Adenosine-to-inosine (A-to-I) RNA editing is a major source of nucleotide diversification that has significant mechanistic implications in cancer progression and treatment response. However, its activity and prevalence have not yet been systematically determined at a single-cell resolution. Chan and colleagues revealed widespread A-to-I RNA editing events in single cancer cells through an in-depth analysis of a public lung adenocarcinoma single-cell transcriptome dataset. Edits significantly enriched in cancer cells compared to other cell types have the potential to inhibit innate immune response and to predict poor therapeutic response and prognosis in patients treated with targeted therapies.
Adenosine-to-inosine (A-to-I) RNA editing is the most prevalent form of posttranscriptional modification of RNA molecules (1). Catalyzed by adenosine deaminases that act on RNA (ADAR), these nucleotide-altering events lead to diverse functional consequences involved in the full life cycle of RNA metabolism. The abundance of RNA sequencing (RNA-seq) data has enabled a quantitative gauge of their presence in the transcriptomes of many species, tissues, and disease conditions, based on a unique biochemical phenomenon through which the inosine is recognized as a guanosine. Early studies largely focused on RNA editing events equivalent to nonsynonymous mutations due to their direct effect on altering protein sequences, thereby inducing the most predictable and significant phenotypic outcome. Later, rapidly increasing annotations of new RNA editing events, as well as a comprehensive investigation of RNA editing in large-scale human sample cohorts [e.g., The Cancer Genome Atlas (TCGA; ref. 2) and GTEx (3)], have shown that the majority of these modifications reside in noncoding regions, including 3′ untranslated regions and introns. These editing activities outside of coding regions potentially exert their effects on the transcriptome through modifying the interactions with trans-factors like RNA binding proteins and miRNAs. A growing body of evidence has supported a multilayer mechanistic role played by A-to-I RNA editing in protein recoding, transcript (de)stabilization, alternative splicing, and translational regulation.
Despite their generally low editing level (<1%), several million high-confidence A-to-I RNA editing sites have been annotated in humans. While it remains controversial whether the majority of these editing events are functionally indispensable (4), a small set of RNA editing events have been reported to show well-defined physiologic consequences. For example, the pre-mRNA of GRIA2 is modified by ADAR2 at amino acid 607 in 100% of its transcripts in the human brain. This edit is found in the second transmembrane domain of the neurotransmitter receptor subunit, causing a Gln-to-Arg conversion and rendering the receptor Ca2+-impermeable (5). Furthermore, several RNA editing events have been identified as tumor-promoting mechanisms in multiple cancer types (6). A specific edit or an editing cluster in the transcript of a cancer-related gene can positively contribute to tumor progression by perturbing diverse downstream protein functions. Such examples include AZIN1 in liver cancer, CDC14B in glioblastoma, RHOQ in colorectal cancer, SLC22A3 and IGFBP7 in esophageal cancer, PODXL in gastric cancer, and GABRA3 in breast cancer. On a global scale, A-to-I RNA editing levels are generally elevated in tumors compared with matched normal tissues, and this can largely be attributed to ADAR1 overexpression. By suppressing the sensing of IFN-inducible double-stranded RNA, A-to-I RNA editing serves as an innate immune checkpoint that when counteracted with genetic deletion or therapeutic inhibition can increase sensitivity to immunotherapy in melanoma (7).
Building upon the findings derived from analyses of bulk RNA-seq data, a natural next step would be to factor in the cellular heterogeneity in terms of both tumor clonal diversity and tumor–stroma interactions. Specifically, this would mean probing the magnitude and prevalence of A-to-I RNA editing events in single-cell transcriptomes. However, this has not been a trivial task even after a decade since the birth of the first single-cell RNA-seq (scRNA-seq) technology. Both technical and biological challenges contribute to this analytic hurdle. The most widely used commercial scRNA-seq platforms are based on automatic microfluidics systems that sacrifice sequencing depth and coverage completeness for high throughput and efficiency, leading to a scarcity of high-confidence editing-supporting reads per cell, especially those bearing a protein recoding potential. Even if an editing event is supported by sufficient coverage, we must have statistical models that take into consideration the dropout effect inherent to the scRNA-seq library for a rigorous interpretation of its actual editing percentage.
