Using Prime Editing to Create Suppressor tRNAs for Genome Editing Across Different Diseases

Therapeutic genome-editing efforts, including more than 70 clinical trials so far, have predominantly used programmable nucleases, base editors or prime editors to disrupt or correct disease-associated genes in an allele-specific manner. These approaches have proven to be effective in patients or in animal models for the treatment of disorders such as sickle-cell disease^6,7, T cell leukaemia⁸, hypercholesterolaemia^9,10, alpha-1-antitrypsin deficiency¹⁰, chronic granulomatous disease¹¹, progeria¹², spinal muscular atrophy¹³, prion disease¹⁴, alternating hemiplaegia of childhood¹⁵ and many other genetic diseases.

Although allele-specific therapeutic genome-editing strategies offer treatments for many serious diseases with few treatment options, the breadth of the global genetic disease crisis, in which more than 8,000 genetic diseases collectively affect hundreds of millions of patients, demands new approaches that more rapidly bring the benefits of therapeutic genome editing to large numbers of patients¹⁶. Allele-specific applications of nucleases, base editors and prime editors require the development of a distinct genome-editing treatment for each of the more than 200,000 known pathogenic mutations, although prime editing also enables corrections of clustered hotspots of mutations. Until substantial streamlining of regulatory, development and manufacturing costs occurs, the development of the thousands of distinct drugs needed to treat any large fraction of patients with genetic disease remains impractical, despite the fact that base editing and prime editing can collectively correct the vast majority of known pathogenic mutations^1–3.

The versatility of prime editing, which uses a programmable nickase and a reverse transcriptase to replace targeted segments of DNA with new sequences of our choosing³, can be used in creative ways that may benefit far larger cross-sections of patients with genetic disease. Nonsense mutations account for 24% of pathogenic alleles in the ClinVar database¹⁶. The common molecular consequence of nonsense mutations—termination of translation before functional protein can be made—suggests more generalized approaches to their therapeutic rescue that might be applied in an allele-agnostic or even disease-agnostic manner.

Nonsense sup-tRNAs have an anticodon that complements a premature termination codon (PTC) and support the installation of an amino acid rather than termination of translation. Although sup-tRNAs can theoretically read through both PTCs and natural termination codons (NTCs), extensive evidence suggests that sup-tRNAs can be well-tolerated in eukaryotes. Both naturally occurring and exogenous sup-tRNAs can be expressed without apparent toxicity^4,5. Multiple biological mechanisms explain the observation of surprisingly low levels of sup-tRNA-mediated readthrough at NTCs. First, the distribution of stop codons for PTCs is distinct from the distribution for NTCs, particularly increasing the safety profile of sup-tRNAs corresponding to the amber stop codon (TAG)¹⁷. Second, the frequent presence of redundant and diverse in-frame stop codons following NTCs reduces the likelihood of extending a protein past a suppressed stop codon by more than a few amino acids¹⁸. Third, recruitment of polypeptide chain release factors to the 3′ untranslated region (3′ UTR) near NTCs can outcompete sup-tRNAs¹⁹. Fourth, if the ribosome continues past the NTC, the RNA and protein are targeted for degradation through the non-stop decay pathway²⁰, and proteins that are translated into the 3′ UTR are further recognized and targeted for degradation²¹. Finally, sup-tRNAs mediate nonsense suppression only in transcripts that are being expressed in a given cell, minimizing the risk of toxicity from ectopic overexpression.

Despite the promise of sup-tRNAs as a potential therapeutic strategy, current approaches such as lipid nanoparticle (LNP)-based or adeno-associated virus (AAV)-mediated delivery of sup-tRNAs^4,5 face the challenge of supporting sup-tRNA production throughout the lifetime of the patient. AAV-based methods, in particular, may be limited to a single dose, owing to the generation of high levels of neutralizing antibodies following dosing, which may not be sufficient to treat genetic diseases that require restoration of protein expression for the lifetime of the patient. In addition, efforts to apply sup-tRNAs as therapeutic agents are stymied by a lack of complete understanding of the determinants of sup-tRNA potency. As a result, recent efforts to rescue nonsense mutations have relied on the overexpression or delivery of high levels of sup-tRNAs with suboptimal suppression efficiency^4,5, even though tRNA overexpression can alter global translation^22–24 and raise the possibility of toxicity. A one-time treatment that would enable permanent expression of endogenous levels of a sup-tRNA optimized for suppression efficiency without apparent toxicity has the potential to transform the treatment of diseases caused by PTCs. Because such a treatment may ameliorate a variety of diseases arising from the targeted type of stop codon, it also has the potential to benefit a large number of patients with multiple disorders caused by many pathogenic PTCs.

Here we evaluated tens of thousands of variants of all 418 high-confidence human tRNAs to engineer highly active sup-tRNAs with sufficient potency to mediate efficient nonsense mutation suppression even when expressed from a single genomic copy with endogenous regulatory elements. By iteratively optimizing: (1) the 40-bp leader sequence of the tRNAs; (2) the tRNA sequence via saturation mutagenesis; and (3) the terminator sequence of the tRNAs, we developed a highly active TAG-targeting sup-tRNA despite its expression remaining at sub-endogenous levels from a single genomic locus. We then used the versatility of prime editing to permanently convert a redundant endogenous human tRNA to these optimized sup-tRNAs. We call this strategy PERT (prime editing-mediated readthrough of premature termination codons).

