Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Comprehensive Network Analysis of Anther-Expressed Genes in Rice by the Combination of 33 Laser Microdissection and 143 Spatiotemporal Microarrays

  • Koichiro Aya ,

    Contributed equally to this work with: Koichiro Aya, Go Suzuki, Keita Suwabe, Tokunori Hobo

    Affiliation Bioscience and Biotechnology Center, Nagoya University, Nagoya, Japan

  • Go Suzuki ,

    Contributed equally to this work with: Koichiro Aya, Go Suzuki, Keita Suwabe, Tokunori Hobo

    Affiliation Division of Natural Science, Osaka Kyoiku University, Kashiwara, Japan

  • Keita Suwabe ,

    Contributed equally to this work with: Koichiro Aya, Go Suzuki, Keita Suwabe, Tokunori Hobo

    Affiliations Graduate School of Life Sciences, Tohoku University, Sendai, Japan, Graduate School of Bioresources, Mie University, Tsu, Japan

  • Tokunori Hobo ,

    Contributed equally to this work with: Koichiro Aya, Go Suzuki, Keita Suwabe, Tokunori Hobo

    Affiliation Bioscience and Biotechnology Center, Nagoya University, Nagoya, Japan

  • Hirokazu Takahashi,

    Affiliation Graduate School of Agricultural and Life Sciences, The University of Tokyo, Tokyo, Japan

  • Katsuhiro Shiono,

    Affiliation Graduate School of Agricultural and Life Sciences, The University of Tokyo, Tokyo, Japan

  • Kentaro Yano,

    Affiliation School of Agriculture, Meiji University, Kawasaki, Japan

  • Nobuhiro Tsutsumi,

    Affiliation Graduate School of Agricultural and Life Sciences, The University of Tokyo, Tokyo, Japan

  • Mikio Nakazono,

    Affiliations Graduate School of Agricultural and Life Sciences, The University of Tokyo, Tokyo, Japan, Graduate School of Bioagricultural Sciences, Nagoya University, Nagoya, Japan

  • Yoshiaki Nagamura,

    Affiliation National Institute of Agrobiological Sciences, Tsukuba, Japan

  • Makoto Matsuoka , (MM); (MW)

    Affiliation Bioscience and Biotechnology Center, Nagoya University, Nagoya, Japan

  • Masao Watanabe (MM); (MW)

    Affiliations Graduate School of Life Sciences, Tohoku University, Sendai, Japan, Faculty of Science, Tohoku University, Sendai, Japan


Co-expression networks systematically constructed from large-scale transcriptome data reflect the interactions and functions of genes with similar expression patterns and are a powerful tool for the comprehensive understanding of biological events and mining of novel genes. In Arabidopsis (a model dicot plant), high-resolution co-expression networks have been constructed from very large microarray datasets and these are publicly available as online information resources. However, the available transcriptome data of rice (a model monocot plant) have been limited so far, making it difficult for rice researchers to achieve reliable co-expression analysis. In this study, we performed co-expression network analysis by using combined 44 K agilent microarray datasets of rice, which consisted of 33 laser microdissection (LM)-microarray datasets of anthers, and 143 spatiotemporal transcriptome datasets deposited in RicexPro. The entire data of the rice co-expression network, which was generated from the 176 microarray datasets by the Pearson correlation coefficient (PCC) method with the mutual rank (MR)-based cut-off, contained 24,258 genes and 60,441 genes pairs. Using these datasets, we constructed high-resolution co-expression subnetworks of two specific biological events in the anther, “meiosis” and “pollen wall synthesis”. The meiosis network contained many known or putative meiotic genes, including genes related to meiosis initiation and recombination. In the pollen wall synthesis network, several candidate genes involved in the sporopollenin biosynthesis pathway were efficiently identified. Hence, these two subnetworks are important demonstrations of the efficiency of co-expression network analysis in rice. Our co-expression analysis included the separated transcriptomes of pollen and tapetum cells in the anther, which are able to provide precise information on transcriptional regulation during male gametophyte development in rice. The co-expression network data presented here is a useful resource for rice researchers to elucidate important and complex biological events.


Recent developments in high-throughput microarray, next-generation sequencing, proteomic analysis, and the accumulated functional genomics data across species have enabled us to utilize integrated large-scale data of gene-expression, protein-protein interaction, and phenotype. There is now an increasing need for integrated analysis at a system biology level, to gain an understanding of the complex relationships between gene-product interactions and biological events (e.g. phenotype). An in-silico-derived co-expression network is constructed from large-scale gene expression profiles, and is based on the assumption that genes with similar expression patterns are likely to interact with each other at the molecular or physiological level. In some model plants as well as animals, this strategy has been broadly used to predict integrated networks in association with protein-protein interaction data [1], the structural information on metabolic pathways [2][4], and the functions of gene products [5].

To date, a model plant, Arabidopsis thaliana, has been subjected to thousands of microarray experiments and the results have been deposited in publicly-available online databases [6]. By using these very large expression datasets, high-resolution co-expression networks have been constructed and these are available as online information resources in Arabidopsis. Such useful information resources have enabled Arabidopsis researchers to identify novel factors involved in cell wall synthesis [3], [4] and the aliphatic glucosinolate biosynthesis pathway [5]. On the other hand, in a model crop for the grass family, rice (Oryza sativa), there are few resources for co-expression network analysis, because of inadequate proliferation of available rice microarray datasets derived from various different platforms and a low number of datasets with few examples of publicly available data [7]. Moreover, in rice, there have been fewer case studies to estimate the effectiveness of co-expression networks, compared to those in Arabidopsis. Thus, a reliable large-scale study of co-expression network analysis is required to drive forward rice research at this time.

Male gametophyte development might be a good target for co-expression network analysis in rice, because pollen transcriptome can be easily and precisely analyzed using larger rice flowers than those in Arabidopsis. In flowering plants, pollen development takes place within a male reproductive organ, the anther, and is controlled precisely by four sporophytic cell layers of the anther (tapetum, middle layer, endothecium and epidermis), which surround the gametophytic pollen grains [8]. After differentiation of the male germline, pollen mother cells form tetrads of haploid microspores via meiosis, and the tetrad microspores are connected to each other by a callose wall. Tapetum cells play an important role in degradation of the callose wall, which allows the microspores to be released into the anther locule. The released microspores subsequently mature into pollen grains through cell division and pollen wall formation. During the course of pollen maturation, the tapetum starts to degenerate and provides the various materials for pollen wall formation on the surface of pollen grains. So far, by using in situ hybridization [9][15] or microarrays [16], [17], many genes expressed in anthers have been identified in Arabidopsis [9], [13], [17], Brassica napus [10], rice [11], [12], [14], and Lotus japonicus [15], [16]. These studies have revealed the complex patterns of gene expression in both the gametophytic pollen/microspore and sporophytic tapetum cells, which are different but influence each other; consequently, it has been difficult to analyze the precise regulatory interactions of gene expression between pollen and tapetum by examination of the whole anther transcriptome In this context, transcriptome data from separated pollen and tapetum cells are necessary to compare with the large sets of other microarray data in order to achieve co-expression network analysis of male gametophyte development in plants. Recently, we have performed transcriptome analysis using 44 K microarrays with RNAs extracted independently from pollen and tapetum cells within the rice anther by laser microdissection (LM) technology [18], [19].

