A cytosine deaminase for programmable single-base RNA editing

See allHide authors and affiliations

Science  26 Jul 2019:
Vol. 365, Issue 6451, pp. 382-386
DOI: 10.1126/science.aax7063

Adding to the RNA editing toolbox

A previously developed RNA editing system called REPAIR can base edit A to I in RNA by fusing the adenine deaminase domain of ADAR2 with catalytically dead CRISPR-Cas13. Using directed evolution, Abudayyeh et al. demonstrated that the ADAR2 deaminase domain can be relaxed to accept other bases. This resulted in cytidine deamination activity, expanding the RNA editing toolbox for C-to-U conversion. This system, referred to as RNA Editing for Specific C-to-U Exchange (RESCUE), can edit on endogenous transcripts and enable modulation of posttranslational protein modification such as phosphorylation.

Science, this issue p. 382


Programmable RNA editing enables reversible recoding of RNA information for research and disease treatment. Previously, we developed a programmable adenosine-to-inosine (A-to-I) RNA editing approach by fusing catalytically inactivate RNA-targeting CRISPR-Cas13 (dCas13) with the adenine deaminase domain of ADAR2. Here, we report a cytidine-to-uridine (C-to-U) RNA editor, referred to as RNA Editing for Specific C-to-U Exchange (RESCUE), by directly evolving ADAR2 into a cytidine deaminase. RESCUE doubles the number of mutations targetable by RNA editing and enables modulation of phosphosignaling-relevant residues. We apply RESCUE to drive β-catenin activation and cellular growth. Furthermore, RESCUE retains A-to-I editing activity, enabling multiplexed C-to-U and A-to-I editing through the use of tailored guide RNAs.

We previously developed a RNA base editing technology called REPAIR [RNA Editing for Programmable A-to-I (G) Replacement], which uses the RNA-targeting CRISPR effector Cas13 (16) to direct the catalytic domain of ADAR2 to specific RNA transcripts to achieve adenine-to-inosine conversion with single-base precision (7). However, REPAIR, along with a number of other RNA editing technologies (815), only allows for A-to-I conversions. Technologies for precise RNA editing of cytidine-to-uridine conversions would greatly expand the range of addressable disease mutations and protein modifications (fig. S1A).

Although natural enzymes capable of catalyzing C-to-U conversion have been harnessed for DNA base editing (16, 17), they only operate on single-stranded substrates (18), exhibit off-targets (1921), and deaminate multiple bases within a window. Therefore, we took a synthetic approach to evolve the adenine deaminase domain of ADAR2 (ADAR2dd), which naturally acts on double-stranded RNA (dsRNA) substrates and preferentially deaminates a target adenine mispaired with a cytidine, into a cytidine deaminase. We fused this evolved cytidine deaminase to catalytically inactive Cas13 (dCas13) to develop programmable RNA Editing for Specific C-to-U Exchange (RESCUE) in mammalian cells (fig. S1B), which we used to activate β-catenin and modulate cell growth. Last, we improved the specificity of RESCUE more than 10-fold via rational mutagenesis, generating a highly specific and precise C-to-U RNA editing tool.

Comparison of the Escherichia coli cytidine deaminase and the human ADAR2dd showed pronounced structural homology between their catalytic cores (22), suggesting the possibility of evolving ADAR2dd into a cytidine deaminase (fig. S1B). We selected residues of ADAR2dd contacting the RNA substrate (23) for three rounds of rational mutagenesis on an ADAR2dd fused to the catalytically inactive Cas13b ortholog from Riemerella anatipestifer (dRanCas13b), yielding RESCUE round 3 (RESCUEr3) with 15% editing activity (Fig. 1, A and B, and figs. S2 and S3). We then began directed evolution across ADAR2dd to identify additional candidate mutations that increase the activity of RESCUE in yeast (see supplementary materials and methods and table S1). Sixteen rounds of evolution, culminating with the final construct RESCUEr16 (hereafter referred to as simply RESCUE), resulted in increased cytidine deamination activity across all target combinations of neighboring 5′ and 3′ bases (Fig. 1C and figs. S4 to S7). We additionally characterized guide features necessary for robust activity, finding that RESCUE is optimally active with C or U base-flips across the target base using a 30-nucleotide guide (Fig. 1C and figs. S8 to S9). Moreover, as dRanCas13b and the catalytically inactive Cas13b ortholog from Prevotella sp. P5-125 (dPspCas13b) were equivalent, the final RESCUE construct used dRanCas13b (fig. S10).