In human cell models of nonsense mutations in genes known to cause Batten disease (TPP1 p.L211X and TPP1 p.L527X), Tay–Sachs disease (HEXA p.L273X and HEXA p.L274X) and Niemann–Pick disease type C1 (NPC1 p.Q421X and NPC1 p.Y423X), treatment with the same prime editor programmed to install an optimized sup-tRNA resulted in restoration of 20–70% of normal enzyme activity. We experimentally quantified the effectiveness of our prime editing-installed sup-tRNA for reading through all clinically relevant TAG PTCs in the ClinVar database, observing PTC readthrough for the vast majority of sequences tested. Finally, we applied PERT in vivo in a mouse with a co-delivered reporter construct containing a nonsense mutation in GFP, as well as in a mouse model of Hurler syndrome, a severe lysosomal storage disease caused by a premature stop codon (IDUA p.W392X). Treatment with a single composition of matter mediated around 25% production of full-length GFP in reporter mice, and therapeutic levels of approximately 6% IDUA enzyme activity restoration, resulting in nearly complete rescue of disease pathology in Hurler syndrome mice. These findings provide a foundation for the development of disease-agnostic therapeutic genome-editing strategies that have the potential to substantially increase the number of patients that can benefit from a single genome-editing drug.

The human genome encodes 47 isodecoder (anticodon) tRNA families comprising 418 high-confidence genes (Supplementary Table 1). This redundancy, coupled with transcription-induced mutagenesis, yields many tolerated tRNA mutations in humans, including anticodon mutations that create natural suppressor or mis-charged tRNAs^25,26. Fourteen isodecoder families have no corresponding tRNA gene and are decoded by pseudocognate tRNAs²⁷, and entire tRNA families have been deleted without consequence²⁸. Therefore, we hypothesized that converting a single endogenous tRNA to a sup-tRNA with prime editing could be well-tolerated and enable readthrough of nonsense mutations regardless of the gene in which they originate. This approach avoids tRNA overexpression that can perturb global translation^22–24 and preserves native tRNA regulation.

To screen endogenous human tRNAs for their ability to become sup-tRNAs, we designed mCherry-STOP-GFP reporters in which GFP expression occurs only after PTC readthrough (Supplementary Fig. 1a–e and Methods). Next, we designed prime editing reagents to convert the anticodon loops of two endogenous tRNAs (tRNA-Gln-CTG-6-1 and tRNA-Arg-CCG-2-1) into sup-tRNAs. For clarity, we follow standard tRNA nomenclature throughout this work. For example, in tRNA-Gln-CTG-6-1: Gln indicates the associated amino acid; CTG indicates the anticodon; 6 is the isodecoder index that distinguishes tRNA genes in the same family with minor sequence differences; and 1 is the locus copy number (Extended Data Fig. 1a).

We installed these sup-tRNAs at the endogenous loci of tRNA-Gln-CTG-6-1 and tRNA-Arg-CCG-2-1 using prime editing in HEK293T cells. We observed on average 29% conversion of each endogenous tRNA into a sup-tRNA (range: 19%–37%) (Extended Data Fig. 1b). We then overexpressed the reporter construct via plasmid transfection, or expressed it from a single gene copy using lentiviral transduction. We characterized PTC readthrough using two metrics: the percentage of GFP-positive cells (% GFP) and the GFP mean fluorescence intensity relative to a wild-type GFP control (relative protein yield) (Extended Data Fig. 1c). Although both sup-tRNAs led to GFP rescue with an overexpressed reporter, neither led to significant PTC readthrough with a single-copy reporter (Extended Data Fig. 1d). Even with an overexpressed reporter, the per cent rescue of GFP relative to a wild-type control was modest, with on average 7.8% GFP-positive cells and a relative GFP protein yield of 10%, and overexpression of the reporter alone without any sup-tRNA yielded 3.9% GFP-positive cells, presumably from pseudocognate tRNA-mediated PTC readthrough.

Previous sup-tRNA studies have heavily relied on high tRNA levels in transfection, LNP or viral settings, often using elevated copy numbers of the sup-tRNA and/or the PTC-containing mRNA^4,5. These findings revealed the need to identify sup-tRNAs that are sufficiently potent to mediate endogenous genomic PTC readthrough at low sup-tRNA levels that are compatible with genomic installation and minimize the risk of perturbing global translation.

We cloned the epegRNA libraries into a lentiviral backbone and transduced them into a corresponding HEK293T cell line containing the mCherry-STOP-GFP reporter. Fluorescence-activated cell sorting (FACS) isolation of GFP-positive cells enriched cells expressing epegRNAs that generate functional sup-tRNAs (Fig. 1b,c and Supplementary Figs. 3 and 4). We identified arginine, leucine, tyrosine and serine tRNA backbones capable of reading through TAG stop codons (Fig. 1d and Extended Data Fig. 2a–f) and leucine and arginine tRNA backbones that read through TGA stop codons (Extended Data Fig. 3a–d and Supplementary Fig. 5). A similar screen for TAA stop codons did not yield effective sup-tRNAs (Supplementary Fig. 6), consistent with previous observations that TAA is the most stringent stop codon and is more resistant to sup-tRNA-mediated readthrough³⁰.

Whereas epegRNAs that simultaneously targeted multiple tRNA sequences generally performed best, half of the most enriched positive hits in the TAG screen were epegRNAs targeting a single tRNA (Extended Data Fig. 2b–d). Although the epegRNA screen successfully identified endogenous tRNAs that could be converted into effective sup-tRNAs, even the best-performing sup-tRNA from this screen achieved less than 5% protein yield relative to a wild-type GFP control (Extended Data Fig. 2b). To address this limitation, we next engineered all major features of each sup-tRNA to enhance readthrough efficiencies.

