The Genomic Landscape and Clinical Relevance of A-to-I RNA Editing in Human Cancers

Abstract

Adenosine-to-inosine (A-to-I) RNA editing is a widespread post-transcriptional mechanism, but its genomic landscape and clinical relevance in cancer have not been investigated systematically. We characterized the global A-to-I RNA editing profiles of 6,236 patient samples of 17 cancer types from The Cancer Genome Atlas and revealed a striking diversity of altered RNA-editing patterns in tumors relative to normal tissues. We identified an appreciable number of clinically relevant editing events, many of which are in noncoding regions. We experimentally demonstrated the effects of several cross-tumor nonsynonymous RNA editing events on cell viability and provide the evidence that RNA editing could selectively affect drug sensitivity. These results highlight RNA editing as an exciting theme for investigating cancer mechanisms, biomarkers, and treatments.

Figure 1. Overview of A-to-I RNA editing patterns in human cancer

(A) Numbers of TCGA tumor and normal samples analyzed in this study. (B) A correlation between the number of total mappable RNA-seq bases and the number of informative RNA editing sites across different cancer types. (C) The editing-level distributions at informative editing sites in different cancer types. Dashed and solid lines denote average and median for each cancer type, respectively. (D) The distributions of informative RNA editing sites in different types of RNA regions. See also Figure S1.

Figure 2. Comparison of the overall A-to-I RNA editing patterns between paired tumor and normal samples

(A) Numbers of over-editing sites and under-editing sites across different cancer types. (B) The correlation between the “net” proportion of over-editing sites (defined as the percentage of over-editing sites minus the percentage of under-editing sites) and the relative mRNA expression of ADAR1 (left), ADAR2 (middle), and ADAR3 (right) (fold change relative to normal tissues). To robustly detect a meaningful relation, the rank-based Spearman correlations were used and plotted. (C) Distribution of editing-level difference in BRCA relative to matched normal breast tissue samples (left panel) and the mRNA expression level of ADAR1 (right panel) (red in tumor and blue in normal). (D) Distribution of editing level difference in KICH samples relative to matched normal kidney samples. (A) and (B) over-editing sites are in red; under-editing sites are in blue. (C) and (D) The paired Wilcoxon test was used to assess the difference between paired tumor and normal samples. The boxes show the median±1 quartile, with whiskers extending to the most extreme data point within 1.5 interquartile range from the box boundaries. See also Table S1.

Figure 3. Identification and patterns of clinically relevant RNA editing sites

(A) The overview of clinically relevant RNA editing sites identified by three complementary computational analyses: differential analysis among tumor subtypes, differential analysis among tumor stages, and correlation analysis with patient overall survivals. An explicative cartoon is shown for illustration purposes. (B-D) Statistical significance for the enrichment or depletion patterns of clinically relevant RNA editing sites through coverage-dependent permutation tests across 12 tumor types for different types of RNA regions: gene annotation (B), non-repetitive (C), non-Alu repetitive and Alu elements, and evolutionary conservation (D). See also Table S2.

Figure 4. Clinical relevance of nonsynonymous A-to-I RNA editing sites

(A) The clinical relevance of 8 nonsynonymous RNA editing sites identified in multiple cancer types. For each cancer type, the grey box indicates not significant, the red box indicates the significant differential editing among tumor subtypes (FDR < 0.2, Diff ≥ 5%), the green box indicates the significant differential editing among stages (FDR < 0.2, Diff ≥ 5%), the blue box indicates the association with the overall survival (FDR < 0.2, Diff ≥ 5%). (B-E) The representative plots showing clinical relevance of nonsynonymous RNA editing events in AZIN1^S367D (CRC subtype: CIN, chromosomal instability; MSI, microsatellite instability) (B), COPA^I164V (STAD subtype: CIN, chromosomal instability; EBV, Epstein–Barr virus (EBV)-positive; GS, genomically stable; MSI, microsatellite instability) (C), COG3^I635V (D) and GRIA2^R764G (E). The boxes show the median±1 quartile, with whiskers extending to the most extreme data point within 1.5 interquartile range from the box boundaries. See also Figure S2, Table S3.

Figure 5. Sequenom validation and functional effects of nonsynonymous RNA editing sites on cell viability

(A) Sequenom validation of AZIN1^S367D. The upper panels show the results of a group of samples at cDNA and gDNA, respectively, where each blue symbol represents the AG genotype of a sample; while the bottom panels show the results of an individual sample in cDNA and gDNA, respectively, where there are one “A” peak and one “G” peak in cDNA but only one “A” peak in gDNA. (B) The effects of AZIN1^S367D, GRIA2^R764G and COG3^I635V in MCF10A cell viability assays. (C) The effects of AZIN1^S367D, GRIA2^R764G and COG3^I635V in BaF3 cell viability assays. Two-sided t-test was used to assess the difference. Error bars denote +/− SEM, * denotes p < 0.05, ** denotes p <0.001, and *** denotes p < 0.0001. See also Figure S3.

Figure 6. Effects of nonsynonymous RNA editing sites on drug sensitivity

(A) Spontaneously transformed Ba/F3 cells (negative control), Ba/F3 cells stably expressing AZIN1 and AZIN1^S367D, GRIA2 and GRIA2^R764G , COG3 and COG3^I635V were screened against the drug library with or without IL-3 for 72 hr. Dose-response curves for the IGF-1R inhibitor BMS536924, the MEK inhibitors CI1040 and trametinib. The drugs were dissolved in DMSO, and only DMSO was added at the drug concentration of 0 as a control. At each drug dosage, the relative cell viability (measured based on three independent replicates) was obtained by normalizing the absolute cell viability to the DMSO control to remove the baseline difference between Ba/F3 cells with and without IL-3. Error bars denote +/− SD. (B) A heatmap showing the correlations of the RNA editing levels of 35 clinically relevant nonsynonymous sites with the IC50 values of 24 clinical drugs across CCLE cell line. The highlighted boxes indicate significant correlations at FDR < 0.1.