Prime Editing Exhibits Limited Genome-Wide Off-Target Effects in Cellular and Embryonic Gene Editing

Highlights

What are the main findings?

• PE5max induces far fewer large deletions and chromosomal translocations than PE3max, as measured by PEM-seq.
• In the GOTI assay using mouse embryos, PE5max was not observed to generate detectable genome-wide, sgRNA-dependent off-target SNVs within the detection limits of the assay.

What are the implications of the main findings?

• The study supports an improved specificity profile of PE5max relative to other PEmax systems in the assays used here, motivating further validation in additional cell types and in vivo settings.
• It provides a framework combining PEM-seq and GOTI for comprehensive, genome-wide safety evaluation of advanced genome editors.

Abstract

Prime editing (PE) is a precise genome-editing technology that avoids double-strand breaks, holding great promise for clinical and agricultural applications. However, its genome-wide off-target effects are not fully understood, raising safety concerns. Here, we systematically compared the safety profiles of four prime editor variants (PE2max, PE3max, PE4max, and PE5max) using PEM-seq and RNA-seq. We further applied an ultra-sensitive method, Genome-wide Off-target analysis by Two-cell embryo Injection (GOTI), to assess PE5max. Our results show that PE5max did not produce detectable sgRNA-dependent off-target single-nucleotide variants (SNVs) in the GOTI assay and induced only limited large deletions and chromosomal translocations. Collectively, this side-by-side benchmarking under matched conditions demonstrates that PE5max achieves an improved specificity profile, with no detectable increase in genome-wide off-target SNVs, advancing its potential for safer therapeutic use.

1. Introduction

PE is an advanced gene-editing technology capable of introducing precise base substitutions, small insertions or deletions, and even large-fragment modifications without generating double-strand breaks [1]. These features give PE broad therapeutic potential, with applicability to roughly 90% of known genetic diseases, and demonstrated efficacy in models of Alternating Hemiplegia of Childhood (AHC) and Phenylketonuria (PKU) [2,3]. PE has also been applied to agricultural and livestock engineering, enabling the creation of porcine disease models and genetic improvements in sheep [4,5]. Building on the original system, several derivatives have expanded editing range and precision, including TwinPE and PASTE for large-fragment deletion or insertion [6,7], and the Amplification Editing (AE) system [8] for megabase-scale genomic replication. These advances highlight the versatility and promise of PE in both biomedical and agricultural fields.

The core PE system (PE2) consists of two components [1]. The first is a fusion protein in which the H840A Cas9 nickase is linked to an engineered Moloney murine leukemia virus reverse transcriptase. The second is the prime editing guide RNA (pegRNA), which contains a genomic-targeting spacer, a Cas9-binding scaffold, a reverse transcription template (RTT) encoding the intended edit, and a primer binding site (PBS). To enhance editing efficiency, PE3 introduces an additional nicking single-guide RNA (nsgRNA) to target the non-edited strand. The PE4 and PE5 systems further improve editing fidelity by inhibiting the DNA mismatch repair (MMR) pathway, for example, by expressing the dominant-negative MLH1dn mutant [9]. More recently, the PEmax system incorporated multiple optimizations, including reverse transcriptase codon optimization, an engineered 34-amino acid linker with a bipartite SV40 nuclear localization signal (NLS), a C-terminal c-Myc NLS, and R221K and N394K mutations in Streptococcus pyogenes Cas9 (SpCas9). In parallel, pegRNA stability was increased by adding structured RNA motifs such as evo-preQ1 to generate engineered pegRNAs (epegRNAs), improving overall editing efficiency [10].

Despite these improvements, the potential off-target activity and broader genomic consequences of PE systems, especially those incorporating MMR inhibition, remain insufficiently characterized. Only a limited number of studies have assessed off-target and genotoxic outcomes of PE3, and the safety profiles of PE4 and PE5 remain largely unexamined [11,12,13,14,15]. Key questions concerning genome-wide specificity, structural variations, and transcriptomic perturbation, therefore, remain unresolved.