A eukaryotic tRNA gene contains three distinct elements: the leader sequence, the tRNA gene body and the termination sequence (Extended Data Fig. 4a). Whereas tRNA expression is typically driven by an internal Pol III promoter, the leader sequence is thought to regulate gene expression³¹. Thus far, most sup-tRNA studies have relied on exogenous promoters for expression⁵, although the role of internal tRNA promoters is being investigated³².

To test tRNA promoter variants, we designed single-copy lentiviral constructs with: (1) an exogenous human U6 promoter (hU6); (2) a minimal human U6 promoter lacking the nucleosome positioning element (minU6)³³; or (3) no exogenous promoter (Extended Data Fig. 4b). We performed three screens using a pool of all 418 mature tRNA sequences, modifying their anticodons to CUA, UCA or UUA to target TAG, TGA or TAA stop codons, respectively. The pooled sup-tRNA libraries were cloned into lentiviral constructs with each promoter variant and transduced into the mCherry-STOP-GFP reporter lines (Supplementary Fig. 2 and Supplementary Table 3).

Unlike the prime editing screen, lentiviral expression of sup-tRNAs with exogenous promoters yielded fewer TAG sup-tRNAs and activity depended on promoter choice (Extended Data Fig. 4c and Supplementary Fig. 7a). Leu-TAA sup-tRNAs were strongest with hU6 (tRNA-Leu-TAA-4-1, tRNA-Leu-TAA-3-1, tRNA-Leu-TAA-1-1 and tRNA-Leu-TAA-2-1, in descending order of activity), whereas Tyr-GTA (tRNA-Tyr-GTA-7-1) performed best with minU6 or no promoter, suggesting that this tRNA may have stronger internal promoter elements (Extended Data Fig. 4c). No effective TGA sup-tRNAs were identified when expressing the sup-tRNAs from a lentiviral construct (Supplementary Fig. 7b). Since exogenous promoters seem to hinder the proper expression of many sup-tRNAs identified in the prime editing screen, we next sought to investigate how endogenous leader sequences and terminator sequences affect sup-tRNA function.

To further evaluate the effect of leader sequences, we analysed the 40 bp upstream of all 418 high-confidence human tRNAs. These sequences are less conserved than the tRNA gene bodies and may influence their expression^31,32. We cloned each leader sequence upstream of a sup-tRNA candidate (tRNA-Leu-TAA-3-1) (Supplementary Fig. 2 and Supplementary Table 4), transduced this pool of constructs into the mCherry-STOP-GFP reporter cell line, and isolated GFP-positive cells via FACS. Eighty-six per cent of the cells expressing a sup-tRNA in this screen were GFP-positive, indicating that most leader sequences support sup-tRNA expression (Extended Data Fig. 4d). We individually tested leader sequences with the highest and lowest enrichment among GFP-positive cells, as well as eight randomly generated control sequences. The top hits performed well (averaging 72% GFP-positive cells) and the bottom hits performed poorly (averaging 6.1% GFP-positive cells), but the random leader controls also performed reasonably well (averaging 57% GFP-positive cells) (Extended Data Fig. 4e). Thus, whereas many diverse leader sequences can support robust sup-tRNA activity, others can nearly abolish activity. No apparent leader consensus sequences explained the observed variation in sup-tRNA activity.

Next, we investigated whether leader sequences could modulate PTC suppression levels across different sup-tRNAs. We selected six types of leader sequences: (1) each tRNA's own endogenous leader; (2) a top-performing leader (Top-6); (3) a bottom-performing leader (Bottom-6); (4) a random leader (Random-4) (Extended Data Fig. 4e); and two leaders upstream of highly expressed tRNAs: (5) tRNA-Cys-GCA-3-1; and (6) tRNA-Ser-GCT-3-1, which could in principle be overwritten with alternative sup-tRNA sequences using prime editing³⁴. In parallel, we explored how downstream termination sequences might affect tRNA expression levels. We identified four potential downstream sequences to test: the endogenous 100-bp sequence downstream of each mature tRNA sequence, and terminators consisting of TTT (3T), TTTT (4T) or TTTTT (5T), which are predicted to terminate transcription from a Pol III promoter with efficiencies of approximately 0%, 75% or 99%, respectively³⁵.

We generated a lentiviral library containing all 6 leader sequences paired with 418 human tRNAs with their anticodons switched to CUA, UCA or UUA. We also assessed the effect of each of the 4 tested terminators placed downstream of sup-tRNAs paired with their endogenous leader sequences, with a total of 11,543 combinations of leader, sup-tRNA and terminator sequences (Supplementary Fig. 2 and Supplementary Table 5). We transduced each reporter cell line with the lentiviral library, isolated GFP-positive cells and sequenced the enriched tRNA variants. Unlike the lentiviral screening with exogenous promoters, this screen successfully identified at least one variant of all TAG sup-tRNAs that had been identified in the prime editing screen. The most effective sup-tRNAs performed best when paired with their own endogenous leader sequence (Extended Data Fig. 4f). Thus leader sequences have a critical regulatory role in tRNA activity.

The screen results also indicated that a terminator is essential for sup-tRNA activity, with a directly adjacent 5T terminator outperforming most endogenous terminator sequences (Extended Data Fig. 4g). We hypothesize that the absence of a terminator may lead to run-off of Pol III, or weakened expression due to a lack of the traditional Pol III recycling mechanism in which Pol III is rapidly reused for a new round of transcription after completing a transcript³⁶. Notably, only 93% and 48% of natural downstream sequences have a 4T or 5T termination sequence, respectively, within 100 bp of their mature tRNA sequence³⁷ (Supplementary Fig. 8a,b).

