The temporal regulation of TEK contributes to pollen wall exine patterning

Pollen wall consists of several complex layers which form elaborate species-specific patterns. In Arabidopsis, the transcription factor ABORTED MICROSPORE (AMS) is a master regulator of exine formation, and another transcription factor, TRANSPOSABLE ELEMENT SILENCING VIA AT-HOOK (TEK), specifies formation of the nexine layer. However, knowledge regarding the temporal regulatory roles of TEK in pollen wall development is limited. Here, TEK-GFP driven by the AMS promoter was prematurely expressed in the tapetal nuclei, leading to complete male sterility in the pAMS:TEK-GFP (pat) transgenic lines with the wild-type background. Cytological observations in the pat anthers showed impaired callose synthesis and aberrant exine patterning. CALLOSE SYNTHASE5 (CalS5) is required for callose synthesis, and expression of CalS5 in pat plants was significantly reduced. We demonstrated that TEK negatively regulates CalS5 expression after the tetrad stage in wild-type anthers and further discovered that premature TEK-GFP in pat directly represses CalS5 expression through histone modification. Our findings show that TEK flexibly mediates its different functions via different temporal regulation, revealing that the temporal regulation of TEK is essential for exine patterning. Moreover, the result that the repression of CalS5 by TEK after the tetrad stage coincides with the timing of callose wall dissolution suggests that tapetum utilizes temporal regulation of genes to stop callose wall synthesis, which, together with the activation of callase activity, achieves microspore release and pollen wall patterning.


Author summary
To develop into mature pollen grains, microspores require formation of the pollen wall. To date, pollen wall developmental events, including production and transportation of pollen wall components, synthesis and degradation of the callose wall, and deposition and demixing of primexine, have been studied in Arabidopsis, and a number of anther-or tapetum-specific genes involved in pollen wall formation have been uncovered. However, whether the specific expression patterns of these genes contribute to pollen wall

Introduction
In angiosperms, the male gametophyte (pollen) is surrounded by a pollen wall. The pollen wall usually comprises two main layers, the outer sporophyte-derived exine and the inner gametophyte-derived intine. Distinctive pollen wall patterns vary between species but are conserved within species. In Arabidopsis, the exine creates a reticulate pattern on the pollen surface, containing layers known as sexine and nexine. The major constituent of sexine is sporopollenin, which is the biopolymer of polyhydroxylated aliphatic chains and aromatic rings [1][2][3][4][5]. Nexine includes arabinogalactan proteins (AGPs) [6], which are hydroxyproline-rich glycoproteins [7]. In turn, intine consists of cellulose, hemicellulose, pectin and proteins [8]. These features and structural components allow pollen wall to protect pollen from environmental stresses, such as desiccation, UV radiation and microbial attack, as well as allow it to provide speciesspecific adhesion to stigma [9][10][11].
The cellular events involved in pollen wall development have been most thoroughly studied in Arabidopsis [9,10,[12][13][14][15][16][17][18][19][20]. The development of the pollen wall initiates in the individual microspores of tetrads after male meiosis. The callose wall first covers the plasma membrane of microspores. The domains where the plasma membrane closely contacts the callose wall are required for the formation of apertures. Later, the primexine matrix follows the undulation of the plasma membrane to guide the siting of sporopollenin. In the early tetrad stage, the callose wall, primexine, and plasma membrane participate in the formation of the sexine template. At the late tetrad stage, nexine starts to gradually accumulate in the primexine matrix. After callose wall degradation and microspore release, nexine completes its formation beneath sexine. Following the expansion of microspores, intine forms below the nexine, and pollen coat fills in the exine cavities. Therefore, these pollen wall components precisely appear at specific time points and sequentially assemble to form the elaborate pollen wall.
Pollen wall biosynthesis is a joint effort on the parts of microspores and the tapetum. The tapetum is the innermost sporophytic cell layer adjacent to microsporocytes and microspores [21]. It plays essential roles in pollen wall development because most pollen wall materials, including sporopollenin, lipids and proteins, are produced by, stored in, and transported from the tapetum [22]. Several tapetum-expressed genes are key players in directing the highly sculptured pollen wall in Arabidopsis and are involved in exine patterning [23][24][25][26][27][28], tapetum development and exine formation [29][30][31][32][33][34], sporopollenin biosynthesis [35][36][37][38], and nexine formation [6,39]. Mutations in these genes lead to abnormal pollen walls and compromised pollen grains. Recent studies have shown that regulatory networks in the tapetum control pollen wall formation [40][41][42][43][44]. Among them, ABORTED MICROSPORES (AMS), a bHLH transcription factor, directly regulates the expression of TRANSPOSABLE ELEMENT SILENCING VIA AT-HOOK (TEK) [39]. Subsequently, TEK, encoding an AT-hook nuclear matrix attachment region (MAR) binding protein, becomes highly expressed in the tapetum to promote nexine formation [39]. The investigations of mutants in which these genes are disrupted have demonstrated that these molecular players are required for the biosynthesis of multilayered pollen walls. However, whether the specific temporal patterns of these regulators contribute to pollen wall development or patterning remains unknown.
Because the AMS transcript is expressed earlier in the tapetum than the TEK transcript (anther stage 5 vs. stage 7, respectively) [39,45], in order to explore whether the alteration in the temporal pattern of TEK disturbs pollen wall development, we used the AMS promoter to prematurely drive TEK expression. In the Arabidopsis pAMS:TEK-GFP transgenic plants with a wild-type background (hereafter referred to as pat), TEK-GFP was precociously expressed in the tapetal nuclei. We found that pat lines exhibited total male sterility and formed defective pollen walls. Ultrastructural observations showed that the callose wall formation was severely blocked and that the abnormal exine pattern existed in all pat microspores. Expression analysis confirmed that expression of the callose synthase CalS5 is strongly decreased in pat, which could account for the reduced callose synthesis. We showed that, in wild-type anthers, TEK negatively regulates CalS5 expression after the tetrad stage, which coincides with the callose wall degradation. Additionally, we found that the precocious TEK-GFP in pat directly represses the expression of CalS5 via H3K9me2. These results showed that the correct regulation of timing of TEK expression is required for pollen wall exine patterning and is coordinated with other pollen developmental events for the processes of pollen wall formation and pollen grain maturation.

