Natural variation in GNP3 determines grain number and grain yield in rice

Abstract

Grain number per panicle critically determines rice yield. Although many underlying genes have been reported, yet the molecular mechanisms linking ethylene to panicle development remain unclear. Here, we identify GRAIN NUMBER PER PANICLE 3 (GNP3) as a regulator of GNP through genome-wide association study (GWAS) combined with map-based cloning. GNP3 encodes a MITOGEN-ACTIVATED PROTEIN KINASE KINASE KINASE 22 (OsMKKK22) that phosphorylates S-adenosyl-L-methionine synthetase 1 (SAMS1), triggering its degradation to suppress ethylene biosynthesis. Ethylene overaccumulation in gnp3−1 mutants reduces grain number, while GNP3 overexpression enhances panicle branching and grain yield by lowering ethylene levels. We demonstrate that a natural haplotype GNP3^Hap-Tprevalent in indica subspecies strengthens GNP3-SAMS1 interaction, accelerating SAMS1 degradation and improving grain number. Furthermore, overexpressing GNP3 increases grain yield by approximately 20% in field plot conditions. Our findings unveil a MAPK-ethylene regulatory module and highlight GNP3^Hap-T as a valuable genetic resource for breeding high-yield rice.

Introduction

Rice (Oryza sativa L.) is a crucial dietary mainstay in many regions globally, most notably in Asia. Increasing rice yield is an important goal to ensure overall food security. Grain yield in rice is a complex trait that is affected by three main components: grain number per panicle (GNP), panicle number per plant (PN), and grain weight (GW)¹. As GNP is more variable and adjustable than the other two characteristics, increasing GNP is an important target to achieve higher rice yield^2,3. Recent studies have identified multiple genes regulating GNP that encode diverse functional proteins⁴. These encompass transcription factors (TFs) like GRAIN NUMBER ASSOCIATED (GNA) and REGULATOR OF GRAIN NUMBER 1 (RGN1), phytohormone metabolic enzymes such as GRAIN NUMBER 1a (Gn1a), LONELY GUY (LOG), and GY3 (encoding a LOG protein), and signaling kinases such as ERECTA 1 (OsER1), among others^5,6,7,8,9,10. Mechanistic investigations reveal that these genes function through distinct pathways: GNA suppresses Gn1a transcription in the inflorescence meristem¹⁰; RGN1 promotes cytokinin biosynthesis by directly activating LOG expression⁸; OsER1 regulates panicle development through the OsMKKK10-OsMKK4-OsMPK6 phosphorylation cascade⁷. Despite these breakthroughs, most characterized alleles originate from mutagenesis screens rather than natural variation, limiting their direct applicability in modern breeding programs.

Mitogen-activated protein kinase (MAPK) cascades, comprising MKKK-MKK-MPK phosphorylation modules, serve as central signaling hubs integrating developmental and environmental cues^11,12. There are a total of 75 MAPKKK, 8 MAPKK, and 17 MAPK genes in rice¹³. In rice, distinct MAPK pathways have been implicated in both yield formation and stress adaptation. For instance, the OsMKKK10-OsMKK4-OsMPK6 cascade dually regulates GNP and grain size¹⁴, while OsMPK6-mediated phosphorylation of DST transcription factor modulates drought responses through cytokinin homeostasis⁶. The OsMKKK70-OsMKK4-OsMPK6 cascade acts upstream of OsWRKY53 to regulate grain size and leaf angle¹⁵. OsMKK6 transcript was induced by cold stress and it could interact with OsMPK3 to enhance chilling tolerance of rice seedlings¹⁶. However, the functional repertoire of rice MKKKs remains underexplored, particularly regarding their downstream targets beyond canonical MKK-MPK partners.

S-adenosyl-L-methionine (SAM) is the precursor of 1-aminocyclopropane-1-carboxylic acid (ACC), the direct precursor of ethylene¹⁷. The conversion of methionine to SAM is catalyzed by SAM synthetases (SAMSs)¹⁸. The rice genome contains three SAM synthetase-encoding genes, among which OsSAMS1 overexpression has been shown to elevate endogenous ethylene levels¹⁹. Recent biochemical characterization identified a putative protein kinase-like protein (OsLCD3) that physically interacts with OsSAMS1 to regulate grain size determination, although no direct phosphorylation was detected between these two proteins²⁰. Maize studies demonstrate that ethylene suppression via ACC OXIDASE 2 (ZmACO2) downregulation enhances kernel number²¹. Emerging evidence suggests intricate crosstalk between ethylene signaling and grain development. Ethylene evolution rates are significantly and negatively correlated with cell division of the endosperm, grain-filling rate and grain weight of inferior spikelets in rice²². Notably, several ethylene-responsive transcription factors associated with the regulation of GNP have been characterized in rice. Among these, FZP, OsAP2-39, and OsERF48 have been demonstrated to negatively regulate panicle architecture and grain formation^23,24,25. However, the molecular mechanisms linking ethylene biosynthesis to GNP accumulation remain incompletely understood, particularly the genetic components that integrate ethylene-mediated pathways with developmental programs governing panicle branching and spikelet differentiation.

Here, we report the cloning of GNP3 through combined genome-wide association study and map-based approaches, revealing its role as a grain number regulator. GNP3 encodes OsMKKK22, which regulates GNP by phosphorylating OsSAMS1 to fine-tune ethylene biosynthesis. Furthermore, we identify a natural elite haplotype GNP3^Hap-T with significant breeding potential, providing insights into the improvement of yield in rice.

Results

Identification of the candidate gene for the QTL qGNP3-3

To search for genes affecting GNP, we phenotyped a panel of 317 cultivated rice accessions for GNP, which ranged from 22.8 to 304. We then used these GNP values for GWAS using a compressed mixed linear model (Supplementary Data 1, Supplementary Fig. 1a). We identified 11 QTLs associated with GNP, distributed over most rice chromosomes (Fig. 1a; Supplementary Fig. 1b; Supplementary Table 1). In this study, we focused on the QTL peak designated qGNP3-3, located on chromosome 3, as it showed the strongest signal in GWAS (Fig. 1a). We defined a 255.59-kb interval using linkage disequilibrium over the 600-kb region around the most significant single nucleotide polymorphisms (SNPs) of qGNP3-3 (Fig. 1b). The annotation of the rice genome at the Rice Genome Annotation Project (RGAP, http://rice.plantbiology.msu.edu/) indicated the presence of 34 genes in this interval, of which only 10 contained significant SNPs (–log10 (P observed) > 4.5) (Supplementary Table 2). We explored the transcriptome data from the Rice Bio-Analytic Resource for Plant Biology website (https://bar.utoronto.ca/eplant_rice/), which revealed that the gene LOC_Os03g49790 was not expressed at any growth stage and that LOC_Os03g49640 was the most highly expressed in inflorescences at different stages compared to the other genes in the interval (Supplementary Fig. 1c). We speculated that LOC_Os03g49640 might play an important role in inflorescence development and GNP formation. LOC_Os03g49640 encodes OsMKKK22, which is conserved in plants (Supplementary Fig. 2). OsMKKK28 is highly homologous to MPKKK22, but the expression of OsMKKK28 is significantly lower than that of MPKKK22 in the developing young panicles (Supplementary Fig. 3). Moreover, loss of OsMKKK22 function produces seedlings with yellow leaves and lower chlorophyll contents before they die, indicating that this gene is indispensable for plant growth²⁶. These results prompted us to further focus on LOC_Os03g49640, which is highly likely to be the candidate gene for qGNP3-3 and was subsequently referred as GNP3.