To address these gaps, we systematically evaluated PE2max, PE3max, PE4max, and PE5max using a comprehensive set of assays. Off-target sites were predicted and validated using Cas-OFFinder [16], and epegRNA sequence tolerance under mismatched conditions was assessed. Structural variations, including large deletions, insertions, and chromosomal translocations, were monitored using PEM-seq [17,18]. Editing specificity and genome-wide off-target effects were further examined in mouse embryos using the GOTI assay [19,20]. Transcriptome-wide changes were evaluated by RNA-seq [12,21]. Across these analyses, PE5max consistently exhibited improved performance, achieving the highest on-target editing efficiencies in HEK293T cells while maintaining a relatively low off-target profile. PEM-seq revealed that PE3max and PE5max could generate detectable structural variations, but GOTI experiments demonstrated that PE5max did not induce significant genome-wide mutations in embryos. Together, these findings provide a side-by-side benchmark of PE2max, PE3max, PE4max and PE5max under matched conditions, thereby identifying that PE5max achieves improved performance.

2. Materials and Methods

2.1. Plasmid Construction

To allow independent expression of PEmax, epegRNA, and nsgRNA, we constructed separate expression plasmids for each component. The backbone plasmid pCMV-PEmax-CMV-mCherry (used as an empty vector control, Addgene ID 174820) contained the red fluorescent protein mCherry as a transfection reporter. Versions of the PE (such as PE2max) were inserted into this backbone to generate CMV-driven expression plasmids (for example, pCMV-PE2max-CMV-mCherry). Three additional engineered PEmax variants were constructed using the same strategy.

For epegRNA expression, we used the pCMV-EGFP-polyA-U6-epegRNA backbone and cloned target-specific sgRNA sequences downstream of the U6 promoter. For nsgRNA expression, we used the pCMV-TagBFP-polyA-U6-epegRNA backbone and inserted the nsgRNA sequences into the U6 expression cassette using the same method. All recombinant plasmids were sequence-verified by Sanger sequencing. epegRNA and nsgRNA sequences are listed in Table S1.

2.2. Cell Culture

In this study, we only used one cell line, HEK293T, which was from the Cell Bank of Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences. HEK293T cells were maintained in DMEM (Yeason, Shanghai, China, 41401ES) supplemented with 10% fetal bovine serum (FBS, BI, Kibbutz Beit-Haemek, Israel, 04-001-1ACS) and 1% Penicillin-Streptomycin (Beyotime, Shanghai, China, C0224) at 37 °C with 5% CO₂. When cultures reached 80–90% confluence, cells were digested with 0.05% Trypsin–EDTA solution (Yeason, 40127ES80) and passaged. Cells were seeded into 96-well plates or 100-mm dishes as required.

2.3. Cell Transfection

Transfections were performed in 96-well plates using polyethylenimine (PEI; Polysciences, Warrington, PA, USA) when cells reached 40-50% confluence. Plasmid DNA and PEI (DNA:PEI mass ratio 1:2) were diluted separately in Opti-MEM (Gibco, Thermo Fisher Scientific, Waltham, MA, USA, 31985-070), incubated for 5 min, mixed, and incubated again for 10 min to form complexes. Complexes were added dropwise to the cells and incubated for 48 h. Cells co-expressing mCherry and EGFP were subsequently isolated by fluorescence-activated cell sorting (FACS; BD FACSAria III, Franklin Lakes, NJ, USA).

2.4. PCR Amplification and Deep Sequencing

Approximately 1 × 10³ sorted cells were lysed in 15 μL buffer (50 mM KCl, 1.5 mM MgCl₂, 10 mM Tris-HCl, pH 8.5, 0.5% NP-40, 0.5% Triton X-100, 10 μg/mL Proteinase K). Target regions were amplified by nested PCR using barcoded primers (Yeasen Hieff^® Ultra-Rapid HotStart Master Mix) (Table S2), and PCR products were purified using a gel extraction kit (Tiangen, Beijing, China). Sequencing libraries were constructed and subjected to high-throughput sequencing on the MGISEQ-T7 platform (MGI, Shenzhen, China).

