Gibberellins (GAs) are plant hormones with a tetracyclic diterpenoid structure involved in various important developmental processes, including those of floral organs and shoots. GAs are synthesized in a stepwise manner by multiple oxidases, and the conversion from direct precursors to bioactive GAs is mediated by GA 3-oxidases (GA3oxs). Bioactive GAs bind to the GA receptor GIBBERELLIN INSENSITIVE DWARF1 (GID1) and induce a downstream GA response1. The synthesized bioactive GAs are inactivated via oxidation by GA 2-oxidases (OsGA2oxs) to regulate their endogenous levels in the plant body2,3. One of the rice GA3ox, OsGA3ox2, undergoes negative expression feedback regulation depending on the endogenous level of bioactive GA, whereas OsGA2oxs undergo positive expression feedback regulation2,3,4,5. Recently, Takehara et al. revealed that OsGA2oxs gradually form multimer structures with increasing GA concentration, which elevates the enzymatic activities through allosteric mechanism6. In this way, rice has elaborate systems to regulate the endogenous levels of bioactive GAs.

Several studies on the relationship between anther development and GAs have been conducted to date. Aya et al. showed that GAs regulate the formation of exine, a component of the pollen wall, and programmed cell death in the tapetum cells of the inner wall of the anther, which are involved in anther development7. They also revealed that the direct activation of a cytochrome P450 hydroxylase CYP703A3 by GAMYB is key to exine formation7. In addition to such sporophytic functions in anthers, GAs are involved in gametophytic functions, such as pollen tube germination and elongation8; however, little is known about their precise roles in the gametophytic processes.

In addition, GAs promote shoot elongation in vegetative tissues. Since it is important to regulate plant height for proper growth, rice has sophisticated mechanisms of GA regulation, such as feedback regulation of gene expression of inactivating enzymes and allosteric GA regulation as described above. If these mechanisms do not work properly, and GA accumulates in excess, rice does not grow normally. One example is the “Bakanae disease” of rice, which is caused by the fungal plant pathogen Gibberella fujikuroi and is a serious agronomical problem throughout Asia9. This fungus has multiple enzyme clusters in its genome and synthesizes large amounts of GA3, one of the bioactive GAs, in the plant body10. GA3 is not structurally inactivated by GA2oxs in plants owing to its structural property of having a double bond between the C1 and C2 positions of the ent-kaurene skeleton11. Another abnormality caused by excessive GA signaling is the slender rice1 (slr1), a loss of function mutant of the SLR1 gene, which encodes the DELLA protein, a GA signal repressor in rice. This plant shows excessive shoot growth and withering, as if the GA response is constantly active12.

Despite the existence of such important systems regulating endogenous GA levels in plants, rice anthers contain more than 100 times higher concentrations of bioactive GAs than do leaves13, far exceeding the need for GA responses by the GID1-GA-DELLA system1. However, it is unclear why such excessive endogenous GA is present in anthers and what enzymes and downstream factors are responsible for such excessive amounts.

GA3ox and GA2ox, members of the 2-oxoglutarate (2OG)- and Fe (II)-dependent dioxygenase (2ODD) family, catalyze the oxidation of substrate GAs using 2OG and molecular oxygen as co-substrates and ferrous Fe(II) cofactors14,15. The crystal structure of OsGA2ox3 has been reported by Takehara et al., and the residues important for its activity have been clarified6. There are nine functional GA2oxs paralogs in rice, and their evolutionary origin has also been reported16. Conversely, the crystal structures of GA3oxs in rice have not been elucidated, and the functional differences and evolutionary origin of the two paralogs in rice are also unknown.

In this study, we report the function and evolution of OsGA3ox1, one of the two OsGA3oxs in rice, and address the synthesis of large amounts of bioactive GAs in anthers. From structural and enzymatic analyses, we found that OsGA3ox1 can generate GA7 as well as GA4 from the precursor GA9 at a higher GA7 production rate than that of OsGA3ox2. GA7 is structurally similar to GA3, is not inactivated by GA2oxs17, and has the highest affinity for the GID1 receptor, making it the most potent bioactive of all forms of GAs. In addition, we show that OsGA3ox1 is expressed only in the later stage of pollen and lodicule development and produces bioactive GAs in pollen, functioning in a gametophytic manner, such as pollen starch accumulation, pollen tube germination, and pollen tube elongation, to increase the reproductive ability. By analyzing GA3ox1 genes of Oryza plants, we found that the characteristics of expression and enzyme activity of OsGA3ox1 have evolved specifically in the genus Oryza. Based on these analyses, we suggest that the evolution of the enzymatic function and gene expression of Oryza GA3ox1 contribute to the reproductive advantage of Oryza plants.

Results and discussion

OsGA3ox1 synthesizes bioactive GA7

First, we determined the endogenous GA levels in rice (cultivar Nipponbare) anthers. The mature anthers in Nipponbare possessed high levels of GA4 (nearly 400 pmol g−1 fresh weight [FW]), which is a bioactive form of GA, whereas relatively less amounts of GA1 (approximately 10 pmol g−1 FW) were detected (Fig. 1a). GA4 and GA1 are both bioactive forms of GA, which lack and possess 13-OH, respectively (Supplementary Fig. 1a, b). These results corroborate previous findings, which indicate that bioactive GAs are produced mainly through the non-13-hydroxylation pathway in rice reproductive organs, whereas the 13-hydroxylation pathway occurs in vegetative organs; also that the abundance of bioactive GAs in anthers is higher than that in leaves18. In addition to GA4, a substantial amount of GA7 (another bioactive GA without 13-OH), was observed (approximately 300 pmol g−1 FW) (Fig. 1a and Supplementary Fig. 1b). Similar to GA3 produced by the “Bakanae” fungus G. fujikuroi, GA7 cannot be inactivated by GA2oxs as it possesses a double bond between the C1 and C2 positions of GA4. We believe that GA7 may have the highest efficiency among bioactive GAs in promoting shoot elongation (Supplementary Fig. 1c–e). Using yeast, the GID1-GA-DELLA interaction for all bioactive GAs was explored (Supplementary Fig. 1f). We deduced that GA7 showed the highest efficiency among all GA ligands, followed by GA4.