By using our LM-microarray [18], [19] and other publicly available microarray datasets [7], in this study, we have conducted comprehensive co-expression analysis and constructed the co-expression subnetworks responsible for two important biological events during anther development, meiosis and pollen wall synthesis. The meiosis network contained many putative meiotic genes as well as known meiotic genes, which could play meiotic roles such as meiosis initiation and recombination. Furthermore, from the pollen wall synthesis network, we efficiently identified several candidate genes involved in the sporopollenin biosynthesis pathway. Taken together, our co-expression network has the potential to be a powerful resource to dissect various important biological events and/or to isolate novel genetic factors regulating development and differentiation of the male reproductive organ in plants at a system biology level. In addition, this study is an important development as it is one of the few demonstrations of the efficiency of co-expression network analysis in rice.


Dataset content

We have previously conducted 44 K agilent LM-microarray analysis of the microspore/pollen and tapetum cells of japonica rice, Oryza sativa cv. Nipponbare, during anther development from meiosis to tricellular pollen stages [18], [19]. In addition to the above study, a LM sample in premeiosis stage was prepared in this study, for more precise gene expression profiling during anther development. The overall LM-microarray data, including three novel microarray datasets, has been deposited at the Gene Expression Omnibus (GEO;; accession number GSE 29217). A detailed procedure for the LM-microarray is described in our previous reports [18], [19].

Recently, co-expression analysis datasets have become a powerful tool to extract the transcriptional networks of genes involved in various biological events in plants [20]. Therefore, we conducted microarray-based co-expression analysis using our LM-microarray data to gain biological insights into gene expression networks mediating rice anther development. Because popular methods that derive regulatory networks from gene expression data, in general, require large amounts of gene expression data, here we used 143 spatiotemporal expression profiles of 48 various vegetative, and reproductive organs and tissues of rice deposited in RicexPro [7], in addition to our 33 LM-microarray data of rice anthers [18], [19] (Table S1).

A 44K agilent microarray slide contains 29,864 distinct rice genes. Using the 176 microarray, first, we examined the spatiotemporal expression pattern by identifying the highest signal intensity for each gene from different stages and organs (Figure 1, Tables S2, S3, S4, S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22, S23, S24, S25). Among the 29,864 genes analyzed, about 10,600 genes were found to be expressed preferentially in at least one of the anther developmental stages or tissues (Figure 1, Tables S7, S8, S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, and S19 for array data of genes preferentially expressed in anther). The remaining approximately 19,300 genes were expressed in the other 44 vegetative or reproductive tissues (Tables S2, S3, S4, S5, S6, S20, S21, S22, S23, S24, S25). Such a large proportion of anther preferentially expressed genes indicates that many genes are involved in anther development and, consequently, analysis of co-expression networks of anther preferentially expressed genes could potentially provide good examples to demonstrate the application of microarray datasets.

Figure 1. Distribution of rice genes with tissue-preferential expression in the 176 microarray datasets.

The 29,864 distinct rice genes on a 44K microarray were classified into 24 spatiotemporal categories (different stages and tissues including LM-separated anther tissues) according to the highest signal intensity among the 176 microarray datasets. Gene number indicates the number of genes in each category. The spatiotemporal categories restricted to the anther are indicated in red. MEI, meiosis; TET, tetrad; UN, uninuclear; BC, bicellular; TC, tricellular.

Biological significance of expression similarity

We first conducted an expression coherence (EC) analysis using gene ontology (GO) terms to examine the biological significance of expression similarity in the 176 microarray datasets used in this study. EC analysis is a statistical analysis to determine whether gene members belonging to a predefined functional gene set are correlated with each other at the transcriptional level [21]. The EC score is a measure of the expression similarity within a set of predefined functional genes [21]; consequently, a higher EC score is obtained, when most of genes with the same GO term are co-expressed with each other. In our EC analysis, the EC scores of all three GO groups (Biological Process, BP; Molecular Function, MF; and Cellular Component, CC) were higher than that expected by random sampling (Figure 2). Among the three GO groups, CC showed the highest expression similarity, in which approximately 17% of categories exhibited higher EC scores than random sampling at threshold EC score of 0.05, while approximately 12% and 8% did so in BP and MF, respectively (dotted line in Figure 2). For genes annotated according to the BP group, the “photosynthesis, light harvesting” category showed the highest EC scores (EC = 0.868, 17 genes), whereas the scores of the categories “small ribosomal subunit” (EC = 0.433, 21 genes) and “threonine-type endopeptidase activity” (EC = 0.322, 24 genes) were highest in the CC and MF groups (Table S26). These results confirm that the levels of expression similarity estimated from the microarray datasets in this study are good reflections of functional similarity.

Figure 2. EC analysis of the 176 microarray data of rice.

BP, MF, and CC represent Biological Process, Molecular Function, and Cellular Component groups of GO, respectively. The total number of categories in each GO group is indicated in parentheses. In the three GO groups (BP, blue; MF, red; and CC, green), the EC score was calculated for each GO category and compared with that of random sampling (purple) to estimate the statistical significant level. Fraction of categories (y axis) indicates the ratio of the number of GO categories with higher EC score than each threshold EC score (x axis) to the total number of GO categories.

Construction of the meiosis-specific network

By using the Pearson correlation coefficient (PCC) method with the mutual rank (MR)-based cut-off (see Materials and Methods), we identified the final dataset of rice co-expression networks generated from all the 176 microarray data used in this study, which contains 24,259 genes (nodes) and 60,449 genes pairs (edges). From this entire dataset of the rice co-expression network, we are able to construct a subnetwork around gene(s) of interest input as guide(s).

