The dg1 mutant of rice (O. sativa) was obtained from the EMS treatment of the japonica cultivar ZH11. ZH11 plants were used as the wild-type strain for phenotypic observation. The dg1 mutant was crossed with indica cultivars Nanjing 6 (NJ6) and Taichung Native 1 (TN1) and the F₁ plants were self-pollinated to generate the F₂ population. Rice plants were cultivated at the experimental paddy in China National Rice Research Institute, Hangzhou from May to October, and in Lingshui, Hainan Province from January to May.

The dg1 mutation was primary mapped in the F₂ population of a cross between the dg1 mutant and the indica cultivar NJ6, with which six simple sequence repeat (SSR) markers from publicly available rice databases flanking the target region. For fine mapping of the dg1 mutation, more F₂ plants were screened and new insertion or deletion (InDel) molecular markers were used. The sequences of primers are listed in Supplementary Table 1.

Fresh spikelet specimens were acquired from the same position of ZH11 and the dg1 mutant grown under the same conditions. Specimens were fixed at 4°C overnight in 50% FAA solution (50% ethanol, 10% formaldehyde, 5% glacial acetic acid, and 35% sterile water). Spikelet specimens were placed under vacuumed for about 30 min prior to fixation to ensure the materials were completely immersed in the FAA solution. The fixed samples were then treated with an ethanol gradient (50, 70, 80, 90, 100, 100, and 100%). Spikelet specimens were infiltrated in xylene and embedded in Paraplast Plus (Sigma). Using a microtome (HM340E), samples were cut into approximately 8-μm thick sections and flattened on glass slides coated with poly-lysine. After deparaffinating the samples in a xylene series and dehydrating them with an ethanol series, the samples were stained using 1% safranine and 1% Fast Green, then dehydrated in an ethanol series and cleared in a xylene series. For microscopy, sections were covered with neutral resins at 42°C for 48 h. Optical microscopy was conducted with a Nikon Eclipse 90i microscope. For electron microscopy, fresh spikelet hull specimens were observed using a HITACHI S-3500 scanning electron microscope that had a –30°C cool stage. We examined cell size, number, and area using the measurement tools of the Nikon Eclipse 90i microscope, and six biological repeats were examined for cell size, cell area, and cell number to obtain the final results.

The total RNA was isolated from floral organs and inflorescences of the wild type and dg1 mutant using the AxyPrep^TM total RNA Miniprep Kit (Axygen) according to the manufacturer’s instructions. The first strand of complementary DNA (cDNA) was reverse transcribed from 500 ng total RNA in a 50 uL reaction volume using the ReverTra Ace^® quantitative PCR RT Master Mix Kit with gDNA remover (Toyobo) and qRT-PCR was conducted with a CFX96 Touch^TM Real-time PCR Detection System and the 2x SsoFast^TM EvaGreen^® SuperMix (Bio-Rad) to amplify the cDNA of all tested genes. All target genes were normalized to the rice internal control gene Actin to detect the relative expression levels. Three biological repeats were conducted to obtain the final results. The primers for the qRT-PCR assays are listed in Supplementary Table 1.

Germinated seeds from ZH11 and dg1 mutant were grown in standard culture solutions of 0, 0.01, 0.1, and 1 uM 24-Epibrassinolide (24-eBL) under darkness for a week, respectively. The angle of leaf inclination was then measured and photographed.

In the vegetative phase, the dg1 mutant was shorter and thinner than the wild type (Figures 1A,B and Supplementary Figure 1). The first internode was distinctly shorter in the dg1 mutant, but the second, third, fourth, and fifth internodes showed no differences when compared with the wild type (Figure 1B and Supplementary Figure 1). The culms and leaves of dg1 mutant were thinner than those of the wild type (Figures 1C,D,F,G and Supplementary Figure 1) and the base of the flag leaves was rolled (Figure 1D). In addition, the leaf angle was enlarged in the dg1 mutant compared with the wild type; the angle of leaf inclination was increased by 51% in the dg1 mutant (Figure 1E and Supplementary Figure 1). Additionally, the average collar length of the adaxial surfaces in the lamina joint of the dg1 mutant was increased by 40%, while the average collar length in the abaxial side was indistinguishable from the wild type (Supplementary Figure 1).

