Using ABE9 and SpRY Cas9 nickase to precisely create mouse models without unwanted mutations

Among all these options, ABE9, which was derived from ABE8e by introducing two point mutations N108Q and L145T in the deaminase domain, appeared to be the most suitable due to its narrowest editing window to date, located between protospacer positions 5–6, and its strong properties for reducing off-target effects in DNA and RNA²⁴. To this end, we compared ABE8e and ABE9 at four previously reported ABE target sites in HEK293T cells to confirm their properties (Supplementary Fig. 1). Indeed, the editing window of ABE9 was specific to positions A5 and A6, with some minimal activity at A4, whereas ABE8e was highly active between A3 and A8 with an extended activity shoulder to A13. Overall, the average activity at A5 was reduced from 43.6 ± 5.9% (mean ± s.d.%) in ABE8e to 34.5 ± 8.8% in ABE9, corresponding to an average reduction of 21% in editing activity, in line with previous work²⁴ (Supplementary Fig. 1a–c). Interestingly, indel frequencies in HEK293T cells were locus-dependent: ABE9 showed higher indel rates than ABE8e at two of the four benchmark sites, while exhibiting reduced bystander C-to-T editing at three sites and achieving greater overall precision across all four sites (Supplementary Fig. 1d–f). Given that our primary goal is to generate precise point-mutation models with minimal bystander outcomes, we prioritised this improved precision over maximal bulk editing efficiency and therefore selected ABE9 for subsequent testing at our loci of interest.

To address this limitation, we next considered alternative Cas9 variants that have different or relaxed PAM requirements, including SaCas9, xCas9, SpCas9-NG, SpCas9-VQR, and SpRY^25,41–46. Instead of an “on-demand” context-dependent approach, we developed a versatile editing strategy that can be applied to all four loci of interest by utilising a single construct and the appropriate sgRNA. SpRY²⁵ and SpRYc⁴⁷ have emerged as potentially universal Cas9 variants with PAMless targetability across NNN PAMs (Table 1). SpRY stands out for having the broadest and most relaxed PAM compatibility and has previously been tested with ABE8e variants^34–36. We fused a SpRY nickase (in the following simply “SpRY”, bears these substitutions compared to SpCas9 nickase: A61R/L1111R/D1135L/S1136W/G1218K/E1219Q/N1317R/A1322R/R1333P/R1335Q/T1337R) to the TadA domains of ABE8e or ABE9 to streamline the experimental workflow and maximise the potential for successful genome editing outcomes (Fig. 2b). These SpRY base editors can, in theory, effectively target spacers with NRN (G or A) PAMs, and to a lesser degree, NYN (T or C) PAMs²⁵. Since the compatibility of deaminase domains with Cas9 variants is not universal, and their efficiency and editing activity can vary with the Cas9 scaffold used^18,48,49, an initial evaluation was necessary. We therefore compared ABE9-SpRY with ABE8e-SpRY in N2a cells using plasmid lipofection with the respective sgRNAs at four loci under identical transfection and enrichment conditions (Fig. 2c; Methods).

