Identification of tapetum-specific genes by comparing global gene expression of four different male sterile lines in Brassica oleracea

The tapetum plays an important role in anther development by providing necessary enzymes and nutrients for pollen development. However, it is difficult to identify tapetum-specific genes on a large-scale because of the difficulty of separating tapetum cells from other anther tissues. Here, we reported the identification of tapetum-specific genes by comparing the gene expression patterns of four male sterile (MS) lines of Brassica oleracea. The abortive phenotypes of the four MS lines revealed different defects in tapetum and pollen development but normal anther wall development when observed by transmission electron microscopy. These tapetum displayed continuous defective characteristics throughout the anther developmental stages. The transcriptome from flower buds, covering all anther developmental stages, was analyzed and bioinformatics analyses exploring tapetum development-related genes were performed. We identified 1,005 genes differentially expressed in at least one of the MS lines and 104 were non-pollen expressed genes (NPGs). Most of the identified NPGs were tapetum-specific genes considering that anther walls were normally developed in all four MS lines. Among the 104 NPGs, 22 genes were previously reported as being involved in tapetum development. We further separated the expressed NPGs into different developmental stages based on the MS defects. The data obtained in this study are not only informative for research on tapetum development in B. oleracea, but are also useful for genetic pathway research in other related species. Electronic supplementary material The online version of this article (doi:10.1007/s11103-015-0287-0) contains supplementary material, which is available to authorized users.


Introduction
Anther development comprises both gametophyte and anther wall development (Engelke et al. 2002;Ge et al. 2010;Ma and Sundaresan 2010). The connection between these two relatively independent processes occurs in the tapetum, a layer of the anther wall. As the innermost of four somatic layers, the tapetum surrounds developing reproductive cells and plays an important role in the male fertility of pollen grains. Its secretory tissue provides proteins, lipids, and other nutrients for pollen development and exine formation (Mariani et al. 1990;Piffanelli et al. 1998).
Anther development is divided into two main phases: microsporogenesis and microgametogenesis, and these are further subdivided into 14 anther stages (Chang et al. 2011;Ma 2005;Sanders et al. 1999;Smyth et al. 1990). In this research, to simplify these complex developmental processes, we divided these anther stages it into four major stages according to anther developmental events: Abstract The tapetum plays an important role in anther development by providing necessary enzymes and nutrients for pollen development. However, it is difficult to identify tapetum-specific genes on a large-scale because of the difficulty of separating tapetum cells from other anther tissues. Here, we reported the identification of tapetum-specific genes by comparing the gene expression patterns of four male sterile (MS) lines of Brassica oleracea. The abortive phenotypes of the four MS lines revealed different defects in tapetum and pollen development but normal anther wall development when observed by transmission electron microscopy. These tapetum displayed continuous defective characteristics throughout the anther developmental stages. The transcriptome from flower buds, covering all anther developmental stages, was analyzed and bioinformatics analyses exploring tapetum development-related genes were performed. We identified 1,005 genes differentially the sporogenesis cell stage, pollen mother cell stage, pretetrad stage, and post-tetrad stage. In the sporogenesis cell stage, the sporogenous cells, which give rise to pollen, are visible within locules of sectioned anthers (Goldberg et al. 1993;Scott et al. 1991Scott et al. , 2004), In the pollen mother cell stage, sporogenous cells develop into pollen mother cells (Owen and Makaroff 1995;Stevens and Murray 1981), In the pre-tetrad stage, microspore mother cells enter meiosis while the middle layer is crushed and degenerates. Tapetum becomes vacuolated and the anther undergoes a general increase in size. Tetrads of microspores are free within each locule when meiosis is completed in the tetrad stage. In the post-tetrad stage, the callose wall surrounding tetrads degenerates and individual microspores are released. Microspores generate an exine wall and become vacuolated (Sanders et al. 1999). After these four stages, tapetum degeneration is initiated. The tapetum undergoes generation, development, and apoptosis, providing enzymes for the release of microspores (Varnier et al. 2005). Eventually, all cell remnants are released into the locules, due to tapetal degeneration, and are integrated into the pollen wall as pollen coat material (Papini et al. 1999). This series of events is completed in a relatively short time, and the progression of stages involves changes in the expression levels of many genes. The identification of genes associated with tapetum-related processes helps build a solid foundation for studying the underlying molecular mechanisms of anther development.