Fig. 1: Activity of rice GA3ox1 and GA3ox2 upon synthetic gibberellic acid, and their structural factors.
figure 1

a Endogenous GA levels in anthers of wild-type (cultivar Nipponbare) and osga3ox1 mutant. Error bars, s.d. n = 3 Student’s t test (**p < 0.05). b Gas chromatogram of the reaction mixture for OsGA3ox2 (left panel), OsGA3ox1 (middle panel), and the OsGA3ox1-F235Y mutant protein (right panel) as a substrate GA9. The positions of the products GA4 and GA7 are indicated. c Structures of OsGA3ox2 around the GA-binding site with GA9 and the co-substrates, Fe (II), and 2OG (see also Supplementary Figs. 4–6). Fe (II) is represented as an orange sphere and 2OG and GA9 are indicated in yellow and blue, respectively. The amino acids establishing interactions with Fe (II) (H227, D229, and H285), 2OG (Y212, R295, and S297), and GA9 (I119, S230, G231, F233, T234, and R337) are shown in orange, yellow, and green, respectively. d Structural comparison of the active sites of OsGA3ox2 and OsGA3ox1 (predicted structure). Important amino acids are shown with the same colors and numbers as in (c) (OsGA3ox2). The site with amino acid Y212 or F235 in the 2OG-interacting site is responsible for the substitution in OsGA3ox1 (shown in e). e Sequence alignment of rice GA oxidase enzymes, OsGA3oxs, OsGA20oxs, OsGA2oxs, and other 2ODD family proteins, ACCO (ACC oxidase), and ANS (anthocyanidin synthase). The black dot indicates F212 in the 2OG-interacting site of OsGA3ox2 as described for (c) and (d).

Next, we examined the causal enzyme for the accumulation of GA4 and GA7 in rice anthers. Rice harbors two GA3ox genes5, which encode GA3-oxidases that catalyze the final step of GA biosynthesis, i.e., the addition of a hydroxyl group at C3 of GA precursors (Supplementary Fig. 1a). OsGA3ox2 is expressed in various organs, such as shoots and roots5 (Supplementary Fig. 2a), and its null mutant alleles exhibit a severe dwarf phenotype19. OsGA3ox1 is expressed specifically in anthers (Supplementary Fig. 2b). OsGA3ox1 promoter-GUS analysis revealed expression in the late developmental stage of pollen, anther filaments, and lodicules, but not in vegetative organs (Supplementary Fig. 2c, d). To examine the role of OsGA3ox1 in planta, we generated osga3ox1 knockout mutant plants using the CRISPR/Cas9 system (Supplementary Fig. 3a–e) and determined endogenous GA levels in the anthers (Fig. 1a). Endogenous GA4 and GA7 levels were significantly decreased, whereas the abundance of precursor GA9 increased in osga3ox1. This suggests that OsGA3ox1 is the major enzyme catalyzing the synthesis of GA4 and GA7 in rice anthers. Next, we determined the GA biosynthesis activities of recombinant OsGA3ox1 and OsGA3ox2 for GA9 as a substrate in vitro. Whereas the main product of OsGA3ox2 was GA4 (Fig. 1b; OsGA3ox2), we found a remarkable peak of GA7 as well as GA4 when we analyzed OsGA3ox1 (Fig. 1b; OsGA3ox1). This result indicates OsGA3ox1 produces GA7 in a higher GA7 synthesis ratio than that of OsGA3ox2.

X-ray crystal analysis of OsGA3ox2

To understand why OsGA3ox1 produced higher levels of GA7 compared to OsGA3ox2, we performed X-ray crystallography analyses of these enzymes. As a result, we succeeded in obtaining the structure of the OsGA3ox2 complex with (co-) substrates; GA9 and 2-oxoglutarate (2OG) at 1.9 Å resolution (Fig. 1c and d, Supplementary Figs. 4, 5, and Supplementary Table 1). Fe (II) was eliminated for crystallization to prevent the enzyme reaction, as previously described6. The carboxyl group of GA9 established interaction with S230, G231, F233, T234, and R337 directly or indirectly through water, and its γ-lactone ring established interaction with the main chain of I119 (Fig. 1c, Supplementary Fig. 5a). These amino acids are conserved between OsGA3ox1 and OsGA3ox2, but not between OsGA2oxs, OsGA20oxs, or even AtGA3oxs (Supplementary Fig. 6), with the exception that I119 is replaced with Y in OsGA3ox1. These residues of OsGA2ox3, whose structures were resolved in our previous study6, were not associated with GA-binding (Supplementary Figs. 5b, c, 6) despite conservation between OsGA3ox2 and OsGA2ox3 (e.g., I119 in OsGA3ox2 for γ-lactone ring interaction and the corresponding residue, I98 in OsGA2ox3 for no interaction with GA). Moreover, upon comparing the structure of OsGA3ox2 with that of OsGA2ox3, we found that the GAs were inverted in the substrate-binding pocket. These data suggest that GA-related 2ODDs acquired various GA-binding features during enzyme evolution. Conversely, the residues H227, D229, and H285 of OsGA3ox2 established interaction with Fe (II); Y212, S297, and R295 established interaction with 2OG (Fig. 1c and Supplementary Fig. 6). These amino acids are highly conserved in almost all 2ODDs, except F235 in OsGA3ox1, which is a conserved Y residue in other 2ODDs (Fig. 1c–e and Supplementary Fig. 6)20,21. Hence, we speculated that this substituted residue could be responsible for GA7 hyperproduction. To confirm this, we produced the mutated OsGA3ox1 protein having the F235Y like other 2ODDs and measured its enzyme activity (Fig. 1b; F235Y-OsGA3ox1). The productivity of GA7 was lower in F235Y-OsGA3ox1 than that in OsGA3ox1. Taken together, these results indicate that the substitution of a highly conserved Y with F in OsGA3ox1 contributes to the hyperproductivity of GA7.

osga3ox1 mutant shows gametophytic male sterility

The osga3ox1 mutant did not exhibit any abnormal phenotype during the vegetative stages (Supplementary Fig. 7a, b). However, compared with the wild type, in the reproductive stages, the osga3ox1 mutant presented with non-elongated anther filaments in the flowering stage, abnormal anther dehiscence, and no spikelet opening; but it showed normal pistils (Supplementary Fig. 7c–f and Supplementary Videos 1 and 2). The osga3ox1 showed abnormal starch accumulation in mature pollen grains (Fig. 2a).