Deep sequencing of transfected N2a cells revealed that the desired editing occurred at all four targets for ABE8e-SpRY, ranging from 14.1 ± 0.7% to 60.6 ± 1.2%. In contrast, only three targets showed targeted A-to-G editing with ABE9-SpRY, ranging from 6.5 ± 0.9% to 13.2 ± 1.5% across all tested sites (Fig. 2d). While ABE8e-SpRY displayed modest editing of 14.1 ± 0.7% at the Tpc2-K188 locus, no desired editing was noted with ABE9-SpRY. A closer examination of adenine editing profiles across the protospacers confirmed a consistent trend across all four loci: ABE8e-SpRY exhibited a broad activity window across multiple adenines (with locus-dependent shoulders extending to more distal positions), whereas ABE9-SpRY showed a more confined editing profile with markedly reduced bystander activity (Fig. 2e). For example, at the Tpc1-I486 locus, ABE8e-SpRY editing extended to distal positions (13.5 ± 0.5% at A2 and 20.8 ± 0.6% at A12), whereas ABE9-SpRY activity was largely restricted to a narrower region. The highest ABE8e-SpRY activities were observed at A5 (Tpc2-K188, Tpc2-L249, Trpm4-L903) and A6 (Tpc1-I486), with 30.8 ± 0.5%, 26.1 ± 6.5%, 41.9 ± 12.8%, and 61.6 ± 12.7% A-to-G editing, respectively. By contrast, ABE9-SpRY demonstrated the enhanced precision previously observed for ABE9, showing a narrower activity window with reduced conversion at flanking adenines across loci; at the Tpc1-I486 locus, activity spanned from A4 to A9, with 9.1 ± 1.3% and 3.3 ± 0.5% editing at A4 and A9, respectively. We also measured the highest efficiencies for ABE9-SpRY at A5 and A6, respectively, with 6.5 ± 0.9%, 12.8 ± 0.8%, and 13.7 ± 1.7% efficiency for Tpc2-L249, Trpm4-L903, and Tpc1-I486, respectively. In N2a cells at four disease-relevant loci, this narrower editing window of ABE9-SpRY also led to a larger fraction of edited alleles containing no bystander mutations compared to ABE8e-SpRY (Fig. 2f), as indicated by the substantially higher product purity, ranging from 50.9 ± 2.4% to 98.6 ± 0.6%, compared to 1.1 ± 0.1% to 69 ± 0.8% for ABE8e-SpRY (Fig. 2g). Product purity calculations were restricted to our predefined CRISPResso2 quantification window (spacer − 7 nt to spacer + PAM (+ 3 nt); Methods). Furthermore, as ABE9-SpRY showed lower overall A-to-G conversion at these loci (Fig. 2d–e), differences in product purity may reflect both window narrowing and reduced activity. Notably, the desired Tpc1^I486T and Trpm4^L903P mutations could not be introduced without bystanders using ABE8e-SpRY (0.7 ± 0.1% and 0.5 ± 0.3%), whereas ABE9-SpRY achieved 6.7 ± 0.4% and 11.9 ± 1.4% precise editing at these sites, respectively. Lastly, ABE9-SpRY demonstrated reduced indel frequencies compared to ABE8e-SpRY across all four sites, peaking at 0.9 ± 0.2% (Tpc1-I486) (Fig. 2h). Since these in vitro data were obtained from a single mouse neuroblastoma cell line and four loci, further testing is needed to determine if the findings apply to other mouse cell types and genomic regions.

Taken together, these in vitro results in N2a cells indicated that ABE9-SpRY can edit three of the four desired target sites, with increased on-target product purity, two of which could not be generated with ABE8e-SpRY due to bystander mutations. Therefore, we next decided to apply these approaches in vivo, testing editing efficiencies in mouse embryos.

Our analysis indicated that ABE8e-SpRY effectively edited target adenines at all four sites, achieving editing efficiencies of up to 100%. The average editing efficiencies were 78.9 ± 25.7% at Tpc1-I486 [median (IQR), 91.4% (67.0–99.9%)], 52.6 ± 30.8% at Tpc2-K188 [median (IQR), 54.5% (32.9–70.1%)], 84.5 ± 20.2% at Tpc2-L249 [median (IQR), 94.1% (75.1–99.9%)], and 99.8 ± 0.6% at Trpm4-L903 [median (IQR), 99.9% (41.7–100.0%)], (Fig. 3b). Editing by ABE9-SpRY was observed at Tpc1-I486T, Tpc2-L249, and Trpm4-L903P, achieving efficiencies of up to 59.1%, 62.3%, and 89.3%, with corresponding averages of 11.0 ± 18.1% [median (IQR), 0.04% (0.02–20.26%)], 11.9 ± 16.8% [median (IQR), 6.71% (0.04–12.95%)], and 23.7 ± 28.6% [median (IQR), 8.50% (0.03–48%)], respectively.

A closer examination of adenine base editing across the protospacer sequences revealed a pattern consistent with previous in vitro findings. Despite the high editing efficiency of ABE8e-SpRY, its activity was broadly distributed, targeting adenines from positions A2 to A14 across all tested loci (Fig. 3c). In contrast, ABE9-SpRY exhibited a narrower editing window, concentrating edits primarily within positions A4 to A8. Notably, low-level editing was detected in a single embryo at the Tpc2-K188 locus, where ABE9-SpRY had shown no in vitro activity. This editing occurred at only one of the two targeted adenines, resulting in a 25.2% conversion that introduced a p.K188E substitution instead of the intended p.K188G mutation.

Using read-level allele classification within the predefined CRISPResso2 quantification window (Methods), we quantified the fraction of desired-only reads (desired substitution without bystander edits within the window). ABE9-SpRY produced higher desired-only fractions at Tpc1-I486 (10.7 ± 17.5% vs 0.3 ± 1.2%) and at Trpm4-L903 (21.3 ± 28.4% vs 0.0 ± 0.0%), whereas ABE8e-SpRY showed higher desired-only fractions activity at Tpc2-K188 (25.5 ± 25.9% vs 0.0%) and Tpc2-L249 (52.8 ± 27.5% vs 11.8 ± 16.2%) (Fig. 3d). Analysing product purity under pooled sgRNA injections within the same quantification window revealed significantly higher purity for ABE9-SpRY at the active loci Tpc1-I486 (85.1 ± 30.6% vs 1.2 ± 5.3%), Tpc2-L249 (94.6 ± 22.3% vs 60.8 ± 27.1%), and Trpm4-L903 (90.5 ± 20.8% vs 0.0%) (Fig. 3e). As these measurements were obtained under pooled sgRNA delivery, single-locus delivery may alter editor occupancy and bystander profiles. Indel frequencies were low and comparable between editors (Fig. 3f).

