Safer and more efficient gene editing using protein-RNA complexes delivered by improved lipid nanoparticles

The potential for therapeutic benefit from administered proteins hinges upon successful delivery across cellular membranes. A number of agents have been demonstrated to enhance intracellular delivery of genome-editing proteins and RNPs, including CPPs. Accordingly, we genetically fused three different CPPs (TAT, CPP5 and ANTP)^44–47 to the N terminus of Cre recombinase and ABE8e-SpCas9-NG (hereafter referred to as ABE8e), which recognizes the NG PAM (Fig. 1a). To facilitate purification, we also fused the 1D4 peptide tag⁴⁸ to the C terminus of ABE8e and the prime editor protein, PE2 (Fig. 1a). Although Cre recombinase and Cas9-based genome-editing proteins possess distinct physicochemical properties, such as size and charge, we hypothesized that insights from the study of intracellular delivery of Cre recombinase would inform the design of delivery vehicles for ABE8e and PE2.

We first purified Cre recombinase to homogeneity, with or without the N-terminal CPP fusion, through nickel metal-affinity chromatography (Ni-NTA) chromatography (Fig. 1b) and size exclusion chromatography (SEC, Fig. 1c), and confirmed the purity of the product through sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) visualized with Coomassie brilliant blue (CBB) (Fig. 1d) and anti-Cre western blot (Fig. 1e). ABE8e was similarly purified through affinity chromatography on TALON metal (Fig. 1e) and 1D4 (Fig. 1f) affinity columns and completed with SEC (Fig. 1g); purity was again assessed by SDS–PAGE with CBB staining and anti-Cas9 western blot (Fig. 1h,i). PE2 was purified in a similar manner as ABE8e, with an additional heparin chromatography step after 1D4 immunoaffinity to maximize separation from impurities (Fig. 1j). We noted that ABE8e after immunoaffinity chromatography still contained contaminating nucleic acids and aggregates; these contaminants were removed effectively by SEC (Fig. 1k). We also noted that ABE8e RNP was more stable than ABE8e protein alone, as determined by differential scanning fluorimetry (DSF), especially when the single guide RNA (sgRNA) was refolded by heating and slow cooling (Fig. 1l and Supplementary Fig. 1a–f). The ABE RNP was further stabilized by 10% (w/v) sucrose. Similar to ABE8e, the PE2 RNP complex with an engineered prime-editing guide RNA (epegRNA) was more stable than PE2 protein alone; however, heat refolding of epegRNA had no additional effect on PE2 RNP stability (Fig. 1m and Supplementary Fig. 1g,h). PE RNP did not require sucrose to remain soluble. For consistency, we heat refolded all sgRNAs and epegRNAs used throughout the study. We also assessed the enzymatic activity of our ABE8e and PE2 RNPs through in vitro activity assays to ensure our proteins maintained activity following purification. ABE8e RNP displayed high levels of activity as assessed by an in vitro deamination assay (Fig. 1n,o), and PE2 RNP similarly displayed high levels of activity as assessed by an in vitro reverse-transcriptase extension assay (Fig. 1p,q).

As all of the CPPs were effective to some degree in delivering Cre recombinase in vitro and in vivo, we then applied them for the delivery of ABE and PE to the rd12 mouse model of Leber congenital amaurosis. In these mice, a nonsense mutation in Rpe65 abolishes RPE65 expression^41,42, leading to a lack of visual chromophore production and photoreceptor light detection; however, successful base or prime editing rescues this phenotype. To enable efficient in vitro screening, we developed a fluorescent rd12 reporter cell line. We retrovirally transduced NIH/3T3 cells with a construct containing a fragment of the Rpe65 rd12 complementary (c)DNA encompassing an in-frame nonsense mutation, which is flanked by mCherry and eGFP. In unedited cells, the stop codon only permits mCherry protein expression and successful transition by base or prime editing results in the expression of the mCherry-eGFP fusion protein (Fig. 3a). We verified that adenine base editing via plasmid transfection of ABE8e with a previously validated rd12 sgRNA⁴² successfully edited the reporter construct and restored eGFP expression, as quantified by next-generation sequencing (22%, Fig. 3b) and flow cytometry (32%, Fig. 3c). Unmodified ABE applied on the rd12 reporter cells caused low but detectable conversion of the fluorescent reporter (Fig. 3c,d), but in contrast to fusion with Cre recombinase, the fused CPP peptides did not enhance the delivery of ABE in vitro (Fig. 3d). However, as we previously observed that sucrose promoted stability of the RNP complex (Supplementary Fig. 1c–f), the addition of sucrose (10% w/v) to the RNP mixture boosted delivery of the ABE8e RNP (Fig. 3d). This was further enhanced 1.5-fold by the addition of the non-covalent 6×His-CM18-PTD4 peptide, but fused CPPs still did not result in improved delivery efficiency (Fig. 3d). In contrast to ABE, we did not observe a colour change in the rd12 reporter cells when PE was delivered with 2% sucrose, but we noted activity when delivered with 10% sucrose. As in the case of Cre and ABE, the 6×His-CM18-PTD4 peptide modestly improved delivery efficiency of PE2 RNP, by ~1.6-fold, although overall efficiencies remained low (Extended Data Fig. 1a,b).

We then investigated whether the non-covalent 6×His-CM18-PTD4 peptide or high sucrose concentration could effectively enable the delivery of ABE8e RNP in vivo. Subretinal injection of ABE8e RNP rescued the scotopic dark-adapted flash electroretinography (ERG) b-wave, indicating successful genome editing and restoration of the visual cycle, as untreated rd12 mice do not exhibit a recordable ERG waveform in response to this light stimulus (Fig. 3e,f). We noted that the low-NaCl, high-sucrose formulation effected a more robust rescue, as measured by ERG b-wave amplitudes (Fig. 3g). Consistent with our in vitro findings, the rescue of the scotopic flash ERG was promoted by the non-covalent 6×His-CM18-PTD4 peptide but was not substantially improved by the fused CPPs (Fig. 3f,g,j,k). We sequenced genomic DNA and transcripts isolated from RPE samples from the treated mice and noted up to 2% on-target genomic editing, corresponding to up to 7% on-target base editing in cDNA. The ABE editing outcomes were distributed approximately equally between precise edits and edits with at least one bystander adenine deaminated (Fig. 3h,i,l,m). Notably, purified RNP led to more precise ABE editing than eVLP packaging of the same ABE8e, which led to multiple deaminated adenines²⁸.

As an alternative to CPPs, lipid reagents are also suitable for CRISPR/Cas9 RNP delivery⁵¹. Our preliminary investigations showed that binding of sgRNA by ABE results in change of the net charge from positive (zeta potential of +4.4 mV) to negative (−7.6 mV), which is in line with data published for Cas9 (ref. ⁵²) and suggests that ABE RNP can be efficiently captured by cationic lipids as a lipoplex after simple mixing, or into a lipid nanoparticle after microfluidic encapsulation. For example, delivery of ABE RNP in vitro via a lipoplex with Lipofectamine 3000 was efficient in the rd12 reporter cell line at concentrations as low as 20 nM (4.5 µg ml⁻¹) (Extended Data Fig. 2b,c). ABE RNP delivered using Lipofectamine 3000 restored expression of RPE65 in the rd12 cDNA reporter cell line, and next-generation sequencing analysis showed improved efficiency and precision of ABE editing mediated by RNP compared with plasmid (Extended Data Fig. 2d,e). However, the formulation of ABE8e RNP that was optimal in vitro (50 pmol (11.2 µg) RNP per µl of Lipofectamine 3000, 5.6 µg RNP per eye) did not restore the ERG responses in the rd12 mice. This result might correspond to toxicity of the high dose of Lipofectamine 3000 (50% by volume, 0.5 µl per eye), as we observed substantially higher rescue when we used a 5-fold lower dose of Lipofectamine 3000 with the same amount of ABE RNP (Extended Data Fig. 2f,g). These results indicate that lipid-mediated delivery of ABE RNP in vivo is a viable approach; however, we envisioned that a chemically defined formulation that minimizes toxicity is required to maximize the potential of RNP delivery in vivo.

