality, and no yield improvement. Thus, only the favorable PCF3-T allele can synergistically interact with Ghd7 to simultaneously achieve the breeding goals of high yield and superior quality in rice.

The discovery of PCF3 provide a groundbreaking perspective for crop breeding theory. For a long time, both the academic community and breeding practices have predominantly focused on genes subjected to artificial selection during domestication(e.g., Ghd7 for yield regulation, GS3 for grain shape control, Chalk5 for chalkiness modulation). These genes are considered as core targets for trait improvement, as they have accumulated favorable alleles through human directional selection. However, our population genetic analyses—including allele frequency distribution, selection signal scanning, Tajima’s D test, and phylogenetic network analysis—clearly demonstrate that the beneficial T allele of PCF3 already exists in wild rice (with a frequency of 37%) and shows no significant directional selection signatures across indica and japonica subspecies of cultivated rice, thus escaping traditional domestication selection. Even so, this allele can still synergistically improve yield (14.92% increase in yield per plant) and eating quality (7.45 increase in taste value, 28.99% decrease in chalkiness rate) by inhibiting OsNF-YC4 to balance starch and protein contents.

This finding challenges the common view that only artificially selected genes possess significant potential for trait improvement. On one hand, it reveals that wild rice and natural cultivated populations may harbor a wealth of silent genetic resources—genes that are not artificially screened but exhibit excellent functional performance. These genes have not lost diversity due to domestication bottlenecks; instead, they may retain regulatory logic more closely aligned with the natural growth laws of crops (e.g., the moderate regulation of carbon-nitrogen metabolism by PCF3, which avoids sacrificing other traits in pursuit of a single trait). On the other hand, it offers a promising strategy for breeding practice: beyond exploring known artificially selected gene loci, systematically screening for unselected but functionally critical genes in natural populations (e.g., mining homologous TCP family genes based on the CCCAC core binding motif of PCF3) can significantly expand the pool of breeding targets. This is particularly instructive for resolving the inherent contradiction of negative correlation between yield and quality in traditional breeding.

Compared with previously reported pathways^30,35 regulating yield and quality in rice, the transcription regulation triggered by Ghd7 exhibits distinct regulatory features and biological functions. The recently reported Ghd7–OsNAC42 module^35,47 also mediates carbon–nitrogen balance, but it primarily operates through antagonistic transcription factor interactions to modulate starch and protein metabolism, whereas a Ghd7-triggered transcriptional regulation in our study that directly represses OsNF-YC4 through PCF3. In contrast, DEP1¹⁴ mainly enhances yield by regulating nitrogen response and panicle development, with relatively minor effects on eating quality. Similarly, OsAAP6¹⁷ improves grain nutritional quality by increasing protein content but often compromises starch accumulation and taste value. The transcriptional regulation reported in this study uniquely integrates high yield and superior eating quality by fine-tuning starch–protein homeostasis, and the favorable allele combination Ghd7-1/4 + PCF3-T provides a promising breeding strategy that overcomes the inherent trade-off between yield and quality.

Importantly, the synergistic improvement of rice yield and quality achieved by utilizing Ghd7 requires precise modulation of its regulatory effects on growth duration to avoid the ecological adaptation risks caused by delayed heading. Particularly for rice production in early-maturing regions, the rational deployment of moderate functional Ghd7 alleles in combination with optimal haplotypes of photoperiod and core quality genes adapted to early-maturing genetic backgrounds serves as a core genetic strategy to break the inherent trade-offs among high yield, superior grain quality, and early maturity, thereby achieving the simultaneous improvement of these three key agronomic traits. Nevertheless, the phenotypic effects of Ghd7 on heading date and grain yield vary substantially across diverse genetic backgrounds. Its transcriptional abundance is also coordinately modulated by environmental cues including photoperiod and temperature. For instance, low-temperature conditions can induce the upregulation of Ghd7, which further aggravates heading delay⁴⁷. Accordingly, the genetic background of recipient varieties and the local photothermal resources of target ecological zones must be fully taken into account in practical breeding programs, so as to avert the adaptive risks arising from excessive prolongation of growth duration.

