Systematic optimization of prime editing for the efficient functional correction of CFTR F508del in human airway epithelial cells

We sought to correct the CFTR F508del mutation soon after we originally reported PE in 2019 (ref. ¹). To efficiently test and optimize PE CFTR F508del correction strategies⁶¹, we first used PE to generate a clonal HEK293T cell line homozygous for this deletion in the endogenous CFTR gene (Supplementary Fig. 1 and Supplementary Tables 1 and 2). Next, we used this cell line with PE2 to screen pegRNAs with combinations of PBS and RTT lengths at two F508del CTT deletion-proximal protospacers with NGG protospacer adjacent motifs (PAMs; NGG1 and NGG2; Supplementary Fig. 2). We observed no detectable correction of CFTR F508del using NGG1 pegRNAs (Extended Data Fig. 1), which we hypothesized was due to the presence of a TTTT sequence on the non-PAM containing strand of the NGG1 protospacer. We speculate that this TTTT may act as an RNA polymerase III transcriptional terminator that prevents complete pegRNA PBS transcription from a U6 promoter, resulting in an incompletely transcribed pegRNA that cannot support PE (Supplementary Fig. 2). We did observe F508del correction when NGG2 pegRNAs were used, but no NGG2 PE2 editing strategies yielded average editing efficiencies greater than 0.2% (Fig. 1b).

To enhance correction of F508del, we performed a PE3 screen of two NGG2 pegRNAs (NGG2 PBS13 RTT29 and NGG2 PBS14 RTT41) against a panel of all available ngRNA protospacers with pegRNA–ngRNA inter-nick distances of approximately 125 bp (Fig. 1c,d). While the best-performing PE3 NGG2 pegRNA offered up to a 2.6-fold improvement in edit installation efficiency over PE2, the maximum mean editing for any NGG2 PE3 edit did not exceed 0.5%, suggesting that PE3 systems alone cannot efficiently correct CFTR F508del.

We hypothesized that the inefficient correction of F508del using NGG1 and NGG2 could be due to inaccessibility of the edit site to Cas effectors due to chromatin state, which has been reported to negatively affect all major forms of mammalian cell genome editing^9,62–64. CFTR resides within a topologically associated domain whose chromatin state is tightly controlled by protein interactions with numerous enhancers and insulators embedded within the gene⁶⁵. To distinguish chromatin accessibility from prime editor-specific editing limitations, we targeted both NGG1 and NGG2 with pegRNA-guided ABE8e-SpCas9(D10A) adenine base editor (ABE)⁶⁶ to test if a base editor could edit either protospacer. Reassuringly, we observed efficient A•T-to-G•C editing at adenine bases within the expected editing window of ABE8e at NGG1 and NGG2, averaging up to 45% and 29%, respectively (Fig. 1e), demonstrating that NGG1 and NGG2 were indeed accessible by pegRNA-guided Cas effectors. These results suggest that suboptimal execution of the PE steps that follow target site engagement was primarily responsible for inefficient CFTR F508del correction.

Several advances in PE were developed during the course of this study (Fig. 2a). The first of these improvements was the development of engineered pegRNAs (epegRNAs), which protect pegRNA RTT and PBS sequences from endogenous exonuclease degradation by appending an RNA pseudoknot motif to pegRNA 3′ ends⁴⁰. We also reported that inhibition or evasion of cellular DNA MMR enhances PE purity and efficiency⁴¹ and that PEmax, an architecture-optimized prime editor protein, further increased editing efficiency⁴¹. Recently, the PE6 suite of laboratory-evolved and engineered RTs and prime editor Cas9 domains with enhanced editing capabilities were reported³⁴. Given that these PE enhancements address distinct bottlenecks in the PE mechanism, we sought to combine their capabilities to improve CFTR F508del correction.

We first screened 178 epegRNAs with combinations of PBS and RTT lengths at seven protospacers to explore correction from several F508del-proximal protospacers (Extended Data Figs. 2 and 3). Four protospacers had NGG PAMs (NGG1–NGG4, including NGG1 and NGG2 from our initial PE2 and PE3 experiments) and three protospacers had NGA PAMs (NGA1–NGA3), which have been demonstrated as targetable by prime editors with SpCas9-VRQR(H840A) Cas9 nickase domains⁶⁷. This first screen was completed using PE4max with epegRNAs (where PE4 denotes PE2 + MLH1dn, a dominant-negative MLH1 variant that impedes MMR). We observed marginal rates of editing with the NGA1, NGA2, NGG1, NGG3 and NGG4 epegRNAs (<0.1% F508del correction) but up to 0.40% average correction of F508del with NGA3 (Extended Data Figs. 2b–d and 3b–d). For NGG2, we were motivated by the detectable editing in our earlier PE2 and PE3 experiments to test a wide panel of 96 epegRNAs with different combinations of PBS and RTT lengths (Fig. 2b and Extended Data Fig. 3e). From the NGG2 PE4max panel we identified the epegRNA NGG2 PBS13 RTT41 strategy, which yielded a mean PE4max editing rate of 0.61%, an approximately 3.8-fold improvement over our previous best PE2 pegRNA outcome. Encouraged by the editing efficiency enhancement offered by epegRNAs, PEmax and MLH1dn, we elected to move forward with further F508del edit optimization with the NGG2 PBS13 RTT41 epegRNA.

To further develop the editing strategy to correct F508del, we tested the ngRNA panel from our earlier PE3 experiments with the epegRNA NGG2 PBS13 RTT41 in a PE5max experiment (where PE5 denotes PE3 + MLH1dn) (Extended Data Fig. 4). We identified a +104 nick that enabled a mean editing rate of 0.86% when used with the epegRNA NGG2 PBS13 RTT41, representing a twofold improvement over the previous best PE3 editing strategy and further demonstrating the utility of epegRNAs, PEmax and MLH1dn when paired with a ngRNA.