To examine whether our co-expression network analysis can identify useful transcriptional networks related to anther development, we first constructed a co-expression subnetwork centered on 9 known meiotic genes (guide genes; red circles in Figure 3A), PAIR1 [22], PAIR2 [23], PAIR3 [24], MEL1 [25], OsRAD214 [26], DMC1A and 1B [27], OsMER3/OsRCK [28], and OsSDS [28], whose functions in meiosis have been experimentally demonstrated in rice. The expression profiles of these nine genes after normalized by the VSN algorism (Figure S1A) are well corresponding to the previous results of their expression patterns: indicating that the VSN normalization was a suitable method for the construction of co-expression network. The meiosis-specific network consisted of a large main cluster with the 7 guide genes, and two independent minor clusters of OsMER3 and PAIR1 (Figure 3A), which in total contained 187 genes and 346 gene interactions (Table S27). The main cluster also contained ZEP1 (rice ZYP1 ortholog) and MEL2 (blue circles in Figure 3A), which have most recently been reported as rice meiotic genes [29], [30]. Moreover, PAIR2 and ZEP1 [23], [29], whose products are both constituents of the synaptonemal complex, were directly connected in the network (red bold lines in Figure 3A). Good association of the 7 meiotic guide genes with each other and with the additional 2 meiotic genes in the main expression cluster indicates that genes involved in meiosis could be strictly co-regulated at the expression level, and thus the meiotic events are potentially good subjects for co-expression analysis. This is also supported by the results of GO analysis (Figures 3B and 3C, Table S28), in which DNA-repair-related GO terms (such as “chromosome organization,” “DNA unwinding involved in replication,” “Mismatch repair,” and “Mismatched DNA binding”) appear frequently in genes in the meiosis-specific network.

Figure 3. The meiosis-specific co-expression network in rice.

(A) The co-expression subnetwork constructed using 9 reported rice meiotic genes as guide genes (red circles). The network contains the most-recently reported meiotic genes (blue circles), putative meiotic genes (green circles) and other genes (open circles). A red bold line indicates the direct connection of PAIR2 and ZEP1 in the co-expression network. A link between two nodes indicates direct interaction with PCC >0.64 and MR <10. The subnetwork vicinity is extracted by taking 2 steps out from a guide gene. (B, C) Percentage of genes classified in enriched GO terms of BP (B) or MF (C) for the genes in the meiosis-specific network (red) relative to for 11,456 genes (blue) that had at least one connection at the PCC (>0.64) and MR (<10) thresholds. Enrichment analysis is performed using the hypergeometric distribution (P value <0.05).

Known or putative meiotic genes in the subnetwork are listed in Table 1. The meiosis-specific network contained cyclin genes and other potential rice orthologs of meiotic genes (Figure 3A). In the network, genes related to meiotic recombination (e.g. DMC1s and OsMSH4/5) are located near a core of the network, and genes functioning in other meiotic events (e.g. MEL1, PAIR3, OsRAD214, and OsSDS) surround them with the small independent cluster of PAIR1 (Figure 3A). The independent small cluster of OsMER3 (a meiotic recombination-related gene) is due to its higher vegetative expression, which could indicate a bifunctional role in meiosis and other biological events.

Table 1. Known or putative meiotic genes in the meiosis-specific subnetwork.

Among genes related to “transition from mitosis into meiosis” in Table 1, MEL1 and MEL2 encode ARGONAUTE and RNA-recognition-motif protein, respectively [24], [30], and putative OsAML1 (Os01g0907900) also encodes an RNA-binding protein [31], which is highly similar to MEI2 of Schizosaccharomyces pombe [32]. Together with PWWP- and SET-domain proteins (Table 1), MEL1, MEL2 and OsAML1 might regulate the initiation of meiosis in rice through molecular mechanisms related to RNA metabolism, chromatin remodeling and/or transcriptional regulation. On the other hand, genes related to “meiotic recombination” are well integrated in the network (Figure 3A, Table 1). Co-expression of recombination-related genes in the network is consistent with previous reports that DMC1A and DMC1B co-ordinate with RAD17 [33], and that MSH4 and MSH5 function as a complex [34].

In addition, it can be concluded from the GO analysis (Figure 3B and 3C, Table S28) that various additional meiotic genes could be included in the network as unknown genes (open circles in Figure 3A). This indicates that novel meiotic genes could be mined from the present data, and that novel interactions between known meiotic genes could also be identified from the unknown-gene-mediated connections in the network. Meiosis is an extremely specialized biological event, thus co-expression network analysis has the potential to achieve maximum success in the identification of gene interactions.

Construction of the pollen wall synthesis network

As a second example system, we analyzed the pollen wall synthesis network to systemically identify the genes that modulate the formation of the outer pollen wall, the exine. Previous biochemical analyses have shown that the exine layer consists mainly of sporopollenin, a polymer of phenylpropanoid and lipidic monomers covalently linked by ether and ester linkages [35][37]. When three genes for rice sporopollenin biosynthesis or transport (CYP703A3, CYP704B2, and Osc6) were selected as guide genes [38][40], a major subnetwork covering all guide genes was identified, which consisted of 108 genes and 278 gene interactions (Figure 4A, Table S29). Again, the expression profiles of these three genes after normalized by the VSN algorism (Figure S1B) are well corresponding to the previous results of their expression patterns. Classification according to GO terms showed that this subnetwork was significantly enriched with fatty acid and several metabolic related terms, including “phospholipid metabolic process”, “lipid metabolic process”, “3-oxoacyl-[acyl-carrier-protein] reductase (NADPH) activity”, and “phospholipase A2 activity” (Figure 4B and 4C, Table S30). This is clearly consistent with previous knowledge of sporopollenin biochemistry [35][37], indicating that this co-expression network reflects the underlying molecular mechanism for exine formation.

Figure 4. The pollen wall synthesis co-expression network in rice.

(A) The co-expression subnetwork constructed using 3 reported rice genes for sporopollenin biosynthesis or transport as guide genes (red circles). The network contains putative genes related to pollen wall synthesis (green circles). Red bold lines indicate the direct connection of CYP703A3, CYP704B2, OsACOS5, OsPKSs, and OsTKPRs in the co-expression network. A blue bold line indicates the direct connection between Os01g0293100 (a bHLH-type transcription factor) and CYP704B2. A link between two nodes indicates direct interaction with PCC >0.64 and MR <10. The subnetwork vicinity is extracted by taking 2 steps out from a guide gene. (B, C) Percentage of genes classified in enriched GO terms of BP (B) or MF (C) for the genes in the meiosis-specific network (red) relative to for 11,456 genes (blue) that had at least one connection at the PCC (>0.64) and MR (<10) thresholds. Enrichment analysis is performed using the hypergeometric distribution (P value <0.05).

