An epigenetically mediated double negative cascade from EFD to HB21 regulates anther development

Epigenetic modifications, which includes DNA methylation and histone modifications etc, play essential roles in plant development. DNA methylation (5-methylcytosine), catalysed by DNA methyltransferases, is an epigenetic modification conserved in eukaryotes. De novo DNA methylation refers to the methylation of cytosine (C5) in an unmethylated DNA strand. Since subsequent DNA methylation is based on the initial methylation on one of the DNA strands, the establishment of de novo DNA methylation is essential for successful whole-genome DNA methylation. To date, de novo DNA methylation has been reported to be established through the RNA-directed DNA methylation (RdDM) pathway in plants¹. DNA methyltransferases are responsible for the catalysis of de novo DNA methylation. In Arabidopsis, de novo DNA methylation is considered to be mediated by DRM1/2², which are the homologues of the mammalian de novo DNA methyltransferase, Dnmt3^3,4,5. The mutation of DRM1/2 causes the retention of the last intron in 28% of the transcripts of PRD2/MPS1, a meiotic gene, which leads to a small portion of polyad production⁶. Histone lysine methylation is another epigenetic modification required for gene expression. Among histone lysine modifications, the trimethylation of H3 lysine 27 (H3K27me3) is a repressive epigenetic mark localised in chromatin that preferentially marks the repressed genes. H3K27me3 is involved in various plant development processes^7,8,9, and it is usually associated with other epigenetic modifications such as DNA methylation. Loss of DNA methylation in Arabidopsis is reported to result in the redistribution of H3K27me3¹⁰. During anther development, H3K27me3 is involved in repressing the expression of several genes encoding master transcription factors that are involved in tapetum and pollen development¹¹. H3K27me3 is also essential for the vegetative cell fate commitment in pollen¹². However, there is still limited information about H3K27me3 repression of genes required for anther development.

In higher plants, the anther is a multilayered male reproductive organ that produces pollen. The anther wall can be divided, from outside to inside, into four diploid concentric cell layers: the epidermis, endothecium, middle layer and tapetum¹³. Pollen mother cells enclosed by these cell layers undergo meiosis to form tetrads which further develop into pollen. Mature pollen is covered by complex walls with a particular wall pattern to protect pollen from biotic or abiotic stresses, such as dehydration and sunlight¹⁴. The pollen mother cell determines the pollen wall pattern¹⁵. Moreover, the materials for the outer pollen wall are biosynthesized and secreted from the tapetum. After microspores develop into mature pollen grains, the anther dehisces in time to release mature pollen for pollination. The diploid endothecium plays a role in anther dehiscence. Anther dehiscence requires lignification of the endothecium cell layer¹⁶. Once released from the anthers, the pollen grains are then carried by wind or other pollinators to fall onto the stigma to achieve pollen hydration and germination¹⁴. Thus, the precise coordination of the different anther cell layers is essential for pollen formation and pollen release.

Exine formation defect (EFD) is a predicted protein with a typical SAM-dependent methyltransferase ___domain. Its knockout lines in Arabidopsis exhibit a defective pollen wall pattern and male sterility. Previous research has shown that after incubating with EFD, a DNA fragment cannot be digested by a DNA methylation sensitive restriction enzyme, suggesting that EFD has DNA methyltransferase activity in vitro¹⁷. In the present study, through complementation analysis, we showed that the efd mutant was rescued by DRM2, a de novo methyltransferase, and discovered epigenetic modifications in HB21, especially at the 3′ terminus of the coding region. EFD-HB21 regulates the expression of several genes involved in pollen wall pattern establishment (in pollen mother cells), pollen wall material synthesis (in the tapetum) and endothecium lignification for anther dehiscence (in the endothecium). Our study revealed an epigenetically mediated double negative cascade, EFD-HB21-anther genes that elaborately coordinates the pollen wall pattern determination, pollen wall material synthesis and anther dehiscence in different anther layers to ensure pollen formation and release during plant reproduction.

Expression of a de novo DNA methyltransferase complements the efd phenotype

To further understand whether EFD performs the function of a de novo DNA methyltransferase in Arabidopsis, we prepared a pEFD:DRM2 construct with the de novo DNA methyltransferase gene DRM2 driven by the EFD promoter and introduced it into the efd homozygous mutant (Fig. 1a). Of sixteen transgenic lines, eleven exhibited full fertility with normal siliques (Fig. 1b, 11/16). Alexander staining revealed a large number of mature pollen grains inside the anthers of these fertile transgenic plants (Fig. 1c; n = 11). The pollen grains of these lines were plump in shape, similar to that of the wild type (Fig. 1d, f; n = 30). The efd pollen surface was rugged and irregularly deposited with materials. The pollen grains of these transgenic plants exhibited a regular and reticulated pollen wall pattern similar to that of the wild type (Fig. 1e; n = 30). Expression analyses of CalS5, RPG1, CYP703A2 and DRM2 showed that their expression was restored to a similar level in the transgenic lines as in the wild type (n = 6; Supplementary Fig. 1a, b). These results indicate that DRM2 rescues both pollen defects and the expression levels of the downregulated genes in efd. It has been reported that CMT3 is a DNA methyltransferase required for maintaining DNA methylation¹⁸ (not de novo DNA methylation). We constructed and transformed pEFD:CMT3 into efd. All the transgenic plants failed to complement the male sterility of efd (n = 12; Supplementary Fig. 1c). In addition, we found that EFD can be detected in various anther cells (Supplementary Fig. 1f), while drm1 drm2 double mutant exhibited no significant difference in pollen production (Supplementary Fig. 1c). We then anaylsed the protein structure of EFD and DRM2. 3D-structure prediction and sequence analyses revealed similar structures in the canonical SAM-MT core area (Supplementary Fig. 1d). Based on the sequence alignment, several amino acids showed identity or similarity in this core area (Supplementary Fig. 1e). Taken together, these results illustrate that the expression of DRM2 restores the efd phenotype including pollen wall formation, supporting its role as a de novo DNA methyltransferase during anther development in Arabidopsis.

a diagram of the pEFD:DRM2 construct. b images of wild type (WT), efd and pEFD:DRM2. The pEFD:DRM2 plants exhibited obvious fertility restoration. c Alexander staining of pollen grains from the WT (n = 30), efd (n = 50) and pEFD:DRM2 anthers (n = 60). Bars = 100 μm. d SEM images of pollen grains of the WT (n = 32), efd (n = 50) and pEFD:DRM2 (n = 52). Bars = 20 μm. e SEM images of magnified pollen grains of the WT (n = 35), efd (n = 50) and pEFD:DRM2 plants (n = 50). The pollen of pEFD:DRM2 plants was plump with a regularly reticulated pollen surface similar to that of the wild type. Bars = 10 μm. f SEM images of dehiscent anthers with pollen grains from the WT (n = 10), efd (n = 25) and pEFD:DRM2 plants (n = 20). A few pollen grains were observed on the surface of the efd anthers. g BF staining of the endothecium cells from wild-type (n = 12), efd (n = 20) and pEFD: DRM2 anthers (n = 15). Bar = 20 μm.