Fig. 1: Identification of GNP3 by GWAS.

a Manhattan plot of GWAS results. b Regional Manhattan plot of qGNP3-3 and pairwise linkage disequilibrium analysis. The lead (most significant) SNP is shown in red. c Representative panicles of wild-type (WT), GNP3-CR1, GNP3-CR2, and GNP3-CR3 plants. Scale bar, 2 cm. d–g Panicle length (PL) (d), primary branch number (No. of PB) (e), secondary branch number (No. of SB) (f), and grain number per panicle (GNP) (g) of WT, GNP3-CR1, GNP3-CR2 and GNP3-CR3 plants (n = 25 plants). h Relative GNP3 expression levels in WT and GNP3 overexpression lines (n = 9, three biological replicates, each biological replicate consists of three technical replicates). i Representative panicles of WT, GNP3-OE1, GNP3-OE6, and GNP3-OE14 plants. Scale bar, 2 cm. j–m Panicle length (j), primary branch number (k), secondary branch number (l), and grain number per panicle (m) of WT, GNP3-OE1, GNP3-OE6 and GNP3-OE14 plants (n = 26/25/25/25 plants). In each boxplot, the center line indicates the median, the box edges represent the first and third quartiles, and the whiskers extend to the minima and maxima. Lowercase letters denote significant differences (one-way ANOVA with Tukey multiple-range test, P < 0.05). Source data are provided as a Source Data file.

GNP3 positively regulates GNP in rice

To assess the function of GNP3, we attempted to generate mutants in this gene via clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated protein 9 (Cas9)-mediated gene editing in the Nipponbare (Nip) background. We obtained four kinds of gene-edited plants: CR1 harbors a 46-bp deletion, CR2 a 41-bp deletion, CR3 a 1-bp insertion, and CR4 a 48-bp deletion in the first exon of the gene (Supplementary Fig. 4a). We failed to obtain homozygous CR1, CR2, or CR3 mutant plants in the T₀ or T₁ generation; we did isolate homozygous CR4 plants, which retained the complete kinase domain but lacked 16 amino acids near the N terminus. We noticed some dead seedlings in CR1, CR2 and CR3 lines while growing the T₂ generation in the paddy field; genotyping and sequencing analysis revealed that these dead seedlings harbored homozygous frameshift mutations (Supplementary Fig. 4b). We concluded that GNP3 is essential for plant survival, in agreement with previous results²⁶, and chose heterozygous GNP3-CR1, GNP3-CR2 and GNP3-CR3 plants for phenotypic characterization. Compared to wild-type (WT) Nip plants, GNP3-CR1, GNP3-CR2, and GNP3-CR3 displayed significantly shorter panicles, fewer secondary branches, and lower GNP, while the number of primary branches was not different from that of WT (Fig. 1c–g). We also noticed the development of a spotted leaf phenotype on the leaves of heterozygous GNP3-CR1, GNP3-CR2, and GNP3-CR3 plants, which began at the late tillering stage (~90 days after planting), gradually spreading from lower leaves to upper leaves (Supplementary Fig. 4c–e). This phenomenon was accompanied by consistently lower chlorophyll content measurements (SPAD values) compared to WT controls (Supplementary Fig. 4f).

To further explore the function of GNP3, we generated GNP3 overexpression lines, using the cauliflower mosaic virus (CaMV) 35S promoter, in the Nip background; we chose three independent lines (GNP3-OE1, GNP3-OE6 and GNP3-OE14) with increased GNP3 expression levels (Fig. 1h). The GNP3-OE1, GNP3-OE6, and GNP3-OE14 plants demonstrated significantly increased secondary branch numbers and GNP compared to WT, while maintaining comparable panicle length and primary branch numbers (Fig. 1i–m). These phenotypes were opposite those seen for the gene-edited GNP3 plants described above. Moreover, the SPAD values significantly increased in overexpression plants relative to the WT (Supplementary Fig. 5). Taken together, these results indicate that GNP3 is indispensable in rice development and positively regulates GNP.

Characterization of the gnp3-1 mutant

We serendipitously identified another mutant generated from tissue culture that showed spotted leaves similar to those seen in the gene-edited GNP3 plants when grown in a paddy field (Fig. 2a). At the seedling stage, the mutant appeared quite similar to the WT (Nip). From the late tillering stage onward, lesion-mimic spots gradually appeared and spread from lower leaves to upper leaves, ultimately expanding to all leaves. The older leaves gradually withered and died; this effect was more severe than seen in GNP3-CR plants (Supplementary Fig. 6a). The formation of spots on leaves did not depend on light exposure, as spots appeared even on a shaded leaf (Supplementary Fig. 6b). Transmission electron microscopy observation of leaf thin sections revealed more starch granules and osmiophilic granules in the chloroplasts of mutant leaves than in WT (Supplementary Fig. 6c).

Fig. 2: Identification of GNP3 by map-based cloning.

a Representative phenotype of WT and the transposon insertion mutant gnp3-1 at the booting stage. Scale bar, 10 cm. b Representative panicle of WT and gnp3-1. Scale bar, 2 cm. c–f Panicle length (c), primary branch number (d), secondary branch number (e), and grain number per panicle (f) of WT and gnp3-1 plants (n = 30 plants). g Fine-mapping of GNP3. The numbers below the chromosome indicate the number of recombinant individuals. h Relative expression levels of ORF6, ORF7, ORF8, and ORF9 in the panicle of WT and gpn3-1 mutant. Data represent means ± SD (n = 3 biological replicates). P1 and P2 are the ORF8 primers used for RT-qPCR analysis, hybridizing to the first and second exons. The green triangle represents the insertion of the 983-bp transposon. i, n Representative phenotypes of panicle (i) and leaf (n) for WT, GNP3 complementation lines (GNP3-CM7, GNP3-CM10, and GNP3-CM2) and gnp3-1. Scale bar, 2 cm (i); Scale bar, 1 cm (n). j–m Panicle length (j), primary branch number (k), secondary branch number (l), and grain number per panicle (m) of WT, GNP3-CM7, GNP3-CM10, GNP3-CM2, and gnp3-1 plants (n = 25/25/26/25/25 plants). In (c–f, h), asterisks indicate statistical significance by two-tailed Student’s t tests (*P < 0.05, **P < 0.01). In (c–f, j–m), the center line indicates the median, the box edges represent the first and third quartiles, and the whiskers extend to the minima and maxima. Lowercase letters denote significant differences (one-way ANOVA with Tukey multiple-range test, P < 0.05). Source data are provided as a Source Data file.

In addition, this mutant had significantly shorter panicles, fewer primary and secondary branches, and a lower GNP compared to WT plants (Fig. 2b–f). These phenotypes were reminiscent of the gene-edited GNP3 mutants, raising the possibility that this spontaneous mutant might have compromised GNP3 function. When we crossed GNP3-CR1 to the spontaneous mutant, the F₁ leaves were light green and covered in spots, indicating that the two mutants are allelic; we therefore named this mutant gnp3-1 (Supplementary Fig. 6d). As an independent confirmation, we crossed gnp3-1 mutant to the rice variety BS1300 to generate a segregating population for mapping. We narrowed down the candidate region for the causal gene to a 119.8-kb region between markers STS20 and STS13 on chromosome 3, containing 19 predicted genes (Fig. 2g). DNA sequencing analysis of this region revealed a 983-bp fragment insertion in the 5′ untranslated region (UTR) of the eighth open reading frame (ORF) in this interval, which markedly decreased the expression level of ORF8 in gnp3-1 mutant compared to the WT (Fig. 2h). According to the RGAP, ORF8 is LOC_Os03g49640. Subsequently, we successfully constructed a complementation (CM) clone and introduced it into the gnp3-1 mutant. The CM lines had normal panicle architecture and GNP and lacked leaf spots (Fig. 2i–n, Supplementary Fig. 6e, f). These results demonstrate that GNP3 is responsible for the gnp3-1 phenotype and controls GNP in rice.

