The Pde6b^nmf137 mice, carrying a missense mutation c.1976T>C, p.L659P (hereafter referred to as the Pde6b-L659P mutation) in the Pde6b gene, exhibit severe photoreceptor degeneration and serve as a valuable model for studying and treating retinal degeneration [18] (Fig. 1). We aimed to perform a parallel comparison of the BE, PE, and CE systems at this disease-associated Pde6b-L659P mutation site. Before applying these systems in mouse in vivo studies, a reliable in vitro model is needed to optimize their parameters and achieve the most efficient editing system. To overcome this challenge, we utilized the ABE system (ABE8e-nSpRY) [19] to introduce the Pde6b-L659P mutation into the mouse Neuro-2a (N2a) cell line (Fig. S1A). After transfecting the ABE system, bulk sequencing results revealed that up to 45% of the desired target mutation was successfully introduced (Fig. S1B). However, this was accompanied by approximately 25% of 2 unintended bystander mutations (Fig. S1B). To address this, we screened edited single-cell colonies and ultimately obtained a stable N2a cell line carrying only the homozygous Pde6b-L659P mutation without bystander mutations (Fig. S1C). This mutation was completely identical to that in Pde6b mutant mice (Fig. S1C). Therefore, this endogenously mutated N2a cell model can be used for in vitro screening and comparison of different gene-editing systems for their repair effects.

Since this Pde6b-L659P mutation is a T-to-C substitution, we hypothesized that the CBE system could be used for C-to-T conversion rescue. The CBE system consists of 2 components: the CBE vector and a single guide RNA (sgRNA) (Fig. 2A). Due to the absence of a classic NGG PAM sequence near the target site, we employed a near-PAMless nSpRY Cas9-mediated CBE, which theoretically recognizes all NNN PAMs (NRN > NYN, R = A/G, Y = C/T) [20]. For CBE vectors, we first tested the nSpRY-CBEs with 3 conventional cytosine deaminases (CDs, including rA1, AID, and CDA1) (Fig. S2A). For sgRNAs, we designed 7 sgRNAs with NNN PAMs, ensuring that the target C was positioned within the primary C2 to C8 editing window (Fig. 2B). To evaluate the BE performance, we introduced CBE vectors carrying sgRNAs 1 through 7 into the N2a cell model and measured the editing outcomes using Sanger sequencing. As a result, although all 3 CBEs achieved effective C-to-T editing at the target site, they also induced a high proportion of bystander edits (Fig. 2C to E). As expected, the presence of multiple adjacent Cs around the target site posed a challenge for precise editing. To address this issue, we tested eA3G-nSpRY-CBE, a system developed by us and other groups (Fig. S2A) [21–23]. This CBE utilizes an engineered A3G (eA3G) deaminase with preferential editing activity in CC motifs (where the preferentially deaminated C is underlined), which could potentially achieve better editing performance at this Pde6b-L659P site within CC contexts. Notably, eA3G-nSpRY-CBE in combination with sg4 achieved an optimal 19.3% targeted editing, with 16.0% bystander3 editing, and no noticeable bystander1 or bystander2 editing (Fig. 2E and Fig. S3A). Theoretically, due to the redundancy of the genetic code, bystander3 editing does not alter the intended amino acid correction (remaining p.P659L) and is therefore deemed acceptable (Fig. S3B). In addition to nSpRY-CBEs, we also tested the performance of these 4 deaminase-mediated nSpNG-CBEs, which are widely used systems capable of recognizing NGN PAMs (Fig. S2B) [24]. However, the results showed that the performance of nSpNG-CBEs did not outperform nSpRY-CBEs (Fig. S4A to D). Additionally, we tested a YE1-rA1-CBE with a narrowed editing window [25], but its efficiency was lower than that of rA1-CBE (Fig. S4E and F). Overall, after screening 10 CBE vectors and 7 sgRNAs in vitro, the combination of eA3G-nSpRY-CBE and sg4 achieved the most efficient targeted editing with the lowest unwanted bystander editing, making it the optimal CBE system for repairing the Pde6b-L659P mutation.