Because of difficulties in separating the tapetum cell layer from other anther wall cell layers, the large-scale identification of tapetum cell-specific genes has not previously been achieved. Tapetum-specific genes have been identified by looking for male sterility (MS) mutants in Arabidopsis thaliana. Only a limited number of tapetumspecific genes were reported, these included ACOS5, A6, LTP12, LAP5, TSM1, TAP35, TAP44, and A9 (Ariizumi et al. 2002;de Azevedo et al. 2009;Fellenberg et al. 2008;Hird et al. 1993;Kim et al. 2010;Ma et al. 2012;Paul et al. 1992). Unfortunately, it is not possible to reveal the whole picture of tapetum gene expression by identifying specific tapetum gene mutants one by one. A large number of MS mutants have been identified from natural and artificial mutations in Brassica. Hybridization of A. thaliana cDNA arrays against close relatives that have bigger anthers, such as Brassica species, is a good approach for studying the genome-wide expression of anther-specific genes in Arabidopsis (Amagai et al. 2003).
Pollen grains can be easily isolated, which allows genes expressed in pollen grains to be easily profiled. A number of Arabidopsis pollen grain transcriptomes have been reported by Becker et al. (2003), Pina et al. (2005). Furthermore, a previous pollen transcriptome study by (Honys and Twell 2003) identified 992 pollen-expressed mRNAs, nearly 40 % of which were detected specifically in pollen. They also (Honys and Twell 2004) developed specific spore isolation procedures for Arabidopsis at the pollen developmental stage, and used Affymetrix ATH1 genomic arrays to identify 13,977 male gametophyte-expressed mRNAs in all stages of microsporogenesis, 9.7 % (1,355) of which were male gametophyte specific. However, comparative studies using the tapetum to identify anther wall-specific genes have not been reported in multiple MS lines in which MS mutants occur at different stages of tapetum development. Non-pollen expressed genes (NPGs), are the genes remaining after the exclusion of pollen-specific expressed genes from the genes expressed specifically in the anther. This provides a narrow range for the identification of potential tapetum-specific expressed genes.
This study employs four types of B. oleracea MS lines: Nigra cytoplasm male sterility (NiCMS), Ogura cytoplasm male sterility (OguCMS), recessive male sterility (RGMS) and dominant male sterility (DGMS) (Kang et al. 2008;Fang et al. 2001). Each MS line has a distinct tapetum abortion phenotype and their abnormal characteristics appear successively during anther development. For the large-scale identification of tapetum-specific genes and to gain further insight into downstream cellular reactions of tapetum development, we compared the anther transcriptomes of the four types of B. oleracea MS lines through the heterologous hybridization of B. oleracea mRNA onto an Arabidopsis whole genome oligonucleotide microarray.

Plant materials
Four B. oleracea MS lines which are different from types and origins were used in this study (Table 1) (Kao et al. 1992;Pearson 1972;Fang et al. 1984Fang et al. , 1995: Nigra Cytological analysis using microscopes Light microscopy and transmission electron microscopy (TEM) were used to investigate the cytological differences between the mutants and wild-type. Different sized buds were fixed overnight in 4 % glutaraldehyde with 200 mM phosphate buffer (pH 7.0) and then rinsed overnight in 200 mM phosphate buffer (pH 7.0). Next, buds were postfixed in a solution containing 1 % osmium tetroxide for 2 h and washed in a dehydrated ethanol series for 30 min at each step (2 × 50, 60, 70, 80, 90, and 2 × 100 %). Dehydrated tissues were embedded in Spurr's epoxy resin (Spurr 1969; Ted Pella, Redding, CA) for 3 days at 60 °C, and sectioned at 60-90 nm using a microtome (Leica Ultratome V, LKB, Bromma, Sweden). Anther transverse sections were stained in 4 % uranyl acetate for 20 min and in lead citrate for 3 min. All specimens were analyzed using TEM (H-8100, Hitachi, Tokyo, Japan). For the light microscopy analysis, buds were fixed overnight in FAA (50 % ethanol, 5.0 % glacial acetic acid, 3.7 % formaldehyde), dehydrated in a graded ethanol series (2 × 50, 60, 70, 85, 95, and 3 × 100 %), embedded in Spurr's epoxy resin, and sectioned in 1 µm thick slices using a microtome. Anther transverse sections were stained in 1 % toluidine blue at 42 °C for 1-2 h and observed under a compound microscope (Olympus Model BH2, Tokyo, Japan).

