Close this search box.

Domain-inlaid Nme2Cas9 adenine base editors with improved activity and targeting scope – Nature Communications

Development and characterization of domain-inlaid Nme2-ABEs

Structural analyses of Nme1Cas9 and Nme2Cas921, including the former in its cleavage-poised ternary complex, hinted that the inconsistent activity observed for the N-terminally fused Nme2Cas9 ABE variant (Nme2-ABE8e-nt) may be due to poor positioning of the deaminase domain relative to the predicted path of the displaced ssDNA target. We hypothesized that re-positioning of the deaminase closer to its target site may lead to increased editing efficiencies. A variety of Cas9 protein engineering approaches have been taken to alter the positioning of Cas9-domain fusions29,30,31. We opted for domain insertion, as several groups have shown that both Spy- and SauCas9 are amenable to this type of engineering22,23,24,25,26,27. Additionally, in the context of Cas9-BEs, the internal placement of a deaminase has been shown to decrease Cas9-independent off-target deamination, improving their safety profiles. We took a structure-guided approach21 to select eight domain insertion sites at surface-exposed loops (Fig. 1a). Our initial panel of inlaid Nme2-ABE8e effectors (Nme2-ABE8e -i1 through -i8) include the TadA8e deaminase flanked by twenty amino acid (AA) flexible linkers inserted into the RuvC-inactivated Nme2Cas9 nickase mutant (nNme2D16A) (Supplementary Fig. 1a). To streamline the initial screening of the Nme2-ABE8e constructs, we used a previously described HEK293T ABE mCherry reporter cell line that is activated upon A-to-G conversion17,32 (Fig. 1b). All domain-inlaid Nme2-ABE8e variants except Nme2-ABE8e-i4 activated the ABE reporter cell line above background levels, with several exhibiting efficiencies greater than that of the N-terminally fused version (Nme2-ABE8e-nt) (Fig. 1b).

Fig. 1: Design of domain-inlaid Nme2-ABEs.
figure 1

a Nme1Cas9/sgRNA/DNA ternary complex structure, PDB:6JDV. Nme2Cas9 is 98% identical to Nme1Cas9 outside of the WED and PAM-interacting domains. Black spheres represent N- and C-termini and colored spheres represent sites of domain insertion. Deaminase domain insertion sites (Nme2Cas9 aa numbers) are specified to the right, with colors matching the sites indicated in the structure. b Activities of Nme2-ABE8e constructs in mCherry reporter cells (activated upon A-to-G editing) after plasmid transfection, measured by flow cytometry (n = 3 biological replicates in technical duplicate; data represent mean ± SD). c A-to-G editing following transfection of Spy-ABE8e vs. Nme2-ABE8e plasmids, using PAM-matched, endogenous HEK293T genomic loci. The editing efficiency at the maximally edited adenine for each target was plotted. Editing efficiencies were measured by amplicon deep sequencing (n = 3 biological replicates; data represent mean ± SD). d Data from (c) were aggregated and replotted, with each data point representing the maximum A-to-G editing efficiency of an individual target site, as measured by amplicon deep sequencing (n = 3 biological replicates; data represent mean ± SEM). e Summary of mean A-to-G editing activities and editing windows for Spy- and Nme2-ABE8e constructs in HEK293T cells. Numbers provided for each position in the protospacer represent the mean A-to-G editing efficiency across eight PAM-matched endogenous target sites, as measured via amplicon deep sequencing (n = 3 biological replicates per target). Crossed-out boxes indicate that no adenine was present at the specified position in the target panel tested. Source data are provided as a Source Data file.