In summary, the core value of this study lies in three key aspects: first, it clarifies the molecular mechanism by which Ghd7-triggered transcriptional regulation controls the starch/protein balance in rice, identifies the dominant epistasis of Ghd7 over PCF3 as the genetic basis for the synergistic improvement of grain quality and yield, and meanwhile reveals that the TA/T variation in the third exon of PCF3 is the key determinant of whether rice can achieve the synergy between good eating quality and high yield; Second, in this study, we confirm that the PCF3‑T allele, which has not undergone domestication selection and naturally exists in wild rice, can synergistically increase rice yield, improve eating quality, and reduce the chalky grain rate. This finding expands our understanding of the conventional view that only artificially selected genes have breeding value and taps into the application potential of silent genetic resources in natural populations. Strong functional dependence of PCF3-T on Ghd7, mismatched selection priorities during early domestication, or linkage drag comparable to that of GmSW19^C48may collectively account for neutral evolution and the lack of intensive artificial selection at the PCF3 locus. Consequently, PCF3‑T persisted as an unselected yet valuable cryptic allele, whose breeding potential is only realized when combined with functional Ghd7 alleles. Third, it proposes a breeding strategy that combines the elite haplotypes of Ghd7 (Ghd7-1/4) with PCF3-T, providing direct and feasible molecular targets and theoretical support for overcoming the bottleneck of the yield-quality negative correlation in rice, and thus promoting efficient and high-quality rice breeding practices.

Methods

Plant materials and field experiments

Kongyu131 (KY131), a japonica rice variety introduced from Japan, currently dominates rice planting areas in Northeast China. It has early maturity, cold resistance, lodging resistance, high tiller count, and high yield. However, it is susceptible to rice blast. Like other varieties, KY131 is lower yielding when grown at low latitudes^37,49. DHX2 is the japonica variety with the best tasting quality in China, comparable to that of the famous Japanese rice variety Koshihikari. It has the highest rice quality standards stipulated by the Chinese Ministry of Agriculture and Rural Affairs. In addition, DHX2 is late-maturing, lacks cold tolerance and is prone to lodging, although resistant to rice blast. Although both KY131 and DHX2 share the Wx^b allele and have similar amylose contents (with only a 1–3% difference as per our measurement), artificial sensory evaluation reveals a distinct difference in taste between them. Guichao2, owing to its high amylose content in grains, is a specialized rice variety used for the production of rice noodles and vermicelli. This variety also exhibits superior plant architecture, uniform panicle emergence, good grain filling color, a high number of effective panicles, high grain setting rate, and high single-plant yield. Since 1982, the cumulative promotion area of this variety has reached more than 100 million mu (from China Rice Data Center). TB154E-TB-2 is an indica rice variety from the 533 mini core collection, and its database number is W218.

The construction of KY131 NILs carrying the Ghd7-2 allele is described in Results. Like NIL(DHX2), NIL(NYZ) also carries the Ghd7-2 allele, whereas NIL(ZS97) has a null allele. The difference is that the NILs were developed in the indica rice variety ZS97³³. Transgenic ZH11 (japonica) lines OX-Ghd7 and Ami-Ghd7 were described elsewhere³⁴. The NILs and recipient and donor parents were planted in two environments for evaluation of quality and yield traits, i.e. Wuchang (Heilongjiang Province, China, N44°4′, 126°33′/E) and Wuhan (Hubei, N29°58′, 113°41′/E) in 2023. The sowing times mid-April and mid-June, respectively, while the sowing time of other materials in Wuhan is May 15^th each year. For fine mapping of qAC7, a BC₄F₂ population consisting of 2400 individuals was grown in Wuhan in summer 2021. The 430 recombinants were progeny tested in Wuhan in summer 2022. Except as specifically stated, other genetic materials were grown in the field at the Experimental Stations of Huazhong Agricultural University. Plants were spaced 16.5 cm in each row, and rows were 26 cm apart. Field management, including irrigation, fertilizer application and pest control, followed local practices.