We next hypothesized that introducing mutations within the tRNA itself could further enhance sup-tRNA activity. We also sought to identify silent mutations that could improve prime editing outcomes by evading cellular mismatch repair (MMR)³⁸ and by preventing rebinding of the prime editor to the target tRNA locus following repair. We selected three candidate sup-tRNAs (tRNA-Arg-CCT-4-1, tRNA-Tyr-GTA-2-1 and tRNA-Leu-TAA-4-1) and a corresponding mouse sup-tRNA (tRNA-Leu-TAA-2-1) and generated lentiviral libraries that include every possible 1-bp deletion, single-nucleotide substitution variant (SNV), and simultaneous base pair substitution in each tRNA's paired bases (Supplementary Fig. 2 and Supplementary Table 6).

We transduced the reporter cell line with these lentiviral libraries and compared their performance to that of the corresponding anticodon-only (ac-only) sup-tRNA sequences. Most mutations reduced sup-tRNA performance (Extended Data Fig. 4h). Deletions were poorly tolerated across sup-tRNAs (Extended Data Fig. 5a). By contrast, for each sup-tRNA, a subset of SNVs and paired-base substitutions improved sup-tRNA performance (Extended Data Fig. 4h). We also found that beneficial mutations in tRNA-Leu-TAA-4-1 could apply to the related family members tRNA-Leu-TAA-1-1, tRNA-Leu-TAA-2-1, and tRNA-Leu-TAA-3-1 (Extended Data Fig. 5b, Supplementary Fig. 9 and Supplementary Table 7).

Of note, we observed that sup-tRNA mutations can be additively combined, supporting the production of up to fivefold higher levels of full-length GFP protein than the ac-only sup-tRNA, yielding up to 35% of the amount of GFP as control cells expressing an uninterrupted GFP gene (Fig. 2d, Extended Data Fig. 5d and Supplementary Table 8). Engineered sup-tRNAs developed in this study substantially outperform previously engineered sup-tRNAs, including a recently reported variant that yielded 16% GFP protein yield (Fig. 2d) but would require changing 83 bp of a native tRNA sequence for genomic installation³². By contrast, the best-performing variants we engineered require the substitution of only 6 bp from the wild-type tRNA-Leu-TAA-1-1 gene sequence.

To evaluate the cellular abundance of prime editing-installed sup-tRNAs compared with a wild-type endogenous tRNA, we performed targeted tRNA sequencing for tRNA-Leu-TAA-1-1 on a polyclonal edited population. By comparing the installation efficiency of each sup-tRNA at the DNA level with the frequency of corresponding tRNA-sequencing reads at the RNA level, we inferred the relative abundance of prime editing-installed sup-tRNAs versus the wild-type tRNA at the same genomic locus. We observed that the ac-only sup-tRNA and the engineered sup-tRNA are 5.6-fold and 2.5-fold less abundant, respectively, than the endogenous tRNA-Leu-TAA-1-1 that they replace (Extended Data Fig. 6a). Introducing mutations in the anticodon stem thus increased the abundance of the engineered sup-tRNA compared with the ac-only sup-tRNA (Extended Data Fig. 6a). Both sup-tRNAs were modified at positions 26 and 37, as expected³⁹ (Extended Data Fig. 6b–e). These findings suggest that mutations in the anticodon stem can increase engineered sup-tRNA performance in part by increasing its cellular abundance.

Across all conditions tested, PE6c yielded the highest mean editing efficiency (Fig. 3b). The mutation combination of hp12ta>cg+hp14gc>ta+mut38a>t yielded the highest mean editing of 24% (Extended Data Fig. 7a) and the inclusion of MLH1dn resulted in greater mean editing efficiencies for most tested variants, suggesting that MMR evasion strategies can enhance editing efficiency for these edits (Fig. 3b and Extended Data Fig. 7c). We observed a high correlation (R = 0.78) between the results from the HeLa and HEK293T datasets (Extended Data Fig. 7d and Supplementary Discussion).

To validate that editing of the lentivirally integrated target site used in the screen reflects editing of the endogenous tRNA-Leu-TAA-1-1 locus, we performed arrayed validation on 122 high-performing and low-performing epegRNAs in HEK293T and HeLa cells. Editing efficiency at the endogenous locus and at the lentiviral target site in both cell types were correlated (Fig. 3d,e, R = 0.80 in HEK293T cells and R = 0.70 in HeLa cells), indicating that the lentiviral target-matched screen reflects prime editing outcomes at the endogenous locus.

On the basis of the optimized editing results and the readthrough efficacy data, we selected two variant combinations to advance: hp13gc>ta+hp12ta>cg and hp13gc>cg+hp12ta>cg. To finalize our editing strategy, we tested whether a PE3 or PE3b strategy—both of which use a nicking guide RNA (ngRNA) to nick the unedited strand of the heteroduplex editing intermediate in order to favour the use of the edited strand as a template for DNA repair—could further improve editing outcomes³. Top-performing PE3 strategies yielded mean editing efficiencies of ~60–80% in HEK293T cells, substantially outperforming earlier designs that yielded <8% editing and completing the optimization of the proof-of-concept editing strategy for PERT (Supplementary Fig. 8c and Extended Data Fig. 7e).