Domain-inlaid base editors have been shown to shift the editing window depending on the site of deaminase insertion24,26. We reasoned that the mCherry reporter, with its single target adenosine at nt 8 of the 24-nt protospacer (A8, counting from the PAM-distal end), would not accurately reflect the editing activity of all Nme2-ABE8e effectors. To provide a more comprehensive view of editing windows, we analyzed the Nme2-ABE8e effectors at 15 endogenous target sites within HEK293T cells via plasmid transfections. Among the target sites tested, we found all domain-inlaid variants improved overall editing efficiencies ranging from a 1.05- to 2.27-fold increase compared to Nme2-ABE8e-nt (describing cumulative average editing for the 15 protospacers, Supplementary Fig. 1c). Encouraged by the activities of the domain-inlaid Nme2-ABE effectors, we explored how they compare to Spy-ABE8e at eight dual PAM targeting sites that have NGGNCC PAM regions (Fig. 1c–e). For this experiment, we focused on Nme2-ABE8e-i1, -i7, and -i8 because they exhibited the highest activities and most varied editing windows. The inlaid Nme2-ABE8e effectors showed comparable activity as Spy-ABE8e at six out of eight of the target sites (Fig. 1c, d). Furthermore, editing hotspots were altered in the inlaid versions in a manner consistent with the sites of deaminase insertion (Fig. 1e). Specifically, Nme2-ABE8e-i1 favored editing of PAM-distal adenosines, whereas the -i7 and -i8 effectors exhibited more PAM-proximal editing windows (Fig. 1e, Supplementary Fig. 1b). These results demonstrated that positioning of the deaminase relative to the targeted R-loop can improve the efficiency and alter the editing window of Nme2-ABEs. Because of their editing efficiencies and distinct editing windows, our subsequent analyses of inlaid Nme2-ABE8e variants focused primarily on the -i1, -i7 and -i8 effectors.

Nme2Cas9 tolerates insertion of alternative deaminases

We next investigated whether Nme2Cas9 tolerates the insertion of cytosine deaminases (Nme2-CBEs). For these experiments, we again focused on the inlaid designs with insertion sites -i1, -i7, and -i8. We first turned our attention to the cytidine deaminase evoFERNY33, which has a similar size as TadA8e (161AA and 166AA respectively). To construct the Nme2-evoFERNY effectors we used the same architecture as the domain-inlaid Nme2-ABEs, with the addition of a C-terminal 10AA linker and a single uracil glycosylase inhibitor (UGI) domain (Supplementary Fig. 2a). In addition to evoFERNY, we also constructed Nme2-CBEs with the larger rAPOBEC1 (rA1, 228AA) cytidine deaminase (Nme2-rA1)3.

We tested the domain-inlaid Nme2-CBEs against their N-terminal fusion counterparts (Nme2-evoFERNY-nt or Nme2-rA1-nt), at three high-activity target sites in HEK293T cells by plasmid transfection. All Nme2-CBE effectors were functional at the genomic sites tested (Supplementary Fig. 2b). Like the domain-inlaid Nme2-ABEs, the Nme2-CBEs exhibited insertion-site-dependent shifts of editing hotspots for the target sites tested (Supplementary Fig. 2c). In addition, we noticed divergent editing patterns occurring at the same genomic loci between two analogous domain-inlaid Nme2-CBE effectors, some of which is likely attributed to the sequence specificity of the distinct cytidine deaminases (Supplementary Fig. 2b, 2c). These results demonstrate that Nme2Cas9 is a flexible scaffold for insertion of a variety of deaminase domains enabling C-to-T as well as A-to-G base editing.

Chimeric Nme2Smu-ABEs enable recognition of N4CN PAMs, increasing their target scope

The activity of Cas9-BEs are limited to editing in specific editing windows specified by the distance from the PAM2. Here, we sought to increase the targeting scope of Nme2-ABEs by altering their PAM recognition properties. Our group previously demonstrated that PID swapping between closely related Type II-C Cas9 orthologs could alter their PAM preferences19. We also discovered and characterized SmuCas9 from Simonsiella muelleri, and found that it has a minimal PAM of N4CN, despite having limited nuclease editing activity in HEK293T cells28. We reasoned that chimeric, domain-inlaid Nme2-ABEs with an SmuCas9 PID (Nme2Smu-ABEs) could alter the PAM preference from N4CC to N4CN, a four-fold increase in the number of target sites available for targeting by Nme2-ABEs.

After constructing the Nme2Smu-ABE8e effectors with -i1, -i7 and -i8 designs (Fig. 2a), we tested their activities at a panel of five N4CC and nine N4CD (D = A, G or T) PAM targets in HEK293T cells by plasmid transfection (Supplementary Fig. 3a). For this experiment, Nme2-ABE8e-i1 was used as a reference. For the N4CC targets, Nme2-ABE8e-i1 exhibited average editing of ~38% (describing the average maximally edited adenine across each protospacer in Supplementary Fig. 3a, 3b). All three Nme2Smu-ABE8e effectors were also active at the N4CC target sites, but with 1.7- to 2-fold reductions in overall activity compared to Nme2-ABE8e-i1 (Supplementary Fig. 3b). As expected, Nme2-ABE8e-i1 had minimal activity at the N4CD target sites (Supplementary Fig. 3b). By contrast, all three PID-swapped, domain-inlaid Nme2Smu-ABE8e effectors effectively installed A-to-G edits at the N4CD PAM target sites, though with varied efficiencies (Supplementary Fig. 3b). These results indicate that Nme2Smu-ABEs can target and install precise edits at sites with N4CN PAMs.

