Engineered ADARs enable single-nucleotide resolution DNA A-to-G editing without bystander effects

Engineered ADARs enable single-nucleotide resolution DNA A-to-G editing without bystander effects | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Biological Sciences - Article Engineered ADARs enable single-nucleotide resolution DNA A-to-G editing without bystander effects Sangsu Bae, Hyeon Woo Im, Bada Jeong, Yeji Lee, Ye Eun Oh, Chanju Jung, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7146806/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract The adenine base editor (ABE), which enables A•T-to-G•C base conversion, has emerged as a powerful tool for therapeutic applications. However, conventional ABEs suffer from bystander nucleotide conversions, limiting their utility for precise editing. Here, we present a single-nucleotide resolution ABE (snuABE) created by fusing a nickase Cas9, nCas9(H840A), with the deaminase domain of ADAR, which acts on DNA:RNA hybrids, instead of TadA, which acts on single-stranded DNA in conventional ABEs. snuABE requires a specially designed target-adenine guide RNA (tagRNA) that introduces a mismatch at the target adenine, enabling highly specific A-to-G editing by ADAR. Engineering ADAR from Pediculus humanus using the in silico protein evolution algorithm EvolvePro, along with 3’-end protection of the tagRNA, further enhances the editing activity of snuABE in human cells. Moreover, snuABE exhibits significantly reduced DNA off-target activity, highlighting its potential as a safer and more precise base editing technology for therapeutic applications. Biological sciences/Biological techniques/Genetic techniques/Gene targeting/CRISPR-Cas9 genome editing Biological sciences/Biological techniques/Genetic engineering Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Main The CRISPR-Cas system has been widely used for genetic engineering with high efficacy. CRISPR nucleases are designed to introduce DNA double-strand breaks (DSBs) at target sites in a guide RNA (gRNA)-dependent manner, primarily resulting in gene disruption 1 . However, this strategy is not suitable for many genetic diseases that require gene correction or base conversion. Additionally, CRISPR nuclease-induced DSBs can lead to serious safety concerns, including large DNA deletions 2,3 , chromosomal aberrations 4 , and p53-mediated cell death 5 . As an alternative, base editors (BEs) have gained significant attention because they enable the modification of one or a few nucleotides with high editing efficiency, without inducing DNA DSBs 6 . BEs typically consist of a nickase Cas9 (nCas9)(D10A) fused to a specific deaminase that catalyzes nucleotide conversions. However, BEs still face notable limitations, which have been partially addressed through several strategies: (i) gRNA-dependent off-target editing in the genome 7 , which can be reduced by employing high-fidelity Cas proteins and modified gRNAs 8 ; (ii) gRNA-independent off-target editing in the genome 9,10 and transcriptome 11-13 , which can be mitigated using engineered deaminases 13 ; and (iii) unintended base conversions near the target site (bystander editing) within the editing window , which can be minimized using BE variants with narrower activity windows 14 . However, despite substantial efforts, completely eliminating bystander editing remains a challenge and often hampers functional recovery 15 . Furthermore, a trade-off exists between editing efficiency and specificity: BEs with enhanced editing activity typically show increased bystander editing. These limitations underscore the need for the development of a highly precise base editing tool that can achieve accurate gene correction without bystander effects. ADARs (adenosine deaminases acting on RNA) selectively recognize and deaminate mismatched adenosines to inosines (A-to-I editing) in double-stranded RNA, a process that can alter protein function, affect RNA stability, and influence gene expression 16 . Leveraging this property, several RNA adenine base editing tools such as CIRTS, LEAPER, RESTORE, REPAIR, and RESCUE have been developed 17-21 . However, to the best of our knowledge, ADAR has not yet been applied for DNA adenine base editing in mammalian cells. In this study, we introduce a single-nucleotide resolution adenine base editor, termed snuABE, in which the deaminase domain of ADAR is fused to the N-terminus of nCas9(H840A). Unlike conventional ABEs that use single-guide RNAs (sgRNAs), snuABE employs a specialized gRNA, target-adenine guide RNA (tagRNA), which creates a mismatch at the target adenine site. For a given DNA target, snuABE can selectively convert specific adenines by designing tagRNAs that introduce unique target-specific mismatches. Notably, the overall editing efficiency of snuABE was enhanced using EvolvePro 22 , an artificial intelligence (AI)-driven directed evolution framework. Furthermore, snuABE demonstrated extremely low levels of DNA off-target editing activity, highlighting its therapeutic potential as a precise and safe base editing tool. Results Operating concept of snuABE utilizing ADAR and a dedicated guide RNA Although ADAR typically acts on double-stranded RNA ( Supplementary Fig. 1a ) 16 , a previous study has shown that it can deaminate DNA adenines within DNA:RNA duplexes in vitro 23 . Based on this finding, we hypothesized that ADAR could be repurposed for DNA adenine editing in human cells, provided that the target adenine is flipped out in an DNA:RNA hybrid conformation ( Fig. 1a ). To test this, we designed six constructs in which the deaminase domain of human ADAR2 (hADAR2dd) was fused to either the N- or C-terminus of various Cas9 variants, including nCas9(H840A), nCas9(D10A), and catalytically inactive Cas9 (dCas9) bearing both D10A and H840A mutations ( Fig. 1b ). To create a specific mismatch at the target adenine, we engineered a tagRNA by extending the 3’-end of a conventional sgRNA. We next evaluated the six snuABE constructs using a tagRNA designed to introduce a mismatch at the 12 th adenine of a FANCF target site in HEK293T cells. While ABE8e, the representative ABE varianta, edited multiple adenines at positions 12, 13, and 18 with efficiencies of up to 61.0%, snuABE exhibited lower but selective editing activity (1.1%) exclusively at the 12 th adenine. These results confirmed that snuABE can function in human cells with high positional specificity. This initial version was named snuABE1.0. ADAR enzymes were screened to identify variants with improved editing activity. Given that ADARs are found across a wide range of eukaryotes, 22 ADAR deaminase domain (ADARdd) orthologs from 20 different species, including primates, insects, and marine organisms, were retrieved from the NCBI database. Previous studies have reported that substituting a specific glutamic acid residue with glutamine (e.g., E488Q in hADAR2dd) enhances deaminase activity 24 . Therefore, we introduced analogous E-to-Q substitutions at conserved residues across all 22 ADARdd orthologs (Fig. 1d and Supplementary Fig. 1) . Remarkably, among the tested ADARdds, the deaminase domain from Pediculus humanus (PhADARdd) demonstrated superior editing activity at both the FANCF target (0.7%) and the AAVS1 target (6.0%) (Fig. 1d) . This improved version was named snuABE2.0. 3’-End protection of tagRNA increase snuABE editing efficiency Although snuABE demonstrated higher specificity than ABE8e, its relatively low editing efficiency prompted further optimization. To improve the performance of snuABE2.0, we adopted several strategies previously used to develop the optimized prime editing (PE) system, PEmax 25 . Specifically, we generated snuABE2.1 by incorporating the SV40 internal nuclear localization signal (NLS), introducing two point mutations (R221K and N394K) into nCas9(H840A), and adding a C-terminal c-Myc NLS ( Fig. 2a ). Given the structural similarity between tagRNA and the PE guide RNA (pegRNA), we hypothesized that protecting the 3’-end of tagRNA might further enhance editing efficiency 26 . To test this, we co-expressed full-length La or truncated La(1–194) protein along with snuABE2.1. At both the FANCF and AAVS1 target sites, snuABE2.1 showed improved editing efficiency over snuABE2.0, and co-expression with either form of La protein further enhanced this activity (up to 8.6% for AAVS1 and 5.1% for FANCF ) ( Fig. 2b , 2c ). Encouraged by these results, we next directly fused the truncated La(1–194) domain to various positions within snuABE2.1, including the N-terminus, internal linker region, and C-terminus ( Fig. 2a ). The highest editing efficiency was observed when La(1–194) was fused to the C-terminus, reaching 16.5% for AAVS1 and 8.5% for FANCF , which was higher than with separate co-expression of La(1–194) ( Fig. 2b , 2 c ). This optimized version was designated snuABE3.0. Based on snuABE3.0, we developed snuABE3.1 by applying additional PEmax-inspired modifications. This final construct achieved robust editing efficiencies of ~20.0% for AAVS1 and 12.1% for FANCF ( Fig. 2 b , 2 c ). We next assessed whether snuABE could selectively edit individual adenines at a given target site by introducing position-specific mismatches using tailored tagRNAs. For this purpose, we designed two tagRNAs targeting either the 13 th or 15 th adenine at the same site in DNMT1 . In parallel, we evaluated a panel of conventional ABEs, including ABEmax, ABE8e, ABE8eWQ, ABE9, and recently developed bystander-reduced variants ABEx1–4 14,27-30 . While all conventional ABEs and ABEx variants edited multiple adenines within their activity window, snuABE3.1 demonstrated clear single-nucleotide resolution. Specifically, with the tagRNA targeting the 13 th adenine, snuABE3.1 efficiently edited the 13 th adenine (5.3%) while sparing the 15 th (0.5 %). Conversely, the 15 th adenine-specific tagRNA yielded high editing at the 15 th position (6.5%), with minimal editing at the 13 th position (1.1 %) ( Fig. 2d ). These results strongly support the unique single-nucleotide selectivity of snuABE3.1. Characterization of optimized conditions of snuABE3.1 We next investigated various characteristics of snuABE3.1, including the optimal tagRNA length, base editing purity, editing window, and sequence motif preference. When testing tagRNAs with 3’ extensions ranging from 15 to 69 nucleotides (nt), the highest editing efficiencies were consistently observed with tagRNA lengths between 24 and 33 nt ( Fig. 3a and Supplementary Fig.2 ). Similar to conventional ABE8e, snuABE3.1 induced minimal insertion and deletion (indel) events (<1.0%) ( Fig. 3b ), suggesting that snuABE3.1 has a low risk of causing genomic instability. Furthermore, A-to-G conversion was the predominant outcome at target sites, with minimal non-specific base conversions ( Fig. 3b ). To investigate base-pairing requirements for efficient editing, we tested various counter-RNA bases at the position opposite the target DNA adenine in the mismatch. Only cytosine led to efficient A-to-G conversion, while other bases contributed little to no editing activity ( Fig. 3c ), consistent with previous observations 19 . To identify the effective editing window, cells were treated with snuABE3.1 using 32 tagRNAs targeting adenines at different positions. The highest editing efficiencies were consistently observed when adenines were located between positions 11 and 16 (Fig. 3d and Supplementary Fig. 3a – m) . Notably, when adenines at positions 18 and 19 were targeted, unintended bystander editing occurred at adenines located within positions 12 and 14 ( Supplementary Fig . 3j – 3 i ), further supporting a confined editing window spanning positions 11 to 16. ADAR is known to exhibit sequence preferences for RNA editing, favoring a 5’ U and 3’ G adjacent to the target adenine 31 . Therefore, we similarly examined the sequence context preferences of snuABE3.1. Using 32 target sites with varied 5’ and 3’ flanking sequences at adenines positioned between 12 and 15, we found that snuABE3.1 generally preferred A or T at the 5’ position and C or G at the 3’ position relative to the target adenine ( Fig. 3c ). Across these arbitrarily selected targets in HEK293T cells, the median editing efficiency was 2.1% ( Fig. 3f ), suggesting that snuABE3.1 is highly target-dependent. We also observed that the broader surrounding sequence context affects snuABE3.1 activity. For example, when editing an AAVS1-derived target containing a non-preferred motif (5’-GAC-3’), the editing efficiency was 17.3%, slightly lower but comparable to that of the preferred motif (5’-AAC-3’), which yielded 20.8% editing ( Supplementary Fig . 4 ). However, altering the flanking distal sequences while retaining the preferred 5’-AAC-3’ led to a significant reduction in efficiency to 6.9%. These results indicate that while snuABE3.1 exhibits a clear motif preference, its editing efficiency is also significantly influenced by the broader sequence context. Determination of off-target effects on genome and transcriptome Conventional ABEs have been reported to induce both gRNA-dependent 7 and gRNA-independent off-target edits in the genome 9,10 , as well as gRNA-independent RNA deamination events 11-13,32 . Similarly, we investigated both tagRNA-dependent and tagRNA-independent off-target effects of snuABE3.1 on the genome and transcriptome. To assess tagRNA-dependent off-target activity, we tested snuABE3.1 at the six most likely off-target sites for AAVS1 , which include sequences with up to two mismatches or one bulge. While ABE8e showed substantial off-target editing at all six sites, no detectable off-target DNA editing was observed with snuABE3.1 ( Fig. 3g and Supplementary Fig. 5a ). To evaluate tagRNA-independent genomic deamination, we employed the orthogonal R-loop assay 33 . This assay involves co-expression of BEs with a catalytically inactive Staphylococcus aureus Cas9 (dSaCas9), which forms artificial R-loop structures, thereby revealing potential off-target deaminase activity independent of gRNA targeting. Notably, while ABE8e induced strong off-target editing at all dSaCas9 binding sites, snuABE3.1 showed no detectable off-target editing at any of the tested sites ( Supplementary Fig . 