Phage-assisted evolution of highly active cytosine base editors with enhanced selectivity and minimal sequence context preference

Abstract

TadA-derived cytosine base editors (TadCBEs) enable programmable C•G-to-T•A editing while retaining the small size, high on-target activity, and low off-target activity of TadA deaminases. Existing TadCBEs, however, exhibit residual A•T-to-G•C editing at certain positions and lower editing efficiencies at some sequence contexts and with non-SpCas9 targeting domains. To address these limitations, we use phage-assisted evolution to evolve CBE6s from a TadA-mediated dual cytosine and adenine base editor, discovering mutations at N46 and Y73 in TadA that prevent A•T-to-G•C editing and improve C•G-to-T•A editing with expanded sequence-context compatibility, respectively. In E. coli, CBE6 variants offer high C•G-to-T•A editing and no detected A•T-to-G•C editing in any sequence context. In human cells, CBE6 variants exhibit broad Cas domain compatibility and retain low off-target editing despite exceeding BE4max and previous TadCBEs in on-target editing efficiency. Finally, we show that the high selectivity of CBE6 variants is well-suited for therapeutically relevant stop codon installation without creating unwanted missense mutations from residual A•T-to-G•C editing.

Fig. 1. Development of a highly active and selective cytosine base editor from a TadA dual base editor using phage-assisted evolution.

a An active and selective cytosine base editor, but not one with residual adenine base editing activity, can cleanly install stop codons into target genes. b Schematic of the evolution of a cytosine base editor from a TadA-derived dual base editor (TadDE). c Diagram depicting phage-assisted continuous evolution (PACE, left) and the selection circuit used in this study (right). A continuous flow of E. coli host cells with the selection circuit and a mutagenesis plasmid (red) are infected by selection phage encoding a deaminase (SP). In the selection circuit, phage propagation is linked with gIII expression (P2), which can only be transcribed with active T7 RNA polymerase. T7 RNA polymerase (P3) is fused to a C-terminal degron, and the deaminase must perform C•G-to-T•A editing to install a stop codon before the degron to generate active T7 RNA polymerase. In the event of phage infection, the full base editor is reconstituted using a split-intein system (P1), and mutations accumulate in the deaminase. Beneficial mutations lead to phage propagation and enrichment in the lagoon, while the less-fit phage are unable to propagate and are washed out by the constant outflow. d Evolutionary trajectory of an active and selective cytosine base editor from TadDE. Phage-assisted non-continuous evolution (PANCE) was performed on TadA-DE until phage titers increased despite higher stringency. The resulting genotypes identified a conserved mutation at position N46 in TadA, so an NNK library was constructed to diversify this position, and PANCE was performed on the resulting variants. PACE was performed for >100 hrs on the resulting variants from both PANCE experiments. Dilution factors are indicated on the right y-axis. Relative promoter units (normalized to proD) for proA, proB, proC, and proD are as follows: 0.030, 0.119, 0.278, and 1.000, respectively. e Mutation table from evolved deaminases showing conserved mutations. f Cryo-EM structure of ABE8e (PDB: 6VPC) with mutations labeled. New mutations are highlighted in magenta, and mutations inherited from TadDE are highlighted in yellow. Source data are provided as a Source Data file.

Fig. 2. Profiling the activity and sequence context specificity of CBE6 variants in E. coli.

a Schematic of a 32-member library that varies the 5’ and 3’ sequence contexts of a target edit at position 6 within the editing window of an E. coli protospacer. b Bar values indicate the average activity of CBE variants when tested on a library of 32 substrates designed to contain the target base (A or C) at protospacer position 6 with all possible combinations of flanking nucleotides. Dots represent the average percentage of sequencing reads containing the specified edit (A•T-to-G•C or C•G-to-T•A) for each of the 16 sequence contexts (each dot represents an average of n = 3 independent biological replicates). The dots are colored according to the 5′ upstream base (A, red; C, green; G, blue; T, yellow). c Bar values indicate the average C•G-to-T•A editing efficiency of CBE variants or A•T-to-G•C activity of ABE8e when tested on a library of 448 substrates designed to contain the target base (A or C) at protospacer positions 1-10 with the 5′ and 3′ base varied as A, T, C, or G, normalized to the highest C•G-to-T•A activity for CBE variants or A•T-to-G•C activity for ABE8e. Dots represent the average percentage of sequencing reads containing the specified edit (A•T-to-G•C or C•G-to-T•A) for each of the 224 sequence contexts (each dot represents an average of n = 2 independent biological replicates). Full data for C•G-to-T•A and A•T-to-G•C activity are provided in Supplementary Figs. 8 and 9. Source data are provided as a Source Data file.

Fig. 3. Comparison of CBE6 variants with existing CBEs in mammalian cells.

a CBE6 variants or existing cytosine base editors, all using SpCas9 nickase domains in the BE4max architecture, were transfected into HEK293T cells with guide RNAs targeting three protospacers. Data are presented as mean values ± SD. Dots represent individual values from n = 3 independent biological replicates. PAM sequences are underlined. HEK293T site 2 is abbreviated HEK2, and HEK293T site 4 is abbreviated HEK4. b CBE6 variants along with existing cytosine base editors using eNme2-C Cas9 nickases in the BE4max architecture were transfected into HEK293T cells with guide RNAs targeting three protospacers. Data are presented as mean values ± SD. Dots represent individual values from n = 3 independent biological replicates. PAM sequences are underlined. Full data for C•G-to-T•A and A•T-to-G•C activity are in Supplementary Figs. 12 and 13. Source data are provided as a Source Data file.

Fig. 4. Cas9-independent DNA and RNA off-target editing by CBE6 variants.

a Average Cas9-independent off-target editing across all cytosines for six orthogonal R-loops (SaR1–SaR6) generated by a dead S. aureus Cas9. Data are presented as mean values ± SD. Dots represent individual values from independent biological replicates. b Off-target RNA editing. Base editors were transfected into HEK293T cells and harvested after 48 hours. cDNA was synthesized, and high-throughput sequencing was used to analyze CTNNB1, IP90, and RSL1D1. Data are presented as mean values ± SD. Dots represent individual values from independent biological replicates. Source data are provided as a Source Data file.

Fig. 5. Stop codon installation at therapeutically-relevant loci by CBE6 variants in fibroblasts.

CBE6 variants were used to install stop codons in PCSK9, a therapeutic strategy for lowering LDL cholesterol levels. The gray boxes indicate the desired location of stop codon installation. Data are presented as mean values ± SD. Dots represent individual values from independent biological replicates. PAM sequences are underlined. Source data are provided as a Source Data file.