Only one C-to-T conversion was noted at the Tpc1-I486 locus in an embryo edited with ABE8e-SpRY, occurring at a frequency of 12.6% at position C6, alongside a single instance of 34.7% unwanted A-to-C editing (Supplementary Fig. 2a). Additionally, an unexpected proximal edit outside the protospacer was found at position A-2 of the Tpc2-L249 locus (Supplementary Fig. 2b). This edit appeared in one ABE9-SpRY and one ABE8e-SpRY injected embryo at frequencies of 12.3% and 21.9%, respectively. A possible explanation is the dynamic nature of the Cas9-sgRNA-DNA complex and conformational changes during editing, which may briefly expose adjacent bases to the deaminase domain, leading to unintended deamination events.

These findings suggest that ABE9-SpRY can enhance editing purity, albeit with locus-dependent activity. Product purity metrics in this study are defined relative to the predefined quantification window and are based on read-level allele classifications within the sequenced amplicon. While this approach captures co-occurrence of desired and bystander edits within the amplicon, it does not provide long-range haplotype phasing beyond the sequenced region; accordingly, additional long-amplicon and phased analyses at selected loci would be valuable to further characterise haplotype-resolved editing outcomes. To further assess precision beyond these target loci, we next analysed Cas9-dependent and Cas9-independent off-target editing to characterise the specificity profiles of these two editors.

Cas-dependent off-target events arise from the tolerance of mismatches between sgRNA and the targeting sequence. Although SpRY can target a broader range of genomic sites, its relaxed PAM tolerance also implies that the potential Cas-dependent off-target sites are substantially expanded, as sgRNA-DNA mismatches become the sole determinant of off-target binding without PAM constraints. To assess specificity, we combined ACEofBASEs-predicted⁵⁶ DNA-site panels in mouse embryos with an orthogonal R-loop assay in HEK293T cells, providing complementary first-line assessments at the examined sites. Accordingly, we selected the top five predicted off-target sites for each locus and analysed these across three embryos per group.

In addition to Cas-dependent off-target effects, base editors have also been reported to exhibit Cas-independent off-target activity, which is evoked by the deaminase activity of the editors^21,22. We therefore asked whether TadA-8e and TadA-9 deaminases cause Cas-independent off-target effects to similar extents, regardless of the Cas9 nickase variant to which they are attached. To this end, we employed the orthogonal R-loop assay⁵⁷ (Fig. 4e), which creates a stable R-loop structure at a specific genomic site, independent of the target region, by using an orthogonal DNA-binding module. This exposes a short ssDNA segment without recruiting the tested base editor via sgRNA/Cas9. The base editor is then expressed in trans, and any A-to-G conversion within the exposed ssDNA reports Cas-independent deamination.

In HEK293T cells, FACS-enrichment of GFP co-transfected cells showed no editing across eight adenines and three R-loop spacers with ABE9-SpRY, similar to the results reported earlier for ABE9²⁴. In contrast, ABE8e-SpRY caused substantial editing from 4.3 ± 1.1% (R loop 1), 7.7 ± 2.7% (R loop 2), to 12.0 ± 1.4% (R loop 3) (Fig. 4f). These results highlight the high specificity of ABE9-SpRY despite the increased non-specific binding of SpRY.

To make read-level co-occurrence of desired and bystander edits explicit, we classified amplicon reads within the CRISPResso2 quantification window into WT, desired-only (“perfect”), desired + bystander(s), bystander-only, and indel-containing categories (Supplementary Fig.), with representative allele frequency spectra (Supplementary Fig.). S9 S10

To determine the F1 transmission rate and establish stable mutant lines for Tpc1 p.I486T and Trpm4 p.L903P, we proceeded to mate two verified adult F0 Tpc1 p.I486T male mice (Fig. 5g) and three verified adult F0 Trpm4 p.L903P male mice (Fig. 5h) with wild-type females. We observed high transmission rates for Tpc1 p.I486T of 55.6% and 62.5% in the two independent matings (Fig. 5g, Supplementary Fig. 7a), and high to complete transmission for Trpm4 p.L903P at rates ranging from 80 to 100% (Fig. 5h, Supplementary Fig. 7b). Finally, we mated selected heterozygous F1 carriers with wild-type mice to establish stable lines (Supplementary Fig. 8).

