Improved cytosine base editors generated from TadA variants

Fig. 1. Directed evolution of CABE-T1 and CABE-T2. a, Schematic overview of directed evolution workflow to identify TadA variants capable of C-to-U deamination. b, Schematic representation of expression and selection plasmids used in directed evolution campaigns. Left: base editor expression plasmid encoding library member and sgRNA, right: selection plasmid encoding a nonfunctional antibiotic resistance gene. Reversion at targeted sites restores gene function. c, Overview of CBE-T editor development through directed evolution and protein engineering. d, Schematic representation of select base editor architectures. ABE12 contains a laboratory-evolved TadA* deaminase, nCas9 (D10A) and a nuclear localization tag (bpNLS). Dual editor (for example, SPACE16) is comprised of an evolved TadA* deaminase, nCas9 (D10A), cytidine deaminase rAPOBEC1 along with two units of UGI and a bpNLS tag. CBEs (for example, BE4 (ref. 3)) are comprised of a naturally occurring cytidine deaminase (for example, rAPOBEC1), nCas9 D10A, two units of UGI and a bpNLS tag. CABE-Ts, reported here, are comprised of a TadA variant capable of A-to-G and C-to-T editing (TADAC), nCas9 (D10A), two units of UGI and a bpNLS tag.…

Fig. 2. Crystal structures of TadA*8.20 and TADAC-1 variants. a, Superpositions between monomers of TadA*8.20 and TADAC-1 variants (RMSD between 0.4 Å and 0.9 Å for all of the Cα atoms). TADAC-1.17 (protein—blue, ssDNA—orange) and TadA*8.20 (protein—green, ssDNA—yellow) structures are similar (RMSD of 0.4 Å for all of the Cα atoms), showing that the substitutions (cyan spheres) neither change the protein structure nor the ssDNA binding mode but may impact catalysis due to the active site substitution S82T. TADAC-1.14 structure (yellow) has a different conformation for the loop between β4 and β5 (magenta) than TadA*8.20 due to the substitution G112H (magenta sphere). TADAC-1.19 structure (pink) has an extended loop between α1 and β1 (orange) with a different conformation than TadA*8.20 due to the substitution E27G (orange sphere). C represents the C-terminus. b, Surface of the active site cavities of TadA*8.20 and TADAC-1 variants. TADAC-1.17 (second panel) shows no difference compared to TadA*8.20 (first panel). TADAC-1.14 loop between β4 and β5 (third panel) alters the shape of the active site cavity, making A109 steric clash with dT(8) of TadA*8.20-ssDNA. TADAC-1.19 loop between α1 and…

Fig. 3. Structure-guided combinatorial screens. a, Structure-guided combinatorial library design workflow. For clarity, only one TadA*8.20 monomer (green) with ssDNA (yellow) is shown. The 23 substitution sites identified in the first round of evolution are shown as violet spheres in the first panel. Based on the structures of TADAC-1 variants (Fig. 2), 10 sites were selected, 8 from TADAC-1 (violet) plus 2 rationally selected (cyan), in the loop between α1 and β1, α2-helix, active site, the loop between β4 and β5 and α5-helix, to generate the TADAC-3 variants (second panel). The substitution sites in TADAC-2 from the second round of evolution are shown as violet (same sites identified in TADAC-1) and orange (new sites) spheres in the TadA*8.20 structure (third panel). To generate the TADC-1 variants, eight substitutions from the second round of evolution were added to TACAC-3.154 (fourth panel). b, Another view of the TadA*8.20 structure (green) showing the substitution sites selected to generate TADAC-3 (violet and cyan spheres). The dashed circles highlight the three regions that we hypothesized to be critical for altering substrate tolerance. c, Substitutions incorporated into…

Fig. 4. CBE-Ts elicit robust C·G to T·A conversions in human cells at levels comparable to or higher than BE4 with a narrower editing window. a,b, Maximum C·G to T·A and A·T to G·C conversion in HEK293T cells transfected with mRNA encoding core CBE-T variants, plus controls, across eight targeted genomic loci via synthetic gRNAs at saturating (500 ng mRNA) (a) and subsaturating conditions (62.5 ng construct mRNA + 437.5-ng nontranslated carrier mRNA) (b). c, Percent change in C·G to T·A (left) and A·T to G·C (right) editing rates between CBE-T1 variants and ABE8.20 at each target site position (PAM = positions 21–23) across eight genomic sites tested (sites 1 to 8; Supplementary Table 3). d, Median C·G to T·A conversion at each target window position as specified on the x-axis, with position numbering defined as the PAM designated as positions 21–23. Values for color maps were determined from mRNA transfections at saturating conditions. Values and error reflect the mean and s.d. of n = 4 (saturating conditions) or 3 (subsaturating conditions) independent biological replicates performed on different days. Source data

