Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity

Abstract

Applications of adenine base editors (ABEs) have been constrained by the limited compatibility of the deoxyadenosine deaminase component with Cas homologs other than SpCas9. We evolved the deaminase component of ABE7.10 using phage-assisted non-continuous and continuous evolution (PANCE and PACE), which resulted in ABE8e. ABE8e contains eight additional mutations that increase activity (k_app) 590-fold compared with that of ABE7.10. ABE8e offers substantially improved editing efficiencies when paired with a variety of Cas9 or Cas12 homologs. ABE8e is more processive than ABE7.10, which could benefit screening, disruption of regulatory regions and multiplex base editing applications. A modest increase in Cas9-dependent and -independent DNA off-target editing, and in transcriptome-wide RNA off-target editing can be ameliorated by the introduction of an additional mutation in the TadA-8e domain. Finally, we show that ABE8e can efficiently install natural mutations that upregulate fetal hemoglobin expression in the BCL11A enhancer or in the the HBG promoter in human cells, targets that were poorly edited with ABE7.10. ABE8e augments the effectiveness and applicability of adenine base editing.

Extended Data Figure 1.. Mutation table of variants from PANCE and PACE.

Data were obtained by sequencing individual plaques. Conserved mutations are bolded. Mutations that are highlighted in the structure in Fig. 2b are highlighted to match the amino acid positions in the structure. Genotypes in red were tested for base editing activity in mammalian cells.

Extended Data Figure 2.. PACE schedule for deoxyadenosine deaminase evolution.

Lagoon L1 contains host cells harboring P1, P2, and P3e. Lagoons L2 and L3 contain host cells harboring P1, P2, and P3g, which form a more stringent selection circuit than the circuit in lagoon L1. For details on plasmids, see Supplementary Table 1. The stringency of the ABE selection was further modulated by increasing the lagoon flow rate (dashed lines). For the first 12 hours, gene III was expressed by the addition of anhydrotetracycline to enable genetic drift in the absence of selection pressure^,.

Extended Data Figure 3.. Titration data at eight editor doses comparing base editing efficiencies for ABE8e and ABE8e-dimer at three sites in HEK293T cells.

Base editing with ABE8e and ABE8e-dimer in HEK293T cells at three genomic sites in HEK293T cells. Transfections were performed with constant amount of sgRNA plasmid but eight varying doses of ABE plasmid. For all plots, dots represent individual biological replicates and bars represent mean±s.d. of three independent biological replicates.

Extended Data Figure 4.. TadA-8e V106W analysis for SaCas9 and LbCas12a.

a, DNA editing comparing SaABE7.10, SaABE8e, and SaABE8e(TadA-8e V106W) at four genomic sites in HEK293T cells. b, DNA editing comparing LbABE7.10, LbABE8e, and LbABE8e(TadA-8e V106W) at six genomic sites in HEK293T cells. For all plots, dots represent individual biological replicates and bars represent mean±s.d. of three independent biological replicates.

Figure 1.. Phage-assisted evolution of a deoxyadenosine deaminase.

a, General PACE overview for base editor evolution^,. E. coli host cells contain a plasmid-based genetic circuit that links expression of gene III (gIII, encoding pIII) to the activity of the base editor encoded in a modified M13 bacteriophage (blue). The production of infectious progeny phage requires expression of gene III, which only occurs in host cells infected by phage variants that encode active base editors. Phage exist in a fixed-volume vessel (the lagoon) continuously diluted with host-cell culture, so only those phage that propagate faster than the rate of dilution can persist and evolve. b, The selection circuit in PANCE or PACE for evolving the deoxyadenosine deaminase component of ABEs. Plasmid P1 (purple) contains gene III driven by a T7 promoter and an sgRNA driven by a Lac promoter. Plasmid P2 (orange) expresses an Npu C-intein fused to a catalytically dead SpCas9 (dCas9), which forms full-length base editor upon trans-intein splicing with TadA fused to a Npu N-intein (encoded on the SP, in blue). Plasmid P3 (green) expresses a T7 RNAP that contains two stop codons that can be corrected to arginine and glutamine upon adenine base editing; this editing event drives expression of gene III. We developed eight P3 variants (P3a-h) with different promoters and ribosome binding sites (RBS) to tune selection stringency. The phage genome is continuously mutated by expression of mutagenic genes from the mutagenesis plasmid (MP, red). c, The gene encoding T7 RNA polymerase (T7 RNAP), which is required for gene III expression from the T7 promoter, contains two stop codons at R57 and Q58. Deamination of both adenines by ABE converts the stop codons back to Arg and Gln, resulting in active T7 RNAP and gene III expression. d, Overnight phage propagation assays to test the activity of phage pools in host cells harboring P1, P2, and eight different variants of P3 (P3a-h) of increasing stringency. We mixed phage pools with an excess of log-phase host cells and allowed the phage to propagate overnight. The fold phage propagation is the output phage titer divided by the input titer. For all plots, dots represent individual biological replicates, bars represent mean values, and error bars represent the standard deviation of three independent biological replicates.

