The commonly used base editors BE3 (cytosine base editor (CBE)) and ABE7.10 (adenine base editor (ABE)) typically generate C·G-to-T·A or A·T-to-G·C conversions, respectively, within a ~5-nucleotide editing window on the target DNA1,2. Using highly active deaminases and nuclease-impaired Cas9 or Cpf1, base editors catalyze nucleotide conversions very potently and rarely generate double-stranded DNA breaks3. However, CBEs or ABEs only catalyze conversion of a single type of nucleotide, either C·G-to-T·A or A·T-to-G·C1–3, which limits product diversity. We assumed that a tool that converts two types of nucleotides on the same allele would drastically modify the target sequence, broadening its base editing capability.We hypothesized that fusing cytidine and adenosine deaminases to Cas9n (SpCas9 D10A mutant) could achieve simultaneous C·G and A·T base editing on the same allele, creating a new base editor, A&C-BE (Fig. 1a). To test this, five constructs were generated with combinations of the two deaminases and Cas9n (Supplementary Fig. 1 and Supplementary Sequence 1) and tested with a reporter system that is activated after simultaneous C/A editing to restore luciferase expression (Supplementary Fig. 2a and Supplementary Sequence 2). A&C-BEs with cytidine deaminase fused to the N terminus, but not the other constructs, significantly stimulated reporter activity (Supplementary Fig. 2b). Further analyzing the editing efficiency at the endogenous site (FANCF-sg3), we found that TadA functioned only when it was fused to the N terminus adjacent to Cas9n, and cytidine deaminases fused to the N terminus of ABE7.10 induced significant levels of simultaneous A/C mutation in the same allele (Fig. 1b,c and Supplementary Fig. 3). Moreover, cytidine deaminase activity was strongly increased when it was fused to the N terminus of ABE7.10, and hAID exhibited higher activity compared to rAPOBEC1 (Fig. 1b,c and Supplementary Fig. 3).On the basis of these results, we further optimized the ABE7.10-N-AID construct to improve its efficiency. Consistent with previous reports4, we optimized codons and added a bipartite nuclear localization signal (NLS) to generate ABE7.10-N-AIDmax, leading to significantly increased CBE activity and slightly elevated ABE activity, as well as simultaneous A/C conversions (Fig. 1b,cand Supplementary Figs. 3 and 4). Through screening five linker sequences, a rigid 15-residue (EAAAKEAAAKEAAAK) linker exhibited better performance (Supplementary Fig. 5). The final version of A&C-BE, A&C-BEmax, was generated by adding two copies of uracil DNA glycosylase inhibitor (UGI)5 (Supplementary Fig. 1). Through the serial optimization steps above, the base conversion efficiency, product purity and A/C simultaneous conversion activity of A&C-BEmax were significantly increased compared to the original construct (Supplementary Fig. 6). Compared to A&C-BEmax, co-transfection of CBE and ABE7.10 with six single-guide RNAs (sgRNAs) yielded a very low simultaneous A/C conversion ratio for most targets (Fig. 1b,c and Supplementary Fig. 7). Most likely, single base editors competed with each other due to occupation of the target site. Once the target has been edited by one editor, the other base editor will not efficiently function because the sgRNA no longer perfectly matches the target DNA. Our initial study showed that, through fusion of cytidine and adenosine deaminase to Cas9n, A&C-BEmax was able to catalyze A/C base conversion on the same allele with increased cytidine deaminase activity compared to CBEs.To unbiasedly characterize the performance of A&C-BEmax compared to ABEmax and AID-BE4max (with optimized codon, NLS and UGI copies; see Supplementary Sequence 1), 28 endogenous targets (including four sites containing only Cs or As, respectively) were investigated in HEK293T cells (Fig. 1d). The A-to-G editing window of A&C-BEmax was consistent, and the A-to-G editing efficiency was similar or slightly decreased compared to ABEmax at most of the targets (Fig. 1d,e and Supplementary Fig. 8a). Notably, the C-to-T editing window for A&C-BEmax was expanded to 16 nucleotides (positions 2–17) compared to positions 3–13 for AID-BE4max (Fig. 1d,f and Supplementary Fig. 8b).