In addition to the above three guide genes, this network contained several fatty acid metabolic genes, such as OsMS2, OsACOS5 (ACYL-COA SYNTHETASE5), OsPKSs (POLYKETIDE SYNTHASE), and OsTKPRs (TETRAKETIDE a-PYRONE REDUCTASE), homologs of which have been demonstrated to be important for sporopollenin biosynthesis in Arabidopsis [41][44] (Figure 4A; Table 2). In the latest model of the sporopollenin biosynthetic pathway [44], [45], fatty acid precursors are esterified to CoA by ACOS5, and then hydroxylated by CYP703A and CYP704B to produce substrates of the subsequent ACOS5 reaction (Figure 5). Subsequently, ACOS5, PKSs, TKPRs, and MS2 mediate the biochemical reactions to produce sporopollenin precursors via fatty alcohols (Figure 5). Consistently, our pollen wall specific network confirmed significant gene interactions between these molecular players (red bold lines in Figure 4A). Furthermore, in order to identify an as-yet-unknown specific thioesterase producing the CYP703A/CYP704B substrates, we compared the expression patterns of 15 rice thioesterase genes with those of the sporopollenin biosynthetic genes, and detected one thioestrase gene (Os09g0517700) that was co-expressed with OsPKSB and OsTKPR2 at the significant PCC level (Table 3). Although Os09g0517700 was not found on the pollen wall synthesis subnetwork (Figure 4A) because of higher MR values, it might play an important role in the sporopollenin biosynthetic pathway.

Figure 5. The current model of the sporopollenin biosynthetic pathway.

Table 2. Known or putative rice genes involved in pollen wall synthesis in the pollen wall synthesis subnetwork.

Table 3. Expression similarities between 15 putative thioesterase genes and 7 sporopollenin biosynthetic genes in rice.

In addition to the previously characterized genes, several novel genes, whose products participate in fatty acid metabolism, secondary metabolism, and lipid transport, were identified within this network (Figure 4A and Table 2). It is probable that these genes are associated with novel metabolic processes related to sporopollenin biosynthesis or sporopollenin transfer from tapetal cells to anther locules (Figure 5).

It has also been reported that some MYB and bHLH transcriptional factors, including GAMYB, AMS, UDT1, and AtMYB103, regulate the expression of sporopollenin biosynthetic genes [38], [46][48]. The 108 co-expressed genes in the pollen wall synthesis network included the two candidates (Os01g0293100 and Os03g0296000) for the transcriptional regulation for sporopollenin biosynthesis (Figure 4A and Table 2). Os03g0296000 is a rice homolog of Arabidopsis MYB35 (TDF1) [49] responsible for tapetum development, whereas Os01g0293100 is a novel bHLH type transcription factor, not previously studied. In this network, Os01g0293100 is directly connected to CYP704B2 (blue line in Figure 4A), suggesting that Os01g0293100 could be involved in sporopollenin biosynthesis via modulating CYP704B2 expression (Figure 5). Thus, the pollen wall synthesis network in the present co-expression analysis efficiently provides us with useful information to identify novel gene functions and interactions.


Co-expression network analysis in plants

In Arabidopsis, useful databases for co-expression analysis are being constructed, and several novel genes have been identified from these data. Aoki et al. [3] and Obayashi et al. [50] listed the available co-expression databases in Arabidopsis, including CressExpress [51], CoP [52], ACT [53], GeneCAT [54], and ATTED-II [55]. As described in these review articles, novel genes for enzymes involved in metabolic pathways, signal factors of biological events (including transcription factors), and components of protein complexes have been identified by using the publicly available databases or analyzing a laboratory's own co-expression networks. For example, irregular xylem8 (IRX8) and irregular xylem13 (IRX13), which are involved in secondary cell wall formation, were identified from co-expression analysis guided by members of a cellulose synthase (CESA) family [4]. IRX8 encodes glycosyltransferase (GT8), which is involved in pectin synthesis, whereas IRX13 encodes an unknown protein similar to an arabinogalactan protein. Involvement of IRX8 and IRX13 in secondary cell wall formation was demonstrated by T-DNA insertion mutagenesis. Similarly, Oikawa and co-workers [56] conducted a comparative co-expression analysis between Arabidopsis and rice, and proposed a model for secondary cell wall formation of Arabidopsis and rice, based on the combined approach of co-expression and predicted cellular localization. More recently, by using Arabidopsis co-expression networks, UGP3, a gene coding for UDP-glucose pyrophosphorylase 3 implicated in sulfolipid biosynthesis [57], NAI2 implicated in endoplasmic reticulum body formation [58], BTS, a BRUTUS gene coding for bHLH-type transcription factor co-expressed with POPEYE in iron homeostasis [59], and STOMAGEN, encoding secretory peptide regulating stomatal density [60], were successfully identified. Thus, co-expression network analysis is a powerful tool for the identification of novel Arabidopsis genes, and such an approach is undoubtedly also applicable to rice research.

In the present study, we performed co-expression network analysis of rice genes by using 176 datasets of 44 K agilent microarrays (including 33 LM-microarray datasets of the male reproductive organs), and constructed two different co-expression subnetworks (meiosis-specific and pollen-wall-synthesis networks) to demonstrate the efficiency of this technique. Integrative application of such comprehensive microarray data enabled us to construct high-resolution networks (Figures 3A and 4A), which included useful information for the dissection of gene regulation in biological processes. This demonstrates that we can now perform co-expression screening to pursue our interests in specific genes or biological events in rice, as well as in Arabidopsis, by the use of our reliable network datasets and the other effective rice databases reported to date [61][64].

To the best of our knowledge, this is one of the few reports of the application of co-expression network analysis to specific biological processes in rice. Four rice co-expression databases are now publicly available. In the most recent version of ATTED-II, the co-expression network can be constructed by using the mutual rank of the Pearson's correlation coefficient (PCC) in rice as well as in Arabidopsis [61]. Ficklin et al. [62] demonstrated application of the Gene Co-expression Network Browser, a website providing a rice co-expression network constructed by random matrix theory and weighted correlation network analysis. In addition, RiceArrayNet [63], based on PCC, and OryzaExpress [64], based on correspondence analysis, are available as rice co-expression databases. These four recent papers [61][64] reported the availability of the rice co-expression databases, but did not provide a demonstration of the identification of genes in relation to specific biological events. Therefore, our present report is a valuable addition to the understanding of the utility of rice co-expression network analysis, together with discussion of specific examples of successful construction of meiosis-specific and pollen-wall-synthesis networks.

