Precise and efficient C-to-U RNA base editing with SNAP-CDAR-S

Abstract Site-directed RNA base editing enables the transient and dosable change of genetic information and represents a recent strategy to manipulate cellular processes, paving ways to novel therapeutic modalities. While tools to introduce adenosine-to-inosine changes have been explored quite intensively, the engineering of precise and programmable tools for cytidine-to-uridine editing is somewhat lacking behind. Here we demonstrate that the cytidine deaminase domain evolved from the ADAR2 adenosine deaminase, taken from the RESCUE-S tool, provides very efficient and highly programmable editing when changing the RNA targeting mechanism from Cas13-based to SNAP-tag-based. Optimization of the guide RNA chemistry further allowed to dramatically improve editing yields in the difficult-to-edit 5′-CCN sequence context thus improving the substrate scope of the tool. Regarding editing efficiency, SNAP-CDAR-S outcompeted the RESCUE-S tool clearly on all tested targets, and was highly superior in perturbing the β-catenin pathway. NGS analysis showed similar, moderate global off-target A-to-I and C-to-U editing for both tools.


INTRODUCTION
Cytidine (C) deamination yielding uridine (U) is a wellknown posttranscriptional reaction that di v ersifies genetic informa tion a t the RNA le v el ( 1 ). The enzymatic base conversion is carried out by hydrolases / deaminases belonging to the class of AID / APOBEC proteins, of which some are specific for RNA, while others can use both RNA and DNA, or only DNA as substrates. The first C-to-U RNA editing enzyme described was APOBEC1 (APO1) ( 2 ), w hich catal yzes the switch from the long ApoB100 to the short ApoB48 isoform by rewriting a glutamine codon (5 -C AA) into a STOP codon (5 -U AA) in the enterocytes of the small intestine ( 3 ). Later, single-strand RNA editing activity of further members of the APOBEC family, including APOBEC3A and 3G, was discov ered. Howe v er, the biological function and targets of their RNA editing activity remained unclear to some extent ( 1 ). C-to-U RNA editing activity is typically found only in a sub-set of tissues, like small intestine and li v er for APO1, or monocytes and macrophages for A3A, but can be up-and down-regulated in various pathologic situations and play a role in tumor e volution ( 4 ), for e xample ( 5 ). AID / APOBEC enzymes are often recruited to their targets by the help of auxiliary proteins, e.g. RBM47 ( 6 ) and A1CF ( 7 ) for APO1, and have a very strong and thus limiting pr efer ence for specific dinucleotides as editing substrates ( 1 ). Highly edited substrates, like the Glutamine-to-STOP site in the ApoB transcript, are placed in specific secondary structures that assist the recruitment and activity of the deaminase ( 8 , 9 ).
Targeted RNA base editing aims at harnessing C-to-U and A(denosine)-to-I(nosine) editing activity for the rewriting of genetic information, including the substitution of amino acids and the formation (C-to-U) or removal (Ato-I) of pr ematur e STOP codons ( 10 ). The approach opens nov el av enues for drug discovery, promising to bypass technical and ethical issues related to genome editing ( 10 ). In this field, our lab contributed an RNA-targeting platform based on fusion proteins of the self-labeling SNAP-tag ( 11 ) (Figure 1 A) ( 12 ). Initially, we engineered a programmable A-to-I RNA base editor by fusing the SNAP-tag ( 11 ) with the catalytic domain of the RNA editing enzyme ADAR (adenosine deaminase acting on RNA) ( 13 , 14 ). In these fusions, the SNAP-tag exploits its self-labeling activity to covalently tether to a guideRNA in a defined 1:1 stoichiometry by recognizing a benzylguanine (BG) moiety, the socalled self-labeling moiety, at the guideRNA. According to simple Watson-Crick base-pairing rules, the guideRNA addresses the editing of one