Measurement of grain quality and agronomic traits

Harvested rice grains were air-dried and stored at room temperature for at least 3 months before testing. Fully filled grains were used for measuring grain quality and yield traits. Polished rice was ground into flour and used to measure amylose content, four storage protein contents, and gel consistency. Briefly, four seed‑storage‑protein fractions (albumin, globulin, prolamin, glutelin) were sequentially extracted from 0.1 g milled rice flour using corresponding solvents: 10 mM Tris‑HCl (pH7.5) for albumin; 1 M NaCl for globulin; 60% n‑propanol with 1 mM EDTA‑2Na for prolamin; and 0.05 M NaOH for glutelin. Each extraction was shaken for 2 h at room temperature, followed by centrifugation (12000 rpm, 15 min, 4°C). The extraction cycle was performed in triplicate. Protein fractions were quantified via the Bradford assay with bovine serum albumin as the standard, measured on an Infinite M200 microplate reader. Amylose content was determined by the iodine colorimetric method. Gel consistency was measured based on the length of cold‑cooled rice gel spread on horizontal surface after gelatinization in dilute alkali solution. Total protein content of dry dehulled grains was measured using an XDS Near-Infrared Rapid Content Analyzer (FOSS), with near-infrared reflectance spectroscopy¹⁷. Chalkiness is evaluated using two indicators: chalky grain rate and chalky area. The chalkiness and grain morphological traits of rice were evaluated using the SC-E rice appearance quality inspection analyzer (Hangzhou Wanshen Testing Technology Co., Ltd., China). For each sample, 200–300 polished rice grains or paddy grains were randomly selected for instrumental detection. The chalkiness rate, chalky area, grain length, and grain width were measured via the image analysis system of the instrument, while the 1000-grain weight was determined using an analytical balance matched and connected to the analyzer.

Total‑starch content of milled rice flour was determined with the total‑starch assay kit (K‑TSTA, Megazyme, Ireland). For eating‑quality value measurement, ~10 g polished rice grains were washed, soaked in distilled water for 30 min, steamed for 30 min and kept warm for 20 min, then cooled for 20 min. Approximately 8 g of cooked rice was shaped into rice balls and assayed by a taste analyzer (Satake, STA1B‑RHS1A‑RFDM1A, Japan) following the manufacturer’s protocol. Textural properties of cooked rice were characterized by texture‑profile analysis using a texture analyzer with a 25‑mm cylinder probe. Instrument parameters: pre‑test speed 60 mm/min, test speed 30 mm/min, post‑test speed 60 mm/min, 60% strain, holding time 2.00 s, trigger force 0.15 N. All assays were performed with three biological replicates.

Plant height was measured from the ground to the tip of the tallest tiller. The heading date was the day of emergence of the first panicle. Agronomic traits were measured as the means of a set of plants with similar growth for each line or genotype.

Generation of constructs and transformation

Various DNA fragments were amplified and inserted into binary vectors for plant transformation. To prepare the constructs for CRISPR-Cas9, we designed target sites at bases 177–196 in the second exon of the Ghd7, bases 100–122 of the first exon of OsNF-YC4 and bases 20–42 of the first exon of PCF3, and bases 51–70 of the third exon of PCF3 (KO-Ghd7, KO-P3Y4 and KO-PCF3), respectively. The online tool CRISPR GE was used to design the target sites⁵⁰. Briefly, the U6-sgRNA cassettes of KO-Ghd7 and KO-PCF3 were PCR-amplified and inserted into KpnI-digested pCXUN. For KO-P3Y4, the PCF3 U6-sgRNA and OsNF-YC4 U3-sgRNA cassettes were sequentially cloned using KpnI and SacI to yield pCXUN-U6-U3-sgRNA. To prepare the complementation construct (CO), we first obtained an ~2 kb promoter fragment and the full-length cDNA of Ghd7 by PCR amplification from DHX2 and inserted them into the plant binary vector pCAMBIA1301. To prepare the overexpression vectors OX-PCF3, we first obtained the full-length cDNA of PCF3 from ZS97 by PCR amplification and then inserted it into the plant binary vector PU2301-flag. The KO-Ghd7 construct was introduced into the rice cultivar Nipponbare, the KO-P3Y4 construct was introduced into the ZH11, the OX-PCF3 construct and KO-PCF3 were introduced into the KY131 NILs, and the remaining construct CO was introduced into the rice cultivar KY131. All the resulting constructs were introduced into Agrobacterium tumefactions strain EHA105 and transferred by Agrobacterium-mediated transformation⁵¹ into recipient genotypes. All primers are listed in Supplementary Data 5.

