A bacterial cytidine deaminase toxin enables CRISPR-free mitochondrial base editing

Abstract

Bacterial toxins represent a vast reservoir of biochemical diversity that can be repurposed for biomedical applications. Such proteins include a group of predicted interbacterial toxins of the deaminase superfamily, members of which have found application in gene-editing techniques^1,2. Because previously described cytidine deaminases operate on single-stranded nucleic acids³, their use in base editing requires the unwinding of double-stranded DNA (dsDNA)-for example by a CRISPR-Cas9 system. Base editing within mitochondrial DNA (mtDNA), however, has thus far been hindered by challenges associated with the delivery of guide RNA into the mitochondria⁴. As a consequence, manipulation of mtDNA to date has been limited to the targeted destruction of the mitochondrial genome by designer nucleases^9,10.Here we describe an interbacterial toxin, which we name DddA, that catalyses the deamination of cytidines within dsDNA. We engineered split-DddA halves that are non-toxic and inactive until brought together on target DNA by adjacently bound programmable DNA-binding proteins. Fusions of the split-DddA halves, transcription activator-like effector array proteins, and a uracil glycosylase inhibitor resulted in RNA-free DddA-derived cytosine base editors (DdCBEs) that catalyse C•G-to-T•A conversions in human mtDNA with high target specificity and product purity. We used DdCBEs to model a disease-associated mtDNA mutation in human cells, resulting in changes in respiration rates and oxidative phosphorylation. CRISPR-free DdCBEs enable the precise manipulation of mtDNA, rather than the elimination of mtDNA copies that results from its cleavage by targeted nucleases, with broad implications for the study and potential treatment of mitochondrial disorders.

Extended Data Fig. 1 |. Analysis of the bactericidal activity of DddA and its activity against dsDNA and RNA substrates.

a, Genomic context of dddA (purple) and dddI_A (blue) in B. cenocepacia H111. b, Viability of B. cenocepacia ΔdddA ΔdddI_A (recipient) over time during competition with B. cenocepacia donor strains carrying wild-type dddA_tox or dddA_tox^E1347A. Values and error bars represent the mean ± s.d. of three technical replicates. The experiment was repeated three times with similar results. c, α-VSV-g western blot analysis of total cell lysates of E. coli expressing the indicated deaminases tagged with VSV-G epitope. RNAP-β was used as a loading control. Results are representative of n = 2 independent biological replicates. d, In vitro DNA cytidine deamination assays using double-stranded 36-nt DNA substrates containing AC, TC, CC, and GC with a FAM fluorophore on the forward (A) or reverse (B) strand. Deamination activity results in a cleaved product (P). Images are representative of n = 2 independent biological replicates. e, f, Poisoned primer extension assay to detect deamination of cytidine in single-stranded (e) or double-stranded (f) RNA substrates. Images are representative of n = 2 independent biological replicates. A mix of RNA substrates containing the sequences GUCG or GUUG at the indicated ratios were incubated with purified DddA_tox and reverse transcriptase. Primer extension was performed in reactions with ddGTP to terminate primer extension at cytidines. Cytidine deamination yields the 31-mer product.

Extended Data Fig. 2 |. DddA_tox deaminates cytidines in bacteria with strong sequence context preference.

a, Number of SNPs from the indicated nucleotide classifications observed in E. coli Δudg following intoxication with DddA_tox or DddA_tox(E1347A). b, c, The position of SNPs on the chromosome of E. coli Δudg isolates intoxicated with DddA_tox (b) or DddA_tox(E1347A) (c). SNPs above the line indicate C-to-T transitions on the plus strand; SNPs below indicate C-to-T transitions on the minus strand. Other mutations are represented on the plus strand. Sequencing coverage was 203–265-fold. d, Deamination assay on DddA_tox with double-stranded DNA substrates containing a single C with different nucleotides (A, T, C or G) at the position immediately 5′ of the C (red) (S, substrate; P, product). Images are representative of n = 3 independent biological replicates.

Extended Data Fig. 3 |. Base-editing efficiencies and indel frequencies of all DddA_tox splits in HEK293T cells.

a–h, Each split was assayed in the aureus-N and aureus-C orientation (see Fig. 2b) across spacing region lengths of 12-bp (a), 17-bp (b), 23-bp (c), 28-bp (d), 33-bp (e), 39-bp (f), 44-bp (g) and 60-bp (h). Cells were collected 3 days post-transfection for DNA sequencing. Colours reflect the mean of n = 2 independent biological replicates.