Fig. 2: Phenotypic analysis of the reproductive organs of osga3ox1 mutant rice.
figure 2

a, b Pollen viability test of Nipponbare, OsGA3ox1+/− (heterozygous), and osga3ox1 mutant performed by conducting staining with the I2-KI solution. Pollen grains stained black are deemed viable; those stained red are deemed sterile (Bars = 0.5 mm) (a). Percentage of viable vs sterile (n = 5) (b). c Pollen viability test of GA-treated osga3ox1 mutant performed by conducting I2-KI staining, same as that for panel (b) (n = 3). P-values by Dunnett’s multiple comparison test are indicated. d Seed setting rate of Nipponbare, OsGA3ox1+/−, osga3ox1, and GA4-treated osga3ox1 mutant plants (n = 3). e In vitro pollen tube germination of Nipponbare and osga3ox1 mutant (n = 15). f Reciprocal crossing test between Nipponbare and osga3ox1 mutant plants. Arrowheads indicate pollen tubes. Bars = 0.5 mm. g, h Endogenous quantities of starch (g) and sucrose (h) contents in anthers of Nipponbare and osga3ox1 mutant (n = 3). i, j Enzyme activity of acid invertase in extractions from Nipponbare and osga3ox1 mutant pollen in the early binucleate stage. Summary of the extracted fractions (i) and their invertase activities (j) (these experiments were repeated three times with similar results). k Western blot analysis of OsCIN3 in extracted fractions in (I) and total extract of anthers using an OsCIN3 specific antibody. CBB staining of total extract is also shown. In all experiments, error bars = s.d. For panels (b) and (d): one-way ANOVA with Tukey’s multiple comparisons test. Different letters denote significant differences (p < 0.001). For panels (e), (g), and (h): ***p < 0.001, **p < 0.01, and *p < 0.05 by two-tailed paired t-tests.

Similar phenotype was observed using two independent mutant alleles, osga3ox1-cr1/osga3ox1-cr1 and osga3ox1-cr7/osga3ox1-cr7 (Supplementary Fig. 3e). We mainly used osga3ox1-cr1/osga3ox1-cr1 for further analyses as osga3ox1. Upon I2-KI staining, 81% of the Nipponbare (wild-type) pollen grains were stained, whereas only 55% of the OsGA3ox1+/− (heterozygous) and 4% of the osga3ox1 mutant (homozygous) pollen grains were stained (Fig. 2a, b, Supplementary Fig. 3e). The abnormality of the osga3ox1 was restored by exogenous GA application; notably, GA7 recovered at a lower concentration than did GA4 (Fig. 2c). osga3ox1 showed a complete loss of seed setting ability (Fig. 2d). In addition, the pollen grains of osga3ox1 did not germinate or elongate a tube, whereas its pistil functioned normally (Fig. 2e and f, Supplementary Fig. 7g, h). These data suggest that osga3ox1 exhibits gametophytic male sterility. The mutant phenotypes were consistent with their tissue-specificities as shown in the promoter-GUS analysis (Supplementary Fig. 2c, d).

Fluorescein diacetate (FDA) and 4′,6-diamidino-2-phenylindole (DAPI) staining revealed that the osga3ox1 mutant pollen shows normal cell development (Supplementary Fig. 7i). The analysis of starch and sucrose in anthers revealed that starch synthesis, in particular the conversion step of glucose from sucrose that is mainly regulated by invertases22, was defective in osga3ox1 mutant pollen (Fig. 2g, h). Rice has nine cell wall invertases (CINs), eight neutral invertases (NINs), and two vacuolar invertases (VINs)23. We speculated that a cell wall invertase, OsCIN3, is likely one of the targets of bioactive GAs synthesized by OsGA3ox1, as its expression shows appreciable correlation with that of OsGA3ox1 in anthers (Supplementary Figs. 2b, 7j, 8)23,24. The robust induction of OsCIN3 was observed upon exogenous GA4 treatment in both wild-type and osga3ox1 mutant (Supplementary Fig. 7k). Despite the induction of OsCIN3 upon GA application, downregulation of OsCIN3 expression was not observed in the osga3ox1 mutant. To examine the detailed relationship between GAs produced by OsGA3ox1 and invertase activities in early maturated pollen, we analyzed the in situ invertase activity (Supplementary Fig. 7l) and enzyme activities of fractionated pollen proteins (Fig. 2i, j). These results indicate that the activities of invertase in fractions D and E, which are associated with the cell wall, were decreased in the osga3ox1 mutants. In addition, we performed a western blot analysis on the anther extract fractions with an OsCIN3 specific antibody (Fig. 2k). These results indicate that in Nipponbare extracts, only a small amount of OsCIN3 protein could be detected in fraction B, which was extracted with sodium phosphate buffer, whereas a large amount of OsCIN3 was detected in same fraction in osga3ox1 extracts. Conversely, a large amount of OsCIN3 was detected in Nipponbare fraction C and D, which was strongly bound to the cell wall. This result indicates that, at least, OsCIN3 is an invertase—properly retained in the cell wall in Nipponbare—with reduced activity due to incorrect retention in the cell wall in osga3ox1. Further research is needed to clarify the mechanism by which GA is involved in proper retaining of OsCIN3. Together, we concluded that a high abundance of GA4 and GA7 produced by OsGA3ox1 in pollen played an important role in gametophytic reproduction through starch synthesis, probably via the regulation of OsCIN3.

Previously, Aya et al. reported that various rice GA-related mutants, such as ent-copanyl diphosphate synthase (cps), gid1, and gamyb, showed defects in programmed cell death of tapetal cells and the formation of exine and Ubisch bodies, resulting in pollen disruption in a sporophytic manner7. We consider that GAs function in various developmental stages of anther and pollen, and the causal genes of the mutants previously reported7 were expressed from the early stage of anther development. As these mutants exhibit the disruption of pollen in the early anther developmental stage through their sporophytic defections, researchers could not analyze the gametophytic GA functions in male organ development in the later stage. In contrast, OsGA3ox1 is a good candidate for the analysis of gametophytic functions of GA at the later stage of anther development25 (Supplementary Fig. 2c).

Evolution of GA7 productivity of Oryza GA3ox1

As OsGA3ox1 can produce substantial amounts of GA7 compared to OsGA3ox2, we speculated that OsGA3ox1 gained a new function in male fertility through the substitution of Y to F in the 2OG-interacting site, that is, GA7 producing activity of OsGA3ox1. Therefore, we examined the mechanism and juncture of acquiring this substitution during enzyme evolution. The phylogenetic analysis of GA3ox orthologs derived from land plants (Supplementary Table 2) indicated that the rice GA3ox orthologous genes underwent duplication after Poaceae emerged (Supplementary Fig. 9a), one is the ancestor of OsGA3ox1 (hereafter referred to as “Poaceae GA3ox1 orthologs”), and the other is OsGA3ox2 (hereafter referred to as “Poaceae GA3ox2 orthologs”). Poaceae GA3ox1 orthologs have been lost in certain Poaceae species, such as barley and wheat, whereas Poaceae GA3ox2 orthologs exist in all Poaceae species examined (Supplementary Fig. 9a, b), as previously reported26. We investigated whether the substitution of Y to F in the 2OG-binding site occurred before the genus Oryza emerged; however, none of the Poaceae GA3oxs orthologs exhibited amino acid substitutions (Supplementary Fig. 9c) except for rice. Therefore, we focused on the GA3ox1 orthologs in the genus Oryza.