5b , 5c ), despite displaying comparable on-target editing efficiency at AAVS1 ( Fig. 3h ). RNA A-to-I editing, mediated by ADARs, is a frequent and functionally important event in human cells, playing a role in post-transcriptional gene regulation 16 . Given this, we hypothesized that snuABE3.1 might induce gRNA-independent RNA deamination. To evaluate this, we first analyzed six well-characterized endogenous ADAR hotspot sites in human mRNAs 34 . Although the editing levels were lower than those typically observed with native hADARs, snuABE3.1 still induced detectable A-to-I conversions at several of these sites ( Fig. 3i and Supplementary Fig. 3a ). To further assess transcriptome-wide RNA off-target activity, we performed whole-transcriptome RNA sequencing in cells expressing snuABE1.0 (containing hADAR2dd), snuABE3.1 (containing PhADARdd), ABE8e (positive control), and nCas9(H840A) (negative control). The results revealed that snuABE3.1 exhibited global A-to-I RNA editing frequencies higher than snuABE1.0 but comparable to ABE8e ( Fig. 3j ), indicating that snuABE3.1 has strong RNA deamination activity. Enhancing editing efficiency of snuABE through AI-based directed evolution Although snuABE3.1 demonstrated usable levels of editing efficiency, its overall activity remained lower than that of conventional ABEs, and its sequence motif preference posed limitations. To address these issues, we focused on engineering the PhADARdd of snuABE3.1. Structural prediction of PhADARdd bound to a DNA:RNA hybrid using AlphaFold 3 35 revealed a prominent N-terminal flap ( Fig. 4a ). We hypothesized that while this flap might be important for the full-length PhADAR protein, it may not be essential for the catalytic PhADARdd. Therefore, we generated a truncated version of PhADARdd lacking the N-terminal 10 amino acids, which resulted in enhanced editing activity ( Fig. 4b ). This improved variant was named snuABE4.0. EvolvePro was used to further improve editing efficiency and reduce sequence bias in snuABE4.0. We first selected five endogenous genomic sites with diverse sequence motifs and target A positions ( Fig. 3e ): 5’-AAT-3’, 5’-AAG-3’, 5’-TAC-3’, 5’-GAC-3’ at position 12 (with editing efficiencies of 0.36%, 10.2%, 3.9%, and 1.02%, respectively), and 5’-AAG-3’ at position 15 (1.2% editing efficiency). In the first round, we generated 187 snuABE variants containing each sinlge random mutation in PhADARdd, and evaluated their performance into HEK293T cells. For each target, the top eight mutations were selected based on predicted editing gains by EvolvePro, resulting in 40 unique variants. These were experimentally tested across all five target sites, and the resulting data were used to iteratively retrain EvolvePro to improve prediction accuracy in subsequent rounds ( Fig. 4c ). Through successive rounds of evolution, we observed a stepwise improvement in editing efficiency. In the final (fifth) round, we utilized EvolvePro’s combinatorial mutation prediction module to identify the top 20 multi-mutant variants, each containing 2–5 mutations predicted to synergistically enhance activity. Experimental validation of these multi-mutants revealed a substantial gain in editing efficiency, with one variant harboring three mutations (E438Q, I536S, E637F) showing the highest activity. This optimized version was designated snuABE4.1 ( Fig. 4d , 4 e ). Notably, snuABE4.1 exhibited a significantly reduced motif bias, showing robust editing even at previously disfavored sequence contexts, such as targets with a 5’-G (Fig. 4f) . Additionally, overall editing efficiency improved markedly, with a median editing rate of 4.8% (Fig. 4g) , compared to 2.1% observed with snuABE3.1. MLH1-SB and additional nick further enhance editing efficiency of snuABE The snuABE system utilizes nCas9(H840A), which introduces a nick on the same DNA strand as the target adenine. This configuration may limit editing efficiency during the repair process. Therefore, similar to the PE system (e.g., PE2 and PE3), we hypothesized that introducing an additional nick on the opposite DNA strand could enhance editing efficiency ( Fig. 4e ) 36 . To test this, we co-delivered nicking guide RNAs (ngRNAs) along with either snuABE3.1 or snuABE4.1, generating the modified versions snuABE3.1n and snuABE4.1n, respectively. These systems were tested across 32 target sites with various sequence motifs. The addition of the opposite-strand nick substantially improved editing outcomes, with snuABE4.1n achieving a median editing efficiency of 5.4% ( Fig. 4g ). Furthermore, motif dependency was alleviated in snuABE4.1n ( Fig. 4f ), supporting the effectiveness of the additional nick in enhancing snuABE performance. We also considered the DNA repair pathways involved in snuABE activity. While conventional BEs primarily engage the base excision repair (BER) pathway, we posited that snuABE may also invoke mismatch repair (MMR) due to mismatches introduced on the nicked strand by nCas9(H840A), which could be targeted and corrected by the MMR system. To test this hypothesis, we co-expressed snuABE4.1 with either a dominant-negative MLH1 (MLH1dn) or a MLH1 small binder (MLH1-SB), both of which act as MMR inhibitors 25,37 . These configurations were termed snuABE4.1-MLH1dn and snuABE4.1-SB (also referred to as snuABE4.2), and were tested at the FANCF and HEK3 loci. In MMR-deficient HEK293T cells 38 , neither MMR inhibitor increased editing efficiency. However, in MMR-proficient HEK293 and HeLa cells, both inhibitors significantly enhanced editing efficiencies ( Fig. 4h ), indicating that suppression of MMR is beneficial for snuABE performance. Further improvement was achieved by combining ngRNAs with snuABE4.2, generating snuABE4.2n. This configuration led to a robust increase in editing activity across all tested loci ( Fig. 4h ). To evaluate its potential in clinically relevant contexts, we assessed snuABE4.2n activity in human fibroblasts and induced pluripotent stem cells (iPSCs) at the AAVS1 and FANCF loci. Notably, editing efficiencies were also enhanced upon ngRNA addition, reaching up to 16.3% at the AAVS1 locus in fibroblasts and 1.4% in iPSCs ( Fig. 4i , 4j ). Precise adenine correction in genetic disease models by snuABE To evaluate the therapeutic potential of snuABE, we first established a HEK293T cell model of Von Hippel-Lindau (VHL) syndrome ( Fig. 5a ) 39 . To correct the target adenine mutation, three gRNAs were designed: one targeting a 5’-NG-3’ protospacer adjacent motif (PAM) and two targeting 5’-NGG-3’ PAMs. While conventional ABEs exhibited high levels of bystander editing within the editing window, snuABE4.1n achieved bystander-free, pinpoint editing with an efficiency of up to 63.6% ( Fig. 5b ). These results demonstrate that snuABE can target previously inaccessible genomic sites where conventional ABEs are limited by bystander effects, broadening the scope for therapeutic genome editing. Low-density lipoprotein (LDL)-related disorders pose a major global health burden, and genome editing of PCSK9 has emerged as a promising strategy to reduce LDL levels by disrupting PCSK9 expression 40 . However, minimizing off-target effects is critical for clinical application. To this end, we investigated the suitability of snuABE, which is characterized by extremely low DNA off-target activity, for precise and safe PCSK9 editing in HeLa cells. Splice donor sites located downstream of exons 1 and 4 in PCSK9 were targeted using snuABE4.2, both with and without additional ngRNAs ( Fig. 5c ). Robust A-to-G conversions were achieved, reaching up to 10.5% at the exon 1 site and 12.3% at the exon 4 site ( Fig. 5d ), supporting the clinical applicability of snuABE for treating LDL-associated disorders. Discussion Although conventional ABEs are widely used due to their ease of application and high editing efficiency, they are significantly limited by the frequent occurrence of bystander editing, which constrains their therapeutic utility. Various strategies have been proposed to address this limitation, including TadA engineering 14,28,41 , sgRNA modification 42 , motif-preferred ABEs 43 , and the use of accessory proteins 44 . However, these approaches have shown limited success in achieving precise single-base editing. In this study, we developed snuABE to achieve single-nucleotide resolution adenine conversion by inducing editing specifically at mismatched adenines formed in DNA:RNA hybrids. This design significantly reduced bystander editing. Furthermore, snuABE exhibited markedly lower DNA off-target activity, likely due to the requirement for perfect complementarity between the tagRNA and the target sequence, similar to the mechanism used by prime editors. These features position snuABE as a promising and safer BE for therapeutic applications. The recent clinical success of ABE-based therapy for carbamoyl-phosphate synthetase 1 (CPS1) deficiency highlights the translational potential of BEs and underscores the need for improved precision and safety 45 . To optimize snuABE up to version 4.2, we adopted strategies from PE development, due to their similar architecture involving nCas9(H840A) and 3’-extended gRNAs. The additional mutations and NLSs used in PEmax were incorporated into snuABE2.1 25 . Likewise, introducing an additional nick on the opposite strand (as in the PE3 strategy) and protecting the 3’-end of tagRNA with La protein (as in PE7) contributed to the creation of the snuABE3 variants 25,36 . Moreover, we applied the AI-based EvolvePro algorithm to engineer the PhADARdd deaminase, generating snuABE4 variants in a manner analogous to the PE6 strategy 22,46 . MMR suppression through MLH1dn or MLH1-SB further enhanced editing efficiency, paralleling the approaches used in PE4 and PE-SB, and led to the generation of snuABE4.2 25,37 . Despite these improvements, snuABE4.2 still faces several limitations: (i) lower editing efficiency compared to conventional ABEs such as ABE8e, (ii) residual sequence preference around the target adenine, and (iii) considerable off-target editing on RNA transcripts. While RNA off-target effects are transient, these issues highlight the need for further refinement. Strategies such as genome-wide CRISPR screening to identify novel cofactors or advanced ADAR engineering, potentially via AI-guided discovery of orthologs, may offer promising paths forward. In addition, it would be beneficial to further develop snuABE to induce cytosine editing without bystander effects. Nonetheless, snuABE provides a compelling alternative to conventional ABEs, offering high specificity and minimal DNA off-target activity. While prime editors are highly versatile and capable of inducing various types of edits (insertions, deletions, substitutions), their clinical utility is often hindered by the complexity of pegRNA design and typically lower editing efficiency. In contrast, snuABE relies on a relatively simple tagRNA design, requiring only optimization of the 3’ extension length (typically 24–32 nucleotides). In a disease-relevant model of VHL syndrome, snuABE achieved bystander-free editing with efficiencies comparable to those of ABE8e. Collectively, these results underscore the potential of snuABE to therapeutically target genomic loci that remain inaccessible using current ABE technologies. Materials and Methods Molecular cloning of tagRNAs, agRNA, sgRNA, and ngRNA tagRNAs, sgRNAs, and ngRNAs used for plasmid transfection were cloned into the pU6-pegRNA-GG-acceptor plasmid (Addgene, 132777). anchor-guide RNAs (agRNAs) used with ABEx1–4 30 were cloned into the pU6-tevopreq1-GG-acceptor plasmid (Addgene, 174038). The vectors were digested with BsaI-HFv2 (New England Biolabs, R3733L). Spacers and extensions were synthesized as oligonucleotides (Macrogen), phosphorylated using T4 Polynucleotide Kinase (enzynomics, M005S), and annealed (37 °C for 30 min; 95 °C for 2 min; cooling to 25 °C at a 0.1 °C/s for 5 minutes). The annealed oligos were ligated into the digested vectors using T4 DNA ligase (enzynomics, M001S) and transformed into E. coli DH5α competent cells prepared with Mix & Go (Zymo Research, T3001). Cloning of snuABE constructs All snuABE constructs were derived from ABE8e (Addgene, 138489). Two rounds of site-directed mutagenesis were performed to introduce the A10D reversion and H840A mutation into Cas9. The resulting plasmid was digested with NotI and BglII (NEB, R0189S, R0144S), and the ADAR deaminase domains were inserted using Gibson Assembly® Master Mix (NEB, E2611L). Human ADAR1 and ADAR2 were obtained from pDY1173_ADAR1p150 (Addgene, 193192) and pmGFP-ADAR2 (Addgene, 117929), respectively. The Drosophila ADAR sequence was obtained from the HyperADAR control plasmid (Addgene, 166969). ADAR deaminase domains from other species were synthesized (Twist Bioscience). The PEmax domain used to construct snuABE2.1 was derived from pCMV-PEmax (Addgene, 174820). The La protein CDS used in snuABE3.1 was synthesized from HEK293T cDNA by PCR. Total RNA was extracted from HEK293T cells (ATCC, CRL-11268) using Trizol (Invitrogen™, 15596026), and cDNA was synthesized using ReverTra Ace (Toyobo, FSK-101F). The CDS was then amplified using KOD plus Neo (Toyobo, TOKOD-401). Mammalian cell culture conditions HEK293T (ATCC, CRL-11268), HEK293 (ATCC, CRL-1573), HeLa (ATCC, CLL-2), and U2OS (Korean Cell Line Bank, KCLB, Seoul, Republic of Korea) cells were cultured in DMEM (Welgene, LM001-05) supplemented with 10% FBS (Welgene, PK004) and 1% antibiotics (Welgene, LS203-1) at 37 °C in a 5% CO 2 incubator. For passaging, cells were washed with DPBS (Welgene, LB001-01), detached using 0.25% Trypsin-EDTA (Welgene, LS015), and resuspended in fresh media before being transferred to new plates. Human fibroblasts were maintained in DMEM supplemented with 10% FBS, 1% antibiotics, and 1× GlutaMAX™ Supplement (Gibco, 35050061) at 37 °C with 5% CO 2 . Cells were cultured in T75 flasks (SPL, 70075) with 20 mL of medium and passaged using TrypLE™ Express (Gibco, 12605010) after washing with DPBS. hiPSCs (CMC-hiPSC-022, Catholic University of Korea) were cultured in Essential 8™ medium (ThermoScientific, A1517001) on 60-mm tissue culture dishes coated with Vitronectin (Gibco™, A14700). For passaging, cells were washed with DPBS and detached using ACCUTASE™ (Stem Cell Technologies, 07922). After detachment, cells were washed twice with DPBS and transferred to Vitronectin-coated plates in Essential 8™ medium supplemented with ROCK inhibitor (Y27632 dihydrochloride, MCE, HY-10583). hiPSCs were maintained at 37 °C with 5% CO 2 . Transfection and transduction protocols and genomic DNA extraction For HEK293T, HEK293, HeLa, and U2OS cells, 0.25 × 10 5 cells were seeded per well in 48-well plates (SPL, 30048) containing 250 µL of DMEM supplemented with 10% FBS and 1% antibiotics, one day prior to transfection. The following day, plasmid vectors were mixed and transfected using 0.5 µL of jetOPTIMUS (Polyplus, 101000006) per well, according to the manufacturer’s instructions. Typically, 375 ng of snuABE and 125 ng of tagRNA were used per transfection. If required, an additional 41.6 ng of ngRNA was included (541.6 ng in total). When co-expressing additional proteins such as La protein or MLH1dn, 100 ng of plasmid was added, bringing the total to 600 ng. After 36 hours, 200 µL of medium was replaced with fresh medium (DMEM, 10% FBS, 1% antibiotics). Cells were harvested 36 hours later using 0.25% Trypsin-EDTA (Welgene, LS015). For transfection of human fibroblasts, electroporation was performed using the Neon® Transfection System (Invitrogen™) with the 10 μL Kit (Invitrogen™, MPK1096). Cells detached using TrypLE™ Express (Gibco™, 12605010) were resuspended in Neon™ R buffer at a density of 1 × 10 5 cells per reaction and mixed with 1 μg of total plasmid DNA. Electroporation was carried out at 1600 V with a 20 ms pulse width and a single pulse. After 72 hours, cells were harvested using TrypLE™ Express. For hiPSC transfection, 1 × 10 5 cells were seeded into a 24-well plate. Plasmid DNA (same total amount as above) was diluted in 25 μL of Opti-MEM™ (Gibco™, 31985-070), then mixed with 25 μL of Opti-MEM™ containing 2 μL of Lipofectamine Stem (ThermoFisher, STEM00015). The resulting 50 μL mixture was transfected into the cells. At 72 hours post-transfection, cells were washed with DPBS and detached using ACCUTASE™ (Stem Cell Technologies, 07922). Harvested cells were centrifuged at 1000 × g for 5 minutes to collect the pellet. Genomic DNA was extracted by adding 100 µL of extraction buffer (40 mM Tris, pH 8.9; 1% Tween-20; 0.2 mM EDTA; Proteinase K [20 mg/mL]; 0.2% Nonidet P-40). Samples were vortexed for 15 seconds and incubated in a thermal cycler at 60 °C for 15 minutes, followed by 98 °C for 5 minutes. Amplicon sequencing and data analysis Genomic DNA was analyzed using an Illumina MiniSeq sequencer. Targeted deep sequencing was performed using a two-step PCR process. In the first step, 2 µL of extracted DNA was amplified using adaptor primers. In the second step, 1 µL of the initial PCR product was used as the template with index primers. PCR was conducted with KOD multi & Epi (Toyobo, KME-101) following the manufacturer’s instructions under the following cycling conditions: 98 °C for 2 minutes; 29 cycles of 98 °C for 10 seconds, 56 °C for 20 seconds, and 68 °C for 30 seconds; and a final extension at 68 °C for 1 minute. The final PCR products were purified using Expin™ PCR SV (GeneAll Biotechnology, 103-102) and sequenced using the MiniSeq Mid Output Kit (Illumina, FC-420-1004). Analysis of A-to-I editing efficiency at target sites was performed using the ABE_detector pipeline (https://github.com/imhw4157/ABE_detector), and total base conversion rates were calculated using the allsub_detector pipeline (https://github.com/imhw4157/allsub_detector). RNA sequencing and data analysis The Neon® Transfection System (Invitrogen™) with the 100 μL Kit (Invitrogen™, MPK10096) was used for mRNA sequencing. HEK293T cells were detached as previously described, resuspended in DPBS, and washed by centrifugation at 500 g for 3 minutes. A total of 1 × 10 6 HEK293T cells were resuspended in 100 µL of sucrose potassium phosphate buffered saline (119 mM KCl, 15.1 mM NaCl, 0.144 mM CaCl 2 , 5 mM MgCl 2 , 9.98 mM KH 2 PO 4 , 250 mM Sucrose, pH 7.4), mixed with 2 µg of vector, and subjected to electroporation (1600 V, 20 ms pulse width, 1 pulse). The transfected cells were transferred to 6-well plates (SPL, 30006) containing 2 mL of DMEM supplemented with 10% FBS and incubated at room temperature for 5 minutes in a clean bench, followed by incubation at 37 °C with 5% CO 2 for 24 hours. After incubation, the media was removed entirely, and total RNA was extracted using 1 mL of TRIzol (Invitrogen™, 15596026) and 200 µL of Phenol:Chloroform:Isoamyl Alcohol (25:24:1, saturated with 10 mM Tris, pH 8.0, and 1 mM EDTA; Sigma-Aldrich, P2069-100ML). For targeted RNA sequencing ( Fig. 3i , j ), cDNA was synthesized using ReverTra Ace (Toyobo, FSK-101F) according to the manufacturer’s protocol and sequenced using Illumina MiniSeq as previously described. For whole-RNA sequencing ( Fig. 3k ), total RNA was processed using the NEBNext® rRNA Depletion Kit v2 (Human/Mouse/Rat) and RNA Sample Purification Beads (NEB, E7405L), followed by library preparation using the NEBNext® Ultra™ II RNA Library Prep Kit with Sample Purification Beads (NEB, E7775S), according to the manufacturer’s instructions. Sequencing was outsourced to Novogene (NovaSeq X Plus), with 6 Gb of sequencing data generated per sample. For RNA sequencing data analysis, FASTQ files were aligned to the GRCh38 reference genome (primary assembly from Ensembl) using HISAT2 47 (https://github.com/DaehwanKimLab/hisat2). The resulting BAM files were sorted and indexed using SAMtools 48 (https://www.htslib.org). A-to-I RNA editing events were analyzed using REDItools2 49 (https://github.com/BioinfoUNIBA/ REDItools2) following the published protocol. To ensure data reliability, editing sites with a read depth of less than 50 were excluded using the -l (MIN_COLUMN_LENGTH) option in REDItools2. Generation of Lenti knock-in HEK293T cells To generate HEK293T cells with an artificial motif knock-in, lentiviral transduction was employed. The lentiviral vector was constructed by digesting pAX198 (Addgene, 173042) with XbaI and XhoI (NEB, R0145S and R0146S), following the same cloning strategy used for tagRNAs, agRNAs, sgRNAs, and ngRNAs. Lenti-X™ 293T cells (Takara, 632180) were seeded at 6 × 10 6 cells per 100 mm cell culture dish (SPL, 20100), and the medium was replaced with fresh DMEM containing 10% FBS the following day. Immediately after media replacement, a mixture consisting of 9.4 µg of the constructed lentiviral vector, 8.6 µg of psPAX2 (Addgene, 12260), 2.6 µg of MD2.G (Addgene, 12259), and 2.0 µg of pRSV-Rev (Addgene, 12253) was prepared and transfected using 65.7 µL of Transporter 5® Transfection Reagent (Polysciences, 26008-50), according to the manufacturer’s instructions. One day after transfection, the media was replaced with fresh DMEM containing 10% FBS. Two days later, the media containing the viral particles was collected, centrifuged at 500 g for 10 minutes to remove cellular debris, and filtered through a 0.45 µm filter (Pall, 4614). The virus was then concentrated using the Lenti-X™ Concentrator (Takara, 631232) following the manufacturer’s protocol. HEK293T cells were infected with the concentrated lentivirus at a multiplicity of infection (MOI) of 0.3. Two days post-infection, cells were selected with 3 µg/mL puromycin (InvivoGen, ant-pr-1) for three days. Directed evolution through EvolvePro Directed evolution using the EvolvePro pipeline was conducted as previously described (https://github.com/mat10d/EvolvePro) 22 . Approximately 200 random single-point mutations were generated and tested in HEK293T cells across five independent target sites. The editing efficiencies of these variants were measured and used as training input for the initial round of EvolvePro predictions. For each target, the top eight mutations predicted to exhibit the highest activity were selected, resulting in a total of 40 variants (8 per target × 5 targets), which were then experimentally validated across all five sites in subsequent rounds. This iterative process was repeated over multiple rounds, with EvolvePro models updated using the most recent experimental data in a target-specific manner. In the final round, all mutation data were compiled, and the average fold-change in editing efficiency across the five targets was calculated. Mutations showing a fold-change greater than 1 were retained and used to generate all possible combinatorial variants containing up to four mutations. These multi-mutant variants were ranked using EvolvePro, and the top 20 predicted variants were selected for final experimental validation. Declarations Acknowledgements The authors thank Dr. Francisco X. Mora (EssayReview) for English-language editing. Most of the sequencing data analysis was conducted using the computing server at the Genomic Medicine Institute Research Service Center. This research was supported by grants from the National Research Foundation of Korea (NRF) (No. 2021M3A9H3015389, No. RS-2024-00451880, No. RS-2024-00455559, and SRC-NRF2022R1A5A102641311) awarded to S.B. Additional support was provided by the Korean Fund for Regenerative Medicine (KFRM) (No. RS-2024-00332601), a grant from the Ministry of Food and Drug Safety (No. 25202MFDS003) in 2025, and the SNUH Lee Kun-hee Child Cancer & Rare Disease Project (No. 25B-001-0700), also awarded to S.B. Author contributions H.W.I. and S.B. conceived the project. H.W.I. and C.-J.J. developed the bioinformatics algorithms. H.W.I., B.-D.J., Y.L., Y.E.O., and Y.-W.K. performed the cell experiments. H.W.I. and H.U. carried out the protein engineering. S.B. supervised the project. H.W.I. and S.B. wrote the manuscript with input from all authors. Additional information Supplementary Information accompanying this paper is available at http:// Declaration of interests H.W.I., B.-D.J., and S.B. have filed a patent application based on this work. The remaining authors declare no competing interests. References Jinek, M. et al. A Programmable Dual-RNA–Guided DNA Endonuclease in Adaptive Bacterial Immunity. Science 337 , 816-821 (2012). https://doi.org/doi:10.1126/science.1225829 Kosicki, M., Tomberg, K. & Bradley, A. Repair of double-strand breaks induced by CRISPR–Cas9 leads to large deletions and complex rearrangements. Nature Biotechnology 36 , 765-771 (2018). https://doi.org/10.1038/nbt.4192 Hwang, G.-H. et al. Large DNA deletions occur during DNA repair at 20-fold lower frequency for base editors and prime editors than for Cas9 nucleases. Nature Biomedical Engineering 9 , 79-92 (2025). https://doi.org/10.1038/s41551-024-01277-5 Cullot, G. et al. CRISPR-Cas9 genome editing induces megabase-scale chromosomal truncations. Nature Communications 10 , 1136 (2019). https://doi.org/10.1038/s41467-019-09006-2 Haapaniemi, E., Botla, S., Persson, J., Schmierer, B. & Taipale, J. CRISPR–Cas9 genome editing induces a p53-mediated DNA damage response. Nature Medicine 24 , 927-930 (2018). https://doi.org/10.1038/s41591-018-0049-z Komor, A. C., Kim, Y. B., Packer, M. S., Zuris, J. A. & Liu, D. R. Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. Nature 533 , 420-424 (2016). https://doi.org/10.1038/nature17946 Kim, D., Kim, D.-e., Lee, G., Cho, S.-I. & Kim, J.-S. Genome-wide target specificity of CRISPR RNA-guided adenine base editors. Nature Biotechnology 37 , 430-435 (2019). https://doi.org/10.1038/s41587-019-0050-1 Kleinstiver, B. P. et al. High-fidelity CRISPR–Cas9 nucleases with no detectable genome-wide off-target effects. Nature 529 , 490-495 (2016). https://doi.org/10.1038/nature16526 Jang, H.-K. et al. High-purity production and precise editing of DNA base editing ribonucleoproteins. Science Advances 7 , eabg2661 (2021). https://doi.org/doi:10.1126/sciadv.abg2661 Jin, S. et al. Cytosine, but not adenine, base editors induce genome-wide off-target mutations in rice. Science 364 , 292-295 (2019). https://doi.org/doi:10.1126/science.aaw7166 Grünewald, J. et al. Transcriptome-wide off-target RNA editing induced by CRISPR-guided DNA base editors. Nature 569 , 433-437 (2019). https://doi.org/10.1038/s41586-019-1161-z Rees, H. A., Wilson, C., Doman, J. L. & Liu, D. R. Analysis and minimization of cellular RNA editing by DNA adenine base editors. Science Advances 5 , eaax5717 (2019). https://doi.org/doi:10.1126/sciadv.aax5717 Zhou, C. et al. Off-target RNA mutation induced by DNA base editing and its elimination by mutagenesis. Nature 571 , 275-278 (2019). https://doi.org/10.1038/s41586-019-1314-0 Jeong, Y. K. et al. Adenine base editor engineering reduces editing of bystander cytosines. Nature Biotechnology 39 , 1426-1433 (2021). https://doi.org/10.1038/s41587-021-00943-2 Lee, S.-H. et al. Bystander editing by adenine base editors impairs vision restoration in a mouse model of Leber congenital amaurosis. Molecular Therapy Methods & Clinical Development 33 (2025). https://doi.org/10.1016/j.omtm.2025.101461 Nishikura, K. Functions and regulation of RNA editing by ADAR deaminases. Annu Rev Biochem 79 , 321-349 (2010). https://doi.org/10.1146/annurev-biochem-060208-105251 Rauch, S. et al. Programmable RNA-Guided RNA Effector Proteins Built from Human Parts. Cell 178 , 122-134.e112 (2019). https://doi.org/https://doi.org/10.1016/j.cell.2019.05.049 Qu, L. et al. Programmable RNA editing by recruiting endogenous ADAR using engineered RNAs. Nature Biotechnology 37 , 1059-1069 (2019). https://doi.org/10.1038/s41587-019-0178-z Cox, D. B. T. et al. RNA editing with CRISPR-Cas13. Science 358 , 1019-1027 (2017). https://doi.org/doi:10.1126/science.aaq0180 Abudayyeh, O. O. et al. A cytosine deaminase for programmable single-base RNA editing. Science 365 , 382-386 (2019). https://doi.org/doi:10.1126/science.aax7063 Merkle, T. et al. Precise RNA editing by recruiting endogenous ADARs with antisense oligonucleotides. Nature Biotechnology 37 , 133-138 (2019). https://doi.org/10.1038/s41587-019-0013-6 Jiang, K. et al. Rapid in silico directed evolution by a protein language model with EVOLVEpro. Science 387 , eadr6006 (2025). https://doi.org/doi:10.1126/science.adr6006 Zheng, Y., Lorenzo, C. & Beal, P. A. DNA editing in DNA/RNA hybrids by adenosine deaminases that act on RNA. Nucleic Acids Research 45 , 3369-3377 (2017). https://doi.org/10.1093/nar/gkx050 Kuttan, A. & Bass, B. L. Mechanistic insights into editing-site specificity of ADARs. Proceedings of the National Academy of Sciences 109 , E3295-E3304 (2012). https://doi.org/doi:10.1073/pnas.1212548109 Chen, P. J. et al. Enhanced prime editing systems by manipulating cellular determinants of editing outcomes. Cell 184 , 5635-5652.e5629 (2021). https://doi.org/https://doi.org/10.1016/j.cell.2021.09.018 Yan, J. et al. Improving prime editing with an endogenous small RNA-binding protein. Nature 628 , 639-647 (2024). https://doi.org/10.1038/s41586-024-07259-6 Richter, M. F. et al. Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity. Nature Biotechnology 38 , 883-891 (2020). https://doi.org/10.1038/s41587-020-0453-z Chen, L. et al. Engineering a precise adenine base editor with minimal bystander editing. Nature Chemical Biology 19 , 101-110 (2023). https://doi.org/10.1038/s41589-022-01163-8 Koblan, L. W. et al. Improving cytidine and adenine base editors by expression optimization and ancestral reconstruction. Nature Biotechnology 36 , 843-846 (2018). https://doi.org/10.1038/nbt.4172 Perrotta, R. M. et al. Machine Learning and Directed Evolution of Base Editing Enzymes. bioRxiv , 2024.2005.2017.594556 (2024). https://doi.org/10.1101/2024.05.17.594556 Eggington, J. M., Greene, T. & Bass, B. L. Predicting sites of ADAR editing in double-stranded RNA. Nature Communications 2 , 319 (2011). https://doi.org/10.1038/ncomms1324 Musunuru, K. et al. In vivo CRISPR base editing of PCSK9 durably lowers cholesterol in primates. Nature 593 , 429-434 (2021). https://doi.org/10.1038/s41586-021-03534-y Doman, J. L., Raguram, A., Newby, G. A. & Liu, D. R. Evaluation and minimization of Cas9-independent off-target DNA editing by cytosine base editors. Nature Biotechnology 38 , 620-628 (2020). https://doi.org/10.1038/s41587-020-0414-6 Zhang, Y., Qian, H., Xu, J. & Gao, W. ADAR, the carcinogenesis mechanisms of ADAR and related clinical applications. Ann Transl Med 7 , 686 (2019). https://doi.org/10.21037/atm.2019.11.06 Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630 , 493-500 (2024). https://doi.org/10.1038/s41586-024-07487-w Anzalone, A. V. et al. Search-and-replace genome editing without double-strand breaks or donor DNA. Nature 576 , 149-157 (2019). https://doi.org/10.1038/s41586-019-1711-4 Park, J.-C., Uhm, H., Kim, Y.