Continuous evolution of base editors with expanded target compatibility and improved activity

Abstract

Base editors use DNA-modifying enzymes targeted with a catalytically impaired CRISPR protein to precisely install point mutations. Here, we develop phage-assisted continuous evolution of base editors (BE-PACE) to improve their editing efficiency and target sequence compatibility. We used BE-PACE to evolve cytosine base editors (CBEs) that overcome target sequence context constraints of canonical CBEs. One evolved CBE, evoAPOBEC1-BE4max, is up to 26-fold more efficient at editing cytosine in the GC context, a disfavored context for wild-type APOBEC1 deaminase, while maintaining efficient editing in all other sequence contexts tested. Another evolved deaminase, evoFERNY, is 29% smaller than APOBEC1 and edits efficiently in all tested sequence contexts. We also evolved a CBE based on CDA1 deaminase with much higher editing efficiency at difficult target sites. Finally, we used data from evolved CBEs to illuminate the relationship between deaminase activity, base editing efficiency, editing window width and byproduct formation. These findings establish a system for rapid evolution of base editors and inform their use and improvement.

Figure 1

Overview of base editing and PACE. (a) Cytosine base editing converts target C•G base pairs to T•A at guide RNA-specified DNA sequences. R-loop formation by the Cas9 domain exposes a small bubble of single-stranded DNA (bases numbered 1–20, with the PAM at positions 21–23) to the fused cytidine deaminase. Cytosines in this bubble are deaminated to uracil, which is protected from excision by uracil glycosylase inhibitor (UGI). To increase editing efficiency in some CBEs, the non-edited strand is nicked, stimulating cellular repair and replication to replace the original C•G base pair with T•A. (b) General PACE schematic. E. coli host cells contain a plasmid-based genetic circuit that links expression of gene III (gIII, encoding pIII) to the activity of the biomolecule of interest 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 with phage variants encoding the desired activity of interest. The phage genome is mutagenized continuously by a mutagenesis plasmid in the host cells. Since the phage exist in a fixed-volume vessel (the lagoon) continuously diluted with host-cell culture, only those phage that propagate faster than the rate of dilution can persist and evolve in the lagoon.

Figure 2

Design and validation of BE-PACE. (a) Schematic for the BE-PACE circuit. T7 RNA polymerase, required to express gene III and luciferase (translationally coupled via overlapping stop and start codons) from a T7 promoter, is fused through a Trp-containing linker to a C-terminal degron that causes the destruction of the protein. The Trp codon (TGG) provides the CBE target CCA on the transcriptional template strand. Deamination of either cytosine by a CBE converts the Trp codon to a STOP codon (UAG, UGA or UAA, arising from transcription of CUA, UCA or UUA on the template strand), preventing translation of the degron, restoring T7 RNA polymerase activity, and activating gene III expression. The 5’ context for the deamination target can be varied by changing the linker sequence. (b) A luciferase assay shows that all components of the BE-PACE system are required for circuit activation. Leaky expression of BE2 during cell growth resulted in significant activation of the circuit even in the absence of induction. Guide RNA targets the CBE to either the T7 RNAP–degron linker or to green fluorescent protein. The H61A mutation in the base editor inactivates its APOBEC1 deaminase. Expression of APOBEC1 and dCas9 as separate polypeptides instead of BE2 does not activate the circuit. Dots represent biological replicates and bars represent mean values. (c) Discrete overnight phage propagation assays to test the prototype BE-PACE circuit. Phage containing the genes shown were mixed with an excess of host cells and allowed to propagate until the host cells grow to saturation. The output phage titer was divided by the input titer to calculate fold phage propagation. Phage with activities that circumvent or short-circuit the selection (gIII, T7 RNAP) enrich strongly (≥10⁴-fold), while BE2 phage propagate weakly (output titer 500-fold lower than input titer) but more than empty phage (5,000-fold lower output titer).

Figure 3

Design and validation of the split intein BE-PACE selection. (a) Plasmids (grey backbones) and phage (orange backbone) in the optimized BE-PACE host-cell selection circuit. In the minimal phage split, only the deaminase is encoded by phage, while dCas9 is encoded on a host plasmid. In the balanced phage split, part of dCas9 is encoded on the phage and the remainder on a host plasmid. In either split, full-length BE2 is reconstituted upon phage infection as a single polypeptide by trans-splicing inteins. (b) Luciferase assays show that split BE2 activates the circuit in an intein-, guide RNA-, and APOBEC1 activity-dependent manner. The C37A mutant of the C-intein disrupts splicing, but not association between the split intein components. Phage backbone numbers refer to which generation phage backbone was used (see text). Bar heights represent mean values for fitted slopes of luminescence per OD₆₀₀ versus time (see Methods) and dots represent slopes for individual biological replicates. (c) BE-PACE competition experiment. Host cells contained the low-stringency selection circuit TCC1 and mutagenesis plasmid. A lagoon, continuously diluted with host cells, was seeded with 99.9% red fluorescent protein (RFP) phage (lower band) and 0.1% APOBEC1–intein phage (upper band). The phage population composition was monitored by PCR using primers flanking the phage insert. L denotes DNA size standard ladder. The total phage titer over time and the lagoon flow rate is shown on the graph at the bottom. This experiment was not repeated.