RNA preparation, probe labeling, and microarray hybridization
For total RNA isolation, all flower buds above the last opened flowers of three flowering branches from three MS and three MF plants were collected in duplicate and combined to reduce plant-to-plant variation. Total RNA was isolated using TRIzol following the manufacturer's instructions (Invitrogen, Beijing, China). RNA was further purified using an RNeasy Mini kit (Qiagen China Co., Ltd., Shanghai, China) and the quality was checked using an Data acquisition, normalization, and gene annotation analysis Hybridized microarrays were scanned sequentially for Cy3and Cy5-labeled probes with a laser scanner (G2655AA, Agilent Technologies China Co., Ltd., Shanghai, China) at a resolution of 10 µm and a PMT of 100. The intensities were normalized by linear LOWESS (Yang et al. 2002). The signal was considered positive when the signal/noise value was >5. To determine pollen stage specificity and co-expression information, our data were compared with array data from two other studies. The pollen transcriptome refers to the dataset from (Honys and Twell 2004), and the stamen transcriptome refers to the dataset from (Wellmer et al. 2004). Microsoft Office Excel (Excel 2010, Microsoft China Co., Ltd., Beijing, China) was used to manage and filter the microarray data. Differently expressed genes (DEGs) were functionally categorized based on the ontology annotation of the Arabidopsis genome from the Arabidopsis Information Resource (http://www.arabidopsis. org). Pollen expression type was determined based on the maximum expression value (MaxP) from the four pollen stages (uninucleate microspore, bicellular pollen, tricellular pollen, and mature pollen), the maximum expression value (MaxS) from seven sporophytic tissues (cotyledon, leaves, petiole, stems, roots, root hair zone, and suspension cell cultures), or the value of MaxP/MaxS according to the pollen transcriptome dataset (Honys and Twell 2004).
Late pollen genes were defined as those expressed after microspore mitosis during male gametophyte development, which had continued transcript accumulation during pollen maturation (McCormick 1993).

Reverse transcription-polymerase chain reaction (RT-PCR)
Total RNA was treated with RNase-free DNase (Promega China Co., Ltd., Beijing, China) to remove genomic DNA. RT-PCR reactions were conducted using first-strand 1 3 cDNA synthesized from 2 μg total RNA with Superscript II Transcriptase (Invitrogen, Beijing, China) and a poly-dT 18 primer (Takara, Tokyo, Japan). The cDNAs were then used as templates for RT-PCR with gene-specific primers designed based on reference sequences of B. oleracea ssp. capitata line 02-12 retrieved from BRAD (http://brassicadb.org/brad; Table 2). The Translation Elongation Factor gene EF-1α was used as a positive control togage the quantity of input cDNA among the different samples. The primers used are listed in Supplementary Table S5.

In situ hybridization
Arabidopsis Col-0 inflorescences were embedded in Paraplast (Sigma-Aldrich, Shanghai, China), sectioned at 8-μm thickness and mounted onto precharged slides. For sense and antisense probe synthesis, five coding regions of the NPGs, MEE48, A9, CYP98A8, EXL6, and GGPS5, resulting in 990-, 895-, 749-, 552-, and 656-bp DNA templates, were PCR amplified from flower cDNA using gene-specific forward and reverse primers. A T7 polymerase binding site was incorporated into the forward primer for sense probe amplification and in the reverse primer for antisense probe amplification. Digoxigenin-labeled probes were transcribed off the template using T7 polymerase (Roche, Shanghai, China). Probes were shortened to 200-bp fragments by limited carbonate hydrolysis, and then quantified and hybridized to slides. Tissue fixation, embedding, hybridization, and signal detection were performed as described by (Hooker et al. 2002).