Extended Data Fig. 4 |. TALE–split DddA_tox proteins mediate efficient base editing in nuclear DNA of U2OS cells.

a, Left–G1333-DddA_tox-N and Right–G1333-DddA_tox-C bind DNA sequences within CCR5. Target cytosines are shown in purple and TALE binding sites are shown in blue. Two copies of UGI proteins (2×-UGI) were fused to the N- or C terminus through a 2- or 16-amino acid linker. Editing efficiencies and indel frequencies for the possible combinations of UGI positions and linker lengths are shown. In the absence of UGI protein, only C₉-to-T₉ edit was observed. b, Architecture of nuclear-targeting CCR5-DdCBE (see Fig. 3c for optimized DdCBE architecture targeting mtDNA). Target cytosines are shown in purple. c, Editing efficiencies and indel frequencies of cells treated with CCR5-DdCBE and ND6-DdCBE 3-days-post transfection are shown. Dead-DdCBEs containing the inactive DddA_tox(E1347A) mutant were used as negative controls. d, Outcomes among edited alleles in which the specified target C is mutated are shown for the indicated base editor. Values and error bars in a, c and d reflect the mean ± s.d. of n = 3 independent biological replicates.

Extended Data Fig. 5 |. Unoptimized mitoTALE–split DddA_tox fusions mediate modest editing of mitochondrial ND6 in HEK293T cells.

a, Architectures of non-UGI containing ND6-mitoTALE–DddA_tox fusion pair. DddA_tox was split at G1333 or G1397, with each half fused to either the left TALE or the right TALE. TALEs bind to mtDNA sequences (blue) that flank a 15-bp spacing region in mitochondrial ND6. Target cytosines are shown in purple. The last TALE repeat (*) did not match the reference genome (see Supplementary Table 4). b, mtDNA editing efficiencies of mitoTALE–DddA_tox pairs in the listed split orientations. The dashed line is drawn at 0.1%. Values and error bars reflect the mean ± s.d. of n = 3 independent biological replicates.

Extended Data Fig. 6 |. DdCBE editing in HEK293T cells persist over multiple divisions while maintaining cell viability and mitochondrial DNA integrity.

a–e, Editing efficiencies for optimized ND6-DdCBE (a), MTND5P1-DdCBE (denoted ND5.1-DdCBE) (b), MTND5P2-DdCBE (denoted ND5.2-DdCBE) (c), ATP8-DdCBE (d) and BE2max and BE4max (e) are shown for each time point. C•G-to-T•A conversions at protein-coding genes that generate missense mutations (green) of the putative amino acid (red) are shown. f, Western blots of ND6-, ND5.1-, ND5.2- and ATP8-DdCBE at various time points. The right halves were FLAG-tagged and the left halves were HA-tagged. Day 3 images are representative of n = 3 independent biological replicates; n = 1 for day 6 and day 12 images (see Supplementary Data 3 for uncropped images and fluorescent tagging of each half). Nuclear β-actin was used as a loading control. g, Cell viability was measured by recording the luminescence at the indicated time points. Luminescence values were normalized to the untreated control. h, DNA gel of PCR-amplified mtDNA captured as two amplicons (red). Images are representative of n = 3 independent biological replicates (see Supplementary Data 4 for uncropped images). i, mtDNA levels of DdCBE-edited cells were measured by qPCR relative to untreated cells. Values and error bars in a–e, g and i reflect the mean ± s.d. of n = 3 independent biological replicates. For a–e, asterisks indicate significant editing based on a comparison between indicated time points. *P < 0.05 and **P < 0.01 by Student’s two-tailed paired t-test. Individual P values are listed in Supplementary Table 7.

Extended Data Fig. 7 |. Stalling mtDNA replication impairs mitochondrial base editing in human cells.

a, Schematic of experimental design. Addition of doxycycline (Dox) induces the stable expression of a dominant-negative mutant of DNA polymerase-gamma containing a D1153A substitution (POLGdn) in a HEK293-derived cell line. Total cell lysate was collected at indicated time points for western blotting of POLGdn in n = 3 independent biological replicates. b, mtDNA levels of uninduced (no Dox) and induced (+Dox) cells treated with indicated DdCBE 48 h post-transfection. mtDNA levels were measured by qPCR and normalized to uninduced cells without DdCBE treatment. c, Editing efficiencies of indicated DdCBE in uninduced and induced cells 48 h post-transfection. All values and error bars in b and c reflect the mean ± s.d. of n = 3 independent biological replicates.

Extended Data Fig. 8 |. Effect of DdCBE editing on mitochondrial function and mtDNA homeostasis.

a, mtDNA levels of ND4-edited cells measured by qPCR relative to cells treated with dead ND4-DdCBE. b, mtRNA levels of ND4-edited cells measured by reverse transcription-qPCRrelative to cells treated with dead ND4-DdCBE. c–f, Confirmation of editing by Sanger sequencing and OCR of cells treated with ND5.1-DdCBE (c), ND5.2-DdCBE (d), MTND5P3-DdCBE (denoted ND5.3-DdCBE) (e) and ND1-DdCBE (f). Untreated cells were used as controls. All cells were collected 6 days post-transfection. For all Sanger sequencing plots, n = 3 independent biological replicates. All values and error bars shown in a, b and OCR plots in c–f reflect the mean ± s.e.m. of n = 3 independent biological replicates. For a and b, Student’s unpaired two-tailed t-test was applied. NS, not significant (P > 0.05).