To this end, we sought to adapt LNP technology, which has successfully delivered mRNA vaccines and Cas9 nuclease RNP^52,53. We found that the ABE RNP was transiently stable at pH 6.0 and in 25% ethanol, conditions which are necessary for LNP formulation. To create ABE and PE RNP LNPs, we tested a panel of ionizable lipids in a prototypical lipid mixture consisting of ionizable lipid, 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), cholesterol and 1,2-dimyristoyl-rac-glycero-3-methoxypolyethylene glycol-2000 (DMG-PEG 2000) at a molar ratio of 50:10:38.5:1.5, respectively. To avoid inactivation of ABE, we utilized ionizable lipids whose pK_a was above 6.0: 8-[(2-hydroxyethyl)[6-oxo-6-(undecyloxy)hexyl]amino]-octanoic acid, 1-octylnonyl ester (SM102): pK_a = 6.68 (ref. ⁵⁴); 9-octadecenoic acid, 1,1′-[7-[4-(dipropylamino)butyl]-7-hydroxy-1,13-tridecanediyl] ester (CL4H6): pK_a = 6.25 (ref. ⁵⁵); 1,2-dioleyloxy-3-dimethylaminopropane (DODMA): pK_a = 6.59 (ref. ⁵⁶), enabling the protonation of the tertiary amine of the ionizable lipids at pH 6.0 to facilitate incorporation of ABE RNP into LNPs during encapsulation (Supplementary Fig. 5 and Fig. 4a). We noted that the resultant LNPs were highly monodispersed with a hydrodynamic diameter between 200 and 250 nm, measured by dynamic light scattering (Fig. 4b). We verified encapsulation of ABE RNP within our LNPs by distinguishing encapsulated RNPs from free ABE RNPs through immunoprecipitation with 1D4 resin and subsequent anti-Cas9 western blot analysis (Fig. 4c,d and Extended Data Fig. 3a,b). While free ABE RNPs were bound by the resin and eluted with 1D4 peptide, ABE RNP LNPs were detected in the non-bound fraction (NB). The binding of ABE was epitope specific, as a B6-30 resin directed against the N terminus of rhodopsin did not bind ABE (Extended Data Fig. 3a)⁵⁷. When the LNPs were tested, the majority of the ABE was found in the NB and in wash fractions (data not shown). Upon elution with Laemmli sample buffer, some RNP was eluted from the 1D4 resin (Fig. 4d), suggesting non-specific binding because no material was eluted from LNP-treated resin when 1D4 peptide was used instead (Extended Data Fig. 3a), and a similar amount of LNP-derived material was eluted with Laemmli sample buffer from non-binding control B6-30 resin (Extended Data Fig. 3b).

We tested the ABE RNP LNPs on the rd12 reporter cells and noted that the LNPs effected rapid delivery of active ABE (Fig. 4e). We also noted high conversion efficiency for all three LNPs, nearing 100% for SM102 LNPs (Fig. 4f,g). The LNPs were highly potent, with as little as 20 nM ABE RNP (4.5 µg ml⁻¹) in SM102 LNPs eliciting nearly total conversion of the reporter cells. This result represented a substantial improvement over the lipoplex with Lipofectamine 3000, which peaked at ~50% of the cells converted (Fig. 4g and Extended Data Fig. 2c). We observed high toxicity of DODMA LNPs, which we alleviated by replacement of the 2.5% DSPC within the LNP with 1-stearoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (SOPS) (Extended Data Fig. 3c). The LNPs remained stable for at least 28 days when stored at −80 °C, and for at least 14 days when stored at 4 °C (Extended Data Fig. 3d). We then verified that these LNPs were active in vivo via subretinal delivery into rd12 mice, with all three LNPs effecting a substantial rescue of the ERG b-wave (Fig. 4h). Importantly, although the concentration of ABE in the LNP preparation (2 µM, 0.45 mg ml⁻¹, 450 ng per eye) was at least 10-fold lower than the RNPs formulated with sucrose or Lipofectamine 3000 in vivo (20–36 µM, 4.5–8.1 mg ml⁻¹, 4.5–8.1 µg per eye), we registered the most substantial response here, with the ERG b-wave amplitude reaching up to 265 µV for the SM102 LNP-RNP, whereas the highest amplitude registered for free RNP was 185 µV with added 6×His-CM18-PTD4 peptide (Fig. 3g), and 200 µV with Lipofectamine 3000 (Extended Data Fig. 2g).

Our experiments in vitro and in vivo suggested that LNPs made with SM102 were the most effective, but we reasoned that we could further improve the LNP formulation. As modulating PEG lipids can influence biodistribution⁵⁸, we investigated the impact of varying the DMG-PEG 2000 concentration. We also investigated whether we could further optimize the lipid:RNA weight ratio, which was 40:1 in our original formulation. We found that DMG-PEG 2000 lipid is indispensable for encapsulation of the RNP, as LNPs without DMG-PEG 2000 exhibited the highest particle size (Fig. 4i) and were not active in vitro (Fig. 4j). We determined that LNPs with 2.5% DMG-PEG 2000 were the most potent in vitro, achieving nearly 100% conversion of cells at 5 nM (1.1 µg ml⁻¹) ABE RNP, a concentration 4-fold lower than that for 1.5% DMG-PEG 2000 (Fig. 4j). The optimal lipid:RNA weight ratio was at least 40:1 (Extended Data Fig. 3e,f). The ABE RNP was fully encapsulated at the optimal formulation of 40:1 lipid:sgRNA ratio and 2.5% DMG-PEG 2000, as shown by 1D4 immunoassay (Extended Data Fig. 3g) and SEC (Extended Data Fig. 3h). We assessed the structure of our optimized RNP LNPs through cryoelectron microscopy (cryoEM) and confirmed intact and uniform particles with an approximate diameter of 100 nm (Fig. 4k). We further characterized our RNP LNPs by absolute quantification of ABE within the LNPs through targeted proteomics (Fig. 4l,m and Supplementary Fig. 6a–e). We noted that Cas9 and TadA* deaminase peptides were detected at the same level, indicative of intact, full-length ABE8e protein within the LNPs (Fig. 4n). The ABE RNP LNPs restored expression of RPE65 in the rd12 cDNA cell line, and rescue via RNP was notably higher via LNP delivery compared with Lipofectamine 3000 (Fig. 4n). Next-generation sequencing revealed that the RNP LNPs mediated higher on-target editing, up to 40%, compared with plasmid- or Lipofectamine-delivered RNP treatment, and lower bystander editing compared with plasmids (Fig. 4o). The off-target activity of ABE in the rd12 cDNA cell line was minimal, as we noted ≤0.3% edited alleles in off-target sites, concurrent with almost 60% editing of the on-target adenine (Fig. 4p). When tested in vivo, the SM102 LNP-RNP with 2.5% DMG-PEG 2000 resulted in a higher maximal rescue of the ERG b-wave compared with SM102 LNP RNPs with 1.5% DMG-PEG 2000 (368 µV and 265 µV, respectively), approaching response levels registered for WT mice (Fig. 4q).

We also successfully encapsulated the prime editor PE2 within our optimized LNP formulation. Similar to ABE RNP, PE2 RNP can be efficiently delivered in vitro as a lipoplex with Lipofectamine 3000, although it was efficient at higher concentrations of RNP (500 nM (149 µg ml⁻¹) for PE compared with 20–100 nM (4.5–22.4 µg ml⁻¹) for ABE; Extended Data Figs. 1a,c,d and 2b,c). Our initial effort to encapsulate PE2 RNP into LNPs with 1.5% DMG-PEG 2000 was partially successful, as some unencapsulated PE2 protein was detected in the 1D4 immunoprecipitation assay (Fig. 5a). The particle size was also larger than for ABE RNP LNPs (Fig. 5b). Nevertheless, the unoptimized PE RNP LNPs delivered PE into the rd12 reporter cells with efficiency surpassing Lipofectamine 3000 (Fig. 5c). By increasing the concentration of DMG-PEG 2000 to 2.5%, we no longer detected unencapsulated PE2 (Fig. 5a) and observed robust delivery of PE into the rd12 reporter cells (Fig. 5c,d). The cryoEM images of optimized PE RNP LNP revealed the presence of homogeneous, well-defined particles (Fig. 5e). We noted that Cas9 and the reverse transcriptase (MMLV RT) were not detected at equimolar concentrations within our PE2 RNP LNPs, indicating that further improvements may be necessary in LNP formulation and in PE2 purification (Fig. 5f). Nevertheless, the PE RNP LNP restored expression of RPE65 in the cDNA-expression cell line (Fig. 5g). Next-generation sequencing revealed that RNP-mediated prime editing led to exceptional purity of editing outcomes, as we did not notice any unwanted editing of the cells bearing the Rpe65 rd12 cDNA, while indels were detected in cells transfected with PE and epegRNA plasmids (Fig. 5h). No off-target editing occurred in the cells treated with PE RNP LNP (Fig. 5i). Importantly, using PE2 RNP LNPs, we achieved a magnitude of rescue of the ERG b-wave in rd12 mice comparable to that of ABE RNP LNPs, with ERG b-wave amplitudes exceeding 300 µV (Fig. 5j).