These results suggest that the pooled embryo analysis has somewhat underestimated the potential editing efficiency of ABE9-SpRY when deployed to individual loci rather than diluted across several for screening. The high editing efficiencies achieved, up to 96% at one target site, underscore the utility of ABE9-SpRY for establishing stable mouse lines at the loci tested. While somatic mosaicism in the F0 generation necessitates the use of F1 offspring for standardised phenotypic characterisation, the observed F0-F1 transmission supports germline contribution, although ear-biopsy allele fractions may not fully reflect germline composition or uniquely determine the degree of mosaicism. Nonetheless, the high frequency of transmissible desired alleles streamlines the transition from zygote injection to the establishment of the experimental cohort.

Information on major resources is provided in Supplementary Table 1.

Wild-type C57BL/6N mice were purchased from Charles River Laboratories (Wilmington, MA, USA). The experimental procedures were approved by the regional council of Karlsruhe, Germany, in accordance with the Animal Welfare Act (AZ35-9185.81/G-60/23). Microinjection was performed in the cytosol of C57BL/6N zygotes. All animals were housed in the Interfaculty Biomedical Faculty (IBF) of Heidelberg University. Mice were maintained under specified pathogen-free conditions with a 12-h light/12-h dark cycle, and water and standard food (Rod18, LASvendi GmbH, Germany) available ad libitum. All mouse experiments described in this study comply with ARRIVE guidelines.

To identify the corresponding mutations in the mouse genome, to four target loci of interest with missense mutations Tpc1 p.I485T, Tpc2 p.L265P, Tpc2 p.K204A, and Trpm4 p.L907P we retrieved the Mus musculus and Homo sapiens gene sequences from Ensembl and performed protein sequence alignments. This allowed us to pinpoint the target amino acids and the corresponding adenines in the mouse coding sequence. Through this alignment, we identified the orthologous amino acid residues in the respective mouse genes and the potential missense mutations we can model by A-to-G transition mutation mediated through ABEs: Tpc1 p.I486T, Tpc2 p.L249P, Tpc2 p.K188G, and Trpm4 p.L903P. Notably, for Tpc2 p.K188G, a glycine substitution was created in silico instead of the alanine. We decided to proceed with the glycine substitution based on the similar chemical and functional properties of glycine and alanine.

sgRNAs were designed manually to enable the precise installation of the defined missense mutations and validated using ACEofBASEs⁵⁶ with the following parameters: select BaseEditor “custom ABE” with Window start "3" and Window end “11” with PAM “NRN” for Tpc1-I486, Tpc2-K188, and Trpm4-L903 target sites or with PAM “NYN” for the Tpc2-L249 target site; for Species “Mouse (Mus musculus) Ensembl V 103”. The full list of sgRNA oligonucleotides to clone and the sgRNAs used in this study are shown in Supplementary Tables 2 and 3.

For in vitro applications of sgRNAs, the 20 bp SpCas9 sgRNA spacer sequences used in this study were designed with the 5’ and 3’ sequences (Supplementary Table 2), as previously described⁸¹, with an additional 5’ “G” added in cases in which the spacer does not start with a G. In brief, the 20 bp target site (N20) sequences were ordered as oligonucleotides with the following sequences 5’-CACC(N20-21)GTTTT-3’, while the sequences complementary to the target site were designed and ordered as oligonucleotides as follows 5’-CTCTAAAAC(N20-21)-3’. sgRNAs were cloned into the BsaI-digested 2.18 kb pU6-pegRNA-GG-acceptor fragment (pU6-pegRNA-GG was a gift from David Liu, Addgene plasmid #132777⁸¹) by oligonucleotide annealing and Golden Gate assembly as previously described⁸¹, including the standard SpCas9 sgRNA scaffold.

R-loop SaCas9 sgRNAs. To generate the SaCas9 sgRNA plasmid, BPK1520, containing a pU6_BsmBIcassette-Sp-sgRNA was used as a backbone (BPK1520 was a gift from Keith Joung, Addgene plasmid #65777⁴²). The SpCas9 sgRNA scaffold was digested using BsmBI and HindIII, and the digested fragment was isolated. The SaCas9 sgRNA scaffold sequence⁴⁵, designed with complementary overhangs (Supplementary Table 2), was ligated into the digested vector backbone. The designed sgRNA spacer sequences were then ordered as oligonucleotides with BsmBI overhangs (Supplementary Table 2), and the annealed oligos were cloned into the BsmBI-digested SaCas9 sgRNA plasmid via Golden Gate assembly. In brief, in a 10 µl reaction volume, 50 ng of SaCas9 sgRNA plasmid, 10 units of BsmBI-v2 (NEB), 20 units of T4 DNA ligase (NEB), 1 µL of 1 µM annealed spacer oligonucleotides, 1X T4 DNA ligase buffer were assembled under the following thermal cycling conditions: digestion at 42 °C for 4 min, followed by ligation at 16 °C for 3 min, repeated for a total of 11 cycles. The reaction was then incubated at 80 °C for 10 min to inactivate BsmBI.