Fig. 2: PAM interacting domain (PID) chimeras expand the targeting scope of Nme2Cas9 effectors.
figure 2

a Schematic of chimeric Nme2-ABE8e-i1 with the SmuCas9 PID (left). Homology model of Nme2SmuCas9 based on PDB:6JDV using the SWISS-MODEL program (right). b A-to-G editing following plasmid transfection of WT or chimeric PID Nme2-ABE8e effectors into HEK293T cells with 183 integrated paired guide-target sites with N4CN PAMs. The editing efficiency at the maximally edited adenine for each target was plotted. Editing activities were measured by amplicon sequencing (n = 3 biological replicates; boxplots represent median and interquartile ranges; whiskers indicate 5th and 95th percentiles and the cross represents the mean). c Summary of mean A-to-G editing activities and editing windows for WT Nme2-ABE8e effectors at N4CC PAM guide-target library members or (d) chimeric PID Nme2-ABE8e effectors and eNme2-C constructs at N4CN PAM guide-target library members in HEK293T cells. Numbers provided for each position in the protospacer represent the mean A-to-G editing efficiency across the guide-target library members, as measured via amplicon deep sequencing (n = 3 biological replicates). Source data are provided as a Source Data file.

Editing windows and activities of Nme2- and Nme2Smu– ABE variants

To further investigate the editing characteristics of the domain-inlaid Nme2- and Nme2Smu-ABE8e effectors, we assessed their activities using a paired guide-target library approach27,34,35,36,37. The library consisted of 200 unique guide-target pairs cloned into a plasmid backbone flanked by Tol2 inverted terminal repeats enabling stable genome integration within HEK293T cells (Supplementary Fig. 4a; Supplementary Data 1, Oligonucleotides). Some guide-target pairs in the library corresponded to previously validated and analyzed sites19,38,39, whereas other were included for their preclinical therapeutic development potential. Following integration, we tested the panel of editors by plasmid transfection and subsequently sequenced the libraries at an average depth of ~1800 per library member (Supplementary Fig. 4b). For this experiment, we also included the recently evolved eNme2-C ABE8e variant as it has a relaxed N4CN PAM preference (akin to that of Nme2Smu-ABE8e) as well as increased activity in comparison to Nme2-ABE8e-nt38.

Consistent with results at endogenous HEK293T target sites, Nme2-ABE variants with a WT PID demonstrated robust activity at N4CC PAM targets, with minimal activity on N4CD target sites. In contrast, inlaid Nme2Smu-ABE8e effectors demonstrated robust activities at N4CN PAM target sites (Fig. 2b, c). For example, Nme2-ABE-i1 exhibited mean maximum editing efficiencies of ~15% at N4CC PAM targets and ~4% at N4CD PAM targets. By contrast, Nme2Smu-ABE-i1 had efficiencies of ~19% at N4CC PAM targets and ~29% at N4CD PAM targets. The observed editing window for eNme2-C across all library members spanned positions 6-14 (referring to activity >50% of the window maximum), with a center of position 9, in agreement with eNme2-C’s previously reported editing window and center38 (Fig. 2d, Supplementary Fig. 5). Consistent with our endogenous HEK293T target site data, domain-inlaid Nme2- and Nme2Smu-ABE8e effectors exhibited wide editing windows of 7-13 nucleotides and with a Tad8Ae insertion-site-dependent shift in editing window (Figs. 2c, d and Supplementary Fig. 5, 6). We next compared editing windows between WT and PID-swapped constructs at N4CC PAM targets. Although window centers were identical between WT and PID-swapped effectors with the same insertion site, Nme2Smu-ABE8e windows were smaller than those of Nme2-ABE8e’s at N4CC PAM targets (Supplementary Fig. 6). Observed window centers for ABE8e effectors with the -i1 insertion site fell between positions 7-8, whereas editing was centered around position 12 for -i8 effectors (Fig. 2c, Supplementary Figs. 5, 6).