Subsequently, we investigated the feasibility of using novel PE systems to correct the Pde6b-L659P mutation. Theoretically, PE offers greater precision than BE, as it directly installs the desired edit using a programmable template, thereby minimizing unintended bystander mutations [13,26,27]. The commonly used PE system typically consists of 3 components: the PE vector, engineered prime editing guide RNA (epegRNA) [14], and nicking guide RNA (ngRNA) (Fig. 3A). Following established PE experimental protocols [28], we systematically designed and tested 5 PE vectors (PE2 [13], PEmax [29], PEmax-hMLH1dn [29], PE6b [30], and PE7 [31]), along with 5 epegRNAs (e1 to e5) and 5 ngRNAs (n1 to n5) featuring diverse nonedited strand nicks (Fig. S5). These epegRNA and ngRNA designs were guided by recommendations from PE design web tools to optimize editing efficiency [32,33]. Evaluation in the N2a cell model revealed that the latest PE7 vector achieved 17% targeted editing with no noticeable bystander edits, outperforming all other PE vectors (Fig. 3B and Fig. S5). Among the 5 tested epegRNAs, e1 exhibited the highest editing efficiency, markedly surpassing the other 4 (Fig. 3C). Similarly, for the 5 ngRNAs, n1, n3, and n4 demonstrated strong efficiency, with n3 achieving the highest at 20.6% (Fig. 3D). In summary, in vitro screening results demonstrated that the combination of PE7 + e1 + n3 achieved the highest targeting efficiency with minimal bystander edits, showing greater precision than CBE for efficient Pde6b-L659P mutation correction.

CE is a newly developed next-generation precise genome editing technology, following BE and PE [8]. In principle, CE combines DNA-dependent polymerase (DDP) with RNA-programmable nCas9 to enable precise genome editing by covalently linking user-defined edits to targeted genomic sites using click DNA (clkDNA) templates. A typical CE system consists of 4 components: CE vector, target CE-gRNA, clkDNA, and ngRNA (Fig. 4A). Here, CE-gRNA and clkDNA are roughly analogous to epegRNA in PE, while ngRNA mirrors the ngRNA in PE, functioning to nick the nontarget strand and enhance efficiency. Given the design similarities between CE and PE, we directly applied the most efficient design parameters from PE to develop the CE system. We designed and tested 1 gRNA (CE-g1), 2 ngRNAs (n1 and n3), and 4 clkDNAs (clk1 to clk4) in the N2a cell model (Fig. 4B). However, Sanger sequencing results showed that none of the tested CE systems achieved obvious targeted editing; there appeared to be a certain proportion of indels, indicating the need for more precise sequencing analysis (Fig. 4C). Therefore, we performed deep sequencing analysis on the optimal CE1 + g1 + n3 + clk1 combination. The results showed that the CE system achieved a low efficiency of 0.7% on-target editing, but was accompanied by a higher efficiency of 10.9% indels (Fig. 4D). Previous studies have also indicated that current CE systems have low editing efficiency at certain sites and induce high levels of indels [8]. Our results demonstrated that the CE system can achieve targeted repair at the Pde6b-L659P locus, but further optimization is required to enhance editing efficiency and reduce indels.