2.5. PEM-Seq for Structural Variation Detection

Seventy-two h after transfection, approximately 1.5 million cells per sample were collected. Genomic DNA was extracted by ethanol precipitation and fragmented to 300–700 bp using a Covaris S220 (Woburn, MA, USA). Primer extension, streptavidin bead enrichment (Dynabeads™ MyOne™ Streptavidin C1, Thermo Fisher Scientific, 65001), on-bead ligation (T4 DNA Ligase, Thermo Fisher Scientific, EL0011), and nested PCR were performed as described (primer sequences in Table S8). Libraries were sequenced on a NovaSeq X Plus platform (PE150 mode).

2.6. Animal Care and Management

Four-week-old female C57BL/6 mice and adult C57BL/6 male mice or Ai9 homozygous males (B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J; JAX Stock No. 007909) were used for embryo collection. ICR females served as embryo-transfer recipients.

2.7. mRNA and sgRNA Preparation

Plasmids encoding PEmax, MLH1dn, or iCre were linearized as templates for IVT using the MESSAGE mMACHINE T7 ULTRA Kit (Thermo Fisher Scientific).

Templates for epegRNA and sgRNA were generated by PCR amplification of px330 (Addgene 42230). IVT was performed using a MEGAshortscript T7 Kit (Thermo Fisher Scientific), and RNA products were purified with a MEGAclear Kit (Thermo Fisher Scientific) and eluted in RNase-free water.

2.8. Embryo Injection, Culture, and Transplantation

Superovulated C57BL/6 females were mated with C57BL/6 or Ai9 males. Fertilized one-cell embryos were collected 24 h after hCG injection.

For embryo editing, we injected PE2max (PEmax + epegRNA), PE3max (PEmax + epegRNA + nsgRNA), PE4max (PEmax + MLH1dn + epegRNA), or PE5max (PEmax + MLH1dn + epegRNA + nsgRNA). PEmax and MLH1dn were injected at 70 ng/μL, and epegRNA/nsgRNA at 30 ng/μL.

For the GOTI experiment, embryos received PE5max components plus iCre (2 ng/μL). Injections were performed in M2 medium containing 5 μg/mL cytochalasin B using a FemtoJet microinjection system (Eppendorf, Hamburg, Germany) in constant flow mode. Embryos were cultured in KSOM with amino acids at 37 °C and 5% CO₂ for 2 h before transfer into the oviduct of ICR females at 0.5 days post coitum (dpc).

2.9. Embryonic Editing Efficiency Detection

At embryonic day 4.5 (E4.5), individual blastocysts were lysed in 4 μL buffer (0.1% Triton X-100, 0.1% TWEEN20, 4 μg/mL Proteinase K) using the following program: 55 °C for 30 min, 95 °C for 10 min, then held at 4 °C. Target genomic regions were amplified by nested PCR. The first round of PCR was performed using Extaq enzyme (Takara, Kusatsu, Shiga, Japan) with an initial denaturation at 95 °C for 3 min, followed by 30 cycles at 95 °C for 30 s, 55 °C for 30 s, 72 °C for 1 min, and a final extension at 72 °C for 5 min. The second round of PCR employed inner nested primers with the same program. PCR products were subsequently purified and submitted for NGS.

2.10. GOTI Assay

GOTI was performed following established protocols. Two-cell embryos from Ai9 x C57BL/6 crosses were injected in one blastomere with IVT-generated PE5max components and iCre. After 2–4 h, embryos were transplanted into pseudo-pregnant ICR recipients.

At E14.5, embryos were dissected, tissues digested with 5 mL of 0.05% Trypsin-EDTA solution at 37 °C for 30 min, and single-cell suspensions were prepared. Digestion was arrested with an equal volume of DMEM containing 10% FBS, and the suspension was triturated to generate single cells. After centrifugation at 1000× g for 6 min, cells were resuspended in DMEM containing 10% FBS and passed through a 40-μm strainer. tdTomato⁺ and tdTomato⁻ cells were sorted by FACS, and genomic DNA was extracted for whole-genome sequencing.

2.11. RNA Extraction and Transcriptome Sequencing

After 72 h post-transfection, 200,000 sorted HEK293T cells were collected for RNA extraction. Total RNA was isolated using Trizol reagent (Ambion, Austin, TX, USA) following the manufacturer’s protocol. RNA quality was assessed, and mRNA-seq libraries were constructed and sequenced using the MGISEQ-T7 platform (MGI).