Analysis of domain-inlaid Nme2-ABE8e specificity

We then sought to determine the specificities of the domain-inlaid Nme2-ABEs. Guide-dependent off-target editing is driven by Cas9 unwinding and R-loop formation at targets with high sequence similarity40. We previously demonstrated that Nme2-ABE8e-nt has a much lower propensity for guide-dependent off-target editing compared to Spy-ABE8e17. Using the most active inlaid variant (Nme2-ABE8e-i1) as a prototype, we examined guide-dependent specificity using a series of double-mismatch guides targeting the mCherry reporter, with Spy-ABE8e and Nme2-ABE8e-nt used for comparison. In all cases, the target adenosine was at the eighth nt of the protospacer (Fig. 3a, b). To account for differences in on-target activity (especially for Nme2-ABE8e-nt), we normalized the activities of the mismatched guides to that of the respective non-mismatched guide. Consistent with our previous results, Spy-ABE8e significantly outperformed Nme2-ABE8e-nt for on-target activity (Fig. 3a), but exhibited far greater activity with mismatched guides (Fig. 3b). Nme2-ABE8e-i1 activated the reporter with a similar efficiency as Spy-ABE8e (Fig. 3a), but with greater sensitivity to mismatches (Fig. 3b). Although the Nme2-ABE8e-i1 variant was less promiscuous than Spy-ABE8e, it exhibited higher activity with mismatched guides than Nme2-ABE8e-nt, illustrating trade-offs between on- and off-target editing efficiencies observed previously elsewhere40. We then assayed the mismatch sensitivity of the Nme2-ABE8e -i7 and -i8 effectors, to determine whether their preference for PAM-proximal editing windows would alter the mismatch sensitivity in comparison to the -nt and -i1 effectors for activating the reporter cell line. In this experiment, Nme2-ABE8e-i7 and -i8 exhibited mismatch sensitivities comparable to Nme2-ABE-nt, while retaining high on-target activity (Fig. 3a, b). A potential explanation for the increased sensitivity of -i7 and -i8 effectors at this site is that the impact of imperfect base pairing between a guide and target may become more apparent as the optimal editing window shifts away from the target adenine. A recent strategy using imperfectly paired guide RNAs to minimize bystander editing relied on a similar concept, providing some support for this hypothesis41.

Fig. 3: Specificities of domain-inlaid Nme2Cas9-ABE8e variants.
figure 3

a Comparison of on-target activity of transfected Spy-ABE8e and Nme2-ABE8e effectors in activating the ABE mCherry reporter, as measured by flow cytometry (n = 3 biological replicates; data represent mean ± SD). b Mismatch tolerance of Spy- or Nme2- ABE8e variants in ABE mCherry reporter cells at an overlapping target site positioning the target adenine for reporter activation at A8. Activities with single-guide RNAs carrying mismatched nucleotides as indicated (MM#, orange) are normalized to those of the fully complementary guides (ON, gray) (n = 3 biological replicates) for each effector, as indicated in the columns to the left. Heatmap data by column represent the normalized mismatched tolerance of the tested effectors. c Comparison of Nme2-ABE8e variants at previously validated genomic targets. A-to-G editing was measured following transfection with WT or chimeric, PID-swapped Nme2-ABE8e plasmids at endogenous HEK293T or mouse N2A genomic loci following transfection. The editing efficiencies at the maximally edited adenine for the On- or Off-target site for each effector were marked in the heatmaps. Off-target mismatches to the spacer are denoted with red nucleotides, whereas dashes correspond to a matched nucleotide. Editing activities were measured by amplicon sequencing (n = 3 biological replicates; data represent mean). Source data are provided as a Source Data file.