Through a series of in vitro screenings in the N2a cell model, we have preliminarily identified the CBE, PE, and CE systems that exhibited the best performance in repairing the Pde6b-L659P mutation. However, their detailed editing characteristics still require further in-depth evaluation and comparison. Therefore, we performed deep sequencing on the genomic DNA of cells edited by the optimal CBE, PE, and CE systems to conduct a detailed and parallel comparison of their editing efficiency and precision. The deep sequencing results showed that CBE primarily edited the target and bystander3, while PE was more precise, mainly editing only the target, consistent with the previous Sanger sequencing results (Fig. 5A and Fig. S6). From a more detailed parallel comparison, PE and CBE exhibited similar repair efficiency on the target (14.1% versus 11.7%), but PE significantly reduced unwanted mutations at bystander3 (1.6% versus 8.8%) (Fig. 5B). In contrast, CE demonstrated lower editing efficiency on the target (0.7%) and induced a higher rate of indels (10.9%) (Fig. 5A and B and Fig. S6). Additionally, we performed deep sequencing on the top 5 predicted off-target sites for these 3 systems [34]. No obvious off-target effects were detected in any of the systems beyond the baseline seen in the controls, confirming the high fidelity of these genome editing systems, which aligns with previous studies [8,21,35,36] (Fig. 5C). Although, as mentioned earlier, the mutation at bystander3 is theoretically unlikely to affect the correct repair of the amino acid, its potential impact on the DNA and mRNA context remains difficult to assess. Therefore, given the similar on-target editing efficiency, we selected the more precise PE system for the next step of in vivo repair evaluation.

Building on the efficient and precise PE correction observed in vitro, we advanced to in vivo evaluation by delivering PE plasmids via electroporation. This well-established technique enables efficient DNA plasmid delivery into the neonatal mouse retina, facilitating the study of retinal gene function and the assessment of gene therapy efficacy [27,37,38]. To achieve this, we combined the 3 plasmids encoding the top-performing PE systems, PE7 + e1 + n3, and introduced them into the retinas of newborn Pde6b-L659P mice (Fig. 6A). The co-expression of enhanced green fluorescent protein (EGFP) served as a marker of successful electroporation. The electroporated mouse retinas were then dissected at P30, a stage when the retinas of Pde6b-L659P mice are fully degenerated [18]. Afterward, we used deep sequencing, Western blotting (WB), and confocal imaging techniques to evaluate the rescue effect (Fig. 6A). Remarkably, electroporation of PE plasmids achieved an average target editing efficiency of 12.4%, with no detectable bystander editing or indels, demonstrating the exceptional precision of the PE system in rescuing the Pde6b-L659P mutation in vivo (Fig. 6B and C). Additionally, the WB results showed that the retinas of Pde6b-L659P mice in the PE-treated group exhibited an obvious restoration of PDE6B protein expression, but it was not as high as in the wild-type (WT) mice group (~11% of WT levels) (Fig. 6D and E). In contrast, the untreated control group showed almost no PDE6B expression (Fig. 6D and E). The WB results strongly supported that PE-mediated repair of the Pde6b-L659P mutation at the DNA level successfully rescued PDE6B protein expression, demonstrating the efficacy of precise gene editing.

PDE6B protein is primarily expressed in the photoreceptor cells of the retina, specifically within the outer segment of rods and cones. It plays a crucial role in the phototransduction pathway, where it is involved in the conversion of light signals into electrical signals in the retina [3,4]. In Pde6b-L659P mutant mice, the photoreceptors gradually degenerate and, within approximately 1 month, are completely lost [18]. We performed immunofluorescence analysis on retinal sections to evaluate whether PE-mediated correction of the Pde6b-L659P mutation maintained photoreceptor integrity. Cryosections of retinas from P30 mice were immunolabeled with an anti-PDE6B antibody. The results demonstrated that in the PE-treated group, PDE6B protein expression was significantly restored and correctly localized to the outer segment of the retina (Fig. 7A and Fig. S7). Conversely, the untreated group exhibited almost no detectable PDE6B expression (Fig. 7A and Fig. S7). Next, we employed antibodies against rhodopsin, a rod-specific marker, and cone arrestin, a cone-specific marker, to assess the therapeutic impact of PE rescue on both rods and cones, the 2 main types of photoreceptors. In untreated retinas, the signals for rhodopsin and cone arrestin were low, as compared to the PE-treated retinas, both photoreceptor markers showed strong expression and were correctly localized—highlighting the remarkable preservation of both rods and cones (Fig. 7B). Furthermore, fluorescence intensity analysis confirmed that retinal photoreceptor cells were effectively preserved in the PE-treated retinas, in contrast to the pronounced degeneration observed in the control group (Fig. 7C). Additionally, at P30, untreated retinas showed only 1 to 2 layers of nuclei in the outer nuclear layer (ONL), while PE-treated retinas exhibited a considerably thicker ONL (Fig. 7A and B). Quantitative analysis of retinal sections showed that the ONL thickness in the PE group was 15.2 μm, a 4.0-fold increase compared to the untreated group (Fig. 7D). These findings collectively suggested that PE treatment effectively preserves retinal photoreceptors in Pde6b-L659P mice.