2.12. Bioinformatic Analysis

Raw sequencing data were demultiplexed using fastq-multx v1.4.2. Editing outcomes were quantified using CRISPResso2 software v2.2.7. FastQC v0.11.8 was used for quality assessment, and adapters were trimmed using Trimmomatic v0.39.

Differential gene expression analysis was performed with DESeq2. Genes were considered significantly differentially expressed when |log₂(FoldChange)| > 1 and p < 0.05. GO term enrichment of DEGs was performed using clusterProfiler v4.14.6.

To identify high-confidence off-target variants specifically present in edited (tdTomato⁺) cells, we implemented an established, stringent analysis pipeline as previously described for the GOTI assay [20]. In brief, somatic SNVs and indels were called by comparing whole-genome sequencing data from tdTomato⁺ cells against the paired, internal control tdTomato⁻ cells from the same embryo. To maximize specificity and minimize false positives, we employed a consensus-calling strategy in which a variant was only considered a high-confidence off-target event if it was independently identified by all three distinct callers (Mutect2, Strelka2, and LoFreq).

All figures were generated in R (v4.3.1). Data are presented as mean ± s.e.m., and statistical significance was determined by two-sided unpaired Student’s t-test, with p < 0.05 considered statistically significant.

3. Results

3.1. PEmax Systems Demonstrate High Specificity Without sgRNA-Dependent Off-Target Effects

To systematically assess editing efficiencies at endogenous loci, we compared PE2max, PE3max, PE4max, and PE5max across three sites (Site 552–554) in HEK293T cells (Table S1). A multi-plasmid co-transfection strategy was used to deliver the editor (PEmax or PEmax–MLH1dn), epegRNA, and nsgRNA, each tagged with distinct fluorescent markers (mCherry, EGFP, and TagBFP) that enabled transfected-cell identification and fluorescence-activated cell sorting (FACS) isolation (Figure 1a and Figure S1a,b). PE2max consisted of PEmax with an epegRNA, while PE4max consisted of PEmax–MLH1dn with an epegRNA, producing dual red and green fluorescence. PE3max (PEmax with epegRNA and nsgRNA) and PE5max (PEmax–MLH1dn with epegRNA and nsgRNA) generated triple red, green, and blue fluorescence (Figure S1a,b, Table S1). Seventy-two h after transfection, fluorescent-positive cells were sorted, genomic DNA was extracted, and target loci were amplified for deep sequencing.

Across all three sites, PE5max consistently produced the highest editing efficiency, averaging 60.63%, with a low indel frequency of 1.15%. PE3max followed with 46.90% efficiency and 1.83% indels, PE4max showed 19.10% efficiency with 0.20% indels, and PE2max displayed 10.24% efficiency with 0.34% indels (Figure 1b and Figure S2a, Tables S2 and S3). To further evaluate specificity, Cas-OFFinder predicted 12 high-similarity potential off-target sites for Site 552, each containing at least two mismatches (Table S4). Deep sequencing demonstrated that none of the PE systems generated detectable editing at these sites, with all efficiencies below 1 percent, no significant differences from wild type controls, and no measurable indels (Figure 1c and Figure S2b–d, Table S5).

To investigate mismatch tolerance between epegRNA and genomic DNA, we designed mismatch epegRNAs (mepegRNAs) containing one to three base substitutions within the spacer region of Site 552 and systematically examined the response of each editor (Table S6).

For single-base mismatches, PE3max and PE5max maintained relatively high editing efficiency when mismatches occurred at positions 4, 5, 8, and 9. PE3max efficiencies at these positions were 65.66%, 58.36%, 65.98%, and 66.57%, while PE5max efficiencies were 66.35%, 66.78%, 62.27%, and 65.10%. In contrast, mismatches at positions 2 and 7 substantially reduced editing: PE3max dropped to 45.93% and 43.01%, whereas PE5max fell to 26.95% and 27.13%. PE2max and PE4max demonstrated reduced activity across nearly all single-base mismatch positions, though editing remained more stable at positions 4, 5, 8, and 9 (Figure S3a,b, Table S7).