specific adenosine residue in a selected transcript with high efficiency, broad codon scope, and very good precision ( 14 ). Competing RNA-targeting platforms have been developed based on Cas proteins ( 15 ), or trans-tethering approaches ( 16 ). While each approach has its specific strength and weakness ( 10 , 12 ), a clear advantage of the SNAP-tag approach is its human origin, its small size, the ease of transfecting of the chemically stabilized guideRNA(s) ( 14 ), the possibility of concurrent editing ( 14 ), and the ready inclusion of small molecule ( 17 ) and photo control ( 18 , 19 ). Furthermore, we have recently shown the concurrent and fully orthogonal usage of two independent RNA editing effectors by complementing a SNAP-tagged C-to-U editing effector with a HALO-tagged A-to-I editing tool within the same cell ( 20 ). In the latter study, we exploited the C-to-U deaminase domain from murine APO1. Other labs have developed C-to-U RNA base editing effectors based on human APO1 or APO3A. In the first example, RNA-targeting was based on the transtethering approach with the MS2 / MCP system ( 21 ). In the la tter, the dCas13 pla tform was applied ( 22 ). Howe v er, none of these approaches is yet working optimally. Our approach with SNAP-tagged APO1 ( 20 ) gave low editing yields on endogenous targets and its programmability, which is the addressing of any gi v en target cytidine with a guide RNA, was somehow limited due to the strong r equir ement ( 8 ) for APO1 substrates to be located in specific secondary structures. While this was better solved for the A3A target ( 22 ), this tool suffers from the strong substrate codon preference for 5 -U C . An e xciting alternati v e came from the engineering of an artificial C-to-U editing enzyme. Specifically, laboratory evolution was used to engineer the A-to-I deaminase domain of the hyperacti v e E488Q mutant ( 23 ) of the ancestor ADAR2 into a C-to-U editing enzyme ( 24 ). With a dCas13-based RNA-targeting mechanism the socalled RESCUE tool was steered to its target RNAs. While the programmability was good, the editing yields remained moderate for 5 -WC (W = A or U) codons and low for 5 -C C , whereas 5 -GC codons were hardly editable ( 24 ), mirroring the well-known codon pr efer ence ( 25 ) of ADAR2. Furthermore, the RESCUE tool retained notable A-to-I off-target editing beside C-to-U off-target editing. A highfidelity variant, RESCUE-S, was de v eloped ( 23 ), that carried an additional point mutation. Howe v er, the point mutation lowered both, the C-to-U on-target and the A-to-I off-target editing yields.
Here, we no w sho w that the high-fidelity cytidine deaminase acting on RNA (CDAR) domain from the RESCUE-S tool works very well when we replace the dCas13 domain with a SNAP-tag for RNA targeting. In particular on endogenous transcripts, the SNAP-CDAR-S outcompetes the RESCUE-S tool clearly and achie v es moderate to good editing yields for all 5 -HC codons (H = C, A, U) under very good control of A-to-I and C-to-U bystander editing.

Generation of guideRNAs
guideRN As (gRN A) were designed and purchased from Eurogentec or Sigma-Aldrich with a 5 -C6-Amino linker or a 3 -C7-Amino linker r eferr ed to as NH 2 -gRNA. sna p / (sna p) 2 -gRN A synthesis was carried out as previousl y described ( 20 ). Briefly, sna p-linker was pre-activated with EDCI for 60 min at 30 • C and (snap) 2 -linker was pre-activated with DIC for 4 h at 45 • C or over-night at 37 • C. Coupling of snap-linker to gRNA was carried out at 30 • C for 90 min and coupling of (snap) 2 -Linker was carried out at 37 • C for 2 h. Purification of gRNA was carried out by 5M urea PAGE and subsequent ethanol precipitation. Concentration and purity were determined by Nan-oDr op ™ 2000 / 2000c Spectr ophotometers (ThermoFisher Scientific). A detailed protocol is also gi v en in the Supplementary Information (pages 3-4).