Following the mismatch sensitivity assay, we evaluated the specificity of domain-inlaid Nme2- and Nme2Smu-ABE8e’s against their respective ABE8e-nt variants at bona fide endogenous off-target sites. Although Nme2Cas9 off-target sites are rare due to its intrinsic accuracy in mammalian genome editing19, a few off-target sites have been identified for both nuclease and ABE variants via GUIDE-seq or in silico prediction. We selected four target sites for assessment, of which three had been validated as detectably edited off-target sites17,19,38 (Fig. 3c). In agreement with the mismatch sensitivity assay, Nme2-ABE8e variants with domain insertion at the -i1 position exhibited the greatest increase in off-target editing efficiencies, reaching above 1% at two out of the four targets and yielding the least favorable specificity ratio [on-target:off-target editing ratio] of ~23:1. Also in agreement with the mismatch sensitivity assay, the -i7 and -i8 effectors displayed increased accuracy in comparison to the -nt effectors (with specificity ratios of ~200:1 for -i7, ~170:1 for -i8, and ~82:1 for -nt) (Fig. 3c).

Next, we turned our attention towards guide-independent off-target editing. We hypothesized that similar to other domain-inlaid BE architectures, the internal positioning of the deaminase would limit the propensity for off-target nucleic acid editing that occurs in trans. We used the orthogonal R-loop assay with HNH-nicking SauCas9 (nSauD10A)20,26,42 to generate off-target R-loops and capture the guide-independent DNA editing mediated by Spy-ABE8e or the Nme2-ABE8e variants (-nt-and i1). We evaluated the on- and off-target activity of these ABE8e effectors by amplicon deep sequencing at the guide-targeted genomic site in addition to three SauCas9D10A-generated R-loops. We found that Nme2-ABE8e-i1 was less prone to editing the orthogonal R-loops compared to Nme2-ABE8e-nt and Spy-ABE8e (Supplementary Fig. 7a). To account for differences in on-target activities of the effectors, we reanalyzed the data by assessing the on-target: off-target editing ratio of each effector. Since Nme2-ABE8e effectors (-nt and -i1) have wider editing windows than Spy-ABE8e, we took the average editing activities across the respective windows of each effector for this target (protospacer positions 1-17nt for Nme2-ABE8e and 3-9nt for Spy-ABE8e), enabling a better comparison between the effectors (Supplementary Fig. 7b). In all cases, Nme2-ABE8e-i1 significantly outperformed Nme2-ABE8e-nt and Spy-ABE8e for guide-independent specificity at the orthogonal R-loops tested (Supplementary Fig. 7c). For this assay, we also investigated whether the TadA8eV106W mutant further increases the guide-independent DNA specificity with the Nme2-ABE8e-i1 architecture (Nme2-ABE8ev106w-i1). As observed previously with other effectors20, we observed increased specificity at all orthogonal R-loops with Nme2-ABE8ev106w-i1 compared to Nme2-ABE8e-i1, though the specificity increase was only significant for R-loop 3 (SSH2) (Supplementary Fig. 7c).

Nme2-ABEs enable correction of common Rett syndrome alleles

Having established several Nme2-ABE variants with varied editing windows and PAM preferences, we sought to demonstrate their use in a disease-relevant context. We previously showed that Nme2-ABE8e-nt can correct the second-most-common Rett syndrome mutation (c.502 C > T; p.R168X)17. The c.502 C > T mutation resides within a pyrimidine-rich region, where the target adenine is not accessible by well-established single-AAV-compatible ABEs (e.g. SauCas9, SauCas9-KKH, and SauriCas9). Although promising, these initial experiments revealed the incidence of bystander editing at an upstream adenine (A16), resulting in a missense mutation of unknown consequence (c.496 T > C; p.S166P).

With the increased activity and shifted editing windows of the domain-inlaid Nme2-ABE8e variants, we hypothesized that we could avoid bystander editing by selecting guide RNAs that shift its position outside the editing window (Fig. 4a). To test whether the domain-inlaid Nme2-ABEs could correct c.502 C > T while avoiding bystander editing, we electroporated mRNAs with synthetic guides into Rett patient-derived fibroblasts (PDFs) bearing the c.502 C > T allele. We first tested the editing activities of the various effectors with our previously validated guide17, denoted 502.G8. Consistent with our previous results, 502.G8 and Nme2-ABE8e-nt effectively corrected the target adenine (~19% efficiency), but with substantial (~10%) bystander editing (Fig. 4b). The domain-inlaid ABE8e variants were also active with 502.G8, with the -i7 and -i8 effectors exhibiting even higher bystander editing (~40–50%), likely reflecting their shifted editing windows, along with potential bystander editing of the wildtype allele (Fig. 4b). We next turned our attention to an additional guide, 502.G6, which places the target and bystander adenine at positions A16 and A22 of the protospacer respectively. With 502.G6, Nme2-ABE8e-nt performed poorly, exhibiting an average on-target editing efficiency of ~4% (Fig. 4b), whereas the -i1 and -i8 variants were somewhat more efficient (~14% and ~11% respectively). Importantly, 502.G6-guided correction by the domain-inlaid Nme2-ABE8e effectors occurred with undetectable editing at the A22 bystander (Fig. 4b).