Premature tapetum-specific expression of TEK-GFP protein in pat
To test whether premature expression of TEK can influence pollen wall development, we fused the AMS promoter with the TEK-GFP chimeric gene and introduced this construct (pAMS: TEK-GFP) into the wild-type Arabidopsis Columbia-0 (Col) plants ( Fig 1A). Of the 26 transgenic plants, 21 plants exhibited male sterility and 5 plants were fertile. All male-sterile lines carried the transgene, as assessed by the amplification of the chimeric fragment, while fertile plants did not show amplified products (S1A and S1B Fig). Among these male-sterile plants, TEK expression was examined in three independent lines (pat-1, pat-2 and pat-3). Quantitative RT-PCR (qRT-PCR) analysis showed that the total transcript levels of TEK in these lines were higher than those in the wild type (S1C Fig). Additionally, other cytological observations were performed in the pat-1, pat-2 and pat-3 lines, and the results were similar to each other (S2 and S3 Figs), leading us to uniformly refer to these lines as pat.
To determine whether this transgene leads to the ectopic expression of TEK in pat lines, we investigated its spatiotemporal patterns in anthers from wild-type and pat plants. Fourteen distinct stages of anther development have been described in Arabidopsis [21]. RNA in situ hybridization showed that TEK was expressed in the tapetum of the wild-type anthers throughout stages 7-8 ( Fig 1B-1E), with the strongest hybridization signal at stage 7 (Fig 1D), in agreement with a previous study [39]. In the pat anthers, however, TEK transcripts were prematurely expressed during stages 5-6, with a strong signal in the tapetum and a relatively weak signal in the microsporocytes (Fig 1G and 1H). From stage 7 to stage 8, the hybridization signal in the tapetum peaked and later decreased after the release of microspores (Fig 1I and  1J), which was the same expression pattern as in the wild type (Fig 1D and 1E). No signal was produced in either wild-type or pat anthers with the sense probe (Fig 1F and 1K). These results indicated that TEK is expressed earlier in pat anthers than in wild-type plants. Additionally, AMS expression level and pattern were not affected in pat anthers, excluding the possibility that male sterility was caused by perturbations in AMS expression (S4 Fig).
To further monitor TEK subcellular localization, we placed the TEK-GFP chimeric gene under its native promoter, and the construct pTEK:TEK-GFP was introduced into wild-type plants. In the pTEK:TEK-GFP transgenic plants, no fluorescence of TEK-GFP was displayed in the microsporocytes and tapetum during stages 5-6 ( Fig 1L and 1M). Then, TEK-GFP specifically exhibited fluorescence in the nuclei of tapetal cells during stages 7-8 ( Fig 1N and 1O). In the pat transgenic plants, the fluorescence of TEK-GFP was similarly restricted to the nuclei of tapetal cells but was prematurely expressed at stage 6 ( Fig 1P and 1Q). The expression signal of TEK-GFP was retained during the tetrad stage (stage 7) (Fig 1R), but only a very weak signal was observed in the tapetum at stage 8 ( Fig 1S). These results demonstrated that TEK-GFP protein driven by the AMS promoter is precociously expressed in the tapetal nuclei of pat anthers.

