RNA-LNP-mediated in vivo prime editing corrects disease phenotypes in a mouse model of citrullinemia type I

Abstract

Citrullinemia type I (CTLN1) is a severe urea cycle disorder caused by pathogenic variants in the ASS1 (argininosuccinate synthetase 1) gene, for which liver transplantation remains the only curative option. Here, we used prime editing to correct the Ass1^fold mouse model of CTLN1. Adeno-associated virus (AAV)-mediated delivery of the PE7 prime editor with an optimized pegRNA achieved 71% and 54% correction of the pathogenic Ass1 mutation in hepatocytes of neonates and juveniles, respectively. Delivery of mRNA-encoded PE7 and synthetic pegRNA via lipid nanoparticles (LNPs) resulted in 24% correction after a single 3 mg kg⁻¹ dose in neonates and 13% after three 4 mg kg⁻¹ doses in juveniles. All treated groups showed full normalization of survival and of blood citrulline and ammonia concentrations, with restored urea cycle function and correction of natural behavior defects. Consistent with these findings, immunostaining demonstrated restoration of wild-type-like ASS1 protein localization in functionally relevant periportal and intermediate-zone hepatocytes. Editing was confined to the liver, with minimal indel formation and off-target activity and only transient elevations in liver enzymes. In a cellular reporter system, 6 of 15 recurrent human pathogenic ASS1 mutations studied, including the most common ASS1^G390R variant, were corrected with similar or higher efficiencies than Ass1^fold. These findings highlight prime editing as a precise and potentially curative treatment strategy for individuals with CTLN1 and potentially other genetic liver diseases.

Introduction

The urea cycle is a vital metabolic pathway in the liver that detoxifies excess nitrogen by converting ammonia into urea for urinary excretion. Disruption of this pathway causes urea cycle disorders (UCDs), a group of inborn errors of metabolism characterized by life-threatening hyperammonemia (1, 2). Citrullinemia type I (CTLN1, OMIM #215700) is one of the most severe UCDs and results from autosomal recessive mutations in the ASS1 gene (OMIM *603470), which encodes argininosuccinate synthetase, an enzyme required for the third step of the urea cycle (3, 4). Affected individuals accumulate toxic levels of plasma ammonia, particularly during catabolic stress, leading to cerebral edema, irreversible neurological injury, and, if untreated, death. The clinical spectrum of CTLN1 ranges from a rapidly progressive neonatal form to milder, late-onset presentations, but all forms require lifelong metabolic management (5).

Current therapies are limited to symptomatic control and include protein-restricted diets and nitrogen-scavenging agents (6). These approaches partially lower blood ammonia but fail to lower citrulline levels, restore metabolic homeostasis, or prevent disease progression (4). Orthotopic liver transplantation is the only curative treatment, but is constrained by donor organ scarcity, surgical risk, and the lifelong need for immunosuppression. In recent years, several gene complementation strategies have been developed to correct UCD mouse models, including adeno-associated virus (AAV)-mediated delivery for correcting CTLN1 (7–9) and mRNA-lipid nanoparticle (LNP)-mediated delivery for correcting ornithine-transcarbamylase deficiency or arginase deficiency (10–12). Although both approaches improved disease-related phenotypes and biochemical markers, they face substantial limitations for clinical translation. AAV vectors trigger adaptive immune responses that prevent repeated dosing, and because they remain episomal, transgene expression diminishes over time, particularly in pediatric patients because liver growth dilutes vector genomes (13, 14). In contrast, mRNA-LNPs can be re-administered. However, the transient expression of mRNA necessitates frequent dosing - every few days to weeks - raising concerns about the long-term feasibility of such treatments due to high costs and potential adverse effects associated with frequently repeated intravenous mRNA-LNP administration (15).

Precise correction of pathogenic ASS1 mutations at the endogenous locus using gene editing offers a more durable treatment option. However, traditional Cas9 nuclease-based editing relies on homology-directed repair (HDR), a pathway largely inactive in non-dividing hepatocytes (16). Moreover, double-strand breaks (DSBs) introduced by Cas9 can generate disruptive insertion/deletion (indel) mutations, potentially converting hypomorphic alleles into null variants, or may also cause large deletions or chromosomal rearrangements, raising safety concerns for therapeutic use (17).

