A total of 1,779 Affymetrix raw data Cel files were downloaded from the AtGenExpress and GEO sites [11–13]. All of them were derived from the Affymetrix ATH1 gene GeneChip microarray for Arabidopsis, and the samples contained at least two replicate chips. The required probe set-to-locus mappings for the ATH chip were obtained from TAIR (ftp://ftp.arabidopsis.org/home/tair/Microarrays/Affymetrix, version 2-5-2007).

Normalization of the raw data Cel files was performed in R using the MAS5 algorithms, which are implemented in the affy package available from the BioConductor project [14, 15]. R scripts were used to calculate non-parametric correlation coefficients (Spearman’s rho) between the expression of SPL genes and the expression of each of the ~22,000 genes represented on the Affymetrix array used to generate this dataset. We ranked the genes according to the correlation coefficients and reported genes that were most positively correlated with SPL gene expression. The p-values were calculated using bivariate normal distribution, with p representing the probability of observing an equal or larger positive or negative correlation by chance. For details of the method, see the article by Wei [16].

To characterize the correlated genes, the web-based ‘Classification SuperViewer’ program (http://bar.utoronto.ca/ntools/cgi-bin/ntools_classification_superviewer.cgi) was used to search for differential distributions of gene ontology (GO) and biological terms within the correlation genes.

The Arabidopsis interactome network was built by the AtPID. The AtPID (Arabidopsis thaliana Protein Interactome Database) represents a centralized platform for depicting and integrating information pertaining to protein-protein interaction networks, domain architecture, ortholog information, and GO annotation in the Arabidopsis thaliana proteome. We mapped the correlated genes to the network and also selected the first neighbors of the genes. To identify the function of SPL-correlated proteins in network clusters, the latest versions of the Biological Network Gene Ontology (BiNGO) [17] and GOlorize tools [18] were used for the statistical evaluation of groups of proteins. This evaluation was carried out using existing annotations of the Gene Ontology Consortium (http://www.geneontology.org).

Applications in the web-based promoter analysis program Promomer [19] were used to analyze the promoters (−1 kb upstream of the predicted translation start site) of the 112 genes showing the strongest correlations to SPL genes. The Promomer program aims to provide a user-friendly, web-based interface to accomplish two goals: (1) to identify statistically over-represented elements in a gene or a group of genes in Arabidopsis using the enumerative method, and (2) to find the number and position of occurrences of an element in genes across the Arabidopsis genome or in a subgroup of genes.

In the first step of the analyses, we extracted and ranked the 112 genes whose expression most tightly correlates with that of SPL genes (Table 1). We took advantage of a visualization tool called Cytoscape to view expression data and analysis results. Using the available data regarding correlations between different Arabidopsis genes, we constructed a network composed of the SPL genes and their co-expressed genes. This network contains 112 Arabidopsis genes (nodes) and 294 correlations (edges) (Figure 1A). Three interesting facts are observed from the co-expression network. First, some SPL genes are absent from the network, including SPL1, SPL2, SPL6, SPL7, SPL8, SPL12, SPL14 and SPL16. This reflects the current lack of knowledge regarding correlations between these SPL genes and other genes. Second, many genes correlate with SPL3, SPL4, SPL5, and SPL9; some genes correlate with SPL10 and SPL13; while SPL11 and SPL15 have only one co-expression gene. Third, within the co-expression network, there is a region comprising four SPL genes that contain 20 Arabidopsis genes (nodes) and 79 correlations (edges) (Figure 1B).

In order to identify a functional role for SPL genes, the correlated genes were analyzed in a web-based gene ontology (GO) analysis tool to identify any bias in GO functional annotation terms in the correlated genes compared to the remainder of the A. thaliana genome (Supplementary Table 1). When GO clustering was applied to the 112 co-expression genes, some GO biological categories were retrieved. Most of these related to stress response, transporter activity, and transcription factor activity (Figure 1A). Other biological processes identified were associated with extracellular activities, development processes, and enzyme activity (Figure 1A). Thus, given previous work [3] suggesting that SPL genes play a role in controlling plant development, we can infer that SPL genes may not only do this but also take part in many other biological processes. In the co-expression network, a complex related to development and the plasma membrane was found (Figure 1B). We presume that the complex is very important for SPL gene function of development (Figure 1B), and SPL proteins may form heterodimers or multimers to carry out their functions.

