Inhibition of base editors with anti-deaminases derived from viruses

Virus-derived anti-deaminases efficiently inhibit APOBEC3-CBEs

The arms race between viruses and APOBEC3 deaminases has generated many natural anti-deaminases. We first selected seven Ades, including EBV-BORF2^17,18, KSHV-ORF61^17,19, HIV-1-Vif²⁰, SIVmac239-Vif^21,22, HSV-1-ICP6¹⁹, EV71-2C²³ and HBV-HBx²⁴ (referred to as Ade1-Ade7), which originate from different viruses and have been reported to antagonize APOBEC3 deaminases in vitro (Table 1). Three APOBEC3-CBEs (A3A, A3B and A3G) were constructed with deaminase-nCas9-2xUGI architecture (BE4max, a state-of-the-art CBE)²⁵ (Fig. 1b). To test Ade inhibition of APOBEC3-CBEs, we transfected CBE and single guide RNAs (sgRNAs) (targeting EMX1-1 and FANCF) into HEK293T cells in the presence or absence of Ade proteins. As a result, four of seven Ades (Ade1-Ade4) showed obvious inhibition of C-to-T activities, while others showed no effect at either target site (Fig. 1c, d). Ade2 was the only effective inhibitor of A3A-CBE, showing slight inhibition, from 1.5- to 2.1-fold (Fig. 1c, d). Notably, Ade1 and Ade2 efficiently inhibited A3B-CBE, from 5.8- to 6.6-fold and from 2.7- to 2.9-fold, respectively (Fig. 1c, d). Both Ade3 and Ade4 exhibited strong inhibition of A3G-CBE, from 3.6- to 4.5-fold and from 3.6- to 4.9-fold, respectively (Fig. 1c, d). Moreover, we observed obvious dose-dependent inhibitory effects of Ade1 and Ade2 (Fig. 1e, f). However, Ade3 and Ade4 exhibited efficient inhibition at low doses, and an increase in doses did not further enhance the inhibition rate (Fig. 1g), which suggests that these Ades may function through two different inhibitory mechanisms. Additionally, we evaluated their inhibitory effect on three A3 deaminases from other species, including RmA3Bctd (rhesus monkey), mA3CDA1²⁶ (mouse) and SsA3Bctd²⁷ (Sus scrofa). These Ades did not show an inhibitory effect on the tested deaminases, except for Ade1, which slightly inhibited RmA3Bctd-CBE, suggesting that these Ades are species-specific and mainly evolved to inhibit human A3 deaminases¹⁵ (Supplementary Fig. 1). These results suggested that the Ade1-Ade4 proteins can efficiently inhibit CBEs with differing APOBEC3 selectivity, holding the potential to be off-switches for CBE applications.

Inhibitory mechanisms of Ades

The A3A protein contains a single catalytically active cytidine deaminase domain. However, some A3 family members, such as A3B and A3G, have a C-terminal catalytic domain (CTD) and an N-terminal pseudocatalytic domain (NTD) that retain the same tertiary folds but are not catalytically effective^15,28. To evaluate whether Ades directly inhibit DNA deamination activity, we first constructed both A3Bctd-CBE and A3Gctd-CBE, which only contain the functional CTD (Fig. 2a). As a result, Ade1 and Ade2 exhibited the same inhibitory effect on A3B-CBE and A3Bctd-CBE, indicating that they may directly interact with A3Bctd and inhibit DNA deamination activity (Fig. 2b). In contrast, Ade3 and Ade4 completely abolished the inhibition of A3Gctd-CBE but not A3G-CBE, suggesting that they did not directly inhibit the deamination activity (Fig. 2b). Subsequently, western blot results indicated a remarkable decrease in A3G-CBE protein levels in the presence of Ade3 and Ade4, indicating that Ade3 and Ade4 inhibit A3G-CBE by activating its degradation (Fig. 2c, d). No obvious changes were observed in other A3-CBEs when the corresponding Ade was added (Supplementary Fig. 2). These results were consistent with previous reports showing that Ade1 interacted preferentially with A3Bctd in vitro and that Ade3 hijacked cellular proteasomal degradation pathways to degrade the whole A3G deaminase^17,20,29.