We next combined public whole-genome data (Supplementary Table 2) and direct sequencing data of GA3ox1s obtained herein using genomic DNA of 19 Oryza species obtained from Wild Core Collection in the National Institute of Genetics. GA3ox1 orthologs from FF to AA genome species27,28,29 demonstrate F at the corresponding residue (referred to as F-type) (Fig. 3a). Amino acid residues for 2OG and Fe (II) binding are highly conserved, except F235. On the other hand, the GA3ox1 orthologs of GG, HHJJ genome species, and L. perrieri, possess Y (referred to as Y-type). These results indicated that the substitution occurred in the ancestor of FF to AA genomes (Fig. 3b). We also isolated full-length cDNA of GA3ox1 orthologs from six Oryza plants and L. perrieri and determined the productive ratio of GA7/GA4 in vitro (Fig. 3c). The ratio of GA3ox1 in O. granulata (GG) was <0.1, whereas that of O. brachyantha (FF) was >0.2. Notably, the ratio increased during evolution from the EE to AA genome, with >0.5 in O. sativa. For L. perrieri, the ratio was about 0.3. We found that the amino acid residues that bind to the substrate GA9, 2OG, and Fe(II) were conserved in GA3ox1 of Oryza genus and L. perrieri except for F235 (Supplementary Fig. 10a). Using structural prediction and the machine learning method of AlphaFold230, we investigated the cause of the high GA7 production rate of L. perrieri GA3ox1. The structural prediction suggested that Arg315 (corresponding to Arg318 in OsGA3ox1) at the interaction site with 2OG was oriented differently only in L. perrieri GA3ox1, resulting in reduced hydrogen bonding (Supplementary Fig. 10b-e). Therefore, the GA7/GA4 ratio of L. perrieri GA3ox1 may also be because its binding to 2OG differs from that of OsGA3ox2.

Fig. 3: Residues in the 2OG-interacting site of Poaceae GA3ox orthologs and enzymatic activity of Oryza GA3ox1.
figure 3

a Amino acid sequence alignment in the 2OG-interacting site around F (in OsGA3ox1) of Poaceae GA3ox orthologs and OsGA20ox2. Residues of F in the interacting site of 2OG are indicated in green, those that are 100% identical are shown in black, and those with more than 50% identity are depicted in gray. The black dot indicates the amino acid in the interacting site with co-substrate 2OG, F (shown in green), or Y. On the right side of the alignment, the letters AA and BB indicate the AA and BB Oryza genomes, respectively. b Phylogenetic tree of Poaceae GA3ox orthologs and OsGA20ox2. The red dot indicates the common ancestor of O. sativa to O. brachyantha that acquired the residue F in the active site. c In vitro enzyme activity of GA3ox1 of Oryza species and L. perrieri; O. sativa GA3ox1, O. rufipogon GA3ox1, O. punctata GA3ox1, O. brachyantha GA3ox1, O. granulata GA3ox1, L. perrieri GA3ox1. Error bars, s.d. n = 3, one-way ANOVA with Tukey’s multiple comparisons test. Different letters denote significant differences (p < 0.05).

Expression patterns of Oryza GA3ox1

Expression data of the Poaceae species, Triticum aestivum, Hordeum vulgare, Brachypodium distachyon, Sorghum bicolor, and Zea mays revealed that several GA3oxs homologs were expressed in anthers but did not exhibit anther-specific characteristics as OsGA3ox1 (Supplementary Fig. 11). Next, we compared the tissue specificity and intensity of GA3ox1 orthologs among six Oryza species and L. perrieri by performing quantitative reverse-transcription polymerase chain reaction (qRT-PCR) (Fig. 4a–g). GA3ox1 orthologs of the Oryza AA genome species (O. sativa and O. rufipogon) showed strict anther specificity and high expression levels (Fig. 4a, b). The GA3ox1 orthologs of O. punctata (BB) and O. australiensis (EE) were expressed in anthers, roots, and whole spikelets (Fig. 4c, d). In O. brachyantha (FF) and O. granulata (GG), GA3ox1 orthologs were also detected in roots or other organs (Fig. 4e, f), and the expression levels in anthers were lower than those in the anthers of AA genome species. In L. perrieri, the expression showed anther specificity, but at extremely low levels (approximately ten‐thousandth of that of OsGA3ox1) (Fig. 4g). These results indicated that the expression pattern of GA3ox1 underwent evolution to gain anther-specific features and underwent enhancements during genus Oryza evolution. The endogenous GA7 levels were consistent with the enzyme activities and the expression analysis data (Fig. 4h).

Fig. 4: Expression analysis of GA3ox1 and endogenous gibberellic acid levels in Oryza species and L. perrieri.
figure 4

ag Gene expression analysis of GA3ox1 orthologous genes in Oryza species and L. perrieri; O. sativa (a), O. rufipogon (b), O. punctata (c), O. australiensis (d), O. brachyantha, (e), O. granulata (f), L. perrieri (g) in leaf blades, leaf sheaths, stems, roots, spikelets, anther, and pistils. Error bars, s.d. n = 3 for each sample. h Endogenous GA levels in anthers of Oryza species and L. perrieri. Error bars, s.d. n = 3 (O. rufipogon, O. punctata, O. australiensis, O. brachyantha, and L. perrieri) or n = 2 (O. granulata).

GA7 is important for the fertility of Oryza

As the pollen grains of osga3ox1 mutant showed abnormality in starch accumulation (Fig. 2a, b, g), we analyzed starch content in anthers obtained from six Oryza and L. perrieri species (Fig. 5a). The further the phylogenetic distance to O. sativa, the lower the levels of endogenous starch in anthers. Similar to the endogenous starch levels, the performance of I2-KI staining in pollen revealed that AA genome species accumulated more starch than other species (Fig. 5b).

