Plant materials, growth conditions and agronomic trait investigations

The rice cultivar Oryza sativa Tainung 71, a japonica-type aroma rice bred by scientists at the Taiwan Agricultural Research Institute (TARI), was used as the wild type in the study. The seeds of wild-type and various osga3ox1 mutants were germinated in a growth chamber at 30°C with a 16 h/8 h light/dark cycle for 14 days and then transferred to a greenhouse or an isolated paddy field at TARI.

The agronomic traits, including the plant height, heading date and yield-related traits, such as panicle numbers/per plant, grain numbers/per panicle, seed-setting rate (SSR) and thousand-grain weight (TGW), were measured and collected for statistical analysis. Plant height: The plant heights were measured at the yellow-ripening/mature stages. Grain numbers/per panicle: total number of seeds per plant divided by the panicle number/per plant to obtain the averaged grain number/per panicle. Seed-setting rate (SSR): the ratio of fertile seeds vs total seeds per plant. Thousand-grain weight (TGW): the weight of 1000 fertile seeds.

Plasmid construction and rice transformation

The sgRNA of OsGA3ox1 was designed using E-CRISPR [22] (http://www.e-crisp.org/E-CRISP/index.html), and the location and DNA sequence of the sgRNA are shown in S1A Fig. Two complementary sgDNA sequences with 5’ added sequences (GGCA for the forward sequence and AAAC for the reverse sequence) (S1 Table) were synthesized, annealed by a thermal cycler and integrated into the BsaI-cut CRISPR/Cas9 vector pRGEB32 [23]. After ligation and E. coli transformation, constructs were screened by BsaI restriction digestion and verified by sequencing. The constructed plasmid vectors were transformed into Agrobacterium tumefaciens strain EHA105 for rice plant transformation as previously described [24].

Regeneration of transgenic rice and selection of gene-edited plants

Transgenic plants containing the T-DNA vector construct were obtained through 50 mg/L hygromycin selection, and the transgene was confirmed by PCR analysis of the construct and/or hygromycin phosphotransferase (hpt) (S1B and S1C Fig). Regenerated T0 transgenic plants from different calli were assigned different line numbers.

For the selection of gene-edited plants, we modified the polyacrylamide gel electrophoresis-based (PAGE) method previously described [25] and used the modified method to select candidates of gene-edited plants. Briefly, the target region of sgRNA was PCR amplified, and the ideal size of the PCR product used in this method was approximately 300 to 400 bp. The PCR products from transgenic plants were subjected to 3% agarose gel electrophoresis for an hour at 100 voltages. After electrophoresis, the PCR product from transgenic plants was isolated into homoduplexes (consistent or smaller than expected size) and heteroduplexes (larger than expected size) (S1C Fig), which were used to identify insertion/deletion (IN/DEL) events from transgenic plants and infer that the target gene in analyzed plants is homozygous or heterozygous [25].

To identify frameshift mutations, PCR products with clear IN/DEL events were subjected to DNA sequencing, and the sequences in PCR products were decoded via CRISPR-ID [26] (http://crispid.gbiomed.kuleuven.be/). Once putative frameshift mutations were found, the PCR products were cloned into the pGEM®-T Easy Vector (Promega, Madison, WI, USA), and ten colonies from the respective PCR products were sent for DNA sequencing for sequence confirmation.

Pollen viability assay

The pollen viability assay was conducted as described previously [8]. Before flowering, anthers in a spikelet were collected and pollens removed from anthers were evenly placed on a glass slide, stained with staining solution (1% (v/v) of I₂ in 3% (v/v) KI), and then examined using a light microscope. Twenty views (12 mm²/each) for the respective samples were selected for counting. The black-stained pollen is referred to as viable pollen.