Editing yield analysis
At endpoint, cells were lysed in 50 l per well (96-well forma t) RLT buf fer (Qiagen). Total RNA was isola ted using the Monarch ® RNA Cleanup Kit 10 g (New England BioLabs) following manufacturer's instructions. Target sites were amplified using either One Step RT-PCR Kit (BiotechRabbit) or One Taq ® One-Step RT-PCR Kit (New England BioLabs) and the appropriate primers. Sanger sequencing was performed by Microsynth or Eurofins. Editing yield was determined as the ratio of peak heights at target sites in the chromatogram of samples.

Functional CTNNB1 assay
Editase expressing cells were transfected as described above in technical duplicates (SNAP-CDAR-S) or quadruplicates (Cas RESCUE-S). Cells were transfected with either M50 Super 8x TOPFlash (Addgene, #12456) or M51 Super 8x FOPFlash (Addgene, #12457) and with pcDNA3.1 expressing Renilla Luciferase for normalization. Samples were also transfected either empty, or with CTNNB1 T41-targeting (sna p) 2 -gRN A or gRN A expressing plasmid (RESCUE-S), or with PPIB R7C-targeting (sna p) 2 -gRN A or gRN A expressing plasmid (RESCUE-S), or with CTNNB1 T41targeting NH 2 -gRNA or plasmid expressing no gRNA (RESCUE-S) in a 96-well format. Cells were lysed with 30 l (SNAP-CDAR-S) or 20 l (Cas RESCUE-S) per well of Passi v e Lysis Buffer (Promega) and shaken for 15 min. at room temper ature. Lucifer ase signal was measured as described before ( 26 ) using the Dual-Luciferase ® Reporter Assay System (Promega) following manufacturer's instructions with a Spark 10 M plate reader (Tecan). Briefly, 10 l of each replicate (two of the four wells for RESCUE-S were pooled) wer e measur ed by addition of 35 l of Luciferase Assay Reagent II (LAR II, Promega) and 35 l of Stop & Glo ® Reagent, and measured for 10 seconds, respecti v ely. Editing yield was determined as described above.
All luminescence measurements and editing yield determinations were conducted in biological triplicates.

Further information
For more detailed protocols and guide RNA sequences, please see additional Supplementary Information and additional Supplementary files. Detailed information on reagents , enzymes , antibodies , and kits as well as cell lines used in this study are presented in the Supporting Information.

The APOBEC1-SNAP tool suffers from low C-to-U editing yields and limited programmability
Recently, we demonstrated the harnessing of the murine APOBEC1 deaminase for site-directed C-to-U RNA base editing. For this, the SNAP-tag was fused to the C-terminus of the full length APOBEC1 enzyme, resulting in an editor called APO1S (Figure 1 A) ( 20 ). In transgenic cell lines that co-express APO1S together with SNAP-ADAR1Q, moderate editing yields were achie v ed on an eGFP reporter gene, but editing yields on the endogenous GAPDH transcript stayed below 20%. By targeting the eGFP reporter, we now tried se v eral means to improve editing yields. On the guide RNA side, the positioning of the guide RNA four to fiv e nucleotides upstream with respect to the target cytidine was most important (Figure 1 B, C). On the protein side, the localization of the editing enzyme to the cytosol was particularly necessary (Figure 1 D, E, Supplementary Figures S1-S6). Ne v ertheless, the APO1S editor suffered overall from low editing yields on endogenous targets and from low programmability (Figure 1 F), meaning that transfer to endogenous transcript was particularly difficult followed by notable guide RNA-dependent bystander editing when ontarget editing was successful (Supplementary Figure S7).