Fig. 5: Effects of GA3ox1 on pollen and seed fertility in Oryza species and L. perrieri.
figure 5

a Endogenous quantities of starch in Oryza species and L. perrieri anthers. Error bars, s.d. n = 3, one-way ANOVA with Tukey’s multiple comparisons test. Different letters denote significant differences (p < 0.05). b Quantification of the staining value of KI-stained Oryza species and L. perrieri pollen. n = 300 pollen grains for each species, statistical analysis as mentioned in panel (a). c Correlation between starch content and endogenous gibberellin levels in Oryza species and L. perrieri anthers. The coefficient of determination (R2) is shown on the upper right side of the plot. The respective values are 0.0085 between starch content and GA4 levels (left panel), and 0.7235 between starch content and GA7 levels (right panel). d Quantification of the staining value of KI-stained pollen grains of transgenic plants described in Supplementary Fig. 12a and vector control. n = 300 pollen grains for each line; statistical analysis as mentioned in panel (a). e Seed setting rate of transgenic plants described in Supplementary Fig. 12a and vector control. n = 3 panicles for each line; statistical analysis as mentioned in panel (a).

Pollen starch accumulation is important for pollen tube germination and elongation and contributes to male reproductivity. Previously, Lee et al. 31 compared the pollination of adp-glucose pyrophosphorylase (osagpl4) mutant, defective in starch synthesis, with that of homozygous (OsAGPL4−/−) and heterozygous (OsAGPL4+/−) mutant anthers. They showed that 10–20% of OsAGPL4−/− inheritors could be detected in the descendants of OsAGPL4−/− plants, whereas none were detected in OsAGPL4+/− plants, suggesting that the pollen grains with high starch accumulation out-competed those with low starch. Analogously, increasing starch content in pollen during the evolution of Oryza GA3ox1 orthologs might have conferred reproductive advantages. We found a correlation between starch content and endogenous GA7 in pollen; however, no correlation between starch and GA4 was observed (Fig. 5c), suggesting that endogenous GA7 is important for starch accumulation in pollen.

To compare the effects of both transcription and enzyme activity on the reproductive phenotypes in the same background, we conducted a complementation test by introducing the genomic region of GA3ox1 orthologs of five Oryza and L. perrieri plants into the osga3ox1 mutant background (Supplementary Figure 12a). K2-KI staining and seed setting rate analyses revealed an appreciable correlation with GA7 productivities of GA3ox1 orthologs (Figs. 4h, 5d, e, Supplementary Fig. 12b, c). The transformants of L. perrieri GA3ox1 showed low expression levels of GA3ox1 in the anthers, while the transformants of O. granulata GA3ox1 showed slightly higher expression levels of GA3ox1 in the leaf blades (Supplementary Fig. 12d). Consistent with each expression level and enzymatic activity, no significant difference was detected in the vegetative phenotype (Supplementary Fig. 12e, f). We also measured the endogenous levels of bioactive GAs in the anthers of all transgenic lines (Supplementary Fig. 12g). The amount of endogenous GA7 in the anthers of the lines that possess GA3ox1 of AA genome plants (O. sativa and O. rufipogon) was higher than that in the anthers of other lines, consistent with their reproductive phenotypes. These results indicate that the ability to synthesize GA7 as well as GA4 may be important, and is likely related to pollen starch accumulation and improved reproductive performance.

Here, we report the analysis of the enzyme and expression evolution of GA3ox1 in wild rice after the establishment of the Oryza species. The Y to F amino acid substitution in the 2OG interaction site accounts for GA7 production in the AA-FF genome species. In line with the evolution of enzymatic activities, gene expression has been localized mainly in the male organ, but not in the shoot where excessive bioactive GAs can promote harmful overgrowth. These enzymatic and expression evolutions might have made it possible for rice to contain high endogenous levels of bioactive GAs in an anther-specific manner. These large amounts of GAs contribute to the accumulation of starch through the regulation of invertase (Fig. 2g, h, j, k). Given the male sterility phenotype of osga3ox1, during evolutionary processes, GA3ox1 may have become crucial for male fertility in rice. This unique GA7 synthesis strategy in Oryza might confer certain reproductive advantages for its successful global dissemination32. Altogether, this study paves the way for a better understanding of the factors of dissemination and the reproductive systems of Oryza.


Plant material, growth conditions, and genotyping

Seeds, whole plants, and genomic DNA extracts of wild rice species were obtained from the Resource Bank of the National Institute of Genetics ( The accession numbers of wild rice used in this study are: Oryza rufipogon, W0106; Oryza punctata, W1514; Oryza officinalis, W1361; Oryza rhizomatis, W1805; Oryza eichingeri, W1527; Oryza latifolia, W1166; Oryza alta, W0017; Oryza grandiglumis, W0613; Oryza australiensis, W0008; Oryza brachyantha, W0654; Oryza granulata, W0003; Oryza meyeriana, W1356; Oryza longiglumis, W1220; Oryza ridleyi, W0001; and Leersia perrieri, W1529.

Oryza sativa L. “Nipponbare”, the osga3ox1 mutant and wild species, seeds were hulled and surface-sterilized with 70% ethanol and 50% sodium hypochlorite. After sterilization, seeds were germinated and subjected to growth conditions on 0.5 × Murashige and Skoog (MS) media containing 0.8% gellan gum (Wako) and 25 mg/l meropenem (Wako) for selection of the osga3ox1 mutant. After germination under continuous light conditions at 30 °C for 2 weeks, seedlings were transplanted to a pot for growth in a greenhouse at a temperature of 28 °C and 60% humidity under 16 h light (long-day treatment) conditions. After approximately 2 months of subjection to the above-mentioned growth condition, plants were subjected to a 10 h light (short-day treatment) condition to induce reproductive growth.

To perform genotyping for osga3ox1, the PCR reactions were conducted using the primers listed in Supplementary Table 3. After the completion of PCR amplification, DNA fragments were subjected to digestion at 37 °C for 2 h with KpnI in 10 μl of the following mixture: 10 units KpnI, 1 μl 10× buffer, and 2 μl of PCR-amplified DNA fragments. Then, the digested DNA fragments were subjected to gel electrophoresis; the presence of a 949 bp fragment indicates the presence of the osga3ox1 mutant, whereas KpnI digestion (presence of both 519 bp and 429 bp bands) indicates the presence of the wild-type (OsGA3ox1+/+); heterozygous (OsGA3ox1+/−) plants were identified based on the presence of all three bands (519 bp, 429 bp, and 949 bp) were present. The four alleles of osga3ox1 were confirmed by sequencing. The T1 generations of homozygous mutants (osga3ox1-cr1/osga3ox1-cr1 and osga3ox1-cr7/osga3ox1-cr7) and heterozygous mutants (OsGA3ox1/osga3ox1-cr1 and OsGA3ox1/osga3ox1-cr7) were obtained by GA treatment of the T0 generations. We mainly used osga3ox1-cr1/osga3ox1-cr1 and OsGA3ox1/osga3ox1-cr1 for the analyses.