Recently developed DSB-free gene editing technologies have broadened the scope of in vivo genome engineering. Base editors (BEs), in which catalytically impaired Cas9 is fused to a deaminase enzyme, enable efficient conversion of single nucleotides in non-dividing cells, but are limited to transition mutations and prone to bystander editing if other adenines or cytosines are near the target base (18, 19). Prime editors (PEs) use a Cas9 nickase fused to a reverse transcriptase (RT) and are guided by a prime editing guide RNA (pegRNA) that encodes the template for the intended DNA modification. PEs can mediate substitutions, insertions, and deletions, offering a powerful platform to correct a wide range of pathogenic mutations (20).

Here, we demonstrated in vivo correction of the Ass1^fold mouse model for CTLN1 via prime editing. PE delivery via AAV or RNA-LNP corrected early lethality, restored urea cycle function, and achieved long-term metabolic normalization in mice. These findings demonstrate the potential of prime editing as a precise and durable therapy for CTLN1.

Results

Correction of the Ass1^fold mutation via prime editing in vitro

To evaluate the therapeutic potential of prime editing for CTLN1, we used the Ass1^fold mouse model, which harbors the clinically relevant p.T389I (c.1166C>T) missense mutation in the Ass1 gene (Fig. 1A). This variant has previously been reported in patients (21, 22), and recapitulates key disease hallmarks in mice, including severe hyperammonemia, elevated plasma citrulline, postnatal growth restriction, and early lethality (23).

Figure 1. Optimization of pegRNAs to correct the T389I Ass1^fold mutation.

(A) Schematic of the murine Ass1 gene and the location of the Ass1^fold mutation in exon 15, where a single C to T base change causes a threonine (Thr) to isoleucine (Ile) amino acid change at position 389. The mutation is targetable by prime editing with two nearby protospacers. Yellow boxes illustrate the 5’ and 3’ UTRs and blue ones the coding exons. (B) Schematic showing the process of the prediction and in vitro testing of pegRNAs. To optimize prime editing efficiencies, silent bystanders were generated ±2 codons of the targeted base and PRIDICT2.0 was used to design optimal pegRNAs, which were then tested in vitro. (C) Prime editing rates (upper panel), indels (middle panel) and PRIDICT2.0 MMR+ (K562) scores (lower panel) of the top 15 pegRNA designs targeting protospacer 2. The best performing pegRNA is highlighted in purple. Data show means ± SD of n=3 replicates. (D) Schematic of the best pegRNA identified, showing the primer binding site (PBS), RT template (RTT), protospacer adjacent motif (PAM), nick site and sequence of the spacer. The pathogenic Ass1^fold mutation is marked in red. Panels A, B and D were created using BioRender.

Although the C>T substitution theoretically permits correction by adenine base editing, the locus does not contain an NG or NGG protospacer adjacent motif (PAM) site to position the target base within the editing window. This constraint led us to pursue prime editing, where two NGG PAM sequences flanking the Ass1^fold mutation (protospacers 1 and 2 - Fig. 1A) allow optimal pegRNA positioning. To identify pegRNA designs that efficiently correct the pathogenic C>T mutation, we used PRIDICT2.0, a machine learning model that is trained on an editing dataset of over 400,000 pegRNAs and allows screening of different pegRNA designs in silico (pegRNAs targeting different spacers with varying lengths of primer binding sites and templates for the RT (24, 25); Fig. 1B,C; Fig. S1A). However, for both protospacers only pegRNA designs with relatively low editing efficiencies (in mismatch repair (MMR) proficient cells) were predicted (max. PRIDICT2.0 MMR+ (K562) score 28.3, Fig. 1C). We therefore extended our design strategy to include pegRNAs that co-introduce silent bystander edits. These multi-base edits are less prone to mismatch repair and are generally more efficient in MMR proficient cells such as hepatocytes (24, 25). Moreover, if silent bystanders are designed to disrupt the PAM, they can further increase precision by preventing re-cutting of the locus after the edit is installed (25, 26). Therefore, we systematically generated all possible combinations of silent, non-coding mutations within ±2 codons of the targeted mutation and used PRIDICT2.0 to design efficient pegRNAs for all edits (Fig. 1B). Using this strategy, we identified pegRNAs with up to two-fold higher predicted editing scores than the best-performing single base replacement pegRNA (max. PRIDICT2.0 MMR+ score 60.3 vs. 28.3), with pegRNAs targeting protospacer 2 consistently showing higher prediction scores (Fig. 1C, Fig. S1A).