To evaluate functional associations between co-expressed proteins, we mapped the genes whose expression was correlated with SPL genes using the Arabidopsis thaliana interactome network (Figure 2A). Since pairs of interacting proteins are more likely to contain similar GO annotations than pairs of unrelated proteins, annotations of neighbor proteins in the network can be used to make predictions of functions. In particular, the GO annotations allow functional assignment of uncharacterized gene products, such when At3g62270 was identified in a large-scale interactome mapping study of anion transport (Figure 2B).

Cluster analysis of this network suggested the molecular mechanisms underlying the function of SPL genes. Analysis of the connected sub-graphs and their GO terms identified functional modules enriched for carbohydrate transport, electron transport, anion transport, flavonoid biosynthetic processes, diaminopimelate metabolic processes, response to hypoxia and stress, and root development (Figure 2B). Many genes and proteins co-expressed with SPL genes are located in these modules, which supports their role in a diversity of physiology processes. These observations support the theory that the network modeled here constitutes a framework that can guide in-depth experimental study of genes and proteins related to SPL gene function in Arabidopsis.

The shared expression profiles of the 112 correlated genes in response to development and stress suggests that these genes are under the same regulatory control and are thus likely to share cis-elements in their promoter regions. Since these genes are co-expressed with SPL genes, we hypothesized that they might be directly regulated by SPL genes. The cis-element of the SBP-box transcription factor is unclear. To reveal which DNA motif can bind to SPL genes, we analyzed the promoter regions of these genes 1 kb upstream of the predicted transcription start site. We also analyzed aspects of shared transcriptional activation sites on these genes for the presence of known plant cis-elements.

Based on previous research [3, 20–22], we hypothesized that the GTAC motif was the cis-element common to the SBP-box genes. Indeed, our analysis showed the GTAC motif to be a statistically over-represented element (Figure 3). Two distributions were created from 1,000 bootstrapped sets, each containing the same number of genes. The first was obtained by sampling for the frequency of occurrence of an element from the correlated genes cluster promoter set bootstrapped 1,000 times; the second was obtained from the same bootstrapping process carried out on 1,000 whole genome promoter data sets containing the same number of sequences as the cluster set. The distribution of occurrence of a given element in both data sets was then obtained and plotted [23] and significant differences were highlighted (Supplementary Figure 1). For comparison, significant tetramers whose expression is unrelated to that of SPL genes are also shown.

The analysis (Supplementary Table 2) indicates that the invariant core GTAC motif is present in 100/112 of our correlated genes a total of 271.9 times, which is higher than the average of 209.3 times across 1,000 randomly-generated promoter sets (Figure 3). The promoter analysis indicates that multiple copies of the SBP-box elements are present in a high percentage of SPL-correlated genes. The core GTAC motifs are significantly enriched compared to expected frequencies, suggesting that they are important regulatory elements in these co-expressed genes.

Transcriptional control of the expression of stress-responsive genes is a crucial part of plant responses to a range of abiotic and biotic stresses. Research carried out in recent years [23] has been productive in identifying transcription factors that are important for regulating plant responses to these stresses. SPL genes have not been shown to be affected by stress in Arabidopsis. Nevertheless, our GO analysis revealed that some genes correlated with SPL genes were involved in the stress response (Supplementary Table 1). These genes include P57 (GO: 0006979, response to oxidative stress), AKT1 (GO: 0009651, response to salt stress), At3g16460 (GO: 0009409, response to cold), ATPPC3 (GO: 0009414, response to water deprivation), and ATCWINV1 (GO: 0009611, response to wounding).

In addition, two transcription factors (TGA1 and WRKY65) that belong to the bZIP and WRKY families, respectively, are co-expressed with SPL genes. bZIPs are a large family of transcription factors in plants, and 75 members are present in Arabidopsis [32]. One class of bZIP proteins that is linked to stress responses comprises the TGA/octopine synthase (ocs)-element-binding factor (OBF) proteins. These bind to the activation sequence-1 (as-1)/ocs element, which regulates the expression of some stress-responsive genes such as PR-1 and GLUTATHIONE S-TRANSFERASE6 (GST6) [33, 34]. In Arabidopsis, seven members of the TGA/OBF family play a role in plant defense, xenobiotic stress responses, and development. WRKY proteins are a novel family of transcription factors that are unique to plants and that form a large, 74-member family in Arabidopsis. WRKY proteins contain either one or two WRKY domains, which is 60-residue region that contains the amino acid sequence WRKYGQK and a motif similar to zinc fingers. Certain WRKY family members show enhanced expression or DNA-binding activity following induction by a range of pathogens, defense signals, and wounding [35]. Since the two transcription families are very important for plant response to stress, and because SPL genes can control some stress-responsive genes, we speculate that SPL genes also play a key role in the stress response.