The average C-to-T conversion efficiency was similar at positions 2–6, but the activity was increased 1.9~14-fold at positions 7–17 in A&C-BEmax-treated cells compared to AID-BE4max (Fig. 1f). Analyzing the first 20 targets containing both As and Cs in the editing window revealed that the simultaneous A/C mutation rate on the same allele ranged from 2% to 30%, and the proportion of alleles bearing only C-to-T or A-to-G varied from 5.3% to 82.6% and from 0.2% to 10%, respectively (Fig. 1g and Supplementary Fig. 9). Higher simultaneous A/C conversion rates were observed in sites containing adenines at position 6–7 within the target (Fig. 1g and Supplementary Fig. 9). Robust base editing efficiency was observed in HeLa cells at all examined targets, suggesting that A&C-BEmax is a general tool functioning in different cell types (Supplementary Fig. 10).By further comparing the mutation spectra created by A&C-BEmax, ABEmax or AID-BE4max, we found that A&C-BEmax generated more mutant allele types (Fig. 1h). Moreover, NGS studies showed that A&C-BEmax did not generate more DNA indels than AID-BE4max (Supplementary Fig. 11) and exhibited similar off-target effects with AID-BE4max and fewer than Cas9 as determined after analyzing 88 potential off-target sites, which were selected either by Cas-OFFinder software prediction6 or experimental identification by GUIDE-seq1, ChIP-seq3 or Digenome-seq7through NGS studies (Supplementary Fig. 12). However, further studies should investigate whether A&C-BEmax also induces significant unpredictable DNA off-targeting effects similar to BE3 (refs. 8,9), because a different cytidine deaminase was incorporated. As recent studies demonstrated that base editors (except for AID-BE3) induced extensive RNA single-nucleotide variants (SNVs)10,11, transcriptome-wide off-target effects of A&C-BEmax were determined. Consistently, BE4max and ABEmax induced tens of thousands of RNA SNVs, whereas AID-BE4max generated fewer RNA mutations. A&C-BEmax had greatly reduced A-to-I RNA off-targeting, less than 20% of that of the ABEmax-treated group (Fig. 1i). The above data demonstrate that A&C-BEmax has some advantages over traditional base editors, including simultaneous A/C mutation on the same allele, higher editing activity, a wider window and less RNA off-targeting.To further investigate the application of A&C-BEmax for gene therapy, a β-hemoglobinopathy model was employed. It is well accepted that reactivation of fetal hemoglobin (HbF) is a feasible strategy for the treatment of sickle cell disease and β-thalassemia12. Patients with β-thalassemia with mutations −114C-to-T or −113A-to-G in the promoter of the γ-globin genes (HBG1 and HBG2) have been identified with increased HbF production and reduced symptoms12. The −114 or −115 C-to-T mutation disrupts the BCL11A binding site, which is a strong transcription repression element responsible for HBG1 and HBG2 inhibition in adults13. The −113A-to-G mutation does not disrupt the BCL11A site but creates a new GATA1 binding site that has been demonstrated to activate HBG1 transcription14. To test the feasibility of A&C-BEmax to disrupt the BCL11A binding site (−114 or −115) and create a GATA1-site (−113) simultaneously, an sgRNA targeting the HBG1and HBG2 promoter was designed and tested in HEK293T cells (Fig. 2a). A&C-BEmax induced variant mutations besides −115 to −113 and exhibited a 1.8-fold and 1.7-fold increase in the editing efficiency at −114C (41.2% versus 14.7%) and −115C (39.2% versus 14.8%), respectively, over AID-BE4max and similar efficiency at −113A with ABEmax, yielding an 8.76% rate of A/C simultaneous mutation (including −113A with either −114C or −115C) on the same allele (Fig. 2b and Supplementary Fig. 13). However, co-transfection of ABE and CBE hardly ever generated A/C simultaneous mutation at either the −115/−113 or the −114/−113 sites (Fig. 2b and Supplementary Fig. 13). Next, an erythroid precursor cell line HUDEP-2 (ref. 15) was employed to test the physiological function of the mutation generated by lentivirus-packaged A&C-BEmax (Supplementary Fig. 14 and Supplementary Sequence 3). A&C-BEmax induced highly efficient base editing, strongly promoting HBG mRNA induction (A&C-BEmax, 73.7% versus AID-BEmax, 45.9%) as a combined effect of several types of mutation in the promoter that disrupted the BCL11A binding site and/or generated a GATA1 site in the promoter (Fig. 2c,d and Supplementary Fig. 15). To further verify whether simultaneous A/C mutation induced higher HBG reactivation, single clones of HUDEP-2(ΔGγ) cells, which carry only one γ-globin gene to facilitate genotype analysis16, were established after A&C-BEmax/sgRNA delivery. After analysis of single clones, we demonstrated that simultaneous −113A/−114C conversion (#B-7) induced the highest HBG expression, and additional mutations at other sites (#B-7 versus #B-1) did not show extra induction (Fig. 2e,f). These data constitute also a proof of principle showing that A&C-BEmax was able to generate sequence variety to investigate the relationship of genome type with function, suggesting that A&C-BEmax could be used as a tool to dissect the function of a given sequence at single-nucleotide resolution.To explore the therapeutic potential of A&C-BEmax to correct pathogenic mutations, we computationally profiled all clinically relevant variants in Clin Var17 and single-nucleotide polymorphisms (SNPs) in dbSNP18 potentially able to be targeted by A&C-BEmax using our recently reported strategy19. Our in silico analysis found 203 target sites containing known pathogenic A-to-G mutation(s) and C-to-T mutation(s) that could potentially be reverted by A&C-BEmax through a single sgRNA, albeit the SNVs might be located on different alleles (Supplementary Fig. 16 and Supplementary Table 1). Thetargeting scope increased by 2.8-fold (~573) when the Cas9-NG variant20 was leveraged for the analysis (Supplementary Fig. 16 and Supplementary Table 1). Moreover, if we expand the category to one pathogenic mutation plus an SNP with unknown function within the editing window, 3,831 targets are potentially restorable by A&C-BEmax (10,784 targets for Cas9-NG variant), suggesting broad applications of A&C-BEmax for gene therapy. Additionally, A&C-BEmax efficiently generates simultaneous A/C conversions that significantly increase the variants number of a given triplet code compared to ABE or CBE (Supplementary Fig. 17 and Supplementary Table 2).