Full size image

efd is defective in sporopollenin synthesis and endothecium lignification

During anther development, the tapetum provides the materials for pollen wall formation, while the endothecium is responsible for anther dehiscence (Fig. 2a). A previous study showed that efd is defective in pollen wall pattern formation¹⁷. During anther development, pollen wall formation is responsible for pollen formation and plant fertility. The outer wall sexine and lipid contents, and the inner wall intine can be stained red and purple by DIOC₂ and Tinopal, respectively¹⁹. We further analysed efd pollen wall defects through Tinopal/DiOC₂ staining. The weak red signal surrounding the efd pollen (Fig. 2b; n = 20) indicates that sporopollenin synthesis was defective in efd. To date, eight genes have been reported to be involved in sporopollenin synthesis²⁰. Quantitative reverse transcription PCR (RT-qPCR) results showed that only CYP703A2 was significantly downregulated in efd (Supplementary Fig. 1b), while the other genes tested exhibited similar expression levels to those of the wild type (Supplementary Fig. 2). Lignification of the endothecium layer is necessary for the anther dehiscence and pollen release²¹. efd was preliminarily shown to be defective in anther dehiscence¹⁷. To further investigate anther dehiscence and endothecium lignification in efd, anthers were stained with Basic Fuchsin (BF, red) solution which can reflect the lignification of endothecium cells. A red fluorescent signal was clearly observed in the wild-type endothecium, but the signal in most efd endothecium cells was significantly reduced (Fig. 2c; n = 20). NAC SECONDARY THICKENING PROMOTING FACTOR1 (NST1) and NST2 are two genes responsible for the endothecium lignification. The nst1 nst2 double mutant lines exhibited anther indehiscence and male sterility²². RT-qPCR results revealed that both NST1 and NST2 were significantly downregulated in efd (Supplementary Fig. 3). In addition, a red signal was observed in the lignified region in the endothecium of the pEFD:DRM2 transgenic lines (Fig. 1g; n = 10). Taken together, these results reveal that in addition to the defective pollen wall pattern formation, efd is also defective in sporopollenin synthesis and anther dehiscence.

a the pollen mother cell, tapetum and endothecium are three diploid sporophytic cells in the anther that provide the materials for both pollen development and anther dehiscence. At stage 6, callose and primexine material are provided by the pollen mother cell (2n) on its surface. At stage 7, in the tapetum (2n), sporopollenin precursors are synthesised and secreted into tetrads which are the fundamental material for the sexine of the pollen wall. At stage 11, the endothecium cells (2n) start to lignify, which is an essential process for anther dehiscence. b Tinopal/DiOC₂ staining of the pollen walls of the wild-type (n = 15) and efd plants (n = 20). Bars = 100 μm. c BF staining of the endothecium cells of the wild type (n = 10) and efd (n = 20). Arrow, the red fluorescent signal indicates the lignified area. En, endothecium cell layer. Bar = 20 μm. d gene rescue analysis and pollen staining of pNPU:RPG1, pAMS:CYP703A2 and pEFD:NST2 transgenic plants showed that pAMS:CYP703A2 can rescue the male sterility of efd.

Full size image

To further investigate which defect is responsible for efd fertility, we prepared three constructs, namely, pAMS:CYP703A2, pNPU:RPG1/SWEET8 and pEFD:NST2 with CYP703A2, RPG1/SWEET8 and NST2 driven by the tapetum-specific promoter pAMS, the microsporocyte promoter pNPU and the EFD promoter, respectively^17,23,24,25. These constructs were subsequently transformed into efd. Neither pNPU:RPG1 nor pEFD:NST2 rescued the fertility of efd (n = 18 and 20; Fig. 2d and Supplementary Figs. 4 and 8). However, the fertility of the pAMS:CYP703A2 transgenic line was significantly restored (n = 22; Fig. 2d). The pollen grains in the transgenic lines were normally produced (Fig. 2d). The exine of the pAMS:CYP703A2 lines were also evidently observed (Supplementary Fig. 4). These results reveal that sporopollenin synthesis is most essential for ensuring the male sterility of efd.

Overaccumulation HB21 leads to anther development defects in efd

DNA methylation usually represses gene transcription. As is expected, the direct target genes of EFD were upregulated in efd. On the other hand, several genes required for anther development and pollen formation including RPG1, CYP703A2, NST1 and NST2 were downregulated in efd (Supplementary Figs. 1b and 3). We propose that EFD might directly methylate some regulators that further affect the expression of these genes. An analysis of reported microarray data revealed that only one regulator, encoding the putative transcriptional factor HB21, was upregulated in efd¹⁷. The RT-qPCR results showed that compared with that in the wild type, the expression of HB21 in the efd buds was upregulated ~11.51-fold (Fig. 3a). To further investigate the details of HB21 upregulation during anther development, in situ hybridisation was performed using an HB21 probe. In efd, the signals of the HB21 transcript in pollen mother cells and tapetum were obviously stronger than those in the wild type during the early anther stages (stage 6 and 7; Fig. 3b).

a examination of the HB21 expression level in both wild-type and efd buds by RT-qPCR. Data are presented as mean values ±SD. The error bars represent the SDs from three independent biological repeats. Two-tailed Student’s t test was used for statistical analysis of the difference between the wild type and mutants, **p < 0.01 (the p-value is shown in the figure). b in situ hybridisation of HB21 transcripts in the wild type and efd during anther development. T, tapetum; M, pollen mother cell; S5–S11, Stage 5–11; eS6, early Stage 6; lS6, late Stage 6. Bars = 50 μm. c fertility observation of wild-type, efd, hb21 and efd hb21 plants. efd was sterile, while hb21 was fertile. The efd hb21 double homozygous line showed obvious restored fertility compared with that of the efd single mutant. Bars = 1 mm. d RT-qPCR identification of the pollen wall developmental related genes in efd hb21. These genes were downregulated in the efd buds. Data are presented as mean values ±SD. The error bars represent the SDs from three independent biological repeats. Student’s t test was used for statistical analysis of the difference between the wild type and mutants, ***p < 0.001 (all the p-values are shown in the figure).