Results and discussion
Cytological defects in the four B. oleracea MS lines The correct spatiotemporal expression of genes in the anther is required for normal tapetum development. We clarified the sequential appearance and characteristics of the cytological defects of the four B. oleracea MS lines by comparing them with the wild-type (Fig. 1I). Light microscopy of the main anther developmental stages revealed that the abortive phenotypes appeared successively in the NiCMS line ( Fig. 1I-7), the RGMS line ( Fig. 1I-14), the OguCMS line ( Fig. 1I-21), and finally in the DGMS line ( Fig. 1I-28). We performed a TEM analysis to characterize the defective tapetum development in the four MS mutant lines (Fig. 1II). In the wild-type line, the sporogenous cells, which give rise to pollen, are visible within locules of sectioned anthers. Concentric rings of other cell types associated with pollen development and release are differentiated around the sporogenous cells during the sporogenesis cell stage ( Fig. 1II-1). Sporogenous cells develop into microspore mother cells and four single distinguishable layers of anther wall and microsporangium could be observed during the microspore mother cell stage ( Fig. 1II-2). Tetrads and tapetum with normal structures, as well as a single microspore tetrad with a central large nucleus, thick cytoplasm and abundant mitochondria develop during the tetrad stage ( Fig. 1II-3).Vacuolated epidermal and endothelial cells, degenerating tapetum, and a free uninucleated microspore, containing a central nucleus, clear nuclear membrane, thick cytoplasm, and abundant plastids, appear in the post-tetrad stage ( Fig. 1II-4). We compared the cytological features of the four male sterility types with those of wild-type. For each, we observed unique defective features. In the NiCMS line, the tapetal cells differentiated inconspicuously, with an indistinguishable middle layer at the sporogenesis cell stage (Fig. 1II-5). In the RGMS line, the tetrad aborted once it was formatted and the tapetal separated from anther wall at the microspore mother stage ( Fig. 1II-6). In the OguCMS line, the tapetums were abnormally activated and thickened continuously when meiosis finished during the early tetrad stage ( Fig. 1II-7).
In the DGMS line, the morphology of the tapetum was not affected (Fig. 1II-8), as reported by (Lou et al. 2007). The development of microspores in the four MS lines was affected at different stages because of the abnormal tapetum development.
Genes with depressed expression levels became the focus of the research because the cytological observations indicated that MS lines were blocked by separate MS proteins. We hypothesized that a sequential developmental interruption model would clarify the gene expression sequence and be in accordance with the cytological results (Fig. 2). Based on the appearance point of the four MS phenotypes (NiCMS earlier than RGMS, RGMS earlier than OguCMS, OguCMS earlier than DGMS), the genes down-regulated only in the NiCMS lines were considered to express earlier than the genes down-regulated in both NiCMS and RGMS lines, and the genes down-regulated in both NiCMS and RGMS lines were considered to express earlier than the genes down-regulated in NiCMS, RGMS, and OguCMS lines. The latest expressing genes would be those that were down-regulated in all four MS lines. Genes involving in anther development mainly express in time series. The accumulation of products produced by early-expressing genes, such as transcription factors and secreted proteins., play important roles in expression of late-expressing genes (Wilson and Zhang 2009).