Extended Data Fig. 9 |. Off-target editing activity of DdCBEs in nuclear DNA of HEK293T cells.

a–c, The on-target editing site in mtDNA and the corresponding nuclear DNA sequence with the greatest homology are shown for ND6-DdCBE (a), ND5.1-DdCBE (b) and ND4-DdCBE (c). TALE binding sites begin at N₀ and are shown in blue. Target cytosines are in purple. Nucleotide mismatches between the mtDNA and nuclear pseudogene are in red. Editing efficiencies are measured by targeted amplicon sequencing 3 days post-transfection (a, b) or six days post-transfection (c) (see Methods for primer sequences). Each amplicon was sequenced at 44,000× coverage. All values and error bars reflect the mean ± s.d. of n = 3 independent biological replicates. Student’s unpaired two-tailed t-test was applied. NS, not significant (P > 0.05).

Extended Data Fig. 10 |. TALE arrays need to bind to mtDNA sequences positioned in close proximity to reassemble catalytically active DddA_tox for off-target editing.

a, The identities and relative binding positions of each mismatched (MM) TALE–DddA_tox half is shown. MM-1 and MM-2 contain a TALE-bound DddA_tox half and a TALE-free DddA_tox half. MM-3 and MM-4 contain DddA_tox halves fused to TALE repeat arrays that bind to distant regions in mtDNA. ND6-Right TALE contains a permissive N-terminal domain (see Supplementary Table 4). b, The average percentage of genome-wide C•G-to-T•A off-target editing in mtDNA by indicated DdCBE and MM pairs are shown. The dashed line represents the percentage of endogenous C•G-to-T•A conversions in mtDNA as measured in the untreated control. Values and error bars reflect the mean ± s.e.m. of n = 3 independent biological replicates.

Extended Data Fig. 11 |. Predicted effects of off-target SNVs on mitochondrial DNA sequence and protein function.

a, Classification of off-target SNVs into noncoding or coding mutations. Mutations occurring in protein-coding regions of mtDNA were further categorized into synonymous, missense or nonsense mutations. b, For nonsynonymous SNVs, SIFT was used to predict the effect of these mutations on protein function. High- or low-confidence calls (indicated in parentheses) were made according to the standard parameters of the prediction software. c, Editing efficiencies of selected off-target TC bases in the indicated sequence contexts are shown. HEK293T cells were treated with the indicated DdCBE and collected 3 days post-transfection for DNA sequencing. Values and error bars reflect the mean ± s.d. of n = 3 independent biological replicates.

Fig. 1 |. DddA is a double-stranded DNA cytidine deaminase that mediates T6SS-dependent interbacterial antagonism.

a, Domains of full-length DddA. PAAR, proline-alanine-alanine-arginine; RHS, rearrangement hotspot; Tox, toxin domain. b, Competitiveness of the indicated donor B. cenocepacia strains (D) towards the B. cenocepacia ΔdddAΔdddI_A recipient strain (R). c, Viability of E. coli populations expressing the indicated deaminases, induced at 300 min (arrow). A3G, APOBEC3G; Cdd, E. coli cytidine deaminase; TadA, tRNA adenosine deaminase A; cfu, colony-forming units. d, Crystal structure of DddA_tox (purple) complexed with DddI_A (grey). e, Structural alignment of DddA_tox (purple) and APOBEC3G (white). The intervening loop of DddA_tox that is absent in APOBEC3G is shown in orange. f, g, In vitro cytidine deamination assays using a single-stranded (f) or double-stranded (g) 36-nt 6-carboxyfluorescein (FAM)-labelled DNA substrate (S), which contains AC, TC, CC and GC as indicated in g. Cytidine deamination leads to products (P) with increased mobility. A3A, APOBEC3A. Gels are representative of three replicates. h, Mutation frequency in E. coli strains expressing DddA_tox or catalytically inactive DddA_tox(E1347A). pBAD24::udg was used for complementation of Δudg (+udg). Values are derived from eight independent biological replicates. Rif^R, rifampicin resistant colonies. i, Probability sequence logo of the region flanking mutated cytosines in five E. coli Δudg isolates serially exposed to a low level of DddA_tox. Values and error bars reflect mean ± s.d. of n = 4 (in b) or n = 3 (in c) independent biological replicates. *P < 0.0001; NS, not significant (P > 0.05) by Student’s unpaired two-tailed t-test.