RNA reverse transcription and RT-qPCR

Total RNA from plant tissues was isolated using Trizol regent (Invitrogen); 4μg of RNA was treated with DNase I and reverse-transcribed with M-MLV reverse transcriptase (Promega Madison, WI, USA) and Oligo (dT) 18 primers (Promega). The concentration of cDNA was measured using a NanoDrop2000 (Thermo Scientific) and diluted to 10ng/μL for use in subsequent qPCR amplification using a mixture containing 6μL template cDNA, 5μL 2× FastStart Universal SYBR Green Master (ROX) Mix (ROCHE), and 0.25mM of forward and reverse primers. Amplifications were carried out on an ABI7500 PCR instrument (Applied Biosystems). Rice constitutive expression gene OsActin (LOC_Os03g50885), was used as the internal control. All primers are listed in Supplementary Data 5.

Electrophoretic mobility shift assays

GST, Ghd7-GST, and MBP-PCF3 were purified using Glutathione Sepharose 4B (GE Healthcare, 17-0756) following the manufacturer’s instructions. DNA probes were synthesized and labeled with biotin by Shenggong Bioengineering Co., Ltd (Shanghai). DNA gel-shift assays were performed using a LightShift Chemiluminescent EMSA Kit (Thermo Fisher Scientific, 20148). Briefly, purified proteins were incubated with biotin-labeled DNA probes. The mixture was separated by native PAGE, and the binding bands were detected via chemiluminescence.

Transactivation activity assays

The promoters of target genes from Nipponbare for transactivation activity assay were cloned into the pGreenII 0800-LUC vector to measure transcriptional activity, and the Renilla luciferase gene was used as an internal control to provide an estimate of the extent of transient expression in the same construct. Full-length cDNA of Ghd7 and PCF3 were amplified from Nipponbare cDNA and inserted into the None vector to generate effector plasmid vectors. Rice protoplasts were prepared from leaf‑sheaths of dark‑grown rice seedlings. Briefly, etiolated leaf‑sheaths were cut into ~0.5 mm strips, vacuum‑infiltrated with enzyme solution containing cellulase R‑10 and macerozyme R‑10 for cell‑wall digestion. After gentle shaking and digestion in darkness, protoplasts were filtered through nylon mesh, harvested by low‑speed centrifugation, washed with W5 solution, and resuspended in MMG solution for use. The rice protoplasts were transfected with different combinations of vectors overnight, then harvested and lysed for detection of firefly luciferase activity (Promega, E1960), according to the manufacturer’s recommendations. Relative luciferase activity was defined by the ratio of LUC to Ren activity. All primers are listed in Supplementary Data 5.