We assessed off-target editing from PERT using two independent methods. First, we identified all sites in the human genome with up to four mismatches from the protospacer of the best-performing engineered sup-tRNA (Extended Data Fig. 8a). We sequenced at depth each of these sites following prime editing to install an engineered sup-tRNA (tRNA-Leu-TAA-1-1+hp12ta>cg+hp13gc>cg) and no significant off-target editing was observed above that of an untreated control (Extended Data Fig. 8b).

Second, as a genome-wide unbiased method for detecting potential off-target edits, we performed a screen in mammalian cells with a large pool of lentiviral integrated putative off-target sites. We computationally identified off-target sites in the genome with up to 6 mismatches from the optimized epegRNA's protospacer or with up to 4 mismatches and up to two bulges, identifying 17,206 candidates (Supplementary Table 11). This coverage is identical to the computational filtering step of circularization for in vitro reporting of cleavage effects by sequencing (CIRCLE-seq)⁴¹, a commonly used genome-wide off-target detection method for RNA-guided nucleases. Following lentiviral integration of the pool of off-target candidate sites and transfection to install an engineered sup-tRNA (tRNA-Leu-TAA-1-1+hp12ta>cg+hp13gc>cg) with prime editing, we sequenced the lentiviral cassette and analysed editing at each off-target site.

This analysis nominated 16 candidate off-target sites with potential prime editing events occurring above the level of the untreated control sample in prime-edited cells (Extended Data Fig. 8c). To further investigate these potential off-target sites, we individually amplified the genomic region corresponding to each of the 16 nominated off-target sites in cells transfected with prime editing reagents to install an engineered sup-tRNA (tRNA-Leu-TAA-1-1+hp12ta>cg+hp13gc>cg). We observed no off-target prime editing or indel formation above background for all amplified genomic loci (Extended Data Fig. 8d). These data are consistent with previous reports of the high sequence specificity of prime editing, which probably arises from its mechanism requiring multiple independent DNA or RNA hybridization events in order to make a productive edit^3,42,43.

We evaluated the potential for several types of abnormalities in human cells following conversion of an endogenous tRNA into a sup-tRNA. To detect transcriptome perturbations, we performed RNA-seq after converting the endogenous tRNA-Leu-TAA-1-1 gene into an engineered sup-tRNA (Leu-TAA-1-1+hp12ta>cg+hp13gc>cg) with prime editing. We observed no differentially expressed genes (FDR < 0.05 and |log₂(fold change)| > 1), suggesting no evident stress response from converting the tRNA-Leu-TAA-1-1 tRNA gene into a sup-tRNA gene in a human cell line (Extended Data Fig. 8e).

Next, we evaluated the effect of repurposing the tRNA-Leu-TAA-1-1 tRNA gene on global tRNA homeostasis by performing quantitative PCR to measure the pooled expression of 28 different tRNA gene families following prime editing-mediated installation of an ac-only or engineered sup-tRNA (Leu-TAA-1-1+hp12ta>cg+hp13gc>cg). We observed no significant difference (adjusted P value > 0.05) in the expression of any of these 28 tRNA gene families in either treatment condition relative to wild-type cells (Extended Data Fig. 8f), suggesting that the tRNA-Leu-TAA-1-1 tRNA gene can be repurposed without significant perturbation to general tRNA homeostasis.

To evaluate potential readthrough of NTCs, we performed mass spectrometry to detect peptides that are translated into the 3′ UTR for protein-coding genes terminating with a TAG stop codon. After confirming that Leu-TAA sup-tRNAs retain 100% fidelity of leucine charging (Extended Data Fig. 9a), we computationally predicted all peptides that could be generated by reading through the 4,036 natural TAG termination codons and extending translation to the second in-frame stop codon. We then examined whether these predicted peptides were present by performing label-free quantitative mass spectrometry on trypsin-digested total protein lysates from both control and prime-edited HEK293T cells expressing and mCherry-STOP-GFP reporter and either the ac-only sup-tRNA or the engineered sup-tRNA (Leu-TAA-1-1+hp12ta>cg+hp13gc>cg).

To test with enhanced sensitivity the ability of a prime editing-installed sup-tRNA to induce NTC readthrough, we identified the 69 most abundant proteins that use a TAG stop codon from the above dataset and performed target-ion mass spectrometry to detect each of 111 predicted peptides that would be generated from reading past the NTC for these highly expressed proteins. We observed a single peptide corresponding to potential readthrough of an NTC (encoding 18 amino acids after the end of the tyrosyl-tRNA synthetase protein), but the abundance of this peptide was not significantly different (adjusted P value > 0.05) between untreated cells and cells that were prime edited to express a sup-tRNA (Extended Data Fig. 9f,g and Supplementary Table 14). Thus both proteome-wide and target-ion protein mass spectroscopy indicate that prime editing-installed sup-tRNAs do not broadly perturb the proteome or induce detected NTC readthrough under the conditions tested, probably owing to cellular mechanisms that safeguard against NTC readthrough.

Several factors influence readthrough efficiency at a given PTC with a sup-tRNA, including the expression level of the PTC-containing transcript and the nucleotide context surrounding the PTC⁴⁴. To assess the potential of PERT to rescue the effects of genomic PTCs with disease relevance, we evaluated readthrough across diverse, endogenously encoded PTCs. We generated models in HEK293T cells of known human pathogenic PTCs corresponding to a TAG stop codon in TPP1, HEXA and NPC1 genes that cause Batten disease, Tay–Sachs disease and Niemann–Pick disease type C1, respectively. We chose six mutations that either originally encoded a leucine amino acid or are known to tolerate a leucine at the indicated position (TPP1 p.L211X, TPP1 p.L527X, HEXA p.L273X, HEXA p.L274X, NPC1 p.Q421X and NPC1 p.Y423X)⁴⁵. As with many loss-of-function genetic diseases, previous studies suggest that even low levels of protein activity for these disorders can provide therapeutic benefits^46–48.