Plasmid construction and rice transformation

PCR amplification for all constructs was performed using high-fidelity PrimeStar-GXL DNA polymerase (Takara Bio Inc., cat# R050A) or KOD FX Neo (TOYOBO, cat# KFX-201). Primer sequences are listed in Supplementary Table 3.

For constructs of the GST-tagged proteins of OsGA3ox1 and OsGA3ox2, the full-length coding sequences for OsGA3ox1 and OsGA3ox2 were amplified, with the full-length cDNA provided by the National Institute of Agrobiological Sciences (NIAS) used as the PCR template and subcloned into pGEX6P1 (Cytiva). F235Y-OsGA3ox1 was constructed by performing PCR using mutagenized primers with pGEX6P1/OsGA3ox1 as the template.

The osga3ox1 mutant was generated using the CRISPR/Cas9 system following the method reported by Mikami et al.33. To identify a suitable protospacer adjacent motif (PAM), CRISPR-P ( was used. Concentrations of 2 μM of the oligo-DNA fragments of sequences corresponding to the guide-RNAs listed in Supplementary Table 3 were incubated at 95 °C for 5 min. Then, they were annealed at 25 °C for 20 min. The cloning vector (pU6gRNA) digested with BbsI and oligo-DNA fragments were ligated and transformed into E. coli competent cells (XL10-gold). Using competent cells, plasmid DNA extraction was performed based on the alkaline lysis method34. The inserted target sequence in the obtained plasmids was sequenced to ensure that no unintentional mutations were introduced. The cloning vectors mentioned above were digested with I-SceI, and the spectinomycin-resistant binary vector pZDgRNA_Cas9ver.2_HPT was digested with AscI and PacI. The digested DNA fragments containing the guide-RNAs sequence and binary vector were ligated and transformed into E. coli competent cells (XL10-gold). Plasmid DNA extraction was then performed using competent cells by adopting the alkaline lysis method34. Then, to confirm the correct insertion of fragments containing the guide-RNA sequence into a binary vector, DNA was digested by BamHI, and products were visualized by gel electrophoresis.

For promoter-GUS constructs, the 3.3 kb region upstream of the 2.6 kb promoter to the middle of the second exon of OsGA3ox1 was amplified by PCR using the primers listed in Supplementary Table 3. The amplified DNA fragments and the kanamycin-resistance vector pBI101-Hm3 were digested with XbaI and SmaI, respectively, and the fragments were inserted into the vector pBI101-Hm3 by ligation. The plasmid DNA was transformed into competent cells (XL10-gold), and extracted using the alkaline lysis method34. The inserted DNA fragment sequence was confirmed by sequencing.

For the wild rice GA3ox1 construct, genomic DNA fragments (O. sativa, 7.8 kb; O. brachyantha, 5.9 kb; O. granulata, 5.6 kb; L. perrieri, 5.2 kb) containing the full-length wild rice GA3ox1 genes were isolated from genomic DNA extract derived from leaves of each plant by PCR using KOD FX Neo (TOYOBO, cat# KFX-201) and primers listed in Supplementary Table 3. These amplified PCR fragments were cloned into the pCR4 Blunt-TOPO vector (Invitrogen, cat# K2875J10) and transformed into E. coli competent cells (XL10-gold). Plasmid DNA extraction was performed using competent cells by adopting the alkaline lysis method34, and successful transformations were confirmed by sequencing. For GA3ox1 obtained from O. sativa and O. rufipogon, to prevent it from being targeted by CRISPR/Cas9, the target sequence (5′-CCTCCGGTACCCGAAGCAGATG-3′) was synonymously substituted as below; 5′-GCTCCGGTACCCGAAACAAATG-3′. Using these plasmids as templates, a second PCR ware conducted using the PrimeSTAR GXL DNA polymerase (Takara Bio Inc., cat# R050A) and the primers listed in Supplementary Table 3. The resulting PCR fragments contained a HindIII linker sequence at the 5′-end and a KpnI linker at the 3′-end; the pSTARA R-5 vectors (Kumiai Chemical Industry Co., Ltd.) were digested with KpnI and HindIII. The digested PCR fragments were inserted into the pSTARA R-5 vector using the NEBuilder HiFi DNA Assembly Kit (New England BioLabs) and were transformed into E. coli One Shot ccdB Survival 2 T1R competent cells (Invitrogen, cat# A10460). The extracted plasmid DNA was introduced into the Agrobacterium tumefaciens strain EHA101 and was used to infect osga3ox1 mutant rice calli35. After subjection to growth continuous of light at 30 °C for 2 weeks, all plants were planted in a pot and grown in a greenhouse.

GA treatment of rice seedlings

For the GA treatment response test, sterilized tangibozu seeds36 were germinated in half-volume Murashige Skoog medium; after 2 days, 1 μL of GA (10−12, 10−11, 10−10, and 10−9 M; in 50% acetone) was added to the coleoptile and leaf-sheath junction of each seedling. For the control, 1 μL of 50% acetone was added. Six days later, the length of the second leaf sheath was measured.

For the effects of gibberellin on shoot elongation of Nipponbare, GA3, GA4, and GA7 were diluted with ethanol to prepare a stock solution, which was further diluted to the described concentration and used to treat the rice plants. The control group was treated with the same volume of ethanol as the stock solution diluted in the same manner.

KI staining of pollen grains and determination of starch content in anthers

For evaluation of pollen viability, two fresh rice anthers per spikelet were randomly collected in 10 μl of KI solution [1% (v/v) of I2 in 3% (v/v) KI] and were crushed well. Then, 2 μl of the solution was sampled on a slide glass and observed with a stereoscopic microscope using a bright field. Pollen grains that were round and exhibited black coloration after staining were assessed as viable, and yellow- or light red-stained pollen grains were considered sterile.

To evaluate the normalized pollen staining value, white balances of micrographs of KI staining were uniformed and converted to hue, saturation, value (HSV) color space, and then gray values of each pollen were evaluated using the ImageJ script (Supplementary Code 1).

Evaluation of starch content in pollen grains

The frozen anther samples were homogenized using 80% ethanol. The insoluble material was collected by centrifugation and was subjected to washing steps using hot 80% ethanol. After centrifugation, the pellet was used for starch analysis. The supernatant was treated with activated charcoal and was filtered to remove ethanol-soluble materials, which could interfere with the assay; then, the sample was used for glucose and sucrose analysis. Sucrose was hydrolyzed into glucose and fructose by invertase (Sigma, cat# I4504), and content was calculated as the difference in the amount of glucose before and after hydrolysis. Glucose was determined via the glucose oxidase method using the glucose autokit (Wako, cat# 439-90901) as per the manufacturer’s instructions. Starch was enzymatically digested to glucose by amylase and amyloglucosidase (Megazyme, cat# K-TSTA) by following the product manual and was quantified using the method described above.