Fig. 1 | Design and optimization of A&C-BE. a, Schematic diagram of A&C-BE construction. Cytidine deaminase refers to rat APOBEC1 or human AID. mTadA, evolved TadA (TadA*). Cas9n, Cas9D10A. b, Heat maps showing C-to-T or A-to-G base editing efficiency for base editors at the endogenous FANCF-sg3 target site in HEK293T cells. Data represent the means from three independent experiments. c, Comparison of the products distribution among edited DNA sequencing reads of the FANCF-sg3 target edited by variant base editors. The individual data points are shown as light green (only C-to-T), black (only A-to-G) and yellow (simultaneous C-to-T and A-to-G) dots. Values and error bars reflect the means and s.d. of three independent experiments. d, Base editing efficiencies of ABEmax, AID-BE4max and A&C-BEmax at 28 endogenous human genomic loci. e, Merged data of average C-to-T editing efficiency at 24 targets in d (except CCR5-sg1, ABE site5, ABE site12 and ABE site13 with no Cs in the editing window) edited by AID-BE4max or A&C-BEmax. Data represent the means from three independent experiments. Values and error bars reflect the means and s.d. of three independent experiments. f, Merged data of average A-to-G editing efficiency at 24 targets in d (except KCNS1-sg1, VEGFA site2, FANCF-M-b and PD-1-sg10 with no As in the editing window) edited by AID-BE4max or A&C-BEmax. Data represent the means from three independent experiments. g, The composition of A&C-BEmax base editing products at 20 endogenous human genomic loci. The individual data points are shown as light green (only C-to-T), black (only A-to-G) and yellow (simultaneous C-to-T and A-to-G) dots. Values and error bars reflect the means and s.d. of three independent experiments. Statistical source data are provided in Source Data Fig. 1. h, Mutation allele types yielded by ABEmax, AID-BE4max and A&C-BEmax at 20 endogenous target sites. Each data point represents average mutation allele types at each target site (except target sites with only As or only Cs in the editing window) calculated from three independent experiments. Data are means ± s.d. P value was determined by two-tailed Student’s t-test. i, Jitter plots from RNA-sequencing experiments in HEK293T cells showing efficiencies of C-to-U or A-to-I conversions (y axis) with BE4max, ABEmax, AID-BE4max and A&C-BEmax expression or a GFP negative control. Total number of modified bases is listed at the top. Each biological replicate is listed on the bottom.

Fig. 2 | Efficient editing of HBG promoter by A&C-BEmax in HUDEP-2 cells. a, Schematic diagram of the critical sequence regulating HBG1 and HBG2expression. The core sequence of the BCL11A binding site is boxed in red. The GATA1 binding site is boxed in green. The PAM sequence of the target site is in blue. b, Editing efficiencies of A&C-BEmax, ABEmax+AID-BE4max, ABEmax or AID-BE4max in the HBG1 and HBG2 promoter in HEK293T cells. Values and error bars reflect the means and s.d. of three independent experiments. c, Base editing efficiency at the HBG1 and HBG2 promoter site in pooled HUDEP-2 cells transduced with lentiviral ABE.7.10-N-AIDmax or AID-BEmax after puromycin selection. Values and error bars reflect the means and s.d. of three independent experiments. d, Comparison of γ-globin mRNA expression relative to β-like globin mRNA via ABE7.10-N-AIDmax or AID-BEmax treatment in HUDEP-2 cells after differentiation. Values and error bars reflect the means and s.d. of five independent experiments. P value was determined by two-tailed Student’s t-test. e, γ-globin mRNA expression relative to β-like globin mRNA in individual single clones of HUDEP-2(ΔGγ) cells treated with the A&C-BEmax and sgRNA vector. Values and error bars reflect the means and s.d. of five independent experiments. P value was determined by two-tailed Student’s t-test. The genotype of each clone is presented in f. The comparison between clone #B-7 and #B-16 indicates that a de novo GATA1 site is critical for HBG induction.

In summary, through tethering two base deaminases, we developed a dual-functional base editor, A&C-BEmax, which can induce simultaneous C-to-T and A-to-G conversions with increased CBE activity and reduced RNA off-targeting (compared to ABEmax). A&C-BEmax is a valuable tool not only for dissecting genomic sequence function at a single base resolution but also for the therapy of genetic disorders.