After we characterized the LNPs in vitro and screened them in vivo for the rescue of ERG response in the rd12 mice, we applied the optimized RNP LNPs to investigate the physiological rescue of the rd12 inherited retinal degeneration phenotype. As transient exposure to genome-editing RNPs improved our editing purities in vitro, we sought to determine the residence time of ABE within the mouse eye. We injected the optimized ABE RNP LNP with SM102 and 2.5% DMG-PEG 2000 subretinally into WT mice and killed individual mice at sequential intervals to determine the kinetics of ABE degradation. We noted that ABE was detectable up to 24 h post injection in the neural retina and up to 48 h in the RPE (Fig. 6a). We then delivered our optimized ABE RNP LNP via subretinal injection into 3–4-week-old juvenile rd12 mice. Two weeks after injection, we noted an average of 0.30% precise correction of Rpe65 genomic DNA by ABE and an average of 0.12% precise correction by PE (Fig. 6b), as well as an average of 1.25% precise correction of Rpe65 transcripts by ABE and 1.28% by PE (Fig. 6c). We noted 0.67% bystander editing by ABE in the transcripts, with no indels with either ABE or PE. The base editing efficiency was similar in eyes treated with ABE RNP LNP with 1.5% DMG-PEG 2000 (not shown). We did not detect off-target editing (Supplementary Fig. 7a,b), and we did not note any editing when ABE and PE RNPs were complexed with guide RNAs that did not target the Rpe65 locus (Supplementary Fig. 8a,b). We also detected restoration of RPE65 in the RPE of the rd12 mice treated with ABE RNP LNP, according to western blot (Fig. 6d) and immunostaining of RPE flatmounts (Fig. 6f). Restoration of RPE65 function would lead to the production of 11-cis-retinal, the chromophore for photoreceptor-mediated vision. Accordingly, we assayed the retinoids in the whole eyes and detected 11-cis-retinal only in rd12 mice that were treated with either ABE or PE RNP LNPs (Fig. 6e). When we performed scotopic flash ERG on the treated rd12 animals, we observed rescue of both the a- and b-wave ERG amplitudes, both of which were not detectable in untreated rd12 animals (Fig. 6g). Again, we noted no rescue of the ERG flash response upon treatment with ABE and PE RNPs with non-targeting guides (Supplementary Fig. 8c,d). We also demonstrated restoration of intact visual signalling from the eye through the optic nerve. The pupillary light reflex (Fig. 6h) and evoked responses from the super colliculus (SC) and primary visual cortex (V1) (Fig. 6i and Extended Data Fig. 4), which both require 11-cis-retinal and intact neuronal connections, were restored in rd12 mice treated with ABE or PE RNP LNP. These results collectively demonstrate that a single dose of ABE or PE RNP LNPs can correct the Rpe65 rd12 mutation and partially restore normal physiology and biochemistry in the rd12 eye. Further development of ABE and PE RNP geared towards increased editing efficiency will result in a blueprint for chemically defined, effective RNP LNP formulations that allow for the repair of genetic mutations causing dysfunction of the RPE and other tissues.

Pigmented Gt(ROSA)26Sor^{tm4(ACTB-tdTomato,-EGFP)Luo} (ROSA^mT/mG) mice were purchased from the Jackson Laboratory (JAX 007676) and crossed with BALB/cJ albino mice (JAX 000651) to establish an albino ROSA^mT/mG line, referred to as ROSA mT/mG mice. The C57BL/6J (‘WT’, JAX 000664) and rd12 (JAX 005379) mouse lines were purchased from the Jackson Laboratory and housed in the vivarium at the University of California, Irvine, where they were maintained on a normal mouse chow diet and a 12 h/12 h light/dark cycle. All animal procedures were approved by the Institutional Animal Care and Use Committee of the University of California, Irvine, and were conducted in accordance with the NIH guidelines for the care and use of laboratory animals, and with the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Visual Research.

The N-terminal Cre recombinase fusion proteins were produced from recombinant plasmids based on pTAT-Cre⁴⁶. pTAT-Cre was purchased from Addgene (35619), and pCPP5-Cre and pN-Cre inserts were synthesized and cloned by GeneWiz and cloned into pET-28b (Supplementary Information, Plasmid DNA sequences). The antennapedia (ANTP) DNA sequence was generated by single-strand oligo annealing, using ANTP-Or_F and ANTP-Or_R primers (Supplementary Table 1). The two oligonucleotides were mixed at a 1:1 molar ratio for a total DNA concentration of 80 µM in 50 mM Tris-HCl, pH 7.4, 62.5 mM NaCl and 10 mM ethylenediaminetetraacetic acid (EDTA). The annealing was done in a thermocycler with an initial heating step of 2 min at 95 °C and gradual cooling for 45 min to 25 °C. The pANTP-Cre plasmid was obtained by replacing the TAT sequence with ANTP between the NcoI and NdeI cloning sites. The plasmids were propagated in Escherichia coli NEB-5α cells (New England Biolabs, C2987H). To produce a positive control Cre eVLP, Cre (Addgene, 123133) was subcloned into pCMV-MMLVgag-3xNES-ABE7.10-NG (Addgene, 181753) via USER cloning (NEB), and eVLPs were prepared as previously described²⁸.

Genes encoding ABE7.10, ABE8e and PE2 constructs were subcloned into pRha rhamnose-inducible expression vectors via USER cloning (NEB). The 1D4 peptide with a preceding tobacco etch virus (TEV) protease recognition site was introduced by PCR site-directed mutagenesis and KLD cloning (NEB, M0554S). Sequences encoding cell-penetrating peptides were introduced into ABE using PAGE-purified single-strand oligonucleotides (Sigma-Aldrich). The oligonucleotides were annealed at a final concentration of 25 µM in 10 mM Tris-HCl pH 8.0, 50 mM NaCl, 1 mM EDTA and 5 mM MgCl₂ by heating at 95 °C for 5 min and slow cooling in a thermocycler. Assembled duplexes were ligated at a 100-fold excess into the protein expression vectors linearized with FastDigest NdeI (Thermo Fisher, FD0584). The ligation mixtures were transformed into NEB-5α cells and plated onto LB-agar plates with 25 µg ml⁻¹ kanamycin. Candidate clones were identified using colony PCR and confirmed by Sanger sequencing (GeneWiz).

The plasmids were transformed into Escherichia coli BL21star (DE3) (Thermo Fisher, C601003), and the selected clones were grown in Terrific Broth (TB; Thermo Fisher, 22711-022) with 50 µg ml⁻¹ kanamycin (Goldbio, K-120-SL25) overnight at 37 °C with mixing at 190 r.p.m. The production cultures were inoculated with overnight cultures and incubated at 37 °C with mixing at 190 r.p.m. After absorbance at 600 nm (A_{600 nm}) reached 0.5, protein expression was induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG; Goldbio, I2481C25) and incubation was continued at 20 °C for 16 h. The cells were collected by centrifugation at 7,000 g at 4 °C for 10 min and the pellets were stored at −80 °C until used further for protein purification.

Cre recombinase was kept on ice or refrigerated in a cold room at 4–8 °C throughout the purification procedure. The cell pellet from a 1.5-l culture was thawed in a room-temperature water bath, resuspended in the Cre-lysis buffer (50 mM Na phosphate pH 7.4, 1 M NaCl, 1 cOmplete EDTA-free protease inhibitor cocktail tablet (MilliporeSigma, COEDTAF-RO) per 50 ml buffer) and lysed by sonication (125-W pulses (5 s on, 5 s off) for 10 min total) or French press (3 passes at up to 15,000 psi). The lysate was centrifuged at 48,500 g for 15 min and incubated with 1 ml of a suspension of Ni-Sepharose High Performance beads (Cytiva, 17526801) in a rotating mixer for 1 h. The resin was centrifuged at 500 g for 5 min, washed with 40 ml of the Cre-wash buffer (25 mM Na phosphate pH 7.4, 500 mM NaCl), centrifuged again, packed in a Tricorn 5/50 column, connected to a Bio-Rad DuoFlow system (Bio-Rad) and perfused at 0.5 ml min⁻¹. The resin was washed with 20 ml of the Cre-wash buffer or until a stable absorbance baseline was observed; then, the proteins were eluted with 30 ml of a continuous gradient of the Cre-elution buffer (25 mM Na phosphate, pH 7.4, containing 500 mM NaCl, 500 mM imidazole). The fractions containing Cre recombinase that were identified by SDS-PAGE and CBB staining (Quick Coomassie stain, Anatrace, GEN-QC-STAIN-1L) were concentrated and subjected to size exclusion chromatography on a Superdex 200 Increase 10/300 GL column (Cytiva, 28990944) or a HiLoad 16/600 Superdex 200 pg column (28989335), with P500G buffer (20 mM sodium phosphate, pH 7.4, 500 mM NaCl, 20% (v/v) glycerol) as the mobile phase. Fractions containing pure Cre recombinase were concentrated, snap frozen in liquid nitrogen and stored at −80 °C.

E. coli BL21star (DE3) cells (Thermo Fisher, C601003) were transformed with the expression plasmids and grown overnight on Luria-Bertani (LB)-agar plates with 25 μg ml⁻¹ kanamycin; single clones were used to inoculate starter cultures in TB with kanamycin and grown overnight. One and a half litre of TB with kanamycin was inoculated with 10 ml of starter culture, and the cells were grown at 37 °C with mixing at 190 r.p.m. until they reached an A_600nm of 1.5. Then, the cultures were cooled in an ice-water slurry for 30 min to 1 h, and protein expression was induced with 0.8% (w/v) rhamnose (Goldbio, R-105-250). The proteins were expressed at 17 °C with mixing at 190 r.p.m. for 16–24 h. The cells were collected by centrifugation at 5,000 g for 10 min at 4 °C and the pellets were stored at −80 °C.