Examination of pollen tube germination and growth

For the in vitro germination test, pollen grains collected from the dehiscent anthers were placed on germination medium [1.6 mM H3BO3, 1.8 mg Ca(NO3)2, 8.5% sucrose, 0.7% agarose] and incubated at room temperature for 2 h. Then, they were observed under an optical microscope.

For the in vivo germination test, 2–4 h after artificial pollination, rice pistils were subjected to fixation using a fixative (ethanol: acetic acid = 3:1) for 30 min and then mixed with 1 N KOH, and incubated at 55 °C for 30 min. The fixed rice pistils were subjected to washing steps using distilled water and stained with 0.1% aniline blue in K3PO4 buffer (pH 8.5), at 25 °C for 2 h. Then, rinsed pistils were visualized under a UV microscope.

DNA sequence analysis, collection of GA3ox1 sequences, and phylogenetic analysis

For examination of the GA3ox1 nucleotide sequence in wild rice, genomic DNA extraction was performed using leaves or samples obtained from the National Institute of Genetics supported by the National Bioresource Project, MEXT, Japan. The DNA fragments of part of the wild rice GA3ox1 gene were amplified using PCR, using KOD FX Neo (TOYOBO, cat# KFX-201) and the primers listed in Supplementary Table 3. The GA3ox1 DNA fragments were sequenced using primers considered for PCR.

For tetraploid Oryza wild-rice GA3ox1 orthologs (O. latifolia, O. alta, O. grandiglumis, O. longiglumis, and O. ridleyi), sequences of two GA3ox1 orthologs of O. ridleyi were obtained; these two orthologs were identical in the region shown in Fig. 3a. For other tetraploid Oryza wild rice, the GA3ox1 orthologs could not be sequenced. To ensure that the sequences were orthologous for the GA3ox1 gene, alignment analysis and phylogenetic analysis were performed with MAFFT version 7.3 using the L-INS-i model37 program for alignment and phylogenetic tree illustrations were generated with the NJ method by using the GENETYX-Mac ( software (version 19).

For GA3ox protein sequence alignment and phylogenetic analysis, GA3ox amino acid and nucleotide sequences of various plants listed in Supplementary Table 2 were subjected to a best-BLAST match search in databases and were aligned with MAFFT version 7.3 using the L-INS-i model37. For the construction of an amino acid phylogenetic tree, Bayesian estimations of phylogenetic topology were conducted using the ML method by MEGA (ver. 10) with the general time-reversible (+I + G) model, which was selected in ProtTest 3.4.238.

GA binding and yeast two-hybrid assays

For performing the yeast two-hybrid assay, the Matchmaker Two-Hybrid System (Clontech Laboratories, Inc., cat# 630489) was used. The yeast strain HA109 was used, and the construct was prepared as per methods described previously1 GID1 in pGBKT7 served as the bait, and SLR1 in pGADT7 was considered as the prey. Plate assays (-His) were performed according to the manufacturer’s instructions with modifications to the plates containing GA1, GA3, GA4, and GA7 concentrations ranging between 10−10 M and 10−5 M. Control plates contained no GA.

RNA isolation and real-time RT-PCR analysis

For RNA isolation from rice and wild rice, the RNeasy plant mini kit (Qiagen, cat# 74904) was used according to the manufacturer’s instructions. First-strand cDNA was synthesized from 10 ng or more of total RNA using the Omniscript reverse-transcription kit (Qiagen, cat# 205111). For Real-time RT-PCR analysis, the KOD SYBR qPCR Mix (TOYOBO, cat# QKD-201) and the primers listed in Supplementary Table 3 were used with the CFX96 real-time PCR detection system (Bio-Rad). Ubiquitin-5 served as the internal control for the normalization of expression levels.

Expression and purification of OsGA3ox1 and OsGA3ox2

For the production of the OsGA3ox1 and OsGA3ox2 proteins, OsGA3ox1/pGEX6P1 or OsGA3ox2/pGEX6P1 constructs were overexpressed in Escherichia coli Rosetta (DE3) pLysS (Novagen). Cells were grown until the optical density reached 0.8–1.0 A600 at 37 °C in Terrific-Broth medium containing 100 μg ml−1 ampicillin, followed by IPTG induction at 18 °C for 16–18 h. The harvested cells were subjected to washing steps using buffer A [10 mM Na-phosphate, 150 mM NaCl, 1 mM DTT (pH 7.5)] containing complete protease inhibitor (Merck) and 1.5 mM phenylmethylsulfonyl fluoride and were then disrupted by sonication. OsGA3ox1 or OsGA3ox2 protein was obtained from the supernatant fraction. The recombinant protein in the supernatant was purified using Glutathione Sepharose 4B resin (Cytiva) equilibrated with buffer A. The column was washed with 30-bed volumes of buffer A and was then eluted with buffer A (plus 20 mM glutathione pH 7.5). To remove the GST-tag, precision protease (Cytiva) was added and incubated overnight at 4 °C. The sample was filtered through a 0.22-μm nylon filter and further purified using a Superdex 200 16/60 column (Cytiva) with buffer A used as a running buffer. The fractions containing the GST and tag-digested OsGA3ox1 or OsGA3ox2 complex were collected and passed through Glutathione Sepharose 4B resin to remove GST. The sample was concentrated and dialyzed with buffer C (10 mM sodium phosphate, pH 7.4) using an Amicon Ultra-4 concentrator unit (10 kDa considered as the molecular weight cut-off) (Millipore). The protein concentrations of OsGA3ox2 were measured by determining UV absorbance at 280 nm. The A0.1% values at 280 nm were calculated based on the amino acid composition data39. The target proteins were detected by comparing with protein standard markers using 12% wt/vol SDS gel, and the optimum conditions for in vitro expression were determined.


Crystallization was performed by the sitting drop vapor diffusion method at 20 °C using a mixture containing 1 μl protein and 1 μl mother liquid solution. Droplets containing 20 mg ml−1 OsGA3ox2 dissolved in 5 mM Na-phosphate (pH 7.5), 10 mM 2-oxoglutarate, and 0.5 mM GA9 and mother liquor, were equilibrated against 50 μl of reservoir solution containing 0.15 M ammonium sulfate, 0.1 M sodium HEPES (pH 7.0), and 20% (w/v) PEG4000. Crystals were flash-frozen in liquid nitrogen with the addition of 20% glycerol in the crystallization mother liquid as a cryoprotectant. X-ray diffraction data were collected using the BL26B1 beamline at SPring-8. All data were processed and scaled using HKL200040. The crystal data are summarized in Supplementary Table 1.