Once we identified a suitable epegRNA and ngRNA pair, we sought to further enhance PE by installing benign silent edits simultaneously with the CTT insertion that corrects F508del. We selected silent edits with two purposes: (1) to disrupt NGG2’s PAM and/or protospacer seed sequence, and thereby enhance edit installation efficiency¹, and (2) to install additional silent edits around the PAM-disrupting edits, thereby creating a contiguous or semi-contiguous tract of mismatches that would cause the PE heteroduplex intermediate to evade MMR, enhancing editing efficiency⁴¹. In PE5max experiments with the NGG2 PBS13 RTT41 epegRNA, we tested four PE strategies that co-installed silent edits with F508del correction (strategies SE1 to SE4). These strategies took advantage of the flexible third nucleotide position in phenylalanine, glycine, valine and tyrosine codons and the ability of serine codons to be completely recoded (Fig. 2c and Extended Data Fig. 5). Compared with correction of F508del alone without any additional silent mutation installation (SE0), SE2’s co-installation of two NGG2 PAM-disrupting edits and three additional silent mutations resulted in a large (fivefold) improvement in F508del correction, with a mean editing rate of 6.8% (Fig. 2d). This result suggests that additional silent edits to disrupt an epegRNA protospacer’s PAM and enhance MMR evasion can offer major increases in PE efficiency, even when the co-installed silent edits are distal from the corrective edit.

The last PE enhancement that we evaluated in the F508del HEK293T cell line was the use of recently evolved PE6 prime editor protein variants³⁴. Using the optimized NGG2 PBS13 RTT41 SE2 epegRNA with MLH1dn and the +104 ngRNA, we compared PEmax with the PE6 variants PE6a–PE6g (Fig. 2e). We observed enhanced F508del correction with PE6b and PE6c compared with PEmax, with mean editing rates of 9.8%, 11% and 7.4%, respectively, consistent with previous observations that PE6b and PE6c enhance editing efficiencies at a variety of targets³⁴. Notably, PE6c offered a 1.5-fold improvement in editing activity over PEmax using the NGG2 PBS13 RTT41 SE2 epegRNA, consistent with our previous observations that PE6c can outperform PEmax at many challenging therapeutic edits that use a long RTT³⁴. These enhancements of PE efficiencies demonstrate that, when matched with edits tailored to their respective strengths, PE6 variants can synergistically improve PE efficiencies in coordination with other PE improvements including epegRNAs, MLH1dn and silent edits.

Having identified a suitable PE6-based strategy for the correction of CFTR F508del in HEK293T cells, next we tested and further optimized the PE strategy in immortalized human bronchial epithelial cells. We used 16HBE14o- cells homozygous for F508del and M470 (16HBEge-F508del) generated by the Cystic Fibrosis Foundation⁶⁸. These cells show reduced CFTR protein expression and impaired chloride channel function⁶⁸. We delivered the prime editor components identified in the above experiments into 16HBEge-F508del cells as RNA (in vitro-transcribed mRNA for PE and MLH1dn proteins; chemically synthesized RNAs for epegRNA and ngRNA) via electroporation. We found that epegRNAs protected with 3′-phosphonoacetate (mP) modifications enhanced editing efficiency by 1.3-fold over epegRNAs protected with 3′-phosphorothioate modifications, consistent with past studies that found the mP modification to enhance editing by Cas9 nucleases⁶⁹ (Extended Data Fig. 6a). We recommend using synthetic epegRNAs protected with 3′-mP modifications instead of 3′-phosphorothioate modifications to maximize editing efficiency when delivering prime editor components using RNA electroporation.

CFTR is actively transcribed in 16HBEge-F508del cells⁶⁸, suggesting that the chromatin state of the gene may make it more accessible to gene editing agents than it is in HEK293T cells. We nonetheless sought to determine whether chromatin accessibility or prime editor-specific editing constraints may be the limiting factor for F508del correction in 16HBEge-F508del cells. To assess these possibilities, we targeted ABEs as well as cytosine base editors to NGG and NGA protospacers adjacent to F508del using standard Cas9 single-guide RNAs (sgRNAs) and pegRNAs. We observed efficient base editing at all protospacers tested (87–98% editing for ABEs and 67–94% editing for cytosine base editors; Extended Data Fig. 6b). However, we observed that editing rates were lower at bases A₁₀ and A₁₂ of protospacer NGG2 when ABE was targeted using a pegRNA or an epegRNA compared with an sgRNA (Fig. 3a). This result suggests that although Cas domains can efficiently engage CFTR in 16HBEge-F508del cells, the pegRNAs or epegRNAs necessary for PE may hinder the editors’ ability to bind the target site, potentially due to the reduced affinity that Cas9 displays for pegRNAs and epegRNAs compared with sgRNAs⁴⁰ resulting in a smaller fraction of prime editors existing in a guide RNA-bound, target-searching state. This observation led us to hypothesize that we may be able to increase the accessibility of prime editors to CFTR by further modulating local chromatin context.