Fig. 5. Guide-independent off-target evaluation of CABE-T and CBE-Ts. a, Odds ratio plot for C to T mutations relative to all other mutation types in cells edited with CABE-T3.155, CABE-T2.19, CBE-T1.14, CBE-T1.46 and CBE-T1.52, BE4, YE1, BE4-PpAPO and BE4-PpAPO-H122A compared to untreated clonally expanded cells, with black bars representing the median odds-ratio for that treatment group (P = 0.8359, 0.7473, 0.9476, 0.8089, 0.9751, 7.770 × 10−5, 0.9859, 0.00148 and 0.7473, respectively; one-sided Mann–Whitney U test). All n = 8 biologically independent single-cell expanded cell populations are shown for each condition. b, Odds ratio plot for A to G mutations relative to all other mutation types in cells edited with CABE-T33.155, CABE-T2.19, CBE-T1.14, CBE-T1.46 and CBE-T1.52 and ABE8.20 compared to untreated clonally expanded cells, with black bars representing the median odds-ratio for that treatment group (P = 0.1641, 0.4796, 0.5204, 0.8607, 0.5204 and 0.2527, respectively; one-sided Mann–Whitney U test). All n = 8 biologically independent single-cell expanded cell populations are shown for each condition. c, In vitro kinetics of A-to-I or C-to-U deamination of the same substrate…

Fig. 6. Evaluation of CABE-Ts and CBE-Ts in therapeutically relevant cell contexts. a, Gene-editing outcomes of primary human hepatocytes transfected with mRNAs encoding CBE-T1.46 and BE4 at three sites within the PCSK9 gene (left, P = 0.2882 (NS); right, P = 0.0017 (double asterisk)). Positional edit within the protospacer indicated. b, Evaluation of relative change in PCSK9 secretion between day 9 (collection time point) and day 0 (transfection time point) through ELISA. The targeted PCSK9 site is indicated on the x axis. P values are as follows: Q555X versus untreated, P = 0.001001 (double asterisk); E4 splice versus untreated, P = 0.001295 (double asterisk). c, Relative change in LDL-R present in supernatant between day 9 and day 0 assessed by ELISA. P values are as follows: Q555X versus untreated, P = 0.001116 (double asterisk), E4 splice versus untreated, P = .009481 (double asterisk). d, Gene-editing efficiencies from sgRNA screens in primary human T cells using CBE-Ts. X axis label indicates the targeted gene and target base within sgRNA. e,f, Percent C·G to T·A conversion (e) and surface protein loss (f) achieved by each base editor or control in multiplex-edited primary human T…

Extended Data Fig. 1. Crystal structure of TadA*8.20 in a complex with ssDNA containing the adenine transition-state analog 2-deoxy-8-azanebularine. a, The top panel depicts hydrolytic deamination of adenosine catalyzed by TadA. The bottom panel shows the hydration of the adenine analog, 2’-deoxy-8-azanebularine (d8Az), forming the transition-state analog that gets trapped in the active site by coordinating with zinc. b, The overall structure of the TadA*8.20 functional homodimer (chain A in dark green; chain B in light green) bound to ssDNA (yellow). c, The overall structure of TadA*8.20 monomer. The monomer (light green) contains five β-strands (β1 to β5) and six α-helices (α1 to α6) that fold into a single domain with a central five-stranded β-sheet surrounded by α-helices. The zinc ion is shown as a gray sphere. d, TadA*8.20 active site with ssDNA-d8Az transition-state analog bound. The catalytic zinc (gray) ion is coordinated to a histidine residue (H57), two cysteine residues (C87 and C90), and the d8Az transition-state analog (yellow). The hydrogen bonds are shown as gray dashed lines. e, The surface of TadA*8.20 dimer (dark and light green) bound to ssDNA (yellow), showing that ssDNA is bound…

Extended Data Fig. 2. Crystal structure of TADAC-1.17 in a complex with ssDNA containing the adenine transition-state analog 2-deoxy-8-azanebularine. a, The overall structure of the TADAC-1.17 functional homodimer (chain A in dark blue; chain B in slate blue) with ssDNA (yellow) bound. The substitutions (T17A, A48G, S82T, and A142E) relative to TadA*8.20 are shown in cyan spheres. b, The overall structure of TADAC-1.17 monomer (slate blue) in a complex with ssDNA (yellow). The monomer contains five β-strands (β1 to β5) and six α-helices (α1 to α6) that fold into a single domain with a central five-stranded β-sheet surrounded by α-helices. C and 5′ represent the C-terminus and 5′-end of the ssDNA, respectively. c, TADAC-1.17 active site with ssDNA-d8Az transition-state analog bound. The catalytic zinc ion (gray sphere) coordinates H57, C87, C90, and the d8Az transition-state analog (yellow). The T82 side chain (cyan) is near the catalytic E59 side chain (3.9-Å; cyan dashed line) and may play a role in deamination by donating/accepting a proton to/from E59. The residue A17 (cyan) is in α1-helix at the protein surface. The residue G48 (cyan) is in α2-helix at the substrate binding pocket. The H-bonds…