Full size image

HB21 encodes a homeobox family (HD-Zip1) transcription factor. The HD-Zip1 family members are widely involved in the developmental processes of roots, stems, axillary buds and embryos^26,27,28,29. However, the function of the HD-Zip1 family in male reproductive growth has not been determined. HB21 was previously shown to be expressed in various plant tissues³⁰. Our RT-qPCR results showed that HB21 was expressed at low levels in roots and floral buds, and almost no expression was detected in leaves and stems (Supplementary Fig. 5). The HB21 gene has three exons and two introns. We obtained one T-DNA-tagged mutant line, CS857242, from the Arabidopsis Biological Resource Center (ABRC). The T-DNA was inserted in the third exon (Supplementary Fig. 6a). This line is a null allele line as HB21 transcripts were not detected (Supplementary Fig. 6b). The vegetative growth of the hb21 mutant was normal, and the mutant was fertile with normal pollen production (Fig. 3c and Supplementary Fig. 6c, d). The semithin sections and SEM observations both revealed that the pollen wall of hb21 was similar to that of the wild type (Supplementary Fig. 6e, g). CalS5, RPG1, CYP703A2 and NST2 were expressed at similar levels in the hb21 single mutant compared with the wild type (Supplementary Fig. 6f). These results indicate that knockout of HB21 does not affect anther or pollen development. To determine whether HB21 overexpression leads to the male sterile phenotype of efd, we generated the efd hb21 double mutant. Interestingly, this line was fully fertile. Approximately 56 seeds per silique were produced in efd hb21, which was close to the production of 58 seeds per silique in wild type (n = 30; Fig. 3c and Supplementary Fig. 7a). In addition, there were an average of 209.7 pollen grains in each efd hb21 anther, which was also similar to the level of the wild type (n = 40; Supplementary Fig. 7b). Furthermore, the pollen wall pattern of efd hb21 was regularly reticulated, similar to that of the wild type (Supplementary Fig. 7c). RT-qPCR analysis revealed that the expression of CalS5, RPG1, CYP703A2, NST1 and NST2 was evidently restored in efd hb21 (Fig. 3d). We then introduced HB21 into the efd hb21 double homozygous mutant (efd hb21^pHB21:HB21). Seven transgenic lines showed evidently reduced male fertility, which was similar to that in the efd mutant (7/10; Supplementary Fig. 8). The expression of CalS5, RPG1, CYP703A2 and NST2 were evidently downregulated in efd hb21^pHB21:HB21(Supplementary Fig. 9). To further confirm the effect of HB21 on anther development, we overexpressed HB21 driven by constitutive and tissue-specific promoters (p35S:HB21, pAMS:HB21, pEFD:HB21 and pNPU:HB21; Supplementary Fig. 10a). The fertility of these transgenic plants was reduced, and their pollen walls were severely disrupted (Supplementary Fig. 10b, c). Expression analyses revealed that HB21 was obviously upregulated, while CalS5, RPG1, CYP703A2 and NST2 were downregulated in the p35S:HB21 overexpression line (Supplementary Fig. 11). All these results demonstrate that the overaccumulation of HB21 leads to the male sterility in efd.

The 3′ terminal region (1458–1478 bp) of HB21 is responsible for its expression

Genetic complementation revealed that the male sterility of the efd mutant was rescued by the introduction of DRM2 (Fig. 1). Knockout of HB21 restored the efd phenotype (Fig. 3), indicating that EFD functions mainly through HB21. Thus, we further analysed whether HB21 is methylated. DNA methylation bisulfite-sequencing was performed using the DNA extracted from the wild-type and efd floral buds. Eight probes covering the entire HB21 genomic region were designed to identify the DNA methylation locus (Fig. 4a). The results showed that the 3′-terminus of the third exon of HB21 in the wild type was highly methylated. However, in efd, DNA methylation in this region was significantly reduced, especially the cytosines in the 1458 ~ 1478 region (Fig. 4b and Supplementary Fig. 12). Besides, the confocal microscopy observations showed that strong GFP signals were detected in the anther layers of the pHB21:GFP transgenic lines in Arabidopsis, while no signal in the pHB21:HB21-GFP lines (Supplementary Fig. 13). These results suggest that there was repressive elements within the HB21 gene. To further understand which region of the genomic HB21 is required for its expression, we prepared several constructs with full-length or truncated gHB21 (the genomic HB21 sequence) fused with eGFP (Fig. 4a), and subsequently transformed them into different tobacco leaves (N. benthamiana). The RT-qPCR of the transgenic truncated gHB21-GFP expression was analysed. Leaves with similar transgenic copy numbers were selected for gene expression analysis. Results showed the expression of the 1–1478 bp of gHB21 fused with GFP (ΔHB21^1~1478-GFP) and pHB21: HB21^1458~1478-GFP were significantly reduced to the level of the full-length HB21 fused with GFP (pHB21:gHB21-GFP; Fig. 4c and Supplementary Fig. 14). In addition, we observed the GFP signals of these transgenic tobacco leaves. GFP signals were detected in pHB21:GFP tobacco leaf cells, suggesting that the HB21 promoter can drive GFP expression in tobacco leaves (Fig. 4d). When the full-length genomic sequence of HB21 was fused with eGFP, the fluorescent signal was barely detectable in the transient transgenic tobacco leaf cells (Fig. 4d). EFD is expressed in the leaves of Arabidopsis plants¹⁷ and tobacco has the EFD homologue (Supplementary Table 2). It is likely that HB21-GFP was methylated in tobacco leaves which further affected its expression. When the 1–1458 bp genomic region of HB21 was fused with GFP (ΔHB21^1~1458-GFP), the HB21 promoter still drove the expression of the fusion protein in tobacco leaves (Fig. 4d). However, the GFP signal was barely detected in the transgenic tobacco leaves when the 1–1478 or full-length genomic sequence of HB21 was fused with eGFP (Fig. 4d). These results indicate that the 1458–1478 region of gHB21 is the target region for EFD to repress HB21 expression.

