Insulin resistance is a condition in which tissues, such as skeletal muscle, liver, and adipocytes, fail to respond to insulin, leading to type 2 diabetes mellitus. Because skeletal muscle is the primary tissue of insulin-mediated glucose uptake [1], it would be important to study the insulin effects on gene expression in skeletal muscle and to identify the insulin-responsive target genes and regulatory genes.

DNA microarray techniques have been widely used in biological research by enabling comprehensive understanding of molecular and systematic mechanism [2,3]. In addition, microarray experiments enable researchers to identify gene sets undergoing a response to a specific stimulus as well as to analyze dynamic response at the level of gene regulatory network [4]. Especially, microarray experiments could be designed to study downstream effects of a specific stimulus or diverse stimuli on gene expression at the level of genomic scale, not a single genetic one [5]. Microarray techniques could be a power tool to discover regulatory factors as most of co-expressed genes are co-regulated by the same regulatory factors [6]. Moreover, because genes co-regulated by the same transcription factor (TF) commonly share binding sites for this TF [6], it would be worth to detect TF based on motif discovery method.

Although diverse motif discovery methods have been proposed and refined [7,8,9,10], most of them are restricted to using human annotated position weight matrices (PWM). On the other hand, iRegulon, a user-friendly Cytoscape plugin, provides approximately nine thousand PWMs that have been collected from various sources and from different species and enable to make regulatory network by linking them to candidate TF [11]. The other approach for pathway analysis is Genomica [12], which yields regulatory modules from gene expression based on probabilistic graphical models and Bayesian networks. It has a benefit in directly mirroring expression of their target genes in a specific condition by comparing expressions between regulators and their target genes [12]. However, this approach requires looping over all regulators, taking too much time in big data.

Here, we tried to construct a regulatory network in order to understand the regulation of insulin-response genes. We identified about 560 genes differentially expressed between before and after treatment in insulin sensitive individuals, and found that those genes act for insulin signaling biological process through Gene Ontology (GO) analysis. Furthermore, we report a reliable regulatory network cont-rolling insulin-mediated genes by integrating motif-based approach using iRegulon and expression-based approach using Genomica; thereby present some benefits for the construction.

Most genes have networks of interactions for appropriate biological functions. Furthermore, regulators do not act alone. Describing a combination of regulatory genes is important to more accurately explain their target genes and understand biological functions. In this regard, construction of regulatory networks can give an insight in the biological interpretation of differential gene set between normal and complex diseases. In order to understand insulin-responsive system, we tried to construct regulatory network with microarray mRNA expression data monitored before and after insulin treatment in skeletal muscle of insulin sensitive samples. For the construction of a reliable network, we carried out integration of two different methods, one based on TF-binding motif and the other on expression variations. Firstly, we defined insulin-mediated gene set using threshold of t test p-value and expression fold change, about 560 genes among total of 12,626 genes were identified. GO enrichment analysis indicated that the gene set is enriched with the biological process terms related with insulin signaling pathway. Especially, among those GO terms, "response to hormone stimulus" and "trans membrane receptor protein tyrosine kinase signaling pathway" have well known links directly to the insulin [8]. To understand regulation of those genes made up of the insulin signaling pathway, 25 TFs were predicted by iRegulon, based on the enrichment of TF-binding motif. iRegulon can infer TFs of target genes, and visualize interactions between TFs and target genes with previously known evidences that what TFs bind to a target gene with sequence information. In this case, expression values of target genes are not considered, and this regulatory network cannot explain how those regulators affect expression variation of target genes. We tried to explain expression variation of target genes with those of TFs identified by iRegulon. However, the expression variation of a TF did not resemble those of target genes. It suggests that expression variations of multiple TFs should be considered collectively to understand those of target genes. With the TFs predicted by iRegulon, we constructed a hierarchical regulatory network among them using Genomica, explaining the expression variation of the target genes. As shown Fig. 4, a combination of multiple TFs instead of a single TF explained the expression variations of target genes. For example, MEF2A which is at the top of the regulatory TF network had the largest influence on the expression variations of the target genes, and YY1, the second highest ranking TF, did not influence as much as MEF2A. However, YY1 affects to increase expression of target genes when MEF2A has low expression value. In fact, MEF2A, YY1, SRF, PDX1, and NFATC4 are known for having relationships with insulin [15,16,17,18,19]. Although expression of downstream TFs in the regulatory network is not perfectly fitted to that of their target genes, the construction based on expression data helps to understand more clearly regulatory network of target genes. Altogether, in this paper, we presented a reliable regulatory network controlling insulin-mediated genes as well as the approach integrating two different methods, iRegulon and Genomica, may allow making a reliable regulatory network.