Figure 4

BE-PACE of APOBEC1, FERNY, and CDA1, and characterization of evolved deaminase CBEs in mammalian cells. (a) PACE of APOBEC1, FERNY, and CDA1 deaminase phage as intein fusions. BE-PACE selection circuits are described in Supplementary Table 2. Solid lines and dots show phage titers (left axis) and dotted lines show flow rate (right axis) during BE-PACE. These experiments were not repeated. (b) Performance of evolved deaminases in the luciferase assay in bacteria (top panel) and in CBEs editing HEK293T cells (bottom panels) at five endogenous genomic sites. Target cytosines are color coded according to the base immediately 5’ of the edited C. In the top panel, deaminases were tested on all four possible ANC₄C₅A target sites with N = A, C, G or T in low-stringency BE-PACE circuits (see Supplementary Table 2). In the lower panels, C•G-to-T•A base editing is shown for cells transfected with each CBE (vertical columns, with evolved deaminase genotypes shown at the bottom) and each of five guide RNAs. Deaminases were not codon-optimized for human cell expression, but the remainder of the BE4max architecture was codon-optimized. Editing byproducts are shown in Supplementary Table 3. Base editing levels are shown for each edited C within the protospacer (positive X-axis numbers, with the PAM at positions 21–23) or upstream of the protospacer (negative X-axis numbers, with −1 being one base upstream). Genotypes for the wild-type deaminase (or the reconstructed ancestral sequence for node 656, “FERNY”, labeled Anc656) and for each mutant are given below each clone name. Evolved clones are named for the time point (in hours) at which they were isolated from PACE (Fig. 4a). The genotypes in grey are evoAPOBEC1-BE4max (left), evoFERNY-BE4max (middle), and evoCDA1-BE4max (right). Dots represent individual biological replicates and bars represent mean values.

Figure 5

Base editing performance of evolved deaminase CBEs, all codon-optimized in the BE4max architecture, in mammalian cells. (a) Editing by wild-type and evolved deaminase CBEs for five endogenous genomic test sites in HEK293T cells. Target cytosines are color-coded by the base immediately 5’ of the edited C as in Fig. 3b. Protospacer positions are specified by X-axis numbers. Dots represent individual biological replicates and bars represent mean values. (b) Base editing activity window plots showing mean C•G-to-T•A editing at all tested protospacer positions across six HEK293T sites. Target cytosines preceded by a 5’ G are excluded for BE4max, AncBE4max, and FERNY-BE4max to avoid misrepresenting their editing windows due to sequence context preference. The dotted horizontal line represents half-maximal peak editing to approximate editing window width. (c) Off-target editing by wild-type and evolved deaminases for a selection of known off-target dCas9 binding sites for the HEK2, HEK3 and HEK3 guide RNAs. The data shown are for off-target sites amplified from the treated cells shown in corresponding panels of Fig. 4a. Protospacer positions are specified by X-axis numbers. Dots represent individual biological replicates and bars represent mean values. Data from all off-target sites examined are in Supplementary Fig. 20 and Supplementary Table 3.

Figure 6

(a-c) Cytosine base editing at disease-relevant sites in primary mammalian cells and cell lines. Protospacer positions are specified by X-axis numbers, with the target C indicated by an arrow, and Cs within the expected editing window of each CBE (based on Fig. 5b) marked with horizontal brown lines. Dots represent individual biological replicates and bars represent mean values. (a) Editing the TMC1 site to revert the Y182C mutation in primary embryonic fibroblasts from the baringo mouse model of recessive hearing loss. (b) Editing the Alzheimer’s disease-associated APOE4 allele into APOE3’ and APOE3 by installing R158C and R112C in immortalized mouse astrocytes. In (a) and (b), the percent of sequencing reads that contain the targeted coding mutation with no other non-silent mutations or indels is shown in grey and labeled “A”. (c) Editing the Wolfram syndrome 1-associated WFS1 gene to install Q1884STOP in HEK293T cells. The grey “A” bar shows the percent of reads with Q1884STOP and no other sequence changes. (d) Model for the relationship between site characteristics, deaminase activity, and editing outcomes in mammalian cells (see Discussion and Supplementary Discussion 6).