Male sterility is associated with microspore degeneration in pat plants
Compared to the wild-type plants, the pat transformants exhibited normal vegetative growth but developed short siliques and were completely male sterile (Fig 2A and 2B). To examine pollen viability in the anthers, we performed Alexander's staining to distinguish the aborted pollen grains from mature ones. In wild-type anthers, the mature pollen grains were stained purple (Fig 2C), whereas in pat anthers, pollen remnants were stained green (Fig 2D), suggesting that the pollen grains were aborted in pat. Anther cross-sections further showed that there was no detectable difference between wild-type and pat anthers until stage 6 ( Fig 2G and 2L). Compared with the wild type, the callose of tetrads in pat seemed to be reduced at stage 7 ( Fig  2H and 2M), which was confirmed by aniline blue staining (Fig 2Q and 2R). At stage 8, the In the wild-type anther (G-K), mature pollen grains were produced. In the pat anther (L-P), round microspores were released from tetrads (N) and gradually degenerated at later stages (O-P). E, epidermis; En, endothecium; ML, middle layer; T, tapetum; MC, meiocytes; Tds, tetrads; Msp, microspore; PG, pollen grain; dMsp, degenerated microspore; dPG, degenerated pollen grains. Scale bars, 5 μm. (Q-R) Aniline blue staining of callose from WT and pat. The peripheral callose wall of pat was slightly thinner (R) than that of the wild type (Q). PC, peripheral callose; IC, interstitial callose. Scale bars, 20 μm.
https://doi.org/10.1371/journal.pgen.1008807.g002 wild-type individual microspores with angular shapes were released from tetrads ( Fig 2I). In contrast, the pat microspores were rounder than those of the wild type (Fig 2N). At stage 9, the wild-type microspores became vacuolated, while the pat microspores began to disintegrate (Fig 2J and 2O). At later stages, the wild-type microspores became noticeably enlarged, condensed their cytoplasm, and finally became mature pollen grains ( Fig 2K). However, the pat microspores further degenerated, and only pollen remnants remained in the locule (Fig 2P). Scanning electron microscopy (SEM) showed the characteristic reticulate pattern on the surface of wild-type pollen, while a defective pollen wall was observed on the collapsed pat pollen (Fig 2E and 2F). These results showed that the failure of microspore production with disordered pollen wall morphology causes complete male sterility in pat plants.

Defective exine development leads to microspore abortion in pat
It is acknowledged that aberrations in exine formation usually cause male sterility; therefore, we investigated the ultrastructure of the exine in pat plants by transmission electron microscopy (TEM). At stage 6, the wild-type microsporocytes underwent meiosis, and the callose wall accumulated around their periphery (Fig 3A and 3F). The accumulation of the callose wall in pat microspores was similar to that in wild-type microspores (Fig 3K and 3P). The tetrad stage is critical for exine formation [46,47], as the stage during which the exine pattern is programmed through cooperation between the tapetum and microspores. In the wild type, primexine accumulated underneath the callose wall, and the plasma membrane became wavy in early stage 7 (Fig 3B and 3G). In contrast, in pat the callose wall was slightly sparse, the plasma membrane remained straight, and the sporopollenin precursors were located beside the primexine (Fig 3L and 3Q). In the middle of stage 7, the invagination of the wild-type plasma membrane was clearer, and the sporopollenin precursors from the tapetum deposited onto the peak to form the probacular within the primexine matrix (Fig 3C and 3H). In contrast, in pat the callose wall was thinner, and the sporopollenin precursors were randomly inserted into the primexine matrix (Fig 3M and 3R). At late stage 7, the probaculae elongated to the callose wall, and their distal ends fused into tectum, which constituted the pro-sexine in the wild type (Fig 3D and 3I). In pat, the sporopollenin precursors submerged into the primexine matrix, forming irregular shapes (Fig 3N and 3S). When the callose wall thoroughly dissolved at stage 8 in the wild type, nexine developed under the sexine, forming an intact exine with a T shape (Fig 3E and 3J). In pat, the irregular sporopollenin surrounded the microspores without any attachment (Fig 3O and 3T). Finally, the pat microspores degenerated because of the lack of pollen wall protection. These observations indicated that pat microspore abortion is due to abnormalities in exine development.