All protein purification steps were conducted in a cold room (~4–8 °C) or on ice. The thawed cells from a 1.5-l culture were homogenized with a 40 ml Dounce homogenizer in lysis buffer (100 mM Bis-Tris propane, pH 8.0, 1 M NaCl, 20% (v/v) glycerol, 5 mM tris(2-carboxyethyl)phosphine (TCEP), 1 cOmplete Ultra EDTA-Free protease inhibitor tablet per 40 ml of the buffer) and lysed by sonication with a Qsonica Q125 sonicator (125 W) with a 1/8-inch microtip at 100% amplitude for a total of 20 min (intermittent pulses: 5 s on, 5 s off). The lysate was clarified two times at 4 °C by centrifugation at 48,500 g for 10 min. ABE was captured on a 3-ml TALON metal-affinity resin (Takara Bio, 635502). The resin was then washed with 100 ml of the lysis buffer without inhibitors and with 10 ml of the wash buffer (100 mM Bis-Tris propane, pH 8.0, 500 mM NaCl, 20% (v/v) glycerol, 1 mM TCEP). The proteins were eluted with the wash buffer supplemented with 150 mM imidazole.

In a second step, ABE was purified by immunoaffinity chromatography using a 1D4 resin (Sepharose CL-4B resin; Cytiva, 17043001) with immobilized 5–10 mg ml⁻¹ 1D4 antibody purified in house. The 1D4 resin (4 ml) was packed in a DWK Life Sciences Kimble Kontes FlexColumn. The column was equilibrated with the wash buffer (described above). ABE was loaded by gravity flow. The column was washed with at least 40 ml of the wash buffer at 0.5 ml min⁻¹, and then ABE was eluted with 1 mg ml⁻¹ 1D4 peptide in the wash buffer at ~1 ml h⁻¹. Fractions containing ABE were pooled, concentrated and further purified to remove contaminating nucleic acids and aggregates, by size exclusion chromatography on a Superdex 200 Increase 10/300 GL column or a Superdex 200 16/60 Prep Grade column. The protein was eluted with the ABE storage buffer (10 mM 4-(2-hydroxyethyl)-1-piperazine ethanesulfonic acid (HEPES), pH 7.0, 500 mM NaCl, 20% glycerol). The fractions containing pure ABE were concentrated using Amicon Ultra centrifugal filters with 30-kDa molecular weight cut-off membranes (Merck, UFC903024), sterilized by passage through 0.22-μm filters, quantified using UV absorption at 280 nm, divided into aliquots and snap frozen in a chilled metal block for storage at −80 °C.

PE was purified using a similar procedure as for ABE. Cells expressing PE were lysed using a single pass through the French press and subsequent sonication for 10 min. Metal affinity and immunoaffinity steps were identical as for ABE, except that the wash buffer contained 400 mM NaCl. After immunoaffinity chromatography, PE was subjected to heparin-affinity chromatography (HiTrap Heparin HP 5 ml; Cytiva, 17040703) at 0.5 ml min⁻¹. The heparin column was equilibrated with the wash buffer containing 400 mM NaCl. After PE was loaded, the column was washed with at least 50 ml of the wash buffer until the UV baseline was stable, and then bound proteins were eluted with a 200-ml NaCl gradient (0.4 M–1.0 M). Purest fractions were selected for size exclusion chromatography on a Superdex 200 16/60 Prep Grade column, with PE-storage buffer (10 mM HEPES, pH 7.0, 500 mM NaCl, 5% (v/v) glycerol) as the eluent. Pure PE was concentrated, filtered, quantified using UV absorption at 280 nm, distributed into aliquots and snap frozen for storage at −80 °C.

The protein purity was assessed using SDS–PAGE in hand-cast Tris-glycine-SDS discontinuous gels with 4% acrylamide in a stacking gel (pH 6.8) and 10% acrylamide in a resolving gel (pH 8.8) with 2.7% crosslinker (acrylamide:bis-acrylamide ratio of 37.5:1; Bio-Rad, 1610158). The samples were mixed with 4×-concentrated Laemmli sample buffer (Bio-Rad, 1610747) and supplemented with 50 mM dithiothreitol (DTT; MilliporeSigma, D9779), denatured at 70 °C for 10 min and centrifuged before applying on gel. The protein concentration was quantified using UV absorption spectroscopy by measuring absorbance at 280 nm with a Nanodrop ND-1000 spectrophotometer. Separation of contaminating nucleic acids was followed by monitoring the ratio of absorbance at 260 nm to absorbance at 280 nm; a ratio <0.60 was used as an indication of protein free from nucleic acids. A typical A_{260 nm}/A_{280 nm} value for purified ABE and PE was 0.55. Extinction coefficients of the constructs were estimated using the ProtParam tool (https://web.expasy.org/protparam/↗).

Synthetic 60-bp-long DNA oligonucleotides (Sigma) labelled with fluorescein on the strand undergoing deamination were annealed at a 1:1 ratio in 10 mM Tris pH 8.0, 50 mM NaCl and 1 mM EDTA by incubation at 95 °C for 5 min and subsequent slow cooling to 20 °C. End-modified sgRNA and epegRNA were ordered from IDT Technologies, with 2′-O-methyl groups on the first three and last three nucleotides, and the first and last three phosphodiester bonds were replaced with phosphorothioate bonds. The guide RNAs were dissolved in nuclease-free water at 37 °C for 15 min at 500 r.p.m. in a thermomixer and folded by incubation at ~75 °C for 5 min, followed by slow cooling. Prepared nucleic acids were quantified using UV absorption spectroscopy by measuring absorbance at 260 nm. Accordingly, an A_260nm value of 1.0 corresponded to a DNA concentration of 50 µg ml⁻¹ or an RNA concentration of 40 µg ml⁻¹. DNA was stored at −20 °C and RNA at −80 °C.

For the ABE enzymatic assay, ABE ribonucleoprotein was assembled by incubation with 1.5-fold molar excess of sgRNA in a reaction buffer (20 mM Bis-Tris propane, pH 7.5, 100 mM KCl, 2.5 mM MgSO₄, 2 mM DTT, 5% (v/v) glycerol) for 15 min at room temperature. Additional 10% (w/v) sucrose was used when RNPs were assembled above 4 µM. Then, ABE was diluted to 1 µM with the reaction buffer, preheated at 37 °C and added with 15 ng of DNA substrate (0.02 µM final). The deamination was conducted for 10–60 min for ABE8e. The 20-µl reactions were quenched by addition of 30 µl of water preheated to 95 °C and incubated for 2 min at 95 °C. After cooling, the mixtures were treated with 1 µl of RNase A (20 mg ml⁻¹) and 1 µl of proteinase K (20 mg ml⁻¹) for 15 min at room temperature; then DNA products were purified using an Oligo Clean and Concentrator kit (Zymo Research, D4061). Purified DNA was nicked with 5 units of Endonuclease V (EndoV, NEB) for 2–3 h at 37 °C, after which the reaction was quenched by addition of TriTrack DNA-loading dye (1× final) (Thermo Fisher, R1161) and incubation at 95 °C for 2 min. The cleavage products were analysed by denaturing polyacrylamide gel electrophoresis with urea (Urea–PAGE) in Bio-Rad MiniProtean continuous hand-cast 15% acrylamide gels in Tris-borate-EDTA (TBE; Bio-Rad, 1610770) with 7 M urea and 5% crosslinker (acrylamide:bis-acrylamide ratio of 19:1; Bio-Rad, 1610144). The voltage was controlled to maintain at least 42 °C in the electrophoresis chamber. Imaging was done using the ChemiDoc MP system (Bio-Rad). For fluorescein, it was done immediately after electrophoresis and for the SYBR Gold (Thermo Fisher, S11494), after 30 min staining with 1× dye in 1×TBE.

For the in vitro reverse-transcriptase elongation assay of PE, the activity assay was carried out in the same reaction buffer as for ABE, supplemented with 0.5 mM deoxyribonucleotides (dNTP). PE2 RNP was assembled with 1.1-fold molar excess of epegRNA, targeting the rd12 locus in the reaction buffer without added sucrose for 15 min at room temperature. Then, reactions were preheated at 37 °C and 45 ng of fluorescein-labelled substrate was added to a final volume of 15 µl. After 15 min, the reactions were quenched with 1 µl each of proteinase K (20 mg ml⁻¹) and RNase A (20 mg ml⁻¹), denatured at 95 °C for 2 min and supplemented with 6×TriTrack loading dye, and 15 ng of substrate was analysed by Urea–PAGE, as for ABE.