Application of co-expression network analysis in sexual reproduction research

Our co-expression analysis included the LM microarray data of male reproductive organs, which can discriminate transcriptomes of gametophytic pollen/microspore cells and sporophytic tapetum cells in the anther, and will provide more precise information on transcriptional regulation of anther genes than has been available previously. None of the examples of gene identification using Arabidopsis co-expression networks described above focus on the microarray databases of reproductive organs, conditions or developmental stages [4], [56][60]. Hence, our co-expression network datasets, focusing on male gametophyte development, shed light on a new and interesting aspect of the potential of co-expression analysis. It may be interesting to apply co-expression network analysis to other LM microarray datasets targeted at specific tissues and cells.

Our meiosis-specific and pollen-wall-synthesis networks reveal that many previously unidentified genes are co-expressed in these specific reproductive events. A reverse genetics approach to the study of these genes will lead to identification of novel meiosis-specific and pollen-wall-synthesis genes. Because there are many interesting and specific biological events during male gametophyte development, the very large co-expression datasets presented here will provide an excellent opportunity for plant researchers to study the molecular biology of sexual reproduction in rice.

Materials and Methods

Plant materials

Rice (O. sativa L. ssp. japonica cv. Nipponbare) plants were grown in a greenhouse under normal conditions. Anthers at the premeiosis stage were collected after their developmental stages were confirmed by DAPI staining using one of the six anthers of each flower.


Anthers were fixed in Farmer's fixative (ethanol: acetate = 3∶1) overnight at 4°C. Dehydration and paraffin embedding were performed using a microwave processor [31]. Paraffin-embedded sections were cut to a thickness of 16 µm and mounted on PEN membrane glass slides (Molecular Devices, Sunnyvale, CA, USA) for LM. To remove paraffin, slides were immersed in 100% xylene (twice), 50% xylene/50% ethanol, and 100% ethanol (v/v), for 5 min at each step and then air-dried completely at room temperature. Three or four individual flowers were used for each LM experiment. LM was performed using the Veritas Laser Microdissection System LCC1704 (Molecular Devices). Selected areas were captured by an infrared (IR) laser onto CapSure Macro LCM Caps (Molecular Devices) and subsequently cut by a UV laser. The target cells that fused to the LCM cap were collected by removing the cap from the tissue section.

Microarray analysis

Total RNAs were extracted from the LM cells with a PicoPure™ RNA isolation kit (Molecular Devices), quantified with a Quant-iT™ RiboGreen RNA reagent and kit (Invitrogen, Carlsbad, CA, USA), and subjected to the rice 44 K oligo microarray (Agilent Technologies, Santa Clara, CA, USA), which contains ∼42,000 oligonucleotides based on the nucleotide sequence and full-length cDNA of the Rice Annotation Project Database (RAP-DB) [65]. Fluorescent probe labeling, hybridization, and scanning were performed according to the manufacturer's instructions (Agilent Technologies), with slight modifications. Each experiment was performed three times using independently isolated samples (three biological replicates). Feature extraction software (Agilent Technologies) was used to delineate and measure the signal intensity of each spot in the array. All microarray data, including our microarray data and publically available data, were statistically normalized by variance stabilization normalization (VSN) using R software ( The normalized signal intensities were then transformed to log base 2. All microarray data is MIAME compliant and the raw data from this study have been deposited in Gene Expression Omnibus (GEO; under accession number GSE 29217.

GO and EC analyses

GO terms were obtained for rice genes to classify their biological functions [66]. GO terms for GO enrichment analysis and EC analysis were retrieved from the RAP-DB. EC analysis was performed as previously described [21], [67], with the following modifications: genes with the same GO term were used as a set of predefined functional genes (each set contained 10 or more genes). EC reports the fraction of gene pairs per GO category that show elevated co-expression. Here, the PCC was utilized as a measure for expression similarity, and we used a PCC threshold of 0.64, corresponding to the 99th percentile of random PCC distribution derived for 1,000 random genes (approximately 1,000*999*0.5 gene pairs). To calculate the random EC for GO categories, random gene sets were sampled with the same size as the category under investigation. For GO enrichment analysis, the statistical significance of GO enrichment within co-expressed genes was evaluated against a background set consisting of 11,456 genes with at least one connection from the dataset of rice co-expression networks using the hypergeometric distribution without a multiple-testing correlation, and P values <0.05 were set as the significant threshold.

Co-expression analysis

To construct co-expression networks, we calculated the PCC values for all combinations of unique 29,864 probes present on the rice 44 K oligo microarray. We estimated two PCC thresholds of 0.48 and 0.65, corresponding to the 95th and 99th percentile of random PCC distribution derived for 1,000 random genes (approximately 1,000*999*0.5 gene pairs). In some previous co-expression studies [3], [68], any two genes with a PCC value greater than 0.6 between their expression profiles were considered as co-expressed genes. Therefore, we set a PCC threshold of 0.64 corresponding to the 99th percentile of random PCC distribution, as described above. For all gene pairs with a significant PCC value, we also calculated MR values between them as another value of co-expression measure to further reduce the number of false positives, according to a previous report [69]. Finally, if the absolute value of MR was lower than 10, the gene pair was considered as a significant connection for the co-expression network. These calculations were conducted using the R/Bioconductor. In extracting meiosis or pollen-wall subnetwork datasets, the network vicinity was extracted by taking 2 steps out from a guide gene, as described previously by Mutwil et al. [70]. The network was illustrated using the program Cytoscape.

Supporting Information

Figure S1.

Representative tissue expression profile of the guide genes after normalization by VSN algorism.Nine meiosis guide genes (A) are preferentially expressed in anther and microspore at meiosis (MEI) stage, whereas three pollen wall guide genes (B) are expressed in microspore and tapetum at tetrad (TET) or uninuclear (UNI) stage. The bar at top represents the average signal intensity (log2). Details on the individual samples can be found in Table S1. All 176 microarray data used for co-expression analysis show a similar expression pattern to those of these guide genes, respectively (Tables S2, S3, S4, S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22, S23, S24, S25).


Table S1.

GEO IDs of the 176 microarray profiles in this study.


Table S2.

Microarray data of Leaf Blade preferentially expressed genes.


Table S3.

Microarray data of Leaf Sheath preferentially expressed genes.


Table S4.

Microarray data of Root preferentially expressed genes.


Table S5.

Microarray data of Stem preferentially expressed genes.


Table S6.

Microarray data of Inflorescence preferentially expressed genes.


Table S7.

Microarray data of Anther_0.3–0.6 mm preferentially expressed genes.