RNA extraction and gene expression analysis

Mature rice anthers (approximately 1.5 to 2.0 mm) from wild-type and osga3ox1 mutant plants were used for RNA extraction. Total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA), and RNA samples were treated with RNase-free Dnase I (Thermo Fisher Scientific, Waltham, MA, USA) to remove possible DNA contaminants. For reverse transcription, 1 μg of total RNA was used in cDNA synthesis using a RevertAid First-Strand cDNA Synthesis Kit (Thermo Fisher Scientific, Waltham, MA, USA), and the reaction was conducted following the protocol of the manufacturer in a 20 μL reaction volume. The quantitative reverse transcription PCR (RT-qPCR) was conducted as previously described [27] using one microliter of synthesized cDNA with the respective primer sets (S1 Table). The expression levels were normalized using the threshold cycle (C_t) values obtained for the reference gene OsActin, and relative quantification was conducted by the 2^-ΔΔCt method [28].

Semithin and ultrathin sections of rice anthers and pollen grains

Mature rice anthers (approximately 1.5 to 2.0 mm) were fixed in 1% glutaraldehyde in 0.1 M phosphate-citrate buffer, pH 7.2, in a microwave tissue processor, Pelco BioWave Pro (TED PELLA, Inc., Redding, CA, USA) at a microwave power of 250 watts for 1 minute twice. After three buffer rinses (at 250 watts for 40 seconds; hereafter the buffer wash was under this condition), the samples were postfixed in 1% OsO4 in the same buffer for 1 minute at 100 watts followed by three washes of buffer. Samples were then dehydrated in an ethanol series under the condition of the power of 100 watts for 40 seconds each step. Subsequently, samples were embedded in a series of Spurr’s resin at room temperature, and polymerized blocks were sectioned with a Lecia Reichert Ultracut S or Lecia EM UC7 ultramicrotome. The semithin sections (1 micrometer) were obtained and stained with 1% toluidine blue, and images of the sections were taken by a Zeiss fluorescence microscope Z1 equipped with a Zeiss Axiocam HRc CCD digital camera (MICRO-OPTICS, Fresh Meadows, NY, USA). The ultrathin sections (70~90 nanometers) were stained with 5% uranyl acetate in 50% methanol and 0.4% lead citrate in 0.1 N NaOH. Images were taken using an FEI G2 Tecnai Spirit Twin transmission electron microscope at 80 kV equipped with a Gatan Orius CCD camera.

Measurement of starch granules

The numerical features of starch granules and pollen grains were quantified by counting and measuring starch granules and pollen grains on histological images of semithin rice anther sections taken by a Zeiss fluorescence microscope Z1 equipped with a Zeiss Axiocam HRC CCD digital camera. Quantification of pollen grains and starch granules within pollen grains was automated using a Cellprofiler ver. 2.4.0rc1 pipeline [29]. The basic steps of the pipeline were the segmentation of objects, including pollen grains and starch granules, identification of objects, and quantification of objects. The outline of the pollen grains was extracted by using duplicated mask images of pollen grains on histological images. The outline of the starch granules on sections was extracted by enhancing brightness and contrast. The pollen grains and starch granules were then segmented using an adaptive local thresholding method based on ellipse fit [30], and identified objects were counted and measured.

Measurement of exine and intine thickness

The exine and intine thicknesses of the pollen wall were measured using Fiji [31]. At least 5 images of pollen grain thin sections and 3 to 60 sampling sites from each image were selected for measurement. Each sampling site was evenly distributed, and the thickness of the exine was measured between two flat lines to avoid spikes (see the bracket indicated in Fig 4); the thickness of the intine was measured from the base to its peak (as indicated by the blue line in Fig 4).

Statistical analysis

The statistical comparisons were performed using one-way ANOVA and then against the WT control by Student’s t test and graphing through the GraphPad Prism 8 algorithm (GraphPad Software Inc. La Jolla, CA, USA). The distributions of each sample dot are shown. Values are the mean±SD. The significant differences are indicated by P values: *p < 0.05, **p < 0.01, ***p < 0.001.