We also performed histocytological analysis to study the cell structures. The cells were shorter in the first internodes of the dg1 mutant (Figures 1H,I and Supplementary Figure 1). Analysis of cross sections of the culms and leaves showed that the total cell numbers were reduced and the number of small vascular bundles was markedly decreased in the dg1 mutant compared with the wild type (Figures 1J,K,L,M,R,S and Supplementary Figure 1). However, the number of big vascular bundles remained unchanged between the wild type and the dg1 mutant (Figures 1J,K,R,S and Supplementary Figure 1). These results suggested that DG1 affects the diameter of culms and the width of leaves mainly by regulating the number of small vascular bundles. In the dg1 leaves, the number and area of bulliform cells were also increased (Figures 1N,O,P,Q and Supplementary Figure 1), resulting in adaxial leaf rolling.

Brassinosteroid plays a vital role in plant growth and development, especially in leaf morphology and leaf angle (Sakamoto et al., 2013; Zhang et al., 2015). Thus, to test whether the dg1 mutant has defects in BR signaling, we conducted a BR sensitivity experiment using different concentrations of 24-eBL and measured the degree of leaf inclination. Notably, the dg1 mutant showed a larger lamina joint angle than that of the wild type after treatment with 0.1 μM 24-eBL (Figures 2A–E). This showed that the dg1 mutant was more hypersensitive to BR than the wild type and suggested that dg1 mutant may be related to BR signaling. We examined the expressions of related genes on BR response and biosynthesis in the BR-treated and untreated wild type and dg1 plants by qRT-PCR. In the untreated plants, the expression of the BR response genes OsBRI1 (brassinosteroid insensitive 1) and OsBZR1 (Brassinazole-resistant 1), was higher in the dg1 mutant than in the wild type (Figure 2F). Similarly, the mRNA levels of BR biosynthetic genes D2 (CYP90D2) and BRD1 (BR-deficient dwarf1) were significantly up-regulated in the dg1 mutant compared with those of the wild type under normal growth condition (Figure 2F). In the treated plants, the expression levels of OsBRI1, OsBZR1, D2 and BRD1 were reduced in the wild type and dg1 mutant.

Compared with the wild type without BR treatment, the expression of D2, BRD1 and OsBZR1 in the wild type under BR treatment was reduced by 70, 73, and 23%, respectively. By contrast, the expression of D2, BRD1 and OsBZR1 in the dg1 mutant under BR treatment was reduced by 75, 77, and 31%, respectively. The decreased ranges of expressions of D2, BRD1 and OsBZR1 were higher in the dg1 mutant than that in the wild type under BR treatment, further revealing that the dg1 mutant was hypersensitive to exogenous BR. These findings indicated that OsBRI1, OsBZR1, D2 and BRD1 were modulated by DG1 under normal growth condition, and confirmed that DG1 is involved in regulating BR signaling.

The wild type spikelet in rice is composed of two pairs of vestigial glumes: rudimentary glumes and sterile lemmas, and one terminal floret comprising a lemma, a palea, two lodicules, six stamens and a central pistil (Figures 3A–E). At heading stage, we found that about 39% spikelets showed longer rudimentary glumes in the dg1 mutant (Figures 3A,B,F,G). The average size of wild type rudimentary glumes was about 0.5 mm length, while the size of rudimentary glumes was variable, from 1.1 to 4.0 mm in length in the 39% spikelets of the dg1 mutant (Supplementary Table 2). About 94% spikelets of the dg1 mutant produced larger sterile lemmas that were similar to the lemmas of the wild type or dg1 mutant (Figures 3A,B,F,G). The length of the dg1 sterile lemmas was varied from 6.0 to 7.9 mm (Supplementary Table 2).