The SNAP-CDAR-S tool combines high editing yields with e x cellent programmability
In contrast, the Cas-13-mediated C-to-U editing tool called RESCUE applies a C-to-U deaminase that was e volv ed from the A-to-I deaminase ADAR2 ( 24 ), and shares with ADAR2 its strong substrate pr efer ence for double-stranded RNA. Indeed, the RESCUE tool seems to have much better programmability and on-target editing was reliably obtained when the target site was positioned inside the guide RN A / mRN A duple x. Howe v er, Cas13-mediated C-to-U editing suffers from global and local C-to-U and A-to-I offtarget editing, and attempts to create more precise tools, like Cas13-RESCUE-S, came along with largely reduced on target editing yields, hardly above 10% on endogenous transcripts ( 24 ). Howe v er, we were wondering how the engineered cytidine deaminase acting on RNA (CDAR) domain w ould w or k in the conte xt of a SNAP-tagged fusion protein ( 12 ). For this, we fused the e volv ed, high-fidelity deaminase of the RESCUE-S tool to the C-terminus of a SNAPtag ( 11 , 14 ) to obtain the SNAP-CDAR-S tool (Figure 2 A). We stably integrated a single copy of the SNAP-CDAR-S transgene into HEK 293 cells by using the Flp-In approach, ( 14 , 19 ) and found homogenous transgene expression under control of doxy cy cline. Similar to the closely related A-to-I editing enzyme SNAP-ADAR2Q (Supplementary Figure  S1), the SNAP-CDAR-S tool was mainly localized to the cytosol (Figure 2 B).
Our initial guide RNA design was inspired from our experience with the SNAP-ADAR tool and was tested for the editing of a 5 -A C A codon in a co-transfected eGFP reporter transcript. Initial guide RNAs were 22 nt long, chemically modified by 2 -O-methylation outside the target base triplet ( 27 ), which is the targeted cytidine plus its two closest neighboring bases, and carried a (snap) 2 self-labeling moiety ( 20 ) at the 5 -end for the recruitment of two SNAP-CDAR-S effectors per guide RNA. The exact composition, sequence and chemistry, of all guide RNAs can be found in the Supplementary Information (Supplementary Table  T1a). While the SNAP-ADAR tool prefers a relati v ely central positioning of the target nucleobase, the SNAP-CDAR-S effector gave clearly better yields when the target cytidine was located near the 5 -terminus of the guide RNA, e.g. design 3-C-18 in Figure 2 C, in good agreement with data from the Cas13-RESCUE ( 24 ) tool. Next, we took a closer look at the guide RNA design for the editing of the endogenous PPIB transcript, specifically, by targeting a 5 -A C G codon in its coding region (ORF). Here, we varied the length of the guide RNA (22 nt and 30 nt) and the positioning of the guide RNA relati v e to the target cytidine, see Figure  2 D. With 60% editing yield, we found the best performing guide RNA to be 30 nt long, positioning the targeted cytidine close to the 5 -end (position 4, 3-C-26) of the guide RNA in the substrate duple x. Howe v er, also other guide RNA designs gave good editing yields and the optimal design might vary to some extent for each target sequence, as suggested for the original Cas13-RESCUE-S tool ( 24 ). As we sought to determine a uni v ersal guideRNA design, we compared editing yields on further endogenous transcripts putting the target C in position 4 (3-C-26) or position 7 (6-C-23, Supplementary Figure S8). In contrast to the target site on PPIB, positioning target C at position 7 showed substantially higher editing yields for all selected sites (Supplementary Figure S8). In addition, we transferred designs pr eviously r eported ( 24 ) ideal f or f our endogenous sites to our SNAP-CDAR-S tool and compared them to our 6-C-23 standard designs (Supplementary Figure S9). For all targets, our standard design gave similar or even better editing yield then the previously described ideal design. This indica ted tha t 6-C-23 can be considered a uni v ersal design for the SNAP-CDAR-S tool. We then continued to analyze the performance further. Editing yields were sa tura ting when ≥2.5 pmol / 96 well (20 nM) guide RNA were transfected (Figure 2 E). To characterize the