Expression pattern of GNP3 and subcellular localization of GNP3

To investigate the spatiotemporal expression pattern of GNP3, we conducted a reverse transcription quantitative PCR (RT-qPCR) analysis, which showed that GNP3 was highly expressed in various tissues, especially in leaves. Relative GNP3 expression levels were significantly lower in gnp3-1 compared to WT (Supplementary Fig. 7a). Histochemical analysis of transgenic plants harboring a ProGNP3:GUS reporter construct (with the GNP3 promoter driving the ß-glucuronidase [GUS] reporter gene) revealed strong GUS activity in various tissues, especially panicles, roots, and leaves (Supplementary Fig. 7b).

To examine the subcellular localization of GNP3, we generated a 35S:GNP3-GFP construct, consisting of the CaMV35S promoter driving a cassette encoding GNP3 fused to the green fluorescent protein (GFP). A confocal microscopy investigation detected green fluorescence from GNP3-GFP protein at the cell periphery and in the cytoplasm of transiently transfected rice protoplasts, while GFP alone was distributed throughout the cell (Supplementary Fig. 8a–c). Additionally, we created transgenic plants expressing the ProGNP3:GNP3-GFP construct and detected similar GFP fluorescence localization in root cells through light-sheet fluorescence microscopy (Fig. 3a).

Fig. 3: GNP3 interacts with SAMS1 in vitro and in vivo.

a Subcellular localization analysis revealed GFP fluorescence patterns in roots of ProGNP3:GNP3-GFP and ProSAMS1:SAMS1-GFP transgenic rice plants using light-sheet fluorescence microscopy. Scale bar, 50 μm. The experiments were performed with at least three independent biological replicates. b, c MbY2H assay showing the interaction of GNP3 with SAMS1. The kinase domain of GNP3 is indicated in green. Black vertical lines indicate a lysine-to-methionine change at aa 410 indicating the inactive version of GNP3 (K410M). SD/–T–L, synthetic defined medium lacking Trp and Leu; SD/–T–L–H–A, synthetic defined medium lacking Trp, Leu, His, and Ade. d GST pull-down assay showing that GNP3 (aa 143–655) interacts with SAMS1 in vitro. e GNP3 interacts with SAMS1 in rice protoplasts. f BiFC assays showing the interaction between GNP3 and SAMS1 in N. benthamiana leaf epidermal cells. As positive controls, GSK3 showed interaction with GL3.1. Scale bar, 50 μm. Source data are provided as a Source Data file.

GNP3 interacts with SAMS1

To identify GNP3-interacting proteins, we constructed a cDNA library based on the split-ubiquitin membrane yeast two-hybrid system (MbY2H). We discovered that SAMS1 could interact with GNP3 in yeast. To determine the interaction domain between GNP3 and SAMS1, we employed truncated forms of GNP3 as bait. Our findings revealed that amino acids (aa) 143-230 of GNP3 are essential for its interaction with SAMS1. A kinase-dead version of GNP3 (harboring the K410M mutation) showed a stronger interaction with SAMS1 than did intact GNP3 (Fig. 3b, c). Moreover, a glutathione S-transferase (GST) pull-down assay indicated that His-tagged GNP3 (aa 143-655) interacted with GST-SAMS1 in vitro (Fig. 3d). We further confirmed the interaction between GNP3 and SAMS1 using co-immunoprecipitation (Co-IP) assays. We co-transfected Nip protoplasts with the constructs 35S:GNP3-HA and 35S: SAMS1-Myc. We immunoprecipitated proteins with HA beads and probed the precipitates by immunoblot analysis using anti-HA and anti-Myc antibodies. SAMS1-Myc co-immunoprecipitated with GNP3-HA, but we detected no SAMS1-Myc from protoplasts transfected with the HA empty vector as negative control (Fig. 3e). The SAMS1-GFP fusion was found to be localized within the cytoplasm and nucleus in transfected rice protoplasts and in the root cells of ProSAMS1:SAMS1-GFP transgenic plants (Fig. 3a; Supplementary Fig. 8a, c). A biomolecular fluorescence complementation (BiFC) assay showed that GNP3 fused to the C-terminal half of the yellow fluorescent protein (cYFP-GNP3) interacted with a fusion between SAMS1 and the N-terminal half of YFP (nYFP-OsSAMS1) in cytoplasm, as evidenced by the reconstitution of YFP fluorescence in N. benthamiana leaf epidermal cells (Fig. 3f). We conclude that GNP3 interacts with SAMS1 in vitro and in vivo.

GNP3 phosphorylates SAMS1 and decreases its stability

As GNP3 is a kinase that interacts with SAMS1, we examined whether GNP3 might phosphorylate SAMS1. We produced and purified recombinant GST-GNP3 (aa 143-655) containing the kinase domain and SAMS1-His from Escherichia coli. We then set up in vitro kinase reactions by incubating recombinant GST-GNP3 (143-655) or GST as control, SAMS1-His, and ATP, followed by separation of the protein on a Phos-tag gel. An immunoblot analysis with an anti-His antibody detected a shift in the mobility of SAMS1-His that was abolished by treatment with lambda phosphatase, suggesting that SAMS1 was a phosphorylation substrate of GNP3 (Fig. 4a). GNP3 also exhibited self-phosphorylation capacity in this assay (Supplementary Fig. 9a). Additionally, we found that SAMS1 phosphorylation levels were markedly decreased in gnp3-1 relative to WT (Fig. 4b). To identify the residue(s) being phosphorylated by GNP3, we performed a liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis of recombinant SAMS1 incubated with GNP3. We identified two possible sites, Thr-137 and Ser-295 (Supplementary Fig. 9b). We also predicted Ser-103 as a phosphorylation site based on the KinasePhos website (http://kinasephos.mbc.nctu.edu.tw). We replaced each predicted phosphorylation site with alanine to generate single, double, and triple mutant variants of SAMS1. An in vitro kinase assay with each recombinant SAMS1 variant and GST-GNP3 (366-655) in the presence of radioactive ATP showed a marked decrease in phosphorylation when Ser-103 and Ser-295 were replaced with Ala (Supplementary Fig. 9c).

Fig. 4: GNP3 phosphorylates SAMS1 to regulate grain number.

a GNP3 phosphorylates SAMS1 in vitro. SAMS1 phosphorylation was detected by mobility shifts on a Phos-tag SDS-PAGE gel. b GNP3 phosphorylates SAMS1 in rice protoplasts. c–e Analysis of the stability of SAMS1 in cell-free assays. The relative abundance of SAMS1 was calculated by ImageJ software, with the 0-min sample set to 1. The experiments (a–e) were performed with at least three independent replicates. f–j Representative panicle (f) and statistical results for Panicle length (g), primary branch number (h), secondary branch number (i), and grain number per panicle (j) for WT, gnp3-1, sams1, gnp3/GNP3, and sams1 gnp3/GNP3 plants. Scale bar, 2 cm (f); (n = 27/29/29/28/28 plants). In each boxplot, the center line indicates the median, the box edges represent the first and third quartiles, and the whiskers extend to the minima and maxima. Lowercase letters denote significant differences (one-way ANOVA with Tukey multiple-range test, P < 0.05). Source data are provided as a Source Data file.