DAP-seq assays and data analysis

Genomic DNA was extracted from 14-day-old hydroponic seedlings (whole seedlings) of Oryza sativa cv. Nipponbare using MICH DNA Clean Beads (Lan Jing Biotechnology) with two biological replicates, and an aliquot of the same DNA was reserved as an input control; libraries for both samples and the input control were constructed using the MICH TLX DNA-Seq Kit (Lan Jing Biotechnology) following the manufacturer’s protocol. DAP-seq was performed as described⁵². In brief, PCF3 protein was synthesized using the TNT SP6 Coupled Wheat Germ Extract System (Promega), immobilized on Magne HaloTag Beads (Promega), and incubated with the DNA libraries; unbound DNA was removed by three washes with 150 mM NaCl buffer containing 0.1% NP-40, bound DNA fragments were eluted and amplified with indexed primers, while Input control underwent the same amplification without protein binding or washing steps. Sequencing was conducted on an Illumina NovaSeq platform with ≥20 million reads per sample, reads were aligned to the O. sativa cv. Nipponbare genome (IRGSP-1.0) using Bowtie2⁵³, and PCF3 binding peaks were called using MACS2⁵⁴ with Input as background, applying thresholds of fold enrichment >2 and q-value < 0.05, with high-confidence peaks (IDR < 0.05) from replicate analyses retained. Peak distribution around transcription start sites (±1 kb) was analyzed using Homer, target genes were identified by overlapping peaks with gene annotations using BEDtools⁵⁵, motif discovery was performed with MEME-ChIP v5.0.5⁵⁶ using Input control for background correction (focusing on TCP family motifs), and visualization was done using IGV.

Yeast one-hybrid assay

For yeast one-hybrid (Y1H) analysis, the coding sequence of Ghd7 harboring the conserved CCT domain was amplified and cloned into the pB42AD vector. Two fragments of the PCF3 promoter were prepared separately: a 192-bp wild-type promoter segment (designated PCF3-WT, primer sequences are listed in Supplementary Data 5) and a mutated PCF3 promoter fragment with artificially substituted cis-element motifs. The full sequence of the mutated promoter (GTTCCACCACCTCCAGCAGCAGCAGCAGCACGACGTCGCCGCCATGCTCGGCTTTTTTTATTTTTATTTTTAGCTGCTCCCTCCTCCGTTTTTGCACCAGCACCCGGAGCCGACGCCCCAAGATCCCGGCGCCGGCGAGTTCATGCGTAAGCGCTACAGGGAAGCCGACGACCTGTTCAAGGACACCTCACG) was chemically synthesized by Zhejiang Shangya Biotechnology Co., Ltd., and subsequently inserted into the reporter vector pLacZi. The recombinant pB42AD-Ghd7 and pLacZi-proPCF3 (WT/mutant) plasmids were co-transformed into the yeast strain EGY48. Transformants were first selected on synthetic dropout medium lacking tryptophan and uracil (SD/−Trp/−Ura). Positive yeast clones were further spotted onto SD/−Trp/−Ura solid medium supplemented with 1× BU salts, 2% (w/v) galactose, 1% (w/v) raffinose, and 80 mg L⁻¹ X-gal to test physical interactions between Ghd7 protein and the PCF3 gene fragment. All primers are listed in Supplementary Data 5.

Subcellular localization

The full-length cDNA of Ghd7 and PCF3 was amplified by PCR using DHX2 cDNA and Zhenshan97 cDNA, respectively. The amplified Ghd7 and PCF3 cDNA fragments were inserted into a premade YFP vector in pM999 via SacI and XbaI restriction enzyme sites to generate Ghd7-YFP and PCF3-YFP. The recombinant plasmids were co-expressed in rice leaf sheath protoplasts together with various marker plasmids by PEG-mediated transfection. Protoplasts were prepared as described above and were observed using a confocal laser scanning microscope and visualized by a Leica Microsystem LAS AF. All fluorescence experiments were repeated at least three times.

Genome-wide association study

A diverse natural rice population was used for genome-wide association analysis. Publicly available genomic sequencing data and systematically collected phenotypic data were applied for GWAS, which was performed using standard bioinformatic software and statistical pipelines.