a structure of HB21. The exons are shown as black boxes, while the introns are shown as black lines. The promoter is shown as grey box. Eight probes were designed to detect cytosine methylation on HB21. Lower, diagram of the different truncations of the HB21 gene fused with GFP. b cytosine methylation pattern of HB21 in wild-type and efd buds. The red bar includes the methylation analyses of the cytosines in 1461–1476 bp. c RT-qPCR expression analyses of the truncations of HB21 in tobacco leaves. Data are presented as mean values ±SD. The error bars represent the SDs from three independent biological repeats. Student’s t test was used for statistical analyses of the difference between the wild type and mutants, **p < 0.01, ***p < 0.001 (p-values are shown in the figure). d GFP signal of the four transgenic constructs with truncations of gHB21 in tobacco leaves. Bars = 40 μm. Three different leaves for each transgenic constructs were selected to be analysed. e diagram of the cytosines in the 1435–1540 region. The methylated cytosines were lost in the efd mutant. The cytosine pattern was shown as the circles (CG), rectangles (CHG) and triangles (CHH) as shown at the top row of HB21 geno. The methylated cytosines are shown as the full circles, rectangles and triangles, while the unmethylated cytosines are shown as empty circles, rectangles and triangles. The red frame indicates that the cytosines in this region were unmethylated in efd. f the genomic sequence and the point mutation in the 1458–1478 region. There are four cytosines: 1461, 1464, 1467 and 1476. g RT-qPCR results of the point mutated HB21 transgenic tobacco leaves. h the GFP signal detection of the point mutations. Data are presented as mean values ±SD. The error bars represent the SDs from three independent biological repeats. Student’s t test was used for statistical analysis. ns, no significant difference, *p < 0.05, **p < 0.01, ***p < 0.001 (p-values are shown in the figure). Bars = 40 μm. i EMSA image of the potential HB21 probe (unmethylated probe) incubated with the EFD protein.

Full size image

The specific cytosines in the 1458–1478 bp region of gHB21 are required for the transcriptional repression of HB21

The 1458 ~ 1478 region of HB21 has four cytosines. In the wild type, these four cytosines are methylated (1461, 1464, 1467 and 1476), while their methylation is lost in efd (Fig. 4e and Supplmentary Fig. 10). To further identify which cytosine residue is responsible for this repression, we created pHB21:gHB21-GFP constructs in which different cytosines were replaced by adenine and injected them into tobacco leaves (Fig. 4f). RT-qPCR analysis revealed that the GFP expression of pHB21:gHB21-GFP was only approximately 1.2% of that of pHB21:GFP (Fig. 4g). The transcription levels of GFP in pHB21:ΔHB21^C1461A-GFP and pHB21:ΔHB21^C1464A-GFP tobacco leaves were only 6.9% and 0.4%, respectively, of the level of GFP in pHB21:GFP leaves (Fig. 4g). Accordingly, GFP signals failed to be detected in these leaves (Fig. 4h). These results suggest that the methylation of these two cytosines is not required for the repression of HB21 expression. In contrast, the transcription of GFP in pHB21:ΔHB21^C1467A-GFP and pHB21:ΔHB21^C1476A-GFP transgenic leaves increased to 7.0% and 38.2%, respectively, of that in pHB21:GFP leaves. When both the 1467th and 1476th cytosines were replaced by adenine, the transcription of GFP in the pHB21:ΔHB21^{C1467A and C1476A}-GFP tobacco leaves was significantly elevated to 60.6% of that in the pHB21:GFP leaf (Fig. 4g). Additionally, the GFP signals were also observed in these tobacco leaves (Fig. 4h). These results indicate that the methylation of the 1467th and 1476th cytosines is essential for the repression of HB21 expression. These results suggest that EFD represses the transcription of HB21 through the methylation of the 1467th and 1476th cytosines of HB21. Besides, we analysed these cytosines in pEFD:DRM2 and pEFD:CMT3 lines. Both 1467th and 1476th cytosines showed higher methylation rate in the pEFD:DRM2 lines than those in the pEFD:CMT3 lines (Supplementary Fig. 15), suggesting the importance of the methylation of these two cytosines. To further confirm whether EFD directly binds to the specific position on HB21, the electrophoretic mobility shift assay (EMSA) analysis of the HB21 was performed. The probe contains the 1467th and 1476th sites. Both EFD and the potential methyltransferase ___domain (1–260 aa) of EFD could bind to the labelled probes (unmethylated or methylated probes), while the unlabelled probe dosely abolished the binding effect of the labelled probe (Fig. 4i and Supplementary Fig. 16a, b). This result confirms that EFD can directly bind to the specific site of HB21 to perform DNA methylation.