Fig. 4: Correction of Rett Syndrome point mutations.
figure 4

a Schematic of a portion of MeCP2 exon 4, highlighting the (c.502 C > T; p.R168X) nonsense mutation in Rett patient-derived fibroblasts. b A-to-G editing of the MeCP2 502 C > T mutation in Rett patient fibroblasts in (a), measured by amplicon deep sequencing, with Nme2-ABE8e effectors delivered as mRNAs with synthetic sgRNAs (n = 3 biological replicates; data represent mean). Protospacer with target adenine (red), bystander adenine (orange), and PAM (bold, underlined). c Schematic of a portion of MeCP2 exon 4, highlighting the (c.916 C > T; p. R306C) missense mutation in Rett patient-derived fibroblasts. d A-to-G editing of the MeCP2 916 C > T mutation in Rett patient fibroblasts in (C), measured by amplicon deep sequencing, with Nme2-ABE8e effectors delivered as mRNAs with synthetic sgRNAs (n = 3 biological replicates; data represent mean). Protospacers are shown with target adenine (red), bystander adenine (orange), and PAM (bold, underlined) indicated. Source data are provided as a Source Data file.

Although the domain-inlaid Nme2-ABE8e effectors enabled correction of the c.502 C > T mutant with undetectable bystander editing, it came at the consequence of lowered on-target activity when using 502.G6 compared to 502.G8. We thus turned to the Nme2Smu-ABE8e variants which allowed targeting the c.502 C > T with an additional four sgRNAs bearing non-canonical N4CN PAMs, two of which (502.G9, 502.G10) placed the target and bystander adenines in favorable positions. Amplicon sequencing of the Nme2Smu-ABE8e edited Rett PDF cells revealed that all the editors were inefficient at installing edits with sgRNA 502.G9 (Source data). Conversely, sgRNA 502.G10 corrected the mutation more efficiently (~18% and ~16% for the -i7 and -i8 effectors respectively) than 502.G6 while avoiding bystander editing (Fig. 4b). We also tested editing of Rett PDF cells bearing the c.916 C > T (p.R306C) missense mutation (Fig. 4c). Using the Nme2Smu-ABE8e variants at an N4CT PAM we were also able to induce correction of this mutation. The -nt and -i1 effectors had average on-target editing rates of ~20%, ~22% respectively with bystander editing below 1% (Fig. 4d). We also tested an additional guide 916.G3 with an N4CC PAM for the Nme2-ABE effectors, and although the on-target editing rate (~17% efficiency) was somewhat comparable to the PID swapped -i1 inlaid ABE variant, bystander editing was more substantial at this site (~5% efficiency) (Fig. 4d).

Installation of therapeutic edits via splice site disruption with Nme2- and Nme2Smu-ABEs

We next explored the installation of additional therapeutically relevant edits by splice site disruption. ABEs can mediate exon skipping by editing consensus splice donor (SDS) or acceptor (SAS) sites, enabling gene disruption or ORF alteration without the introduction of double-strand breaks. These approaches are particularly advantageous for ABEs with wide editing windows, as the issue of bystander editing is minimized due to their presence within the intron or skipped exon.

First, we tested the ability of Nme2Smu-ABE8e variants and eNme2-C editors to edit the SDS of DMD exon 50, an approach enabling DMD Δexon 51 reading frame restoration and a potential therapeutic approach for ~8% of DMD patients (Supplementary Fig. 8a)43. In this experiment, the domain-inlaid Nme2Smu-ABEs performed similarly to eNme2-C depending on the target site, reaching SDS editing rates between 40-45% in HEK293T cells following plasmid transfection (Supplementary Fig. 8b).