hiPSC sgRNAs. Cloning of targeting sgRNA into the pDT-sgRNA-XMAS-1 × plasmid was performed as previously described (pDT-sgRNA-XMAS-1 × was a gift from Xiao Wang, Addgene plasmid #164413⁶⁰).

All sgRNAs for mouse experiments were ordered via Integrated DNA Technologies (IDT) as custom Alt-R CRISPR-Cas9 sgRNA (Supplementary Table 3).

All plasmids constructed here were cloned via Gibson Assembly using the NEBuilder HiFi DNA Assembly (NEB) kit with PCR products amplified from designated template plasmids with specific primers (Supplementary Table 4) with Q5 Hot Start DNA polymerase (NEB).

To obtain pCMV_ABE9, we used the ABE9 editor sequence form²⁴, and recloned pCMV_ABE8e (a gift from David Liu, Addgene plasmid #138489²³) by introducing the two defining mutations N108Q and L145T by PCR, altogether assembled via four fragments (Supplementary Table 4): pCMV_backbone, ABE9_fragment_1_TadA*_Nterm_N108Q, ABE9_fragment_2_TadA*_N108Q_L145T, ABE9_fragment_3_Cas9n-Nterm. To clone pCMV_ABE8e-SpRY, we assembled the following PCR products: pCMV_backbone, ABE8e_fragment1_TadA*, and ABE9-SpRY_fragment_4_SpRY-Cas9n (amplified from pCMV-T7-ABE8e-nSpRY-P2A-EGFP (KAC1069), a gift from Benjamin Kleinstiver, Addgene plasmid #185912⁸²). Similarly, pCMV_ABE9-SpRY was constructed by assembling: pCMV_backbone, ABE9_fragment_1_TadA*_Nterm_N108Q, ABE9_fragment_2_TadA*_N108Q_L145T, ABE9_fragment_3_Cas9n-Nterm, and ABE9-SpRY_fragment_4_SpRY-Cas9n.

For the transfection of hiPSCs, pCMV_ABE8e-SpRY and pCMV_ABE9-SpRY were recloned to be expressed under the control of the EF1α promoter. In brief, two fragments were amplified, first the pEF1α_backbone (from pEF1α-hMLH1dn, a gift from David Liu, Addgene Plasmid #174824⁸³) and then either ABE8e-SpRY or ABE9-SpRY, for subsequent Gibson assembly cloning.

The entire sequence of all constructs assembled using PCR was verified with the PlasmidEZ (Azenta/GeneWiz) full-plasmid sequencing service. All plasmids used for the transfection experiments were prepared using the EndoFree Plasmid Maxi Kit (Qiagen).

mRNA for the microinjection of mouse zygotes for ABE8e-SpRY and ABE9-SpRY was generated by in vitro transcription from SalI-HF linearised pCMV_ABE8e-SpRY and pCMV_ABE9-SpRY plasmid templates, respectively, using the HiScribe T7 ARCA mRNA Kit. The mRNA was purified using the RNeasy Mini Kit. mRNA quality was confirmed using the TapeStation RNA ScreenTape.

Human HEK 293 T cells and Neuro2a cells were purchased from ATCC and maintained in Dulbecco’s modified Eagle’s medium (DMEM; Gibco) supplemented with 10% FBS (Gibco) at 37 °C with 5% CO₂, up to 20 passages, by passaging cells every two to three days. The hiPSC line SCVI15 (Supplementary Table 1) was obtained from Greenstone Biosciences, Inc. via Timon Seeger; beyond the information listed in Supplementary Table 1, no additional donor background or genomic characterisation information was provided by the source. The hiPSCs were cultured on hESC-Qualified Matrix Matrigel-coated (Corning) 6-well plates (Sarstedt) in StemMACS PSC-Brew XF medium (Miltenyi Biotec). Medium was changed daily, and cells were passaged every three days using StemMACS Passaging Solution XF (Miltenyi Biotec). The culture medium was supplemented with 2.5 µM Rock Inhibitor (Y27632, Miltenyi Biotec) on passaging days for 24 h. The supernatant media from cell cultures were routinely analysed for mycoplasma presence using the MycoSPY Master Mix (Biontex).