5′-Truncated guide RNAs with only 14–16 nucleotides of protospacer complementarity enable Cas9 to bind to a target sequence but do not support DNA cleavage⁷⁰. We hypothesized that leveraging one such catalytically ‘dead’ sgRNA (dsgRNA) to direct the prime editor protein to a site proximal to F508del would disrupt local chromatin state and increase the accessibility of the prime editor to the site of the therapeutic edit (Fig. 3b); this strategy has been shown to improve PE efficiency in mouse cells⁹. We designed eight NGG-PAM dsgRNAs with 14-nucleotide spacers that bind to sequences within 110 bases of NGG2 and tested their effect on F508del correction when co-delivered with PE6c into 16HBEge-F508del cells. We found that three of the eight tested dsgRNAs substantially improved editing efficiency by an average of 1.9-fold. The best-performing dsgRNA, binding at position −40 relative to the nick location specified by the NGG2 epegRNA, enabled a twofold increase in CFTR F508del correction (to a mean efficiency of 42.6%) compared with a control that used a non-targeting dsgRNA (Fig. 3c). We included the −40 dsgRNA in subsequent experiments with 16HBEge-F508del cells and primary CF airway epithelial cells from patients.

Thus, over the course of our efforts to maximize PE-mediated F508del correction in HEK293T and 16HBEge-F508del cell lines, we identified four enhancements—epegRNAs, silent edits, PE6 variants and dsgRNAs—that each substantially increased editing efficiency. Since these enhancements each improve editing efficiency by a different hypothesized mechanism, we tested how the inclusion of each enhancement, both alone and in combination, contributes to efficient editing in 16HBEge-F508del cells. We found that all four enhancements are essential for maximizing editing efficiency in 16HBEge-F508del cells (Fig. 3d). In the absence of these four key enhancements, PE5max F508del correction averaged 1.8% with 0.57% indel formation. With all four enhancements implemented, we observed a mean F508del correction rate of 51% with 13% indel formation and a 28-fold improvement in editing efficiency, while also increasing the edit-to-indel ratio from 3.1 to 3.9. In a separate experiment, we sought to characterize the influence of MLH1dn on editing efficiency in the presence of these enhancements. We found that omitting MLH1dn led to a modest reduction in F508del correction efficiency from 58% to 52% and a decrease in edit-to-indel ratio from 4.8 to 3.2 (Fig. 3e). These findings indicate that mechanistically distinct improvements to PE systems contribute synergistically to enhanced editing efficiencies.

Given the inefficient editing rates we observed in our initial pegRNA optimization efforts using PE2 and PE4max (Figs. 1b and 2b), we sought to determine whether the epegRNA characteristics we selected remained optimal in the presence of other enhancements. We rescreened epegRNA PBS and RTT lengths at NGG2 in CFTR F508del HEK293T cells using PE6c with MLH1dn, SE2 silent edits, the −40 dsgRNA and the +104 ngRNA and found that PBS13 RTT41 was still among the most optimal PBS and RTT combinations (Extended Data Fig. 7a). Rescreening nicking sgRNAs also indicated that the +104 nsgRNA remained optimal (Extended Data Fig. 7b). These results suggest that optimal epegRNA and ngRNA designs for the CFTR F508del corrective edit remain the same in the context of diverse PE enhancements, and therefore, for at least some target sites, these characteristics can be optimized independently of other improvements to the PE system.

We considered that the marginal PE2 and PE4max editing rates (Figs. 1b and 2b) may be the result of epegRNA PBS–RTT instability or cis-acting interactions between the spacer and the PBS. To circumvent these potential issues, we tested the prime editing template RNA (petRNA) system in which epegRNAs are split into separate sgRNAs and circularized RTT–PBS transcripts⁷¹. Using combinations of an NGG2 sgRNA and petRNAs with varied PBS and RTT lengths, we rescreened F508del correction in CFTR F508del HEK293T cells using MLH1dn, SE2 silent edits, the −40 dsgRNA and the +104 ngRNA (Extended Data Fig. 7c). As previously described⁷¹, an nCas9 (here based on PEmax) and an MS2 coat protein (MCP)–RT (here using a PE6c-based RT) were used for petRNA editing. While we were able to recapitulate efficient petRNA editing at DNMT1 and RUNX1 positive control targets (Extended Data Fig. 7d), we did not observe CFTR F508del correction greater than 4%, indicating that the use of petRNAs did not enhance PE correction of F508del.

We also sought to assess whether epegRNA screening could be accelerated using computational tools for pegRNA design. Using PE6c with MLH1dn, the −40 dsgRNA and the +104 ngRNA, we tested the F508del correction efficiencies of the top 24 epegRNAs predicted by PRIDICT⁷² and the top 24 epegRNAs predicted by DeepPrime⁷³. However, neither model predicted NGG2 PBS13 RTT41 within their top 24 designs; this epegRNA outperformed all others generated from the suggestions of both models in the context of both SE0 and SE2 (Extended Data Fig. 8). On the basis of these results, we recommend experimental epegRNA screening to identify epegRNAs that yield the highest editing efficiencies, rather than relying solely on current computational models.

Encouraged by the development, optimization and characterization of an efficient PE CFTR F508del correction strategy in 16HBEge-F508del cells, we next tested our F508del correction strategy in primary human airway epithelial cells from three patients with homozygous CFTR F508del alleles. Using PE6c with MLH1dn, the −40 dsgRNA, the +104 ngRNA and the NGG2 PBS13 RTT41 SE2 epegRNA delivered by RNA electroporation, we observed a mean F508del correction rate of 25% with 6.5% indels (Fig. 4a) among all treated cells with no sorting.