-W., Oh, Y. E. & Bae, S. AI-generated small binder improves prime editing. bioRxiv , 2024.2009.2011.612443 (2024). https://doi.org/10.1101/2024.09.11.612443 Panigrahi, G. B., Slean, M. M., Simard, J. P. & Pearson, C. E. Human mismatch repair protein hMutLα is required to repair short slipped-DNAs of trinucleotide repeats. J Biol Chem 287 , 41844-41850 (2012). https://doi.org/10.1074/jbc.M112.420398 Zhang, J. et al. Germline mutations in the von Hippel-Lindau disease (VHL) gene in mainland Chinese families. J Cancer Res Clin Oncol 134 , 1211-1218 (2008). https://doi.org/10.1007/s00432-008-0399-x Ding, Q. et al. Permanent Alteration of PCSK9 With In Vivo CRISPR-Cas9 Genome Editing. Circulation Research 115 , 488-492 (2014). https://doi.org/doi:10.1161/CIRCRESAHA.115.304351 Qin, W. et al. ABE-ultramax for high-efficiency biallelic adenine base editing in zebrafish. Nature Communications 15 , 5613 (2024). https://doi.org/10.1038/s41467-024-49943-1 Zhao, D. et al. Imperfect guide-RNA (igRNA) enables CRISPR single-base editing with ABE and CBE. Nucleic Acids Research 50 , 4161-4170 (2022). https://doi.org/10.1093/nar/gkac201 Shang, M. et al. A motif preferred adenine base editor with minimal bystander and off-targets editing. bioRxiv , 2024.2011.2023.624961 (2024). https://doi.org/10.1101/2024.11.23.624961 Wang, Z. et al. Decreasing predictable DNA off-target effects and narrowing editing windows of adenine base editors by fusing human Rad18 protein variant. International Journal of Biological Macromolecules 253 , 127418 (2023). https://doi.org/https://doi.org/10.1016/j.ijbiomac.2023.127418 Musunuru, K. et al. Patient-Specific In Vivo Gene Editing to Treat a Rare Genetic Disease. New England Journal of Medicine 392 , 2235-2243 (2025). https://doi.org/doi:10.1056/NEJMoa2504747 Doman, J. L. et al. Phage-assisted evolution and protein engineering yield compact, efficient prime editors. Cell 186 , 3983-4002.e3926 (2023). https://doi.org/https://doi.org/10.1016/j.cell.2023.07.039 Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nature Biotechnology 37 , 907-915 (2019). https://doi.org/10.1038/s41587-019-0201-4 Danecek, P. et al. Twelve years of SAMtools and BCFtools. GigaScience 10 (2021). https://doi.org/10.1093/gigascience/giab008 Flati, T. et al. HPC-REDItools: a novel HPC-aware tool for improved large scale RNA-editing analysis. BMC Bioinformatics 21 , 353 (2020). https://doi.org/10.1186/s12859-020-03562-x Additional Declarations Yes there is potential Competing Interest. H.W.I., B.-D.J., and S.B. have filed a patent application based on this work. The remaining authors declare no competing interests. Supplementary Files SupplementaryInformation0717.pdf Supplementary Information Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7146806","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Biological Sciences - Article","associatedPublications":[],"authors":[{"id":493416544,"identity":"a4a6eec1-2ee0-47f5-b1b0-c7cc414f77ee","order_by":0,"name":"Sangsu Bae","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAtUlEQVRIiWNgGAWjYDACCQYG5j8VMN4BIrUw8JwhWQtvGyla5Gc3P3wgOa8uz+AA88MPDGfuEdZicOeYsYHhtsPFBgfYjCUYbhQToUUiwUwicduBxA0HGMwYGD4kEOGwGenffxycUwfUwv6NOC0MN3LMGBsbmIFaeIC23CBCi8GdM8XSDMcOJ848zFMskXCGGIfNbt/4maGmLrHvePvGDx+OEeMwOGAGYpI0jIJRMApGwSjADQCwQjxRnnLEGwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-3615-8566","institution":"Seoul National University College of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Sangsu","middleName":"","lastName":"Bae","suffix":""},{"id":493416545,"identity":"5e6b36a5-1d50-4712-b257-f5403b513e31","order_by":1,"name":"Hyeon Woo Im","email":"","orcid":"","institution":"Seoul National University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hyeon","middleName":"Woo","lastName":"Im","suffix":""},{"id":493416546,"identity":"9eda36a8-55f9-49c6-9479-5c8f7f1bd2e2","order_by":2,"name":"Bada Jeong","email":"","orcid":"https://orcid.org/0009-0003-0805-2825","institution":"Seoul National University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Bada","middleName":"","lastName":"Jeong","suffix":""},{"id":493416547,"identity":"27a605b3-c67e-4c9b-b1ff-cc64cd223501","order_by":3,"name":"Yeji Lee","email":"","orcid":"","institution":"Seoul National University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yeji","middleName":"","lastName":"Lee","suffix":""},{"id":493416548,"identity":"93b1ad65-79d7-4d1c-a888-49d2191ac722","order_by":4,"name":"Ye Eun Oh","email":"","orcid":"","institution":"Seoul National University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Ye","middleName":"Eun","lastName":"Oh","suffix":""},{"id":493416549,"identity":"1133e488-b451-466e-8856-c9dc87a6f5bc","order_by":5,"name":"Chanju Jung","email":"","orcid":"","institution":"Seoul National University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Chanju","middleName":"","lastName":"Jung","suffix":""},{"id":493416550,"identity":"c9876abf-0ffe-4f80-9dce-c0427549edc8","order_by":6,"name":"Yong-Woo Kim","email":"","orcid":"","institution":"Seoul National University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yong-Woo","middleName":"","lastName":"Kim","suffix":""},{"id":493416551,"identity":"a756f3d9-fb3c-42b0-b887-70b34eece3cc","order_by":7,"name":"Heesoo Uhm","email":"","orcid":"https://orcid.org/0000-0001-8993-3955","institution":"University of Oxford","correspondingAuthor":false,"prefix":"","firstName":"Heesoo","middleName":"","lastName":"Uhm","suffix":""}],"badges":[],"createdAt":"2025-07-17 08:46:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7146806/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7146806/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88889066,"identity":"9cc122fe-6450-40b6-a5a5-ffcabd7bca90","added_by":"auto","created_at":"2025-08-12 12:34:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":766366,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConceptualization of a novel base editor (snuABE) utilizing ADAR and cross-species ADAR screening. a,\u003c/strong\u003e Illustration comparing conventional ABEs and snuABEs, the latter utilizing an ADAR deaminase domain. \u003cstrong\u003eb\u003c/strong\u003e, Schematics of conventional ABE8e and snuABEs based on the human ADAR2 deaminase domain (ADAR2dd). \u003cstrong\u003ec\u003c/strong\u003e, Comparison of editing efficiencies at the human \u003cem\u003eFANCF\u003c/em\u003e locus using the constructs shown in \u003cstrong\u003eb\u003c/strong\u003e. \u003cstrong\u003ed\u003c/strong\u003e, Editing efficiency comparison of different ADAR proteins tested at the human \u003cem\u003eAAVS1\u003c/em\u003e and \u003cem\u003eFANCF\u003c/em\u003e loci. Cell values in heat maps represent the mean, while bar graphs show the mean ± s.d. with error bars (\u003cem\u003en\u003c/em\u003e = 3 independent biological replicates). All experiments were performed in HEK293T cells.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/24ac39b29580ff4697e81ba7.png"},{"id":88889065,"identity":"00d14fd9-d9ec-4618-a878-1f83a96e8827","added_by":"auto","created_at":"2025-08-12 12:34:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":903767,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEngineering of snuABEs and target adenine selectivity through tagRNA switching. a\u003c/strong\u003e, Engineering schemes of snuABE constructs from versions 1.0 to 3.1. \u003cstrong\u003eb\u003c/strong\u003e,\u003cstrong\u003ec\u003c/strong\u003e, Comparison of editing efficiencies at the human \u003cem\u003eAAVS1\u003c/em\u003e and \u003cem\u003eFANCF\u003c/em\u003e loci using constructs from \u003cstrong\u003ea\u003c/strong\u003e. \u003cstrong\u003ed\u003c/strong\u003e, Comparison of snuABEs with conventional ABEs at the endogenous \u003cem\u003eDNMT1\u003c/em\u003e locus. ABEx1–x4 were used with agRNAs, other ABEs with sgRNAs, and snuABEs with tagRNAs. Cell values in heat maps represent the mean, and bar graphs display the mean ± s.d. with error bars (\u003cem\u003en\u003c/em\u003e = 3 independent biological replicates). All experiments were performed in HEK293T cells.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/8a4e6fb4aecf37421faf1c00.png"},{"id":88891375,"identity":"32371a5c-13d3-46ee-8da0-1b0d05438483","added_by":"auto","created_at":"2025-08-12 12:50:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1377378,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional characterization of snuABE in genomic and transcriptomic contexts. a\u003c/strong\u003e, Comparison of snuABE editing efficiency based on tagRNA length at the human \u003cem\u003eAAVS1\u003c/em\u003e locus. \u003cstrong\u003eb\u003c/strong\u003e, Indel frequency and purity of the edited target adenine (A) derived from \u003cstrong\u003ea\u003c/strong\u003e.\u003cstrong\u003ec\u003c/strong\u003e, Editing efficiency comparison based on the base opposite the target adenine at the \u003cem\u003eAAVS1\u003c/em\u003e and \u003cem\u003eFANCF\u003c/em\u003e loci. The symbol (–) denotes a bulge. \u003cstrong\u003ed\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eEffective targeting range of snuABE across 13 distinct sites within the \u003cem\u003eAAVS1\u003c/em\u003e locus. \u003cstrong\u003ee\u003c/strong\u003e,\u003cstrong\u003ef\u003c/strong\u003e, Editing efficiency across all three-base motifs (NAN) at 32 endogenous loci. Sequence logos were generated using the ‘seqLogo’ package in R, based on normalized base editing efficiency rather than nucleotide frequency. \u003cstrong\u003eg\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eTotal base conversion rates of ABE8e and snuABE3.1 at six off-target sites identified by Cas-OFFinder.\u003cstrong\u003e h\u003c/strong\u003e, On-target vs. off-target editing activities of ABE8e and snuABE3.1 as measured by an orthogonal R-loop assay, based on the pooled data from three independent biological replicates, shown in \u003cstrong\u003eSupplementary Fig. 5b \u003c/strong\u003efor each target and off-target site. \u003cstrong\u003ei\u003c/strong\u003e, Comparison of A-to-I editing effects on seven endogenous mRNA sites using full-length human ADAR1, human ADAR2, snuABE1.0, and snuABE3.1. \u003cstrong\u003ej\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eWhole-RNA sequencing analysis of A-to-I editing ratios for ABE8e, H840A, snuABE1.0, and snuABE3.1, each performed in a single replicate. The cell values in the heat map represent mean values; bar graphs show mean ± s.d. with error bars (\u003cem\u003en\u003c/em\u003e = 3 independent biological replicates). All experiments were performed in HEK293T cells.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/1d0dd524725137052c8fc77c.png"},{"id":88890137,"identity":"cfdb2b76-9e8d-4eeb-936a-77c5c68b337d","added_by":"auto","created_at":"2025-08-12 12:42:12","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1389400,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEngineering and optimization of snuABE variants via AI-assisted evolution. a\u003c/strong\u003e, Predicted structure of the \u003cem\u003eP. humanus\u003c/em\u003eADAR deaminase domain bound to an DNA:RNA hybrid, modeled using AlphaFold3. DNA is shown in orange, RNA in blue. \u003cstrong\u003eb\u003c/strong\u003e, Editing efficiencies of truncated N-terminal flap variants of snuABE3.1. Each point represents a single replicate performed on a distinct target. \u003cstrong\u003ec\u003c/strong\u003e, Schematic of the directed evolution strategy using EvolvePro. Each point indicates the mean of two independent biological replicates per target. \u003cstrong\u003ed\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eFold-change in editing efficiency from predicted mutations identified by EvolvePro, tested across five endogenous targets. \u003cstrong\u003ee\u003c/strong\u003e, Schematic representations of snuABE4.0, snuABE4.1, snuABE4.1n, snuABE4.2, and snuABE4.2n, incorporating an additional nicking gRNA and MLH1-SB fusion. \u003cstrong\u003ef\u003c/strong\u003e,\u003cstrong\u003eg\u003c/strong\u003e, Motif-wide editing efficiencies of snuABE3.1, snuABE3.1n, snuABE4.2, and snuABE4.2n. Each data point shown in \u003cstrong\u003eg\u003c/strong\u003e individually represents the mean of each 35 targets presented in \u003cstrong\u003eSupplementary Fig. 6a,b\u003c/strong\u003e. \u003cstrong\u003eh\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eEditing efficiency comparisons of these tools at the \u003cem\u003eFANCF\u003c/em\u003e and \u003cem\u003eHEK3\u003c/em\u003e loci across multiple cell lines (HEK293T, HEK293, and HeLa). \u003cstrong\u003ei\u003c/strong\u003e,\u003cstrong\u003ej\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eA-to-G editing efficiencies of snuABE4.2(n) at the \u003cem\u003eAAVS1\u003c/em\u003e and \u003cem\u003eFANCF\u003c/em\u003e loci in human fibroblasts and induced pluripotent stem cells (iPSCs).\u003cstrong\u003e \u003c/strong\u003eThe bar graphs show the mean ± s.d. with error bars; experiments shown in \u003cstrong\u003ef\u003c/strong\u003e,\u003cstrong\u003ej\u003c/strong\u003ewere performed with n = 3 independent biological replicates, \u003cstrong\u003eh\u003c/strong\u003e with n = 4, and \u003cstrong\u003ei\u003c/strong\u003e with n = 2. Unless otherwise indicated, experiments were performed in HEK293T cells.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/ed3ab24e6541ed6fb3e08a8c.png"},{"id":88890135,"identity":"e581890b-4a05-4312-8783-fbced7a606db","added_by":"auto","created_at":"2025-08-12 12:42:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":748125,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eApplication of snuABE variants for therapeutic A-to-G correction at disease-relevant genomic loci.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e,\u003cstrong\u003eb\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eComparison of editing outcomes between conventional ABEs and snuABEs in a VHL syndrome-mimicking disease model using HEK293T cell lines. \u003cstrong\u003ec\u003c/strong\u003e,\u003cstrong\u003ed\u003c/strong\u003e Evaluation of snuABE-mediated A-to-G editing at the splicing acceptor region of the \u003cem\u003ePCSK9\u003c/em\u003e gene in HeLa cells. The cell values in the heat map represent the mean, while the bar graphs show the mean ± s.d. with error bars (\u003cem\u003en\u003c/em\u003e = 3 independent biological replicates).\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/ff70ad8ca32800cc1d8794ec.png"},{"id":88893784,"identity":"98ce5ed1-32c5-48d1-8e58-256ead0c652d","added_by":"auto","created_at":"2025-08-12 12:58:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6475841,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/052ede34-71ae-465a-80a1-57d499d6555c.pdf"},{"id":88889064,"identity":"1f15260a-0594-441c-9ab4-9f5a7cfdb284","added_by":"auto","created_at":"2025-08-12 12:34:12","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2122449,"visible":true,"origin":"","legend":"Supplementary Information","description":"","filename":"SupplementaryInformation0717.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7146806/v1/d58106930313b4977cbe31e9.pdf"}],"financialInterests":"\u003cb\u003eYes\u003c/b\u003e there is potential Competing Interest.\nH.W.I., B.-D.J., and S.B. have filed a patent application based on this work. The remaining authors declare no competing interests.","formattedTitle":"Engineered ADARs enable single-nucleotide resolution DNA A-to-G editing without bystander effects","fulltext":[{"header":"Main","content":"\u003cp\u003eThe CRISPR-Cas system has been widely used for genetic engineering with high efficacy. CRISPR nucleases are designed to introduce DNA double-strand breaks (DSBs) at target sites in a guide RNA (gRNA)-dependent manner, primarily resulting in gene disruption\u003csup\u003e1\u003c/sup\u003e. However, this strategy is not suitable for many genetic diseases that require gene correction or base conversion. Additionally, CRISPR nuclease-induced DSBs can lead to serious safety concerns, including large DNA deletions\u003csup\u003e2,3\u003c/sup\u003e, chromosomal aberrations\u003csup\u003e4\u003c/sup\u003e, and p53-mediated cell death\u003csup\u003e5\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs an alternative, base editors (BEs) have gained significant attention because they enable the modification of one or a few nucleotides with high editing efficiency, without inducing DNA DSBs\u003csup\u003e6\u003c/sup\u003e. BEs typically consist of a nickase Cas9 (nCas9)(D10A) fused to a specific deaminase that catalyzes nucleotide conversions. However, BEs still face notable limitations, which have been partially addressed through several strategies: (i) gRNA-dependent off-target editing in the genome \u003csup\u003e7\u003c/sup\u003e, which can be reduced by employing high-fidelity Cas proteins and modified gRNAs\u003csup\u003e8\u003c/sup\u003e; (ii) gRNA-independent off-target editing in the genome\u003csup\u003e9,10\u003c/sup\u003e and transcriptome\u003csup\u003e11-13\u003c/sup\u003e, which can be mitigated using engineered deaminases\u003csup\u003e13\u003c/sup\u003e; and (iii) unintended base conversions near the target site (bystander editing) within the editing window , which can be minimized using BE variants with narrower activity windows\u003csup\u003e14\u003c/sup\u003e. However, despite substantial efforts, completely eliminating bystander editing remains a challenge and often hampers functional recovery\u003csup\u003e15\u003c/sup\u003e. Furthermore, a trade-off exists between editing efficiency and specificity: BEs with enhanced editing activity typically show increased bystander editing. These limitations underscore the need for the development of a highly precise base editing tool that can achieve accurate gene correction without bystander effects.