All HEK293T and N2a transfection experiments were conducted with at least three independent biological replicates. HEK293T and N2a cells were seeded at a density of roughly 110,000 and 180,000 cells per well in 24-well plates, respectively, 20–24 h before transfections. Transfections were performed with 750 ng of base editor plasmid, 250 ng of sgRNA expression plasmid, and, for experiments involving GFP co-transfection, 200 ng of pMAX-GFP (Lonza), combined with 2 μl of Lipofectamine 2000 (Thermo Fisher Scientific) in a total volume of 50 μl Opti-MEM (Thermo Fisher Scientific), according to the manufacturer’s instructions. Medium was exchanged to culture medium approximately 24 h after transfection. For the Orthogonal R-loop assay, HEK293T cells were seeded at a density of 110,000 cells per well in 24-well plates, as described above. However, the cells were co-transfected with a mixture containing 500 ng of plasmids encoding base editor plasmids, 350 ng of SpCas9 sgRNA (HEKsite2), 500 ng of dSaCas9, 350 ng of saCas9 sgRNAs targetting separate genomic loci unrelated to the on-target site, and 200 ng of pMAX-GFP (Lonza).

hiPSCs were seeded on 6-well plates and transfected after approximately 24 h at around 60% confluency. Before transfections, the medium was changed to StemMACS PSC-Brew XF supplemented with StemMACS PSC-Support XF (Miltenyi Biotec). Transfections were performed with 300 ng pEF1α-XMAS-1xStop (a gift from Xiao Wang, Addgene plasmid #164411⁶⁰), 300 ng pU6-SpCas9-XMAS-1 × plasmid, 2100 ng of pEF1α_ABE8e-SpRY or pEF1α_ABE9-SpRY, and 10 µl Lipofectamine Stem (Thermo Fisher Scientific) per well, according to the manufacturer’s instructions. Medium was exchanged to culture medium approximately 24 h after transfection.

Approximately 48 h after transfections, HEK293T, N2a or hiPSCs were harvested for sorting. HEK293T and N2a cells were treated with TrypLE Express Enzyme (Thermo Fisher Scientific) to singularise the cells, followed by a washing step with D-PBS. The cells were resuspended in buffer containing 1 × PBS without Mg²⁺/Ca²⁺ (Thermo Fisher Scientific), 0.5% FCS (Thermo Fisher Scientific) and 1% Penicillin–Streptomycin and filtered through a 35 µm cell strainer (Corning). hiPSCs were dissociated with Accutase (Thermo Fisher Scientific) and washed with D-PBS, followed by resuspending the cells in 1 × PBS without Mg²⁺/Ca²⁺, 0.5% FCS, 2.5 µM Rock Inhibitor and passed through a 35 µm cell strainer (Corning). HEK293T and N2a cells co-transfected with pMAX-GFP were sorted for the top 30% GFP-positive population. For the XMAS-TREE reporter assay in hiPSCs, only cells co-expressing GFP and mCherry were sorted, and the data were analysed using a sequential gating strategy. Events were first gated as “Cells” based on forward- and side-scatter properties to exclude debris, followed by selection of “Singlets” using FSC-A versus FSC-H to exclude doublets. Within the singlet population, mCherry⁺ cells were identified using thresholds defined by negative and single-colour controls, and the mCherry⁺GFP⁺ double-positive population was subsequently quantified. Representative gating plots and population statistics (%Parent and %Total as reported by the acquisition software) for all samples are shown in Supplementary Fig. 11–13 and Supplementary Table 12. FACS sorting was performed with either a BD FACSAriaIII or BD FACSAria Fusion at the Flow Cytometry & FACS Core Facility (FFCF), Zentrum für Molekulare Biologie der Universität Heidelberg (ZMBH) or FACS Core Facility (dFCCU), Uniklinikum Heidelberg, respectively.

After FACS HEK293T, N2a cells were collected in 500 µl DMEM GlutaMAX supplemented with 10% FBS. hiPSCs were collected in 500 µl StemMACS PSC-Brew XF medium supplemented with 2.5 µM Rock Inhibitor. Cells were collected by centrifugation at 20,000 × g for 2 min, washed with 500 µl D-PBS, and lysed in mammalian gDNA lysis buffer (10 mM Tris–HCl, 0.05% SDS, pH 8, 800 units proteinase K (NEB)) for 2 h at 37 °C to extract genomic DNA. Afterwards, proteinase K was inactivated for 30 min at 80 °C. Cells transfected without fluorophore plasmid were washed once 72 h after transfection, and then lysed in mammalian gDNA lysis buffer, as described above.