In addition, we used electroretinography (ERG) to measure the electrical activity of photoreceptors and optokinetic tracking response (OKR) to quantify visual acuity at P30. As a result, the PE-treated group showed a slight increase in scotopic ERG response compared to the untreated group (34.8 ± 5.8 μV vs. 16.0 ± 1.7 μV, at 1 cd⋅s/m²) (Fig. S8A and B). However, the OKR test indicated that PE plasmid electroporation did not lead to a notable recovery of visual function (Fig. S8C and D). We speculate that electroporation itself may cause structural damage to the eye, thereby affecting the restoration of visual function [37,39]. Therefore, using less invasive delivery methods such as adeno-associated virus (AAV) to administer PE may be a better therapeutic option in the future.

The CBE plasmids used in this study, including rA1-nSpRY-CBE (#139999), rA1-nSpNG-CBE (#125617), YE1-nSpNG-CBE (#138159), CDA1-nSpNG-CBE (#125612), AID-nSp-CBE (#174696), A3G-nSp-CBE (#163636), and ABE8e-nSpRY (#185671), were obtained from Addgene. Additional CBE plasmids were generated through site-directed mutagenesis and DNA recombination cloning, using pre-existing plasmids as templates. These modifications were performed with the Fast Site-Directed Mutagenesis Kit (Vazyme, #C214) and the ClonExpress Ultra One Step Cloning Kit (Vazyme, #C115). For PE plasmids, the PE2 (#132775), PEmax (#180020), PEmax-hMLH1dn (#174828), PE6b (#207852), and PE7 (#214812) were obtained from Addgene. The epegRNA plasmids were constructed based on the pU6-tevopreq1-GG-acceptor (#174038), following published protocols [13,28]. For CE plasmids, the Click editor-CE1 (#208942) was obtained from Addgene. The clkDNAs were designed according to the original guidelines [8] and synthesized by Integrated DNA Technologies. The sequences of all gRNAs and clkDNAs are listed in Table S1.

The N2a cell line was maintained in Dulbecco’s Modified Eagle’s Medium (Corning, #10013CV) supplemented with 10% fetal bovine serum (FBS) under standard conditions of 37°C with 5% CO₂. For transfection, cells were plated in 24-well plates and treated with PolyJet In Vitro DNA Transfection Reagent (SignaGen Laboratories, #SL100688). Specifically, 1.5 μl of PolyJet reagent was mixed with a total of 500 ng of BE, PE, or CE plasmids per well. After 72 h, cells were lysed for subsequent sequencing. Target sequences were amplified using primers listed in Table S2, and editing efficiency was assessed through Sanger sequencing, analyzed via EditR [57].

N2a cells were engineered to carry the Pde6b (c.1976T>C, p.L659P) mutation by transfecting them with ABE8e-nSpRY and ABE-gRNA-PGK-Puro plasmids. After transfection, cells underwent puromycin selection (2 μg/mL) for 5 days. Surviving cells were then diluted and plated into 96-well plates for further expansion. Individual colonies were isolated and verified by Sanger sequencing. A successfully mutated single-cell clone was expanded and used as the N2a cell model for subsequent in vitro screening.