In this issue of Cancer Research, for the first time, Chan and colleagues address these challenges by performing an interrogation of A-to-I RNA editing in cancers at a strictly single-cell resolution (8). Three features of this study made it possible to measure single-cell RNA editing events (Fig. 1A): (i) they utilized a unique cancer scRNA-seq dataset that has a close-to-bulk sequencing coverage generated through the SMART-seq2 platform while also containing a large number of cells (>23,000) spanning all three compartments of the tumor microenvironment, namely epithelial, stromal, and immune cells; (ii) they limited the search space to the records of REDIportal, a public catalog of high-confidence RNA editing sites derived from thousands of bulk RNA-seq data of human tissues (9); and (iii) they implemented a robust edit calling and filtering pipeline that has been well tested in previous studies (10).
On the basis of this comprehensive single-cell dissection of RNA editing in numerous cell types of lung adenocarcinoma, the authors were able to answer many intriguing questions that may not have been sufficiently addressed before. First, they asked what the cell type sources of widely observed elevated editing in bulk tumor RNA-seq data were. The difference between bulk tumor and matched normal tissue is not only the presence of a large number of cancer cells but also infiltrated immune cells and altered stromal cells, and they possibly contribute to the RNA editing activation. Surprisingly, the authors found that cancer cells almost exclusively covered the editing sites that showed increased editing levels in bulk comparisons. Further corroboration came from the observation that when directly comparing the editing signals in the same cell types between tumor and normal samples (where cancer cells are matched with epithelial cells), only cancer cells showed a dramatic increase in editing levels and not any other cell type. Meanwhile, these cancer-specific overedited sites were enriched in genes with known cancer progression functions, such as TGFβ signaling and cytoskeleton organization. Given such a strong bias of editing activation in cancer cells, it is reasonable to speculate that editing is a mechanism hijacked by cancer cells for immune evasion through the IFN-constraining effect. Indeed, viewing the relationship between the overall editing level and the expression of IFN-stimulated genes (ISG) in single cancer cells, the authors found a consistently negative association using different ISG signatures (Fig. 1B).
Dampened IFN production by cancer cells would then lead to the malfunction of immune recruitment and activation, which was captured by the authors’ analyses on a sample-level correlation between cancer-editing and immune composition that showed a significantly restrained natural killer cell presence. Once an immunosuppressive effect had been implicated in cancer cell–specific editing, the authors looked into whether this was manifested as differential clinical outcomes of patients with drastically different editing activities (Fig. 1C). Thanks to the rich clinical information associated with the original SMART-seq2 dataset, the authors were able to directly compare editing events in treatment-naïve samples with those of progressed disease or residual disease and found that editing levels in cancer cells were significantly diminished in residual disease samples while elevated in progressed disease samples. In addition, they established a metric “RNA editing load” as a significant, independent predictor for patient prognosis even when integrated with other clinical variables. Importantly, RNA editing load possessed an even stronger predictive power than tumor mutation burden and ISG expression.
The study by Chan and colleagues is a starting point of an exciting paradigm. Similar systematic analyses are warranted for other cancer types and more clinically relevant cohorts. Two questions are of interest: (i) whether RNA editing–derived features can provide robust and additional power in prognostic or predictive utility, and (ii) whether editing activities in cancer cells have a unique advantage in capturing overall tumor immunity status.
H. Liang reports grants from NIH/NCI during the conduct of the study and personal fees from Precision Scientific outside the submitted work. No disclosures were reported by the other author.
This study was supported by the NIH (U24CA264128, R01CA251150, U01CA253472, U01CA217842, P50CA221703, and P30CA016672) and Barnhart Family Distinguished Professorship in Targeted Therapies at MD Anderson Cancer Center. The authors also thank K. Mojumdar for editorial assistance.