ABE and PE proteins were complexed with 1.5-fold molar excess of guide RNA in phosphate-buffered saline (PBS) at room temperature for 15 min to obtain 10 µM RNP. ABE RNP contained an additional 10% (w/v) sucrose. Subsequently, the RNPs were diluted with PBS to 1 µM, SYPRO Orange probe was added to a final concentration of 5×, the samples were aliquoted (10 µl per well) into a 384-well plate (Applied Biosystems, 4483319) and the plate was sealed with optical foil (Applied Biosystems, 4360954). After an additional 15 min incubation, the plate was centrifuged at 1,000 g for 1 min at room temperature and installed in a pre-equilibrated Bio-Rad CFX384 thermocycler. Fluorescence was measured in Förster resonance energy transfer (FRET) mode every 0.2 °C from 20 °C to 95 °C. The rate of change of fluorescence (−dF/dT) was used to estimate the melting temperature (T_m). All samples were run in triplicate and plots of averaged data are reported.

HEK293-loxP-GFP-RFP cells (referred to as ‘CS cell line’; GenTarget, SC018-Bsd), NIH/3T3 rd12 cells⁴¹ and rd12 reporter cells were maintained in DMEM/F12 medium with GlutaMAX supplement (Thermo Fisher, 10565018) or in DMEM with glutamine (Thermo Fisher, 11965092), both supplemented with 10% FBS (Genesee Scientific, 25-514H) and optional 100 U ml⁻¹ penicillin-streptomycin (Thermo Fisher, 15140122) (complete medium) in a humidified incubator at 37 °C and 5% CO₂.

Primary fibroblasts were isolated from the skin of P0-P3 ROSA mT/mG mice. The mice were euthanized, the skin separated and washed with PBS containing 100 U ml⁻¹ penicillin-streptomycin and 40 µg ml⁻¹ gentamicin (Thermo Fisher, 15710072). The skins were digested in a 100-mm cell culture dish using a 1:1 mixture of 0.25% trypsin without EDTA (Thermo Fisher, 15050065) and 5 U ml⁻¹ dispase (STEMCELL Technologies, 07913) for 1 h at 37 °C, after which the dermis was collected to a new dish and digested with 0.25% collagenase I (Thermo Fisher, 17018029) in serum-free DMEM/F12 for 1 h at 37 °C. The tissue fragments were thoroughly resuspended, filtered through a 70-µm strainer and extensively washed with DMEM/F12 with 15% FBS and 100 U ml⁻¹ penicillin-streptomycin by two centrifugation steps (180 g, 5 min, room temperature). The cells were maintained in DMEM/F12 with 15% FBS and passaged every 3–4 days using 0.05% trypsin-EDTA (Thermo Fisher, 25300054). After the first passage, the cells were filtered through a 40-µm strainer to remove undigested aggregates that permeated through the 70-µm strainer. Both freshly isolated and cryopreserved cells that were passaged at least twice were used for the experiments. For Cre delivery, medium with 10% FBS was used and the procedure was the same as for the CS cell line. For protein delivery experiments, the cells were seeded 24 h before the experiment in 24-, 48- and 96-well plates to reach 50–70% confluency (~100,000, 50,000 and 25,000 cells per well, respectively, for CS cell line and NIH/3T3 rd12 cells, and 50,000, 24,000 and 10,000 cells per well, respectively, for rd12 reporter cells and primary fibroblasts).

The CS cells and primary ROSA mT/mG fibroblasts were plated on 48-well treated tissue-culture plates in complete medium with 10% FBS. Cre recombinase, fused separately with each of the cell-penetrating peptides (CPPs: CPP5 (KLPVM)⁴⁷, TAT (RKKRRQRRR)⁹¹ or ANTP (RQIKIWFQNRRMKWKK))⁹², was prepared at various final concentrations (0.10, 0.25, 0.50, 0.75 or 1.00 µM; 4.2, 10.5, 21.0, 31.4 or 41.9 µg ml⁻¹, respectively) in 250 µl each of OptiMEM medium (Thermo Fisher, 31985070). Only the 0.5 µM Cre proteins were tested in the primary fibroblasts. For non-covalent complexation with 6×His-CM18-PTD4 peptide (Genscript)⁴⁹, an aliquot of the non-covalent peptide was added to Cre recombinase (free or fused with one of the CPPs) and incubated for 15 min at room temperature. The reporter cells were washed with 180 µl of OptiMEM medium, which was then exchanged with OptiMEM plus one of the various preparations of Cre recombinase. After 3 h, the medium was exchanged with complete medium. The CS cells and ROSA mT/mG fibroblast cells were maintained post treatment for a total of 24 h and 72 h, respectively. For imaging, the medium was exchanged to FluoroBrite DMEM (Thermo Fisher, A1896701) and the cells were imaged using a Keyence BZ-X810 microscope with GFP and Texas Red optical filters.

Cells were washed with PBS (Thermo Fisher, 10010023), detached with 0.05% trypsin (Thermo Fisher, 25300054), transferred to a 96-well round-bottom plate and centrifuged at 180 g for 5 min at room temperature. Centrifuged cells were washed with a FACS buffer (PBS with 2% FBS, 100 U ml⁻¹ penicillin-streptomycin), centrifuged again and resuspended in the FACS buffer with 1 µg ml⁻¹ 4′,6-diamidino-2-phenylindole (DAPI; Thermo Fisher, 62248). The cells were analysed using a Novocyte Quanteon (Agilent) flow cytometer with Pacific Blue (445/45 nm), FITC (530/30 nm) and PE (586/20 nm) optical filters. Cells were gated on forward and side scatter, viability via DAPI exclusion and single cells (Supplementary Fig. 9).

Cre proteins with or without fused cell-penetrating peptides were diluted in OptiMEM to 10 µM. 6×His-CM18-PTD4 peptide was optionally added to Cre in 10-fold excess and incubated for at least 15 min before injection. A lipoplex of Cre with Lipofectamine 3000 was injected at 10 µM, with 2% Lipofectamine 3000 by volume (419 ng Cre per eye, 0.02 µl Lipofectamine 3000 per eye). AAV2/1-Cre (Addgene, 105537-AAV1; 1.8 × 10¹⁰ vg) or vesicular stomatitis virus G glycoprotein (VSV-G) pseudotyped eVLP-Cre (concentrated by ultracentrifugation 1,000-fold as previously described²⁸) were injected as a positive control for ROSA mT/mG mice. Mice were anaesthetized by intraperitoneal injection of a cocktail consisting of 20 mg ml⁻¹ ketamine and 1.60 mg ml⁻¹ xylazine in PBS at a dose of 100 mg kg⁻¹ of ketamine and 8 mg kg⁻¹ of xylazine, and their pupils were dilated by topical administration of 1% tropicamide ophthalmic solution (Akorn, 17478-102-12) and 10% phenylephrine ophthalmic solution (MWI Animal Health, 054243). The corneas were hydrated with GenTeal Severe Lubricant Eye Gel (0.3% hypromellose, Alcon). Subretinal injections were performed using an ophthalmic surgical microscope (Zeiss). An incision was made through the cornea adjacent to the limbus at the nasal side using a 27-gauge needle. A 34-gauge blunt-end needle (World Precision Instruments, NF34BL-2) connected to an RPE-KIT (World Precision Instruments, RPE-KIT) with SilFlex tubing (World Precision Instruments, SILFLEX-2) was inserted through the corneal incision while avoiding the lens and advanced into the subretinal space. Each mouse received a 1-μl injection per eye, and volume and rate were controlled with a UMP3 UltraMicroPump (World Precision Instruments, UMP3-4). After surgery, the mice were placed on a heating pad and anaesthesia was reversed with intraperitoneal atipamezole (2.5 mg kg⁻¹; MWI Animal Health, 032800). Triple antibiotic ophthalmic ointment (neomycin, polymyxin and bacitracin) was administered to the cornea to promote recovery.

After killing, intact enucleated mouse eyes were submerged in room temperature PBS. Pulsing infrared light from a Ti:sapphire laser (Coherent, Vision S; tunable between 690 and 1,050 nm) was set to 950 nm and attenuated in a controlled, variable manner with an electro-optic modulator. To image and spectrally separate GFP and tdTomato, two internal spectral detectors were used with their detection bandwidths set to 490–545 nm for GFP and 590–680 nm for tdTomato. A 1.0 NA ×20 Leica objective was used for the imaging^93,94.