Decreased CalS5 expression leads to reduced callose synthesis in pat
Since the callose wall plays an important role in determining the exine pattern [48,49], combined with the aberration of callose deposition in pat at the tetrad stage (Fig 3Q-3S), we explored the expression of some callose-related genes in pat inflorescences. CalS5, encoding a callose synthase (CalS), is required for callose synthesis around microsporocytes [48]. The mutations in CalS5 affect the amount of callose deposition, leading to a disrupted exine pattern [48,49]. CYCLIN-DEPEDENT KINASE G1 (CDKG1) facilitates callose wall formation via the regulation of CalS5 splicing [50]. AUXIN RESPONSE FACTOR 17 (ARF17) regulates the expression of CalS5 [43]. qRT-PCR analysis showed that only the CalS5 transcript was greatly downregulated in the pat lines, while the others were not altered (Fig 4A). In addition, because primexine represents a decisive factor in exine ornamentation [24,26,27,51], the expression of genes involved in primexine formation was also detected. We found that while the expression of NO PRIMEXINE AND PLASMA MEMBRANE UNDULATION (NPU) was slightly decreased, the levels of other genes were similar to those in wild type (Fig 4B), consistent with our TEM observation that primexine could still be formed in pat (Fig 3Q-3S).
To further confirm the reduction of callose synthesis in pat, we used aniline blue to stain callose and performed a callose fluorescence bleaching test. We randomly observed the tetrads from 12 independent tetrad-stage buds in the 12 independent WT and pat lines, recording the quenching time of each set ( Fig 4C). In the wild type, the callose wall fluorescence of tetrads gradually quenched under UV. However, the callose wall fluorescence in pat quenched much faster. The average quenching time of callose wall fluorescence in pat was significantly less than in wild type. Therefore, these results showed that the defective callose synthesis in pat correlates with the severely reduced CalS5 expression; this, in turn, may induce other defects in later pollen wall placement and patterning.