scope, programmability and precision of the SNAP-CDAR-S tool, we targeted fiv e different guide RNAs (all 30 nt, 6-C-23, and 2 -OMe) against fiv e differ ent cytidine bases, which wer e all located in close proximity in the ORF of the endogenous GAPDH transcript and determined on-target as well as C-to-U and A-to-I bystander editing yields. We found excellent programmability, with good on-target yields (8% to 41%) and with lacking bystander editing (detection limit Sanger sequencing ca. 5%) at neighboring cytidine or adenosine bases (Figure 2 F, the same was found for an alternati v e 3-C-18 guide RNA design, see Supplementary Figure S10). 2 -O-Methylation was shown in the past to block bystander Ato-I editing very efficiently in SNAP-ADAR tools ( 14 , 27 ), and this may contribute here to the high precision of the targeted editing too. Howe v er, we were not fully satisfied with the editing yield at the 5 -C C A codon (GAPDH S51S), which achie v ed only 8% with the best design (6-C-23, a 3-C-18 design gave < 5%). This limited scope was also reported for the Cas13-based RESCUE tool ( 24 ) and resembles the codon pr efer ence of the ancestor ADAR2 ( 25 ) protein. A r ecent structural analysis of the ADAR2 deaminase bound to a dsRNA substrate re v ealed a steric clash between the peptide backbone of glycine 489 and the minor groove face of G = C base pairs residing at the 5 neighboring position to the target adenosine ( 28 ). This steric clash could be r elax ed by replacing the 5 -neighboring G = C base pair with sterically less demanding I = C base pair (lacking an e xocy clic amino group), simply by pairing the 5 -C C A target codon with a 5 -UCI sequence in the guide RNA (Figure 2 G). Indeed, we found a 3-fold improved editing yield of 32% for the respecti v e site in GAPDH (Figure 2 H, Supplementary Figure S10B). We then systematically tested the principle f or all f our potential 5 -C C N codons (N = A, U, G, C) and found that an inosine base opposite the 5 -neighboring cytidine always improved editing at the targeted cytidine base (Figure 2 H). Even for the well-edited 5 -C C C codon (34%), we could still achie v e a notable gain in editing yield (66%).

SNAP-CDAR-S clearly outperforms Cas13-RESCUE-S on endogenous targets
To benchmark the SNAP-tagged tool with the Cas13-based tool, we generated an analogous 293 Flp-In T-REx cell line stab ly e xpressing the Cas13-based RESCUE-S on doxycycline induction. In the original work ( 24 ), RESCUE-S has always been applied by means of transient over expr ession, howe v er, this often leads to high variability in editing yields and artifacts in off-target analyses ( 12 ). We tested both editing tools side-by-side for the editing of eight different sites on fiv e different endogenous transcripts (GAPDH, PPIB, CTNNB1, STAT3, STAT1) and one disease-relevant cDNA (APOE). Most target sites were taken from the original Cas13-RESCUE-S publication ( 24 ) so that optimal Cas guide RNAs have already been reported for each of them (Supplementary Table T1b). We repeated these experiments by transfecting 300 ng / 96 well of the plasmid-borne optimal guide RNAs into the stable Cas13-RESCUE-S cells. The guide RNAs for the SNAP-CDAR-S cell lines were not optimized, but we simply transfected 5 pmol / 96 well chemically stabilized, 30 nt guide RNAs of the 6-C-23 standard design. Ne v ertheless, the SN AP-CDAR-S tool clearl y outcompeted the Cas13-RESCUE-S tool on all eight targets, achieving editing yields between 10% and 50% (Figure 3 A), while the editing yields of the Cas13-RESCUE-S tool did not achie v e editing yields abov e 10%, in accor dance with the original report ( 24 ). For four targets, only SNAP-CDAR-S, but not Cas13-RESCUE-S, was able to achieve detectable editing (PPIB S21G, CTNNB1 H63Y and P44L, STAT1 S727F). Inter estingly, thr ee out of these four examples target 5 -C C N codons, w hich is readil y done by the SNAP-CDAR-S approach with inosine-containing guide RNAs, highlighting that the SN AP-CDAR-S a pproach does not only gi v e generally higher editing yields but also increases the codon scope towards 5 -C C N sites.