In dual-base consecutive mismatch experiments, editing efficiencies declined across all systems. However, PE3max and PE5max still retained measurable activity at the 9–10 mismatch (12.65% and 16.45%, respectively). PE2max and PE4max showed modest editing peaks at the 7-8 mismatch (3.02% and 6.87%, respectively). With three consecutive mismatches, activity of all editors dropped to background levels (Figure S3a,c, Table S7).

Mismatch patterns also influenced indel frequencies. For example, PE3max displayed slightly elevated indels at positions 6, 8, and 12 (1.63%, 1.66%, and 1.90%, respectively), relative to the 1.62% with a perfect match. PE5max showed modest increases at the 8–10 and 12 mismatch positions but remained below the 3.04% indel rate seen with the perfect match. These results indicate that specific mismatch configurations reduce editing efficiency and can alter outcome precision (Figure S3d, Table S7).

3.2. PE5max Mitigates Structural Variations Induced by Prime Editing

We applied primer extension-mediated sequencing (PEM-seq) to profile genome-wide structural variations generated by PE systems (Figure S4a, Table S8). PEM-seq uses biotin-labeled primers upstream of double-strand breaks (DSBs) to capture and amplify ligated repair products, enabling sensitive detection of deletions greater than 100 bp, large insertions greater than 20 bp, and chromosomal translocations.

Using this assay, we found that PE3max and PE5max produced significantly higher frequencies of large deletions compared with the wild type control, at 2.29% (p < 0.05) and 0.78% (p < 0.05), respectively. PE2max and PE4max generated only 0.02% deletions, which did not differ from control (Figure S4b). For large insertions, PE3max and PE5max reached 1.53% and 0.72% (p < 0.05), respectively. PE2max produced 0.35% insertions (p < 0.05), while PE4max showed 0.15%, which was not significant (Figure S4c, Table S9).

Chromosomal translocations were detected in all editors. Frequencies were 0.24% for PE2max, 0.32% for PE3max, 0.10% for PE4max, and 0.17% for PE5max, all significantly elevated relative to wild type (p < 0.05) (Figure 1d). Thus, even at low frequencies, PEmax systems like PE3max and PE5max remain capable of inducing detectable structural variations. Genome-wide mapping showed that translocations were broadly distributed, with PE2max and PE3max producing the most hotspots (29 and 31, respectively), whereas PE4max and PE5max exhibited fewer hotspots (10 and 9, respectively) (Figure 1e, Table S9). Notably, PE3max displayed a detectable but low-frequency preference for G-quadruplex-associated (G4) translocation events (0.01%) (Figure S4d, Table S9).

3.3. Prime Editors Avoid RNA Off-Target Editing but Activate Innate Immune Responses

To determine whether PEs introduce transcriptome-wide off-target activity, we transfected HEK293T cells with each PEmax system and enriched high-expression populations by FACS (Figure S5a). RNA-seq-based analysis of SNVs, after removing variants present in wild-type controls, showed no significant increase in RNA SNV burden in any PE-treated cells (Figure 2a and Figure S5b, Table S10). All groups displayed similar mutation spectra dominated by A-to-G and T-to-C substitutions, which matched the wild type pattern and indicated that PE systems did not induce RNA off-target editing (Figure 2b and Figure S5c, Table S11). Motifs and chromosomal distributions of A-to-I substitutions were likewise comparable across systems (Figure 2c,d and Figure S6a,b, Table S11).

We next examined transcriptional perturbations. Using EGFP-transfected cells as a negative control, gene set enrichment analysis (GSEA) revealed that all PE systems significantly activated innate immune and inflammation-associated signaling pathways (Figure S7a). The TNFα signaling_via_NFκB gene set was strongly upregulated across editors (Normalized Enrichment Score, NES > 3). In contrast, interferon-α- and interferon-γ-response pathways were broadly suppressed. Among the systems, PE4max and PE5max produced the strongest NF-κB activation, whereas PE2max and PE3max showed more pronounced suppression of interferon-associated pathways (Figure S7b).