We transfected these six cell models with the same optimized PE3 editing components to convert the endogenous genomic tRNA-Leu-TAA-1-1 locus into an ac-only sup-tRNA or engineered sup-tRNAs (tRNA-Leu-TAA-1-1+hp13gc>ta+hp12ta>cg or tRNA-Leu-TAA-1-1+hp13gc>cg+hp12ta>cg). At 3 d after transfection, mean editing efficiencies ranged from 67–80% for installation of the ac-only sup-tRNA, 56–84% for tRNA-Leu-TAA-1-1+hp13gc>ta+hp12ta>cg and 64–83% for tRNA-Leu-TAA-1-1+hp13gc>cg+hp12ta>cg (Fig. 4c, Supplementary Fig. 10a and Supplementary Fig. 12). Engineered sup-tRNA installation led to substantial (17–70%) restoration of enzyme activity in the four TPP1 and HEXA pathogenic PTC models, and partial restoration of full-length protein expression in both models of PTCs in NPC1 (which encodes a cholesterol transporter rather than an enzyme) using the same prime editing reagents, establishing the disease-agnostic therapeutic potential of this approach (Fig. 4b–d, Extended Data Fig. 10b–d and Supplementary Fig. 13). In each case, treatment resulted in full-length protein expression that meets or exceeds the documented therapeutic threshold for each disease target^46–48. Together, these results establish that a single prime editing agent to convert an endogenous genomic tRNA gene into a sup-tRNA can rescue a substantial fraction of enzyme activity and full-length protein production lost to nonsense mutations.

To evaluate the performance of PERT at a broad spectrum of nonsense mutations, we cloned 14,746 pathogenic PTCs in the ClinVar database flanked on either side by the 18 nucleotides present in the native mRNA sequence into a nonsense-mediated decay-sensitive reporter construct (Supplementary Fig. 11a,b, Supplementary Fig. 14 and Supplementary Table 15). Transcripts with PTCs are often destroyed by nonsense-mediated decay during the pioneer round of translation⁴⁹; however, if the PTC is read through, the transcript will be stabilized for the remainder of its lifetime. This phenomenon allows quantification of the ability of a sup-tRNA to read through a given PTC by comparing the abundance of mRNA transcripts for each PTC-containing sequence (Methods). We transduced this library of PTC-containing sequences into a clonal ac-only sup-tRNA cell line and observed an average readthrough score of 69 ± 30% across 14,746 pathogenic PTCs, indicating stabilization of the mRNA transcript and readthrough of the PTC for most sequences. We independently confirmed readthrough of full-length CFTR cDNAs containing 15 distinct pathogenic PTC mutations that cause cystic fibrosis using a prime editing-installed ac-only sup-tRNA. Protein yields across the variants were moderately correlated (R = 0.49) with the readthrough scores measured in the pooled ClinVar screen (Supplementary Fig. 11c,d and Supplementary Table 16), with a maximum observed protein yield of 14% relative to wild-type CFTR. Together, these results demonstrate that PERT can effectively rescue diverse pathogenic genomic PTCs and can be used to generate therapeutically relevant levels of full-length protein and active enzymes in human cells.

At 3 weeks after injection, we collected the brain tissue for analysis of editing and readthrough efficiency. We observed an average editing efficiency of 11% for the TAG sup-tRNA and 23% for the TGA sup-tRNA in the bulk brain cortex of treated mice (Extended Data Fig. 10a). To measure readthrough efficacy, we dissociated the cells from whole hemisphere halves of treated mice and measured GFP expression by flow cytometry. We observed an average relative protein yield of 24% of the co-delivered TAG GFP reporter and 26% of the co-delivered TGA GFP reporter (Fig. 5b). Therefore, although editing efficiency would benefit from further optimization in mice, cells that receive prime editing agents can achieve readthrough levels that could ameliorate many diseases.

To assess the tolerability of PERT in vivo, we performed whole-proteome mass spectrometry on tissues that had been prime edited to express an ac-only sup-tRNA. Similar to our observations in human cell lines (Fig. 4a,b), we did not detect any peptides corresponding to readthrough of natural stop codons in tissues from mice treated with PERT (Supplementary Table 17). In addition, we monitored a separate cohort of mice over the course of 15 weeks and observed no significant difference in body weight in the PERT-treated mice (n = 7) relative to untreated controls (n = 9) (Extended Data Fig. 10b,c).

Finally, we investigated the in vivo therapeutic application of PERT in a mouse model of the human lysosomal storage disease mucopolysaccharidosis type I (MPS-I). MPS-1 is caused by a lack of α-L-iduronidase, leading to the pathogenic accumulation of glycosaminoglycans (GAGs). This model harbours a TAG stop codon, Idua p.W392X, which is orthologous to the most common pathogenic human mutation, IDUA p.W402X, and has been used previously to evaluate PTC readthrough in vivo^5,51. For this experiment we further engineered the mouse sup-tRNA using mutations identified via saturation mutagenesis screening of mouse tRNA-Leu-TAA-2-1, generating an engineered sup-tRNA (mouse tRNA-Leu-TAA-2-1+hp13gc>ta+mut38a>t) (Supplementary Fig. 15 and Supplementary Discussion).