Considering that GNP3 interacted with and phosphorylated SAMS1, we wondered whether GNP3 might affect SAMS1 stability. We examined the degradation rate of SAMS1 in a cell-free protein degradation assay with protein extracts from WT (Nip), gnp3-1, and GNP3-OE panicles. We added recombinant SAMS1-His to each extract and followed its abundance over time by immunoblot analysis with an anti-His antibody. SAMS1-His was gradually degraded in the WT protein extracts. The degradation of SAMS1-His was markedly slower when added to gnp3-1 protein extracts but accelerated when added to GNP3-OE protein extracts (Fig. 4c–e). These data suggest that GNP3 phosphorylates SAMS1 and promotes its degradation in rice.

To explore the role of SAMS1 in regulating GNP in rice, we generated SAMS1 knock-out transgenic plants named sams1 via CRISPR/Cas9-mediated gene editing in the Nip background (Supplementary Fig. 10a). sams1 plants exhibited large panicles and higher GNP compared to WT (Fig. 4f–j). To explore the genetic interaction between GNP3 and SAMS1, we generated sams1 gnp3/GNP3 plants that are homozygous for an SAMS1 mutation and heterozygous for a gnp3 mutant allele via CRISPR/Cas9-mediated gene editing in the Nip background. Our analyses revealed that sams1 gnp3/GNP3 plants exhibited significantly longer panicles, increased number of secondary branches, and higher GNP compared to gnp3/GNP3 plants, yet showed similar secondary branching and GNP relative to the sams1 single mutant (Fig. 4f–j, Supplementary Fig. 10b). Furthermore, the sams1 gnp3/GNP3 plants displayed higher SPAD values than gnp3/GNP3 plants, reaching levels comparable to those of sams1 plants (Supplementary Fig. 10c–e). Collectively, these findings demonstrate that SAMS1 acts as a negative regulator of GNP and functions at least partially within the same genetic pathway as GNP3.

GNP3–SAMS1 module participates in ethylene biosynthesis

The SAMS1 encoded SAM synthase plays a pivotal role in SAM biosynthesis, where SAM serves as a methyl (Me) donor and critical precursor for ethylene (ET) and polyamines (PAs) production (Fig. 5a). Comparative analysis of panicles from WT, gnp3-1 mutants, and GNP3-OE plants revealed that gnp3-1 exhibited significantly elevated SAM synthase activity, increased SAMS content, and enhanced SAM accumulation (Fig. 5b-d). Consistent with SAM dynamics, ethylene emission was higher in panicles of gnp3-1 but reduced in GNP3-OE compared to WT (Fig. 5e; Supplementary Fig. 11a). Quantitative analysis of major PAs (putrescine, spermidine, spermine) showed no systematically concentration differences across genotypes (Supplementary Fig. 11b–d). Phenotypic analyses demonstrated that treatment with ethephon (ETH), an ethylene-releasing agent, significantly inhibited GNP during the panicle development stage. Notably, GNP3-OE and SAMS1-CR plants showed heightened sensitivity to ethephon compared to the WT. In contrast, treatment with polyamines did not significantly impact GNP (Fig. 5f). Whole-genome bisulfite sequencing revealed comparable DNA methylation levels between WT and gnp3-1 panicles (Fig. 5g, h). Furthermore, the sams1 single mutants and sams1 gnp3/GNP3 plants double mutants displayed similarly reduced SAM synthase activity, SAMS content, SAM accumulation, and ethylene production (Fig. 5b–e). These findings collectively establish that the GNP3-SAMS1 module regulates grain number mainly through SAM-mediated ethylene biosynthesis.

Fig. 5: GNP3–SAMS1-mediated ethylene synthesis regulates the expression of GNP-related genes.

a Schematic of SAMS1 function as enzyme in SAM biosynthesis, which serves as a methyl donor and precursor for ethylene and polyamines. b–e SAMS activity (b), SAMS content (c), SAM accumulation (d), and ethylene emission (e) in young panicles of WT, gnp3-1, sams1, and sams1 gnp3/GNP3 plants. Data represent means ± SD (n = 3 biological replicates). f GNP in WT, gnp3-1, GNP3-overexpression (OE1, OE6), and SAMS1-knockout (CR1, CR2) plants treated with ethephon (ETH) or polyamines (PAs). Data represent means ± SD (n = 12/16/18 plants). g Global DNA methylation levels in WT and gnp3-1 panicles via whole-genome bisulfite sequencing. Data represent means ± SD (n = 3 biological replicates). h DNA methylation profiles (mCG, mCHG, mCHH) across gene regions (TSS to TES) in WT and gnp3-1. Transcription start site (TSS) and transcription end site (TES) are indicated. i Volcano plot of differentially expressed genes in gnp3-1 vs. WT panicles. j Heatmap showing the expression of ethylene-related genes in WT and gnp3-1. k Heatmap showing the expression of GNP-related genes in WT and gnp3-1. In (b–g), asterisks indicate statistical significance by two-tailed Student’s t tests (*P < 0.05, **P < 0.01). Source data are provided as a Source Data file.

To delineate GNP3-mediated transcriptional regulation, we performed RNA-seq analysis of young panicles from WT and gnp3-1 mutants. Differential expression analysis (P value < 0.05) identified 2381 downregulated and 2916 upregulated genes in gnp3-1 (Fig. 5i; Supplementary Data 2). GO enrichment analysis revealed significant overrepresentation of biological processes including “floral organ development” (GO:0009908), “reproductive structure development” (GO:0048608) (Supplementary Fig. 12). Notably, the ethylene biosynthetic pathway showed coordinated transcriptional changes, with four AMINOCYCLOPROPANE-1-CARBOXYLATE OXIDASE genes (OsACO1, OsACO2, OsACO3, OsACO7) exhibiting upregulation, accompanied by differential expression of 30 ethylene-responsive factors (Fig. 5j). Furthermore, we observed regulation of key grain number regulators: eight characterized negative regulators (including OsLAC, OsERF48) were upregulated, while 33 characterized positive regulators (including RGN1, APO2, OsSNB) were downregulated in gnp3-1 (Fig. 5k). These transcriptional signatures indicate that GNP3 might modulate panicle architecture by coordinately influencing ethylene biosynthesis capacity and multiple grain number regulators.

Natural variation in GNP3 modulates GNP

We performed a haplotype analysis for GNP3 using a panel of 3142 rice accessions, comprising 142 Or-I, 105 Or-II, 145 Or-III, 1715 indica, 283 temperate japonica, 479 tropical japonica, 199 aus, and 74 aromatic accessions. We identified only one non-synonymous SNP (S28274865, G-to-T, Ala-to-Glu) in GNP3 coding sequence (CDS) (Table 1, Supplementary Data 3). Based on this SNP, GNP3 could be divided into two haplotypes, GNP3^Hap-G and GNP3^Hap-T. We observed that accessions with the GNP3^Hap-T haplotype had higher GNP values than did those with GNP3^Hap-G among aromatic accessions (Fig. 6a). GNP3^Hap-G was mainly present in the japonica subpopulation, whereas GNP3^Hap-T was mainly present in the indica subpopulation. Phylogenetic analysis and haplotype network analysis based on 12 common SNPs between 2696 cultivars and 446 wild rice accessions showed that GNP3^Hap-G and GNP3^Hap-T were present in both cultivated and wild rice accessions, suggesting that the differentiation of GNP3 into its two haplotypes originated in wild rice, with GNP3^Hap-T spreading into indica, while GNP3^Hap-G was restricted to temperate japonica (Fig. 6b, c, Supplementary Table 3, Supplementary Data 3). However, we saw no evidence for selection on GNP3 (Supplementary Fig. 13; Supplementary Table 4).