\u003c/p\u003e\n\u003cp\u003eADARs (adenosine deaminases acting on RNA) selectively recognize and deaminate mismatched adenosines to inosines (A-to-I editing) in double-stranded RNA, a process that can alter protein function, affect RNA stability, and influence gene expression\u003csup\u003e16\u003c/sup\u003e. Leveraging this property, several RNA adenine base editing tools such as CIRTS, LEAPER, RESTORE, REPAIR, and RESCUE have been developed \u003csup\u003e17-21\u003c/sup\u003e. However, to the best of our knowledge, ADAR has not yet been applied for DNA adenine base editing in mammalian cells. In this study, we introduce a\u0026nbsp;single-nucleotide resolution adenine base editor, termed snuABE, in which the deaminase domain of ADAR is fused to the N-terminus of nCas9(H840A). Unlike conventional ABEs\u0026nbsp;that use\u0026nbsp;single-guide RNAs (sgRNAs),\u0026nbsp;snuABE employs a specialized gRNA, target-adenine guide RNA (tagRNA), which creates a mismatch at the target adenine site. For a given DNA target, snuABE can selectively convert specific adenines by designing tagRNAs that introduce unique target-specific mismatches. Notably, the overall editing efficiency of snuABE was enhanced using EvolvePro\u003csup\u003e22\u003c/sup\u003e, an artificial intelligence (AI)-driven directed evolution framework. Furthermore, snuABE demonstrated extremely low levels of DNA off-target editing activity, highlighting its therapeutic potential as a precise and safe base editing tool.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eOperating concept of snuABE utilizing ADAR and a dedicated guide RNA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAlthough ADAR typically acts on double-stranded RNA (\u003cstrong\u003eSupplementary Fig.\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;1a\u003c/strong\u003e)\u003csup\u003e16\u003c/sup\u003e, a previous study has shown that it can deaminate DNA adenines within DNA:RNA duplexes \u003cem\u003ein vitro\u003c/em\u003e\u003cem\u003e\u003csup\u003e23\u003c/sup\u003e\u003c/em\u003e. Based on this finding, we hypothesized that ADAR could be repurposed for DNA adenine editing in human cells, provided that the target adenine is flipped out in an DNA:RNA\u0026nbsp;hybrid conformation\u0026nbsp;(\u003cstrong\u003eFig. 1a\u003c/strong\u003e). To test this, we designed six constructs in which the deaminase domain of human ADAR2 (hADAR2dd) was fused to either the N- or C-terminus of various Cas9 variants, including nCas9(H840A), nCas9(D10A), and catalytically inactive Cas9 (dCas9) bearing both D10A and H840A mutations (\u003cstrong\u003eFig. 1b\u003c/strong\u003e). To create a specific mismatch at the target adenine, we engineered a tagRNA by extending the 3\u0026rsquo;-end of a conventional sgRNA. We next evaluated the six snuABE constructs using a tagRNA designed to introduce a mismatch at the 12\u003csup\u003eth\u0026nbsp;\u003c/sup\u003eadenine of a\u0026nbsp;FANCF target site in HEK293T cells. While ABE8e, the representative ABE varianta, edited multiple adenines at positions 12, 13, and 18 with efficiencies of up to 61.0%, snuABE exhibited lower but selective editing activity (1.1%) exclusively at the 12\u003csup\u003eth\u003c/sup\u003e adenine. These results confirmed that snuABE can function in human cells with high positional specificity. This initial version was named snuABE1.0.\u003c/p\u003e\n\u003cp\u003eADAR enzymes were screened to identify variants with improved editing activity. Given that ADARs are found across a wide range of eukaryotes, 22 ADAR deaminase domain (ADARdd) orthologs from 20 different species, including primates, insects, and marine organisms, were retrieved from the NCBI database. Previous studies have reported that substituting a specific glutamic acid residue with glutamine (e.g., E488Q in hADAR2dd) enhances deaminase activity \u003csup\u003e24\u003c/sup\u003e. Therefore, we introduced analogous E-to-Q substitutions at conserved residues across all 22 ADARdd orthologs \u003cstrong\u003e(Fig. 1d\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Supplementary Fig. 1)\u003c/strong\u003e. Remarkably, among the tested ADARdds, the deaminase domain from \u003cem\u003ePediculus humanus\u003c/em\u003e (PhADARdd) demonstrated superior editing activity at both the FANCF target (0.7%) and the AAVS1 target (6.0%) \u003cstrong\u003e(Fig. 1d)\u003c/strong\u003e. This improved version was named snuABE2.0.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3\u0026rsquo;-End protection of tagRNA increase snuABE editing efficiency\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAlthough snuABE demonstrated higher specificity than ABE8e, its relatively low editing efficiency prompted further optimization. To improve the performance of snuABE2.0, we adopted several strategies previously used to develop the optimized prime editing (PE) system, PEmax\u003cem\u003e\u003csup\u003e25\u003c/sup\u003e\u003c/em\u003e. Specifically, we generated snuABE2.1 by incorporating the SV40 internal nuclear localization signal (NLS), introducing two point mutations (R221K and N394K) into nCas9(H840A), and adding a C-terminal c-Myc NLS (\u003cstrong\u003eFig. 2a\u003c/strong\u003e). Given the structural similarity between tagRNA and the PE guide RNA (pegRNA), we hypothesized that protecting the 3\u0026rsquo;-end of tagRNA might further enhance editing efficiency\u003csup\u003e26\u003c/sup\u003e. To test this, we co-expressed full-length La or truncated La(1\u0026ndash;194) protein along with snuABE2.1. At both the \u003cem\u003eFANCF\u003c/em\u003e and \u003cem\u003eAAVS1\u003c/em\u003e target sites, snuABE2.1 showed improved editing efficiency over snuABE2.0, and co-expression with either form of La protein further enhanced this activity (up to 8.6% for \u003cem\u003eAAVS1\u003c/em\u003e and 5.1% for \u003cem\u003eFANCF\u003c/em\u003e) (\u003cstrong\u003eFig. 2b\u003c/strong\u003e, \u003cstrong\u003e2c\u003c/strong\u003e). Encouraged by these results, we next directly fused the truncated La(1\u0026ndash;194) domain to various positions within snuABE2.1, including the N-terminus, internal linker region, and C-terminus (\u003cstrong\u003eFig. 2a\u003c/strong\u003e). The highest editing efficiency was observed when La(1\u0026ndash;194) was fused to the C-terminus, reaching 16.5% for \u003cem\u003eAAVS1\u003c/em\u003e and 8.5% for \u003cem\u003eFANCF\u003c/em\u003e, which was higher than with separate co-expression of La(1\u0026ndash;194) (\u003cstrong\u003eFig. 2b\u003c/strong\u003e, \u003cstrong\u003e2\u003c/strong\u003e\u003cstrong\u003ec\u003c/strong\u003e). This optimized version was designated snuABE3.0. Based on snuABE3.0, we developed snuABE3.1 by applying additional PEmax-inspired modifications. This final construct achieved robust editing efficiencies of ~20.0% for \u003cem\u003eAAVS1\u003c/em\u003e and 12.1% for \u003cem\u003eFANCF\u003c/em\u003e (\u003cstrong\u003eFig. 2\u003c/strong\u003e\u003cstrong\u003eb\u003c/strong\u003e, \u003cstrong\u003e2\u003c/strong\u003e\u003cstrong\u003ec\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eWe next assessed whether snuABE could selectively edit individual adenines at a given target site by introducing position-specific mismatches using tailored tagRNAs. For this purpose, we designed two tagRNAs targeting either the 13\u003csup\u003eth\u003c/sup\u003e or 15\u003csup\u003eth\u003c/sup\u003e adenine at the same site in \u003cem\u003eDNMT1\u003c/em\u003e. In parallel, we evaluated a panel of conventional ABEs, including ABEmax, ABE8e, ABE8eWQ, ABE9, and recently developed bystander-reduced variants\u0026nbsp;ABEx1\u0026ndash;4\u003csup\u003e14,27-30\u003c/sup\u003e. While all conventional ABEs and ABEx variants edited multiple adenines within their activity window, snuABE3.1 demonstrated clear single-nucleotide resolution. Specifically, with the tagRNA targeting the 13\u003csup\u003eth\u003c/sup\u003e adenine, snuABE3.1 efficiently edited the 13\u003csup\u003eth\u003c/sup\u003e adenine (5.3%) while sparing the 15\u003csup\u003eth\u003c/sup\u003e (0.5 %). Conversely, the 15\u003csup\u003eth\u003c/sup\u003e adenine-specific tagRNA yielded high editing at the 15\u003csup\u003eth\u003c/sup\u003e position (6.5%), with minimal editing at the 13\u003csup\u003eth\u003c/sup\u003e position (1.1 %) (\u003cstrong\u003eFig. 2d\u003c/strong\u003e). These results strongly support the unique single-nucleotide selectivity of snuABE3.1. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCharacterization of optimized conditions of snuABE3.1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe next investigated various characteristics of snuABE3.1, including the optimal tagRNA length, base editing purity, editing window, and sequence motif preference. When testing tagRNAs with 3\u0026rsquo; extensions ranging from 15 to 69 nucleotides (nt), the highest editing efficiencies were consistently observed with tagRNA lengths between 24 and 33 nt (\u003cstrong\u003eFig. 3a\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Supplementary Fig.2\u003c/strong\u003e). Similar to conventional ABE8e, snuABE3.1 induced minimal insertion and deletion (indel) events (\u0026lt;1.0%) (\u003cstrong\u003eFig. 3b\u003c/strong\u003e), suggesting that snuABE3.1 has a low risk of causing genomic instability. Furthermore, A-to-G conversion was the predominant outcome at target sites, with minimal non-specific base conversions (\u003cstrong\u003eFig. 3b\u003c/strong\u003e). To investigate base-pairing requirements for efficient editing, we tested various counter-RNA bases at the position opposite the target DNA adenine in the mismatch. Only cytosine led to efficient A-to-G conversion, while other bases contributed little to no editing activity (\u003cstrong\u003eFig. 3c\u003c/strong\u003e), consistent with previous observations\u003csup\u003e19\u003c/sup\u003e. To identify the effective editing window, cells were treated with snuABE3.1 using 32 tagRNAs targeting adenines at different positions. The highest editing efficiencies were consistently observed when adenines were located between positions 11 and 16 \u003cstrong\u003e(Fig. 3d\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Supplementary Fig. 3a\u003c/strong\u003e\u0026ndash;\u003cstrong\u003em)\u003c/strong\u003e. Notably, when adenines at positions 18 and 19 were targeted, unintended bystander editing occurred at adenines located within positions 12 and 14 (\u003cstrong\u003eSupplementary Fig\u003c/strong\u003e\u003cstrong\u003e. 3j\u003c/strong\u003e\u0026ndash;\u003cstrong\u003e3\u003c/strong\u003e\u003cstrong\u003ei\u003c/strong\u003e), further supporting a confined editing window spanning positions 11 to 16.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eADAR is known to exhibit sequence preferences for RNA editing, favoring a 5\u0026rsquo; U and 3\u0026rsquo; G adjacent to the target adenine\u003cdel cite=\"mailto:HyeonWoo%20Im\" datetime=\"2025-07-17T16:42\"\u003e\u0026nbsp;\u003c/del\u003e\u003csup\u003e31\u003c/sup\u003e. Therefore, we similarly examined the sequence context preferences of snuABE3.1. Using 32 target sites with varied 5\u0026rsquo; and 3\u0026rsquo; flanking sequences at adenines positioned between 12 and 15, we found that snuABE3.1 generally preferred A or T at the 5\u0026rsquo; position and C or G at the 3\u0026rsquo; position relative to the target adenine (\u003cstrong\u003eFig. 3c\u003c/strong\u003e). Across these arbitrarily selected targets in HEK293T cells, the median editing efficiency was 2.1% (\u003cstrong\u003eFig. 3f\u003c/strong\u003e), suggesting that snuABE3.1 is highly target-dependent. We also observed that the broader surrounding sequence context affects snuABE3.1 activity. For example, when editing an AAVS1-derived target containing a non-preferred motif (5\u0026rsquo;-GAC-3\u0026rsquo;), the editing efficiency was 17.3%, slightly lower but comparable to that of the preferred motif (5\u0026rsquo;-AAC-3\u0026rsquo;), which yielded 20.8% editing (\u003cstrong\u003eSupplementary Fig\u003c/strong\u003e\u003cstrong\u003e. 4\u003c/strong\u003e). However, altering the flanking distal sequences while retaining the preferred 5\u0026rsquo;-AAC-3\u0026rsquo; led to a significant reduction in efficiency to 6.9%. These results indicate that while snuABE3.1 exhibits a clear motif preference, its editing efficiency is also significantly influenced by the broader sequence context. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDetermination of off-target effects on genome and transcriptome\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConventional ABEs have been reported to induce both gRNA-dependent\u003csup\u003e7\u003c/sup\u003e and gRNA-independent off-target edits in the genome\u003csup\u003e9,10\u003c/sup\u003e, as well as gRNA-independent RNA deamination events\u003csup\u003e11-13,32\u003c/sup\u003e. Similarly, we investigated both tagRNA-dependent and tagRNA-independent off-target effects of snuABE3.1 on the genome and transcriptome. To assess tagRNA-dependent off-target activity, we tested snuABE3.1 at the six most likely off-target sites for \u003cem\u003eAAVS1\u003c/em\u003e, which include sequences with up to two mismatches or one bulge. While ABE8e showed substantial off-target editing at all six sites, no detectable off-target DNA editing was observed with snuABE3.1 (\u003cstrong\u003eFig. 3g\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Supplementary Fig. \u0026nbsp;5a\u003c/strong\u003e). To evaluate tagRNA-independent genomic deamination, we employed the orthogonal R-loop assay \u003csup\u003e33\u003c/sup\u003e. This assay involves co-expression of BEs with a catalytically inactive \u003cem\u003eStaphylococcus aureus\u003c/em\u003e Cas9 (dSaCas9), which forms artificial R-loop structures, thereby revealing potential off-target deaminase activity independent of gRNA targeting. Notably, while ABE8e induced strong off-target editing at all dSaCas9 binding sites, snuABE3.1 showed no detectable off-target editing at any of the tested sites (\u003cstrong\u003eSupplementary Fig\u003c/strong\u003e\u003cstrong\u003e. 5b\u003c/strong\u003e, \u003cstrong\u003e5c\u003c/strong\u003e), despite displaying comparable on-target editing efficiency at \u003cem\u003eAAVS1\u003c/em\u003e (\u003cstrong\u003eFig. 3h\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eRNA A-to-I editing, mediated by ADARs, is a frequent and functionally important event in human cells, playing a role in post-transcriptional gene regulation\u003csup\u003e16\u003c/sup\u003e. Given this, we hypothesized that snuABE3.1 might induce gRNA-independent RNA deamination. To evaluate this, we first analyzed six well-characterized endogenous ADAR hotspot sites in human mRNAs\u003csup\u003e34\u003c/sup\u003e. Although the editing levels were lower than those typically observed with native hADARs, snuABE3.1 still induced detectable A-to-I conversions at several of these sites (\u003cstrong\u003eFig. 3i\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Supplementary Fig. 3a\u003c/strong\u003e). To further assess transcriptome-wide RNA off-target activity, we performed whole-transcriptome RNA sequencing in cells expressing snuABE1.0 (containing hADAR2dd), snuABE3.1 (containing PhADARdd), ABE8e (positive control), and nCas9(H840A) (negative control). The results revealed that snuABE3.1 exhibited global A-to-I RNA editing frequencies higher than snuABE1.0 but comparable to ABE8e (\u003cstrong\u003eFig. 3j\u003c/strong\u003e), indicating that snuABE3.1 has strong RNA deamination activity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEnhancing editing efficiency of snuABE through AI-based directed evolution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAlthough snuABE3.1 demonstrated usable levels of editing efficiency, its overall activity remained lower than that of conventional ABEs, and its sequence motif preference posed limitations. To address these issues, we focused on engineering the PhADARdd of snuABE3.1. Structural prediction of PhADARdd bound to a DNA:RNA\u0026nbsp;hybrid using AlphaFold 3\u003csup\u003e35\u003c/sup\u003e revealed a prominent N-terminal flap (\u003cstrong\u003eFig. 4a\u003c/strong\u003e). We hypothesized that while this flap might be important for the full-length PhADAR protein, it may not be essential for the catalytic PhADARdd. Therefore, we generated a truncated version of PhADARdd lacking the N-terminal 10 amino acids, which resulted in enhanced editing activity (\u003cstrong\u003eFig. 4b\u003c/strong\u003e). This improved variant was named snuABE4.0.\u003c/p\u003e\n\u003cp\u003eEvolvePro was used to further improve editing efficiency and reduce sequence bias in snuABE4.0. We first selected five endogenous genomic sites with diverse sequence motifs and target A positions (\u003cstrong\u003eFig. 3e\u003c/strong\u003e): 5\u0026rsquo;-AAT-3\u0026rsquo;, 5\u0026rsquo;-AAG-3\u0026rsquo;, 5\u0026rsquo;-TAC-3\u0026rsquo;, 5\u0026rsquo;-GAC-3\u0026rsquo; at position 12 (with editing efficiencies of 0.36%, 10.2%, 3.9%, and 1.02%, respectively), and 5\u0026rsquo;-AAG-3\u0026rsquo; at position 15 (1.2% editing efficiency). In the first round, we generated 187 snuABE variants containing each sinlge random mutation in PhADARdd, and evaluated their performance into HEK293T cells. For each target, the top eight mutations were selected based on predicted editing gains by EvolvePro, resulting in 40 unique variants. These were experimentally tested across all five target sites, and the resulting data were used to iteratively retrain EvolvePro to improve prediction accuracy in subsequent rounds (\u003cstrong\u003eFig. 4c\u003c/strong\u003e). Through successive rounds of evolution, we observed a stepwise improvement in editing efficiency. In the final (fifth) round, we utilized EvolvePro\u0026rsquo;s combinatorial mutation prediction module to identify the top 20 multi-mutant variants, each containing 2\u0026ndash;5 mutations predicted to synergistically enhance activity. Experimental validation of these multi-mutants revealed a substantial gain in editing efficiency, with one variant harboring three mutations (E438Q, I536S, E637F) showing the highest activity. This optimized version was designated snuABE4.1 (\u003cstrong\u003eFig. 4d\u003c/strong\u003e, \u003cstrong\u003e4\u003c/strong\u003e\u003cstrong\u003ee\u003c/strong\u003e). Notably, snuABE4.1 exhibited a significantly reduced motif bias, showing robust editing even at previously disfavored sequence contexts, such as targets with a 5\u0026rsquo;-G \u003cstrong\u003e(Fig. 4f)\u003c/strong\u003e. Additionally, overall editing efficiency improved markedly, with a median editing rate of 4.8% \u003cstrong\u003e(Fig. 4g)\u003c/strong\u003e, compared to 2.1% observed with snuABE3.1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMLH1-SB and additional nick further enhance editing efficiency of snuABE\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe snuABE system utilizes nCas9(H840A), which introduces a nick on the same DNA strand as the target adenine. This configuration may limit editing efficiency during the repair process. Therefore, similar to the PE system (e.g., PE2 and PE3), we hypothesized that introducing an additional nick on the opposite DNA strand could enhance editing efficiency (\u003cstrong\u003eFig. 4e\u003c/strong\u003e)\u003csup\u003e36\u003c/sup\u003e. To test this, we co-delivered nicking guide RNAs (ngRNAs) along with either snuABE3.1 or snuABE4.1, generating the modified versions snuABE3.1n and snuABE4.1n, respectively. These systems were tested across 32 target sites with various sequence motifs. The addition of the opposite-strand nick substantially improved editing outcomes, with snuABE4.1n achieving a median editing efficiency of 5.4% (\u003cstrong\u003eFig. 4g\u003c/strong\u003e). Furthermore, motif dependency was alleviated in snuABE4.1n\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e(\u003cstrong\u003eFig. 4f\u003c/strong\u003e), supporting the effectiveness of the additional nick in enhancing snuABE performance.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe also considered the DNA repair pathways involved in snuABE activity. While conventional BEs primarily engage the base excision repair (BER) pathway, we posited that snuABE may also invoke mismatch repair (MMR) due to mismatches introduced on the nicked strand by nCas9(H840A), which could be targeted and corrected by the MMR system. To test this hypothesis, we co-expressed snuABE4.1 with either a dominant-negative MLH1 (MLH1dn) or a MLH1 small binder (MLH1-SB), both of which act as MMR inhibitors\u003csup\u003e25,37\u003c/sup\u003e. These configurations were termed snuABE4.1-MLH1dn and snuABE4.1-SB (also referred to as snuABE4.2), and were tested at the \u003cem\u003eFANCF\u003c/em\u003e and \u003cem\u003eHEK3\u003c/em\u003e loci. In MMR-deficient HEK293T cells \u003csup\u003e38\u003c/sup\u003e, neither MMR inhibitor increased editing efficiency. However, in MMR-proficient HEK293 and HeLa cells, both inhibitors significantly enhanced editing efficiencies (\u003cstrong\u003eFig. 4h\u003c/strong\u003e), indicating that suppression of MMR is beneficial for snuABE performance. Further improvement was achieved by combining ngRNAs with snuABE4.2, generating snuABE4.2n. This configuration led to a robust increase in editing activity across all tested loci (\u003cstrong\u003eFig. 4h\u003c/strong\u003e). To evaluate its potential in clinically relevant contexts, we assessed snuABE4.2n activity in human fibroblasts and induced pluripotent stem cells (iPSCs) at the \u003cem\u003eAAVS1\u003c/em\u003e and \u003cem\u003eFANCF\u003c/em\u003e loci. Notably, editing efficiencies were also enhanced upon ngRNA addition, reaching up to 16.3% at the \u003cem\u003eAAVS1\u003c/em\u003e locus in fibroblasts and 1.4% in iPSCs (\u003cstrong\u003eFig. 4i\u003c/strong\u003e,\u003cstrong\u003e\u0026nbsp;4j\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePrecise adenine correction in genetic disease models by snuABE\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo evaluate the therapeutic potential of snuABE, we first established a HEK293T cell model of Von Hippel-Lindau (VHL) syndrome (\u003cstrong\u003eFig. 5a\u003c/strong\u003e) \u003csup\u003e39\u003c/sup\u003e. To correct the target adenine mutation, three gRNAs were designed: one targeting a 5\u0026rsquo;-NG-3\u0026rsquo; protospacer adjacent motif (PAM) and two targeting 5\u0026rsquo;-NGG-3\u0026rsquo; PAMs. While conventional ABEs exhibited high levels of bystander editing within the editing window, snuABE4.1n achieved bystander-free, pinpoint editing with an efficiency of up to 63.6% (\u003cstrong\u003eFig. 5b\u003c/strong\u003e). These results demonstrate that snuABE can target previously inaccessible genomic sites where conventional ABEs are limited by bystander effects, broadening the scope for therapeutic genome editing.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLow-density lipoprotein (LDL)-related disorders pose a major global health burden, and genome editing of \u003cem\u003ePCSK9\u003c/em\u003e has emerged as a promising strategy to reduce LDL levels by disrupting \u003cem\u003ePCSK9\u003c/em\u003e expression\u003csup\u003e40\u003c/sup\u003e. However, minimizing off-target effects is critical for clinical application. To this end, we investigated the suitability of snuABE, which is characterized by extremely low DNA off-target activity, for precise and safe \u003cem\u003ePCSK9\u003c/em\u003e editing in HeLa cells. Splice donor sites located downstream of exons 1 and 4 in \u003cem\u003ePCSK9\u003c/em\u003e were targeted using snuABE4.2, both with and without additional ngRNAs (\u003cstrong\u003eFig. 5c\u003c/strong\u003e). Robust A-to-G conversions were achieved, reaching up to 10.5% at the exon 1 site and 12.3% at the exon 4 site\u0026nbsp;(\u003cstrong\u003eFig. 5d\u003c/strong\u003e), supporting the clinical applicability of snuABE for treating LDL-associated disorders.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eAlthough conventional ABEs are widely used due to their ease of application and high editing efficiency, they are significantly limited by the frequent occurrence of bystander editing, which constrains their therapeutic utility. Various strategies have been proposed to address this limitation, including TadA engineering\u003csup\u003e14,28,41\u003c/sup\u003e, sgRNA modification\u003csup\u003e42\u003c/sup\u003e, motif-preferred ABEs\u003csup\u003e43\u003c/sup\u003e, and the use of accessory proteins\u003csup\u003e44\u003c/sup\u003e. However, these approaches have shown limited success in achieving precise single-base editing. In this study, we developed snuABE to achieve single-nucleotide resolution adenine conversion by inducing editing specifically at mismatched adenines formed in DNA:RNA\u0026nbsp;hybrids. This design significantly reduced bystander editing. Furthermore, snuABE exhibited markedly lower DNA off-target activity, likely due to the requirement for perfect complementarity between the tagRNA and the target sequence, similar to the mechanism used by prime editors. These features position snuABE as a promising and safer BE for therapeutic applications. The recent clinical success of ABE-based therapy for carbamoyl-phosphate synthetase 1 (CPS1) deficiency highlights the translational potential of BEs and underscores the need for improved precision and safety\u003csup\u003e45\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eTo optimize snuABE up to version 4.2, we adopted strategies from PE development, due to their similar architecture involving nCas9(H840A) and 3\u0026rsquo;-extended gRNAs. The additional mutations and NLSs used in PEmax were incorporated into snuABE2.1 \u003csup\u003e25\u003c/sup\u003e. Likewise, introducing an additional nick on the opposite strand (as in the PE3 strategy) and protecting the 3\u0026rsquo;-end of tagRNA with La protein (as in PE7) contributed to the creation of the snuABE3 variants \u003csup\u003e25,36\u003c/sup\u003e. Moreover, we applied the AI-based EvolvePro algorithm to engineer the\u0026nbsp;PhADARdd deaminase, generating snuABE4 variants in a manner analogous to the PE6 strategy\u003csup\u003e22,46\u003c/sup\u003e. MMR suppression through MLH1dn or MLH1-SB further enhanced editing efficiency, paralleling the approaches used in PE4 and PE-SB, and led to the generation of snuABE4.2\u003csup\u003e25,37\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDespite these improvements, snuABE4.2 still faces several limitations: (i) lower editing efficiency compared to conventional ABEs such as ABE8e, (ii) residual sequence preference around the target adenine, and (iii) considerable off-target editing on RNA transcripts. While RNA off-target effects are transient, these issues highlight the need for further refinement. Strategies such as genome-wide CRISPR screening to identify novel cofactors or advanced ADAR engineering, potentially via AI-guided discovery of orthologs, may offer promising paths forward. In addition, it would be beneficial to further develop snuABE to induce cytosine editing without bystander effects.\u003c/p\u003e\n\u003cp\u003eNonetheless, snuABE provides a compelling alternative to conventional ABEs, offering high specificity and minimal DNA off-target activity. While prime editors are highly versatile and capable of inducing various types of edits (insertions, deletions, substitutions), their clinical utility is often hindered by the complexity of pegRNA design and typically lower editing efficiency. In contrast, snuABE relies on a relatively simple tagRNA design, requiring only optimization of the 3\u0026rsquo; extension length (typically 24\u0026ndash;32 nucleotides). In a disease-relevant model of VHL syndrome, snuABE achieved bystander-free editing with efficiencies comparable to those of ABE8e. Collectively, these results underscore the potential of snuABE to therapeutically target genomic loci that remain inaccessible using current ABE technologies.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003eMolecular cloning of tagRNAs, agRNA, sgRNA, and ngRNA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003etagRNAs, sgRNAs, and ngRNAs used for plasmid transfection were cloned into the pU6-pegRNA-GG-acceptor plasmid (Addgene, 132777). anchor-guide RNAs (agRNAs) used with ABEx1\u0026ndash;4\u003csup\u003e30\u003c/sup\u003e were cloned into the pU6-tevopreq1-GG-acceptor plasmid (Addgene, 174038). The vectors were digested with BsaI-HFv2 (New England Biolabs, R3733L). Spacers and extensions were synthesized as oligonucleotides (Macrogen), phosphorylated using T4 Polynucleotide Kinase (enzynomics, M005S), and annealed (37\u0026thinsp;\u0026deg;C for 30 min; 95\u0026thinsp;\u0026deg;C for 2 min; cooling to 25\u0026thinsp;\u0026deg;C at a 0.1 \u0026deg;C/s for 5 minutes). The annealed oligos were ligated into the digested vectors using T4 DNA ligase (enzynomics, M001S) and transformed into \u003cem\u003eE. coli\u003c/em\u003e DH5\u0026alpha; competent cells prepared with Mix \u0026amp; Go (Zymo Research, T3001).