Injections were performed as previously described⁸⁴, with minor modifications. In brief, the injection mix contained adenine base editor mRNA (100 ng/μl) and sgRNAs (50 ng/μl each; 12.5 ng/µl each, for pooled sgRNA injections) in injection buffer (10 mM Tris–HCl, 0.25 mM EDTA, pH 7.4) to a final volume of 50 μl with molecular biology-grade water. To prevent clogging of the cannula, the injection mix was filtered shortly before microinjection using a Corning Costar Spin-X 0.22 μm centrifuge tube filter. Zygotes were collected from timed matings and cultured in vitro in M2 medium during the injection. Injection into the cytoplasm of zygote-stage embryos was conducted by carefully inserting the injection needle at the equatorial level into the cytoplasm of each embryo, delivering a volume of 1–2 picolitres. Subsequently, the embryos were placed in preincubated M16 medium to select lysed and intact embryos. Embryos were transferred on the same day via oviduct transfer to 0.5-day pseudopregnant foster mothers to increase the overall implantation (nidation) rates.

Embryos were collected at embryonic day 14 (E14) from surrogate mothers to ensure sufficient tissue for genomic DNA isolation and downstream amplicon sequencing. For DNA extraction, each embryo was sectioned into smaller pieces, and genomic DNA was isolated using the DNeasy Blood & Tissue Kit (Qiagen) following the manufacturer’s protocol, with minor modifications. All reagents were applied at double the recommended volume since the starting material exceeded 10 mg per embryo. Proteinase K from New England Biolabs (NEB) was utilised instead of the kit-supplied enzyme.

For genotyping of founder mice, ear biopsies were obtained from the Interfakultäre Biomedizinische Forschungseinrichtung of Heidelberg University. Genomic DNA was extracted from ear biopsy samples using a lysis buffer consisting of DirectPCR Lysis Reagent (Mouse Tail) (Viagen Biotech) and 0.25 µg/µl Proteinase K (AppliChem). For lysis, each sample was incubated in 100–150 µl of this lysis buffer at 55 °C with vigorous shaking for a minimum of 4–5 h, followed by inactivation of Proteinase K at 85 °C for 45 min. The lysate is subsequently used directly for PCR.

The top five predicted off-target sites for each target locus were identified using ACEofBASEs⁵⁶ as described under sgRNA design (Supplementary Table 7–10). ACEofBASEs provides links to Ensembl gene IDs; therefore, the locus information for these sites was extracted from Ensembl. Importantly, the column “distance” in the off-target list was used when exporting Ensembl sequence information. In brief, when navigating the ENSEMBL genome browser, to export the locus information, ensure the value of distance and an additional 1 kb is added to the “5' Flanking sequence” and “3' Flanking sequence”. Primers were designed (Primer3 2.3.7) to flank the predicted off-target loci. For cumulative editing, all reads containing A-to-G edits in the spacer sequence were added together.

Samples were prepared for Amplicon-EZ-based NGS analysis (GeneWiz/Azenta Life Sciences) as previously described⁵⁶. All locus-specific primers contained the 5’ partial Illumina adapter sequences (Fwd: 5’-ACACTCTTTCCCTACACGACGCTCTTCCGATCT-3’, Rev: 5’-GACTGGAGTTCAGACGTGTGCTCTTCCGATCT-3’), and are listed in Supplementary Table 5. Typically, samples corresponding to different target loci from each biological replicate were pooled into a single sample. The pooled NGS data were demultiplexed by mapping to the respective reference sequences later.

For a higher level of sample multiplexing, a sequencing library with indexing primers that include unique i5 and i7 indexes, as well as P5 and P7 adapter sequences, was established. In brief, following the preparation of PCR amplicons with locus-specific primers as described above (PCR1), a second round of PCR was performed on these amplicons to barcode each sample, thereby allowing for the pooling of samples from both the same and different target sites (PCR2). The primer pairs for the second PCR contain complementary regions to the partial Illumina adapters along with the unique i5 and i7 index combinations (Supplementary Table 6). For multiplexing, PCR1 samples were amplified using Q5 Hot Start DNA polymerase (NEB) for 27 cycles, validated by agarose gel electrophoresis of a subsample, and diluted between 1:5 and 1:20 based on gel band density. PCR2 was then set up at a volume of 20 µl, containing 1 µl of the diluted PCR1 product, 0.5 µM of each forward (i5) and reverse primer (i7) to create a combinatorial dual-index library, and Phusion U Green Multiplex PCR Master Mix (Thermo Fisher Scientific); it was amplified for an additional 13 cycles. Individual PCR2 amplification was validated by agarose gel electrophoresis of a subsample, and 24 PCR2 samples were pooled, run on a 1% agarose gel, with specific bands excised and cleaned up using the Monarch DNA Gel Extraction Kit (NEB), measured by Qubit 4 Fluorometer (Thermo) using the Qubit dsDNA BR Assay Kit, and pooled to equimolarity totalling 1 µg DNA. Samples were sequenced by GeneWiz (Azenta Life Sciences) using the Sequencing only service with a 5% PhiX spike-in on an Illumina MiSeq with 2 × 250 bp, paired-end sequencing, at a depth of approximately 10 million reads.