The rd12 reporter construct was synthesized by GenScript according to the following strategy: 198 bp of the mouse Rpe65 cDNA was flanked by 5’-mCherry and 3’-eGFP, and the whole construct was inserted into the pcDNA3.1/(Zeo)+ backbone with BamHI and XhoI restriction sites (mCherry-rd12-eGFP). The mCherry-rd12-eGFP construct was then subcloned into pMXs-IRES-blasticidin via double digestion of the backbone with BamHI and XhoI. The downstream sequence of the internal ribosomal entry site (IRES) and blasticidin-resistance gene enabled co-expression of the reporter and selectable marker. The rd12 reporter cell line was generated by transduction of NIH/3T3 cells with retrovirus obtained from Phoenix-Eco cells transfected with pMXs-mCherry-rd12-eGFP-IRES-blasticidin^95,96 according to a previously published protocol⁹⁷. Transduced cells were then selected with blasticidin for 10 days (5 μg ml⁻¹; Thermo Scientific, R25001↗). The surviving cells were sorted using flow cytometry to select high-expressing mCherry-positive clones and then seeded into 96-well plates for clonal selection. Single colonies were screened for proper expression and editability via Gene Juice (MilliporeSigma, 70967-3) co-transfection of pCMV-NG-SpCas9-ABE7.10max and pSPgRNA-rd12-A6, and via proper co-expression of mCherry and eGFP. Finally, the clones were sequenced. No additional characterization was performed on the sorted cells.

ABE RNP was assembled by incubation of up to 20 µM ABE for at least 15 min at room temperature with 1.1-fold excess of sgRNA in OptiMEM medium supplemented with 10% (w/v) sucrose. The sgRNA used for experiments in cell lines and in vivo contained additional modifications as described (Supplementary Information, Supplementary Sequences, rd12-A6-sgRNA highly modified) and was supplied by IDT⁷⁵. The 6×His-CM18-PTD4 peptide was dissolved in 100 mM HEPES pH 8.0 to achieve a peptide concentration of 10 mM; the final pH of this stock solution was ~7.0. The 6×His-CM18-PTD4 peptide was diluted to 400 µM with water, added to ABE RNP and incubated for an additional 15 min. CPP fusions of ABE with or without 6×His-CM18-PTD4 peptide were diluted to their final concentrations in OptiMEM containing 2% or 10% sucrose, and 100 µl of each of the mixtures was applied on the cells in a 48-well plate that had been washed with OptiMEM medium. After 3 h, the medium was exchanged with 250 µl of complete medium.

To assemble Lipofectamine 3000 lipoplexes, 0.5 µl of Lipofectamine 3000 per well of a 48-well plate or 0.2 µl per well of a 96-well plate was diluted with OptiMEM containing 10% (w/v) sucrose to 12.5 and 5 µl per well, respectively; the ABE RNPs diluted with OptiMEM with 10% (w/v) sucrose to 12.5 µl (48-well) or 5 µl (96-well) were then added to the diluted lipids and incubated for 15 min at room temperature. A lipoplex of ABE with Lipofectamine 3000 in a volume of 25 µl was added to the cells with 225 µl fresh complete medium in a 48-well plate, or 10 µl was added to the cells with 90 µl medium in a 96-well plate. The cells were incubated for 48 h before analysis. Plasmid DNA transfections were done in 48-well plates using 160 ng pCMV-NG-SpCas9-ABE, 80 ng pSPgRNA-rd12-A6 and 0.75 µl Lipofectamine 3000 per well, following manufacturer protocol. PE was delivered in vitro, similarly as ABE, but without sucrose, unless otherwise noted. Activities of ABE and PE in the rd12 reporter cell line were analysed by fluorescence microscopy and flow cytometry, as described above for Cre delivery.

Rescue of RPE65 expression was analysed using NIH/3T3 cells stably expressing Rpe65 rd12 cDNA, as previously described⁴¹. The ABE and PE were applied on these cells, and after 48 h, the cells were detached using trypsin and washed three times with PBS. One-tenth of the cell suspension was lysed with 10 mM Tris pH 7.5 and 0.05% SDS with 0.02 mg ml⁻¹ proteinase K for 1 h at 37 °C, then the proteinase K was inactivated at 85 °C for 45 min. The lysate was used as a template for PCR and the products were subjected to next-generation sequencing. The remaining cells were lysed in 1× RIPA buffer (Cell Signaling Technology, 9806) with 1× cOmplete Ultra EDTA-free protease inhibitors (Roche) for 1 h on a rotator in a cold room and then centrifuged at 17,000–21,400 g for 20 min at 4 °C, and the supernatant was used for analysis. Protein concentration in the extracts was measured using the BCA assay (Thermo Fisher, 23252) and the extracts were subjected to western blotting.

Complementary DNA was synthesized from RNA with the High Capacity RNA-to-cDNA kit (Thermo Fisher, 4387406) according to manufacturer instructions. Of the isolated genomic DNA or cDNA, 0.5–1 μl was used as input for the first of two PCR reactions (PCR1). Genomic loci were amplified in PCR1 using Phusion Plus polymerase (Thermo Fisher, F631S). PCR1 primers for genomic loci are listed in Supplementary Table(marked as HTS_fwd and HTS_rev). PCR1 was performed as follows: 98 °C for 30 s; 30 cycles at 98 °C for 10 s, 60 °C for 20 s and 72 °C for 30 s; 72 °C for 5 min. PCR1 products were confirmed on a 2% agarose gel. One microlitre of PCR1 was used as input for PCR2 to install Illumina barcodes. PCR2 was conducted for 9 cycles of amplification using a Phusion HS II kit (Life Technologies). Following PCR2, samples were pooled and gel purified in a 1% agarose gel using a Qiaquick Gel Extraction kit (Qiagen). Library concentration was quantified using the Qubit High-Sensitivity Assay kit (Thermo Fisher). Samples were sequenced on an Illumina MiSeq instrument (paired-end reads, read 1: 200–280 cycles, read 2: 0 cycles) using an Illumina MiSeq 300 v.2 kit (Illumina). 2

Sequencing reads were demultiplexed using the MiSeq Reporter software (Illumina) and analysed using CRISPResso2 as previously described^27,98. Batch analysis mode (one batch for each unique amplicon and sgRNA combination analysed) was used in all cases. Reads were filtered according to minimum average quality score (Q > 30) before analysis. The following quantification window parameters were used: -w 20 -wc -10. Base editing efficiencies are reported as the percentage of sequencing reads containing a given base conversion at a specific position. Prism 10 (GraphPad) was used to generate dot plots and bar plots.

Protein was analysed via western blotting and detected using mouse anti-RPE65 antibody (produced in house)⁹⁹ or mouse anti-SpCas9 (clone 7A9, Biolegend, 844302) diluted 1:1,000 in 5% non-fat dry milk in PBS (Bio-Rad, 1610780) with 0.1% Tween 20 (MilliporeSigma, P9416) (PBST). The lysates (7–10 µg total protein) were separated by SDS–PAGE, transferred to a polyvinylidene fluoride membrane (MilliporeSigma, IPVH00010), blocked for 1 h with 5% non-fat dry milk in PBST at room temperature and incubated with the primary antibodies overnight in the cold room. After washing 4 times with PBST for 5 min each, the blots were incubated for 1 h at room temperature with a horseradish peroxidase (HRP)-linked horse anti-mouse antibody (Vector Laboratories, PI-2000-1) diluted 1:2,500 in 5% non-fat dry milk in PBST. The signals were detected with SuperSignal West Pico Plus Chemiluminescent substrate (Thermo Fisher, 34577). Next, the antibodies were stripped from the membrane with 0.2 M glycine pH 2.2, 0.1% SDS and 1% Tween 20; then, the membranes were washed, blocked and re-probed for 1 h at room temperature using rabbit anti-β-actin polyclonal antibody (1:2,000; Cell Signaling Technology, 4970S). Goat anti-rabbit IgG with HRP (1:2,500; Cell Signaling Technology, 7074S) was used as the secondary antibody before developing the blots, as described above.

ABE was diluted into OptiMEM containing additional sucrose (10 or 25% (w/v) final, as indicated); or into the ABE storage buffer containing high salt concentration (390 mM NaCl final) without added sucrose. To form RNPs, ABE was added to guide RNA dissolved in water and incubated for 15 min at room temperature. For combination with 5-fold molar excess of 6×His-CM18-PTD4 peptide, ABE RNPs were added to a peptide stock solution diluted to 1 mM. In the case of Lipofectamine 3000, ABE RNPs were added to the undiluted reagent. RNPs were incubated with the reagents for at least 15 min at room temperature before subretinal injection into rd12 mice as described.

Mouse eyes were dissected under a light microscope to separate the posterior eyecup (containing RPE, choroid and sclera) from the retina and anterior segment. Each posterior eyecup was immediately immersed in RLT Plus (Qiagen). RPE, choroid and scleral cells were detached from the posterior eyecup by gentle pipetting, followed by removal of the remaining posterior eyecup. Cells were then processed for genomic DNA and RNA using the AllPrep DNA/RNA Micro kit according to manufacturer instructions (Qiagen, 80284). To prepare the protein lysate from the mouse RPE tissue, the dissected mouse posterior eyecup was transferred to a microcentrifuge tube containing 40 μl of ice-cold RIPA buffer with protease inhibitors and homogenized with a motorized tissue grinder (Fisher Scientific, K749540-0000), incubated on ice for 20 min and then centrifuged for 20 min at 21,000 g at 4 °C. The resulting supernatant was pre-cleared with Dynabeads Protein G (Thermo Fisher, 10003D) by rotation at 4 °C for 15 min to remove immunoglobulin contaminants from blood before loading on the gel.

