Systematic characterization of A-to-I RNA editing hotspots in microRNAs across human cancers

Abstract

RNA editing, a widespread post-transcriptional mechanism, has emerged as a new player in cancer biology. Recent studies have reported key roles for individual miRNA editing events, but a comprehensive picture of miRNA editing in human cancers remains largely unexplored. Here, we systematically characterized the miRNA editing profiles of 8595 samples across 20 cancer types from miRNA sequencing data of The Cancer Genome Atlas and identified 19 adenosine-to-inosine (A-to-I) RNA editing hotspots. We independently validated 15 of them by perturbation experiments in several cancer cell lines. These miRNA editing events show extensive correlations with key clinical variables (e.g., tumor subtype, disease stage, and patient survival time) and other molecular drivers. Focusing on the RNA editing hotspot in miR-200b, a key tumor metastasis suppressor, we found that the miR-200b editing level correlates with patient prognosis opposite to the pattern observed for the wild-type miR-200b expression. We further experimentally showed that, in contrast to wild-type miRNA, the edited miR-200b can promote cell invasion and migration through its impaired ability to inhibit ZEB1/ZEB2 and acquired concomitant ability to repress new targets, including LIFR, a well-characterized metastasis suppressor. Our study highlights the importance of miRNA editing in gene regulation and suggests its potential as a biomarker for cancer prognosis and therapy.

Figure 1.

Overview of bioinformatic pipeline and RNA editing profiles in miRNAs across cancer types. (A) Schematic of bioinformatic pipeline. For each cancer type, sample and read processing, miRNA editing calling (filtering DNA variants using various resources), and identification of miRNA editing hotspots. (B) Proportions of A-to-I RNA editing events among all RNA editing events observed at hotspots in different cancer types; average proportion across 20 cancer types is 73.4%. (C) Heat map of miRNA editing hotspots. Edited miRNA expression amounts (log₂RPM) are in color; editing frequencies (% of samples with editing signals) are indicated by circle size.

Figure 2.

Validation and molecular profiles of cancer miRNA editing hotspots. (A) Numbers of miRNA editing hotspots in the analysis of ADAR-perturbation experiments in 786O, HeyA8, and Hs578T cell lines. (B) The editing level changes after transfection of wild-type ADAR enzymes (ADAR WT), and inactive ADAR enzymes (ADAR mut). Inferred ADAR determinant(s) of each miRNA editing hotspot are shown in each miRNA subpanel. (C) For each edited miRNA, the highest rank of the edited miRNA expression amount (upper-quartile value across samples in a cancer type) relative to the WT miRNA expression amounts (median value) across cancer types is shown. (D) A heat map showing editing-level differences of 19 miRNA editing hotspots between cancerous and matched normal samples (two-sided paired Wilcoxon rank test, n ≥ 5). Red indicates overediting and blue indicates underediting in cancer samples; boxes highlight significant differences (FDR < 0.05).

Figure 3.

Correlation of miRNA editing hotspots with clinical features, molecular drivers, and signaling pathways. (A) A heat map showing the clinical relevance of miRNA editing hotspots. Significant correlations with tumor subtype (orange, two-sided Wilcoxon or Kruskal-Wallis test), disease stage (blue, two-sided Wilcoxon or Kruskal-Wallis test), and patient survival times (green, univariate Cox proportional hazards model or log-rank test for median-based two-group comparison) (FDR < 0.2); after considering tumor purity, boxes highlight remaining significant correlations. (B) Significant correlations between the miRNA editing levels and significantly mutated genes (the gene nodes are colored according to the fold change between the mutated samples and the WT sample groups; two-sided Wilcoxon test, FDR < 0.05). The specific cancer types showing the significant correlations are listed below gene names. (C) Correlations between the miRNA editing levels and frequent SCNAs (Spearman rank correlation, |R_S| > 0.5 and FDR < 0.05). (D) Correlations between the miRNA editing levels and signaling pathway scores (derived from TCGA protein expression data; Spearman rank correlation, |R_S| > 0.3 and FDR < 0.05). Red lines indicate positive correlations, while blue lines are negative correlations.

Figure 4.

Effects of RNA editing in miR-200b on cell migration and invasion, and correlation with clinical outcomes. (A) Cartoon of A-to-I editing in stem–loop structure of pre-mir-200b. (B) Summary of correlations of WT miR-200b expression and miR-200 editing level with patient survival times across cancer types. Circle size represents statistical significance; color represents direction. In general, high expression of miR-200b is associated with better patient survival; high editing level in miR-200b is associated with worse patient survival. (C) Kaplan-Meier plots of patients grouped by miR-200b expression in individual cancer types. (D) Kaplan-Meier plots of patients grouped by editing level in miR-200b in individual cancer types. (E,F) Effects of miR-200b mimics on (E) migration and (F) invasion in MCF10A, MDAMB-231, SLR25, and OVCAR8 cells (n = 2 or 3). Two-sided t-test was used to assess the difference. Error bars denote ±SEM; (*) P < 0.05, (**) P < 0.01, (***) P < 0.001. Scale bar length is 500 µm.

Figure 5.

RNA editing in miR-200b redirects the target genes. (A) Sequence motif identified in 3′ UTRs of down-regulated genes upon transfection with WT miR-200b mimics (vs. negative control), corresponding to seed match of WT miR-200b. (B) Sequence motif identified in 3′ UTRs of down-regulated genes upon transfection with edited miR-200b mimics (vs. negative control), corresponding to seed match of edited miR-200b. (C) High-confidence predicted target genes of WT miR-200b and edited miR-200b by integrating gene expression and sequence motif data.

Figure 6.

Molecular mechanisms of WT and edited miR-200b in cancer cells. (A) 3′ UTR representation of WT miR-200b target gene ZEB1. (B) qRT-PCR of ZEB1 upon 24-h transfection with WT miR-200b and edited miR-200b mimics in MCF10A, MDAMB-231, SLR25, and OVCAR8 cells. (C) Western blots of ZEB1 upon 48-h transfection with WT miR-200b and edited miR-200b mimics in MDAMB-231, SLR25, and OVCAR8 cells. (D) 3′ UTR representation of edited miR-200b target, LIFR. (E) qRT-PCR of LIFR upon 24-h transfection with WT miR-200b and edited miR-200b mimics in MCF10A, MDAMB-231, SLR25, and OVCAR8 cells. (F) Western blots of LIFR upon 48-h transfection with WT miR-200b and edited miR-200b mimics in MCF10A, MDAMB-231, SLR25, and OVCAR8 cells. Blots with short-time (SE) and long-time exposure (LE) are shown. (G) Luciferase reporter assays that contain two predicted binding sites of edited miR-200b (F1 and F2) in LIFR. In B, E, and G, two-sided t-test was used to assess the difference, n = 2 or 3, and error bars denote ±SEM; (*) P < 0.05. (**) P < 0.01, (***) P < 0.001. (H) Proposed mechanistic model in which WT miR-200b inhibits key EMT regulators ZEB1 and ZEB2, thereby suppressing cell migration and invasion, whereas edited miR-200b (catalyzed by both ADAR1 and ADAR2) inhibits a new target LIFR, a well-characterized metastasis suppressor, thereby promoting cell migration and invasion.