Premature TEK-GFP controls CalS5 expression via increasing H3K9me2 presence at the CalS5 gene
Taken together, these results suggest that TEK-GFP proteins are prematurely expressed ( Fig  1Q) and that the expression of CalS5 is greatly decreased (Fig 4A). Is there a possibility that premature TEK-GFP affects the CalS5 expression in pat anthers? To determine this, we investigated the CalS5 expression pattern by in situ hybridization. In wild-type anthers, hybridization signals were predominantly observed in microsporocytes at stages 5-6 ( Fig 5A and 5B). At stage 7, the signals were highly expressed in both the tapetum and tetrads (Fig 5C). No signal was found in the tapetum or microspores at stage 8 ( Fig 5D). In the pat anthers, the signals at stage 5 showed no obvious difference from those in the wild-type anthers (Fig 5F). However, the signals in pat were reduced during stages 6-7 (Fig 5G and 5H and S5 Fig), consistent with the qRT-PCR data (S1C Fig). In tek anthers, the expression patterns of CalS5 at early stages were similar to those in wild-type anthers (Fig 5K-5M). However, hybridization signals were still detected in both the tapetum and microspores at stage 8 (Fig 5N), suggesting that, in the wild-type, TEK specifically suppresses CalS5 expression at this stage. Control hybridizations with the sense probe for CalS5 did not show any signals in wild-type, pat and tek anthers ( Fig  5E, 5J and 5O). Thus, considering that TEK-GFP was expressed precociously (Fig 1Q and 1R), we supposed that premature expression of TEK decreased CalS5 expression in the pat anther.
As a putative MAR binding protein, TEK functions in silencing transposable elements (TEs) and repeat-containing genes by regulating histone dimethylation on H3K9 during flowering time regulation in Arabidopsis Landsberg erecta-0 (Ler) [52]. We wondered whether this precocious TEK could repress CalS5 expression via the similar mechanism in pat anthers. To test this hypothesis, we performed a chromatin immunoprecipitation (ChIP) assay using inflorescences from pat plants. It has been reported that MARs are AT-rich sequences of high affinity [53], and TEK binds MARs through the AT-hook motif [6]. We searched all putative MARs (A-box motif, WADAWAYAWW motif and AATATT motif) throughout the CalS5 promoter and genomic sequence (Fig 6A). Primers were designed including or near the identified motifs to generate fragments of approximately 200 bp (Fig 6A and S6 Fig). Quantitative ChIP-PCR (qChIP-PCR) on putative TEK-GFP-binding sites showed that a genomic region of CalS5 amplified by the primer set P11 was particularly enriched compared to the mock control (without the GFP monoclonal antibody) (Fig 6B). In contrast, a promoter region with the predicted MARs represented by primer set P3 showed no specific enrichment (Fig 6B). To further confirm the qChIP-PCR results in vitro, we performed an electrophoretic mobility shift assay (EMSA). First, the recombinant TEK protein fused to glutathione S-transferase (GST-TEK) was expressed in and purified from Escherichia coli (S7 Fig). Then, this GST-TEK protein and probes containing the P11 fragment (+2147 to +2332) were incubated together, which resulted in a specific band shift. No band shift appeared when only GST and the probe were co-incubated as a negative control. When the unlabeled probes were added, the intensity of shifted bands was gradually reduced in a concentration-dependent manner, confirming the binding specificity ( Fig 6C). These results suggested that TEK directly binds to the specific MARs of CalS5 in vivo.
We then performed a ChIP assay to determine whether the repression of CalS5 is associated with the change in H3K9me2. The inflorescences of wild-type and pat plants were collected, and antibody against H3K9me2 was used. qChIP-PCR showed that compared to the mock ChIP was performed on pat inflorescences with (gray bars) or without (black bars) the GFP monoclonal antibody. Error bars represent SD (n = 2). � p < 0.05 (t-test). (C) EMSA assay was performed with the GST-TEK fusion protein, biotin-labeled probe that spanned the P11 fragment (+2147 to +2332), and a 25-fold, 100-fold and 300-fold excess of unlabeled competitor probes. Unlabeled competitors with the same P11 sequence are able to reduce the visible shift significantly (arrowhead). Glutathione S-transferase (GST) protein was expressed as a negative control. (D) ChIP performed on Col (black bars) or pat (gray bars) inflorescences with H3K9me2 antibody. Fold enrichment calculations from two replicate qPCR assays in three independent ChIP experiments. Error bars represent SD (n = 2). �� p < 0.01 (t-test).
https://doi.org/10.1371/journal.pgen.1008807.g006 control, H3K9me2 was specifically enriched in pat plants in the fragment of CalS5 represented by the primer set P11 (Fig 6D and S2 Table). These results suggest that the premature appearance of TEK represses CalS5 expression by modulating H3K9 dimethylation in pat.