Fig. 1 Evolution of an ADAR2 deaminase domain for cytidine deamination.

(A) Schematic of RNA targeting of the catalytic residue mutant (C82R) of Gaussia luciferase reporter transcript. Single-letter abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr. (B) Heatmap depicting the percent editing levels of RESCUEr0 to RESCUEr16 on cytidines flanked by varying bases on the Gluc transcript. More favorable editing motifs are shown in the top map, whereas less favorable motifs (5′ C) are shown at the bottom. (C) Editing activity of RESCUE on all possible 16 cytidine flanking-bases motifs on the Gluc transcript with U-flip or C-flip guides. (D) Activity comparison between RESCUE, ADAR2dd without Cas13, full-length ADAR2 without Cas13, or no protein. (E) Editing efficiency of RESCUE on a panel of endogenous genes covering multiple motifs. The best guide for each site is shown, with the entire panel of guides displayed in fig. S19. Error bars in (C) to (E) indicate ±SEM (n = 3 replicates).

The 16 mutations in RESCUE are distributed throughout the structure of ADAR2dd (fig. S11A), indicating both direct interactions of the evolved residues with the RNA target within the catalytic pocket as well as indirect effects (fig. S11B). These mutations enable fitting of either adenosine or cytidine, as RESCUE is capable of both adenosine and cytidine deamination (fig. S12). We evaluated the role of each mutant by individually adding them to REPAIR or removing them from RESCUE (fig. S13). We found that mutations in the catalytic core [V351→G (V351G), K350I] and contacting the RNA target (S486A, S495N) were integral to RESCUE activity, whereas others had minor effects. Biochemical characterization of RESCUE mutations on purified ADAR2dd showed no activity on double-stranded DNA, single-stranded DNA, or DNA-RNA heteroduplexes, with the evolved mutations improving the kinetics of C-to-U editing on dsRNA substrates in vitro (fig. S14). We also found that ADAR2 or alternative RNA editing platforms without Cas13 (8, 9, 11, 13, 24) with introduced RESCUE mutations had markedly reduced editing functionality compared with Cas13b-based RESCUE (Fig. 1D and figs. S15 to S18).

We next evaluated the efficiency of RESCUE on endogenous transcripts in human embryonic kidney (HEK) 293FT cells via bulk sequencing of cell populations. We tested a variety of guide designs across 24 different sites across nine genes, as well as on 24 synthetic disease-relevant mutation targets from ClinVar, and found editing rates up to 42% (Fig. 1E, figs. S19 to S22, and table S2). Across the guides tested (tables S3 to S5), we found multiple guide design rules, most notably related to features of the motif (5′ U or A preferred) and guide mismatch position (see supplementary materials and methods).

We next applied RESCUE to alter activation of the STAT and Wnt/β-catenin pathways via modulation of key phosphorylation residues, which inhibits ubiquitination and degradation (25) (Fig. 2, A and B, and fig. S23). We tested a panel of guides targeting the β-catenin transcript (CTNNB1) at known phosphorylation residues and observed editing levels between 5 and 28% (Fig. 2C), resulting in up to fivefold activation of Wnt/β-catenin signaling (Fig. 2D) and increased cell growth in HEK293FT (Fig. 2, E and F) and human umbilical vein endothelial cells (HUVECs) (fig. S24). Given that therapeutic applications with RESCUE will require shorter constructs for viral delivery, we also evaluated RESCUE activity with C-terminal truncations of dRanCas13b and found either similar or improved deaminase activity (fig. S25).