Our GO analysis shows transcription activity to be an important function for co-expressed genes. SPL genes can activate other transcription factor families, including B3, bZIP, WRKY, MYB, bHLH, and MADs-box. We believe that SPL genes function by controlling the expression of these transcription families. For example, the expression of AGL8 (FUL), a MADs-box gene, correlates with SPL gene expression. In another study, the MADs-box gene AP1 was also found to be activated by SBP-box genes [2]. The proteins AP1 and FUL may serve as hubs between the flower induction pathway, which is comprised of interacting proteins such as SVP, SOC1, and AGL24, and the floral organ identity proteins. Both AP1 and FUL have dual functions in floral meristem identity (early function) and floral organ determination (late function) [36, 37]. This is consistent with the fact that the flowering proteins and floral homeotic proteins come together to form dimers.

Analysis of cis-elements suggests an even more complex picture: (1) SPL genes may be activated by other transcription families (Figure 4A), and (2) the 112 genes correlated to SPL genes in our analysis may be controlled by other transcription families (Fig 4B). We therefore hypothesize a complex control network linking SPL genes with other transcription families.

An integral membrane protein (IMP) is a protein molecule (or assembly of proteins) that is permanently attached to the biological membrane. In the co-expression network of SPL genes, some IMP genes correlate with SPL genes. These include ATK1 and ATK2 (ARABIDOPSIS K TRANSPORTER), ATVGT1 (sugar transporter family protein), PGP4 (P-GLYCOPROTEIN 4), SULTR1;2 (SULFATE TRANSPORTER 1;2), At3g55110 (ABC transporter family protein), and some genes encoding ligands of membrane proteins. The IMPs encoded by these genes include transporters, channels, receptors, enzymes, structural membrane-anchoring domains, proteins involved in accumulation and transduction of energy and sugar, and proteins responsible for the stress response. Our analysis of genes correlated with SPL genes leads us to speculate that SPL genes may participate in development, stress responses, and other biological processes by controlling the expression of IMPs. In addition, some SPL proteins have a transmembrane domain, which suggest that SPL may have interaction with IMPs.

Our analysis indicates that the cis-element shared by SPL genes, the GTAC motif, is necessary for binding of SPL proteins to DNA. In the co-expression network, the expression of some SPL genes correlates with that of other SPL genes; these genes may form a complex within the network (Figure 1B). Therefore, we propose that SPL genes regulate themselves. Using promoter analysis, we found the GTAC motif to be present in the 1 kb upstream region in 13/16 of SPL genes a total of 56 times (Supplementary Table 3), which was significantly higher than the occurrence in tetramer sets of randomly selected promoters (Figure 5A). In the AtPID database [38], we found that SPL9 can interact with SPL15, and SPL4 can interact with SPL5. SPL genes may therefore regulate themselves using a feedback loop based on protein interactions and transcriptional regulation (Figure 5B).

The SBP-box gene family belongs to a group of plant-specific zinc finger protein genes that encodes plant-specific transcription factors. The proteins encoded by SBP-box genes bind specifically to promoters of the floral meristem identity gene SQUAMOSA and its orthologous genes [2, 3, 6]. Except for the presence of a conserved SBP-box/domain, the SPL genes vary substantially in their genomic organization, transcript size, and the size and sequences of their encoded proteins. The position of the SBP domain within the protein also varies [39]. The different gene structures and the divergence of amino acid sequences among different SPL genes provide us with some hints that SBP transcription factors may have a variety of physiological functions. To date, several important and divergent biological processes regulated by SBP-box genes have been reported. These include flower and fruit development [2, 4, 8, 9], architecture formation [40], sporogenesis [6], response to copper and fungal toxin [7, 41], and control of GA level [42].

Our co-expression analysis has revealed that SPL genes may play a role in the stress response. Network analysis suggests that SPL genes function by controlling other transcription factor families, and that they have relatives encoding proteins involved in membrane protein transport. The interactome of the correlated genes has shown that SPL genes may also take part in glucose, inorganic salt, and ATP production. Notably, some SPL genes are not detected in our co-expression analysis. Among them, SPL8 can affect pollen sac development [6], and SPL14 plays a role in plant development and sensitivity to fumonisin B1 [7]. All of the available evidence suggests that SPL genes have varied functions in Arabidopsis.