WT C57BL/6 mice (Charles River Laboratories, #027) and Pde6b^nmf137 mutant mice (The Jackson Laboratory, #004766) were used in this study. This Pde6b^nmf137 mutant mice were generated by ethylnitrosourea mutagenesis in C57BL/6J genetic background by The Jackson Laboratory [18]. The Pde6b^nmf137 mutants showed extensive degeneration of the ONL of the retina starting at P16, and no ONL was left in the retina by P30 [18]. The Pde6b^nmf137 mice used in this study were confirmed by Sanger sequencing to carry a homozygous Pde6b-L659P mutation, and retinal dissection further validated that they exhibited a clear retinal degeneration phenotype. Experimental procedures were conducted in accordance with protocol #32223, approved by the Institutional Animal Care and Use Committee of Stanford University School of Medicine.

The in vivo retinal electroporation in Pde6b-L659P neonatal mice (P0) was performed following established protocols [27,37,58]. In brief, pups were anesthetized by ice chilling, and their eyelids were gently opened using a 30-gauge needle. A 0.3- to 0.5-μl PE DNA solution (containing 1000 ng/μl of each plasmid and 0.1% Fast Green dye for visualization) was carefully delivered into the subretinal space using a FemtoJet 4i microinjector (Eppendorf, #5252000021). Following injection, the pup’s head was positioned between a 10-mm tweezer electrode, and five 80-V pulses (50 ms each, 950-ms intervals) were administered using a NEPA21 electroporator (Bulldog Bio). Subsequent EGFP expression was used as a marker of successful electroporation.

The detailed procedure has been previously published [27]. Briefly, genomic DNA was isolated from N2a cells or GFP-positive regions of electroporated mouse retinas. Potential off-target sites for CBE-sgRNA, PE-epegRNA, and CE-gRNA were identified using Cas-OFFinder [34]. Targeted deep sequencing was performed using the Amplicon-EZ service from Azenta Life Sciences, with data analysis conducted via CRISPResso2 [59]. The primers for on-target and off-target amplification are listed in Tables S2 and S3.

The detailed procedure has been previously published [27]. Primary antibodies included anti-PDE6B (Proteintech, #22063-1-AP, 1:1000) and anti-alpha tubulin (Proteintech, #11224-1-AP, 1:5000) as a loading control. Chemiluminescence signals were captured directly using a digital imaging system (Amersha Imager 600). The images were quantitatively analyzed using Fiji ImageJ.

The detailed procedure has been previously published [27]. The following primary antibodies were used: rabbit anti-PDE6B (Proteintech, #22063-1-AP, 1:500), mouse anti-rhodopsin (Abcam, #ab5417, 1:1000), and rabbit anti-cone arrestin (Millipore, #AB15282, 1:500). For secondary antibodies, Alexa-Fluor-555-conjugated anti-rabbit IgG (Invitrogen, #A21428, 1:500) and Alexa-Fluor-647-conjugated anti-mouse IgG (Invitrogen, #A-31571, 1:500) were used. Images were captured using a Zeiss LSM880 inverted confocal microscope. All images were acquired under nonsaturating conditions using identical acquisition settings to ensure comparability. Fluorescence intensity within defined regions of interest was quantitatively analyzed using Fiji (ImageJ). Mean fluorescence values were obtained through the integrated “Analyze > Measure” function.

The detailed procedure has been previously published [27]. Briefly, the ERG was performed with an ERG stimulator (Celeris, Diagnosys LLC) according to the manufacturer’s instructions. Mice were stimulated with flashes of 0.01, 0.1 and 1 cd⋅s/m² light intensity.

The detailed procedure has been previously published [60,61]. Briefly, the OKR was assessed using the OptoMotry system (CerebralMechanics Inc.), a virtual-reality platform designed to swiftly quantify visuomotor behavior according to the manufacturer’s instructions.

All data are expressed as mean ± SEM (standard error of the mean) of at least 3 individual determinations for all experiments. Statistical differences between the values were tested using 2-tailed unpaired Student’s t tests by GraphPad Prism software 8.0.1. The exact sample size was clarified in each figure legend.