We packaged prime editing agents to install this sup-tRNA into the v3em dual-AAV9 architecture⁵⁰ and treated homozygous and heterozygous P0 neonatal Idua^W392X littermates via intracerebroventricular injection (Fig. 5c). At 7 weeks after treatment, we evaluated prime editing efficiencies across tissues and observed mean desired prime editing efficiencies of 6.3%, 6.1%, 6.8% and 0.05% in the cortex, heart, liver and kidney, respectively (Fig. 5d), consistent with the known tissue tropism of AAV9 (ref. ⁵²). Editing led to marked restoration of IDUA enzymatic activity compared with untreated controls, with mean enzymatic activity relative to wild-type mice of 7.6%, 6.3%, 1.7% and 0.31% in the heart, cortex, liver and kidney, respectively (Fig. 5d and Extended Data Fig. 10d). These data demonstrate the ability of in vivo PERT to restore enzyme function that has been abrogated by pathogenic endogenous PTCs.

To assess whether restoration in enzymatic activity was sufficient to rescue disease phenotypes, we performed histological analysis of brain, liver and spleen tissue from homozygous Idua^W392X mice. We scored phenotypic severity on the basis of MPS-I related pathology, including hepatic and splenic foam cell accumulation, heart vessel-associated foam cell accumulation, cerebellar Purkinje cell vacuolization, splenic stromal vacuolization and pathogenic GAG accumulation.

MPS-I GAG storage pathology was observed across all tissues analysed from untreated Idua^W392X homozygous mice, but was minimal or absent in treated counterparts. Purkinje cells of the cerebellum and thalamic neurons displayed moderate vacuolization in untreated homozygous Idua^W392X mice and was resolved following PERT treatment with absent or minimal vacuoles (Fig. 5e,f and Supplementary Note). We observed a similar pattern of attenuated vacuolization in the spleen, liver and heart in treated cohorts (Fig. 5e,f). Pathogenic accumulation of foam cells observed in the liver, heart and spleen in untreated mice was absent in the treated cohort (Fig. 5e,f). Finally, the liver of untreated mice displayed mild Alcian blue staining that was minimal in the treated mice, indicating depletion of GAG accumulation (Fig. 5f). These data are consistent with previous reports that restoration of only 1% or more IDUA enzyme activity is sufficient to rescue most or all observed MPS-1 disease phenotypes⁵. Together, these results illustrate the in vivo applicability of PERT for PTC readthrough and rescue of disease phenotypes in a clinically relevant mouse model.

Clinical genome-editing technologies, including nucleases, base editors and prime editors, have advanced the treatment of many genetic disorders. These therapeutic approaches have predominantly been disease-specific, with most efforts requiring the design of distinct gene-correction strategies for each pathogenic mutation. Given that more than 200,000 distinct mutations collectively affect hundreds of millions of individuals, developing, manufacturing and meeting regulatory requirements for thousands of individual therapies needed to treat even a modest fraction of patients with genetic disease is a monumental challenge, creating an urgent need for allele- and disease-agnostic therapeutic approaches that can address large classes of mutations with a single therapeutic composition.

As one potential disease-agnostic strategy, we developed PERT as a prime editing-mediated approach to convert redundant endogenous tRNAs into optimized sup-tRNAs. This strategy combines the durable, one-time treatment potential of genome editing with the broad applicability of sup-tRNAs. Through high-throughput evaluation of thousands of variants spanning all 418 high-confidence tRNA genes, we identified sup-tRNAs that are capable of reading through 32% of all possible nonsense mutations with precise amino acid correction (installing the wild-type amino acid in response to the PTC), including Arg and Leu sup-tRNAs that can read through TGA stop codons and Leu, Ser and Tyr sup-tRNAs that can read through TAG stop codons. Whereas precise amino acid correction is the most conservative strategy, many nonsense mutations may still be rescued with non-wild-type amino acid substitutions, although characterization of the missense protein variant in these cases is necessary⁵³.

Among the sup-tRNAs tested, tRNA-Leu-TAA-1-1 exhibited the highest baseline readthrough efficiency. By swapping the anticodon of the endogenous tRNA-Leu-TAA-1-1 gene to CTA and introducing prime editing-accessible mutations in the anticodon stem, we engineered sup-tRNAs variants capable of restoring more than 35% of wild-type protein activity and/or expression from a gene containing a TAG nonsense mutation, surpassing the therapeutic threshold known for many loss-of-function genetic diseases^46–48. In mouse models, including those for Hurler syndrome, PERT treatment demonstrated substantial protein restoration and near-complete rescue of disease pathology.

During the course of this study, we did not observe any evident toxicity from PERT in any cultured cells or mice, changes in the transcriptome of cells, including the abundance of other tRNAs, readthrough of natural stop codons in human cells or in tissues from PERT-treated mice using proteome-wide mass spectrometry, or any off-target editing using two genome-wide detection methods. Since the stoichiometry of cellular tRNAs can influence global translation outcomes^22–24, we hypothesize that conversion of a single endogenous genomic tRNA gene into a sup-tRNA may be less perturbative to global translation outcomes than previous approaches to introduce sup-tRNAs by exogenous overexpression or delivery^4,5.