NGS data (in fastq.gz format) were processed and analysed using CRISPResso2 version 2.2.11⁸⁵. In brief, two types of analyses were run in batch mode. First, all samples were analysed in base editing mode with the following custom input parameters: –min_average_read_quality, 30; –quantification_window_size, 20; –quantification_window_center, -10; –base_editor_output. Downstream analysis was conducted using R version 4.2.1 in R Studio (packages: tidyverse, ggplot, ggpubr, cowplot), with data used to plot Adenines across the protospacer sourced from the “Nucleotide_percentage_summary_around_sgRNA.txt” output file. “Precision” (Supplementary Fig. 1) was calculated as previously described²⁴ by dividing the second highest adenine-based editing frequency by the highest (A5 or A6); for this purpose, the indel frequency was extracted from the “CRISPRessoBatch_quantification_of_editing_frequency.txt” file and calculated as previously described⁵⁶.

Second, to quantify precise gene editing outcomes, the HDR mode was employed with the following custom input parameters: –min_average_read_quality, 30; –expected_hdr_amplicon_seq; –quantification_window_coordinates, “spacer-7nt (upstream)” – “spacer + PAM sequence (+ 3nt)”; –discard_indel_reads, TRUE. Downstream analysis was conducted using R, as described above, with data sourced from the “CRISPRessoBatch_quantification_of_editing_frequency.txt” file, with slight modifications following⁸⁶. In short, for each sample’s HDR amplicon, values under “Reads_aligned_all_amplicons”, “Reads_aligned”,“Unmodified”, “Modified”, and “Discarded” (HDR) were collected; additionally, the values under “Discarded” (REF) from the Reference amplicon were obtained. Specific values were then calculated as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} & {\mathrm{Specified}}\;{\mathrm{mutation}},{\mathrm{incl}}.\;{\mathrm{bystander}}\;{\mathrm{edits}}\left( \% \right) \\ & \quad = \frac{{``{\mathrm{Reads}}\_{\mathrm{aligned}}{{\mathrm{''}}}}}{{``{\mathrm{Reads}}\_{\mathrm{aligned}}\_{\mathrm{all}}\_{\mathrm{amplicons}}{\mathrm{''}}}} \times 100 \\ \end{aligned}$$\end{document}2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} & {\mathrm{Specified}}\;{\mathrm{mutation}}\;{\mathrm{without}}\;{\mathrm{bystander}}\;{\mathrm{edits}}\left( \% \right) \\ & \quad = \frac{{\text{``Unmodified{\mathrm{''}}}}}{{{\mathrm{``Reads}}\_{\mathrm{aligned}}\_{\mathrm{all}}\_{\text{amplicons{\mathrm{''}}}}}} \times 100 \\ \end{aligned}$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Purity}}\left( \% \right) = \frac{{{\mathrm{``Unmodified}}{\mathrm{''}}}}{{{\mathrm{``Reads}}\_{\mathrm{aligned}}{\mathrm{''}}}} \times 100$$\end{document}4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Indels}}\left( \% \right) = \frac{{{\mathrm{``Discarded}}{\mathrm{''}}\left( {{\mathrm{HDR}}} \right) + {\mathrm{``Discarded}}{\mathrm{''}}\left( {{\mathrm{REF}}} \right)}}{{{\mathrm{``Reads}}\_{\mathrm{aligned}}\_{\mathrm{all}}\_{\mathrm{amplicons}}{\mathrm{''}}}} \times 100$$\end{document}

All data reported in the text are presented as mean ± standard deviation (SD). Where appropriate (n ≥ 20), we also reported the median (Q2) and interquartile range (IQR) between the first quartile (Q1) and the third quartile (Q3). Statistical comparisons were conducted using two-tailed Welch’s t-test, as specified in the figure legends. Details on data reporting, including sample sizes and the statistical methods employed in each experiment, are provided in the figure legends. Data demonstrating statistical significance are depicted in the figures. Analysis and graphical data visualisation were performed in R version 4.2.1 with the Tidyverse, ggplot2, ggpubr, and cowplot packages^87–90. Figures were assembled in Adobe Illustrator, with some icons and schemes imported from Biorender.