Se v eral of the indicated targets are of clinical interest. The r emoval of thr eonine 41 from the ␤-catenin protein inactivates a degron and thus stabilizes the protein. ( 29 ) Enhanced le v els of ␤-catenin could be applied to boost li v er regeneration or wound healing transiently ( 30 ). We benchmarked the activation of the Wnt pathway by the SNAP-CDAR-S versus its analog Cas13 tool in a plasmid-borne luciferase assa y, f ollowing a protocol reported before ( 24 ). While the SNAP-tagged tool achie v ed 21% editing yield and an 8-fold increase in ␤-catenin activity (Figure 3 B, Supplementary Figure S11), the Cas13-RESCUE-S tool gave only 7% editing yield and 1.6-fold increase. The STAT3 protein (signal transducer and activator of transcription 3) is a multifunctional signaling molecule, which acts as a transcription factor in the nucleus, or translocates to the mitochondrium, and modulates immune response and metabolism. Its hyperactivation plays an important role in autoimmune disease, sterile inflammation, and cancer ( 31 ). Here, we  removed serine 727 from STAT3, a functionally important phosporylation site. We could achie v e up to 41% serine-togl ycine editing, w hich was accompanied by a visible reduction in S727 phosphorylation as determined by Western blot (Figure 3 C, Supplementary Figure S12). Finally, we aimed at introducing a protecti v e genotype into the apolipoprotein E (APOE) transcript, introducing the rs7412 SNP (R158C), which could transfer the neutral ε 3 allel (ca. 78% caucasian carriers) into the protecti v e ε 2 allel, which was shown to largely reduce the risk for ather oscler osis ( 32 ). Howe v er, gi v en the non-pr eferr ed natur e of the codon (5 -G C G), and the very high GC content of the surrounding sequence space, an editing yield of only 14% was achie v ed, still clearly outcompeting the Cas13 tool (Figure 3 A).

Both tools show moderate global A-to-I and C-to-U offtarget editing
We used next-generation sequencing of the poly(A)+ transcriptome (10 GB per condition) to assess the transcriptome-wide A-to-I and C-to-U off-target editing of the SNAP-CDAR-S versus its analog Cas13-RESCUE-S. We took RNA from cells expressing the respecti v e editing effector in the presence and absence of the respecti v e guide RNA and compared them to Flp-In T-REx cells not expressing an engineered effector ( 14 ). First, we compared the editing reactions against the empty Flp-In T-REx cell line and were able to detect the on-target editing event (PPIB Arg7Cys) with editing yields of 48% for SNAP-CDAR-S and 14% for Cas13-RESCUE-S, which were confirmed by Sanger sequencing (Figure 4 A). Beside the on-target editing, we found around 1000 A-to-I and three to four hundred C-to-U off-target e v ents for both effectors (Figure  4 B). As seen for ADAR-based effectors before, ( 10 , 12 , 33 ) A-to-I off-target editing was a combination of enhanced editing at known sites and editing at novel sites, whereas the large majority of C-to-U off-target editing were novel sites. We further analyzed the outcomes of the editing reactions and found that only a moderate number of all editing e v ents resulted in missense mutations (Figure 4 C).
At less than ten missense sites, the change in the off-target editing yield was increased above 25%, indicating that most missense sites are onl y marginall y affected (Figure 4 D).
The patterns between the two effectors were very similar, which was expected given that the CDAR domain of both editing tools is identical. The presence of the editing tools gave no larger changes in gene expression (Supplementary Figure S13), and both editing effectors were expressed to similar TPM le v els (Figure 4 e). Finall y, we anal yzed the guide RNA-dependent changes in editing. Clearly, the vast majority of off-target editing came from the presence of the editing enzymes and was guide RNA-independent ( Figure  4 F, G). The guide RNA-dependent C-to-U editing was very clean for the SNAP-CDAR-S tool. In contrast, the on-target editing with Cas13-RESCUE-S was covered by a small number of further editing e v ents. This became also visible when we plotted all guide RNA-dependent changes in editing yields (Figure 4 H). While the on-target site gave the largest Editing value for the SNAP-CDAR-S tool, the Cas13-RESCUE-S tool gave six A-to-I and another six C-to-U off-target e v ents with higher change in editing le v el. Ov erall, both enzymes were expressed to comparable TPM le v els, gav e v ery similar patterns and le v els of global off-target editing and mainly differed in the 5-fold higher on-target editing yield of the SNAP-CDAR-S tool.