Figure 2.. Mutations and kinetics of TadA-8e, and editing characteristics of ABE8e in human cells.

a, Conserved mutations after 25 passages of PANCE (PANCE 1 = passages 1–15; PANCE 2 = passages 11–25), and genotypes of five TadA variants emerging from 84 hours of PACE (surviving ~10¹⁷⁴-fold total dilution). For a list of all evolved TadA genotypes, see Extended Data Figure 1. b, E. coli TadA deaminase (green, PDB 1Z3A) aligned with the structure of S. aureus TadA (not shown) complexed with tRNA^Arg (grey, PDB 2B3J). Mutations evolved during PANCE and PACE are colored to correspond to those in Extended Data Figure 1. c, The architecture of ABE7.10 (ABEmax) and ABE8e. d, Left: representative denaturing polyacrylamide gels of 5’-radiolabeled dsDNA deamination reactions performed with in vitro reconstituted ABE7.10 and ABE8e RNPs, followed by treatment with E. coli EndoV, which cleaves DNA 3’ of deoxyinosine. Middle: the fraction of deaminated dsDNA plotted as a function of time in hours. Right: the fraction of deaminated dsDNA plotted as a function of time in the first 60 minutes. The data were fit to a single exponential equation to extract apparent first-order deamination rate constants for ABE7.10 (black) and ABE8e (orange). Data are represented as the mean±s.d. from three independent experiments. e, Base editing in HEK293T cells by SpABE7.10 versus SpABE8e, SaABE7.10 versus SaABE8e, and LbABE7.10 and enAsABE7.10 versus LbABE8e and enAsABE8e, for the two nucleotides with the highest editing efficiency within each protospacer. Bars represent mean values, and error bars represent the standard deviation of three independent biological replicates. For editing across the entire protospacer for each site and indel frequencies, see Supplementary Figs. 4-6. f, Base editing in HEK293T cells by NG-ABE7.10 versus NG-ABE8e and SaKKH-ABE7.10 versus SaKKH-ABE8e, for the two nucleotides with the highest editing within each protospacer. For editing efficiencies across the entire protospacer and indel frequencies, see Supplementary Figs. 7, 8. g, Base editing in HEK293T cells within the protospacer by CP–ABE7.10 and CP–ABE8e variants, compared to SpABE7.10 and SpABE8e. For editing efficiencies across the entire protospacer and indel frequencies, see Supplementary Fig. 4. For all plots, bars represent mean values, and error bars represent the standard deviation of three independent biological replicates.

Figure 3.. Off-target analysis of ABE8e.

a, DNA off-target analysis comparing ABE7.10 plasmid delivery, ABE8e plasmid delivery, and ABE8e RNP delivery at site 5 (HBG), site 6 (VEGFA3), and EMX1. Editing efficiencies and on-target:off-target editing ratios are shown. b, Off-target transcriptome-wide A-to-I conversion analysis in cellular RNA. c, DNA editing comparing ABE7.10, ABE8e, and ABE8e(TadA-8e V106W) at seven genomic sites in HEK293T cells. d, Orthogonal R-loop assay overview. e, Cas9-independent off-target A•T to G•C editing frequencies detected by the orthogonal R-loop assay at each R-loop site with dSaCas9 and a SaCas9 sgRNA. Each R-loop was performed by cotransfection of ABE7.10, ABE8e, or ABE8e (TadA–8e V106W), and a SpCas9 sgRNA targeting site 3 with dSaCas9 and a SaCas9 sgRNA targeting R-loops 1–5, respectively. For all plots, bars represent mean values, and error bars represent the standard deviation of three independent biological replicates.

Figure 4.. Adenine base editing with ABE8e at disease-relevant loci in human cells.

a, Base editing efficiency and indel frequencies in HEK293T cells at a GATA1 binding site of the BCL11A enhancer by ABE7.10 and ABE8e. Editing efficiencies at each adenine in the protospacer individually and efficiencies of editing both adenines within the same allele. b, Base editing efficiencies and indel frequencies in HEK293T cells of the HBG1/2 promoter with ABE7.10 and ABE8e. Data are shown for each sgRNA provided individually, and for dual editing of both target sites within a single allele when both sgRNAs are provided simultaneously. Protospacers are named based on the position of the target adenine relative to the HBG transcription start site. For all plots, bars represent mean values, and error bars represent the standard deviation of three independent biological replicates.