We then experimentally validated the 15 pegRNA designs with the highest PRIDICT2.0 MMR+ scores in MMR proficient K562 reporter cells harboring the pathogenic Ass1^fold locus (100 bases up-and downstream of the targeted mutation), including the best pegRNA design for the single T>C correction. Although this pegRNA only yielded modest editing efficiencies of 3.1%, the best-performing multi-base replacement pegRNA achieved 19.7% editing when tested in combination with the conventional PEmax (27) editor (Fig. 1C). Editing was also highly specific, with no detectable indels above baseline (Fig. 1C). Beyond optimization of the pegRNA design, several additional strategies have been developed in recent years to enhance prime editing efficiency, including PE3, which is based on PEmax but co-introduces a second nick on the non-edited DNA strand (20), template-jumping prime editing, which mimics the mechanism of retrotransposon reverse transcription (28), and twin prime editing, in which two pegRNAs simultaneously target opposite DNA strands (29). However, when we applied these approaches to the Ass1^fold locus, correction efficiencies were only improved for low efficiency pegRNAs, but not for the most active pegRNA6 (Fig. S1, B and C). Based on these data, we selected pegRNA6, which targets protospacer 2 and co-introduces two silent mutations that disrupt the PAM, for in vivo experiments (Fig. 1D).

Correcting Ass1^fold mice via AAV-mediated prime editing

In a previous study, we demonstrated that the PE7 system - with the La small RNA binding protein fused to the RT domain of PEmax (30) - considerably enhances prime editing efficiencies in the liver (31). To assess whether this also applies for the Ass1^fold locus, we compared the efficiency of PE7 to conventional PEmax in heterozygous Ass1^fold mice. We generated AAV vectors encoding pegRNA6 and intein-split PEmax or PE7 (Fig. 2A). To ensure that both constructs remained within the AAV packaging limit (~4.5 kb), Cas9 was split at amino acid position 712-713 (31) for PEmax and 1068-1069 for PE7, and the RNase-H-domain was removed from the M-MLV-RT. Constructs were packaged into AAV9 capsids and administered to neonate mice at a dose of 2.5 × 10¹³ vg/kg. In line with previous in vivo prime editing results on the phenylalanine hydroxylase (Pah) and DNA methyltransferase 1 (Dnmt1) loci (31), as well as in vitro results in Ass1^fold reporter cells (Fig. S2A), PE7 also yielded higher editing efficiencies in heterozygous Ass1^fold mice (32 ± 2% vs. 26 ± 2%; Fig. S2A). We therefore proceeded with PE7 for subsequent experiments and treated homozygous Ass1^fold mice by systemically administering the PE7 AAV either to neonates (postnatal day 1) or to juveniles (4 weeks old) at a dose of 2.5 × 10¹³ vg/kg (Fig. 2B). Both treatment regimens fully rescued the early lethality phenotype observed in saline- or mock AAV-injected animals, which showed a median survival of only 12 weeks (Fig. 2C, Fig. S2C). In addition, in both treatment groups, body weights normalized to those comparable to wild-type littermates (Fig. 2D, Fig. S2, D-G).

Figure 2. AAV-mediated prime editing in the livers of Ass1^fold mice normalizes survival and body weights.