Creation of OsGA3ox1 mutants

To study the biological functions of OsGA3ox1 in rice, the plasmid pRGEB32 containing an sgRNA, which targets the first exon of OsGA3ox1 (S1A and S1B Fig), and the CRISPR/Cas9 construct was used to create loss-of-function transgenic plants. More than 10 transgenic rice lines (T0 plants) were successfully regenerated via Agrobacterium-mediated transformation. The transgenic rice lines showing targeted gene editing were screened through a modified gel-based screening method. Among these T0 transgenic lines, 8 independent lines with different editing types, revealed by the various PCR band patterns, were selected for further characterization (S1C Fig). Their DNA sequences at the OsGA3ox1 gene target site were sequenced, and 7 lines of different combinations of edited sequences were identified (S1D Fig).

Except for the “-3” allelic variant (a 3-bp deletion), the amino acids derived from other edited allelic variants showed frame-shift and premature stops (S1E Fig), resulting in knocked-out function of the OsGA3ox1 gene. Two lines (#4 and #5) showed biallelic “-1/-2” and “-1/-4” variants that had both allelic genes knocked out and produced no seeds (S1C Fig). Another two lines, #7 (-3/-16/?) and #8 (-3/-31/?), each contains the “-3” allele, a knockout allele and a not-yet-identified allelic variant could produce fertile seeds, but due to its complications in editing, no further characterization was pursued. The remaining four heterozygotic biallelic lines, line #1 (“-3/-19”), lines #2 and #3 (“-3/-2”) and line #6 (“-3/-10”), which also contained the “-3” allele, could produce fertile seeds and were used for further phenotypic and molecular analyses to investigate the function of OsGA3ox1.

Mutation of OsGA3ox1 affects the seed-setting rate but not plant height

Although the “-3” allele-containing heterozygotic lines could produce T1 seeds, their seed-setting rate (SSR) was significantly reduced (S1C Fig). To confirm that the low SSR was due to the loss of function of OsGA3ox1, T1 progenies from line #1 (“-3/-19”) were first grown in a greenhouse and characterized. The “-3/-19” variant segregated into “-3/-3” homozygous, “-19/-19” homozygous and “-3/-19” heterozygous plants that were recognized by the PCR band patterns and confirmed by sequencing (S2A Fig). Our results showed that the plant heights were not different among these three genotypes; however, their SSRs were reduced, and the “-19/-19” knockout mutant produced no fertile seeds (S2B Fig).

To further confirm these observations, the T2 progenies segregated from the biallelic “-3/-19” plants, and T1 progenies segregated from biallelic “-3/-2” (line # 3) and “-3/-10” (line # 6) lines were grown in the paddy field for characterization. Genotypes of segregated progenies were identified by sequencing (Fig 1A–1C). All different variants/genotypes showed the same plant height as WT (Fig 1D–1F), similar to the plant height observed in the greenhouse. However, all homozygotic knockout variants (“-19/-19”, “-2/-2” and “-10/-10”) showed unfertilized panicles and produced few fertile seeds (Fig 1G, the fertile seeds are circled). The SSRs of “-19/-19”, “-2/-2” and “-10/-10” were greatly reduced to only ~2.8%, 3.4% and 2.2%, respectively (Fig 1H and 1I).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Characterizations of Cas9-induced osga3ox1 “-3/-19”, “-3/-2” and “-3/-10” biallelic variants and analysis of their progenies.

A) to C) PCR product analysis of progeny plants derived from three Cas9-induced osga3ox1 variants. Eight representative progeny plant samples from each of the three biallelic variants are shown. The progeny genotypes derived from “-3/-19” (A) and “-3/-2” (B) and “-3/-10” (C) were confirmed by sequencing and are indicated below the gel samples. The PCR primers used for construct and target gene detection are provided in S1A and S1B Fig and S1 Table. Genotypes are presented as sequence I/D (I, insertion “+” or D, deletion “-”; WT, wild-type plant or gene). Sequence variations indicated by Arabic numerals separated by a slash are used to represent the diploid allelic genotypes. D) to F) Representative matured plants of different genotypes segregated from three Cas9-induced osga3ox1 variants. The knockout variants “-19/-19” (D), “-2/-2” (E) and “-10/-10” (F) showed unfertilized erected panicles. Their average plant height and relative plant height (RPH) are listed below. Sample size n≧5. G) Matured panicle comparison in WT and different knockout variants. The very few fertilized seeds in the knockout variants are circled. H) and I) Comparisons of the seed-setting rates in WT and three Cas9-induced osga3ox1 variants. The data of H) and I) were collected from different crop seasons with different sample sizes (H, n = 5; I, n = 3). The significant differences between WT and variants were determined by Student’s t test. *p<0.05, **p<0.01, ***p<0.001, NS; no significant difference.