a Schematic representation of the A3B, A3Bctd, A3G and A3Gctd-CBE architectures. nCas9, D10A. b Base editing of A3B, A3Bctd, A3G and A3Gctd-CBE in the presence or absence of Ade1-Ade4 at the EMX1-1 site. Plasmids expressing CBE, sgRNA, and each Ade (1:1:1) were cotransfected into HEK293T cells. c Immunoblots of A3Bctd- and A3G-CBE in the presence or absence of the corresponding Ades. Tubulin was used as a loading control. d Quantification of the relative CBE protein contents normalized to tubulin in c. e, f Fold inhibition of base editing by Ade1 (e) and Ade2 (f) at the FANCF site. g, h Representative images of HEK293T cells expressing the A3Bctd-CBE-mCherry (h) and A3A-CBE-mCherry (g) constructs alone or in combination with Ade-EGFP constructs. The immunofluorescence microscopy experiment was repeated three times independently with similar results. Values and error bars reflect the mean ± s.e.m. and n = 3 biologically independent experiments. All p values were calculated by two-sided t tests. **p < 0.01, ****p < 0.0001. Source data are provided as a Source Data file.

Ade1 inhibited only A3B-CBE, while Ade2 inhibited both A3A-CBE and A3B-CBE. To understand the different inhibition mechanisms of Ade1 and Ade2, we aligned the amino acid sequences of A3A, A3Bctd and A3Gctd (Supplementary Fig. 3a). The protein sequence of A3A is highly homologous with that of A3Bctd, but their key DNA binding loop 1 (L1) is quite different (Supplementary Fig. 3a). In addition, it has been reported that Ade1 interacts specifically with loop7 (L7) residues of A3Bctd in vitro¹⁷. The structural data also demonstrated that L1 together with L7 control substrate access to the open catalytic pocket of deaminase³⁰. Therefore, we speculated that L1 and L7 might be the key regions controlling Ade inhibition of A3-CBEs. To validate our hypothesis, we constructed a series of A3-CBE chimeras by exchanging L1 and L7 in A3A, A3Bctd and A3Gctd (Supplementary Fig. 3b). As a result, Ade1 showed higher fold inhibition to A3A-L1B than to A3A (3.26-fold vs. 1.06-fold) and remarkable inhibition to A3Gctd-L1B compared with that to A3Gctd (5.67-fold vs. 0.96-fold). Its fold inhibition to A3Bctd-L1A was significantly decreased compared with that to A3Bctd (1.47-fold vs. 5.20-fold), suggesting that the L1 region of A3Bctd is indispensable for the inhibitory action of Ade1 (Fig. 2e and Supplementary Fig. 3c). In addition, Ade1 showed no inhibition of A3Bctd-L7G but significantly suppressed A3Bctd (1.06-fold vs. 5.20-fold). It also slightly improved the fold inhibition of A3Gctd-L7B compared with that of A3Gctd (1.32-fold vs. 0.96-fold), suggesting that the L7 region is also a prerequisite for the inhibitory action of Ade1 (Fig. 2e and Supplementary Fig. 3c). Furthermore, A3Gctd-L1L7B was also significantly inhibited compared with A3Gctd (5.66-fold vs. 0.96-fold) (Fig. 2e and Supplementary Fig. 3c). In contrast, Ade2 showed obvious inhibition of all A3A- and A3Bctd-chimeras but exhibited no impact on any of the A3Gctd-chimeras, indicating that Ade2 may rely on regions other than L1 and L7 to co-exert inhibitory effects (Fig. 2f and Supplementary Fig. 3c). These A3 chimeras showed different editing efficiencies, editing windows and sequence context specificities, consistent with previous reports that L1 and L7 influence deaminase activities and intrinsic sequence preferences^30,31 (Fig. Supplementary Fig. 3d).