Differentially expressed gene (DEG) analysis and Gene Ontology (GO) enrichment further demonstrated that upregulated genes clustered in immune and antiviral processes, including response to virus (ZC3H12A, BCL3), response to type I interferon (IRF7, PTPN1), response to lipopolysaccharide (FOS, CXCL2), and regulation of innate immunity (FOSL1, NFKBIA) (Figure S6c,d). These results align with the GSEA findings and indicate that PE system delivery activates a transcriptional program resembling an antiviral innate immune response (Figure S7a–d).

3.4. PE5max Presents Undetectable Off-Target Effects in the GOTI Assay

To assess the editing efficacy and safety of PE systems in vivo, we evaluated their performance in mouse embryos. Using an optimized in vitro transcribed (IVT) mRNA backbone, constructs were engineered with a 5′ cap to mimic endogenous transcripts and incorporated a eukaryotic initiation factor 4G (eIF4G) aptamer into the 5′-untranslated region. The 3′ end contained a WPRE and a 120 nt poly(A) tail to enhance mRNA stability (Figure S8a). EpegRNAs targeting Site 1 and Site 3 were also designed and synthesized (Table S1). Fertilized C57BL/6×C57BL/6 eggs were assigned to groups including uninjected controls, iCre-mRNA only controls, and experimental groups injected with combinations of PEmax-mRNA, MLH1dn-mRNA, site-specific epegRNA, and nsgRNA. All injections were performed at the one-cell stage, and blastocyst formation was evaluated 4.5 days post-fertilization.

Amplification sequencing of embryos that reached the blastocyst stage revealed locus-specific editing. At Site 1, average editing efficiencies were 6.98% (PE2max), 3.80% (PE3max), 0.00% (PE4max), and 8.75% (PE5max), with PE5max exhibiting a numerically higher value, which was not statistically significant compared to the control (p = 0.074) (Figure S8b, Table S12). At Site 3, editing efficiencies were 3.80% (PE2max), 2.66% (PE3max), 1.79% (PE4max), and 10.53% (PE5max), with PE5max demonstrating the strongest effect and reaching high statistical significance (p < 0.05) (Figure S8c, Table S12). Indel analysis showed that all PE systems produced a small number of indels at Site 1 (p < 0.05), while no significant indels were detected at Site 3 (Figure S8d,e, Table S12).

Because PE5max showed the highest and most consistent editing efficiencies across sites in embryos, we used the GOTI method to determine whether it generated genome-wide off-target effects [19,20,22]. The GOTI assay was performed in two-cell embryos derived from C57BL/6 × Ai9 (CAG-LoxP-Stop-LoxP-tdTomato) crosses. One blastomere of each embryo was microinjected with iCre-mRNA, PEmax mRNA, MLH1dn mRNA, and the site-specific epegRNA and nsgRNA, while the second blastomere served as an internal control. Cre-mediated recombination removed the Stop cassette, enabling tdTomato expression. Thus, tdTomato-positive cells reliably marked the lineage derived from the injected blastomere and confirmed co-delivery of PE components and Cre. Injected embryos were transferred into surrogate mothers, with a subset retained for developmental monitoring. At E14.5, tdTomato⁺ and tdTomato⁻ cells were isolated by FACS, genomic DNA was extracted, and whole-genome sequencing was performed (Figure 3a and Figure S8f).

Despite a limited sample size inherent to embryo microinjection, the paired-control GOTI design (comparing tdTomato⁺ and tdTomato⁻ cells within the same embryo) minimized confounding genetic variation. Combined with stringent bioinformatic criteria (consensus across three variant callers), this approach ensured reliable detection. Whole-genome sequencing confirmed minimal on-target editing in control (tdTomato⁻) cells (0–2.7%) versus efficient editing (7.59–100%) in targeted (tdTomato⁺) cells (Figure 3b, Table S13). Genome-wide SNV analysis revealed only 5–18 background-level SNVs per embryo—comparable to CRE-only controls, unique to each embryo, non-recurrent across samples, and absent from predicted off-target sites (Figure 3d,e, Table S14). Together, these data indicate that PE5max does not produce detectable genome-wide off-target SNVs under the sensitivity of the GOTI assay.