Extended Data Fig. 3. Crystal structure of TADAC-1.14 without ssDNA and structural comparisons with TadA*8.20. a, The overall structure of the TADAC-1.14 functional homodimer (chain A in dark yellow; chain B in light yellow). The substitutions (I49K, Y76I, and G112H) relative to TadA*8.20 are shown in magenta spheres. The residue H2 is disordered and not visualized in the structure. The zinc ion is shown as a gray sphere. The dashed line represents the partially disordered (A109 to A114) loop between β4 and β5 (R107 to V130). b, The overall structure of TADAC-1.14 monomer. The superposition between chains A (dark yellow) and B (light yellow) shows two different conformations for the loop between β4 and β5 (R107 to V130). c, TADAC-1.14 active site with water (red sphere) bound to the zinc ion (gray sphere). The residues H57, C87, and C90 coordinate with the zinc ion. The water molecule (red sphere) H-bonds (gray dashed lines) to the catalytic residue E59. In the first step of the TadA reaction, this water is added to the substrate to form a transition state specie (Extended Data Fig. 1a). d–f, Structural comparisons between substrate-free TADAC-1.14 (dark yellow) and ssDNA-bound TadA*8.20 (dark green…

Extended Data Fig. 4. Crystal structure of TADAC-1.19 without ssDNA and structural comparisons with TadA*8.20. a, Overall structure of the TADAC-1.19 functional homodimer (dark and light pink). E27G and I49N substitutions relative to TadA*8.20 are shown in orange spheres. Zinc ion is shown as a gray sphere. b, Overall structure of TADAC-1.19 monomer. C represents C-terminus. c, TADAC-1.19 active site with water (red sphere) bound to the zinc ion. H57, C87, and C90 coordinate with the zinc ion. The water molecule H-bonds (dashed lines) to the catalytic residue E59. d–h, Structural comparisons between TADAC-1.19 and TadA*8.20 structures. (d) Superposition between substrate-free TADAC-1.19 (pink) and ssDNA-bound TadA*8.20 (green, yellow) monomers, showing high structural similarity (RMSD of ~0.9-Å for all of the Cα atoms). The main structural differences are in α1-helix, the loop between α1 and β1, and C-terminal α5- and α6- helices. (e) TadA*8.20 has E27 side chain H-bonding (black dashed lines) to the main chains of A48, I49, and G50, and E27G substitution removes these interactions. To compensate for these critical contacts, TADAC1.19 places E25 at a similar position to the formerly occupied by E27 in…

Extended Data Fig. 5. Biochemical characterization of CABE-Ts and CBE-Ts. a, In vitro 24-hour end-point deamination assay to detect relative A-to-I (hAAG + APE1) and C-to-U (USERII) deamination by BE4, ABE8.20, CABE-Ts, and CBE-Ts programmed with the same guide and acting on the same dsDNA substrate. Endonuclease V, Endo V; human Alkyl Adenine DNA glycosylase, hAAG; Apurinic/apyrimidinic Endonuclease 1, APE1. Error bars represent standard deviation from the mean (plotted) of three independent replicates. Data were normalized to untreated sample. Endo V detects both A-to-I and C-to-U deamination. b, Left, Single-turnover rates of A-to-I or C-to-U deamination of the same dsDNA substrate by BE RNP. Right, Single-turnover rates of nicking by BE RNP in the same experiment as shown on left. Pseudo-first order kapp rate constants obtained by fitting to single exponential are reported (mean ± s.d., n = 3 independent replicates). Source data

Extended Data Fig. 6. Product purity of CBE-Ts relative to BE4. a, Product distribution of sequencing reads mapped as edited for core CBE-Ts and BE4, in which the specified target cytosine (highlighted in red) is mutated. Values were determined from transfection of HEK293T with mRNA at saturating conditions. Values and error bars reflect the mean and SD at n = 3 independent biological replicates performed on different days. b, Color map of maximum C·G to T·A conversions outside and 5′ of the protospacer target window. Target site positions where >0.8% C-to-T editing was detected for any editor are included. Values were determined from transfection of HEK293T with mRNA encoding editors or controls at saturating conditions, with n = 4 independent biological replicates performed on different days. Source data