CMTCN: a web tool for investigating cancer-specific microRNA and transcription factor co-regulatory networks

Identification of miRNA-TF co-regulatory interactions

CMTCN pools TF-gene, TF-miRNA, miRNA-gene, miRNA-TF regulatory relationships. Based on these relationships, CMTCN focuses on co-regulatory pairs and three-node FFLs identified with the help of the network motif detection algorithm. A co-regulatory pair includes a TF and miRNA that regulate a gene simultaneously. There are three types of three-node FFLs: TF-FFLs, miRNA-FFLs, and composite-FFLs. In a TF-FFL, the TF is the master regulator, which regulates a partner miRNA and their joint target. In a miRNA-FFL, the miRNA is the master regulator, repressing its partner TF and their joint target gene. A TF-FFL and miRNA-FFL can combine to form a composite-FFL in which the miRNA and TF regulate each other. CMTCN can query and display each co-regulatory pattern in detail, and can incorporate expression data from TCGA to refine co-regulatory interactions (Fig. 2D).

Co-regulatory interactions refinement

CMTCN capitalizes on TCGA expression data to select important co-regulatory interactions. TCGA RNA-Seq data (run date 2016-01-28) provided in the Firehose data repository are accessed using the R package RTCGAToolbox (Samur, 2014). CMTCN calculates pairwise Spearman correlation values between TFs, miRNAs, and genes. Users can refine co-regulatory pairs or FFLs on the basis of correlation p-values and correlation coefficients. For instance, when the user sets the p-value cutoff to 0.05 and the correlation coefficient cutoff to 0.2, CMTCN displays a TF-target edge of p < 0.05 with correlation coefficients whose absolute values are ≥0.2. Since most miRNAs are assumed to inhibit the expression of their targets (Beermann et al., 2016), CMTCN shows miRNA-target edge p < 0.05 and correlation coefficient ≤−0.2. Users can indicate which types of TF-target regulation they need. CMTCN gives the user the ability to select to differentiate between positive and negative TF regulation. Thus, if the user needs to examine only positive or negative regulation, CMTCN can retain only positive or negative correlation coefficient interactions, respectively.

Network visualization

CMTCN utilizes major co-regulatory motifs to form a cancer-specific miRNA-TF co-regulatory network. It uses the D3.js to depict an interactive and intuitive co-regulatory network map in which genes, TFs, and miRNAs are represented by green, yellow, and red nodes, respectively. To improve the presentation of the force-directed graph, users can adjust link distance, node repulsion, and the number of co-regulatory relationships displayed. CMTCN network graphs presented can be saved as images (Fig. 2B).

Network topology analysis

The key nodes in a co-regulation network have biological significance because they are signal convergence sites with pronounced control and influence over the network; accordingly, they represent potential candidates for biomarker prediction, clinical prognosis, and treatment (Barabási, Gulbahce & Loscalzo, 2011). CMTCN uses three indicators in its key node analysis: node degree, hub score, and authority score (Kleinberg, 1999) (Fig. 2C). Node degree represents the number of edges that meet at a vertex. A node with a high hub score contains a large number of outgoing links, and a node with a high authority score is pointed to by many other nodes with high hub scores. Letting A be the adjacency matrix of the graph, the hub score is defined as the principal eigenvector of AA^T, and the authority score is the principal eigenvector of A^TA. CMTCN uses appropriate pictures to produce a vivid presentation of scoring results.

Gene/miRNA enrichment analysis

To better capture and mine biological roles of a co-regulatory network, CMTCN takes advantage of annotated gene/miRNA sets from GSEA (Subramanian et al., 2005) and miEAA (Backes et al., 2016), thereby enabling functional enrichment analysis of genes, TFs, and miRNAs involved in the co-regulatory network. CMTCN enables detailed gene-ontology association analyses with a variety of biological and biomedical ontologies, extending beyond GO (Consortium et al., 2000) and KEGG (Kanehisa & Goto, 2000), thereby providing clues for follow-up studies (Fig. 2E).

Uncovering and analyzing the miRNA-TF co-regulatory network in THCA

THCA is a common endocrine malignancy with an increasing worldwide incidence (Cabanillas, Mcfadden & Durante, 2016). In CMTCN, we chose the THCA cancer set, selected the validated regulation information confidence level, and built a full miRNA-TF co-regulatory network for THCA. For each type of co-regulatory pattern, we required a p-value <0.05, an absolute value of correlation coefficient ≥0.2, and both types of TF regulation. CMTCN established a THCA-specific miRNA-TF co-regulatory network comprised of 391 nodes and 518 links, with 710 co-regulatory pairs, 7 TF-FFLs, 1 miRNA-FFL, and 2 composite-FFLs.

CMTCN then used network topology analysis to reveal the top-five genes in terms of authority score (MELK, PIGR, SNX5, CLU, and DAPK2) (Table S4). A comprehensive literature review of these genes confirmed their implicated roles in cancer diagnosis and therapy. MELK has been reported to be potential therapeutic targets for malignancies (Pitner et al., 2017). PIGR has the potential to be a candidate prognostic biomarker (Fristedt et al., 2014). Regulation of CLU by oncogenes and epigenetic factors has important consequences for mammalian tumorigenesis (Sala et al., 2009). The aberrant methylation, and hence silencing, of DAPK2 has been reported to play a critical role in thyroid cancer tumorigenesis and progression (Hu et al., 2006). Finally, reduced expression of SNX5 was shown recently to be related to promotion of thyroid tumorigenesis (Jitsukawa et al., 2017) and SNX5 expression studies can be used to support a pathology diagnosis of thyroid cancer (Ara et al., 2012). Additionally, CMTCN carried out a functional enrichment analysis for genes and TFs in the THCA-specific miRNA-TF co-regulatory network. With KEGG pathway enrichment, CMTCN found 11 significant pathways, all of which were related to cancer (Table S5). CMTCN pinpointed four TFs (E2F4, TFDP1, SP1, MYC) and one gene ACVR1 in the transforming growth factor-β signaling pathway, a negative regulator of thyroid follicular cell growth (Geraldo, Yamashita & Kimura, 2012).

MiRNA-TF co-regulatory subnetwork of top mutated genes in OV

OV is highly aggressive gynecological cancer (Sung et al., 2017). We used CMTCN to establish an OV-specific miRNA-TF subnetwork encompassing the top-100 mutated genes in OV. Again, we set the confidence level to validated and required a p-value <0.05, an absolute correlation coefficient ≥0.2 and both types of TF regulation. Our goal was to use CMTCN to reveal the miRNAs and TFs related to the top mutated genes in OV, as well as the regulatory effects of these miRNAs and TFs.

We obtained six co-regulated pairs and one TF-FFL related to the top-100 mutated genes in OV, which revealed six miRNAs and three TFs with possible associations with these top mutated genes. The sole TF-FFL obtained was comprised of a TF (TP53), a miRNA (hsa-mir-29c), and a joint target gene (PTEN). In this TF-FFL, the TF regulates both the miRNA and the target gene, with the miRNA repressing the target gene. Regarding OV pathogenesis, the loss function of PTEN, together with TP53 alteration is a common event (Martins et al., 2014). Interestingly, hsa-mir-29c, an effector of regulator TP53, can also suppress cancer development (Li et al., 2018). The possibility that abnormal expression of the two cross-talking regulators and their co-target gene may be predictive of OV risk is worth further careful study in future experiments.