Fig. 2 Phenotypic outcomes of RESCUE on cell growth and signaling.

(A) Schematic of β-catenin domains and RESCUE targeting guide. (B) Schematic of β-catenin activation and cell growth via RESCUE editing. (C) Percent editing by RESCUE at relevant positions in the CTNNB1 transcript. (D) Activation of Wnt/β-catenin signaling by RNA editing as measured by β-catenin–driven (TCF/LEF) luciferase expression. (E) Representative microscopy images of RESCUE CTNNB1 targeting and nontargeting guides in HEK293FT cells. (F) Quantitation of cellular growth due to activation of CTNNB1 signaling by RNA editing in HEK293FT cells. Error bars in (C), (D), and (F) indicate ±SEM (n = 3). *P < 0.05; ns, not significant; NT, nontargeting guide.

Because RESCUE retains adenosine deaminase activity (fig. S12), the native pre–CRISPR RNA (pre-crRNA) processing activity of Cas13b (4) enables multiplexed adenine and cytosine deamination. By delivering RESCUE along with a pre-crRNA targeting an adenine and a cytosine in the CTNNB1 transcript (Fig. 3A), we found that RESCUE could edit both targeted residues S33F and T41A at rates of ~15 and 5%, respectively (Fig. 3B). However, in these experiments, as well as single-plex assays, we found A-to-I off-targets near the targeted cytosine (figs. S26 and S27). To eliminate these off-targets, we introduced disfavorable guanine mismatches in the guide across from off-target adenosines (Fig. 3C), significantly reducing off-target editing while minimally disrupting the on-target editing (Fig. 3D).

Fig. 3 RESCUE and REPAIR multiplexing and specificity enhancement via guide engineering.

(A) Schematic of multiplexed C-to-U and A-to-I editing with pre-crRNA guide arrays. (B) Simultaneous C-to-U and A-to-I editing on CTNNB1 transcripts. (C) Schematic of rational engineering with guanine base-flips to prevent off-target activity at neighboring adenosine sites. (D) Percent editing at on-target C and off-target A sites for Gaussia luciferase (left) and KRAS (right) using rational introduction of disfavored base-flips. Error bars in (B) and (D) indicate ±SEM (n = 3).

Because of off-targets observed near the target site, we profiled off-targets with whole-transcriptome RNA-sequencing, finding that although RESCUE performed ~80% C-to-U editing on the Gluc transcript (Fig. 4A), it created 188 C-to-U off-targets and 1695 A-to-I off-targets, comparable to A-to-I off-targeting with REPAIRv1 (7) (Fig. 4, A and B). We thus performed rational mutagenesis of ADAR2dd at residues interacting with the RNA target (Fig. 4C), resulting in RESCUE mutants with improved specificity (Fig. 4, D to G). The mutant with the highest specificity, S375A on RESCUE (hereafter referred to as RESCUE-S), maintained ~76% on-target C-to-U editing (Fig. 4E) and created only 103 C-to-U off-targets and 139 A-to-I off-targets (Fig. 4, E to G), with reduced missense mutations and differentially regulated transcripts (figs. S28 to S31). We also found that RESCUE-S retained efficiency similar to that of RESCUE at endogenous sites with higher specificity (figs. S32 to S34).

Fig. 4 Transcriptome-wide specificity of RESCUE.

