Dynamic and Modularized MicroRNA Regulation and Its Implication in Human Cancers

Abstract

MicroRNA is responsible for the fine-tuning of fundamental cellular activities and human disease development. The altered availability of microRNAs, target mRNAs, and other types of endogenous RNAs competing for microRNA interactions reflects the dynamic and conditional property of microRNA-mediated gene regulation that remains under-investigated. Here we propose a new integrative method to study this dynamic process by considering both competing and cooperative mechanisms and identifying functional modules where different microRNAs co-regulate the same functional process. Specifically, a new pipeline was built based on a meta-Lasso regression model and the proof-of-concept study was performed using a large-scale genomic dataset from ~4,200 patients with 9 cancer types. In the analysis, 10,726 microRNA-mRNA interactions were identified to be associated with a specific stage and/or type of cancer, which demonstrated the dynamic and conditional miRNA regulation during cancer progression. On the other hands, we detected 4,134 regulatory modules that exhibit high fidelity of microRNA function through selective microRNA-mRNA binding and modulation. For example, miR-18a-3p, -320a, -193b-3p, and -92b-3p co-regulate the glycolysis/gluconeogenesis and focal adhesion in cancers of kidney, liver, lung, and uterus. Furthermore, several new insights into dynamic microRNA regulation in cancers have been discovered in this study.

Figure 1

Discovery of dynamic and modularized miRNA regulation during cancer progression. (A) The identification pipeline of conditional miRNA regulatory interactions; (B) Meta-Lasso Regression approach to the detection of microRNA regulators of each gene in each cancer stage.

Figure 2

Overview of the interactions detected by the CLASH analysis. (A) Distribution of the miRNA-gene interactions (y-axis) with respect to their abundances (x-axis). (B) Distribution of the corrected p-values of each interaction from Binomial test for quality filtering. (C) Projection of each interaction based on the abundance of the corresponding miRNA (x-axis), mRNA (y-axis), and interaction (z-axis). (D) The relationship between the abundances of each miR-92a-3p interaction and the corresponding mRNA. The colored points indicated the three levels of binding affinities based on MFE values: High (red, [−31.2, −21.8]], Medium (green, [−21.8, 12.3]), and Low (blue, [−12.3, −2.9]).

Figure 3

Overview of the miRNA-mRNA interactions identified in nine cancers. (A) Number of interactions detected in a specific stage (Blue) versus multiple stages (Orange), in each cancer. (B) Pie chart shows the percentages of interactions shared by different cancers.

Figure 4

Illustration of the dynamic miRNA-mediated gene regulation (miRNA: pink diamonds and genes: purple nodes). (A) Three miRNAs regulate HOXA11 gene in different stages of KIRC. (B) Five miRNAs repress the same KIF1C gene in KIRC while the miR-484 and miR-769-3p (pink diamonds with red border) consistently interact with KIF1C across stages.

Figure 5

MiRNA-mediated gene regulation involved in two functional pathways in the early stage of KIRC and KIRP (miRNAs: pink diamonds; genes: ellipses). (A) PI3K-Akt Signaling Pathway; (B) Cell Cycle Pathway. Purple nodes denote the common genes between KIRC and KIRP; pink diamonds with red border indicate the common miRNA regulators; green ellipses are the condition-specific cancer genes.

Figure 6

Illustration of the cooperative regulation of a miRNA module that contains four miRNAs: miR-18a-3p, -320a, -193b-3p, and -92b-3p (pink diamonds). This miRNA module consistently regulates the Focal Adhesion pathway in KIRC, LUSC, and UCEC. The target genes are color-coded (each color represents one cancer condition and the bi-colored nodes indicate the corresponding genes have been regulated by miRNAs in two conditions). The detailed miRNA-gene interactions are listed in Table 3. The network is visualized using Cytoscape 3.2.0  with KEGGscape plugin.

Figure 7

Consistency assessments of regulator selection regarding difference sample sizes of three genes: CARD11 (red, 28 potential regulators), JAK2 (blue, 73), and HMGA1 (green, 109). The lines denote the transition Jaccard’s similarity of the regulators selected at different sample sizes of these genes. (A) Consistency of all regulator selection across difference sample sizes. (B) Consistency of microRNA regulator selection across difference sample sizes.

Figure 8

(A) Method comparison on the identified miRNA-target interactions on the test set. MiRDR, RACER and MCMG were applied on the Hepatocellular carcinoma (LIHC) data. The x-axis represents the top ranked interaction predicted by miRDR while the y-axis represents the count of validated interactions. The p-values were calculated by one-sided Wilcoxon singed rank test; (B) Method comparison based on the counts of identified miRNA targets. X-axis: number of target genes; y-axis: miRNA percentage. Outer plot shows the overall distribution and the inner shows the zoom-in view of partial distribution in the range of [0, 200].

Figure 9

Four matrices constructed for each gene j and each condition, i.e., a specific cancer stage. Y represents the vector of expression changes of gene j in the corresponding cancer patients under this condition; X ₁ represents the matrix of background adjustment factors (CNV and DNA methylation); R represents a RS vector that contains the regulatory scores of all r ₁ TF- and r ₂ miRNA- regulators of gene j; P is the matrix that includes all the expression changes of each regulator in each patient under the corresponding condition.