H3K27me3 deposition on HB21 is reduced in efd

H3K27me3 has been shown to repress the expression of functional genes. Previous research has shown that H3K27me3 is highly localised on the HB21 gene in the wild-type plants¹⁰ (Supplementary Fig. 17). Loss of DNA methylation causes the redistribution of H3K27me3¹⁰. To determine whether the H3K27me3 distribution on HB21 is altered in efd, we conducted chromatin immunoprecipitation (ChIP) assays using the floral buds of the wild type and efd. Four probes with lengths of 150–200 bp (P1–P4) were designed to determine the enrichment of H3K27me3 on HB21 (Fig. 5a). We found that the deposition of H3K27me3 was evidently lower on HB21 in efd than on the wild type (Fig. 5b). Specifically, compared with those of the wild type, the probe 1 on the promoter region (P1: −132 to −300 bp) and the P3 on the 3′ terminus (P3: 1380–1551 bp) of HB21 showed significantly reduced H3K27me3 deposition in efd (Fig. 5b). These results indicate that H3K27me3 is largely reduced in efd, suggesting that EFD may be involved in H3K27me3 deposition on HB21 to repress its transcription. It has been reported that the Polycomb Repressive Complex 2 (PRC2) catalyses the trimethylation of H3K27³¹. SWINGER (SWN), CURLY LEAF (CLF) and (FIE) are the subunits of the PRC2 complex that are responsible for the modification of H3K27me3^32,33. SWN, CLF and FIE (members of PRC2) were responsible for this trimethylation. We searched the reported Ch-IP sequencing data to compare H3K27me3 deposition on the HB21 gene between the wild type and the swn clf and fie mutants^34,35. The deposition of H3K27me3 was strongly reduced in the swn clf double mutant and amiRfie knockdown mutant (Supplementary Fig. 18), indicating that SWN, CLF and FIE are closely associated with the trimethylation of H3K27me3 on HB21. In addition, we obtained the swn-1 clf-50 double mutant. Unfortunately, no homozygous plants could be obtained. However, the swn-1^-/- clf-50^+/- plants presented reduced pollen fertility to different extents (n = 6; Supplementary Fig. 19a). A portion of the pollen grains were shrunken (182/240; Supplementary Fig. 19b) and the pollen wall was defective compared with that of the wild type (Supplementary Fig. 19c). These results further suggest that H3K27me3 deposition on HB21 is required for anther development.

a structure of the genomic HB21 gene (gHB21) and the distribution of ChIP probes (P1–P4). b ChIP analysis and comparison of H3K27me3 deposition on the HB21 genomic sequence between wild type and efd. Data are presented as mean values ±SD. The error bars represent the SDs from three independent biological repeats. Student’s t test was used for statistical analysis of the differences between the wild type and mutants, *p < 0.05, **p < 0.01, ***p < 0.001 (p-values are shown in the figure).

Full size image

HB21 represses the expression of genes involved in anther development and pollen formation

Several genes responsible for anther development and pollen formation including RPG1, CYP703A2, CalS5, NST1 and NST2 were downregulated in efd (Fig. 2). Genetic and expression analyses demonstrated that EFD is required for anther development through HB21. It repressed the expression of HB21 through DNA methylation. We subsequently analysed whether HB21 directly regulates the expression of these genes. First, to reveal whether HB21 can bind to the promoters of these genes, an EMSA experiment was performed. The HD-Zip1 family members are reported to bind to the conserved cis-element ATTA³⁶. The promoter regions of CYP703A2, CalS5, RPG1/SWEET8 and NST2 contain several predicted binding sites: CalS5 (−104 to −107), RPG1/SWEET8 (−214 to −223), CYP703A2 (−211 to −214) and NST2 (−297 to −317; Fig. 6a). The maltose binding protein (MBP) tag was fused to HB21 at its N-terminus, expressed in Escherichia coli, and then purified using an affinity column. The EMSA results showed that the MBP-HB21 protein could bind to the biotin-labelled probe containing the putative binding sites, while the unlabelled competitor dose-dependently competed with the wild-type probe (Fig. 6b). To further analyse the binding of HB21 to these sites in vivo, we performed ChIP analysis. As the overaccumulation of HB21 repressed the expression of downstream genes, we used the p35S:HB21-GFP line as the plant material for the ChIP experiment. Probes spanning the potential HB21 binding sites on the promoters of CalS5, RPG1, CYP703A2 and NST2 were designed and synthesised for analysis. The results showed that HB21 was more strongly enriched on CalS5, RPG1, CYP703A2 and NST2 when incubated with the antibody (+AB) than when incubated without the antibody (−AB; Fig. 6c). These results indicate that HB21 can directly bind to the promoters of these genes. Subsequently, we used the tobacco transient expression system to analyse whether HB21 can repress the expression of these genes. The firefly luciferase (LUC) reporter gene driven by the CalS5, RPG1/SWEET8, CYP703A2 and NST2 promoters was used as a reporter, while the p35S:HB21 was used as an effector. When the pRPG1/SWEET8:LUC, pCYP703A2:LUC and pNST2:LUC constructs were transformed into tobacco leaves, a strong expression signal was observed (Fig. 6d). However, when these constructs were cotransformed with the effector p35S:HB21 into tobacco leaf cells, the LUC signals were obviously weakened (Fig. 6e). These results demonstrated that HB21 directly binds to the promoters of these genes to repress their expression.

a cis-elements of the potential HB21 binding sites on the CalS5, RPG1/SWEET5, CYP703A2 and NST2 promoters. b the results of gel shift analysis showed that HB21 can directly bind to the probes on the CalS5, RPG1/SWEET5, CYP703A2 and NST2 promoters. c ChIP analyses of the direct binding of HB21 to the potential binding sites on the promoter of the downstream genes (CalS5, RPG1, CYP703A2, NST2) in the p35S:HB21-GFP line. The probe on the β-TUBULIN promoter was used as a negative control. Data are presented as mean values +/- SD. The error bars represent the SDs from three independent biological repeats. Student’s t test was used for statistical analysis of the difference between the wild type and mutants, ns, no significant difference, *p < 0.05, **p < 0.01, ***p < 0.001 (p-values are shown in the figure). d tobacco transient expression assay for luciferase activity detection. The results showed that HB21 could repress the expression of CYP703A2, RPG1/SWEET5 and NST2. e the working model shows that HB21 globally regulates pollen wall pattern establishment (in pollen mother cells), sporopollenin material synthesis (in the tapetum) and endothecium lignification (in the endothecium) by regulating the expression of CalS5, RPG1/SWEET5, CYP703A2 and NST2.