Phylogenetic and haplotype analyses, nucleotide diversity and neutrality tests, selective sweep analysis

Single-nucleotide polymorphism (SNP) and insertion-deletion (Indel) data of the PCF3 gene and its flanking regions from 533 rice accessions were retrieved from the RiceVarMap v2.0 database. These accessions constitute a globally representative mini-core collection covering major rice-growing regions worldwide, and were systematically classified into four cultivated subgroups: Indica (295 accessions, 55.5%), Japonica (156 accessions, 29.4%), Aus (46 accessions, 8.7%), and VI (intermediate subgroup, 32 accessions, 6.0%)⁴². The resequencing data of 106 wild rice accessions were obtained from the online resource (http://viewer.shigen.info/oryzagenome/). Haplotype analysis was performed using WPS Excel (a product under Kingsoft Office Software). Haplotype networks were constructed with PopART 1.7 software. Phylogenetic trees were built via the Neighbor Joining method using Phylip 3.697 software, with 1000 bootstrap replications for reliability assessment, and visualized using the online tool iTOL⁵⁷. Nucleotide diversity, F_st, and Tajima’s D test were estimated using the PopGenome package in R programming language.

Principal component analysis (PCA) was conducted based on all SNPs using EIGENSOFT software (v4.2)⁵⁸. This software package employs PCA on genetic data to analyze population structure. Here, the three-dimensional scatter plot was generated using the first three principal components (PC1, PC2, and PC3), which explained 6.66%, 3.36%, and 2.39% of the variance, respectively. The plotting was carried out with the built-in R functions prcomp() and princomp(), and samples were color-coded to distinguish among the T, TA, and Wild groups. In the selective sweep analysis, the log₂ value of the θπ ratio (the ratio of θπ,533Admix to θπ,Wild) and Fₛₜ values were first calculated. Subsequently, regions were classified according to the thresholds of log₂ θπ ratio (<−0.888, −0.888 to 0.878, > 0.878) and Fₛₜ threshold (0.202). With these criteria, selected regions in the 533Admix (blue points) and Wild (red points) populations were identified, while gray points represented non-selected regions. Meanwhile, density plots and cumulative distribution plots were used to present the distribution patterns of these values^59,60.

Statistical analysis

Experimental data were analyzed using Excel 2016 or IBM SPSS Statistics 22.0 for basic statistical analyses, including t tests, correlation coefficients and ANOVA. One-way ANOVA and Duncan’s multiple range tests were used to assess significant differences in the data. The schematic diagram in Fig. 8 was created using Adobe Illustrator 26.0 (64‑bit) and Microsoft PowerPoint (Microsoft PowerPoint for Microsoft 365MSO, Version 2607 (Build 16.0.20228.20190), 64‑bit).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

G.L. and H.G. performed most of the experiments; the preliminary positioning of qAC7 was completed by W.R., and the fine positioning and cloning of qAC7 was completed by Y.T.; G.J. constructed and provided the dihaploid population derived from KY131 and DHX2 used in this study; genome-wide association analysis and partial drawing were performed by Y.T., M.L., S.C., Y.C., and R.Y.; P.C. created the near-isogenic line material of Ghd7 in the Zhenshan 97 background; Z.Z., Y.Z., B.W., and J.C. participated in some quality determination tests; Y.L. provided guidance for determination of some quality traits; H.S., G.G., Y.W., and P.Y. participated in field experiments; Y.L. generously provided some of the macromolecular interaction vectors used in this experiment; Y.L., X.L., D.Z., and L.X. improved the paper; A.Y. supply partial fund, Y.H. designed experiments; Y.H. and G.L. wrote the manuscript; all authors discussed and commented on the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by grants from the Ministry of Science and Technology (2022YFD1200100, 2021YFF1000200), the Innovation Program of Chinese Academy of Agricultural Sciences (No. CAAS-CSCB-202402), Hubei Science and Technology(2024BBB008, 2024BBA005) and Hainan Seed Industry Laboratory and China National Seed Group(B24YQ0006, B24CQ006P).

Data availability

The ChIP-seq data reported previously³⁵ are available in the National Genomics Data Center database under accession CRA010211. The DAP-seq data generated in this study have been deposited in the National Genomics Data Center database under accession CRA049929. Source data are provided with this paper.

Competing interests

Authors declare that they have no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-78253-x.