To determine the phenotypic impact of correcting F508del in primary CF airway epithelial cells, we quantified CFTR-mediated anion secretion in prime-edited donor cells under short-circuit conditions following differentiation for 3 weeks at an air–liquid interface (ALI). The change in short-circuit current (I_sc) was recorded following stimulation with the CFTR-activating agents forskolin and 3-isobutyl-1-methylxanthine (IBMX; ‘forskolin and IBMX’ is abbreviated as F&I) and inhibition with the CFTR-specific channel inhibitor CFTR(inh)-172. Donor-matched mock-treated epithelial cells (electroporated without RNA), epithelial cells treated with elexacaftor, tezacaftor and ivacaftor (ETI) and non-CF epithelial cells were included as electrophysiological controls. Compared with mock-treated CF donor epithelial cells, F&I stimulation resulted in substantially increased CFTR-dependent Isc in the prime editor-treated cells and CFTR(inh)-172 administration likewise decreased the CFTR-dependent Isc in the prime editor-treated cells (Fig. 4b,c). Compared with treatment with ETI, prime editor-treated cells exhibited a similar change in CFTR-dependent Isc when stimulated with F&I or after CFTR(inh)-172 administration (Fig. 4b,c). These results demonstrate substantial rescue of CFTR channel activity following correction by PE, with CFTR function in prime editor-treated donor cells exceeding 50% of current levels in healthy (non-CF) airway epithelial cells (Fig. 4c).

To assess the persistence of F508del correction during cellular differentiation, we evaluated F508del correction rates in airway epithelial cells from two ALI-cultured populations. We found that rates of F508del correction remained consistent when compared at 3 days post-electroporation and after 3 weeks of differentiation at ALI (Supplementary Fig.), demonstrating the edit’s stability during cellular expansion. These findings demonstrate that the optimized PE strategy not only restores expression of functional CFTR capable of much-improved anion transport in primary CF airway epithelial cells from patients, but also that the corrective edit persists through cellular proliferation and differentiation. 3

Having established that enhanced PE systems can efficiently correct CFTR F508del and restore CFTR function in primary CF airway epithelial cells, we sought to characterize unintended editing outcomes of our PE strategy—including epegRNA scaffold integration, partial silent edit incorporation and genomic off-target editing—in detail. The highly processive Tf1 RT used in PE6c can generate indels by reverse transcribing the epegRNA scaffold, which can be incorporated into the target genomic DNA³⁴. To determine if this scaffold incorporation occurs in prime-edited primary CF airway epithelial cells, we analysed high-throughput sequencing (HTS) reads for partial and complete scaffold integration events. The first base of the epegRNA scaffold, indexed from the 3′ end, is complementary to the first base of genomic DNA encoded beyond the RTT; we therefore cannot distinguish between 0 and 1 bases of scaffold integration, as these lead to identical (scaffold-free) sequence outcomes. Across three edited donor cell lines, we observed no scaffold incorporation in 98.8–99.9% (mean 99.2%) of reads containing F508del CTT insertion (Extended Data Fig. 9a). We also observed no additional scaffold incorporation in the three replicates of 16HBEge-F508del cells edited with high-processivity PE6c (percentage of CTT insertion events that are scaffold-free: 97.9–98.0%, mean 98.0%) compared with PEmax (percentage of CTT insertion events that are scaffold-free: 97.3–98.7%, mean 98.0%) (Extended Data Fig. 9b). These findings are consistent with the previous observation that highly processive PE6 variants do not increase rates of scaffold incorporation when carefully matched with edits suited to their capabilities (such as using PE6c to enhance editing with the long 3′ extension of the NGG2 PBS13 RTT41 SE2 epegRNA)³⁴. As predicted by NUPACK⁷⁴, the folding energy of the PBS and RTT of the NGG2 PBS13 RTT41 SE2 epegRNA is −7.80 kcal mol⁻¹. The choice of PE6c for this edit is consistent with our prior findings that Tf1-based PE variants (PE6b and PE6c) frequently outperform alternatives for epegRNAs with predicted secondary structure folding energies smaller in magnitude (less stable) than −23 kcal mol⁻¹ (ref. ³⁴).

The CFTR F508del correction strategy developed above installs silent edits—most notably, converting a serine codon from TCC to AGT—to evade cellular MMR mechanisms. Partial incorporation of these silent edits could lead to non-synonymous changes in protein sequence. We therefore searched for the occurrence of all possible combinations of SE2 partial silent edit incorporation, both with and without concomitant CTT insertion, in treated primary airway epithelial cells. We found that partial silent edits are incorporated at a low frequency without CTT insertion, and we did not observe the incorporation of any partial edits that would lead to outcomes encoding non-synonymous protein sequences (Extended Data Fig. 9c,d). Importantly, some of these silent edits also disrupt Cas9 protospacer and PAM sequences and thus should prevent Cas9 re-engagement after editing. We observed the incorporation of protospacer and PAM edits without concomitant CTT insertion in primary CF airway epithelial cells with a mean frequency of 3.6% (compared with a mean CTT insertion frequency of 23%). This editing outcome effectively renders a small fraction of F508del-containing sequences un-targetable by PE, eliminating their ability to be corrected. The increase in CTT insertion frequency we observe when installing silent edits therefore greatly outweighs the fraction of un-correctable sequences generated due to partial silent edit incorporation. We recommend that partial silent edit incorporation be investigated in future therapeutic PE studies, especially if some silent edits encode changes to the protospacer or PAM sequences.

To characterize off-target editing frequencies of the optimized PE strategy, we used CIRCLE-seq⁷⁵ to identify sites within the human genome that are cleaved in vitro by Cas9 guided with the epegRNA, ngRNA or dsgRNA used in our CFTR F508del PE strategy. We assessed editing outcomes in primary CF airway epithelial cells from three separate donors at the top 32 candidate sites for each of the three guide RNAs, representing 96 total CIRCLE-seq-nominated off-target candidate sites investigated (Supplementary Tables 3–5). We observed few differences in off-target indel or substitution formation in PE-treated donor cells compared with untreated donor cells (Fig. 4d and Extended Data Fig. 10a–c), consistent with the observation from multiple laboratories that PE induces few detected off-target modifications, probably due to its mechanism requiring multiple DNA hybridization events^1,3,5–17. Of the 96 total CIRCLE-seq nominated off-target sites we assessed in primary CF airway epithelial cells, 4 showed a mean 1.5-fold or higher increase in indel formation in the treated cells compared with the untreated cells (Fig. 4d), potentially suggestive of off-target editing. No off-target sites showed a mean 1.5-fold or higher increase in substitution formation in treated cells compared with untreated cells.