Table S8.

Microarray data of Premeiosis anther preferentially expressed genes.


Table S9.

Microarray data of Meiosis (MEI)-microspore preferentially expressed genes.


Table S10.

Microarray data of Meiosis (MEI)-tapetum preferentially expressed genes.


Table S11.

Microarray data of Anther_0.7–1.0 mm preferentially expressed genes.


Table S12.

Microarray data of Tetrad (TET)-microspore preferentially expressed genes.


Table S13.

Microarray data of Tetrad (TET)-tapetum preferentially expressed genes.


Table S14.

Microarray data of Anther_1.2–1.5 mm preferentially expressed genes.


Table S15.

Microarray data of Uninuclear (UN)-microspore preferentially expressed genes.


Table S16.

Microarray data of Uninuclear (UN)-tapetum preferentially expressed genes.


Table S17.

Microarray data of Anther_1.6–2.0 mm preferentially expressed genes.


Table S18.

Microarray data of Bicellular (BC)-pollen preferentially expressed genes.


Table S19.

Microarray data of Tricellular (TC)-pollen preferentially expressed genes.


Table S20.

Microarray data of Pistil preferentially expressed genes.


Table S21.

Microarray data of Lemma preferentially expressed genes.


Table S22.

Microarray data of Palea preferentially expressed genes.


Table S23.

Microarray data of Ovary preferentially expressed genes.


Table S24.

Microarray data of Embryo preferentially expressed genes.


Table S25.

Microarray data of Endosperm preferentially expressed genes.


Table S26.

EC scores for different GO categories.


Table S27.

PCC and MR values of all gene pairs in the meiotic subnetwork.


Table S28.

Frequency of GO categories for genes in the meiotic subnetwork.


Table S29.

PCC and MR values of all gene pairs in the pollen wall synthesis subnetwork.


Table S30.

Frequency of GO categories for genes in the pollen wall synthesis subnetwork.



We thank M. Miyano (Tohoku University), S. Tashiro (Soma Agricultural School), and H. Kamakura (The University of Tokyo) for their helpful assistance. We are also grateful to K. Yano (Meiji University), M. Kawagishi-Kobayashi (National Institute of Crop Science) and A. Higashitani (Tohoku University) for their useful discussion.

Author Contributions

Conceived and directed the project: MW MM. Designed and performed laser microdissection experiment: HT K. Shiono NT MN. Designed and performed microarray analysis: YN. Analyzed microarray data and performed GO, EC and co-expression network analyses: KA GS K. Suwabe KH KY TH. Discussed the results and commented on the manuscript: KA GS K. Suwabe TH HT K. Shiono KY NT MN YN MW. Wrote the manuscript: KA GS K. Suwabe MW MM.