(A) Schematic illustration of the PE7 AAV constructs selected for in vivo experiments. Indicated AAV genome lengths include inverted terminal repeats (ITRs). nSpCas9, nickase Streptococcus pyogenes Cas9; epegRNA, trimmed engineered pegRNA; M-MLV-RT^dRnH, M-MLV RT-delta RNase H; Syn. pA, Synthetic short poly-adenylation signal; NLS, nuclear localization signal; La, La RNA binding protein; P3, P3 promoter; hU6, human U6 promoter. (B) Schematic illustration of the experimental setup. Mice were treated with PE7 AAVs as neonates (postnatal day 1) or as juveniles (4 weeks old). Mice were followed until the 24-week endpoint when editing rates were assessed. (C and D) Survival rates (C) and body weight gain (D) of Ass1^fold mice injected with the PE7 AAV as neonates (n=17, green) or juveniles (n=9, blue). Controls included PBS treated Ass1^fold mice (n=16, orange) and healthy WT mice (n=16, grey). (E) Editing rates at the Ass1^fold locus in hepatocytes and different tissues assessed by NGS of Ass1^fold mice treated as neonates (n=11) or juveniles (n=9). (F) Editing rates in hepatocytes of Ass1^fold mice treated as neonates (n=11) or juveniles (n=9) at GUIDE-seq-nominated off-target sites. (G) RNA editing rates assessed by cDNA sequencing in hepatocytes from Ass1^fold mice treated as neonates (n=11) or juveniles (n=9). (H) Schematic showing the in situ sequencing (ISS) method. ISS was done after the experimental endpoint. To identify edited or unedited sequences, barcoded padlock probes were used. Reverse transcription of Ass1 was followed by padlock hybridization and ligation, rolling circle amplification (RCA), and sequencing-by-synthesis. Amplified RCA colonies were imaged and computationally segmented. (I) Representative images showing a liver section sequenced by ISS using barcoded padlock probes for the Ass1 pathogenic (yellow) and corrected base (red). Sections from Ass1^fold mice treated as neonates or juveniles were counterstained with E-cadherin (green) to mark periportal zone 1 and glutamine synthetase (GS, white) to mark pericentral zone 3. Scale bars: 50 μm. See Fig. S3D for a dot-only version with IHC outlines. (J) Representative images of liver sections of wild type, PBS- and Ass1^fold mice treated as neonates or juveniles treated. Sections were stained with ASS1 (magenta) and GS (red) antibodies (red), and DAPI (blue) for nuclei. Blue arrows show examples of uncorrected cells with speckled ASS1 aggregation, and white arrows of corrected cells with wild-type-like cytosolic ASS1 distribution. Scale bars: 20μm. Data in (D to G) show means ± SD. Fig 2, A and B were created using BioRender.

To assess editing rates in treated animals, we performed deep sequencing of DNA isolated from bulk liver tissue or isolated hepatocytes at experimental endpoints (six months). Editing rates in neonate-treated mice reached 51 ± 8% in bulk liver tissue and 71 ± 4% in isolated hepatocytes (Fig. 2E). In juvenile-treated mice, editing rates reached 35 ± 8% in bulk liver tissue and 54 ± 11% in isolated hepatocytes (Fig. 2E). Analysis of DNA isolated from other tissues confirmed that editing was largely restricted to the liver (Fig. 2E), consistent with the hepatotropic nature of AAV9 and hepatocyte-specific expression from the P3 promoter (32, 33). Furthermore, indel formation rates at the Ass1^fold locus remained below 0.2% in both treatment groups (Fig. 2E). To next characterize unintended off-target editing at other sites in the genome, we experimentally identified off-target binding sites for pegRNA6 via GUIDE-seq (34) in a murine cell line harboring the pathogenic Ass1 locus (Fig. S3, A and B). Targeted deep sequencing of these sites in hepatocytes of treated mice revealed that off-target editing was limited to one locus: a non-expressed pseudogene copy of Ass1 that differs from the on-target sequence of pegRNA6 only by the pathogenic mutation (Fig. 2F). Moreover, in depth analysis of this site revealed that only silent mutations encoded by the pegRNA but no indels were introduced (Fig. S3C, Fig. 2F).

The urea cycle is not uniformly active across the liver but is confined to ASS1 expressing hepatocytes in the periportal zone 1 and intermediate zone 2. We therefore performed RNA sequencing to quantify editing specifically within this functionally relevant subset of hepatocytes. Editing efficiencies in Ass1-expressing hepatocytes were comparable to those measured in total hepatocyte DNA, with 72 ± 5% editing observed in neonate-treated mice and 62 ± 9% in juvenile-treated mice (Fig. 2G). Consistent with these findings, in situ sequencing (ISS - Fig. 2H) of liver sections counterstained for E-cadherin (for zone 1 periportal hepatocytes) and glutamine synthetase (GS - for zone 3 pericentral hepatocytes) confirmed correction of the Ass1^fold mutation in zone 1 and 2 hepatocytes (Fig. 2I, Fig. S3, D and E). Lastly, correction of Ass1 was further confirmed by ASS1 immunostaining. The characteristic speckled aggregation pattern of ASS1 observed in untreated Ass1^fold mice was converted into a diffuse, wild-type-like cytosolic distribution in the majority of hepatocytes in treated animals (Fig. 2J).