Full size image

EFD is likely a de novo DNA methyltransferase

The EFD protein contains a typical SAM-dependent methyltransferase ___domain. In the present study, we provide data to support the function of this gene as a putative de novo DNA methyltransferase. (1) We found that HB21 is the major target of EFD (Fig. 3). The DNA methylation of HB21 is in agreement with its transcriptional repression (Fig. 4). (2) Expression of the de novo DNA methyltransferase DRM2 restored the efd phenotype, as indicated by pollen formation and plant fertility, the DNA methylation pattern of HB21 and the expression of the downstream genes, while CMT3 failed to restore the defects, the DNA methylation pattern and the gene expression in efd (Fig. 1 and Supplementary Figs. 1 and 15). De novo DNA methylation is involved in many developmental processes, such as the organ development and cancer cell suppression. DRM1/2 are the de novo DNA methyltransferases in Arabidopsis², but the drm1 drm2 mutant has no obvious developmental defect³. EFD is also expressed in vegetative tissues in addition to the anther¹⁷. It is likely that EFD plays roles in other developmental processes except the anther development. Further studies are still needed to confirm the de novo DNA methyltransferase activity of EFD.

The epigenetic modifications of HB21 repress its expression

DNA methylation and H3K27me3 are both the essential epigenetic modifications for repressing gene expression. DNA methylation occurs in the promoter, gene body and the 3’-terminus after the gene terminator^37,38. Whole genome methylation analysis and transcriptome data from the met1 mutant (lack of CG methylation) revealed that 20% of the total genes are methylated in Arabidopsis, most of which are methylated in the gene body³⁹. The loss of methylation in these gene bodies leads to transcriptional upregulation³⁹. In the present study, the loss of EFD led to the upregulation of HB21 transcription (Fig. 3a). In eukaryotes, the gene transcription is generally catalysed by RNA polymerase II (Pol II). DNA methylation in the coding frame can reduce the binding of Pol II to DNA strands⁴⁰. Our work showed that methylation in the HB21 exon led to the repression of HB21 transcription in efd (Fig. 4c, d, g). This result suggested that EFD may affect the transcriptional elongation of Pol II on HB21 through methylation, thereby repressing the transcription of HB21. DNA methylation is usually associated with histone H3 methylation⁴¹. In Arabidopsis, H3K27me3 usually marks genes with high tissue specificity^7,8,9. H3K27me3 is generally considered to be associated with stalling of the elongation of Pol II, negatively controlling gene transcription^42,43. The loss of DNA methylation leads to redistribution of methylated H3¹⁰. The Ch-IP data showed that H3K27me3 was highly enriched on HB21¹⁰. Our work showed that the deposition of H3K27me3 on HB21 was evidently reduced in efd (Fig. 5b). Mutations in SWN, CLF and FIE, three PRC2 complex members required for the trimethylation of H3K27, led to the loss of H3K27me3 deposition on the HB21 gene (Supplementary Fig. 18). The swn-1 clf-50 double mutant exhibited defective pollen wall development (Supplementary Fig. 19). Thus, DNA methylation at HB21 is proposed to enhance the distribution of H3K27me3 on HB21, which represses the transcriptional elongation of Pol II on HB21.

EFD plays a role in the global regulation of anther development

The anthers have multiple cell layers closely associated with each other to support the pollen formation and anther dehiscence. The pollen mother cell, tapetum and endothecium are three key cell layers in the anther. These genes have unique functions during anther development. The pollen mother cell undergoes meiosis to form microspores which further develop into mature pollen. It is also responsible for the establishment of the pollen wall pattern. The tapetum is responsible for the synthesis of materials for pollen wall formation while the endothecium is required for anther dehiscence to release mature pollen. A previous study showed that the pollen wall pattern is defective in efd, and the expression of RPG1, a pollen wall pattern related gene, is downregulated¹⁷. In this study, we discovered that EFD was also essential for sporopollenin synthesis and endothecium lignification (Fig. 2). Therefore, EFD regulates all three anther layers for both pollen wall development and anther dehiscence. However, the introduction of sporopollenin synthesis gene CYP703A2 into efd rescued the fertility of efd, while the introduction of RPG1 and NST2 failed to rescue efd fertility (Fig. 2d and Supplementary Fig. 20). These results indicate that sporopollenin synthesis is the major factor leading to male sterility in efd. In addition to EFD, ARF17 also plays a role in regulating these three anther layers. In pollen mother cells, ARF17 regulates the expression of CalS5 to ensure tetrad callose wall formation⁴⁴. In the tapetum, ARF17 plays a role in its development, but the specific function of ARF17 in the tapetum is still unclear⁴⁵. In the endothecium, ARF17 regulates the expression of MYB108 which is required for anther dehiscence⁴⁶. ARF17 is a transcriptional activator regulated by miRNA160⁴⁷, while EFD is a putative de novo DNA methyltransferase that represses gene transcription. Further comparisons of these two genes will be helpful for understanding the mechanism of anther development.

The EFD-HB21 double negative cascade regulates anther development

Plant reproductive development is essential for the successful production of offspring. Our genetic data showed that EFD functions mainly through the repression of HB21 transcription (Figs. 3–5). HB21 belongs to the homeobox family, the members of which are transcriptional regulators indispensable for various processes of vegetative growth^26,27,28,29. We found that HB21 further repressed the expression of the genes essential for the pollen wall pattern, pollen sporopollenin synthesis and anther dehiscence (Fig. 6). Such multistep repression cascades were proposed to generate robust temporal delays in the transcriptional network during development^48,49. In Arabidopsis, such a double repressive cascade was reported to be involved in light-dependent seed germination and leaf senescence^50,51. In the present study, the EFD-HB21 double negative cascade regulated the expression of genes in pollen mother cells, the tapetum and endothecium to maintain the pollen wall pattern, pollen sporopollenin synthesis and anther dehiscence (Fig. 2b). This type of regulation may effectively coordinate these different cell layers for pollen formation and pollen release. Plants have evolved from mosses to lycopodium, ferns and seed plants. Sporopollenin is a conserved component of the spores or pollen of all land plants⁵². Some species of moss and ferns do not produce primexine⁵³. In addition, we recently reported that the epidermis of fern sporangia can be lignified in a manner similar to that of endothecium in higher plants⁵⁴, suggesting that sporangium lignification and dehiscence may have originated from ferns (early vascular plants). Our sequence alignment revealed genes homologous to EFD (60 ~ 70%), HB21 (60 ~ 80%), CYP703A2 (60 ~ 70%), NST1/2 (70 ~ 80%) and RPG1 (50 ~ 60%) in various bryophytes and ferns (Supplementary Table 2). A double negative cascade is likely present after plant landing. Although the primexine structure for pollen wall pattern formation is present in seed plants, the conserved sequences of EFD, HB21 and RPG1 across land plants suggest that EFD-HB21-RPG1 regulation may have occurred exist much earlier during plant evolution.