  1. 1. Lee HK, Hsu AK, Sajdak J, Qin J, Pavlidis P (2004) Coexpression analysis of human genes across many microarray data sets. Genome Res 14: 1085–1094.
  2. 2. Ihmels J, Levy R, Barkai N (2004) Principles of transcriptional control in the metabolic network of Saccharomyces cerevisiae. Nat Biotechnol 22: 86–92.
  3. 3. Aoki K, Ogata Y, Shibata D (2007) Approaches for extracting practical information from gene co-expression networks in plant biology. Plant Cell Physiol 48: 381–390.
  4. 4. Persson S, Wei H, Milne J, Page GP, Somerville CR (2005) Identification of genes required for cellulose synthesis by regression analysis of public microarray data sets. Proc Natl Acad Sci U S A 102: 8633–9638.
  5. 5. Hirai MY, Sugiyama K, Sawada Y, Tohge T, Obayashi T, et al. (2007) Omics-based identification of Arabidopsis Myb transcription factors regulating aliphatic glucosinolate biosynthesis. Proc Natl Acad Sci U S A 104: 6478–6483.
  6. 6. Swarbreck D, Wilks C, Lamesch P, Berardini TZ, Garcia-Hernandez M, et al. (2008) The Arabidopsis Information Resource (TAIR): gene structure and function annotation. Nucleic Acids Res 36: D1009–1014.
  7. 7. Sato Y, Antonio BA, Namiki N, Takehisa H, Minami H, et al. (2011) RiceXPro: a platform for monitoring gene expression in japonica rice grown under natural field conditions. Nucleic Acids Res 39: D1141–1148.
  8. 8. Suzuki G (2009) Recent progress in plant reproduction research: the story of the male gametophyte through to successful fertilization. Plant Cell Physiol 50: 1857–1864.
  9. 9. Koltunow AM, Truettner J, Cox KH, Wallroth M, Goldberg RB (1990) Different temporal and spatial gene expression patterns occur during anther development. Plant Cell 2: 1201–1224.
  10. 10. Scott R, Dagless E, Hodge R, Paul W, Soufleri I, et al. (1991) Patterns of gene expression in developing anthers of Brassica napus. Plant Mol Biol 17: 195–207.
  11. 11. Tsuchiya T, Toriyama K, Ejiri S, Hinata K (1994) Molecular characterization of rice genes specifically expressed in the anther tapetum. Plant Mol Biol 26: 1737–1746.
  12. 12. Hihara Y, Hara C, Uchimiya H (1996) Isolation and characterization of two cDNA clones for mRNAs that are abundantly expressed in immature anthers of rice (Oryza sativa L.). Plant Mol Biol 30: 1181–1193.
  13. 13. Rubinelli P, Hu Y, Ma H (1998) Identification, sequence analysis and expression studies of novel anther-specific genes of Arabidopsis thaliana. Plant Mol Biol 37: 607–619.
  14. 14. Jeon J-S, Chung Y-Y, Lee S, Yi G-H, Oh B-G, et al. (1999) Isolation and characterization of an anther-specific gene, RA8, from rice (Oryza sativa L.). Plant Mol Biol 39: 35–44.
  15. 15. Masuko H, Endo M, Saito H, Hakozaki H, Park J-I, et al. (2006) Anther-specific genes, which expressed through microsporogenesis, are temporally and spatially regulated in model legume, Lotus japonicus. Genes Genet Syst 81: 57–62.
  16. 16. Endo M, Matsubara H, Kokubun T, Masuko H, Takahata Y, et al. (2002) The advantages of cDNA microarray as an effective tool for identification of reproductive organ-specific genes in a model legume, Lotus japonicus. FEBS Lett 514: 229–237.
  17. 17. Amagai M, Ariizumi T, Endo M, Hatakeyama K, Kuwata C, et al. (2003) Identification of anther-specific genes in a cruciferous model plant, Arabidopsis thaliana, by using a combination of Arabidopsis macroarray and mRNA derived from Brassica oleracea. Sex Plant Reprod 15: 213–222.
  18. 18. Suwabe K, Suzuki G, Takahashi H, Shiono K, Endo M, et al. (2008) Separated transcriptomes of male gametophyte and tapetum in rice: validity of a laser microdissection (LM) microarray. Plant Cell Physiol 49: 1407–1416.
  19. 19. Hobo T, Suwabe K, Aya K, Suzuki G, Yano K, et al. (2008) Various spatiotemporal expression profiles of anther-expressed genes in rice. Plant Cell Physiol 49: 1417–1428.
  20. 20. Usadel B, Obayashi T, Mutwil M, Giorgi FM, Bassel GW, et al. (2009) Co-expression tools for plant biology: opportunities for hypothesis generation and caveats. Plant Cell Environ 32: 1633–1651.
  21. 21. Pilpel Y, Sudarsanam P, Church GM (2001) Identifying regulatory networks by combinatorial analysis of promoter elements. Nat Genet 29: 153–159.
  22. 22. Nonomura K-I, Nakano M, Fukuda T, Eiguchi M, Miyao A, et al. (2004) The novel gene HOMOLOGOUS PAIRING ABERRATION IN RICE MEIOSIS1 of rice encodes a putative coiled-coil protein required for homologous chromosome pairing in meiosis. Plant Cell 16: 1008–1020.
  23. 23. Nonomura K-I, Nakano M, Eiguchi M, Suzuki T, Kurata N (2006) PAIR2 is essential for homologous chromosome synapsis in rice meiosis I. J Cell Sci 119: 217–225.
  24. 24. Yuan W, Li X, Chang Y, Wen R, Chen G, et al. (2009) Mutation of the rice gene PAIR3 results in lack of bivalent formation in meiosis. Plant J 59: 303–315.
  25. 25. Nonomura K-I, Morohoshi A, Nakano M, Eiguchi M, Miyao A, et al. (2007) A germ cell specific gene of the ARGONAUTE family is essential for the progression of premeiotic mitosis and meiosis during sporogenesis in rice. Plant Cell 19: 2583–2594.
  26. 26. Zhang L, Tao J, Wang S, Chong K, Wang T (2006) The rice OsRad21-4, an orthologue of yeast Rec8 protein, is required for efficient meiosis. Plant Mol Biol 60: 533–554.
  27. 27. Deng ZY, Wang T (2007) OsDMC1 is required for homologous pairing in Oryza sativa. Plant Mol Biol 65: 31–42.
  28. 28. Chang L, Ma H, Xue HW (2009) Functional conservation of the meiotic genes SDS and RCK in male meiosis in the monocot rice. Cell Res 19: 768–782.
  29. 29. Wang M, Wang K, Tang D, Wei C, Li M, et al. (2010) The central element protein ZEP1 of the synaptonemal complex regulates the number of crossovers during meiosis in rice. Plant Cell 22: 417–430.
  30. 30. Nonomura K-I, Eiguchi M, Nakano M, Takashima K, Komeda N, et al. (2011) A novel RNA-recognition-motif protein is required for premeiotic G1/S-phase transition in rice (Oryza sativa L.). PLoS Genet 7: e1001265.
  31. 31. Kaur J, Sebastian J, Siddiqi I (2006) The Arabidopsis-mei2-like genes play a role in meiosis and vegetative growth in Arabidopsis. Plant Cell 18: 545–559.
  32. 32. Watanabe Y, Shinozaki-Yabana S, Chikashige Y, Hiraoka Y, Yamamoto M (1997) Phosphorylation of RNA-binding protein controls cell cycle switch from mitotic to meiotic in fission yeast. Nature 386: 187–190.
  33. 33. Shinohara M, Sakai K, Ogawa T, Shinohara A (2003) The mitotic DNA damage checkpoint proteins Rad17 and Rad24 are required for repair of double-strand breaks during meiosis in yeast. Genetics 164: 855–865.
  34. 34. Higgins JD, Vignard J, Mercier R, Pugh AG, Franklin FC, et al. (2008) AtMSH5 partners AtMSH4 in the class I meiotic crossover pathway in Arabidopsis thaliana, but is not required for synapsis. Plant J. 55: 28–39.
  35. 35. Rozema J, Broekman RA, Blokker P, Meijkamp BB, de Bakker N, et al. (2001) UV-B absorbance and UV-B absorbing compounds (para-coumaric acid) in pollen and sporopollenin: the perspective to track historic UV-B levels. J Photochem Photobiol B 62: 108–117.