Signal extraction of the microarrays
To reduce plant to plant and inflorescence branch-to-branch variations, we pooled the RNA obtained from three different plants. Pooling RNA before labeling has the advantage  of reducing the variation due to biological replication and sample handling. Two replicate hybridizations were performed using pooled RNA from three different sterile or fertile plants and the labels were swapped for the second slide, with biological replicates. After the quantification of the signal intensities, the data were normalized to compensate for the nonlinearity of intensity distributions and differences in probe labeling (Fig. 3). Using a signal/noise value >5 to select for positive signals, we detected 12,838 positive signals both in NiCMS control samples and NiCMS  (Fig. 3a) and 0.9808 for the NiCMS samples (Fig. 3b); 0.9754 for the RGMS control samples (Fig. 3c) and 0.9663 for the RGMS samples (Fig. 3d); 0.9708 for the OguCMS control samples (Fig. 3e) and 0.9834 for the OguCMS at different developmental stages (5-8), revealing that the sequence appearance and characteristics of tapetum cytological defects at the sporogenesis cell stage, microspore mother cell stage, pre-tetrad stage and post-tetrad stage. Bar = 2 μm. A the sporogenesis cell stage; B the microspore mother cell stage; C the early tetrad stage; D the uninucleate microspore stage; E the mature pollen stage; and F the dehiscence stage. Sc sporogenesis cell, Ta tapetum, Ml middle layer, Mmc microspore mother cell, En endothecium cell, Ep epidermis cell, Te tetrad, Ms microspore  (Fig. 3f); and 0.9772 for the DGMS control samples (Fig. 3g) and 0.9738 for the DGMS samples (Fig. 3h). The consistency of the two slides, together with the large number of detectable genes, indicates the feasibility of using the Agilent Arabidopsis 2 Oligo array to analyze the B. oleracea transcriptome.

Microarray analysis
An Agilent Arabidopsis 2 Oligo array was used to compare gene expression profiles of the four B. oleracea MS lines to their MF control lines. The array contained 21,500 probes for genes or transcripts of Arabidopsis. Hybridizations of two replicates were performed using pooled RNA from each of three different sterile or fertile plants, and labels were swapped for the second slide with biological replicates. A signal/noise value of >5 was used to select positive signals; this identified 12,837, 13,036, 13,082, and 11,580 genes in NiCMS, RGMS, OguCMS, and DGMS lines, respectively. After combining these four datasets 13,984 genes (65.0 %) were detected as positive signals in at least one of the MS lines. This percentage of identified genes was similar to previous research that identified 14,660 (64.5 %) genes in six organs and structures, including the inflorescences, at different Arabidopsis floral stages (Zhang et al. 2005). Genes with differential mRNA abundance levels (ratios >3, or <0.33) in the two replicate slides were selected for further analysis. To confirm the microarray profiling data, nine genes were randomly selected for semi-quantitative RT-PCR analyses, and their expression patterns were found to be consistent with the microarray results (Fig. 4, Supplementary Table S1).
There were 544, 338, 526, and 209 down-regulated genes, and 5, 45, 151, and 15 up-regulated genes detected in the NiCMS, RGMS, OguCMS, and DGMS lines, respectively (Supplementary Table S2 and S3). After redundant genes were merged, 838 DEGs were down-regulated (group 1) and 188 up-regulated (group 2) in at least one of the four MS lines. The overlap of group 1 and group 2 contained 21 DEGs, including the two tapetum-related genes TDF1 and MYB 103. These genes play vital roles in tapetum synthesis and degeneration, and the sporopollenin monomer biosynthesis process (Higginson et al. 2003). These genes were down-regulated in the NiCMS line but up-regulated in RGMS and OguCMS lines. Following the removal of these 21 DEGs, 1,005 non-redundant DEGs were identified in this research.

Identification of tapetum-specific genes
The tapetum has been studied in many kinds of plants using MS defects, and, due to the difficulty of tapetum isolation, only four genes have been reported to be tapetum specific by MS mutant gene cloning (Suwabe et al. 2008). Therefore, we developed an approach to identify tapetumspecific genes on a large-scale, and at the developmental stages in which they are expressed. Fortunately, the separation of male gametophytes is easy, and numerous male gametophyte-specific genes have been identified from multiple MS mutants and global transcriptome analyses.