Table 1 Haplotype analysis of GNP3

Fig. 6: Natural variation in GNP3 modulates grain number.

a Comparison of GNP between accessions carrying GNP3^Hap-G and GNP3^Hap-T in aromatic rice (n = 41/48/38/48 plants). HN-R1: accessions cultivated in Sanya, Hainan province (replicated 1);  HN-R2: accessions cultivated in Sanya, Hainan province (replicated 2). b, c Phylogenetic tree (b) and minimum spanning network (c) of GNP3 haplotypes across eight rice subgroups: Or-I, Or-II, Or-III, indica (ind), temperate japonica (tej), tropical japonica (trj), aus, and aromatic (aro). d, e Representative panicles (d) and GNP (e) of WT, gnp3-1, GNP3^Hap-G-CM, and GNP3^Hap-T-CM plants grown in Nanning (n = 19/19/16/15 plants). f, g Representative panicles (f) and grain number per panicle (g) of WT, gnp3-1, GNP3^Hap-G-OE, and GNP3^Hap-T-OE plants grown in Beijing (n = 20/15/17/20 plants). h Frequency of GNP3 alleles in improved varieties (IMP) and landraces (LAN). i, j Representative panicles (i) and GNP (j) of NIL-GNP3^Nip and NIL-GNP3⁹³¹¹ (n = 30 plants). Scale bar, 2 cm (d, f, i). In (a, j), asterisks indicate statistical significance by two-tailed Student’s t tests (*P < 0.05, **P < 0.01). In (a, e, g, j), the center line indicates the median, the box edges represent the first and third quartiles, and the whiskers extend to the minima and maxima. Lowercase letters denote significant differences (one-way ANOVA with Tukey multiple-range test, P < 0.05). Source data are provided as a Source Data file.

To investigate the differential effects of GNP3^Hap-G and GNP3^Hap-T on GNP, we generated two types of transgenic plants: (1) genomic complementation lines (GNP3^Hap-G-CM and GNP3^Hap-T-CM) in the gnp3-1 mutant background, and (2) overexpression lines (GNP3^Hap-G-OE and GNP3^Hap-T-OE) in the Nip background. Phenotypic analysis showed that GNP was significantly higher in both GNP3^Hap-T-CM compared to GNP3^Hap-G-CM plants and in GNP3^Hap-T-OE compared to GNP3^Hap-G-OE plants (Fig. 6d–g). Furthermore, analysis of allele frequencies revealed that the GNP3^Hap-T allele was higher in improved varieties (IMPs) than in landraces (LANs) among aromatic accessions (Fig. 6h). To functionally characterize this potentially elite haplotype associated with higher GNP, we developed near-isogenic lines (NILs): NIL-GNP3^Nip (carrying GNP3^Hap-G) and NIL-GNP3⁹³¹¹ (carrying GNP3^Hap-T) (Supplementary Fig. 14a–c). Comparative analysis demonstrated that NIL-GNP3⁹³¹¹ exhibited longer panicles, increased secondary branch number, and higher GNP relative to NIL-GNP3^Nip (Fig. 6i, j; Supplementary Fig. 14d–f). Collectively, these findings demonstrate that natural allelic variation in GNP3 influences grain number in rice.

SNP S28274865 mapped to the region encoding the interaction domain (aa 143-230) between GNP3 and SAMS1. To determine whether this SNP affected the interaction between GNP3 and SAMS1, we conducted an MbY2H assay using the GNP3^Hap-G or GNP3^Hap-T CDS as bait. We determined that the interaction between GNP3^Hap-T and SAMS1 was stronger than that between GNP3^Hap-G and SAMS1 (Fig. 7a). We also detected a stronger interaction when conducting a GST pull-down assay between GNP3^Hap-T and GST-SAMS1 (Fig. 7b). Furthermore, we observed a higher phosphorylation level for recombinant SAMS1 when co-incubated with recombinant GNP3^Hap-T compared to GNP3^Hap-G (Fig. 7c). In addition, the cell-free protein degradation assay demonstrated that the degradation of SAMS1-His occurred at a faster rate when incubated with GNP3^Hap-T protein extracts (Fig. 7d). We further discovered that the SAM synthetase activity, the SAMS and SAM contents, as well as ethylene production in the panicles of GNP3^Hap-G-OE plants, were significantly higher than those in GNP3^Hap-T-OE plants (Fig. 7e–h). These findings suggest that the natural variation in GNP3, in the form of the SNP S28274865, affects the interaction between the encoded GNP3 protein and its phosphorylation of SAMS1. Consequently, this has an impact on ethylene biosynthesis and panicle development.

Fig. 7: Natural variation in GNP3 enhances its interaction with SMAS1.

a, b Stronger interaction between GNP3^Hap-T and SAMS1 compared to GNP3^Hap-G in MbY2H assay (a) and GST pull-down assays (b). c GNP3^Hap-T exhibits higher phosphorylation activity toward SAMS1 in vitro. d Accelerated degradation of SAMS1 in cell-free assays with GNP3^Hap-T extracts. SAMS1 abundance (normalized to 0 min) was quantified using ImageJ software. e–h SAMS activity (e), SAMS content (f), SAM accumulation (g), and ethylene emission (h) in young panicles of GNP3^Hap-G-OE and GNP3^Hap-T-OE plants. Data represent means ± SD (n = 3 biological replicates). Asterisks indicate statistical significance by two-tailed Student’s t tests (*P < 0.05, **P < 0.01). i A predicted working model of GNP3 in regulating grain number in rice. GNP3^Hap-T enhanced SAMS1 interaction and phosphorylation suppress ethylene biosynthesis, thereby increasing grain number. Source data are provided as a Source Data file.

Overexpression of GNP3 simultaneously increases grain yield in field

To evaluate the potential value of GNP3 for breeding, we investigated other important agronomic traits in addition to GNP. We established that overexpression of GNP3 simultaneously increased grain length and thousand-grain weight without affecting panicle number, tiller number, plant height or leaf morphology (Fig. 8a–i). Field trials showed that GNP3-OE1 and GNP3-OE6 plants showed significantly improved grain yield per plant compared to WT, with increases of 18.4% and 17.6%, respectively. The plot yield of GNP3-OE1 and GNP3-OE6 plants increased by 23.7% and 22.4% over that of WT plants under paddy conditions (Fig. 8j–l). We determined whether this increased grain yield came at the cost of grain quality by checking the chalky grain percentage and chalkiness degree of grains. The chalkiness degree, chalky grain percentage and chalkiness area of GNP3-OE1 and GNP3-OE6 grains were significantly lower compared to those of the WT (Fig. 8m–p). The head rice rate, gel consistency, gelatinization temperature, and amylose content were also comparable between the GNP3-OE lines and WT (Supplementary Fig. 15). These results demonstrate that GNP3 overexpressing improves grain yield and grain quality.

Fig. 8: GNP3 overexpression enhances both grain yield and quality.

a, b Morphology of grains for WT and GNP3-OE plants. Scale bar, 1 cm. c–i Statical results for grain length (c), grain width (d), thousand grain weight (e), panicle number (f), plant height (g), flag leaf length (h), and flag leaf width (i) in WT and GNP3-OE plants (n = 10 plants). j Field performance of WT, gnp3-1, and GNP3-OE lines in a paddy field. k, l Plant yield (k) and plot yield (l) of WT, gnp3-1, and GNP3-OE lines (n = 23/22/22/25 plants in (k); n = 3 biological replicates in (l). m Milled rice from WT and GNP3-OE lines. Scale bar, 2 cm. n–p Chalkiness degree (n), percentage of chalky grains (o) and chalkiness aera (p) of WT and GNP3-OE plants (n = 7 plants). In each boxplot, the center line indicates the median, the box edges represent the first and third quartiles, and the whiskers extend to the minima and maxima. Lowercase letters denote significant differences (one-way ANOVA with Tukey multiple-range test, P < 0.05). Source data are provided as a Source Data file.