https://doi.org/10.1371/journal.pone.0292400.g001

Detailed agronomic traits of the WT host and progenies segregated from the transgene-free “-3/-19” variant/line were compared, and the plant height, panicle number and grain number/per panicle were the same for WT and all mutant genotypes (Table 1). For the SSR, “-3/-3” was similar to the WT, “-3/-19” was slightly lower than the WT, and the “-19/-19” knockout mutant showed a significant reduction. The 1000-grain weight and grain yield per plant in the knockout mutant were significantly affected (Table 1). Taken together, our findings indicate that OsGA3ox1 plays a critical role in the development of seeds but not in the development of vegetative organs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Agronomic trait comparisons between TNG71 and progenies segregated from the biallelic “-3/-19” variant.

https://doi.org/10.1371/journal.pone.0292400.t001

Mutation of OsGA3ox1 affects the pollen number and pollens were stained lighter in osga3ox1 mutants

To determine whether the low SSR was due to defects in any gametophytic features, the flower organs of the three genotypes segregated from the progeny of the transgene-free “-3/-19” variant were examined. Unlike the abnormally developed floral organs observed in GA-deficient mutants reported previously [9, 32, 33], the anthers and pistils in the three genotypes showed no difference from WT (Fig 2A). However, the total pollen and the pollen viability by IKI staining in the knockout mutant “-19/-19” were significantly reduced (Fig 2B). The numerical quantification data of the collected samples showed that the pollen numbers and the pollen viability of “-3/-19” and “-19/-19” were approximately 72.8% vs 42.6%, and 79.1% vs 51.3%, respectively, compared to the WT (Fig 2C and 2D), whereas no significant difference between “-3/-3” and WT was observed (Fig 2C and 2D). These observations suggest that OsGA3ox1 affects pollen development and reduces SSR, but does not have any discernable effect on other floral organs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Phenotypes of floral organs and characterization of pollen viability and pollen number analysis in the “-3/-19” variant.

A) Comparisons of floral organs. Flower samples from WT and different genotypes segregated from the “-3/-19” biallelic variant are shown. No phenotypic difference was observed. Bar = 1 mm. B) Comparisons of pollen viability. Representative views (2 mm²) for each genotype are shown. The pollen viability was based on KI starch staining. Healthy pollen is stained black. Bar = 150 μm. C) Comparisons of total analyzed pollen numbers. Twenty views (12 mm²/each) of pollen slides from different genotypes were sampled. The graphic bar of each genotype shows the total analyzed pollen numbers that were accumulated by each of their 20 samples. D) Comparisons of the pollen viability/stained pollens in WT and mutant genotypes. The significant differences between WT and variants were determined by Student’s t test. *p<0.05, **p<0.01, ***p<0.001; NS, no significant difference. (n = 20).

https://doi.org/10.1371/journal.pone.0292400.g002

While the SSR of the knockout mutants were dramatically decreased (Fig 1H), a high proportion (~51.3%) of the stained pollen were observed in the present study (Fig 2D) which differs from the low proportion (~5%) reported previously [21]. This variation could be due to the different rice cultivars (TNG71 vs Nipponbare) and culture environments (Taichung Taiwan 24°0′N 120°7′E vs Nagoya Japan 35°2′N 136°9′E). Furthermore, besides a few heavy stained pollen, most pollens in the knockout mutant were stained lighter than WT (Fig 2B) and these pollens were counted as stained pollen even though they might not be viable. Additionally, the total pollen numbers in osga3ox1 mutant were dramatically decreased to ~42.6% of the WT (Fig 2C), and together with these less viable pollens, low SSR was observed in our knockout mutants.