To identify genes specifically involved in tapetum development, we removed genes known to be expressed in sporophytic tissues based on the findings of Honys and Twell 2004. Their research identified 17,677 sporophytic genes,  Table S4). This left 280 DEGs thought to be expressed specifically in the anthers. Furthermore, 176 male-gametophyte expressed DEGs identified by (Honys and Twell 2004) were removed from the 280 DEGs specifically expressed in the anthers, leaving 104 NPGs (Fig. 5). As the four MS lines had tapetums aborted at successive developmental stages, we were able to identify 104 NPGs as anther wall-specific genes, the vast majority of which were considered tapetum-specific genes because their anther walls developed normally with the exception of the distinct abortion of the tapetums (Table 2). We cannot completely rule out that some genes expressed from other tissues are included in the 104 NPGs, although this probability is very low.
Male sterility genes block the developmental pathways of anthers at a certain stage, and lead to abnormal anther development. Because each of the four MS lines showed distinct tapetum abortive phenotypes, and their abnormal characteristics appeared successively during anther development, we could deduce the expression sequence of the NPGs by their expression patterns. The expression sequence of the 55 NPGs could be determined according to the sequential appearance of the tapetum abortive phenotypes (Table 2). First, 28 NPGs that were only down-regulated in the NiCMS line were expressed, followed by seven NPGs down-regulated in both the NiCMS and RGMS lines, then 16 NPGs down-regulated in the NiCMS, RGMS, and OguCMS lines, and finally four NPGs down-regulated in the four MS lines were expressed. It is thought these 55 NPGs constitute the main stream of tapetum development, while the remaining 49 NPGs belonged to bypass ways which also play roles in tapetum development. These results showed that the tapetum development is strongly correlated with gene expression patterns and anther developmental timing. In the eight previously reported tapetumspecific genes, five (ACOS5, A6, LTP12, LAP5, and A9) were detected in our filtered microarray results, although the other three tapetum-specific genes (TSM1, TAP35, and TAP44) were detected in our microarray analysis at variable expression levels in the four MS lines. Because of the strict filter these three tapetum-specific genes were not included in our set of 104 NPGs. There were many noticeable features in the up-regulated genes, with 13 NPGs up-regulated in the OguCMS line, five NPGs up-regulated in both the RGMS and OguCMS lines, and one NPG up-regulated in both the RGMS and DGMS lines (Table 2). This is consistent with the cytological observations of different abnormal tapetum development in the four MS lines.
The tapetum developmental network is regulated by many genetic pathways (Wilson and Zhang 2009). MS mutants occurring at different developmental stages lead to abnormal downstream reactions, including altered tapetum structure and gene expression patterns. These changes are evoked by the presence of MS-associated proteins (Fujii et al. 2010). Only a few previous studies have analyzed gene expression patterns in B. oleracea MS lines (Kang et al. 2008). These studies compared anther gene expression profiles in MS lines with their corresponding fertile lines, allowing for the preferentially expressed anther genes to be identified. Despite their clear contribution to anther developmental pathways, downstream expressed anther wall-specific genes have been overlooked because of the removal of male gametophyte-specific expressed genes.