Next, we tested the activities of eNme2-C, Nme2- and Nme2Smu-ABE8e variants at 12 sites in Neuro2A cells targeting either the SAS or SDS of mouse Cln3 exon 5. Deletion or skipping of Cln3 exon 5 has been demonstrated to ameliorate disease phenotypes in a validated Cln3 Δex7/8 mouse model44,45 (Supplementary Fig. 8c). We observed on-target editing up to 15%, with domain-inlaid variants outperforming eNme2-C with 6 out of the 7 guides that exhibited significant activity (Supplementary Fig. 8d).

Domain-inlaid Nme2-ABE8e enables in vivo base editing with a single AAV vector

We previously developed and optimized a compact AAV design that enables all-in-one delivery of Nme2-ABE8e-nt with a sgRNA for in vivo base editing17. At 4996 bp, the cassettes harboring the domain-inlaid Nme2-ABE8e variants and a guide RNA are also within the packaging limit of some single AAV vectors, allowing us to test whether they outperform Nme2-ABE8e-nt in an in vivo setting. For our in vivo experiments we designed AAV genomes containing Nme2-ABE8e-nt, Nme2-ABE8e-i1 or Nme2-ABE8eV106w-i1 with an sgRNA targeting the Rosa26 locus (Fig. 5a).

Fig. 5: In vivo editing with AAV9.Nme2-ABE8e-nt vs. –i1 vs. –i1V106W.
figure 5

a Schematic of the AAV constructs for the Nme2-ABE8e effectors. b Editing with AAV Nme2-ABE vectors in mouse liver (left) and striatum (right). Left, quantification of the editing efficiency at the Rosa26 locus by amplicon deep sequencing using liver genomic DNA from mice that were tail-vein-injected with the indicated vector at 4 × 1011 vg/mouse (n = 3 mice per group; data represent mean ± SD). Nme2-ABE8e-i1 (p = 0.04), Nme2-ABE-i1V106W (p = 0.015). Right, quantification of the editing efficiency at the Rosa26 locus by amplicon deep sequencing using striatum genomic DNA from mice intrastriatally injected with the indicated vector at 1 × 1010 vg/side (n = 3 mice per group; data represent mean ± SD). One-way ANOVA analysis: ns, p > 0.05; *, p ≤ 0.05. c Protospacer of the Rosa26 on-target site (“ON”) and a previously validated Nme2-ABE8e off-target site (OT1, “OFF”). Adenines are in red, mismatches in OT1 have asterisks, and PAM regions are bold and underlined. The bar graph shows quantification of A-to-G edits in amplicon deep sequencing reads at the OT1 site using liver genomic DNA from mice tail-vein injected in (b), with vectors indicated in the inset (n = 3 mice per group; data represent mean ± SD). Source data are provided as a Source Data file.

We conducted two in vivo editing experiments with 9-week-old mice. First, we focused on systemic [intravenous (i.v.)] injection and editing in the liver, whereas the second experiment tested editing in the brain after intrastriatal injection. In both cases, mice were sacrificed 6 weeks after their respective injections and editing was quantified by amplicon sequencing. Within the liver, Nme2-ABE8e-i1 and Nme2-ABE-i1V106W had editing efficiencies of ~49% (p = 0.015) and ~46% (p = 0.04) respectively, outperforming Nme2-ABE8e-nt (editing efficiency ~34% at A6 of the Rosa26 target site), (one-way ANOVA) (Fig. 5b). Within the striatum the trend continued, with both Nme2-ABE8e-i1 and Nme2-ABE-i1V106W exhibiting improved editing activities (~37% and ~34% at A6 of Rosa26), compared to Nme2-ABE-nt (~25%), albeit this improvement did not reach statistical significance (p = 0.26 and 0.5, for Nme2-ABE8e-i1 and Nme2-ABE-i1V106W respectively) (Fig. 5b).

We next sought to determine whether the boost in on-target activity in the liver was also accompanied by increased sgRNA-dependent off-target activity. The Rosa26 sgRNA used in this study is unusual among Nme2Cas9 guides in having a previously validated off-target site (Rosa26-OT1)19. We conducted amplicon sequencing at Rosa26-OT1 on genomic DNA extracted from the mouse livers used for our on-target analysis. We found that both Nme2-ABE8e-i1 and the V106W variant increased off-target A-to-G editing (up to ~7% and ~5% respectively) compared to Nme2-ABE8e-nt (~0.2%) (Fig. 5c). Collectively these results demonstrate that the increased activity of the domain-inlaid ABEs can translate to an in vivo setting, though this increase can come at the cost of increased off-target editing.