Structure determination and refinement

The structure of OsGA3ox2 was determined by performing molecular replacement with Molrep41,42 using the refined structure and OsGA2ox3 (6KU3) as the search model. Model building and refinement were performed using WinCoot (version 0.8.9)43, and restrained refinement was performed using REFMAC541 from the CCP4 package (version 7) and PHENIX program (version 1.17.1)44. The statistically analyzed data after collection and refinement are shown in Supplementary Table 1. Molecular graphics of OsGA3ox2 and predicted models by Alphafold2 were illustrated using PyMOL (

GA quantification

Extraction and semi-purification of GAs were performed as previously described45,46. Quantification of GAs was carried out using an ultra-high performance liquid chromatography (UHPLC)-ESI quadrupole-orbitrap mass spectrometer (UHPLC/Q-Exactive™; Thermo Scientific) as described previously46,47 with an ODS column (AQUITY UPLC HSS T3, 1.8 mm, 2.1 × 100 mm; Waters).

In situ staining of invertase activity

Wash buffer [500 μl; 200 mM HEPES-KOH (pH 7.0), 3 mM MgCl2, 1 mM EDTA, 2% glycerin] was added to the anthers, and the anthers were broken in tubes. These were centrifuged at 8000 × g for 5 min at room temperature. Fresh 500 μl wash buffer was added to the precipitate, mixed by vortexing, and centrifuged again at 8000 × g for 5 min at room temperature to remove as much as possible of the material other than pollen. This was repeated once more to remove the soluble sugars. These were resuspended in wash buffer and dispensed in 20 μl portions, and Incubation buffer [20 U ml−1 glucose oxidase, 0.014% phenazine methosulphate, 0.024% nitroblue tetrazolium salt, 0.5% Sucrose], adjusted to each pH using 100 mM citric acid and 200 mM K2HPO4, was added. Sucrose was not added to the Incubation buffer in the control experiment. The supernatants were removed by centrifugation at 8000 × g for 5 min at room temperature after shaking for 16 h at 26 °C in the dark. Then, 200 μl 70% ethanol was added, and the ethanol was removed by centrifugation at 8000 × g for 5 min at room temperature. The pollen grains were suspended in 20 μl of sterile water and observed and photographed under an optical microscope.

DAPI and FDA staining

For DAPI staining, anthers were washed with 80% ethanol after fixation, placed on a glass slide, and 5 μl 2 μg ml−1 DAPI staining solution (2 μl of 1 mg ml−1 DAPI scale-up to 1 ml in PBS) was added. The anthers were crushed in DAPI solution. Pollen was placed on a glass slide and covered with a cover glass for observation.

For FDA staining, a 10 μl drop of 0.5 μg ml−1 FDA solution (0.5 mg ml−1 FDA in DMSO diluted 1000-fold with PBS solution) was placed on a glass slide, and two anthers were crushed in the solution to release the pollen. As a negative control, PBS solution containing pollen was heated in a block incubator at 100 °C for 5 min, and then 0.5 μg ml−1 FDA solution was added. After covering with a cover glass, the cells were incubated at room temperature for 5 min and observed with a ZEISS LSM 700 confocal microscope.

Anther fraction isolation and invertase enzyme assays

For anther fraction, randomly collected rice anthers (each sample contained approximately 250 pollen grains) were sonicated in 5 mM sodium phosphate buffer (pH 7.0), centrifuged at 15,000 × g for 30 min at 4 °C to separate them into soluble invertase (supernatant) (A fraction in Fig. 2i and j) and wall-bound invertase (precipitate). Furthermore, 1.3 M NaCl was added to the precipitated fraction, incubated overnight, separated into the supernatant, and precipitated by centrifugation (D and E fractions in Fig. 2i, j). Each sample, including the one incubated overnight in 5 mM phosphate buffer as a control (B and C fractions in Fig. 2i, j), was used for subsequent enzyme test.

For invertase enzyme assays, invertase activity was calculated according to a previously described protocol48,49, with some modifications. For each sample, 50 μL of solution were diluted in 0.2 M sodium acetate buffer (pH 5.5 for wall-bound invertase assays) to a final volume of 600 μL. The enzymatic reaction was triggered by adding 800 μL of 0.225 M sucrose. An aliquot of 700 μL (t 1 h sample) was transferred to a new tube, whereas the remaining volume (t 0 h sample) was mixed with 200 μL Somogyi’s solution (Sigma, cat# 28-4730) and heated for 15 min at 100 °C. The t 1 h sample was incubated at 30 °C for 1 h, and the reaction was stopped as described before. Samples were then placed on ice for 5 min, and 200 μL Nelson’s solution (Sigma, cat# 21-0925) was added and left for 5 min. After 20 min, the absorbance at 660 nm was measured. The invertase activity was calculated as the difference in reduced sugar levels between t 1 h and t 0 h and was represented as μg of glucose formed in 1 ml of extract per hour and mg of total protein.

For western blotting, randomly collected rice anthers (each sample contained approximate three thousand pollen grains) were used. After adding 2× sample buffer to the protein fractions, the samples were boiled for 5 min. Protein samples were separated by 6% SDS-PAGE and transferred to a PVDF membrane using the Trans-Blot Turbo transfer system (Bio-Rad Laboratories, cat #1704156). To detect OsCIN3 protein, blots were blocked with 5% skim milk in TBST buffer [0.1% Tween 20 in 2 mM Tris-HCl, pH 7.6, 13.7 mM NaCl] for 1 h, followed by incubation with rabbit-grown anti-OsCIN3 antiserum (1:5,000 dilution, Cosmo Bio) at 4 °C overnight. The blots were washed three times with TBST buffer for 15 min each. The membrane was incubated with goat anti-rabbit IgG horseradish peroxidase-conjugated secondary antibody (1:10,000 dilution) for 45 min, and then washed following the same procedure above. All reactions, except the primary antibody reaction, were performed at room temperature.

Statistics and reproducibility

Results are presented as the mean ± SD. Statistical tests were selected on a case-by-case basis, depending on the nature of the data collected. More information on the selection of tests can be found in the figure legends. A two-tailed probability of less than 0.05 was considered statistically significant.

Reporting summary

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