Fig. 2 |. Non-toxic split-DddA_tox halves reconstitute activity when co-localized on DNA in HEK293T cells.

a, DddA_tox was split at the peptide bond between each labelled amino acid and the following residue. b, Architectures of split-DddA_tox–Cas9 fusions. DddA_tox-N and DddA_tox-C contain the N terminus and C terminus of DddA_tox, respectively. Two fusion orientations (aureus-N or aureus-C) are possible for a given split. sgRNA, single guide RNA. c, Fusions of split-DddA_tox halves to orthogonal Cas9 variants enable reassembly of active DddA_tox, without creating non-functional homodimers. PAM, protospacer adjacent motif. d, Heat maps showing C•G-to-T•A conversion and indel frequencies for G1333 and G1397 splits at the nuclear EMX1. The split orientations and positions of dSpCas9 (pink) and SaKKH-Cas9(D10A) (blue) protospacers are shown. Colours reflect the mean of n = 2 independent biological replicates.

Fig. 3 |. TALE–split DddA_tox fusions for mitochondrial base editing in HEK293T cells.

a, Top, candidate TALE–split DddA_tox fusions to target MT-ND6. Target cytosines and TALE-binding sites are shown in purple and blue, respectively. Bottom, MT-ND6 editing efficiencies from fusions containing 1×- or 2×-UGI at the N- or C terminus 3 days post-transfection. b, Fluorescence imaging of HA- and FLAG-tagged halves of UGI–TALE–split DddA_tox and TALE–split DddA_tox–UGI pairs in HeLa cells 24 h after plasmid transfection. Mitochondrial localization was followed using MitoTracker (magenta). Scale bars, 10 μm. Images are representative of 3 independent biological replicates. c, Top, optimized DdCBE architecture containing one UGI fused to the C terminus of each TALE–split DddA_tox fusion. Bottom, editing and indel frequencies at MT-ND6 (mtDNA) and EMX1 (nuclear DNA) 3 days post-transfection.BE2max, rAPOBEC1–dSpCas9–2×-UGI. For a and c, the last TALE repeat (*) does not match the reference genome(see Supplementary Table 4). d, Outcomes among edited alleles in c are shown for the indicated DdCBE variants. e, Frequencies of MT-ND6 alleles in c. Edited cytosines are boxed. Values and error bars for a, c–e reflect the mean ± s.d. of n = 3 independent biological replicates.

Fig. 4 |. DdCBE editing at five mtDNA genes in HEK293T cells.

a–g, Target spacing regions and the split DddA_tox orientation that resulted in the highest editing efficiencies are shown for ND1-DdCBE (a), ND5.1-DdCBE (b), ND4-DdCBE (c), ND5.2-DdCBE (d), ND5.3-DdCBE (e), ATP8-DdCBE (f) and ND2-DdCBE (g). Editing efficiencies are shown on the right. Genomic DNA was collected 3 days (b, d, f) or 6 days (a, c, e, g) post-transfection. h, DdCBE orientations and corresponding approximate windows (red and blue) within which target cytosines are edited. i, Mitochondrial DNA editing efficiencies in untransformed human primary fibroblasts 5 days after nucleofection of mRNA encoding the DdCBEs shown; n = 2 independent biological replicates. j, k, Oxygen consumption rate (OCR) (j) and relative values of respiratory parameters (k) in ND4-DdCBE-treated HEK293T cells. FCCP, carbonyl cyanide-4 (trifluoromethoxy) phenylhydrazone. l, Blue-native PAGE of HEK293T mitochondrial lysates treated with ND4-DdCBE, visualized with antibodies against the indicated subunits of mitochondrial complexes; n = 3 independent biological replicates. m, The activities of complex I (left) and complex IV (right). mOD: absorbance at optical density of 450 nm. Values and error bars in a–g, j, k and m reflect the mean ± s.d. of n = 3 independent biological replicates. *P < 0.05; **P < 0.01; NS, not significant (P > 0.05) by Student’s unpaired two-tailed t-test.

Fig. 5 |. Mitochondrial genome-wide off-target DNA editing by DdCBEs.

a, HEK293T cells were transfected with plasmids encoding active DdCBE, dead-DdCBE or TALE-free MTS–split DddA_tox–UGI. The average coverage of each base was 5,100- to 9,900-fold (see Supplementary Data 1). b, Average percentage of genome-wide C•G-to-T•A off-target editing in mtDNA for each DdCBE and for controls. The vertical line represents the percentage of endogenous C•G-to-T•A conversions in mtDNA in untreated cells. Values and error bars reflect the mean ± s.e.m. of n = 3 independent biological replicates. c, Sequence logos generated from off-target C•G-to-T•A conversions by each indicated DdCBE. Bits reflect sequence conservation at a given position.