The indel frequencies at all four of these potential off-target sites were ≤0.1% in treated samples across all replicates, indicating that the off-target modification at these sites occurs at a very low frequency. We nonetheless analysed these sites more closely to assess their potential physiological relevance (Fig. 4d). One of the four sites is engaged by the epegRNA (epegRNA-OT01) and targets the EPCAM-DT long non-coding RNA, which has no known function. The remaining three sites are engaged by the ngRNA. Two of these sites lie within intergenic regions and are unlikely to have strong physiological consequences if disrupted at a low frequency. The final site (ngRNA-OT18) is situated within an intronic region of RUNX1⁷⁶. Because ngRNA-OT18 is located ~60 kb from the nearest exon and showed a low mean indel formation rate of 0.048% in treated primary cells, we do not anticipate this off-target site to carry substantial physiological consequences. No potential off-target sites were detected from the dsgRNA, which may be explained by the inability of Cas9 nuclease to cleave DNA targets when guided with a 6-bp truncated guide sequence⁷⁰. These observations collectively suggest that the PE CFTR F508del correction strategy induces few off-target edits in the human genome under the tested conditions.

Guide RNA expression plasmids (pegRNAs, epegRNAs, ngRNAs, dsgRNAs and sgRNAs) were cloned as previously described using isothermal assembly and synthetic gene fragments⁶¹. Guide RNA sequences are provided in Supplementary Table 2. Guide RNA expression plasmids were purified using PureYield Plasmid Miniprep kits (Promega). Prime editor and MLH1dn plasmids were purified using Qiagen Plasmid Plus Midi kit (Qiagen). DNA PCR amplification was completed using Phusion U Green Multiplex PCR master mix (Thermo Fisher Scientific). Primers and gene fragments were ordered from Integrated DNA Technologies. New prime editor plasmids were cloned using isothermal assembly, and sequences are provided in Supplementary Information.

Synthetic pegRNAs and epegRNAs were ordered from Agilent Research Labs and contained 2′-O-methyl modifications at the first three nucleotides, 2′-O-methyl modifications at the third-to-last and second-to-last nucleotides, 3′-phosphorothioate linkages between the first three nucleotides and 3′-mP linkages between the last two nucleotides. Synthetic ngRNAs were ordered from Synthego and contained 2′-O-methyl modifications at the first three and last three nucleotides and 3′-mP linkages between the first three and last two nucleotides. Synthetic dsgRNAs were ordered from Integrated DNA Technologies and contained 2′-O-methyl modifications at the first three and last three nucleotides and phosphorothioate linkages between the three first and last three nucleotides. Guide RNA sequences are provided in Supplementary Table. 2

The prime editor, ABE, cytosine base editor and MLH1dn-encoded mRNA were generated using in vitro transcription as previously described⁶¹. Briefly, the prime editor, base editor or MLH1dn transcripts, containing a 5′ untranslated region, Kozak sequence, prime editor or MLH1dn open reading frame, and 3′ untranslated region were PCR amplified from a template plasmid containing an inactive T7 (dT7) promoter. PCR primers repaired this dT7 promoter and also installed a 119 nt poly(A) tail. The purified PCR double-stranded DNA amplicon was used as an in vitro transcription template using the HiScribe T7 high-yield RNA synthesis kit (NEB). In vitro transcription followed the manufacturer’s optional protocol to include CleanCap reagent AG (Trilink) and substitute N1-methylpseudouridine-5′-triphosphate (Trilink) for uridine triphosphate. Lithium chloride precipitation was used to purify mRNA from complete in vitro transcription reactions, and mRNA transcripts were reconstituted in nuclease-free water.

HEK293T (ATCC CRL-3216) cells were purchased from American Type Culture Collection and cultured in Dulbecco’s modified Eagle medium with GlutaMax (Thermo Fisher Scientific) supplemented with 10% foetal bovine serum (Thermo Fisher Scientific) at 37 °C with 5% CO₂, as previously described⁶¹.

16HBEge-F508del cells homozygous for CFTR F508del were a gift from the Cystic Fibrosis Foundation Therapeutics Lab and were cultured as previously described⁶⁸. Briefly, 16HBEge-F508del cells were cultured in minimum essential medium (Thermo Fisher Scientific) supplemented with 10% foetal bovine serum (Thermo Fisher Scientific) and 1× penicillin–streptomycin (Thermo Fisher Scientific) at 37 °C with 5% CO₂. For routine cell culture, standard tissue culture flasks were pre-treated with a thin layer of coating solution at 37 °C 5% CO₂ for 2 h, followed by coating solution removal and coated flask storage at 4 °C. Coating solution composition was LHC-8 basal medium (Thermo Fisher Scientific) supplemented with 1.34 µl ml⁻¹ bovine serum albumin 7.5% (Thermo Fisher Scientific), 10 μl ml⁻¹ bovine collagen solution type 1 (Advanced BioMatrix) and 10 µl ml⁻¹ fibronectin from human plasma (Thermo Fisher Scientific).