Not all endogenous tRNA genes could be converted into effective sup-tRNAs with a simple anticodon change. In some cases, only specific isodecoders within a family were functional sup-tRNAs (for example, Tyr-GTA but not Tyr-ATA). Some sup-tRNAs favoured TAG stop codon readthrough (for example, Tyr-GTA), whereas others supported readthrough of only TGA stop codons (for example, Arg-TCG). We were unable to identify an effective prime editing-installed sup-tRNA for TAA stop codons with low-copy screening, although future efforts could explore reporter sensitivity adjustments or additional tRNA mutations in order to achieve this goal. Although certain patterns emerge in these data—such as the identification of Ser and Leu sup-tRNA backbones, which do not rely on the anticodon loop for recognition by their aminoacyl-tRNA synthetases^54,55—it remains unclear why a functional Ala sup-tRNA, which also is thought to lack synthetase recognition of the anticodon, was not identified. Our data suggest that the determinants of sup-tRNA effectiveness, despite extensive study, remain incompletely understood.

Efficiency of sup-tRNA readthrough also depends on the sequence context of the nonsense mutation. Across the sequence contexts of 14,746 pathogenic TAG stop codons in ClinVar, more than 70% showed readthrough from PERT. However, additional factors outside of the immediate sequence context are also likely to influence sup-tRNA activity, including ribosome velocity and PTC position in the mRNA. Further improvements may come from mutations beyond the anticodon stem, such as alterations in the D- or T-loops, which may enhance aminoacylation, transcription or ribosome accommodation of sup-tRNAs. Alternative methods for integrating a sup-tRNA sequence into the genome, such as twin prime editing (twinPE)³⁴, eePASSIGE⁵⁶, CASTs^57,58 or methods based on homology-directed repair⁵⁸, may facilitate the incorporation of more advanced combinations of mutations. Combining PERT with readthrough-promoting small molecules may further boost functional protein restoration. We also anticipate that, although our primary goal was to optimize a sup-tRNA specifically for one-time genomic installation, the best-performing variants we identified could improve readthrough efficiency when delivered through tRNA-supplementation approaches such as AAV- or LNP-mediated delivery of sup-tRNAs.

Additional studies are needed to investigate and optimize the safety, efficacy and ability of the strategy developed in this study to be delivered to relevant tissues through viral and non-viral in vivo delivery methods. Future work could explore tissue-targeted capsid engineering, receptor-specific LNP-mediated or engineered virus-like particle-mediated^59,60 delivery, and regulated or transient expression systems that avoid unnecessary long-term exposure to editing agents. Nevertheless, the findings of this initial study highlight the potential of PERT as a disease-agnostic therapeutic strategy in which a single composition of matter may offer a one-time treatment that benefits many different cohorts of patients with diverse genetic diseases. For example, approximately 8,000 people with cystic fibrosis⁶¹, 252,000 people with Stargardt disease⁶², 31,000 people with phenylketonuria⁶³ and 43,500 people with Duchenne muscular dystrophy⁶⁴ have nonsense mutations in CFTR, ABCA4, PAH or DMD, respectively, that could in principle be treated with common PERT prime editing agents after further optimization.

Although direct DNA-level correction of pathogenic mutations remains the most straightforward approach to precision genetic medicine, the large effort currently required to develop, meet regulatory requirements and manufacture many gene editing agents needed to correct many mutations underscores the importance of complementary allele-agnostic and disease-agnostic approaches such as PERT to maximize the ability of patients to benefit from advances in gene editing. By enabling a single treatment to address a broad cross-section of nonsense mutation-related diseases, PERT represents an advance toward making therapeutic genome editing accessible to a much larger patient population. We hope that PERT and the flexibility of prime editing will stimulate additional allele-agnostic and mutation-agnostic therapeutic gene editing strategies that may provide broadly applicable treatments for diverse serious disorders.

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-025-09732-2.

Supplementary Information This file contains Supplementary Discussion including additional references, Supplementary Note, Supplementary Tables and Supplementary Figs. 1–15. Reporting Summary Supplementary Table 1 Oligonucleotides, epegRNA sequences, ngRNA sequences, and plasmid sequences. Supplementary Table 2 Design and results for PE2 epegRNA screens to convert endogenous tRNAs into sup-tRNAs with TAG, TGA, and TAA reporters. Supplementary Table 3 Design and results for initial lentiviral sup-tRNA screen. Supplementary Table 4 Design and results for sup-tRNA leader sequence screen. Supplementary Table 5 Design and results for follow-up sup-tRNA leader and terminator screen. Supplementary Table 6 Design and results for saturation mutagenesis screening of individual sup-tRNAs. Supplementary Table 7 Design and results for saturation mutagenesis screening of the Leu-TAA family of sup-tRNAs. Supplementary Table 8 Mutation combinations, tRNA sequences, and and statistics related to Figure 3d. Supplementary Table 9 Design and results for epegRNA screening of Leu-TAA-1-1 into a sup-tRNA with an adjacent synthetic target site in HeLa cells. Supplementary Table 10 Design and results for epegRNA screening of Leu-TAA-1-1 into a sup-tRNA with an adjacent synthetic target site in HEK293T cells. Supplementary Table 11 Design and results for lentiviral off-target screening of the epegRNA. Supplementary Table 12 Evaluation of first and second stop codons for each protein in the human proteome. Supplementary Table 13 Whole proteome mass spectrometry on human cells treated with PERT. Supplementary Table 14 Target-ion mass spectrometry on potential 3′UTR peptides for the top 69 proteins with a TAG stop codon in cells treated with PERT. Supplementary Table 15 Design and results for sequence contexts amenable to readthrough with a CUA sup-tRNA. Supplementary Table 16 CFTR cDNA constructs used in validation of sequence context screening results. Supplementary Table 17 Whole proteome mass spectrometry on mouse livers and cortices treated with PERT.

Source Data Fig. 5