MAPK cascades also play an important role in response to environmental stresses²⁷. We also conducted an MbY2H assay to look for interaction between GNP3 and six OsMKKs: GNP3 interacted with OsMKK1, OsMKK6, and OsMKK10.2 in vitro (Supplementary Fig. 16a). We confirmed these results with a BiFC assay in N. benthamiana leaf cells (Supplementary Fig. 16b). Notably, OsMKK6 has been previously implicated in cold stress tolerance¹², and we observed significant sequence differentiation in GNP3 between indica and japonica subspecies (Table 1), which typically exhibit differential cold adaptation. Functional analyses demonstrated that GNP3 positively regulates cold tolerance: under cold stress conditions, GNP3-overexpression lines (GNP3-OE1 and GNP3-OE6) showed higher survival rates compared to WT plants, while gnp3-1 mutants exhibited lower survival rates (Supplementary Fig. 17a, b). To further elucidate this pathway, we generated MKK6^DD (a constitutively active form of OsMKK6) overexpression lines in the gnp3-1 background. These plants showed restoration of cold tolerance, with survival rates comparable to WT (Supplementary Fig. 18), demonstrating that GNP3 and OsMKK6 function in the same pathway for cold stress response.

Discussion

In this study, we identified GNP3 as a pivotal regulator governing grain number determination in rice, unveiling molecular mechanisms underlying yield enhancement. Mechanistic investigations revealed that GNP3 encodes the mitogen-activated protein kinase kinase kinase OsMKKK22, which directly phosphorylates SAMS1, triggering its degradation. This post-translational regulation effectively modulates endogenous ethylene biosynthesis, subsequently orchestrating the transcriptional network of panicle development-related genes to optimize inflorescence architecture. Notably, we discovered a functional natural allelic variation (S28274865, G-to-T, Ala-to-Glu) within the coding sequence of GNP3 that significantly enhances its protein interaction affinity with SAMS1. This strengthened molecular association results in suppressed ethylene production and consequently elevated GNP (Fig. 7i).

The MAPKKK family serves as crucial signaling nodes coordinating plant growth, developmental programs, and environmental adaptation^11,27. In this study, we established the essential biological function of GNP3 through comprehensive genetic analyses. Targeted knockout via CRISPR/Cas9-mediated genome editing failed to yield homozygous frameshift mutants, with genotyping-by-sequencing revealing lethality at the seedlings stage (Supplementary Fig. 4b). Notably, surviving heterozygous mutants (gnp3/GNP3) exhibited small panicles and spotted leaves (Fig. 1c–g; Supplementary Fig. 4e). This lethal phenotype mirrors that of the oscsl1 mutant harboring a T-DNA insertion in the seventh intron and exhibiting a chlorotic seedling phenotype before died²⁶. Interestingly, we isolated a viable gnp3-1 mutant carrying a 983 bp transposon insertion in the 5′ UTR of GNP3. RT-qPCR confirmed a significant reduction in GNP3 transcript levels (Fig. 2h). The survival of gnp3-1 homozygotes further demonstrates GNP3 as an indispensable regulator of rice viability.

The role of MAPK cascades in plant growth and stress responses is well-documented, but their direct involvement in ethylene biosynthesis and grain yield regulation has remained unclear. A high concentration of ethylene induces spikelet degradation or seed abortion in maize and rice^28,29. Lower ZmACO2 expression reduces ethylene emissions but increases the number of fertile florets and kernels, thus enhancing grain yield in maize²¹. These results show that ethylene may negatively regulate GNP formation. Here, we establish that GNP3 phosphorylates SAMS1, a rate-limiting enzyme in SAM synthesis, to accelerate its degradation. This phosphorylation-dependent destabilization reduces SAM availability, thereby limiting ethylene production (Figs. 3 and 4). Ethylene, a phytohormone known to suppress panicle branching and grain formation, was significantly elevated in gnp3-1 mutants but reduced in GNP3-overexpression plants (Fig. 5e). The genetic interaction between GNP3 and SAMS1 further supports this mechanism: sams1 mutants phenocopy GNP3-overexpression phenotypes, while sams1 gnp3/GNP3 double mutants exhibit similar traits as sams1 mutants (Fig. 4f–j). These findings reveal a regulatory pathway linking MAPK signaling to ethylene-mediated panicle development. RNA-seq analysis identified 41 grain number regulators that might be regulated by GNP3 (Fig. 5k). However, the precise molecular mechanisms by which ethylene regulates panicle branching require further elucidation.

The identification of SNP S28274865 as a functional allele underlying GNP3 haplotype differentiation provides a compelling example of how natural variation shapes agronomic traits. The GNP3^Hap-T allele, prevalent in indica subspecies, enhances GNP3-SAMS1 interaction and phosphorylation efficiency, leading to lower ethylene levels and higher grain number compared to GNP3^Hap-G allele (common in japonica) (Figs. 6, 7). This allelic divergence likely reflects adaptation to distinct ecological niches, with indica varieties prioritizing yield potential under favorable conditions, while japonica accessions may retain alleles balancing yield with stress tolerance. Notably, the increased frequency of GNP3^Hap-T in improved varieties of aro underscores its breeding value (Fig. 6i). The development of NILs and haplotype-specific overexpression lines confirmed that GNP3^Hap-T significantly outperforms GNP3^Hap-G in grain number and yield, offering a promising target for marker-assisted selection.

Beyond its role in panicle development, GNP3 overexpression improved cold tolerance via interaction with OsMKK6, a key component of the MAPK cascade in stress signaling (Supplementary Fig. 17). It is worthy of further study to explore whether stronger cold tolerance observed in GNP3^Hap-G-OE lines, implying that allelic variation in GNP3 may reflect trade-offs between yield and cold stress adaptation. While GNP3^Hap-T maximizes grain production, GNP3^Hap-G appears to favor stress resilience, possibly explaining its prevalence in japonica varieties adapted to colder climates. This pleiotropic effect highlights the versatility of GNP3 in addressing both yield and environmental challenges—a critical advantage in the context of climate change. Moreover, the concurrent improvement in grain quality (reduced chalkiness) in GNP3-overexpression lines demonstrates that yield gains need not compromise nutritional or market value, further enhancing its applicability in breeding programs. However, field trials across diverse environments will be essential to validate the stability of GNP3-mediated yield improvements under varying climatic conditions.

Currently, it has been confirmed that the MKKK10-MKK4-MAPK6 pathway plays an important role in regulating GNP in rice^7,14,30. In this study, we not only elucidated the mechanism of the GNP3-SAMS1 molecular module in determining GNP in rice but also discovered the physical interaction between GNP3 and MKK1, MKK6, and MKK10.2. It is worth noting that after single-gene knockout of MKK1, MKK6, and MKK10.2, there was no statistically significant change in GNP of rice (Supplementary Fig. 19). This result points to the possible functional redundancy of these MAPKK genes in regulating GNP. In view of this, subsequent research needs to deeply explore the specific functions of these genes interacting with GNP3 and their related pathways in regulating GNP or stress responses.

Methods

Plant materials and growth conditions

A panel of 317 Oryza sativa accessions was cultivated in Nanning, Guangxi, China, in 2018. For each accession, five mid-row plants were randomly selected to measure the GNP. Phenotypic data for all materials used in the GWAS are provided in Supplementary Data 1. The gnp3-1 mutant was isolated from tissue culture-derived populations. Plants were grown under natural paddy field conditions in Beijing, Sanya or Naning city. Phenotypic evaluations were performed using T3 or later generations of transgenic lines. For yield assessment, WT, gnp3-1, and GNP3-OE seedlings were transplanted into 2 × 2 m plots, with each plot containing 10 rows of 21 seedlings per row. Three biological replicates were analyzed per sample.