Viral ribonucleotide reductases (Ade1 and Ade2) were reported to inhibit APOBEC3 by relocating it from the nucleus to the cytosol¹⁹. To characterize how they affect base editors, we used an immunofluorescence microscopy approach to locate Ades and A3-CBEs in human cells after transfection. A3Bctd- or A3A-CBE alone showed obvious nuclear localization; however, they were also partially relocated to the perinuclear region by both Ade1 and Ade2 (Fig. 2g, h and Supplementary Fig. 4). Our results were different from those of a previous report showing that A3 deaminases were relocated from nuclear to cytoplasmic bodies¹⁹, which may have been a result of the presence of a nuclear localization signal (NLS) in our CBE constructs. Both Ade1 and Ade2 relocated A3-CBEs from the nucleus to the perinuclear region, while they showed different inhibitory effects on A3-CBEs, suggesting that they might adopt other mechanism of inhibition.

Therefore, we tested that if Ade1 and Ade2 proteins directly inhibit deaminase activity of A3 deaminases by an in vitro deaminase activity assay. As a result, Ade1 potently inhibited A3Bctd and Ade2 moderately inhibited both A3A and A3Bctd, consistent with the results in human cells (Supplementary Fig. 5). Interestingly, the Ade1 slightly inhibited A3A in vitro, while no evident inhibition of A3A-CBE was observed in human cells (Supplementary Fig. 5). We suspected that there is a certain threshold of inhibition intensity to achieve effective inhibition in vivo, which also explains why some inhibitors (Ade5-Ade7) that are effective in vitro while defunct in vivo.

Taken together, these results demonstrated that Ade1 and Ade2 affect base editors by inhibiting deaminase activity and relocation. Specifically, Ade1 interacts with the L1 and L7 regions, and both Ade1 and Ade2 relocate A3-CBEs from the nucleus to the perinuclear region. Ade3 and Ade4 inhibit A3G-CBE by degrading the whole A3G deaminase.

Expanding the application of Ades to CGBE, A&CBE and A&CGBE

It is well known that rat APOBEC1 (rA1)-CBE is the most widely used CBE system, followed by the A3A-, eAID-, eCDA1-, A3G- and other CBE systems^32,33,34. However, non-A3-CBEs cannot be inhibited by the Ades used in this study (Supplementary Fig. 6). We intended to replace rA1-CBE with A3B- and A3Bctd-CBE since they have excellent C-to-T editing efficiency and can be greatly inhibited by Ade1. Therefore, we compared the base editing characteristics of the rA1-, A3B- and A3Bctd-CBE systems at seven target sites (Supplementary Fig. 7a–c). Notably, the average efficiency of A3Bctd-CBE (56.7 ± 2.1%) was comparable to that of classical rA1-CBE (56.4 ± 1.4%), and A3B-CBE (51.8 ± 2.6%) was slightly less efficient (Supplementary Fig. 7d). In addition, A3Bctd-CBE and A3B-CBE showed a similar but slightly broader editing window than rA1-CBE (Supplementary Fig. 7e). Importantly, Ade1 and Ade2 exhibited strong and moderate inhibition of A3Bctd-CBE at all seven tested sites, respectively (Supplementary Fig. 7f). These data demonstrated that A3Bctd-CBE can replace conventional rA1-CBE for efficient base editing.

Other BE systems, except for CBE, including CGBE, which induces C-to-G transversion, and A&CBE, which simultaneously induces C-to-T and A-to-G conversions, were developed based on the rA1 deaminase^{35,36,37,38,39,40}. Here, we constructed nine new A3A variant- and A3Bctd variant-CGBEs by introducing a series of mutations into the A3A-nCas9 and A3Bctd-nCas9 architectures^33,35,41,42 (Fig. 3a). As a result, the C-to-G efficiencies of the newly designed A3A-N57G, A3A-Y130F, A3Bctd-R211A and A3Bctd-R211K systems were comparable to those of the classical rA1-R33A system when tested at four sites (Supplementary Fig. 8). Each CGBE performed differently in specific sites, as reported in a previous study⁴³. Moreover, the four best-performing A3A and A3Bctd variant-CGBEs were significantly inhibited by Ade1 and Ade2 (Fig. 3b, c). We observed that engineering deaminase is an efficient method to improve C-to-G editing efficiency, consistent with a previous report of rA1-R33A³⁵. Although the mechanism is not clear, we reasoned that narrowing the deamination window of the deaminase might increase the C-to-G editing outcome.