We next investigated the floral organs of four whorls in the dg1 flowers. Interestingly, the dg1 mutant showed slightly whiter and smaller stamens, and highly bushy stigmas compared with the wild type (Figures 3C,D,E,H,I,J). However, the number of floral organs was not changed in the dg1 mutant (Figures 3A,C,F,H). To investigate the cellular morphology, we performed paraffin section and scanning electron microscopy (SEM). In the wild type, the lemma had five vascular bundles and two unique inward hook-like structures (Supplementary Figures 2A,C). The palea had three vascular bundles and exhibited a uniquely smooth epidermis that lacked outer silicified epicuticular cells (Figures 4A–C and Supplementary Figures 2A,C). The sterile lemma developed one vascular bundle (Supplementary Figure 2D) and had a smooth epidermis with regularly arranged cells and rare trichomes (Figures 4A,D). The epidermis of the rudimentary glumes displayed irregularly arranged cells and bore numerous small protrusions and trichomes (Figures 4A,E). No obvious vascular bundles were observed in the rudimentary glumes of the wild type (Supplementary Figure 2E). We also found that the stigmas of the wild-type pistil were sparse and the pistils had normal ovaries (Figures 4F,G and Supplementary Figures 2B,F,G). By contrast, the sterile lemmas and rudimentary glumes were elongated in the dg1 mutant (Figures 4H–J). The sterile lemmas and rudimentary glumes had more vascular bundles, big protrusions and trichomes, and inward hook-like structures (Figures 4K,L and Supplementary Figures 2H–K), which was similar with that of the lemma in the wild type (Figure 4B and Supplementary Figures 2A,C). In addition, the dg1 mutant displayed denser stigmas and an abnormal ovary (Figures 4M,N and Supplementary Figures 2B,G).

To clarify the identities of the floral organs, we examined the expression of related marker genes in the dg1 spikelets. The expressions of OsMADS1, OsMADS14, OsMADS15, and DROOPING LEAF (DL) was detected in the sterile lemmas and rudimentary glumes of the dg1 spikelets, implying that the sterile lemmas and rudimentary glumes were converted to lemma-like organs (Figure 5A). No OsMADS6 transcript was found in the sterile lemmas and rudimentary glumes of the wild type and dg1 spikelets, suggesting that the sterile lemmas and rudimentary glumes did not acquire the palea identity (Figure 5A). The expression of OsMADS2, OsMADS4, and OsMADS16 was lower in the dg1 stamens (Figure 5B), and the expression of OsMADS3, OsMADS58, and DL was lower in the dg1 pistils (Figure 5C), which was consistent with the phenotype defects in the dg1 mutant. These results revealed that the glumes of the dg1 mutant partly acquired the lemma identity, and the identities of stamens and pistils were altered.

We examined spikelet development in the dg1 mutant by SEM. The wild-type spikelets formed sterile lemma and rudimentary glume primordia, and the palea and lemma primordia were growing at the spikelet 4 stage (Sp4) (Figure 6A). At the Sp5 and Sp6 stages, six stamen primordia were observed in the wild-type spikelet and the growth of the stamen primordium on the lemma side was delayed (Figure 6B). Also at this stage the rudimentary glume ceased growing and the sterile lemma continued growing (Figures 6B,C). During the Sp7 and Sp8 stages in the wild-type spikelet, the pistil primordium was observed and the sterile lemma was further enlarged and was much longer than the rudimentary glume (Figures 6C,D). By contrast, no obvious differences were detected in the rudimentary glume and sterile lemma of the dg1 spikelets at the Sp4–Sp6 stages (Figures 6E,F). At the Sp7 and Sp8 stages, the rudimentary glume appeared normal in the dg1 spikelets, while the sterile lemma was longer (Figures 6G,H). We concluded that the dg1 glumes continue growing after the Sp8 stage, thus producing further elongated glumes at heading stage, and the rudimentary glume develops at later stages. Additionally, the development of the dg1 stamens was not synchronous and other floral organs appeared normal in the dg1 florets. These results revealed that the enlargement of glumes was strongly affected and the development of stamens was disrupted in the dg1 mutant.