Primary airway epithelial cells from non-CF donors were isolated from trachea or bronchi from post-mortem lungs unsuitable for transplant. Primary CF airway epithelial cells were isolated from lung tissue obtained following lung transplant. Studies were approved by the University of Iowa Institutional Review Board, under the US Department of Health and Human Services registration number IRB00000099. The genotypes of the CF donors were F508del/F508del (n = 3 donors). Cells were expanded on culture plates coated with human collagen IV (Sigma) in PneumaCult-Ex Plus medium (STEMCELL Technologies). To establish ALI cultures, cells were seeded onto collagen-coated, semi-permeable membranes (0.33 cm², polycarbonate, Costar) and differentiated in PneumaCult-ALI medium (STEMCELL Technologies) for ≥3 weeks before functional assays.

All cell lines were verified to be free of mycoplasma and identity-authenticated by their suppliers.

HEK293T transfections were conducted as previously described⁶¹. Briefly, approximately 16,000 HEK293T cells were plated per well on a 96-well plate in complete culture media. After 18–24 h, cells were transfected at 70–80% confluency with variable amounts of plasmid DNA and 0.5 µl of Lipofectamine 2000 diluted in Opti-MEM (Thermo Fisher Scientific), following the manufacturer’s instructions.

For a single 96-well transfection using standard prime editor systems, the following plasmids amounts were transfected: 200 ng of prime editor; 50 ng of pegRNA, epegRNA or sgRNA; 15 ng of ngRNA (if included in experiment); 15 ng of dsgRNA (if included in experiment); and 100 ng of MLH1dn (if included in experiment).

For a single 96-well transfection using the petRNA system with a split prime editor⁷¹, the following plasmid amounts were transfected: 100 ng of the PEmax nCas9, 100 ng of the MCP–RT(PE6c), 25 ng of petRNA, 25 ng of petRNA-paired sgRNA, 15 ng of ngRNA, 15 ng of dsgRNA and 100 ng of MLH1dn.

Following transfection, cells were incubated at 37 °C with 5% CO2 for 72 h.

Electroporation of 16HBEge-F508del cells and primary airway epithelial cells from patients with CF was completed with the SG Cell Line 4D-Nucleofector X Kit (Lonza) and a 4D-Nucleofector (Lonza). Cells from maintenance culture were dissociated using TrypLE (Thermo Fisher Scientific), and 200,000 cells were resuspended in 20 µl complete SG nucleofector solution per reaction. PE RNA reagents were added the nucleofector solution cell mix as follows: 1 µg prime editor mRNA, 1 µg MLH1dn mRNA, 90 pmol pegRNA or epegRNA, 30 pmol ngRNA, and 30 pmol dsgRNA. The cell–RNA mixture was transferred into a nucleofector 16-well strip and electroporated using the programme CM-137. Following electroporation, 80 µl of pre-warmed growth media was added to each cuvette and the 100 µl cell mix was used to seed tissue culture plates.

For 16HBEge-F508del cells, cells were transferred onto a 24-well tissue culture plate with pre-warmed 16HBEge-F508del cell growth media and incubated at 37 °C in 5% CO₂ for 6 days. On day 4, cells were rinsed with phosphate-buffered saline to eliminate dead cells and subsequently incubated with fresh growth media.

For primary CF airway epithelial cells, cells were transferred onto collagen-coated six-well plates with pre-warmed Ex Plus medium and incubated at 37 °C in 5% CO₂. Cells were rinsed with phosphate-buffered saline to eliminate dead cells and subsequently incubated with fresh Ex Plus medium the following day. Seventy-two hours after electroporation, cells were dissociated with TrypLE and seeded onto collagen-coated semi-permeable Transwell membranes at density of 1.25 × 10⁵ per filter in Ex Plus medium. Cells were maintained in Ex Plus medium for 2–4 days and subsequently switched to ALI medium. The cells were differentiated at ALI for 3 weeks before functional analysis. A cell aliquot obtained during filter seeding was used for genomic DNA extraction and HTS analysis.

Genomic DNA (gDNA) was isolated from HEK293T and 16HBEge-F508del cell culture following a custom lysis protocol and paramagnetic bead extraction. To lyse cells, growth medium was carefully removed from cell culture plates and 100 µl or 250 µl of lysis buffer (100 mM Tris-HCl pH 8.0, 200 mM NaCl, 5 mM EDTA, 0.05% SDS, 4.0 mg ml⁻¹ proteinase K (New England Biolabs), and 12.5 mM dithiothreitol) was added to each well of a 96-well or 24-well plate, respectively. Plates with lysis buffer were incubated for 20 h at 55 °C with shaking. To extract gDNA, one volume of lysate was thoroughly mixed with one volume of Ampure XP beads (Beckman Coulter Life Sciences) and incubated for 5 min, separated on plate magnet, washed with 70% ethanol three times with bead resuspension and eluted in 50 µl nuclease-free water.

Genomic DNA from primary CF airway epithelial cells was extracted using QuickExtract (LGC Biosearch Technologies) lysis solution, following the manufacturer’s protocol.

Briefly, for a single confluent well of HEK293T cells in a 96-well plate: 72 h after transfection, medium was carefully removed and 30 µl TrypLE Express (Thermo Fisher Scientific) was used to coat the cell monolayer, which was then incubated at 37 °C with 5% CO2 for 5 min. Following incubation, 70 µl complete HEK293T cell growth medium was added and cells were resuspended by pipetting. The cell–medium mix was centrifuged at 250g for 5 min, medium supernatant was removed and pelleted cells were resuspended in 500 µl phosphate-buffered saline (Thermo Fisher Scientific), and the cell–medium mix was passed through a 35-μm cell strainer (Corning). Cells were sorted using a MA900 Cell Sorter (Sony Biotechnology) at the Broad Institute Flow Cytometry Core. See Supplementary Fig. 5 for the fluorescence-activated cell sorting gating strategy. Cells were sorted into custom HEK293T cell lysis buffer (listed above), and gDNA was extracted as described for the custom lysis buffer (procedure listed above).