Unlike both of the GA deficiency mutant rpe1 and GA signaling mutant Slr1-d3 that impaired the pollen viability but not the number of pollen grains [8], the knockout mutant osga3ox1 reduces the total number of pollen to ~42.6% of the WT (Fig 2B and 2C). Although OsGA3ox1 is mainly expressed in anthers at the late stage of pollen development, its expression could also be detected in the tapetum at the early stage of pollen development [16, 20]. The expression of OsGA3ox1 has been observed in anthers from the meiosis to the tricellular stages, increasing gradually and peaking at the binucleate stage, then declining at the tricellular stage [14]. In the present study, the tapetum layer in osga3ox1 and WT were all degenerated in the mature anther (Fig 3A), but we could not rule out the possibility that the bioactive GAs required for early pollen development could partly contributed by OsGA3ox1. Therefore, the bioactive GAs in osga3ox1 mutant might be reduced to retard the progression of the tapetum program cell death (PCD), causing less nutrients for early pollen development and resulting in early pollen degeneration, which leads to pollen number reduction in osga3ox1 mutant (Fig 2C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The pollen grain phenotypes, sizes and starch granule distributions in “-3/-19” variants.

A) Phenotypical pollen grain comparisons in anther microscopic sections. Anther samples were from WT, and different genotypes segregated from the “-3/-19” heterozygous/biallelic variant. Two types of pollen grains, “-3”-like (blue arrow) and “-19”-like (red arrow), observed in anthers of the “-3/-19” genotype are indicated. Bar = 20 μm. B) TEM sections of pollen grains from WT, “-3” and “-19”. Different types of pollen grains from “-19” are presented to show various amounts of starch granules. Bar = 5 μm. C) Comparisons of pollen grain sizes. Sizes were measured on the surface areas of the largest pollen section images available from each genotype. Sample sizes were n = 13, 13 and 28 for WT, “-3” and “-19”, respectively. The significant differences between WT and mutant genotypes were determined by Student’s t test. *p<0.05, **p<0.01, ***p<0.001; NS, no significant difference. D) Quantification of the area ratio of starch grains to pollen grains from anther samples of WT and different genotypes. Sample sizes were n = 13, 16, 18 and 28 for WT, “-3/-3”, “-3/-19” and “-19/-19”, respectively. The sample dot distribution is shown in the plot.

https://doi.org/10.1371/journal.pone.0292400.g003

When analyzing progeny segregation of the heterozygous variants, we always observed a very low ratio of the gene knockout plants in the populations. An example of 24 progenies with a segregation ratio of 11:11:2 = “-3/-3”:”-3/-19”:”-19/-19” was observed from the progeny of the “-3/-19” heterozygous variant (S3 Fig), which did not follow the 1:2:1 Mendelian segregation ratio. This low transmission frequency of the osga3ox1 knockout plants in their progeny population could be due to its low pollen viability. It is similar to the observation of GA-deficient mutants, such as cps, ks, ko2 and kao of GA biosynthesis genes, and their mutation effects are transmitted to progeny in a gametophytic manner that affects their transmission frequency [8].

Mutation of OsGA3ox1 affects pollen starch granule accumulations

To pursue the cause of low pollen viability in osga3ox1 mutant, we investigated the anatomic features of mature anthers from different genotypes. We found no difference in sporophytic features, including epidermal, endothecium and tapetum cells of anthers (Fig 3A and S4B Fig). Interestingly, there was a significant difference in the gametophytic development of pollen (Fig 3A). In contrast to WT and “-3/-3” variants, which had larger pollens and a great amount of starch granule accumulation, the pollens from the homozygotic knockout variant “-19/-19” were smaller and had less starch granule accumulation (Fig 3A and 3B), and pollens which contain small amounts of starch granules were lightly stained (Fig 2B). The quantification data showed that the pollen collected from the knockout variant had its size reduced to ~ 70% - 80% of the size of the WT (Fig 3C).