(A) On-target C-to-U editing and summary of C-to-U and A-to-I transcriptome-wide off-targets for RESCUE compared to REPAIR. (B) Manhattan plots of RESCUE A-to-I (left) and C-to-U (right) off-targets. The on-target C-to-U edit is depicted in orange. (C) Schematic of ADAR2dd interactions with RNA. Residues mutated for improving specificity are depicted in red. (D) Luciferase values for C-to-U activity with a targeting guide (y axis) and A-to-I activity with a nontargeting guide (x axis) shown for RESCUE and 95 RESCUE mutants. RESCUE is depicted in red and mutants with better specificity in blue. The T375G mutant (REPAIRv2) is shown in orange. (E) On-target C-to-U editing and summary of C-to-U and A-to-I transcriptome-wide off targets of RESCUE, REPAIR, and highest-specificity mutants. (F) Manhattan plots of RESCUE-S (+S375A) A-to-I (left) and C-to-U (right) off-targets. The on-target C-to-U edit is shown in orange. (G) Representative RNA-sequencing reads surrounding the on-target Gluc editing site (blue triangle) for RESCUE (left) and RESCUE-S (right). A-to-I edits are depicted in red; C-to-U (T) edits are shown in blue; sequencing errors are depicted in yellow. Error bars in (D) and (E) indicate ±SEM (n = 3).

RESCUE is a programmable base editing tool capable of precise cytidine-to-uridine conversion in RNA. Using directed evolution, we demonstrate that adenosine deaminases can be relaxed to accept other bases, resulting in a cytidine deamination mechanism that facilitates dsRNA editing. The larger targetable amino acid codon space of RESCUE enables modulation of more posttranslational modifications, such as phosphorylation, glycosylation, and methylation, as well as expanded targeting of common catalytic residues, disease mutations, and protective alleles, such as ApoE2 (figs. S1 and S35). Overall, RESCUE extends the base editing functionality of the RNA-targeting toolkit, allowing for expanded modeling and potential treatment of genetic diseases.

Supplementary Materials

Materials and Methods

Figs. S1 to S35

Tables S1 to S8

References (2636)

Structure S1

References and Notes

Acknowledgments: We thank R. Munshi, A. Baumann, A. Philippakis, and E. Banks for computational guidance; J. Strecker, N. Gaudelli, S. Kannan, M. Alimova, R. Wu, A. Lee, R. Wu, and I. Slaymaker for helpful discussions; P. Rogers, C. Otis, S. Saldi, and N. Pirete for flow cytometry assistance; A. Farina for help with yeast supplies; and R. Macrae, R. Belliveau, and the entire Zhang laboratory for discussions and support. Funding: O.O.A. is supported by a NIH F30 NRSA (1F30-CA210382). F.Z. is a New York Stem Cell Foundation–Robertson Investigator. F.Z. is supported by NIH grants (1R01-HG009761, 1R01-MH110049, and 1DP1-HL141201), the Howard Hughes Medical Institute, the New York Stem Cell and G. Harold and Leila Mathers foundations, the Poitras Center for Psychiatric Disorders Research at MIT, the Hock E. Tan and K. Lisa Yang Center for Autism Research at MIT, J. and P. Poitras, and the Phillips family. Author contributions: O.O.A., J.S.G., and F.Z. conceived the study. O.O.A., J.S.G., and B.F. designed and participated in all experiments. J.J. and P.K. performed off-target analysis. A.L. performed HUVEC experiments; M.J.K. and J.K. performed biochemistry assays. O.O.A, J.S.G., D.B.T.C., and F.Z. designed the early mutagenesis experiments. O.O.A., J.S.G., and F.Z. wrote the manuscript with help from all authors. Competing interests: J.S.G., O.O.A, and F.Z. are coinventors on patent applications filed by the Broad Institute relating to work in this manuscript. J.S.G., O.O.A., and F.Z. are cofounders of Sherlock Biosciences. F.Z. is a cofounder and advisor of Beam Therapeutics, Editas Medicine, Arbor Biotechnologies, and Pairwise Plants. O.O.A. and J.S.G. are advisors for Beam Therapeutics. Data and materials availability: Sequencing data are available at Sequence Read Archive under BioProject accession number PRJNA525232. Expression plasmids are available from Addgene under a uniform biological material transfer agreement; support forums and computational tools are available via the Zhang lab website (

Correction (31 October 2019): The Phillips family was omitted from the list of funding sources in the Acknowledgments. This error has been corrected.

View Abstract

Stay Connected to Science


Navigate This Article