a Schematic representation of engineered A3A variant- and A3Bctd variant-CGBEs. nCas9, D10A. b Bar plots showing the base editing frequencies induced by rA1 variants, A3A variants-, and A3Bctd variants-CGBEs at the RNF2 site. Target C, red. c C-to-G editing frequencies of four representative CGBEs in the presence or absence of Ade at the RNF2 site. Plasmids expressing CGBE, sgRNA, and Ade (1:1:1) were cotransfected into HEK293T cells. d Schematic representation of engineered A3Bctd-A&CBE. nCas9, D10A. e, f Frequencies of single C-to-T and A-to-G conversions using A3Bctd-A&CBE in the presence or absence of Ade1 at RNF2 (e) and ABE site7 (f). Target C, red triangle. g Schematic representation of engineered A3Bctd-A&CGBE. nCas9, D10A. h, i Frequencies of single C-to-G and A-to-G conversions using A3Bctd-A&CGBE in the presence or absence of Ade1 at RNF2 (h) and ABE site7 (i). Target C, red triangle. Values and error bars reflect the mean ± s.e.m. and n = 3 biologically independent experiments. All p values were calculated by two-sided t tests. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Source data are provided as a Source Data file.

Subsequently, we constructed an A3Bctd-A&CBE using the A3Bctd-8e-nCas9-2xUGI architecture, in which 8e is a state-of-the-art ABE⁴⁴ (Fig. 3d). Both C-to-T and A-to-G editing was efficient at two target sites (Fig. 3e, f). Interestingly, Ade1 inhibited both cytosine and adenine base conversions produced by A&CBE, but C-to-T editing was inhibited to a larger extent (Fig. 3e, f). Furthermore, we designed a novel A&CGBE using the A3Bctd-R211A-8e-nCas9 architecture (Fig. 3g), which induced both C-to-G and A-to-G editing efficiently at the RNF2 site but not at ABE site7 (Fig. 3h, i). Notably, Ade1 significantly suppressed C-to-G editing and only slightly attenuated A-to-G editing at the RNF2 site (Fig. 3h). Taken together, these results suggested that various BE systems comprising A3A and A3Bctd enable various types of base editing and can be regulated by Ade-off switches.

Expanded inhibitory ranges of Ades to rA1-CBE and ABE8e

The A3A- and A3Bctd-BEs significantly expanded the inhibitory range, while the classical rA1-CBE and ABE system still could not be inhibited by current Ades (Supplementary Fig. 6). In addition, the method of exchanging L1 and L7 in A3 deaminases to achieve inhibition is not practical for rA1 or ABE due to their significant sequence differences. However, we noticed that Ade3 and Ade4 use a different inhibitory mechanism, degrading the A3G deaminase. It has been reported that the N-terminal domain of A3G (A3Gntd) binds to Ade3, and the smallest fragment of A3G containing only amino acids 105–245 can trigger Ade3-induced degradation^45,46. On the basis of these considerations, we hypothesized that fusing A3Gntd to other BE systems can extend the Ade3 inhibitory ranges. Therefore, we first constructed a series of A3Gntd-rA1-CBEs by connecting A3Gntd segments of various lengths (1–196, 1–245, 105–196 and 105–245) to the N-terminus of rA1-CBE (Fig. 4a). As a result, Ade3 and Ade4 showed no obvious inhibition of all four A3Gntd-rA1-CBEs, while full-length A3G-CBE was significantly inhibited at the two tested sites (Fig. 4b, c). We speculated that the divergence may lie in the difference between the in vitro and in vivo experimental environments and the influence of the architecture of cytidine deaminase-nCas9 fusion. Subsequently, we inactivated full-length A3G (dead A3G, dA3G) by introducing an E259A mutation to abolish its catalytic activity and constructed dA3G-rA1-CBE⁴⁷ (Fig. 4a). Notably, the C-to-T editing frequencies of dA3G-rA1-CBE were moderately decreased in the presence of Ade3 and Ade4, compared with those of the control group (Fig. 4b, c). Encouraged by this result, we fused dA3G to the N-terminus of ABE8e (Fig. 4d). Notably, dA3G-ABE8e was significantly inhibited by both Ade3 and Ade4, while ABE8e was not inhibited at the two target sites (Fig. 4d). These results indicated that the inhibitory ranges of Ades were further expanded to rA1-CBE, particularly ABE8e, with an engineered dA3G domain.