Discussion
To successfully grow into pollen grains, microspores require formation of the pollen wall. To date, multiple genes involved in formation of pollen wall layers have been identified and characterized [54]. However, whether the accurate timing of the expression of these genes is associated with pollen wall patterns and formation remains unknown. A previous study reported that the nexine layer is absent in the knockout mutant of TEK [39]. In the present study, we revealed that TEK specifically represses the expression of CalS5 in wild-type anthers after the tetrad stage (Fig 5D and 5N). We further demonstrated that prematurely expressed TEK represses CalS5 expression at the tetrad stage via histone modifications (Figs 1Q, 1R, 4A, 6B  and 6D). This disturbs normal callose wall synthesis and exine patterning (Figs 2F, 3K-3T and 4C). Combined with previous studies, these results show that TEK not only activates the genes for nexine formation at the tetrad stage but also represses the expression of CalS5 after the tetrad stage. Thus, these findings suggest that TEK, as a pollen wall regulator, executes its different functions in pollen wall patterning via differential temporal regulation.
In this study, defective callose synthesis resulting from reduced CalS5 expression led to aberrant exine deposition (Fig 3K-3T). This result demonstrates that appropriate callose synthesis is essential for exine patterning during microspore development. Several theories on the biological functions of the callose wall have been advanced. Our TEM observation showed that primexine initially coupled to the contours of the plasma membrane and that the peaks of the undulating plasma membrane determined the sites for sporopollenin accumulation at the early tetrad stage. Meanwhile, almost no sporopollenin precursors derived from the tapetum were transported to primexine passing through the callose wall (Fig 3B and 3G). In contrast, in pat tetrads, the sporopollenin precursors penetrated through the callose wall and aggregated around the primexine (Fig 3L and 3Q). It is likely that the lower callose content in pat does not provide a sufficient barrier for sporopollenin precursors when the primexine matrix is not ready for their deposition. Therefore, in combination with the similar TEM data in cdkg1 tetrads [50], these results seem to support the biological functions of the callose wall proposed by Heslop-Harrison (1964): that the callose wall acts as a chemical barrier to filter the molecules and isolate the haploid microspores in tetrads from the influence of the surrounding diploid tissues, ensuring normal pollen wall patterning [55].
Both primexine formation and membrane undulation are involved in pollen wall ornamentation [56]. Thin or absent primexine is usually found in callose deficient mutants, such as cals5, cdkg1 and arf17 mutants [43,48,50], suggesting that callose may provide a surface against which primexine is deposited or act as a source of glucose for primexine formation. In pat anthers, the primexine matrix still thickens normally at stage 7 (Fig 3Q-3S), and the expression of genes required for primexine formation is not affected (Fig 4B), suggesting that the reduced callose wall of pat may affect primexine for exine patterning through other routes. The primexine is described to be a polysaccharide material [15,26], and the mixture of these polysaccharides is not stable [57]. Recently, it has been reported that polysaccharide materials tend toward demixing in the primexine, leading to spatially modulated phase separation. When primexine separation is in different states, it will form different templates for pollen wall deposition. It is speculated that membrane undulation in the vicinity of the callose wall induces this phase separation, and that components, including sporopollenin polymers and cellulose fibrils, arrest the phase separation [58]. We speculate that the reduced callose wall, straight membrane, and early appearance of sporopollenin precursors in pat may alter the phase-separation process in the primexine, leading to a change in exine patterning.
During pollen wall development, microsporocytes and tapetum work in tandem. Initially, the callose wall is produced by microsporocytes during meiosis, and tetrads are encased by the callose wall, which marks the initiation of the exine pattern [59]. Subsequently, sporopollenin precursors produced from the tapetum are transported to specific places within the primexine [15], forming exine via self-assembly [60]. When the pro-sexine is formed, callase secreted from the tapetum dissolves the callose wall and releases microspores [61], marking the end of the exine patterning process. It has been reported that engineering callase activity to prematurely dissolve the callose wall produces microspores lacking normal pollen walls and causes male sterility in transgenic tobacco [62]. Therefore, the timing of callase secretion is critical for normal pollen wall development. In this study, we found that TEK, a tapetum-specific transcription factor, represses the expression of CalS5 at stage 8 (Fig 5D), which coincides with the timing of callose wall degradation. This result suggests that tapetum not only provides the callase to dissolve the callose wall but also utilizes the temporal regulation of genes to stop callose wall synthesis, which together achieve microspore release and pollen wall patterning. In conclusion, we propose that the temporal control of pollen wall regulators exists in coordination with other processes to establish the intrinsic developmental timing scheme for whole pollen wall building and patterning.

Generation of constructs and transgenic plants
To generate the constructs pAMS:TEK-GFP and pTEK:TEK-GFP, a 771-bp genomic fragment of TEK was amplified from the wild type by KOD Neo Plus polymerase (Toyobo Co., Ltd., Osaka, Japan). The PCR product was cloned into a modified GFP-pCAMBIA1300 vector. Then, a 1500-bp AMS promoter and a 920-bp TEK promoter were amplified. These PCR products were individually digested by restriction enzymes (Takara Biotechnology) and ligated into the plasmid. After confirmation by restriction digestion and DNA sequencing, the resulting constructs were transformed into Agrobacterium tumefaciens GV3101, and the plants were transformed using the floral dip method [63]. The transformants were screened on Plant Nutritional Solution (PNS) media containing 20 mg/L hygromycin B and later transferred into the soil for PCR identification. Primer sequences are presented in S1 Table.

Expression analysis
Total RNA extraction was performed using TRIzol (Life Technologies) following the protocol in the user's manual. cDNAs of wild-type and pat inflorescences were used for the expression analysis of selected genes. Real-time quantitative PCR was performed using gene-specific primers and SYBR Green Master Mix (TOYOBO) on the ABI 7300 platform (Applied Biosystems). The experiments were repeated three times, and the data were averaged. The β-tubulin gene was used as an internal normalization control [65]. Fold changes in gene expression were calculated using the ΔΔCt (cycle threshold) values. The relevant primers are listed in S1 Table.

