RNA editing in cancer: Mechanistic, prognostic, and therapeutic implications

ABSTRACT

We have recently provided a comprehensive analysis of A-to-I RNA editing events in various cancer types, revealing many clinically relevant RNA editing sites and demonstrating that RNA editing can selectively affect cancer drug sensitivity. Our results unveil mechanistic, prognostic, and therapeutic implications for RNA editing in cancer.

RNA editing is an evolutionarily conserved RNA modification mechanism that introduces specific nucleotide changes into certain RNA transcripts.^1,2 The most common type of RNA editing in humans is adenosine to inosine (A to I); the newly introduced I is recognized as guanosine (G) by the translational machinery.³ As a result of advances in next-generation sequencing technology, more than one million A-to-I RNA editing sites have been detected. The majority of these editing sites are in non-coding and repetitive element regions, and their functions remain largely unknown. The functional role of A-to-I RNA editing in human cancer has been only reported for individual cases, such as AZIN1 in liver cancer.⁴ We recently performed a systematic analysis of A-to-I RNA editing in 6,236 samples across 17 cancer types of The Cancer Genome Atlas, aiming to assess the global pattern, functional importance, and clinical relevance of RNA editing in cancer.⁵

To quantify the global RNA-editing profiles of cancer samples, we first developed a computational pipeline on the basis of approximately 1.4 million high-confidence annotated RNA editing sites.⁶ To ensure the quality of our analysis, we applied a series of filters to remove potential contamination of SNPs or somatic mutations. For a certain cancer type, we then defined “informative” RNA editing sites based on detected editing signals and sample coverage. We found that the number of informative editing sites was highly correlated with the total number of mappable bases or the total number of mappable reads, suggesting that sequence depth is a key factor for the survey of RNA editing sites. The majority of these RNA editing sites were located in 3′-UTR, intronic, and intergenic regions.

To examine the patterns of RNA editing in tumors relative to normal samples, we performed a comprehensive comparison based on 12 tumor types with matched normal samples. Interestingly, we observed a diversity of altered RNA editing patterns across these cancer types: there were significant numbers of over-editing sites in head and neck, breast, thyroid, and lung cancer and under-editing sites in kidney cancers. Our further analysis suggested that among 3 RNA-editing enzymes, adenosine deaminase, RNA-specific 1, 2, and 3 (ADAR, ADARB1, ADARB2, also known as ADAR1, ADAR2 and ADAR3, respectively), ADAR1 is likely to be the major molecular determinant for the observed patterns.

To identify RNA editing events that may be functionally tumorigenic or clinically valuable, we employed a 3-step strategy (Fig. 1). First, we performed within-disease analysis to detect RNA editing sites whose editing levels correlate with established tumor subtypes, clinical stages, or patient survival. In total, we identified 3,899 clinically relevant RNA editing events across cancer types, which accounts for approximately 3.5% of the total informative editing sites examined. Our permutation tests further showed that these clinically relevant RNA editing events are depleted in Alu elements, and tend to be conserved in evolution. Among various RNA editing events, the effects of RNA editing events at nonsynonymous sites are of particular interest since they can directly change the encoded amino acids. We identified 35 nonsynonymous editing sites with a clinical relevant pattern. Pan-cancer analysis indicated that some of these sites might play an important role in more than one cancer type. We prioritized the top candidates across cancer types, including S367G in AZIN1, I635V in COG3, and R764G in GRIA2.

Figure 1.

Overall strategy for the identification of clinically relevant “driver” RNA editing events across cancer types. We first used 3 complementary correlation analyses (stage, subtype, and patient survival) to identify clinically relevant RNA editing events, and then prioritized the top candidates across cancer types, and finally performed functional characterization of these candidates using cell viability and drug sensitivity assays.

To examine the functional consequences of these RNA editing sites on tumor growth, we performed cell viability assays showing that overexpression of the mutated AZIN1^S367G, GRIA2^R764G, and COG3^I635V genes significantly increased cell survival relative to the wild-type gene in both MCF10A cells and Ba/F3 cells. To examine the potential effects of RNA editing on cancer therapy we performed a drug screening assay to evaluate whether the nonsynonymous RNA editing events affected drug sensitivity of cell lines. We found that the edited genes selectively affected the sensitivity of Ba/F3 cells to several targeted therapeutics relative to the wild-type genes, including AZIN1^S367G for the IGF-1R inhibitor BMS536924, GRIA2^R764G for MEK inhibitors CI1040 and PD0325901, and COG3^I635V for MEK inhibitors CI1040, PD0325901, and trametinib. Furthermore, we extended our analysis to the 35 clinically relevant nonsynonymous editing sites based on drug sensitivity data from the Cancer Cell Line Encyclopedia. We found that 16 of these RNA editing sites showed a significant correlation with the IC50 value of drugs across the cancer cell lines, suggesting a more common effect of RNA editing events on drug sensitivity.

Taken together, our study presents a comprehensive view of A-to-I RNA editing events in the cancer genome and provides the first evidence that RNA editing could selectively affect drug sensitivity. Thus, just like somatic mutations, some RNA editing events may act as “drivers” for tumor growth and serve as prognostic or predictive markers for patient stratification. Further efforts are required to investigate RNA editing together with other emerging players in cancer.^7,8

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Funding

This study was supported by the National Institutes of Health (R01CA175486); NIH/NCI Uterine SPORE Career Development award; the R. Lee Clark Fellow Award from The Jeanne F. Shelby Scholarship Fund; the Lorraine Dell Program in Bioinformatics for Personalization of Cancer Medicine to HL; grants from the Cancer Prevention and Research Institute of Texas (RP140462 to HL, RR150085 to LH).