Mouse eyes were enucleated and fixed with 4% paraformaldehyde in PBS for 20 min at room temperature and washed three times in PBS for 5 min each. To make RPE flatmounts, the anterior segment and retina were removed from the posterior eyecup under a dissecting microscope, and four radial cuts were made towards the optic nerve to flatten the eyecup into an RPE flatmount. Samples were permeabilized and blocked in 0.1% Triton X-100 (Sigma-Aldrich, T8532) with 3% normal goat serum (NGS) in PBS for 30 min and incubated with the following primary antibodies in PBS, 0.1% Triton X-100 and 3% NGS: mouse anti-RPE65 antibody (1:100; in house) and rabbit anti-ZO-1 polyclonal antibody (1:100; Invitrogen, 61-7300) overnight at 4 °C. The next day, samples were washed three times in PBS for 5 min each and then incubated with the appropriate secondary antibodies in PBS, 0.1% Triton X-100 and 3% NGS, including Alexa Fluor 555-conjugated goat anti-mouse IgG (1:200; Thermo Fisher, A11032) and Alexa Fluor 647-conjugated goat anti-rabbit IgG (1:200; Thermo Fisher), for 2 h at room temperature in the dark. The secondary antibodies were then removed and the flatmounts were incubated in DAPI (1 µg ml⁻¹; Thermo Fisher, 62248) in PBS for 10 min. Samples were washed three times in PBS for 5 min each. The samples were then mounted with VECTASHIELD HardSet Antifade Mounting Medium (Vector Labs, H-1400-10) and imaged on a Keyence BZ-X810 All-in-One fluorescence microscope.

Before ERG recording, mice were dark adapted for 1 week. Under a safety light, mice were anaesthetized by isoflurane inhalation, and their pupils were dilated with topical administration of 1% tropicamide ophthalmic solution (Akorn, 17478-102-12) and 10% phenylephrine ophthalmic solution (MWI Animal Health, 054243), followed by hypromellose (Akorn, 9050-1) for hydration. The mouse was placed on a heated Diagnosys Celeris rodent-ERG device (Diagnosys). Ocular stimulator electrodes were placed on the corneas, the reference electrode was positioned subdermally between the ears, and a ground electrode was placed in the rear leg. The eyes were stimulated with a green-light flash stimulus (peak emission 544 nm, bandwidth ∼160 nm) of −0.3 log (cd s m⁻²) light intensity. The responses for 10 stimuli with an inter-stimulus interval of 10 s were averaged, and the a- and b-wave amplitudes were acquired from the averaged ERG waveform. Data were analysed with the Espion V6 software (Diagnosys). For the RNP LNP optimization study, eyes that received LNP and had no ERG response after treatment were excluded from the analyses.

The PLR was characterized in mice (n = 5 for each group) using the A2000 computerized pupillometer (Neuroptics). Mice were dark adapted for 6 h before recordings in a dark room. This pupilometer system consists of a sensing device equipped with two infrared cameras that independently record and track dynamics of each pupil. The light profile consisted of four white-light stimuli (1.2 log, 10^1.2 W m⁻²), each for 500 ms. For the duration of the PLR testing routine, mice were kept under isoflurane anaesthesia. The experiments were carried out under scotopic conditions, with no background illumination from the pupillometer, with the infrared cameras as the primary light source. The maximum size of the pupil after dark adaptation was quantified at 2 min after anaesthesia and was used to establish baseline size. The pupil response was expressed as percent constriction of the pupil when compared to baseline. Captured digital movies of pupil responses were recorded using the Active Presenter software (v.9.1.3, Atomi Systems), and the videos were subsequently decomposed into individual frames using the Adobe Premiere Rush programme (v.10.0.1, Adobe Systems) for manual verification of pupil dynamics and calculation of absolute pupil diameters from the recorded images.

Mice were dark adapted for 2 days before eye enucleation. Eyes were homogenized in 1 ml of a 10 mM sodium phosphate buffer (pH 8.0) containing 50% (v/v) methanol (Sigma-Aldrich, 34860-1L-R) and 100 mM hydroxylamine (pH 8.0) (Sigma-Aldrich, 159417-100G). After 15 min incubation at room temperature, 2 ml of 3 M NaCl was added. The resulting sample was extracted twice with 3 ml ethyl acetate (Fisher Scientific, E195-4). Then, the combined organic phase was dried in vacuo and reconstituted in 250 μl hexane. Extracted retinoids (100 µl) were separated on a normal-phase HPLC column (Zorbax Sil, 5 µm, 4.6 mm × 250 mm; Agilent Technologies) connected to an Agilent Infinity 1260 HPLC system equipped with a diode-array detector. Separation was achieved with a mobile phase of 0.6% ethyl acetate in hexane (Fisher Scientific, H302-4) at a flow rate of 1.4 ml min⁻¹ for 17 min, followed by a step increase to 10% ethyl acetate in hexane for an additional 25 min. Retinoids were detected by monitoring absorbances at 325 nm and 360 nm using Agilent ChemStation software.

RNPs were assembled by mixing 10 μM purified ABE or PE with synthetic sgRNA (ABE) or epegRNA (PE) at a 1:1.1 molar ratio. The buffer for ABE was 10 mM HEPES (pH 7.0), 500 mM NaCl, 20% (v/v) glycerol; for PE, the buffer was 10 mM HEPES (pH 7.0), 500 mM NaCl, 5% (v/v) glycerol; and guide RNA was dissolved in water. Proteins were diluted, supplemented with 10% (w/v) sucrose from 50% (w/v) stock in water and added to guide RNA. Final composition of the buffer in which RNPs were assembled was 2.8 mM HEPES (pH 7.0), 140 mM NaCl, 10% (w/v) sucrose, 5.6% (v/v) glycerol (ABE); or 1.4% (v/v) glycerol (PE). The RNPs were incubated at room temperature for at least 15 min. Transient turbidity, which cleared during incubation, was observed. Immediately before encapsulation, RNPs were diluted to 0.711 μM in a 50 mM Tris-acetate buffer (pH 6.0) with 10% (w/v) sucrose to achieve a final NaCl concentration of 10 mM. The lipids used to encapsulate pre-formed RNPs comprised an ionizable cationic lipid (either CL4H6 (Cayman Chemical, 37279), SM102 (Broadpharm, BP-25499) or DODMA (Avanti Polar Lipids, 890899), all of which have pK_as > 6.0 at 6.25, 6.68 and 6.59, respectively) and co-lipids DSPC (Avanti Polar Lipids, 850365), cholesterol (Avanti Polar Lipids, 700100) and DMG-PEG 2000 (Avanti Polar Lipids, 880151) (Supplementary Fig. 5) at a molar composition of 50/10/38.5/1.5, respectively, to encapsulate pre-formed RNPs. For optimization of DMG-PEG 2000 lipid content from 1.5 to 10 mole%, the concentration of cholesterol was decreased accordingly. In some formulations, 2.5 mole% of SOPS (18:0–18:1 PS, Avanti Polar Lipids, 840039C) was used with 7.5 mole% of DSPC. In brief, the lipids were dissolved in ethanol and rapidly combined with pre-formed RNP at a volume ratio of 1:3 (ethanol:aqueous) and a total lipid:guide RNA weight ratio of 40:1 (approximate total lipid:protein weight ratio of 7.75:1). The combination was performed by microfluidic mixing using a Precision NanoSystems Ignite device (Precision NanoSystems). Immediately after mixing, the formed LNPs were dialysed two times for 2 h each at room temperature against 20 mM Tris, 4.3 mM Na acetate (pH 7.4) and 10% (w/v) sucrose (TAS buffer) to remove the ethanol and deprotonate the ionizable cationic lipid at neutral pH. Final concentration of protein was ~0.53 µM (0.10 mg ml⁻¹ ABE8e, 0.13 mg ml⁻¹ PE), and that of guide RNA was 0.59 µM (0.02 mg ml⁻¹ sgRNA, 0.03 mg ml⁻¹ epegRNA). The LNPs were transferred to ice and concentrated to no less than one-fifth of the initial volume using an Amicon Ultra centrifugal filter with molecular weight cut-off of 30 kDa (Merck, UFC903024); the concentrated LNPs were distributed into aliquots, quickly frozen on a pre-cooled metal block and stored at −80 °C. The percent recovery and concentration factor of the LNPs were estimated using SDS–PAGE electrophoresis and CBB staining. Fluorescence intensity of Coomassie dye bound to protein was measured using ChemiDoc MP imager and analysed using ImageLab software v.6.1.0 (Bio-Rad). Doses of RNP LNP are reported in vitro as a final concentration of RNP and in vivo as RNP concentration and injected volume.

Particle-size distribution was measured using a Malvern Zetasizer Advance Nano (Malvern Panalytical). Twenty microlitres of freshly dialysed LNP were diluted to 200 μl with TAS buffer and subjected to particle-size distribution measurement in triplicates. The particle size distributions were processed by the accompanying software to calculate average particle diameter and polydispersity index (PdI).