a Schematic representation of rA1-, A3G-, A3Gntd-rA1- and dA3G-rA1-CBEs. nCas9, D10A. b, c C-to-T editing of seven CBEs in the presence or absence of Ade3 and Ade4 at the EMX1-1 site (b) and FNACF site (c). Plasmids expressing CBE, sgRNA, and each Ade (1:1:1) were cotransfected into HEK293T cells. d Top: Schematic representation of dA3G-ABE8e. Bottom: A-to-G editing of ABE8e and dA3G-ABE8e in the presence or absence of Ade3 and Ade4 at the FANCF and RNF2 sites. Values and error bars reflect the mean ± s.e.m. and n = 3 biologically independent experiments. All p values were calculated by two-sided t tests. *p < 0.05, **p < 0.01, ***p < 0.001. Source data are provided as a Source Data file.

Inhibition of Cas9-dependent and Cas9-independent off-target editing by Ades

The recognition of cognate protospacer sequences of CBEs can cause severe Cas9-dependent OT effects^3,5. In addition, it has been reported that CBEs can potentially create stochastic Cas9-independent OT edits on the genome and transcriptome^6,7,8,9. The potential safety risk caused by DNA off-target edits is higher than that caused by RNA off-target edits, so we focused on the DNA OT editing of CBEs. First, a classical EMX1-1 target site was selected to evaluate on-target editing and Cas9-dependent OT editing (Fig. 5a, b). A recently reported sensitive and cost-effective orthogonal R-loop assay was used to assess the Cas9-independent OT effect on the SaCas9-induced R-loop region, including the Sa site5, site6, site14 and site17, which had been detected in previous reports^4,27,48 (Fig. 5c). Conventional AcrIIC1 (no inhibition of SpCas9, as a negative control), AcrIIA5 (strong inhibition of SpCas9) and Ade1 and Ade2 were used for parallel comparisons⁴⁹. As expected, AcrIIA5 efficiently inhibited the on-target editing of both A3A-CBE (3.0-fold) and A3Bctd-CBE (4.8-fold) (Fig. 5a). Ade1 inhibited A3Bctd-CBE (5.0-fold), and Ade2 inhibited both A3A-CBE (1.3-fold) and A3Bctd-CBE (2.3-fold) (Fig. 5a). However, AcrIIA5 significantly reduced Cas9-dependent OT editing but maintained a high level of Cas9-independent OT editing, consistent with the inhibition mechanism by which AcrIIA5 interacts with the Cas9 domain (Fig. 5b, c). Notably, Ade1 almost completely suppressed both Cas9-dependent and Cas9-independent OT editing of A3Bctd-CBE, and Ade2 significantly reduced the OT editing of both A3A- and A3Bctd-CBE (Fig. 5b, c). Furthermore, both Ade1 and Ade2 exhibited stronger inhibition of Cas9-independent OT activities than AcrIIA5 (Fig. 5c). Similar results were observed at the HEK site4 (Supplementary Fig. 9). Ade1 also showed a slight decrease in OT editing of A3A-CBE without compromising on-target editing (Fig. 5b, c). This might be explained by our observations showing that Ade1 slightly inhibited A3A deaminase activity and relocates A3A-CBE from the nucleus to perinuclear bodies. A3Gctd-CBE is of high specificity naturally in both Cas9-dependent and Cas9-independent OT sites, as reported in previous reports^31,50,51 (Supplementary Fig. 10). Overall, these results suggested that Ades can suppress both Cas9-dependent and Cas9-independent OT editing, holding the potential of being safer off switches of CBEs than current Acrs.