Another observation was that both the “-3” and “-19” types of pollen can be observed in the “-3/-19” variant (Fig 3A). The area ratio of starch granules to a pollen grain in different genotypes was measured to indicate the amount of starch granule accumulation. Relatively high levels of starch accumulation (area ratio of 0.3 and above) were observed in the pollen of WT and “-3/-3”, while the “-19/-19” genotype revealed an area ratio of less than 0.1. The starch accumulation ratio in pollen collected from the “-3/-19” genotype (Fig 3D) had a distribution from 0 to ~ 0.5 and formed two subgroups, in which the higher subgroup was equivalent to the group of WT and “-3/-3” and the lower subgroup was similar to the group in the “-19/-19” genotype (Fig 3D), indicating that both normal and abnormal pollen could be generated in the same anther sac. The above phenotypic observations in the “-3/-19” variant were also found in the “-3/-10” and “-2/-2” variants (S4 Fig). The pollens from “-2/-2” variant contain less starch granule and are smaller, the same as those from “-19/-19”. The pollens produced from “-3/-10” variant contain two types of pollens, the same as those generated from “-3/-19” variant (S4 Fig). These data support that the defect in pollen development was indeed caused by the defect of OsGA3ox1.

Unlike pollen development in the GA-deficient mutant oscps1-1, which showed an enlarged tapetum layer and collapsed microspores [7], no morphological difference in the tapetum layer was observed in osga3ox1 (Fig 3A). Our findings suggest that mutation of OsGA3ox1 does not inhibit the tapetum PCD, but is involved in the regulation of starch synthesis during late pollen development. This notion is supported by the fact that OsGA3ox1 is expressed mainly in the late stage of pollen development [14, 16, 20] and the accumulation of pollen starch granules can be enhanced in pollens of the GA₃-treated almond anther [34]. Therefore, OsGA3ox1 is suggested to be expressed in the pollen grains and is involved in late pollen development, mainly affecting starch synthesis (Fig 3).

OsGA3ox1 affects pollen wall development and starch synthesis

The rice pollen wall is elaborated and contains different substructures, including exine (outer layer), tryphine (in the cavities of exine) and intine (inner layer). Exine is further divided into two components, sexine (tectum and baculum) and nexine. Since GA has been proposed to regulate exine formation and GA-deficient mutants showed defective pollen cell walls [7], we compared the pollen wall ultrastructure between mutant osga3ox1 and WT. From TEM image observations, the knockout mutant “-19/-19” revealed a thinner exine, but the thickness of the intine was similar to those of “-3/-3” and WT (Fig 4A). The quantification data measured from 30 to 186 samples showed that the exine thickness in the knockout mutant was ~ 81 to 89% of that in the WT, the reduction being statistically significant (Fig 4B), but no significant difference was observed in the thickness of the intine (Fig 4C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The ultrastructure of pollen walls and expression analysis of pollen development genes in osga3ox1 mutants.

A) The ultrastructure, including the exine and intine, of the pollen walls from WT, “-3” and “-19” are indicated. The structure of exine including tectum (Te), tryphine (Ty), baculum (Ba) and nexine (Ne) are indicated in WT. The abnormal formation of baculum (blue arrow) and tryphine (red arrow) in “-19” pollen are pointed. Bar = 500 nm. B) Comparison of exine thickness in WT, “-3” and “-19”. Sample sizes were n = 186, 53 and 30 for WT, -3 and -19, respectively; sample dot distribution is shown in the plot. C) Comparison of intine thickness in WT, “-3” and “-19”. Sample sizes were n = 90, 27 and 15 for WT, -3 and -19, respectively; sample dot distribution is shown in the plot. D) Expression analysis of the pollen development genes in WT and the ga3ox1 knockout mutant. The significant differences between WT and variants were determined by Student’s t test. *p<0.05, **p<0.01, ***p<0.001.

https://doi.org/10.1371/journal.pone.0292400.g004

In addition to the thinner exines, the bacula in the knockout mutant is not well organized, showing either fewer or incompletely developed bacula with abnormally connected tryphines (Fig 4A and S5 Fig, blue arrows). While the tryphines in the WT and “-3” pollen walls are well arranged with smooth boundaries, the tryphines in the mutant show distorted and wiggly structures (Fig 4A and S5 Fig, red arrows). The ultrastructure of the “-19”-like pollen from “-3/-19” anther was phenotypically similar to that of the “-19” pollen from “-19/-19” anther (S5 Fig).