Plant material and growth conditions

Experiments were performed using Arabidopsis thaliana Columbia (Col-0). An efd T-DNA insertion line was obtained from Meng-Xiang Sun’s laboratory¹⁷. The hb21 T-DNA line (CS857242) was obtained from ABRC²⁷. The swn-1 clf-50 double mutant was obtained from Dr Justin Goodrich’s laboratory. The drm1 drm2 double mutant was obtained from Dr Hua Jiang’s laboratory. To generate the rescued lines, the genes were respectively cloned and inserted into pCAMBIA1300. The constructs were subsequently transformed into Agrobacterium tumefaciens GV3101, which was subsequently used to transfect the Columbia (Col-0) wild-type plants, efd, hb21 and efd hb21 mutants as described. The Arabidopsis seeds were sown on soil and grown in a temperature-controlled chamber with 24 °C and 16-h-light: 8-h-dark cycle.

Vector construction

The 1.5 kb promoters of NPU, AMS and EFD and the genomic sequences of EFD and HB21 were respectively amplified from the wild-type genomic DNA and subsequently cloned and inserted into pCAMBIA1300-NOS or pCAMBIA1300-eGFP-NOS. The EFD, HB21, CalS5 and RPG1 coding sequences (CDS) were amplified from the wild-type floral bud cDNA. The pEFD:DRM2, pEFD:CMT3, pHB21:HB21 and pHB21:ΔHB21-GFP were also cloned as described above. The primers used are listed in Supplementary Table 1. pNPU:HB21, pAMS:HB21, pEFD:HB21 and p35S:HB21 were introduced into wild type. pNPU:RPG1, pAMS:CYP703A2, pEFD:DRM2 and pEFD:CMT3 were introduced into efd. pHB21:HB21 was introduced into efd hb21^-/-. p35S:EFD, pCYP703A2:LUC, pCalS5:LUC, pRPG1:LUC, pNST2:LUC and pHB21:ΔHB21-GFP were introduced into tobacco leaves. The plasmids were subsequently transformed into Agrobacterium GV3101, which were used to infect the Arabidopsis plants.

Real-time PCR

Total RNA was extracted from flower buds using TRIzol (Invitrogen) and purified using a Qiagen RNeasy Kit. The first strand of cDNA was synthesised using a reverse transcription kit from Transgenes Company. The expression levels of the target genes were analysed using RT-qPCR. The SYBR Green FastMix ROX was used for all RT-qPCRs, which were performed using ABI7300 Sequence Detection System (Applied Biosystems). For the expression analyses of the tobacco leaves, the transgene copy numbers in tobacco leaves were first estimated using qPCR with genomic DNA, and multiple plants with similar transgenic copy numbers (the Ct value is between 17 and 17.5) for each transgene were used for the expression analyses. Each experiment included three independent biological replicates. All the RT-qPCR experiments were performed under the following programs: 95 °C for 5 min, 40 cycles of 95 °C for 10 s and 60 °C for 1 min. The samples were normalised using β-TUBULIN as control. The relative quantification was calculated using the 2^-ΔΔCt method. The fold changes in the expression of the targeted genes were normalised to that of the internal control, β-TUBULIN. Three independent experiments of biological replicates were performed.

Expression analysis and in situ hybridisation

The floral buds of wild type and efd were fixed in the formaldehyde solution and vacuum infiltrated for 10 min on ice. The sample dehydration was performed in a graded ethanol series and stained with safranine in the xylene/ethanol solutions. Samples were put into a 60 °C constant temperature incubator for a week and finally embedded in Paraplast. The transverse sections (8 μm thick) were transferred on the poly-L-lysine coated glass slides and were prepared for hybridisation. RNA in situ hybridisation was performed using the Digoxigenin RNA Labelling Kit (Roche, USA) and the PCR DIG Probe Synthesis Kit (Roche). A 302 bp probe of HB21 was amplified using the floral bud as template and the specific HB21 primers. The PCR products were subsequently cloned and inserted into the pBluescriptSK vector. The HB21 RNA probe was transcribed in vitro using T3 or T7 RNA polymerase and the linearised pBluescriptSK-HB21 was used as the transcription template.

Cytological observations

For microscopic analysis, the plants were photographed with a Nikon digital camera. Flowers and siliques were dissected from floral buds under an Olympus SZX stereomicroscope. Pollen grains were analysed using Alexander staining solution⁵⁵. For semi-thin sectioning, the flower buds were fixed in FAA buffer and embedded in resin⁵⁶. The inflorescences from different genotypes were fixed for 2-3 d by FAA and were embedded in Spurr’s resin. The 0.01% toluidine blue/sodium borate solution were used to stain the semi-thin sections (1 μm thick) 45 °C for 5 min. For Tinopal/DiOC₂ and BF staining, the semi-thin sections on glass slides were immersed in Tinopal/DiOC₂ or BF dyeing liquid for ten minutes. Then, the slides were placed and observed under an Olympus microscope (BX51).

Bisulfite DNA conversion, sequencing and analysis

The genomic DNA of wild-type and efd floral buds (n = 120) was extracted using the PlantZol Kit from TRANSGENE. The bisulfite conversion of DNA was conducted using the EZ DNA Methylation-Gold™ Kit (D5005) from ZYMO RESEARCH and the EpiTect Bisulfite Kit from QIAGEN, (59,104) according to the kit protocols (three-time independent experiment). Eight probes were designed to cover the entire HB21 genomic region from the initial codon (P1: 0–337 bp; P2: 293–543 bp; P3: 442–662 bp; P4: 589–827 bp; P5: 761–1045 bp; P6: 954–1275 bp; P7: 1164–1472 bp; P8: 1352–1602 bp). High-purity gene fragments were amplified through PCR. All the PCRs used the 2xEs Taq Master Mix (Dye) and were performed using a ProFlex PCR System. The DNA fragments were ligated to the pMD19-T vector (TaKaRa) and subsequently transformed into DH5α competent cells. To search for potential DNA methylation sites in HB21, approximately 40-100 typical colonies were identified for each probe through PCR and subsequently sent to be sequenced. The cytosine methylation of the HB21 genomic region of pEFD:DRM2 and pEFD:CMT3 were also detected using P7 and P8. Approximately 150 typical colonies were identified for P7 and P8 through PCR and DNA sequencing.