Biochemically, Ass1^fold mice recapitulate several hallmark features of patients with CTLN1, including markedly elevated plasma citrulline and ammonia concentrations. In addition, plasma arginine concentrations are altered, but in contrast to patients who typically have reduced arginine concentrations, Ass1^fold mice show elevated plasma arginine (8). Supporting the therapeutic efficacy of prime editing, blood citrulline concentrations were reduced from 1,851 ± 250 µM in untreated mice to 216 ± 61 µM and 268 ± 61 µM in neonate- and juvenile-treated mice, respectively (Fig. 3A). Plasma ammonia similarly normalized from 820 ± 138 µM in untreated mice to 258 ± 52 µM and 239 ± 37 µM in the treated groups, reaching values within the wild-type range (265 ± 62 µM, Fig. 3B). Elevated blood arginine concentrations in Ass1^fold mice (183 ± 22 µM) were likewise normalized to wild-type levels (88 ± 4 µM) in all treated groups (70 ± 9 µM in neonate and 98 ± 13 µM in juvenile treated mice; Fig. S4A). Functional restoration of the urea cycle to wild-type rates was independently confirmed in a ureagenesis assay, which measures the incorporation of isotope-labeled ammonia into urea (Fig. 3C, Fig. S4B).

Figure 3. AAV-mediated prime editing in Ass1^fold mice corrects biochemical and behavioral phenotypes.

(A and B) Blood citrulline (A) and plasma ammonia (B) concentrations in wild type (grey, n=10; n=8), PBS-injected Ass1^fold (orange, n=10; n=5), and Ass1^fold mice treated as neonates (green, n=11; n=11) or juveniles (blue, n=9; n=9). (C) Urea enrichment rates in the ureagenesis assay in wild type (grey, n=12), PBS-injected Ass1^fold (orange, n=12), neonate (green, n=17) or juvenile (blue, n=9) AAV-treated animals. (D) Number of stereotypic jumps per minute during the open field test in wild type (grey, n=22), PBS injected Ass1^fold (orange, n=15), neonate (green, n=17) or juvenile (blue, n=9) AAV treated animals. (E) Number of uncovered marbles during the marble burying test in wild type (grey, n=15), PBS injected Ass1^fold (orange, n=11), neonate (green, n=17) or juvenile (blue, n=9) AAV treated animals. Ureagenesis and behavioral experiments were done with 12-week-old mice, citrulline and ammonia levels were measured at experimental endpoints. Data show means ± SD. Statistical significance was assessed by Brown-Forsythe and Welch ANOVA with Dunnett’s T3 multiple-comparisons test. Only significant comparisons are shown: * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001

Consistent with the neurological involvement observed in CTLN1, Ass1^fold mice exhibit abnormalities, including increased stereotypic jumping, reduced marble-burying behavior, and mildly decreased exploratory activity in open field tests (9). Although we did not observe significant differences between wild-type and Ass1^fold mice in overall movement or zone preference in the open field test (Fig. S4, C and D), stereotypic jumping (Fig. 3D) and marble-burying behaviors (Fig. 3E, Fig. S4E) were impaired in Ass1^fold mice. Both phenotypes were fully normalized following prime editing in neonate- and juvenile-treated animals (Fig. 3, D and E).

Altogether, these findings demonstrated that systemic AAV9-mediated delivery of intein-split PE7 together with pegRNA6 enabled efficient correction of the Ass1^fold mutation in hepatocytes in vivo, enabling correction of CTLN1-related phenotypes including survival, body weights, behavior, and blood markers of metabolic function. In addition, long-term monitoring of a cohort of neonate-treated Ass1^fold mice for up to one year confirmed the durability of the therapeutic effect, with editing efficiencies (Fig. S5) and biochemical markers (Fig. S6) remaining indistinguishable from treated mice analyzed at six months.

Correcting Ass1^fold mice via RNA-LNP-mediated prime editing

Although AAV vectors enable efficient in vivo delivery of prime editors to the liver, they are associated with several inherent limitations, including that adaptive immune responses against the viral capsid hinder efficient re-dosing and prolonged expression of the editor increases the risk of accumulating off-target edits over time (35). Lipid nanoparticles (LNPs) are non-viral, can be administered repeatedly, and, due to the rapid degradation of mRNA and pegRNA, result in transient editor expression. LNP-based delivery of RNA encoding the PE offers a potentially safer alternative to AAV-based administration, and we therefore investigated whether RNA-LNP-mediated PE7 delivery could achieve sufficient prime editing in hepatocytes to correct CTLN1 in Ass1^fold mice.