\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eCloning of snuABE constructs\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll snuABE constructs were derived from ABE8e (Addgene, 138489). Two rounds of site-directed mutagenesis were performed to introduce the A10D reversion and H840A mutation into Cas9. The resulting plasmid was digested with NotI and BglII (NEB, R0189S, R0144S), and the ADAR deaminase domains were inserted using Gibson Assembly\u0026reg; Master Mix (NEB, E2611L). Human ADAR1 and ADAR2 were obtained from pDY1173_ADAR1p150 (Addgene, 193192) and pmGFP-ADAR2 (Addgene, 117929), respectively. The \u003cem\u003eDrosophila\u003c/em\u003e ADAR sequence was obtained from the HyperADAR control plasmid (Addgene, 166969). ADAR deaminase domains from other species were synthesized (Twist Bioscience). The PEmax domain used to construct snuABE2.1 was derived from pCMV-PEmax (Addgene, 174820). The La protein CDS used in snuABE3.1 was synthesized from HEK293T cDNA by PCR. Total RNA was extracted from HEK293T cells (ATCC, CRL-11268) using Trizol (Invitrogen\u0026trade;, 15596026), and cDNA was synthesized using ReverTra Ace (Toyobo, FSK-101F). The CDS was then amplified using KOD plus Neo (Toyobo, TOKOD-401).\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eMammalian cell culture conditions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHEK293T (ATCC, CRL-11268), HEK293 (ATCC, CRL-1573), HeLa (ATCC, CLL-2), and U2OS (Korean Cell Line Bank, KCLB, Seoul, Republic of Korea) cells were cultured in DMEM (Welgene, LM001-05) supplemented with 10% FBS (Welgene, PK004) and 1% antibiotics (Welgene, LS203-1) at 37 \u0026deg;C in a 5% CO\u003csub\u003e2\u003c/sub\u003e incubator. For passaging, cells were washed with DPBS (Welgene, LB001-01), detached using 0.25% Trypsin-EDTA (Welgene, LS015), and resuspended in fresh media before being transferred to new plates. Human fibroblasts were maintained in DMEM supplemented with 10% FBS, 1% antibiotics, and 1\u0026times; GlutaMAX\u0026trade; Supplement (Gibco, 35050061) at 37 \u0026deg;C with 5% CO\u003csub\u003e2\u003c/sub\u003e. Cells were cultured in T75 flasks (SPL, 70075) with 20 mL of medium and passaged using TrypLE\u0026trade; Express (Gibco, 12605010) after washing with DPBS. hiPSCs (CMC-hiPSC-022, Catholic University of Korea) were cultured in Essential 8\u0026trade; medium (ThermoScientific, A1517001) on 60-mm tissue culture dishes coated with Vitronectin (Gibco\u0026trade;, A14700). For passaging, cells were washed with DPBS and detached using ACCUTASE\u0026trade; (Stem Cell Technologies, 07922). After detachment, cells were washed twice with DPBS and transferred to Vitronectin-coated plates in Essential 8\u0026trade; medium supplemented with ROCK inhibitor (Y27632 dihydrochloride, MCE, HY-10583). hiPSCs were maintained at 37 \u0026deg;C with 5% CO\u003csub\u003e2\u003c/sub\u003e.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eTransfection and transduction protocols and genomic DNA extraction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor HEK293T, HEK293, HeLa, and U2OS cells, 0.25 \u0026times; 10\u003csup\u003e5\u003c/sup\u003e cells were seeded per well in 48-well plates (SPL, 30048) containing 250 \u0026micro;L of DMEM supplemented with 10% FBS and 1% antibiotics, one day prior to transfection. The following day, plasmid vectors were mixed and transfected using 0.5 \u0026micro;L of jetOPTIMUS (Polyplus, 101000006) per well, according to the manufacturer\u0026rsquo;s instructions. Typically, 375 ng of snuABE and 125 ng of tagRNA were used per transfection. If required, an additional 41.6 ng of ngRNA was included (541.6 ng in total). When co-expressing additional proteins such as La protein or MLH1dn, 100 ng of plasmid was added, bringing the total to 600 ng. After 36 hours, 200 \u0026micro;L of medium was replaced with fresh medium (DMEM, 10% FBS, 1% antibiotics). Cells were harvested 36 hours later using 0.25% Trypsin-EDTA (Welgene, LS015). \u003c/p\u003e\n\u003cp\u003eFor transfection of human fibroblasts, electroporation was performed using the Neon\u0026reg; Transfection System (Invitrogen\u0026trade;) with the 10 \u0026mu;L Kit (Invitrogen\u0026trade;, MPK1096). Cells detached using TrypLE\u0026trade; Express (Gibco\u0026trade;, 12605010) were resuspended in Neon\u0026trade; R buffer at a density of 1 \u0026times; 10\u003csup\u003e5\u003c/sup\u003e cells per reaction and mixed with 1 \u0026mu;g of total plasmid DNA. Electroporation was carried out at 1600 V with a 20 ms pulse width and a single pulse. After 72 hours, cells were harvested using TrypLE\u0026trade; Express.\u003c/p\u003e\n\u003cp\u003eFor hiPSC transfection, 1 \u0026times; 10\u003csup\u003e5\u003c/sup\u003e cells were seeded into a 24-well plate. Plasmid DNA (same total amount as above) was diluted in 25 \u0026mu;L of Opti-MEM\u0026trade; (Gibco\u0026trade;, 31985-070), then mixed with 25 \u0026mu;L of Opti-MEM\u0026trade; containing 2 \u0026mu;L of Lipofectamine Stem (ThermoFisher, STEM00015). The resulting 50 \u0026mu;L mixture was transfected into the cells. At 72 hours post-transfection, cells were washed with DPBS and detached using ACCUTASE\u0026trade; (Stem Cell Technologies, 07922).\u003c/p\u003e\n\u003cp\u003eHarvested cells were centrifuged at 1000 \u0026times; g for 5 minutes to collect the pellet. Genomic DNA was extracted by adding 100 \u0026micro;L of extraction buffer (40 mM Tris, pH 8.9; 1% Tween-20; 0.2 mM EDTA; Proteinase K [20 mg/mL]; 0.2% Nonidet P-40). Samples were vortexed for 15 seconds and incubated in a thermal cycler at 60 \u0026deg;C for 15 minutes, followed by 98 \u0026deg;C for 5 minutes.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eAmplicon sequencing and data analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGenomic DNA was analyzed using an Illumina MiniSeq sequencer. Targeted deep sequencing was performed using a two-step PCR process. In the first step, 2 \u0026micro;L of extracted DNA was amplified using adaptor primers. In the second step, 1 \u0026micro;L of the initial PCR product was used as the template with index primers. PCR was conducted with KOD multi \u0026amp; Epi (Toyobo, KME-101) following the manufacturer\u0026rsquo;s instructions under the following cycling conditions: 98 \u0026deg;C for 2 minutes; 29 cycles of 98 \u0026deg;C for 10 seconds, 56 \u0026deg;C for 20 seconds, and 68 \u0026deg;C for 30 seconds; and a final extension at 68 \u0026deg;C for 1 minute. The final PCR products were purified using Expin\u0026trade; PCR SV (GeneAll Biotechnology, 103-102) and sequenced using the MiniSeq Mid Output Kit (Illumina, FC-420-1004). Analysis of A-to-I editing efficiency at target sites was performed using the ABE_detector pipeline (https://github.com/imhw4157/ABE_detector), and total base conversion rates were calculated using the allsub_detector pipeline (https://github.com/imhw4157/allsub_detector).\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eRNA sequencing and data analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Neon\u0026reg; Transfection System (Invitrogen\u0026trade;) with the 100 \u0026mu;L Kit (Invitrogen\u0026trade;, MPK10096) was used for mRNA sequencing. HEK293T cells were detached as previously described, resuspended in DPBS, and washed by centrifugation at 500 g for 3 minutes. A total of 1 \u0026times; 10\u003csup\u003e6\u003c/sup\u003e HEK293T cells were resuspended in 100 \u0026micro;L of sucrose potassium phosphate buffered saline (119 mM KCl, 15.1 mM NaCl, 0.144 mM CaCl\u003csub\u003e2\u003c/sub\u003e, 5 mM MgCl\u003csub\u003e2\u003c/sub\u003e, 9.98 mM KH\u003csub\u003e2\u003c/sub\u003ePO\u003csub\u003e4\u003c/sub\u003e, 250 mM Sucrose, pH 7.4), mixed with 2 \u0026micro;g of vector, and subjected to electroporation (1600 V, 20 ms pulse width, 1 pulse). The transfected cells were transferred to 6-well plates (SPL, 30006) containing 2 mL of DMEM supplemented with 10% FBS and incubated at room temperature for 5 minutes in a clean bench, followed by incubation at 37 \u0026deg;C with 5% CO\u003csub\u003e2\u003c/sub\u003e for 24 hours. After incubation, the media was removed entirely, and total RNA was extracted using 1 mL of TRIzol (Invitrogen\u0026trade;, 15596026) and 200 \u0026micro;L of Phenol:Chloroform:Isoamyl Alcohol (25:24:1, saturated with 10 mM Tris, pH 8.0, and 1 mM EDTA; Sigma-Aldrich, P2069-100ML).\u003c/p\u003e\n\u003cp\u003eFor targeted RNA sequencing (\u003cstrong\u003eFig. 3i\u003c/strong\u003e,\u003cstrong\u003ej\u003c/strong\u003e), cDNA was synthesized using ReverTra Ace (Toyobo, FSK-101F) according to the manufacturer\u0026rsquo;s protocol and sequenced using Illumina MiniSeq as previously described. For whole-RNA sequencing (\u003cstrong\u003eFig. 3k\u003c/strong\u003e), total RNA was processed using the NEBNext\u0026reg; rRNA Depletion Kit v2 (Human/Mouse/Rat) and RNA Sample Purification Beads (NEB, E7405L), followed by library preparation using the NEBNext\u0026reg; Ultra\u0026trade; II RNA Library Prep Kit with Sample Purification Beads (NEB, E7775S), according to the manufacturer\u0026rsquo;s instructions. Sequencing was outsourced to Novogene (NovaSeq X Plus), with 6 Gb of sequencing data generated per sample.\u003c/p\u003e\n\u003cp\u003eFor RNA sequencing data analysis, FASTQ files were aligned to the GRCh38 reference genome (primary assembly from Ensembl) using HISAT2\u003csup\u003e47\u003c/sup\u003e (https://github.com/DaehwanKimLab/hisat2). The resulting BAM files were sorted and indexed using SAMtools\u003csup\u003e48\u003c/sup\u003e (https://www.htslib.org). A-to-I RNA editing events were analyzed using REDItools2\u003csup\u003e49\u003c/sup\u003e (https://github.com/BioinfoUNIBA/ REDItools2) following the published protocol. To ensure data reliability, editing sites with a read depth of less than 50 were excluded using the -l (MIN_COLUMN_LENGTH) option in REDItools2.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eGeneration of Lenti knock-in HEK293T cells\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo generate HEK293T cells with an artificial motif knock-in, lentiviral transduction was employed. The lentiviral vector was constructed by digesting pAX198 (Addgene, 173042) with XbaI and XhoI (NEB, R0145S and R0146S), following the same cloning strategy used for tagRNAs, agRNAs, sgRNAs, and ngRNAs. Lenti-X\u0026trade; 293T cells (Takara, 632180) were seeded at 6 \u0026times; 10\u003csup\u003e6\u003c/sup\u003e cells per 100 mm cell culture dish (SPL, 20100), and the medium was replaced with fresh DMEM containing 10% FBS the following day. Immediately after media replacement, a mixture consisting of 9.4 \u0026micro;g of the constructed lentiviral vector, 8.6 \u0026micro;g of psPAX2 (Addgene, 12260), 2.6 \u0026micro;g of MD2.G (Addgene, 12259), and 2.0 \u0026micro;g of pRSV-Rev (Addgene, 12253) was prepared and transfected using 65.7 \u0026micro;L of Transporter 5\u0026reg; Transfection Reagent (Polysciences, 26008-50), according to the manufacturer\u0026rsquo;s instructions. One day after transfection, the media was replaced with fresh DMEM containing 10% FBS. Two days later, the media containing the viral particles was collected, centrifuged at 500 g for 10 minutes to remove cellular debris, and filtered through a 0.45 \u0026micro;m filter (Pall, 4614). The virus was then concentrated using the Lenti-X\u0026trade; Concentrator (Takara, 631232) following the manufacturer\u0026rsquo;s protocol. HEK293T cells were infected with the concentrated lentivirus at a multiplicity of infection (MOI) of 0.3. Two days post-infection, cells were selected with 3 \u0026micro;g/mL puromycin (InvivoGen, ant-pr-1) for three days.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eDirected evolution through EvolvePro\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDirected evolution using the EvolvePro pipeline was conducted as previously described (https://github.com/mat10d/EvolvePro)\u003csup\u003e22\u003c/sup\u003e. Approximately 200 random single-point mutations were generated and tested in HEK293T cells across five independent target sites. The editing efficiencies of these variants were measured and used as training input for the initial round of EvolvePro predictions. For each target, the top eight mutations predicted to exhibit the highest activity were selected, resulting in a total of 40 variants (8 per target \u0026times; 5 targets), which were then experimentally validated across all five sites in subsequent rounds. This iterative process was repeated over multiple rounds, with EvolvePro models updated using the most recent experimental data in a target-specific manner. In the final round, all mutation data were compiled, and the average fold-change in editing efficiency across the five targets was calculated. Mutations showing a fold-change greater than 1 were retained and used to generate all possible combinatorial variants containing up to four mutations. These multi-mutant variants were ranked using EvolvePro, and the top 20 predicted variants were selected for final experimental validation.\u003c/p\u003e\n"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank Dr. Francisco X. Mora (EssayReview) for English-language editing. Most of the sequencing data analysis was conducted using the computing server at the Genomic Medicine Institute Research Service Center. This research was supported by grants from the National Research Foundation of Korea (NRF) (No. 2021M3A9H3015389, No. RS-2024-00451880, No. RS-2024-00455559, and SRC-NRF2022R1A5A102641311) awarded to S.B. Additional support was provided by the Korean Fund for Regenerative Medicine (KFRM) (No. RS-2024-00332601), a grant from the Ministry of Food and Drug Safety (No. 25202MFDS003) in 2025, and the SNUH Lee Kun-hee Child Cancer \u0026amp; Rare Disease Project (No. 25B-001-0700), also awarded to S.B.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eH.W.I. and S.B. conceived the project. H.W.I. and C.-J.J. developed the bioinformatics algorithms. H.W.I., B.-D.J., Y.L., Y.E.O., and Y.-W.K. performed the cell experiments. H.W.I. and H.U. carried out the protein engineering. S.B. supervised the project. H.W.I. and S.B. wrote the manuscript with input from all authors.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdditional\u0026nbsp;information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupplementary Information accompanying this paper is available at http://\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration\u0026nbsp;of\u0026nbsp;interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eH.W.I., B.-D.J., and S.B. have filed a patent application based on this work. The remaining authors declare no competing interests.\u003c/p\u003e\n"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eJinek, M.\u003cem\u003e et al.\u003c/em\u003e A Programmable Dual-RNA\u0026ndash;Guided DNA Endonuclease in Adaptive Bacterial Immunity. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e337\u003c/strong\u003e, 816-821 (2012). https://doi.org/doi:10.1126/science.1225829\u003c/li\u003e\n\u003cli\u003eKosicki, M., Tomberg, K. \u0026amp; Bradley, A. Repair of double-strand breaks induced by CRISPR\u0026ndash;Cas9 leads to large deletions and complex rearrangements. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 765-771 (2018). https://doi.org/10.1038/nbt.4192\u003c/li\u003e\n\u003cli\u003eHwang, G.-H.\u003cem\u003e et al.\u003c/em\u003e Large DNA deletions occur during DNA repair at 20-fold lower frequency for base editors and prime editors than for Cas9 nucleases. \u003cem\u003eNature Biomedical Engineering\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 79-92 (2025). https://doi.org/10.1038/s41551-024-01277-5\u003c/li\u003e\n\u003cli\u003eCullot, G.\u003cem\u003e et al.\u003c/em\u003e CRISPR-Cas9 genome editing induces megabase-scale chromosomal truncations. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 1136 (2019). https://doi.org/10.1038/s41467-019-09006-2\u003c/li\u003e\n\u003cli\u003eHaapaniemi, E., Botla, S., Persson, J., Schmierer, B. \u0026amp; Taipale, J. CRISPR\u0026ndash;Cas9 genome editing induces a p53-mediated DNA damage response. \u003cem\u003eNature Medicine\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 927-930 (2018). https://doi.org/10.1038/s41591-018-0049-z\u003c/li\u003e\n\u003cli\u003eKomor, A. C., Kim, Y. B., Packer, M. S., Zuris, J. A. \u0026amp; Liu, D. R. Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e533\u003c/strong\u003e, 420-424 (2016). https://doi.org/10.1038/nature17946\u003c/li\u003e\n\u003cli\u003eKim, D., Kim, D.-e., Lee, G., Cho, S.-I. \u0026amp; Kim, J.-S. Genome-wide target specificity of CRISPR RNA-guided adenine base editors. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 430-435 (2019). https://doi.org/10.1038/s41587-019-0050-1\u003c/li\u003e\n\u003cli\u003eKleinstiver, B. P.\u003cem\u003e et al.\u003c/em\u003e High-fidelity CRISPR\u0026ndash;Cas9 nucleases with no detectable genome-wide off-target effects. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e529\u003c/strong\u003e, 490-495 (2016). https://doi.org/10.1038/nature16526\u003c/li\u003e\n\u003cli\u003eJang, H.-K.\u003cem\u003e et al.\u003c/em\u003e High-purity production and precise editing of DNA base editing ribonucleoproteins. \u003cem\u003eScience Advances\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, eabg2661 (2021). https://doi.org/doi:10.1126/sciadv.abg2661\u003c/li\u003e\n\u003cli\u003eJin, S.\u003cem\u003e et al.\u003c/em\u003e Cytosine, but not adenine, base editors induce genome-wide off-target mutations in rice. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e364\u003c/strong\u003e, 292-295 (2019). https://doi.org/doi:10.1126/science.aaw7166\u003c/li\u003e\n\u003cli\u003eGr\u0026uuml;newald, J.\u003cem\u003e et al.\u003c/em\u003e Transcriptome-wide off-target RNA editing induced by CRISPR-guided DNA base editors. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e569\u003c/strong\u003e, 433-437 (2019). https://doi.org/10.1038/s41586-019-1161-z\u003c/li\u003e\n\u003cli\u003eRees, H. A., Wilson, C., Doman, J. L. \u0026amp; Liu, D. R. Analysis and minimization of cellular RNA editing by DNA adenine base editors. \u003cem\u003eScience Advances\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, eaax5717 (2019). https://doi.org/doi:10.1126/sciadv.aax5717\u003c/li\u003e\n\u003cli\u003eZhou, C.\u003cem\u003e et al.\u003c/em\u003e Off-target RNA mutation induced by DNA base editing and its elimination by mutagenesis. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e571\u003c/strong\u003e, 275-278 (2019). https://doi.org/10.1038/s41586-019-1314-0\u003c/li\u003e\n\u003cli\u003eJeong, Y. K.\u003cem\u003e et al.\u003c/em\u003e Adenine base editor engineering reduces editing of bystander cytosines. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 1426-1433 (2021). https://doi.org/10.1038/s41587-021-00943-2\u003c/li\u003e\n\u003cli\u003eLee, S.-H.\u003cem\u003e et al.\u003c/em\u003e Bystander editing by adenine base editors impairs vision restoration in a mouse model of Leber congenital amaurosis. \u003cem\u003eMolecular Therapy Methods \u0026amp; Clinical Development\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e (2025). https://doi.org/10.1016/j.omtm.2025.101461\u003c/li\u003e\n\u003cli\u003eNishikura, K. Functions and regulation of RNA editing by ADAR deaminases. \u003cem\u003eAnnu Rev Biochem\u003c/em\u003e \u003cstrong\u003e79\u003c/strong\u003e, 321-349 (2010). https://doi.org/10.1146/annurev-biochem-060208-105251\u003c/li\u003e\n\u003cli\u003eRauch, S.\u003cem\u003e et al.\u003c/em\u003e Programmable RNA-Guided RNA Effector Proteins Built from Human Parts. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e178\u003c/strong\u003e, 122-134.e112 (2019). https://doi.org/https://doi.org/10.1016/j.cell.2019.05.049\u003c/li\u003e\n\u003cli\u003eQu, L.\u003cem\u003e et al.\u003c/em\u003e Programmable RNA editing by recruiting endogenous ADAR using engineered RNAs. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 1059-1069 (2019). https://doi.org/10.1038/s41587-019-0178-z\u003c/li\u003e\n\u003cli\u003eCox, D. B. T.\u003cem\u003e et al.\u003c/em\u003e RNA editing with CRISPR-Cas13. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e358\u003c/strong\u003e, 1019-1027 (2017). https://doi.org/doi:10.1126/science.aaq0180\u003c/li\u003e\n\u003cli\u003eAbudayyeh, O. O.\u003cem\u003e et al.\u003c/em\u003e A cytosine deaminase for programmable single-base RNA editing. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e365\u003c/strong\u003e, 382-386 (2019). https://doi.org/doi:10.1126/science.aax7063\u003c/li\u003e\n\u003cli\u003eMerkle, T.\u003cem\u003e et al.\u003c/em\u003e Precise RNA editing by recruiting endogenous ADARs with antisense oligonucleotides. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 133-138 (2019). https://doi.org/10.1038/s41587-019-0013-6\u003c/li\u003e\n\u003cli\u003eJiang, K.\u003cem\u003e et al.\u003c/em\u003e Rapid in silico directed evolution by a protein language model with EVOLVEpro. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e387\u003c/strong\u003e, eadr6006 (2025). https://doi.org/doi:10.1126/science.adr6006\u003c/li\u003e\n\u003cli\u003eZheng, Y., Lorenzo, C. \u0026amp; Beal, P. A. DNA editing in DNA/RNA hybrids by adenosine deaminases that act on RNA. \u003cem\u003eNucleic Acids Research\u003c/em\u003e \u003cstrong\u003e45\u003c/strong\u003e, 3369-3377 (2017). https://doi.org/10.1093/nar/gkx050\u003c/li\u003e\n\u003cli\u003eKuttan, A. \u0026amp; Bass, B. L. Mechanistic insights into editing-site specificity of ADARs. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e \u003cstrong\u003e109\u003c/strong\u003e, E3295-E3304 (2012). https://doi.org/doi:10.1073/pnas.1212548109\u003c/li\u003e\n\u003cli\u003eChen, P. J.\u003cem\u003e et al.\u003c/em\u003e Enhanced prime editing systems by manipulating cellular determinants of editing outcomes. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e184\u003c/strong\u003e, 5635-5652.e5629 (2021). https://doi.org/https://doi.org/10.1016/j.cell.2021.09.018\u003c/li\u003e\n\u003cli\u003eYan, J.\u003cem\u003e et al.\u003c/em\u003e Improving prime editing with an endogenous small RNA-binding protein. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e628\u003c/strong\u003e, 639-647 (2024). https://doi.org/10.1038/s41586-024-07259-6\u003c/li\u003e\n\u003cli\u003eRichter, M. F.\u003cem\u003e et al.\u003c/em\u003e Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 883-891 (2020). https://doi.org/10.1038/s41587-020-0453-z\u003c/li\u003e\n\u003cli\u003eChen, L.\u003cem\u003e et al.\u003c/em\u003e Engineering a precise adenine base editor with minimal bystander editing. \u003cem\u003eNature Chemical Biology\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 101-110 (2023). https://doi.org/10.1038/s41589-022-01163-8\u003c/li\u003e\n\u003cli\u003eKoblan, L. W.\u003cem\u003e et al.\u003c/em\u003e Improving cytidine and adenine base editors by expression optimization and ancestral reconstruction. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 843-846 (2018). https://doi.org/10.1038/nbt.4172\u003c/li\u003e\n\u003cli\u003ePerrotta, R. M.\u003cem\u003e et al.\u003c/em\u003e Machine Learning and Directed Evolution of Base Editing Enzymes. \u003cem\u003ebioRxiv\u003c/em\u003e, 2024.2005.2017.594556 (2024). https://doi.org/10.1101/2024.05.17.594556\u003c/li\u003e\n\u003cli\u003eEggington, J. M., Greene, T. \u0026amp; Bass, B. L. Predicting sites of ADAR editing in double-stranded RNA. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 319 (2011). https://doi.org/10.1038/ncomms1324\u003c/li\u003e\n\u003cli\u003eMusunuru, K.\u003cem\u003e et al.\u003c/em\u003e In vivo CRISPR base editing of PCSK9 durably lowers cholesterol in primates. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e593\u003c/strong\u003e, 429-434 (2021). https://doi.org/10.1038/s41586-021-03534-y\u003c/li\u003e\n\u003cli\u003eDoman, J. L., Raguram, A., Newby, G. A. \u0026amp; Liu, D. R. Evaluation and minimization of Cas9-independent off-target DNA editing by cytosine base editors. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 620-628 (2020). https://doi.org/10.1038/s41587-020-0414-6\u003c/li\u003e\n\u003cli\u003eZhang, Y., Qian, H., Xu, J. \u0026amp; Gao, W. ADAR, the carcinogenesis mechanisms of ADAR and related clinical applications. \u003cem\u003eAnn Transl Med\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 686 (2019). https://doi.org/10.21037/atm.2019.11.06\u003c/li\u003e\n\u003cli\u003eAbramson, J.\u003cem\u003e et al.\u003c/em\u003e Accurate structure prediction of biomolecular interactions with AlphaFold 3. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e630\u003c/strong\u003e, 493-500 (2024). https://doi.org/10.1038/s41586-024-07487-w\u003c/li\u003e\n\u003cli\u003eAnzalone, A. V.\u003cem\u003e et al.\u003c/em\u003e Search-and-replace genome editing without double-strand breaks or donor DNA. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e576\u003c/strong\u003e, 149-157 (2019). https://doi.org/10.1038/s41586-019-1711-4\u003c/li\u003e\n\u003cli\u003ePark, J.-C., Uhm, H., Kim, Y.-W., Oh, Y. E. \u0026amp; Bae, S. AI-generated small binder improves prime editing. \u003cem\u003ebioRxiv\u003c/em\u003e, 2024.2009.2011.612443 (2024). https://doi.org/10.1101/2024.09.11.612443\u003c/li\u003e\n\u003cli\u003ePanigrahi, G. B., Slean, M. M., Simard, J. P. \u0026amp; Pearson, C. E. Human mismatch repair protein hMutL\u0026alpha; is required to repair short slipped-DNAs of trinucleotide repeats. \u003cem\u003eJ Biol Chem\u003c/em\u003e \u003cstrong\u003e287\u003c/strong\u003e, 41844-41850 (2012). https://doi.org/10.1074/jbc.M112.420398\u003c/li\u003e\n\u003cli\u003eZhang, J.\u003cem\u003e et al.\u003c/em\u003e Germline mutations in the von Hippel-Lindau disease (VHL) gene in mainland Chinese families. \u003cem\u003eJ Cancer Res Clin Oncol\u003c/em\u003e \u003cstrong\u003e134\u003c/strong\u003e, 1211-1218 (2008). https://doi.org/10.1007/s00432-008-0399-x\u003c/li\u003e\n\u003cli\u003eDing, Q.\u003cem\u003e et al.\u003c/em\u003e Permanent Alteration of PCSK9 With In Vivo CRISPR-Cas9 Genome Editing. \u003cem\u003eCirculation Research\u003c/em\u003e \u003cstrong\u003e115\u003c/strong\u003e, 488-492 (2014). https://doi.org/doi:10.1161/CIRCRESAHA.115.304351\u003c/li\u003e\n\u003cli\u003eQin, W.\u003cem\u003e et al.\u003c/em\u003e ABE-ultramax for high-efficiency biallelic adenine base editing in zebrafish. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 5613 (2024). https://doi.org/10.1038/s41467-024-49943-1\u003c/li\u003e\n\u003cli\u003eZhao, D.\u003cem\u003e et al.\u003c/em\u003e Imperfect guide-RNA (igRNA) enables CRISPR single-base editing with ABE and CBE. \u003cem\u003eNucleic Acids Research\u003c/em\u003e \u003cstrong\u003e50\u003c/strong\u003e, 4161-4170 (2022). https://doi.org/10.1093/nar/gkac201\u003c/li\u003e\n\u003cli\u003eShang, M.\u003cem\u003e et al.\u003c/em\u003e A motif preferred adenine base editor with minimal bystander and off-targets editing. \u003cem\u003ebioRxiv\u003c/em\u003e, 2024.2011.2023.624961 (2024). https://doi.org/10.1101/2024.11.23.624961\u003c/li\u003e\n\u003cli\u003eWang, Z.\u003cem\u003e et al.\u003c/em\u003e Decreasing predictable DNA off-target effects and narrowing editing windows of adenine base editors by fusing human Rad18 protein variant. \u003cem\u003eInternational Journal of Biological Macromolecules\u003c/em\u003e \u003cstrong\u003e253\u003c/strong\u003e, 127418 (2023). https://doi.org/https://doi.org/10.1016/j.ijbiomac.2023.127418\u003c/li\u003e\n\u003cli\u003eMusunuru, K.\u003cem\u003e et al.\u003c/em\u003e Patient-Specific In Vivo Gene Editing to Treat a Rare Genetic Disease. \u003cem\u003eNew England Journal of Medicine\u003c/em\u003e \u003cstrong\u003e392\u003c/strong\u003e, 2235-2243 (2025). https://doi.org/doi:10.1056/NEJMoa2504747\u003c/li\u003e\n\u003cli\u003eDoman, J. L.\u003cem\u003e et al.\u003c/em\u003e Phage-assisted evolution and protein engineering yield compact, efficient prime editors. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e186\u003c/strong\u003e, 3983-4002.e3926 (2023). https://doi.org/https://doi.org/10.1016/j.cell.2023.07.039\u003c/li\u003e\n\u003cli\u003eKim, D., Paggi, J. M., Park, C., Bennett, C. \u0026amp; Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. \u003cem\u003eNature Biotechnology\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 907-915 (2019). https://doi.org/10.1038/s41587-019-0201-4\u003c/li\u003e\n\u003cli\u003eDanecek, P.\u003cem\u003e et al.\u003c/em\u003e Twelve years of SAMtools and BCFtools. \u003cem\u003eGigaScience\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e (2021). https://doi.org/10.1093/gigascience/giab008\u003c/li\u003e\n\u003cli\u003eFlati, T.\u003cem\u003e et al.\u003c/em\u003e HPC-REDItools: a novel HPC-aware tool for improved large scale RNA-editing analysis. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 353 (2020). https://doi.org/10.1186/s12859-020-03562-x\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7146806/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7146806/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"The adenine base editor (ABE), which enables A•T-to-G•C base conversion, has emerged as a powerful tool for therapeutic applications. However, conventional ABEs suffer from bystander nucleotide conversions, limiting their utility for precise editing. Here, we present a single-nucleotide resolution ABE (snuABE) created by fusing a nickase Cas9, nCas9(H840A), with the deaminase domain of ADAR, which acts on DNA:RNA hybrids, instead of TadA, which acts on single-stranded DNA in conventional ABEs. snuABE requires a specially designed target-adenine guide RNA (tagRNA) that introduces a mismatch at the target adenine, enabling highly specific A-to-G editing by ADAR. Engineering ADAR from Pediculus humanus using the in silico protein evolution algorithm EvolvePro, along with 3’-end protection of the tagRNA, further enhances the editing activity of snuABE in human cells. Moreover, snuABE exhibits significantly reduced DNA off-target activity, highlighting its potential as a safer and more precise base editing technology for therapeutic applications.","manuscriptTitle":"Engineered ADARs enable single-nucleotide resolution DNA A-to-G editing without bystander effects","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-12 12:34:07","doi":"10.21203/rs.3.rs-7146806/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-biotechnology","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"nbt","sideBox":"Learn more about [Nature Biotechnology](http://www.nature.com/nbt/)","snPcode":"","submissionUrl":"","title":"Nature Biotechnology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"c435fdcb-70f8-44ee-923f-262d7c8c4088","owner":[],"postedDate":"August 12th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":52398968,"name":"Biological sciences/Biological techniques/Genetic techniques/Gene targeting/CRISPR-Cas9 genome editing"},{"id":52398969,"name":"Biological sciences/Biological techniques/Genetic engineering"}],"tags":[],"updatedAt":"2025-08-12T12:34:07+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-12 12:34:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7146806","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7146806","identity":"rs-7146806","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}