DISCUSSION
In comparison to the APOBEC1 enzyme, the CDAR domain, e volv ed from ADAR2, performs considerably better in targeted RNA base editing tools. While the CDAR domain was taken from the Cas13-mediated RESCUE approach ( 24 ), we could show here that this deaminase domain works particularly well when the self-labeling SNAPtag is applied as the RNA-targeting mechanism. Compared to the Cas13-RESCUE-S, the SNAP-CDAR-S gave reliably higher on-target yields with less bystander editing, while the global A-to-I and C-to-U off-target effects were comparable. A reason for the superior efficiency of the SNAP-tagged tool might be the chemical design and the covalent bond that tethers the guide RNA to the SNAP-tag and may foster the encounter of guide RNA, target RNA and editing effector ( 12 , 14 ). Regarding the design of the guide RNAs with respect to chemical modifications, we found that lessons learned from the engineering of the closely related SNAP-ADAR tool ( 14 , 27 ) could be largely transferred. In particular, the general guide RNA design with 2 -O-methylation at the ribose moieties outside the base triplet and the usage of non-encodable bases like inosine opposite 5 -C C N codons have contributed to the improved performance so that editing yields between 10% and 50% are regularly achie v ed in 5 -H C N (H = C, A, U) codons, and only 5 -G C N codons remain challenging. To our knowledge, our data is the first report of stable integration of the Cas13-RESCUE-S tool and shows that it functions as well under genomic integration as it does via plasmid ov ere xpression. Ev en though Cas13-RESCUE-S was presented as a high-fidelity enzyme with largely reduced global A-to-I off-target editing before ( 24 ), there still remains notable A-to-I as well as C-to-U off-target editing, which might have been underestimated in the prior study, where off-target analysis was done on reporter cDNA under co-transfection of the editing enzyme.
Our work further allows to compare the A-to-I off-target effects of the SNAP-CDAR-S deaminase directly to the rela ted, stably integra ted wildtype and hyperacti v e (E488Q) m utant of SN AP-ADAR2 ( 14 ). W hile the of f-target A-to-I editing of the SNAP-CDAR-S tool is clearly below that of the hyperacti v e, off-target-prone SNAP-ADAR2 E488Q mutant ( 14 ), the SNAP-CDAR-S tool still has notably frequent off-target A-to-I editing when compared to the wildtype SNAP-ADAR2 enzyme (see Supplementary Figure  S14), w hich is clearl y not yet optimal for a C-to-U editing enzyme and may r equir e further engineering efforts to generate a pure C-to-U editing enzyme. Recently, improved C-to-U editing tools based on the Cas13-RESCUE platform have been described ( 34 , 35 ). Engineering efforts allo wed to lar gely reduce the size of the Cas protein, e.g. to enab le AAV-mediated deli v ery. Howe v er, these tools remain built on the original C-to-U deaminase domain taken from the RESCUE(-S) tool so that global C-to-U and in particular A-to-I off-target editing remains observable. In contrast, a split version of the CDAR domain has recently been described to strongly improve editing precision, e.g. by largely abolishing such global off-target editing. ( 36 ) Specifically, the tool uses an orthogonal trans-tethering approach, steering one CDAR half via an MS2 / MCP and the other half via a N / BoxB interaction to the target mRNA. While this was working in principle, the C-to-U editing efficiency remained very low (e.g. around 5%). Finally, C-to-U editing tools have been constructed based on the RN A / DN A editing enzyme APOBEC3A, including the Cas13-based tool CURE ( 22 ) and the PUF domain-based tool REWIRE ( 37 ). While both tools enable programmable editing, they also come with specific limitations. These include global offtarget C-to-U editing at the RNA and DNA le v el, but also a strongly limited codon scope, e.g. 5 -U C R (R = A, G). Overall, the SNAP-CDAR-S tool adds a reliable and efficient enzyme with large codon scope to the tool box for