At maturity, the dg1 mutant had shorter and smaller panicles (Supplementary Figure 3). Compared with the wild type, the panicles were 33% shorter in the dg1 mutant (Supplementary Figure 3). The number of spikelets from the dg1 mutant was only 66% of that of the wild type (Supplementary Figure 3). The pollen viability of the stamens was 52% and the seed-setting rate was 45% in the dg1 mutant (Supplementary Figures 3, 4), compared with 96 and 85% in the wild type, respectively (Supplementary Figures 3, 4). This suggested that the low seed-setting rate in the dg1 mutant may be due to defective pollen grains. The wild-type grains and brown rice averaged 7.1 and 5.4 mm long, and 3.5 and 3.0 mm wide, respectively (Figures 7A–E and Supplementary Figure 3). The dg1 grains were longer and thinner than those of the wild type (Figures 7A,D,E and Supplementary Figure 3). The dg1 grains and brown rice averaged 7.9 and 5.6 mm long and 3.3 and 2.8 mm wide, respectively (Supplementary Figure 3). Compared with the wild type, the 1,000-grain weight was not affected in the dg1 mutant, but the weight of 1,000-brown rice from the dg1 mutant was decreased markedly (Supplementary Figure 3), suggesting that the proportion of hulls in the dg1 grains was higher than that of the wild type, which supports the ideal that the hulls can affect grain yield. We also examined the expression of genes involved in the regulation of grain size. Compared with the wild type, BG1, BG2, GS3, GW2, GW8, and TGW6 were down-regulated in the dg1 mutant (Figure 7F). These data supported the phenotypic observations and the ideal that DG1 may affect grain size by negatively regulating the related genes expression.

To investigate the abnormal grains in the dg1 mutant at the cellular level, we performed histocytological analysis, including SEM and paraffin section. The observations of the outer epidermal cells of the lemmas showed that the cells were compressed longitudinally in the dg1 mutant (Figures 8A,B,E,F). However, the total cell number and the number of cells per millimeter along the longitudinal axis of the lemmas from the dg1 mutant were significantly increased compared to the wild type (Figures 8B,F,I,J), resulting in longer grains in the dg1 mutant. Cross sections of central parts of the hulls revealed that the dg1 spikelet contained shorter length of total cells and smaller cells than the corresponding region in the wild type (Figures 8C,D,G,H,K,L); this differences likely were responsible for the slender grains observed in the dg1 mutant. Taken together, our results implied that DG1 controls grain size by regulating cell number and size.

Because cell number and size were altered in the dg1 mutant, we next investigated the expression of several genes that regulate cell cycle and cell expansion in rice. The expressions of eight cell cycle-related genes was altered. Compared with the wild type, CYCD3, CDKA2, and CYCB2.1 were up-regulated in the dg1 mutant, and CAK1A, CYCT1, MCM4, MCM5, and CDC20 were down-regulated in the dg1 mutant (Figure 8M). The expression of three cell expansion-related genes (OsEXPA2, OsEXPA32, and OsEXPB5) was reduced in the dg1 mutant compared to the wild type (Figure 8M). These results were consistent with the phenotypic observations and implied that DG1 mainly affects grain size by regulating cell proliferation and cell expansion.