HTS of genomic loci was completed as previously described⁶¹. All HTS primers are provided in Supplementary Table 1. Briefly, two rounds of PCR amplification were performed using Phusion U Green Multiplex PCR master mix (Thermo Fisher Scientific). In the first round of PCR amplification, approximately 150 ng of gDNA was used to template a PCR with primers containing Illumina adaptor overhangs (Supplementary Table 1) and cycled under the following conditions: 95 °C for 3 min, 27–30 cycles of 95 °C for 10 s, 58–61 °C (corresponding to the experimentally optimized Tm) for 20 s and 72 °C for 30 s, followed by 72 °C for 5 min. In the second round of PCR, combinations of Illumina-barcoded forward and reverse primers were used to uniquely identify each sample and, with 1–2 µL of PCR round 1 as a template, were amplified under the following conditions: 95 °C for 3 min, seven cycles of 95 °C for 10 s, 61 °C for 20 s and 72 °C for 30 s, followed by 72 °C for 5 min. PCR round 2 reactions from the same genomic locus were pooled and gel extracted with QIAquick gel extraction kit (Qiagen) to eliminate primer dimers. Gel-extracted amplicons were quantified using Qubit double-stranded DNA high-sensitivity assay kit (Thermo Fisher Scientific), and 4 nM libraries were run on an Illumina MiSeq 300 v2 Kit with 200–300 cycles.

Sequencing reads were de-multiplexed using MiSeq Reporter (Illumina) and amplicon sequences were aligned to a reference sequence using CRISPResso2¹⁰², as previously described⁶¹. For PE experiments, the parameter --quantification_window_coordinates (QWC) was set as the amplicon coordinates 10 bp 5′ upstream from the nick position of the most 5′ epegRNA, pegRNA, ngRNA or dsgRNA and 10 bp 3′ downstream from the nick position of the most 3′ epegRNA, pegRNA, ngRNA or dsgRNA. If the end of the 3′ flap generated by the RT plus 10 bp extended past one of these defined QWC bounds, the amplicon coordinate of the 3′ flap generated by the RT plus 10 bp was used in place of the superseded QWC bound. For all CRISPResso2 runs the discard_indel_reads parameter was set as ‘TRUE’, the q parameter set as ‘30’ and, for base editing experiments, the parameter w was used in place of QWC and set to ‘10’. Additionally, in all CRISPResso2 runs, indels for a given sample were calculated as ((‘Discarded’/‘Reads_aligned_all_amplicons’) × 100) using data from the ‘CRISPRessoBatch_quantification_of_editing_frequency.txt’ output file. For single-base changes, editing efficiency at a given position was calculated as ((frequency of specified point mutation from the ‘Nucleotide_percentage_summary.txt’ file) × (‘Reads aligned’/‘Reads_aligned_all_amplicons’) × 100), where the ‘Reads aligned’ and ‘Reads_aligned_all_amplicons’ values were collected from the ‘CRISPRessoBatch_quantification_of_editing_frequency.txt’ output file. For precise PE insertion or deletion edits, CRISPResso2 was run in HDR mode and editing efficiency was calculated as ((HDR amplicon ‘Reads_aligned’/‘Reads_aligned_all_amplicons’) × 100) using the ‘CRISPRessoBatch_quantification_of_editing_frequency.txt’ output file.

The monoclonal CFTR F508del HEK293T cell line was isolated by limiting dilution. Briefly, after identifying the most efficient PE2 strategy to install the CFTR F508del CTT deletion into the endogenous CFTR gene of HEK293T cells (Supplementary Fig. 1), HEK293T cells were transfected with this PE2 strategy following the methods listed above. After 72 h, transfected cells were dissociated from adherent culture, counted and plated across ten 96-well plates in 100 µl of medium per well at a concentration of 0.5 cells per each well. Cells were grown for 10 days and monitored for the development of monoclonal cell populations. Identified monoclonal cultures were expanded and genotyped following the HTS methods listed above. One clonal HEK293T cell line homozygous for CFTR F508del was isolated.

Web servers for DeepPrime (https://deepcrispr.info/DeepPrime/↗) and PRIDICT 2.0 (https://pridict.it↗) were used to design pegRNA sequences for CTT insertion at CFTR F508del. Default settings were used for both models. The desired edited sequence used as a model input was simple CTT insertion without concomitant silent edits (that is, SE0). The 24 pegRNA sequences with the highest predicted editing efficiency were synthesized as epegRNAs and tested experimentally. SE2 epegRNAs were designed by incorporating the SE2 silent edits into the SE0 pegRNA sequences generated by each model.

A total of 3 weeks following electroporation, primary airway epithelial cell cultures were treated for 24 h with 10 µM forskolin (Cayman) and 100 µM IBMX (Sigma) before bioelectric studies. Donor-matched naive CF cultures treated for 24 h with ETI (2 μM of elexacaftor (VX-445, Cayman), 18 μM of tezacaftor (VX-661, Cayman) and 1 μM of ivacaftor (VX-770, Cayman) in the presence of 10 mg ml⁻¹ normal human serum (Sigma), as described previously⁵⁴, were used as a control for CFTR modulator treatment. Non-CF cultures treated with 10 µM forskolin and 100 µM IBMX were used as non-CF positive control.