Fluorescence microscopy
For callose staining, anthers at the tetrad stage were squeezed onto the slide and stained by aniline blue solution (0.1 g/L in 50 mM K3PO4 buffer, pH 7.5) [64]. Aniline blue was observed under UV illumination on an Olympus BX51 fluorescence microscope. The TEK-GFP localization in anthers of pTEK:TEK-GFP and pat lines was detected by a Carl Zeiss confocal laser scanning microscope (LSM 5 PASCAL; Zeiss, http://www.zeiss.com).

RNA in situ hybridization
The probe fragment was amplified from the wild-type cDNA using primers (see S1 Table for primers). The PCR products were cloned into the pBluescript II SK (-) vector (Strata gene; http://www.stratagene.com) and confirmed by sequencing. Plasmid DNA was completely digested with EcoRI or BamHI. Antisense and sense digoxigenin-labeled probes were prepared using T3 or T7 RNA polymerase by the PCR DIG Probe Synthesis Kit (Roche, USA). Images were obtained with an Olympus BX-51 microscope. More details were described previously [45].

Transmission electron microscopy (TEM) and scanning electron microscopy (SEM)
Arabidopsis buds from wild type and pat were fixed in 2.5% glutaraldehyde in 10 mM phosphate buffer (pH 7.4). Samples were post-fixed in 1% osmium tetroxide and dehydrated in an ethanol/water series (30,50,75,85,90, 95 and 100%). Then, the samples were dehydrated twice with 100% propylene oxide. Samples were subsequently transferred to 1:1, 1:3 and 3:1 propylene oxide/Spurr's resin mixtures and kept overnight. Later, samples were embedded in Spurr's resin and polymerized at 65˚C for 48 h. Ultrathin sections (70-nm thick) were cut using diamond knives and stained in a solution of uranyl acetate and lead citrate. The images were viewed on a Hitachi H-600 transmission electron microscope (Hitachi Ltd, http://www. hitachi.com). A scanning electron microscopy assay was performed as described by [33].

Chromatin immunoprecipitation assay
In the T1 generation, the pat transformants were confirmed by PCR identification and all of them showed male sterility. Then, WT pollen was used to cross-fertilize several independent pat stigmas to produce F1 seeds. In the F1 generation, the male-sterile plants were confirmed by PCR to have transgene insertion. Inflorescences from the male-sterile pat plants with transgene insertion were the materials used for the ChIP assay. A total of 0.8-1.0 g inflorescences from wild-type and pat plants were collected and crosslinked in the formaldehyde-containing buffer. After isolating the nuclei and shearing the chromatin with ultrasonication, most DNA fragments had a size between 200-800 bp. After preimmunization with sheared salmon sperm DNA/protein A agarose mix (Millipore, USA) for 1 h, supernatants were incubated at 4˚C overnight with monoclonal antibodies (1:125 dilution) against either GFP or dimethylated H3K9. Seventy microliters of magnetic beads coupled with protein G (Invitrogen) were added to precipitate the antibody-protein/DNA complexes. The DNA fragments were eluted after reverse crosslinking at 65˚C overnight. The remaining steps for DNA purification were carried out according to the manufacturer's instructions. Real-time PCR was performed on an ABI PRISM 7300 detection system (Applied Biosystems, USA) with SYBR Green I master mix (TOYOBO, Japan). All PCR experiments were performed under the following conditions: 95˚C for 5 min, 40 cycles of 95˚C for 10 s and 60˚C for 1 min. Under the same conditions, we calculated the ΔCt values and used 2 -ΔCt as the fold enrichment. The relevant primers are listed in S1 Table. Electrophoretic mobility shift assay To obtain purified TEK protein, the full-length fragment of TEK was amplified and ligated into the pGEX-4T vector (GE Healthcare, http://www.gehealthcare.com) to generate the construct GST-TEK. Expression and purification of the fusion protein were performed according to the manufacturer's instructions. Corresponding primers (P11-F/P11-R) were used to amplify the probes with 5' biotin labeling. The competitor probes contained the same DNA sequence but lacked the 5' biotin labeling. EMSA was performed according to the manufacturer's instructions (Thermo Scientific, Waltham, MA, USA). Primer sequences are listed in S1 Table. Supporting information