Integrative identification of deregulated miRNA/TF-mediated gene regulatory loops and networks in prostate cancer

Abstract

MicroRNAs (miRNAs) have attracted a great deal of attention in biology and medicine. It has been hypothesized that miRNAs interact with transcription factors (TFs) in a coordinated fashion to play key roles in regulating signaling and transcriptional pathways and in achieving robust gene regulation. Here, we propose a novel integrative computational method to infer certain types of deregulated miRNA-mediated regulatory circuits at the transcriptional, post-transcriptional and signaling levels. To reliably predict miRNA-target interactions from mRNA/miRNA expression data, our method collectively utilizes sequence-based miRNA-target predictions obtained from several algorithms, known information about mRNA and miRNA targets of TFs available in existing databases, certain molecular structures identified to be statistically over-represented in gene regulatory networks, available molecular subtyping information, and state-of-the-art statistical techniques to appropriately constrain the underlying analysis. In this way, the method exploits almost every aspect of extractable information in the expression data. We apply our procedure on mRNA/miRNA expression data from prostate tumor and normal samples and detect numerous known and novel miRNA-mediated deregulated loops and networks in prostate cancer. We also demonstrate instances of the results in a number of distinct biological settings, which are known to play crucial roles in prostate and other types of cancer. Our findings show that the proposed computational method can be used to effectively achieve notable insights into the poorly understood molecular mechanisms of miRNA-mediated interactions and dissect their functional roles in cancer in an effort to pave the way for miRNA-based therapeutics in clinical settings.

Figure 1. Three-node regulatory motifs considered by IntegraMiR.

The Type I FFL consists of triplets (miRNA, TF, mRNA) such that a miRNA simultaneously targets a mRNA and its TF mRNA. The Type II FFL consists of triplets (miRNA, TF, mRNA) such that a TF simultaneously regulates a miRNA and its target mRNA. Finally, the Type III loop consists of triplets (miRNA, G-1, G-2) such that the miRNA simultaneously targets two transcripts in a given KEGG pathway, one from each gene G-1 and G-2, whose corresponding proteins could potentially interact with each other based on a pathway map provided in the KEGG database.

Figure 2. General description of IntegraMiR.

The method assigns biological roles to miRNAs by integrating five major sources of information together with state-of-the-art statistical techniques to reliably infer specific types of miRNA-target interactions in the context of regulatory loops from mRNA and miRNA expression data.

Figure 3. Consistency of deregulated loops.

A deregulated loop is deemed to be consistent if the expression pattern of its nodes are in agreement with its regulatory edge structure. Any deregulated loop that does not satisfy this property is said to be inconsistent.

Figure 4. Predicted FFL-based transcriptome deregulation in PCa.

(A) Distribution of the fraction of deregulated FFL subtypes grouped in terms of consistent and inconsistent deregulation based on expression data. (B) Percentages of transcriptome change due to significantly upregulated (in green) and downregulated (in red) miRNAs. (C) Cumulative percentages of transcriptome change due to significantly upregulated (in green) and downregulated (in red) miRNAs.

Figure 5. Comparison of miRNA-Target predictions for miRNAs in the same family versus those not in one family.

(A) Venn diagram depicting the number of mRNA targets of six significantly upregulated miRNAs, miR-17 and miR-20a (from the miR-17/92 cluster), miR-106b and miR-93 (from the miR-106b/25 cluster), and miR-106a and miR-20b (from the miR-106a/363 cluster), which belong to the same family. (B) Venn diagram depicting the number of mRNA targets of three significantly downregulated tumor suppressor miRNAs, miR-24, miR-29a, and miR-145, which do not belong to one family.

Figure 6. Predicted FFL-based miRNA-TF co-regulation.

(A) Numbers of coherent and incoherent deregulated FFLs for each type of miRNA-TF interaction. (B) Percentages of consistently and inconsistently deregulated FFLs under each miRNA-TF interaction type depicted in (A).

Figure 7. TP53 miRNA-mediated network model for apoptosis.

IntegraMiR identifies two deregulated FFLs in PCa that model regulatory interactions among miR-125b, TP53 (p53), and BBC3 (PUMA). (A) Type I coherent and Type II-A coherent FFLs. (B) TP53 miRNA-mediated network model for apoptosis obtained by combining the two FFLs in (A).

Figure 8. MYC-E2F1 miRNA-mediated network model for cell proliferation.

A network of proliferative and anti-proliferative miRNAs interacting with MYC and E2F1 predicted by IntegraMiR. This network consists of 18 distinct FFLs: 8 Type I coherent, 2 Type II-A coherent, and 8 Type II-A incoherent. Green edges depict true-positive miRNA-target interactions identified by the predictive module of IntegraMiR, the brown edge predicts a false-negative miRNA-target interaction, red edges depict novel miRNA-target interactions, and black edges represent known interactions.

Figure 9. Predicted deregulated Type III regulatory loops in the Prostate Cancer Pathway.

Portion of the Prostate Cancer Pathway, adopted from the KEGG database , , with the targets of miR-24, miR-29a and miR-145 that participate in deregulated Type III loops being color-coded. One example of a deregulated Type III loop is shown for each miRNA. All depicted Type III loops are novel and consistent, in the sense that the corresponding miRNA-target interactions are anti-correlated according to the data.

Figure 10. Predicted regulatory circuits controlling EMT.

(A) An initial regulatory circuit, predicted by IntegraMiR, controlling EMT in PCa through regulation of CDH1 (E-cadherin) transcriptional repressors. This network consists of 14 distinct FFLs: 2 Type I coherent, 5 Type I incoherent, 2 Type II-A coherent, and 5 Type II-B incoherent. (B) The five FFLs predicted to be (consistently) deregulated in PCa by IntegraMiR comprising miR-200b, miR-200c, or miR-141, and GATA3 and TGFBR3. (C) The nine deregulated miRNA-target interactions involving miR-200b, miR-200c, and miR-141 as well as the TGFB ligands and receptors. (D) An extended integrated regulatory circuit, predicted by IntegraMiR, controlling EMT through TGF- signaling and regulation of CDH1 transcriptional repressors. In these figures, green edges depict true-positive miRNA-target interactions identified by the predictive module of IntegraMiR, brown edges represent false-negative miRNA-target interactions, red edges depict novel miRNA-target interactions, and black edges depict known interactions.

Figure 11. Integrative miRNA-mediated model for PCa development.

A snapshot of a high-level integrative miRNA-mediated model for PCa development which encapsulates major sources of deregulation at the transcriptional, post-transcriptional, and signaling levels, coupled with genetic and epigenetic alterations.

Figure 12. Examples of consistently and inconsistently deregulated FFLs identified by IntegraMiR.

(A) A consistently deregulated Type I coherent FFL. (B) An inconsistently deregulated Type I coherent FFL. Green edges represent true-positive predictions, the red edge represents a novel prediction, and black edges represent known interactions. The red edges emanating from the miRNAs that target the two signaling pathways represent the novel interactions depicted in Figure 9.

Figure 13. Complex regulatory motifs can be constructed from results obtained by IntegraMiR.

(A) SIM motif of GF, GFR, and PI3K genes targeted by miR-29a in the KEGG prostate cancer pathway. (B) DOR motif of GFR and PI3K co-targeting by miR-29a, miR-24, and miR-145 in the KEGG prostate cancer pathway.