Epithelial cell cultures were mounted in Ussing chambers (Physiologic Instruments) in symmetrical Cl⁻-buffered Ringer’s solutions consisting of (mM): 135 NaCl, 5 HEPES, 0.6 KH₂PO₄, 2.4 K₂HPO₄, 1.2 MgCl₂, 1.2 CaCl₂ and 5 dextrose, with pH titrated to 7.40 at 37 °C with NaOH. The command voltage was set to 0 mV and I_sc was monitored. The following drugs were used during the assay: apical amiloride (Sigma-Aldrich, 100 μM from 100 mM DMSO stock) to inhibit ENaC, apical 4, 4’-dilsothiocyano-2, 2’-stilbenedifulonic acid (DIDS, Sigma-Aldrich, 100 μM from 100 mM DMSO stock) to inhibit non-CFTR Cl^– channels, apical forskolin (Cayman Chemical, 10 μM from 10 mM DMSO stock) and IBMX (100 μM from 100 mM ethanol stock) to stimulate CFTR channel activity, and CFTR_inh-172 (Sigma-Aldrich, 10 µM from 10 mM DMSO stock) to block CFTR.

NUPACK⁷⁴ (https://old.nupack.org/↗) was used to predict the secondary structure and folding energies of pegRNAs. The pegRNA RTT and PBS (for epegRNA NGG2 PBS13 RTT41 SE2: UGGCACCAUUAAAGAAAAUAUCAUCUUUGGUGUGAGUUACGAUGAAUAUAGAUA) sequence was queried using default parameters (type: RNA, temperature: 37 °C, maximum complex size: one strand).

Quantification of pegRNA scaffold insertion was completed using a custom Python script as described previously¹. The script searches for occurrences of sequences of the pegRNA scaffold adjacent to on-target edits within HTS data, beginning with a single base at the 3′ end of the scaffold (proximal to the pegRNA RTT, referenced in the analysis as base 1). The script iterates through the entire scaffold sequence; in its ith iteration, insertion events representing integration of scaffold sequences 1 − i are quantified. Scaffold insertion was quantified using genomic DNA from three prime-edited primary CF airway epithelial cell cultures as well as prime-edited 16HBEge-F508del cells. The script used to quantify scaffold integration is available in Supplementary Note 1.

Partial edit incorporation was quantified using CRISPResso2 in HDR mode, as described above (see ‘HTS and data analysis’ section). For each sample, a separate CRISPResso2 run was executed for every possible combination of partial SE2 edit incorporation, both with and without concomitant CTT insertion. Partial edit frequency was calculated as ((HDR amplicon ‘Reads_aligned’/‘Reads_aligned_all_amplicons’) × 100) using the ‘CRISPRessoBatch_quantification_of_editing_frequency.txt’ output file. Partial edit quantification was performed using genomic DNA from three prime-edited primary CF airway epithelial cell cultures as well as three unedited controls.

Off-target site nomination was performed using CIRCLE-seq as previously described¹⁰³. The top 32 CIRCLE-seq nominated off-target sites, as defined by nuclease read count, for each of the three guide RNAs used to edit primary CF airway epithelial cells (epegRNA NGG2 PBS13 RTT41 SE2, ngRNA +104 and dsgRNA −40) were selected for deep sequencing analysis. The nominated sites are listed in Supplementary Tables 3–5. The primers were designed to amplify each of the 96 nominated off-target sites within a 240–280 bp amplicon using National Center for Biotechnology Information Primer-BLAST (Supplementary Table 1). Off-target sites were amplified from purified gDNA and prepared for HTS as described above (see ‘HTS and data analysis’ section). Off-target amplicon sequencing was performed using gDNA from three prime-edited primary CF airway epithelial cell cultures as well as three unedited controls.

Indels and substitutions at nominated off-target sites were quantified from HTS data using CRISPResso2¹⁰². Indels were quantified within a window from 10 bp 5′ upstream of the end of the epegRNA homology template to 10 bp 3′ downstream of the Cas9 nick position; this was achieved by setting the parameter WC to ‘18’ and the parameter W to ‘31’. Substitutions were quantified within a window from 0 bp 5′ upstream of the end of the epegRNA homology template to 0 bp 3′ downstream of the Cas9 nick position; this was achieved by setting the parameter WC to ‘18’ and the parameter W to ‘21’. For both indel and substitution quantifications, the discard_indel_reads parameter was set as ‘TRUE’ and the q parameter was set as ‘30’. Indels were calculated as ((‘Discarded’/‘Reads_aligned_all_amplicons’) × 100) using data from the ‘CRISPRessoBatch_quantification_of_editing_frequency.txt’ output file from the indel CRISPResso2 run. Substitutions were calculated as ((‘Substitutions’/‘Reads_aligned_all_amplicons’) × 100) using data from the ‘CRISPRessoBatch_quantification_of_editing_frequency.txt’ output file from the substitution CRISPResso2 run. Off-target sites were further investigated if the mean indel or substitution rate at the site for prime-edited samples was 1.5-or-more-fold greater than mean indel or substitution rate for the same site in untreated samples.

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

Nature Biomedical Engineering thanks Maarten Geurts, Stephen Hart and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

is available for this paper at 10.1038/s41551-024-01233-3.

The online version contains supplementary material available at 10.1038/s41551-024-01233-3.

All data supporting the results of this study are available within the paper and its Supplementary Information. High-throughput sequencing data are available from the NCBI Sequence Read Archive database (PRJNA1055086↗). Key plasmids are available from Addgene (depositor: D.R.L.), or from the corresponding author on request. Source data are provided with this paper.

The code used for the analysis of the high-throughput sequencing data is available via GitHub at https://github.com/pinellolab/CRISPResso2↗. The assessment of epegRNA scaffold insertion was completed using a custom Python script described previously¹ and provided as Supplementary Information.