Genes encoding fatty acid biosynthetic genes and seed storage reserve associated proteins are located in different subnetworks
While the entire coexpression network is useful for network topology analysis, isolation of a subnetwork (or cluster) makes it more accessible to biologists [40,58]. More importantly, a subnetwork in the large coexpression network is often more biologically relevant in a pathway context. Hence, we extracted subnetworks from this gene coexpression network for genes relevant to the accumulation of seed storage reserves (Figure 4). Of the 48 genes known to encode enzymes involved in FA biosynthesis [17,59], we identified 44 (or ~92%) genes represented on the ATH1 array, and all of them were found in one subnetwork (Figure 4A). This subnetwork cluster consists of 1854 genes (Additional File 1), which is in general agreement with an interactive correlation network generated genome-wide in Arabidopsis using a heuristic clustering algorithm [41]. Such a gene list can be used to identify interactors of genes in FA synthesis in developing seeds. Consistent with the coexpression subnetwork analysis, the majority of genes involved in FA biosynthesis were associated with Cluster 1 (Figure 3). Their expression levels increased steadily from the globular embryo stage, generally reached the peak at the expanded cotyledon stage, and dramatically declined subsequently throughout late seed maturation (Figure 4B). Such a unified expression pattern for most FA biosynthetic genes supports earlier studies showing that FA supply can be a limiting factor for triacylglycerol (TAG) accumulation in developing embryos of Brassica napus [60], olive (Olea europaea L.) and oil palm (Elaeis guineensis Jacq.)[61], as well as cuphea lanceolata and other oil species [62]. Recent studies of metabolic flux in developing embryos of B. napus, however, indicated that TAG assembly was more limiting than FA biosynthesis in regulating the flow of carbon into TAG [63]. The majority of genes encoding oilbody oleosins and SSPs were found in another subnetwork with a distinct expression pattern (Figure 4C). The subnetwork encompassing genes encoding oleosins and SSPs is comprised of 1392 genes (Additional File 2). Genes encoding oleosins and SSPs were in Cluster 2 (Figure 3), and their expression profiles were strikingly similar. These genes were virtually unexpressed at the globular stage, increased rapidly (>1000-fold in many cases) from the globular stage to the bilaternal stage, and remained at the elevated expression level throughout the remaining stages of seed maturation (Figure 4D). Transcripts for OLEOSIN and SSP genes are most abundant in the seed transcriptome late during seed development. In contrast, most genes in the TAG assembly pathway were found in different subnetworks, exhibiting various expression profiles during seed development (Figure 5). DIACYLGLYCEROL ACYLTRANSFERASE 1 (DGAT1), FATTY ACID DESATURASE 2 (FAD2), FATTY ACID ELONGASE 1 (FAE1) and STEAROYL DESATURASE (SAD) genes were identified in this subnetwork, albeit expressed at substantially lower levels compared to genes encoding oleosins and SSPs (Additional File 3). DGAT catalyzes the acyl-CoA-dependent acylation of sn-1,2-diacylglycerol to produce TAG and CoA [64]. FAD2 catalyzes the introduction of a second double bond into acyl groups in phospholipid whereas SAD catalyzes the formation of monounsaturated FA in the plastid [65]. FAE1 catalyzes the elongation oleoyl-CoA in the endoplasmic reticulum [65]. Our analysis determined that AT1G48300, which was named DGAT3, is the putative gene encoding a cytosolic DGAT in Arabidopsis. The amino acid sequence of AT1G48300 has a significantly high degree of similarity (expect value < 1 × 10-21) to the soluble DGAT in peanut (Arachis hypogaea), where the cytosolic DGAT gene in plants was first discovered [66]. Notably, DGAT3 exhibited a similar expression pattern with DGAT1, but expressed higher during late seed maturation. In earlier studies, quantification of DGAT activity during seed maturation in B. napus indicated that enzyme activity was maximal during the rapid phase of oil accumulation with a substantial decrease in activity occurring as oil levels reached a plateau [67,68]. Assuming DGAT activity shows a similar profile during seed development in Arabidopsis, this suggests that DGAT may be down-regulated post-transcriptionally and/or post-translationally during the latter stages of seed development.
Figure 4  Subnetwork and temporal expression profiles for genes involved in seed storage reserve accumulation in developing Arabidopsis seeds. A is the subnetwork for genes including those in fatty acid (FA) biosynthesis, and B depicts the expression profiles of FA biosynthetic genes identified in the analysis. C is another subnetork including genes encoding oleosins and seed storage proteins (SSP), and D depicts the expression profiles of genes encoding oleosin and SSP. In B and D, the expression values, AGI identifiers of the genes depicted are listed in Additional File 3, and the log2 expression values were standardised by subtracting the value at the first S3 stage for each gene. Dashed red, blue lines indicate 2-fold up- or down-regulation, respectively.
Figure 5  Expression profiles of genes including homologues in the triacylglycerol assembly pathway. The dash line at 6.0 is often used as the cutoff for present (expressed; above the line) or absent (unexpressed; below the line). All expression data were transformed to the log2 scale for plotting the profiles. Genes and homologs in the triacylglycerol (TAG) assembly pathway were identified based on an early survey of Arabidopsis genes involved in acyl lipid metabolism [59], and their AGI identifiers listed in Additional File 3. Refer to [64] for their roles in TAG assembly. The abbreviations of these genes and their encoded enzymes (EC numbers) are as follows: GPAT: sn-glycerol-3-phospahte acyltransferase (EC 2.3.1.15); LPAAT: lysophosphatidic acid acyltransferase (EC 2.3.1.15); PAP: PA phosphatase (EC 3.1.3.4), including LIPIN (PAP1) and LPP (PAP2); AAPT: Aminoalcoholphosphotransferases (EC 2.7.8.1 and EC 2.7.8.2); CPT: cytidine diphosphate (CDP)-choline: 1, 2-diacylglycerol cholinephosphotransferase (EC 2.7.8.2); LPCAT: lysophosphatidylcholine acyltransferase (EC 2.3.1.23); PLA2: Phospholipase A2 (EC 3.1.1.4); PDAT: phospholipid:diacylglycerol acyltransferase (EC 2.3.1.158); LCAT: lechitin:cholesterol acyltransferase (EC 2.3.1.43), these three shown here are PDAT homologs; DGAT: Diacylglycerol acyltransferase (E.C. 2.3.1.20).   In summary, our new results suggest that genes acting in a biological process (FA biosynthesis) can be indicated by their presence in the same coexpression network cluster, but genes involved in the same pathway (TAG assembly) may not necessarily exhibit expression coherence. As a result, computational approaches using coexpression network to predict gene function, such as in [40], will undoubtedly have limitations.