  36. 36. Bubert H, Lambert J, Steuernagel S, Ahlers F, Wiermann R (2002) Continuous decomposition of sporopollenin from pollen of Typha angustifolia L. by acidic methanolysis. Z Naturforsch C 57: 1035–1041.
  37. 37. Ahlers F, Lambert J, Wiermann R (2003) Acetylation and silylation of piperidine solubilized sporopollenin from pollen of Typha angustifolia L. Z Naturforsch C 58: 807–811.
  38. 38. Aya K, Ueguchi-Tanaka M, Kondo M, Hamada K, Yano K, et al. (2009) Gibberellin modulates anther development in rice via the transcriptional regulation of GAMYB. Plant Cell. 21: 1453–1472.
  39. 39. Li H, Pinot F, Sauveplane V, Werck-Reichhart D, Diehl P, et al. (2010) Cytochrome P450 family member CYP704B2 catalyzes the {omega}-hydroxylation of fatty acids and is required for anther cutin biosynthesis and pollen exine formation in rice. Plant Cell. 22: 173–190.
  40. 40. Zhang D, Liang W, Yin C, Zong J, Gu F, et al. (2010) OsC6, encoding a lipid transfer protein, is required for postmeiotic anther development in rice. Plant Physiol. 154: 149–162.
  41. 41. Aarts MG, Hodge R, Kalantidis K, Florack D, Wilson ZA, et al. (1997) The Arabidopsis MALE STERILITY 2 protein shares similarity with reductases in elongation/condensation complexes. Plant J 12: 615–623.
  42. 42. de Azevedo Souza C, Kim SS, Koch S, Kienow L, Schneider K, et al. (2009) A novel fatty Acyl-CoA Synthetase is required for pollen development and sporopollenin biosynthesis in Arabidopsis. Plant Cell 21: 507–525.
  43. 43. Kim SS, Grienenberger E, Lallemand B, Colpitts CC, Kim SY, et al. (2010) LAP6/POLYKETIDE SYNTHASE A and LAP5/POLYKETIDE SYNTHASE B encode hydroxyalkyl α-pyrone synthases required for pollen development and sporopollenin biosynthesis in Arabidopsis thaliana. Plant Cell 22: 4045–4066.
  44. 44. Grienenberger E, Kim SS, Lallemand B, Geoffroy P, Heintz D, et al. (2010) Analysis of TETRAKETIDE α-PYRONE REDUCTASE function in Arabidopsis thaliana reveals a previously unknown, but conserved, biochemical pathway in sporopollenin monomer biosynthesis. Plant Cell 22: 4067–4083.
  45. 45. Ariizumi T, Toriyama K (2011) Genetic regulation of sporopollenin synthesis and pollen exine development. Annu Rev Plant Biol 62: 437–460.
  46. 46. Sorensen AM, Kröber S, Unte US, Huijser P, Dekker K, et al. (2003) The Arabidopsis ABORTED MICROSPORES (AMS) gene encodes a MYC class transcription factor. Plant J 33: 413–423.
  47. 47. Jung KH, Han MJ, Lee YS, Kim YW, Hwang I, et al. (2005) Rice Undeveloped Tapetum1 is a major regulator of early tapetum development. Plant Cell 17: 2705–2722.
  48. 48. Zhang ZB, Zhu J, Gao JF, Wang C, Li H, et al. (2007) Transcription factor AtMYB103 is required for anther development by regulating tapetum development, callose dissolution and exine formation in Arabidopsis. Plant J 52: 528–538.
  49. 49. Zhu J, Chen H, Li H, Gao JF, Jiang H, et al. (2008) Defective in Tapetal development and function 1 is essential for anther development and tapetal function for microspore maturation in Arabidopsis. Plant J 55: 266–277.
  50. 50. Obayashi T, Kinoshita K (2010) Coexpression landscape in ATTED-II: usage of gene list and gene network for various types of pathways. J Plant Res. 123: 311–319.
  51. 51. Srinivasasainagendra V, Page GP, Mehta T, Coulibaly I, Loraine AE (2008) CressExpress: a tool for large-scale mining of expression data from Arabidopsis. Plant Physiol. 147: 1004–1016.
  52. 52. Ogata Y, Suzuki H, Sakurai N, Shibata D (2010) CoP: a database for characterizing co-expressed gene modules with biological information in plants. Bioinformatics. 26: 1267–1268.
  53. 53. Manfield IW, Jen CH, Pinney JW, Michalopoulos I, Bradford JR, et al. (2006) Arabidopsis Co-expression Tool (ACT): web server tools for microarray-based gene expression analysis. Nucleic Acids Res. 34: W504–W509.
  54. 54. Mutwil M, Obro J, Willats WG, Persson S (2008) GeneCAT--novel webtools that combine BLAST and co-expression analyses. Nucleic Acids Res. 36: W320–W326.
  55. 55. Obayashi T, Kinoshita K, Nakai K, Shibaoka M, Hayashi S, et al. (2007) ATTED-II: a database of co-expressed genes and cis elements for identifying co-regulated gene groups in Arabidopsis. Nucleic Acids Res. 35: D863–D869.
  56. 56. Oikawa A, Joshi HJ, Rennie EA, Ebert B, Manisseri C, et al. (2010) An integrative approach to the identification of Arabidopsis and rice genes involved in xylan and secondary wall development. PLoS One 5: e15481.
  57. 57. Okazaki Y, Shimojima M, Sawada Y, Toyooka K, Narisawa T, et al. (2009) A chloroplastic UDP-glucose pyrophosphorylase from Arabidopsis is the committed enzyme for the first step of sulfolipid biosynthesis. Plant Cell 21: 892–909.
  58. 58. Yamada K, Nagano AJ, Nishina M, Hara-Nishimura I, Nishimura M (2008) NAI2 is an endoplasmic reticulum body component that enables ER body formation in Arabidopsis thaliana. Plant Cell 20: 2529–2540.
  59. 59. Long TA, Tsukagoshi H, Busch W, Lahner B, Salt DE, et al. (2010) The bHLH transcription factor POPEYE regulates response to iron deficiency in Arabidopsis roots. Plant Cell 22: 2219–2236.
  60. 60. Sugano SS, Shimada T, Imai Y, Okawa K, Tamai A, et al. (2010) Stomagen positively regulates stomatal density in Arabidopsis. Nature 463: 241–244.
  61. 61. Obayashi T, Nishida K, Kasahara K, Kinoshita K (2011) ATTED-II updates: condition-specific gene coexpression to extend coexpression analyses and applications to a broad range of flowering plants. Plant Cell Physiol. 52: 213–219.
  62. 62. Ficklin SP, Luo F, Feltus FA (2010) The association of multiple interacting genes with specific phenotypes in rice using gene coexpression networks. Plant Physiol. 154: 13–24.
  63. 63. Lee TH, Kim YK, Pham TT, Song SI, Kim JK, et al. (2009) RiceArrayNet: a database for correlating gene expression from transcriptome profiling, and its application to the analysis of coexpressed genes in rice. Plant Physiol. 151: 16–33.
  64. 64. Hamada K, Hongo K, Suwabe K, Shimizu A, Nagayama T, et al. (2011) OryzaExpress: an integrated database of gene expression networks and omics annotations in rice. Plant Cell Physiol. 52: 220–229.
  65. 65. Rice Annotation Project (2008) The Rice Annotation Project Database (RAP-DB): 2008 update. Nucleic Acids Res 36: D1028–1033.
  66. 66. The Gene Ontology project in 2008 (2008) Gene Ontology Consortium. Nucleic Acids Res 36: D440–444.
  67. 67. Vandepoele K, Quimbaya M, Casneuf T, De Veylder L, Van de Peer Y (2009) Unraveling transcriptional control in Arabidopsis using cis-regulatory elements and coexpression networks. Plant Physiol 150: 535–546.
  68. 68. Fu FF, Xue HW (2010) Coexpression analysis identifies Rice Starch Regulator1, a rice AP2/EREBP family transcription factor, as a novel rice starch biosynthesis regulator. Plant Physiol. 154: 927–938.
  69. 69. Obayashi T, Kinoshita K (2009) Rank of correlation coefficient as a comparable measure for biological significance of gene coexpression. DNA Research 16: 249–260.
  70. 70. Mutwil M, Ruprecht C, Giorgi FM, Bringmann M, Usadel B, et al. (2009) Transcriptional wiring of cell wall-related genes in Arabidopsis. Mol Plant 2: 1015–1024.