LNP samples were concentrated 3–4 times in an Amicon Ultra 0.5 device with a 10 kDa molecular weight cut-off. LNP solution (2.5 μl) was applied onto a Quantifoil 200 mesh grid coated with a thin carbon film (Ted Pella). Grids were blotted for 2 s with filter paper, then plunged into liquid ethane using a manual plunger. The image was collected on an FEI Tecnai TF20 high resolution transmission electron microscope equipped with a K2 Direct-Detection Camera at an accelerating voltage of 200 kV.

LNPs were diluted 1:3 in an immunoprecipitation buffer (10 mM HEPES (pH 7.0), 150 mM NaCl, 10% (w/v) sucrose), and 80 μl of diluted LNPs were incubated with 20 μl of 1D4 resin for 30 min in a cold room in an overhead mixer. As a control, pre-formed free RNP was diluted in the TAS buffer to the approximate concentration of RNP in a prepared LNP and also incubated with 20 μl of 1D4 resin. Samples were then centrifuged at 700 g for 5 min at 4 °C. The supernatant was filtered with centrifugation at 200 g for 2 min through a 30-µm polyethylene filter. Pelleted resin was resuspended in 500 μl of the immunoprecipitation buffer and washed by centrifugation at 700 g for 5 min. The supernatant was discarded and the washing process was repeated four separate times to ensure thorough washing. To elute the RNP, 80 μl of the elution buffer (1 mg ml⁻¹ of 1D4 peptide in the immunoprecipitation buffer) was introduced to the resin and the samples were incubated overnight in a cold room in an overhead mixer. After incubation, the samples were centrifuged and filtered as described above. Filtered samples were then analysed by western blot, as described.

The LNPs containing ABE8e RNP or free ABE8e RNP (112 µg RNP in both) were diluted into 500 µl of 1× PBS (Corning, 46-013-CM) with 0.001% Pluronic F-68 (Gibco, 24040032), filtered on a pre-washed 0.22-µm cellulose acetate centrifugal filter (Corning, 8160) and resolved on a HiPrep 16/60 Sephacryl S-500 HR column (Cytiva, 28935606) at a flow rate of 0.4 ml min⁻¹. One millilitre fractions were collected and analysed by western blot with anti-Cas9 antibodies as described.

Deionized water in all experiments was generated using a Milli-Q water-purification system (Millipore). Formic acid (FA), ammonium bicarbonate and acetonitrile of MS grade were purchased from Fisher chemical. Iodoacetic acid and DTT were of analytical grade and supplied by Millipore. Sequencing-grade modified trypsin was provided by Promega. Stable-isotope-labelled peptides (SIL peptides) were synthesized with alkylated cysteines by GenScript. The stock solutions of all the peptides were prepared by accurately weighing the synthetic peptides and then dissolving them in water or dimethylsulfoxide following manufacturer instructions. The SIL peptides were diluted in water before adding to the samples (Supplementary Table). 3

The samples were diluted with 50 mM ammonium bicarbonate and reduced with 10 mM DTT for 1 h at 56 °C and alkylated with 20 mM iodoacetic acid for 30 min at room temperature in the dark. Then, the SIL peptides were spiked into protein samples, and then free trypsin was added at a trypsin to protein ratio of 1:50 and incubated overnight at 37 °C. Trypsin activity was inhibited by acidification with 0.1% FA and the samples were then desalted using a C18 desalting column (Nest). After drying completely by speed vacuum, peptides were dissolved in 0.1% FA. The samples were analysed by LC–MS/MS using a Vanquish LC instrument (Thermo Fisher) coupled in-line with a Q Exactive mass spectrometer (Thermo Fisher) with an ESI source. Mobile phase A was composed of 0.1% FA in water and mobile phase B was composed of 0.1% FA in acetonitrile. The total flow rate was 0.4 ml min⁻¹. Peptides were separated with a 25-min gradient on an Acquity UPLC BEH C18 column (1.7 μm, 2.1 mm × 100 mm; Waters). The acquisition method combined a full scan method with a time-scheduled sequential parallel-analysis monitoring (PRM) method. For PRM, MS2 scan parameters were set to select the m/z ratio of the natural peptides of Cas9, TadA deaminase and reverse transcriptase, and their corresponding SIL peptides with defined elution time windows. MS1 scans were acquired at the m/z range of 300–1,000, mass resolution of 70,000, automatic gain control (AGC) target of 1 × 10⁶ and maximum ion injection time of 50 ms. The PRM scans were acquired at a resolution of 17,500, AGC target value of 1 × 10⁵, maximum ion injection time of 50 ms and isolation window of 2.0 m/z.

Mice were initially anaesthetized with 2% isoflurane in a mixture of N₂O/O₂ (70%/30%) and then placed into a stereotaxic apparatus. A small, custom-made plastic chamber was glued (Vetbond) to the exposed skull. After 1 day of recovery, re-anaesthetized animals were placed in a custom-made hammock, maintained under isoflurane anaesthesia (1–2% in O₂), and multiple single tungsten electrodes were inserted into a small craniotomy above the V1 and SC. Once the electrodes were inserted, the chamber was filled with sterile agar and sealed with sterile bone wax. During recording sessions, animals were kept under isoflurane anaesthesia (0.5–1% in 30% O₂). EEG and EKG scans were monitored throughout the experiments and body temperature was maintained with a heating pad (Harvard Apparatus).

Data were acquired using a 32-channel Scout recording system (Ripple). The local field potential (LFP) from multiple locations was bandpass filtered from 0.1 Hz to 250 Hz and stored with spiking data on a computer with a 1-kHz sampling rate. The LFP signal was cut according to stimulus time stamps and averaged across trials for each recording location to calculate visually evoked potentials (VEP)^41,100–102. The evoked potential across all layers was recorded and the most robust response was used for comparisons between groups at the same SC or V1 layer.

The spike signal was bandpass filtered from 500 Hz to 7 kHz and stored in a computer hard drive at a 30 kHz sampling frequency. Spikes were sorted online in Trellis software (Ripple) while performing visual stimulation. Visual stimuli were generated in Matlab (Mathworks) using the Psychophysics Toolbox^103–105 and displayed on a gamma-corrected LCD monitor (55 inches, 60 Hz; 1,920 × 1,080 pixels; 52 cd m⁻² mean luminance). Stimulus onset times were corrected for LCD-monitor delay using a photodiode and microcontroller (in-house design).

Vision was assessed using protocols published in our previous work^{41,101,106,107}. Cells were first tested with 100 repetitions of a 500-ms bright flash of light (105 cd m⁻²) for the presence of the visually evoked responses. When cells showed signs of robust visually driven activity, we used further drifting grating stimuli to assess the properties of the spatiotemporal receptive fields. Briefly, each cell was evaluated for orientation selectivity, optimal stimulus size, optimal spatial frequency, optimal temporal frequency and contrast sensitivity. Recorded tuning curves were further normalized between 0 and 1 for visual purposes and plotted together for comparison. Data are presented as mean ± s.e.m. The level of statistical significance was set at P < 0.05 for two-tailed Mann–Whitney U-tests. The figures show single recording locations from the SC and V1 recordings as examples. Offline data analysis and statistics were performed in Matlab (Mathworks).

Tuning curves were calculated on the basis of the average spike rate. Optimal visual parameters were chosen as the maximum response value. Orientation tuning was measured in degrees at the half-width at half-height (HWHH; 1.18 × σ) on the basis of fits to Gaussian distributions using equation (): 1

where O_s is the stimulus orientation, R_Os is the response to different orientations, O_p is the preferred orientation, R_p and R_n are the responses at the preferred and non-preferred direction, σ is the tuning width and ‘baseline’ is the offset of the Gaussian distribution. Gaussian fits were estimated without subtracting spontaneous activity¹⁰⁷.

The optimal spatial and temporal frequency was extracted from the data fitted to Gaussian distributions using equation (2)¹⁰⁷:

where R_SF/TF is the estimated response and R_pref indicates response at a preferred spatial or temporal frequency. SF/TF indicates spatial or temporal frequency, σ is the standard deviation of the Gaussian and the baseline is the Gaussian offset.

Unless otherwise stated, data are presented as mean ± s.d. and statistical analyses were performed using GraphPad Prism 10.0, with *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 and ^NSP ≥ 0.05.

Further information on research design is available in thelinked to this article. Nature Portfolio Reporting Summary

Nature Biomedical Engineering thanks Michael Mitchell and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

is available for this paper at 10.1038/s41551-024-01296-2.

The online version contains supplementary material available at 10.1038/s41551-024-01296-2.

High-throughput sequencing data are available from the National Center for Biotechnology Information Sequence Read Archive database, under accession PRJNA1124167↗. Source data for the figures are provided with this paper. The raw and analysed datasets generated during the study are available for research purposes from the corresponding authors on reasonable request.