The principle of the assay is illustrated on the left, and the results are shown as bar graphs on the right. a On-target editing frequencies of A3A and A3Bctd-CBE in the presence or absence of AcrIIC1 (as a negative control), AcrIIA5, Ade1 and Ade2 at the EMX1-1 site. b Cas9-dependent OT editing frequencies of A3A and A3Bctd-CBE in the presence or absence of AcrIIC1 (the negative control), AcrIIA5, Ade1 and Ade2 at the EMX1-1 OT1 and OT2 sites. c Cas9-independent OT editing frequencies of A3A and A3Bctd-CBE in the presence or absence of AcrIIC1 (the negative control), AcrIIA5, Ade1 and Ade2 at Sa site5, site6, site14 and site17. The dSaCas9 together with a gRNA was used to induce a stable ssDNA region (orthogonal R loop) at a specific locus, thus artificially magnifying Cas9-independent deamination. Values and error bars reflect the mean ± s.e.m. and n = 3 biologically independent experiments. All p values were calculated by two-sided t tests. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Source data are provided as a Source Data file.

Cell-specific CBE editing by microRNA-responsive Ade switches

To render CBE-based gene modifications and therapies precise and safe, strategies that confine the activity of a CBE to specific cells and tissues are highly desired. In recent studies, Acr proteins have been used to achieve cell-specific CRISPR-Cas9 editing, activation and even tissue-restricted genome editing in vivo by microRNA-dependent expression of Acr proteins^52,53,54. Given that Ades can inhibit both Cas9-dependent and Cas9-independent OT activities, they make safer switches for CBEs than traditional Acrs. Therefore, we tried to develop a cell type-specific Cas-ON switch based on miRNA-regulated expression of Ade proteins. We inserted binding sites of miR-122 (miR-122BS), which are abundant specifically in liver cells, into the 3′UTR of Ade transgenes (Fig. 6a). Coexpressing these with A3Bctd-CBE and sgRNAs resulted in Ade knockdown and released CBE activity exclusively in hepatocytes (such as in Huh7 cells), while CBE was inhibited in miR-122-free cells (such as HEK293T cells) (Fig. 6a). In HEK293T cells, Ade1 robustly inhibited C-to-T editing by A3Bctd-CBE at two tested sites (Fig. 6b, c). The presence or absence of miR-122BS showed no significant effect on editing inhibition in the miR-122-free cells (Fig. 6b, c). The editing frequency was decreased in Huh7 cells compared with that in HEK293T cells due to the lower transfection efficiencies in Huh7 cells, as previously reported^52,54. Ade1 also prevented editing in Huh7 cells when expressed by constructs that lacked miR-122BS. In contrast, Ade1 plasmids incorporating miR-122BS in the 3′UTR failed to inhibit CBE editing in Huh7 cells, as indicated by editing efficiencies that were similar to the no-Ade control (Fig. 6b, c). These results illustrated that the developed CBE-ON switch can confine CBE activity to the selected cell types, facilitating safe and precise CBE-based gene modifications and therapies.

a Schematic of the CBE-ON switch. The miR-122 binding sites were inserted into the 3′UTR of an Ade-encoding transgene. Upon delivery, Ade expression is selectively knocked down within the target cells, permitting CBE activity. In OFF-target cells lacking the miR-122 signature, the Ade protein is translated and inhibits A3Bctd-CBE. b, c HEK293T and Huh7 cells were transfected with plasmids encoding A3Bctd-CBE, sgRNA and Ade1 with or without 3xmiR-122BS at the EMX1-1 site (b) and FANCF site (c). Values and error bars reflect the mean ± s.e.m. and n = 3 biologically independent experiments. All p values were calculated by two-sided t tests. **p < 0.01, ***p < 0.001. Source data are provided as a Source Data file.