GWAS and QTL gene annotation

Sequencing data were obtained from the 3000 Rice Genomes Project (3K-RG). Genome-wide association analysis was performed using the compressed mixed linear model (CMLM) in the GAPIT package in an R environment. Based on 1,000 permutation tests, a significance threshold of −log10(P) > 4.5 was applied at the genome-wide level. A QTL was defined as a genomic region containing at least three significant SNPs clustered within a 170-kb window. All detected QTLs are listed in Supplementary Table 1. Candidate genes within the qGNP3-3 locus were annotated and provided in Supplementary Table 2.

Plasmid construction and plant transformation

To construct the CRISPR/Cas9 vector, two 20 bp PAM sequences from the CDS of target genes were selected for specific recognition and cloned into the vector pHUE411-2gR³¹. To construct Pro35S:Flag-GNP3 overexpression vectors, full-length GNP3 CDS without stop codons were amplified by PCR and cloned into the SpeI and KpnI sites of binary plant expression vector pCM1307. To produce the complementation construct pMDC163-GNP3, the full-length genomic sequence of GNP3 was amplified from Nip and 9311, respectively, including 3121 bp upstream and 859 bp downstream, and cloned into the plant binary vector pMDC163. To construct the ProGNP3:GUS vector, a genomic fragment of the GNP3 promoter region starting at 3.0 kb upstream of the ATG initiation codon was amplified and cloned into the PmeI and AscI sites of the binary plant expression vector pMDC162. For subcellular localization, GNP3 CDS without stop codons were amplified by PCR and cloned into the KpnI and HindIII sites of the binary plant expression vector pSuper1300-GFP. To construct the ProGNP3:GNP3-GFP and the ProSMAS1: SMAS1-GFP vectors, the native promoter and CDS without stop codons were amplified by PCR and cloned into binary plant expression vector pCambia-1300-GFP. The DNA constructs used in this study were generated using Tks Gflex™ DNA Polymerase (R060Q, Takara, Japan). All constructs were confirmed by sequencing. The vectors were introduced into Agrobacterium tumefaciens strain EHA105 and subsequently transformed into rice calli from mature embryos of rice cultivars by Agrobacterium-mediated transformation³².

RNA extraction and RT-qPCR

Total RNA was extracted from fresh tissues using the RNApure Total RNA Kit (RN0302, Aidlab, China). First-strand cDNA was synthesized from 1 μg of total RNA using HiScript II Reverse Transcriptase (R211, Vazyme, China) according to the manufacturer’s protocol. Quantitative real-time PCR (RT-qPCR) was performed on an ABI 7500 Real-Time PCR System (Applied Biosystems, USA) using SYBR® Premix Ex Taq II (RR820A, TaKaRa, Japan). Each reaction was performed in triplicate. The rice ubiquitin gene (LOC_Os03g13170) was used as an internal reference for normalization.

GUS staining and subcellular localization of GNP3

The solution buffer for GUS staining contained 50 mM Na₂HPO4, 10 mM Na₂EDTA, 0.5 mM K₃Fe (CN)₆, 0.5 mM K₄Fe (CN)₆, 0.1% Trition X-100, and 1 mg/mL 1,5-bromo-4-chloro-3-indolyl β-D-glucuronic acid. Tissues from ProGNP3:GUS transgenic plants were collected and submerged in GUS-staining solution for 12 h at 37 °C in the dark. The chlorophyll was then removed with 95% alcohol.

For subcellular localization analysis, Pro35S:GNP3-GFP and Pro35S:SAMS1-GFP constructs were transiently expressed in rice protoplasts. The following marker proteins were co-expressed as controls: (1) GSN1-mCherry (cell periphery and cytoplasm marker³⁰) and (2) RGN1-mCherry (nuclear marker⁸). Fluorescence signals were detected using an LSM 880 confocal laser scanning microscope (Zeiss, Germany) with the following excitation/emission settings: GFP (488/500–550 nm) and mCherry (587/600–650 nm).

The subcellular localization of GNP3 and SAMS1 was analyzed in transgenic plants expressing ProGNP3:GNP3-GFP and ProSAMS1:SAMS1-GFP constructs, respectively. Root tissues were imaged using two complementary microscopy systems: (1) Confocal microscopy: LSM 880 system (Zeiss, Germany) with GFP excitation/emission at 488/500–550 nm; (2) Light-sheet microscopy: Zeiss Lightsheet 7 for improved 3D reconstruction of fluorescence patterns.

Split-ubiquitin membrane yeast two-hybrid assays

MbY2H assays were performed following the manufacturer’s instructions (Oebiotech, Shanghai, China). Yeast two-hybrid assays were conducted using the NMY51 system with the DUAL membrane starter kit (Dualsystems Biotech AG, Schlieren, Switzerland). DNA fragments containing GNP3 and SAMS1 genes were cloned into pBT3-STE (bait) and pPR3-N (prey) vectors, respectively. Yeast cells were cultured on SD/-Trp-Leu or SD/-Trp-Leu-His-Ade medium at 30 °C in darkness for 3–5 days. The optical density (OD) of yeast suspensions was normalized prior to interaction intensity assessment.

Bimolecular fluorescence complementation (BiFC) assays

For BiFC assays, the full-length CDS of related genes were cloned into the binary vectors pSPYCE(M) and pSPYNE173, respectively. These constructs were co-transformed into Nicotiana benthamiana mesophyll cells via Agrobacterium-mediated transformation using acetosyringone. After infiltration, plants were kept in the dark for 48 h before imaging. Fluorescence signals were detected using an LSM 880 confocal microscope (Zeiss, Germany). GSK3 combined with GL3.1 was used as positive controls³³.

Co-immunoprecipitation assays

GNP3-HA and SAMS1-Myc constructs were transformed into Nip protoplasts for transient expression. Total protein was isolated with extraction buffer (150 mM NaCl, 50 mM Tris-HCl, 10 mM MgCl₂, 1% NP-40, 1× complete protease inhibitor cocktail, and 3 mM Dithiothreitol) and incubated with HA-beads for 1 h at 4 °C. After incubation, the beads were washed five times, and then SDS loading buffer was added. The samples were denatured at 98 °C for 10 min and finally subjected to SDS-PAGE analysis. We used anti-HA (Sigma, H3663) and anti-Myc (Sigma, M4439) at a 1:5000 dilution to detect the input and immunoprecipitates, respectively, followed by secondary goat antimouse IgG (light chain specific) (Easybio, BE0105). Images were scanned using the Chemiluminescence detector Fusion X7 (Vilber Lourmat, France).

GST pull-down assays

The CDS of GNP3 was cloned into the pGEX-4T-2 vector. The CDS of SAMS1 was cloned into the pET-28a vector. Proteins were expressed and purified in Escherichia coli strain BL21. GST-GNP3 and SAMS1-His protein were incubated with GST-beads for 2 h at 4 °C. After incubation, the beads were washed five times using PBS buffer, and then SDS loading buffer was added. The samples were denatured at 98 °C for 10 min and finally subjected to SDS-PAGE analysis. Then, anti-GST (Abcam, ab92) and anti-His (Sigma, H1029) at 1:5000 dilution were used to detect the input and immunoprecipitates, respectively. Followed by secondary goat antimouse IgG (light chain specific) (Easybio, BE0105). Images were scanned using the Chemiluminescence detector Fusion X7 (Vilber Lourmat, France).