Function of genes arrested by the four types of B. oleracea MS lines
The distribution of the NPGs was determined in the gene ontology data set (MAS 3.0, http://bioinfo.capitalbio.com/ mas3/) and found to cover virtually all functional categories (Fig. 6). The classification of functional categories revealed that some were enriched in DEGs that had reduced expression levels, including structural molecules, transporters, and physiological processes. These categories are associated with metabolic activities that are dynamic in the tapetum, suggesting a positive role of the tapetum in the regulation of metabolic functions. As we are interested in the genetic mechanism of tapetum abortion in the four MS lines, genes specifically expressed in the tapetum were further analyzed to identify tapetum abortive phenotypes. Comparison between differentially expressed genes (DEGs) and genes expressed in sporophytic and gametophytic tissues to identify non-pollen expressed genes (NPGs) of B. oleracea. A total of 17,677 sporophytic-expressed genes and 13,977 male gametophyteexpressed genes were detected in Honys' research (Honys and Twell 2004). Of the DEGs, 725 genes were also sporophytic-expressed genes and 613 were also male gametophyte-expressed genes Special attention was paid to abortive mechanisms leading to tapetum dysfunction by gene regulation networks within the 1,005 DEGs in any of the MS lines. According to previous studies, 22 DEGs were proven to be related to tapetum development (Supplementary Table S6). Among these 22 genes, 11 of the DEGs were reported to play important roles in tapetum development (ULT2, TDF1, PGA4, PAB3, TKPR2, PAB5, SHT, ACT12, LAP6, HMA4, and ATA20). More importantly, the other 11 DEGs were identified as NPGs in this study (LTP12, CYP703A2, CYP704B1, LAP5, ACOS5, ABCG26, MYB103, MYB99, WBC27, ATBHLH089, and ATBHLH091) ( Table 2). To determine the expression patterns of NPGs, we compared them with other mutant transcriptomes that have been analyzed by bioinformatics filtering. The results showed that 27 (26 %) NPGs were also detected in an anther-specific expressed gene set (Xu et al. 2014), 32 (31 %) NPGs were detected in a stamen-specific expressed gene set that Five Brassica oleracea NPG mRNAs were localized by in situ hybridization to sections taken from developing anther locules of wild type (Col-0) Arabidopsis flowers. Top line using each of the five non-pollen expressed genes (NPGs) control sense probes. Bottom line using each of the five NPGs anti-sense probe. Dark precipitate indicates hybridization of the probe. T tapetum, PG pollen grain, PMC pollen mother cell. Bar = 70 µm excluded pollen-specific expressed genes (Ma et al. 2012). To validate microarray results in anthers, we performed in situ hybridization using five randomly chosen NPG-derived probes and developing wild-type A. thaliana flowers (Fig. 7). We used MEE48, A9, CPY98A8, and EXL6 probes that hybridized with anthers when they had developed to the uninucleated microspore stage, respectively. GGPS5 probes were hybridized with anthers when they developed to the pollen mother cell stage. No hybridization signal was observed in locules with microspores and tapetum using any of the NPG sense control probes. However, NPG antisense probes resulted in varying degrees of hybridization to the tapetum. These results showed that our data set, which was generated by bioinformatics filtering, was reliable. Therefore, the NPGs can be seen as potential tapetum-specific expressed genes.

Conclusions
In this study, four different types of B. oleracea MS lines, covering the whole of anther development, were sampled for microscopy and microarray analysis. A systematic study of the tapetum developmental and molecular phenotypes of the four B. oleracea MS lines was performed using the Agilent Arabidopsis 2 Oligo array, which contains 25,000 probes to known or predicted genes. This is the first large-scale trial to explore the spatial and temporal expression patterns of tapetum-specific gene in B. oleracea MS lines. The results revealed that most tapetum-specific genes were expressed in a stage-specific manner.
The most significant phenotype of the four B. oleracea MS lines is that the anther walls are normally developed, with the exception of the distinct tapetum abortive phenotypes. These abnormal characteristics appeared successively during anther development (Fig. 1). The abortive phenotypes first appears in NiCMS, then in RGMS, followed by OguCMS, and finally in DGMS. A sequential developmental interruption model was proposed to clarify the expression order of the DEGs in plants. MS-associated proteins might directly or indirectly regulate some of the genes involved in tapetum synthesis and degeneration, and then block the expression of a large number of genes involved in normal microspore development. Eliminating sporophytic-and male gametophyte-specific expressed genes identified in the NPGs as anther wall-specific genes, resulted in the majority of remaining genes being tapetum specific. We defined the expression sequence of 55 NPGs based on the sequential developmental interruption model and determined that they are involved in structural molecules, transporters, and physiological processes related to cell wall modification and catalytic activities. The categories suggested the metabolic role of the tapetum in the regulation of anther development. These results outline a methodology to retrieve information on hard to isolate tissues through the comparison of global expression with gene expression in easily obtained mutants. The relationship of anther-expressed genes can be clarified by comparing the sequential cytological appearance of defects in multiple independent MS lines with the same tissues in the wild-type.