4. Discussion

This study provides a benchmarking of the editing efficiency and genome-wide safety profiles of four prime editing systems. In aggregate, our results demonstrate that PE5max exhibits a comparatively improved safety profile, with minimal off-target activity and lower structural variation compared to the other systems, including reduced structural variation compared with PE3max. In mouse embryos, the GOTI assay did not detect an increased genome-wide SNV burden following PE5max editing, supporting a favorable specificity profile within the sensitivity and scope of this assay. Because GOTI pairs edited cells with genetically matched internal controls, these data strengthen the evaluation of prime editors that incorporate transient MMR inhibition [23].

Our mismatch-tolerance analysis showed that PE3max and PE5max maintain substantial editing activity when mismatches occur in central spacer positions of epegRNA. This suggests some flexibility in epegRNA –DNA pairing; however, the strict requirement for nsgRNA activity in these systems likely prevents such mismatches from producing meaningful sgRNA-dependent off-target editing [24]. These results indicate that sequence tolerance does not translate into appreciable off-target genomic activity.

It is also important to critically evaluate the structural variations induced by PE5max. While PEM-seq data showed that PE5max generated significantly fewer large deletions (0.78%) and insertions (0.72%) than PE3max (2.29% and 1.53%, respectively), these events are not negligible. Structural rearrangements are generally considered to pose greater clinical risks than SNVs due to their potential to disrupt gene structure, cause chromosomal instability, or lead to oncogenic fusion events. Therefore, although PE5max represents a substantial improvement in specificity, the low but detectable level of structural variation underscores the need for further optimization and careful risk assessment in therapeutic contexts.

At the transcriptome level, we confirmed that none of the PE systems caused RNA off-target editing, a finding that is consistent with previous studies [12]. Instead, all editors activated innate immune and inflammatory signaling, most prominently through strong NF-κB pathway induction. This transcriptional response is more consistent with cellular stress induced by the delivery of exogenous editing components (e.g., Cas9 mRNA, reverse transcriptase, and pegRNA) than with a direct consequence of on-target genomic modifications. While such transient immune activation may reflect a general reaction to editor expression, its potential to provoke adverse inflammatory responses in vivo underscores the importance of monitoring immune reactions and refining delivery strategies in future therapeutic applications. Although the long-term impact of this transcriptomic stress response is not yet clear, it represents an important consideration for the future therapeutic development of prime editors.

Several limitations should be noted. First, our assessments of editing outcomes were conducted primarily in HEK293T cells, a transformed cell line with inherently unstable DNA repair mechanisms and a well-documented deficiency in the MMR pathway [25]. It remains uncertain how broadly these findings extend to primary, stem, or disease-relevant cell types. Furthermore, although transient MMR inhibition was employed to enhance editing efficiency, the potential long-term risks associated with MMR suppression, including microsatellite instability, elevated mutation rates, and increased cancer risk, warrant careful consideration in any future therapeutic application [26]. Although PE5max achieves highly efficient and precise editing through MMR inhibition, its potential long-term risks (e.g., microsatellite instability) need to be further validated in primary cells and long-term animal models to guide clinical translational applications. The chromosomal translocations detected across all systems occurred at low frequency, but their functional impact is unknown. In addition, while GOTI provides a sensitive readout for genome-wide SNV burden, it is not optimized to comprehensively quantify large structural variants or epigenetic alterations. Finally, this work primarily assessed short-term editing outcomes; long-term analyses will be essential to determine the durability, stability, and safety of PE5max-mediated edits.

5. Conclusions

Our results support PE5max as a highly efficient prime editor with an improved specificity profile relative to other PEmax systems in the assays used here, including reduced structural variation in cells and no detectable increase in genome-wide SNVs in the GOTI setting. However, low-frequency structural variations and consistent transcriptomic stress responses were observed, and the long-term consequences of transient MMR inhibition remain to be fully evaluated in non-transformed and disease-relevant models. Continued optimization of guide design, editor components, and delivery strategies will be important to further improve precision and to assess durability and safety for future applications.