The exine and tryphine components such as sporopollenin, fatty acid and long-chain fatty acid derivatives are from the degradation of the sporophytic tapetum cells during pollen development [1]. Therefore, the exine development could be affected by the degeneration of tapetum cells in a sporophytic manner as in the mutation of GAMYB [7], which affects tapetum degradation and results in no pollen grain development. However, the thinner exine in the “-19” pollen (Fig 4A and 4B) can be attributed to the defective expression of the mutated osga3ox1 gene in the “-19” pollen, but not in the “-3” pollen, by a gametophytic manner (Fig 3A, the “-3/-19” variant). The phenomena of the partially affected thinner exine (Fig 4A and 4B) and the various degrees of starch granule reduction in the “-19” pollen (Fig 3B and 3D) imply that the GA biosynthesis was significantly downregulated by the mutation of osga3ox1, but not completely absent, owing to the residual minor expression of other GA biosynthesis genes [14, 16, 20], such as OsGA3ox2. Moreover, the thickness of exine of almond pollen can be enhanced by GA₃-treatment [34]. These lines of evidence support that the exine formation is modulated by the availability of GA and also GA regulation, and the thinner exine in “-19” pollen was caused by the mutation of OsGA3ox1 (Fig 4A and 4B), which reduced the amount of bioactive GA in the “-19” pollen at the gametophytic phase.

In term of starch synthesis, a recent study proposed that the cell wall invertase OsCIN3, which was not properly retained in the cell wall, might affect pollen development in osga3ox1 [21]. OsINV4 (same as OsCIN3, LOC_Os04G33720) was expressed in mature pollen grains, and downregulation of OsINV4 disrupted starch formation in pollen [35]. We therefore propose that improper retention of OsCIN3/OsINV4 in osga3ox1 might be due to structural changes in the exine of pollen walls, which caused impaired starch synthesis and finally led to defective accumulation of starch granules in osga3ox1 pollen.

Expression of OsGAMYB is downregulated in anther of the osga3ox1 mutant

To explore the molecular regulation of pollen wall development, the expression of pollen development-related TF genes, such as OsGAMYB, OsTDR and OsbHLH142, in anthers of the osga3ox1 mutant was investigated. These TF-encoding genes are directly or indirectly regulated by GA and are crucial for pollen wall development [7, 12, 13]. GAMYB, a GA-regulated TF [36], is a positive regulator of OsTDR and OsbHLH142 [7, 11] and has been reported to regulate exine formation via activation of CYP703A3 [7]. In addition, the lipid transporter proteins OsC4 and OsC6 are important for the development of pollen exine [37, 38], and downregulation of OsC4 and OsC6 was reported in mutants of OsGAMYB, OsTDR and OsbHLH142 [7, 13, 39], suggesting that OsTDR and OsbHLH142 might also be involved in exine development. Our qPCR data showed that the expression levels of OsGAMYB, OsTDR and OsbHLH142 in osga3ox1 were significantly downregulated (Fig 4D and S6 Fig), which might result in the thinner exine in osga3ox1.

GAMYB is a pivotal TF involved in the regulation of almost all GA-regulated genes during pollen development and it functions mainly in tapetum [7]. However, the expression of GAMYB and its target gene CYP703A3 were also observed in microspores [7] suggesting that GAMYB and CYP703A3 are also involved in pollen development at the gametophytic phase. Given that GA regulates anther development and exine formation via GAMYB signaling [7], the reduction of GA biosynthesis in pollen of osga3ox1 mutant would downregulate the expression of GAMYB and possibly many of its downstream genes, such as OsTDR, OsbHLH142, OsCYP703A3, OsC4 and OsC6, resulting in abnormal exines in the mutant.