Electronic microscopic observations

For the SEM observation, the anthers of the wild type, efd, efd hb21 and the transgenic lines were mounted on SEM stubs. The mounted samples were coated with palladium-gold in a sputter coater and examined with a JSM-840 instrument. For TEM observation, the floral buds of the wild-type, mutant and transgenic plants were fixed and embedded⁵⁷. For the sample preparation of TEM, buds from wild-type, mutant or the transgenic inflorescences were placed in 2.5% glutaraldehyde (v/v) in the ice-cold 10 mM phosphate buffer (pH 7.2–7.6). Samples were then fixed in 1% osmium tetroxide. The sample dehydration was performed in an ethanol/water series (30%, 50%, 75%, 85%, 90%, 95% and 100%). The floral buds were dehydrated twice with 100% propylene oxide, which were then transferred to the propylene oxide/Spurr’s resin mixtures with ratios of 1:1, 1:3 and 3:1, respectively. The sample polymerisation was performed to embed the floral buds in Spurr’s resin and place them at 65 °C for 48 h. The 50-70 nm thick resin slices were stained in the uranyl acetate and lead citrate solutions. The JEM-1230 transmission electron microscope was used to observe the samples.

ChIP analysis

The chromatin immunoprecipitation (ChIP) analysis of H3K27me3 in wild type and efd and the GFP in the p35S:HB21 transgenic lines were conducted as previously described⁵⁸ with slight modifications. Approximately one gram of the floral buds from the wild-type, efd and the p35S:HB21-GFP transgenic plants were frozen in liquid nitrogen, and they were then respectively crosslinked in sucrose-formaldehyde buffer. The nuclei isolation of the transgenic buds were performed in the extraction buffer and then lysed with a lysis buffer. After the chromatin shearing with ultrasonic, the DNA fragments mostly were between 200–800 bp. After pre-immuneserum with sheared salmon sperm DNA/protein A agarose mix for 1 h, the DNA–protein complex was immune-precipitated at 4 °C overnight using the antibody. Then, the magnetic beads coupled with protein G (Invitrogen, USA) were added to perform the antibody–protein/DNA complexes precipitation. The samples were incubated at 65 °C overnight after washing to reverse the crosslinking. The antibody against H3K27me3 was a rabbit polyclonal antibody that was purchased from ABclonal (A2363) and with 1:100 dilution in the ChIP assay. ChIP analysis of HB21 was conducted using the p35S:HB21-GFP transgenic plants. The rabbit polyclonal antibody was purchased from Agrisera (As184227) and was diluted with 1:60 in the ChIP assay. The DNA from the early floral buds of these transgenic plants was extracted and purified. Nuclei were isolated with extraction buffer and lysed with lysis buffer. The precipitated DNA was then purified and analysed by the RT-qPCR. We calculated the ΔCt values and used the 2^-ΔCt method to analyse the fold enrichment. All samples were run with at least three biological replicates.

Electrophoretic mobility shift assay

The construction of plasmids for the expression of recombinant MBP-EFD and MBP-HB21 proteins in the Rossetta E. coli strain and the purification processes were performed according to the manufacturer’s instructions. The Light Shift Chemiluminescent EMSA Kit (Thermo Scientific) was used in the EMSA experiment. The DNA fragments containing the consensus sequence (ATTA) in the regulatory region of the promoters of the pollen wall development genes were generated by PCR amplification with the following specific primers that were used to generate a biotin label. The EMSA was performed with a Light Shift Chemiluminescent EMSA Kit (Thermo Scientific, http://www.thermoscientific.com). The binding reactions containing 10X binding buffer and 0.05 mg mL⁻¹ poly (dI-dC)], MBP-EFD, MBP-EFD (1–260aa) or MBP-HB21 recombinant fusion protein and biotin-labelled DNA were performed at room temperature for 20 min. The subsequent processes were performed according to the manufacturer’s instructions. Each EMSA experiment was performed three times with consistent results.

Dual transient expression assay in Tobacco leaves

To generate the luciferase reporter constructs, the promoters of CYP703A2, RPG1/SWEET5 and CalS5 were amplified from Col-0 genomic DNA and cloned and inserted into pLL00R-LUC, with CYP703A2 promoter driving firefly luciferase (LUC) gene and a CaMV 35S promoter driving the HB21 gene. The primers used for all the constructs are listed in Supplementary Table 1. To generate the effector constructs of pCYP703A2:LUC, pRPG1:LUC and pNST2:LUC, the coding sequences of these three genes were amplified and cloned into pCAMBIA1300 vector with the 35S promoter in the multiple cloning site. The tobacco leaves grown for three to five weeks were chosen to be injected with the GV3101 Agrobacterium containing these constructs. The transient signals were detected using a Tanon-5500 Chemiluminescent Imaging System (Tanon, Shanghai, China) within the same exposure time. The experiments were repeated three times.

Data collection and analysis

CellSens Entry (OLYMPUS) is used for pollen and anther imaging. Tanon-5500 chemiluminescent imaging system (Tanon) is used for the luciferase signal detection. DNA gel imaging system is AIIDoc-x (Tanon). 7300 system SDS software (ABI) is used for qRT-PCR. The statistics data are measured by Prism 9 and Excel 2010.

Accession numbers

Genes referred in this work are as follows: EFD (At3g54150), HB21 (At2g18550), CalS5 (At2g13680), RPG1/SWEET5 (At5g40260), CYP703A2 (At1g01280), MYB103/MS188 (At5g56110), NPU (At3g51610), AMS (At2g16910), CYP704B1 (At1g69500), MS2 (At3g11980), AcoS5 (At1g62940), LAP5 (At4g34850), LAP6 (At1g02050), TKPR1 (At4g35420), TKPR2 (At1g68540), DRM1 (At5g15380), DRM2 (At5g14620), SWN (At4g02020), CLF (At2g23380), FIE (At3g20740).

Reporting summary

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