To survey whether the dg1 mutant phenotype is controlled by multiple genes or a single gene and whether it is dominant or recessive, we performed the reciprocal crosses between the dg1 mutant and two indica cultivars, NJ6 and TN1 to obtain F₁ and F₂ plants (Table 1). The F₁ offspring of all combinations showed a normal phenotype (Table 1), indicating that the dg1 mutation is recessive. In the four F₂ populations, the segregation ratio of the wild type to dg1 phenotype was 3:1 (Table 1), indicating that the mutated phenotype is controlled by a single, recessive nuclear gene.

To map the locus responsible for the dg1 mutant phenotype, we selected the cross of dg1 mutant and NJ6. We then tested 312 pairs of SSR primers evenly distributed on the 12 rice chromosomes and found that 165 of these markers showed polymorphisms between the two parental lines. Using the polymorphic markers, we screened DNA pools from 15 wild type individuals and 15 recessive dg1 individuals and compared the pools with the parental lines. Two markers, C2 and C21 on chromosome 8, had the same amplified fragment length in the recessive DNA pool and the dg1 mutant, and the fragment length differed from that of the wild-type DNA pool. Using this small F₂ population (70 plants), the mutated locus was primarily mapped to chromosome 8 between markers C2 and C21 (Figure 9A).

To further access the dg1 locus, we designed a large number of SSR and InDel markers flanking C2 and C21. The markers C5, C7, C8, C9, C17, C18, In10, In11, In13, In14, and In15 exhibited polymorphisms between the two parents (Figure 9A). Using 518 recessive individuals, the DG1 gene was finally delimited between InDel markers In13 and In14, an approximately 31-kb physical distance in Nipponbare (Figure 9B). In the 31-kb target region, there were eight predicted opening reading frames (ORFs) (Figure 9B). ORF1 encodes an acyl CoA binding protein that was reported to be related to plant growth and stress tolerance. ORF2 encodes a ribosomal protein and ORF3 and ORF4 encode retrotransposon proteins. ORF5 encodes an expression protein and ORF6 encodes a nuf2 family protein. ORF7 and ORF8 encode an Mps one binder kinase activator-like 1A and a DNA polymerase, respectively. We next sequenced all the annotated genes, but no nucleotide changes were detected between the wild type and dg1 mutant. We also investigated the expression of all annotated genes in the leaf and panicles by qRT-PCR. Compared with the wild type, only LOC_Os08g06550 (ORF1) showed a lower level of expression in the dg1 mutant; no differences were detected in the other genes between the wild type and dg1 mutant (Figure 9C). More work is needed to identify the gene responsible for the dg1 mutant phenotype.

Since the dg1 mutant has defects in spikelet development, we examined whether DG1 is involved in the regulation of genes associated with floral development in rice. We detected several MADS-box and non-MADS-box genes including A-class genes (OsMADS14, and OsMADS15), B-class genes (OsMADS2, OsMADS4, and OsMADS16), C-class genes (OsMADS3 and OsMADS58), E-class genes (OsMADS1, OsMADS6, and OsMADS34), and non-MADS-box genes (DL, G1, SNB, IDS1, and MFS1). Except for OsMADS6 and OsMADS34, most of the tested MADS-box genes showed lower expression levels in the dg1 mutant (Figure 10A). Lower expression of OsMADS1, OsMADS14, and OsMADS15 may be attributed to abnormal heading and short panicles in the dg1 mutant. Abnormal expression of OsMADS2, OsMADS4, OsMADS16, OsMADS3, OsMADS58, and DL likely caused the defective stamens and pistils in the dg1 mutant.

We also observed more abundant transcripts of G1 and OsMADS34 which restrained the elongation of glumes (rudimentary glume and sterile lemma) (Figure 10B) and fewer transcripts of SNB and IDS1 which positively regulated fate of glumes in the dg1 mutant (Figure 10B). These results supported the phenotypic observations and suggested that DG1 negatively regulated the identities of glumes, resembling the functions of G1 and OsMADS34. Taken together, our findings revealed pleiotropic roles of DG1 in the floral development and DG1 regulated the expression of these genes associated with floral development.