In vitro and in vivo phosphorylation assays

For in vitro phosphorylation assays, 4 µg recombinant proteins SAMS1-His and 6 µg GST-GNP3 were incubated in kinase reaction buffer (20 mM Tris-HCl pH 7.5, 20 mM MgCl₂, 1 mM DTT, 50 mM ATP) at 30 °C for 30 min. Proteins were separated on a Phos-tag PAGE gel containing 50 µM Phos-tag and 0.1 mM MnCl₂ Acrylamide. The gel was washed three times with washing buffer (38.63 mM Glycine, 47.88 mM Tris, 20% Methanol, and 10 mM EDTA), followed by western blot analysis using anti-His and anti-GST antibodies. For in vivo phosphorylation assays, SAMS1-Myc constructs were transiently expressed in rice protoplasts derived from WT (Nip) and gnp3-1 mutant plants. Total protein was extracted and subjected to western blot analysis using anti-Myc (1:5000, Sigma-Aldrich, M4439) and anti-phosphoserine/threonine antibodies (1:500, abmart, T91067S).

To detect the phosphorylation sites of SAMS1, phosphorylated proteins were cut from the SDS-PAGE and digested using trypsin (Sigma-Aldrich) at 37 °C overnight. LC-MS/MS was used to analyze the tryptic peptides. LC-MS/MS data were queried against the Oryza_Sativa_Uniprot database using the following search parameters: fixed modification, carbamidomethylation (C); variable modifications, oxidation (M), phosphorylation (ST), and phosphorylation (Y); enzyme, trypsin; peptide mass tolerance, 20 ppm; fragment mass tolerance, 0.02 Da; and maximum missed cleavages, 2. For in vitro phosphorylation assays of mutant phosphorylation sites, 1 µg recombinant proteins GST-GNP3 (366-655) and 10 µg SAMS1-His were incubated in kinase reaction buffer (20 mM Tris-HCl pH 7.5, 20 mM MgCl2, 1 mM DTT, 50 mM ATP) at 30 °C for 30 min with or without 1µCi [γ-³²P] ATP. The proteins were separated by SDS-PAGE and visualized by autoradiography.

Cell-free degradation assays

Total proteins were extracted from young panicles with native buffer (50 mM Tris-MES pH 8.0, 10 mM EDTA pH 8.0, 0.5 M sucrose, 1 mM MgCl₂, and 5 mM DTT). Purified recombinant SAMS1-His was added to an equal amount of WT, gnp3-1, and GNP3-OE total proteins with 10 mM ATP and incubated at 25 °C for different time periods. SAMS1-His was separated on SDS-PAGE and detected with an anti-His antibody (1:5000, Sigma, H1029). The proteasome inhibitor MG132 was used at a concentration of 50 μM.

Endogenous SAMS, SAM, and ethylene measurement

Endogenous SAMS activity and SAM content were measured using the plant SAMS ELISA kit (Shanghai Shuangying Biotechnology Co., Ltd, SY-P09712) and the plant S-adenosyl Methionine (SAM) ELISA kit (Shanghai Shuangying Biotechnology Co., Ltd, SY-P02166) according to the suppliers’ protocols.

To measure endogenous ethylene production, we collected young panicles and placed them in a 10 mL container with 2 mL of double distilled water and sealed them with a rubber stopper. After incubation at 28 °C for 24 h in the dark, 1 mL of gas was withdrawn using a gas-tight syringe from the headspace and injected into a gas chromatograph (SHIMADZU GC-2010Plus, Japan). The ethylene production rate (microliters per gram of fresh weight per hour) was calculated on the basis of the initial fresh weight.

Endogenous polyamine measurement

Polyamine analysis was performed using an optimized HPLC-based method. Rice panicle samples (1 g) were liquid nitrogen-ground and extracted with 3 mL ice-cold 5% perchloric acid (1 h, 4 °C). After centrifugation (15,000 × g, 20 min, 4 °C), 0.5 mL supernatant was benzoylated with 1 mL 2 M NaOH and 10 μL benzoyl chloride (37 °C, 20 min). The reaction mixture was phase-separated using saturated NaCl and diethyl ether (1500 × g, 5 min, 4 °C). The ether phase was vacuum-dried, reconstituted in methanol, and filtered (0.2 μm). Chromatographic separation was achieved using an Agilent 1290 HPLC system with C18 column (150 × 4.6 mm, 5 μm) at 30 °C. The mobile phase (acetonitrile:water = 44:56, v/v) flowed at 0.7 mL/min, with detection at 230 nm.

RNA-seq analysis

RNA samples with three biological replicates were collected from Nip and gnp3-1 young panicles. Sample extraction and Illumina sequencing were performed by the Beijing Genome Institute (BGI, Wuhan). Clean reads were mapped to the rice genome (MSU-RGAP 7.0) using Bowtie2 (v2.2.5). Gene expression levels were calculated based on Fragments Per Kilobase of exon model per Million mapped fragments (FPKM).

Whole genome bisulfite sequencing (WGBS)

Young panicles from WT and gnp3-1 mutants (triplicates per genotype) were subjected to genomic DNA extraction. Libraries were prepared using the MGIEasy WGBS Kit (BGI) with 1% lambda phage spike-in controls. Fragmentation (Covaris LE220) was followed by end repair, adapter ligation, and bisulfite conversion (EZ DNA Methylation Gold Kit, D5020, Zymo Research, USA). Post-PCR amplification libraries were quality-controlled and processed through: (1) Single-strand circularization with linear DNA digestion; (2) Phi29-mediated rolling circle amplification generating DNA nanoballs (DNB, >300 copies); (3) DNB loading onto patterned nanoarrays. Sequencing was performed on BGI’s G400/T7/T10 platform (PE150). Methylation profiling (CG/CHG/CHH contexts) and DMR identification were conducted by BGI-Shenzhen’s bioinformatics pipeline.

Phylogenetic and haplotype network analysis

SNPs data of 2750 cultivated accessions and 392 wild rice accessions were obtained from Huang et al.³⁴ and the 3K-RG project³⁵. Information on cultivated rice accessions for haplotype analysis is given in Supplementary Data 3. MEGA 11.0 was used to conduct the neighbor-joining phylogenetic tree³⁶ and then visualized and annotated using iTOL³⁷. For evolutionary analysis, the GNP3 gene region and up or down 100 kb region were chosen to calculate Pi and Tajima’s D value using VCF tools.

Cold stress evaluation

The grains were imbibed on moist filter paper at 37 °C for 2 days. The seedlings were replant to PCR plates in normal Hoagland solution (pH 5.8) in a greenhouse under 14 h light and 10 h dark and 28 °C/25 °C with a relative humidity of 60%. The medium was replaced every two days to keep fresh. Two weeks old seedlings were used for treatments. For cold treatment, seedlings were transferred into a 4 °C low-temperature incubator for 4 days, followed by a 7-day recovery under normal conditions. The survival rate of each line was surveyed after recovery.

Grain quality measurement

Grains harvested from each plot were dried. After being decorticated, the rice was polished using a grain polisher (Kett electric laboratory, PEARLEST TP-3000, Japan). Polished rice grain appearance, including chalkiness rate and chalkiness degree, were evaluated using a rice appearance quality analyzer (SC-E, Hangzhou Wanshen Test Technology Corporation, China). Gel consistency and amylose content were measured using a modification of American Association of Cereal Chemists Method and UV5 spectrophotometer (Mettler-Toledo). Gelatinization temperature was measured using differential scanning calorimetry (Mettler-Toledo).

Primers

The primers used in the study are listed in Supplementary Data 4.

Statistical analyses

Data are presented as mean ± standard deviation with error bars. A two-tailed Student’s t-test implemented in Microsoft Excel 2016 was used to compare differences between two groups, with P < 0.05 considered statistically significant. More than two experimental groups were analyzed by one-way ANOVA combined with Tukey multiple range test in IBM SPSS software (version 21.0).

Reporting summary

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