We performed in vitro transcription of pseudouridine (Ψ)-modified mRNA encoding PE7 and synthesized pegRNA6 protected with 2′-O-methyl-3′-phosphorothioate modifications at the 5′ and 3′ ends. In addition, we added a 3′ poly(U) tail (…UU*mU*mU*mUU-3’) to the pegRNA, which facilitates binding to the La domain of PE7 that protects the pegRNA from degradation (Fig. 4A). The RNAs were encapsulated in hepatotropic LNPs composed of 1,2-distearoyl-sn-glycero-3-phosphocholine, cholesterol, a PEG–lipid, and an ionizable cationic lipid, resulting in particles that are near-neutral at physiological pH. Encapsulation efficiencies reached 94% for the mRNA and 98% for the pegRNA (Fig. S7A). The resulting LNPs had mean particle diameters of 92.1 nm (mRNA-LNP) and 83.8 nm (pegRNA-LNP) (Fig. S7B) and exhibited narrow size distributions, with polydispersity indices of 0.06 and 0.04, respectively (Fig. S7C). Formulations were then systemically administered to 4-day-old neonates at a total dose of 3 mg kg⁻¹ (1.5 mg kg⁻¹ mRNA and 1.5 mg kg⁻¹ pegRNA; Fig. 4B). In addition, we administered RNA-LNP systemically to 4-week-old juvenile mice at a dose of 4 mg kg⁻¹ (2 mg kg⁻¹ mRNA and 2 mg kg⁻¹ pegRNA), either as a single injection or in a triple-dose regimen with five-day intervals (Fig. 4B). All three treatment groups exhibited full correction of the early lethality phenotype (Fig. 4C, Fig. S7D). In addition, we observed partial normalization of body weight in single-dosed juveniles (Fig. S7E), and full normalization to wild-type values in neonate-treated and re-dosed juvenile animals (Fig. 4D).

Figure 4. RNA-LNP-mediated prime editing normalizes survival and body weights in Ass1^fold mice.

(A) Schematic illustration of the mRNA and pegRNA packaged into LNP. The in vitro transcribed pseudouridine (Ψ) modified mRNA encodes the PE7, containing following domains: nSpCas9, nickase of Streptococcus pyogenes Cas9; M-MLV-RT^dRnH, M-MLV RT-delta RNase H; NLS, nuclear localization signal; La, La RNA binding protein. The chemically synthesized pegRNA contains: 2′-O-methyl (m) -3′-phosphorothioate (*) modifications, spacer, primer binding site and reverse transcriptase template. (B) Schematic illustration of the experimental setup. (C) Survival rate and (D) body weight gain of Ass1^fold mice injected with RNA-LNP as neonates (n=5) or as juveniles (n=8). Controls included PBS treated Ass1^fold mice (n=16) and healthy WT mice (n=16). (E) Editing rates at the Ass1^fold locus in hepatocytes and different tissues assessed by NGS of either neonate (n=5) or juvenile (n=8) treated animals. (F) Editing rates in hepatocytes of either neonate (n=5) or juvenile (n=8) treated animals at GUIDE-seq nominated off-target sites. (G) RNA editing rates assessed by cDNA sequencing in hepatocytes from neonate (n=5) or juvenile (n=8) treated animals. (H) Representative images showing a liver section sequenced by ISS using barcoded padlock probes for the Ass1 pathogenic (yellow) and corrected base (red). Sections from neonate or juvenile treated Ass1^fold animals were counterstained with E-cadherin to mark periportal zone 1 (green) and glutamine synthetase (GS) to mark pericentral zone 3 (white). Scale bar: 50μm. See Fig. S7G for a dot-only version with IHC outlines. (I) Representative images of liver sections of wild type, PBS or RNA-LNP-treated neonate or juvenile treated Ass1^fold animals. Sections were stained with ASS1 (magenta) and GS (red) antibody, and DAPI (blue) for nuclei. Scale bar: 20μm. Data on graphs D to G show means ± SD. Fig. 4, A and B were created using BioRender.

We next performed targeted deep sequencing of the Ass1^fold locus at the 6-month experimental endpoint. In neonate-treated animals, correction efficiencies reached 14.2 ± 6.1% in the liver tissue and 24.3 ± 14.7% in isolated hepatocytes (Fig. 4E). In juvenile animals, editing rates were lower, with 0.9 ± 0.6% editing in the liver and 3.5 ± 1.5% editing in hepatocytes after a single RNA-LNP dose (Fig. S7F), and 4.6 ± 1.8% editing in the liver and 13.3 ± 2.6% editing in hepatocytes after three RNA-LNP doses (Fig. 4E). Across all treatment groups, indel frequencies remained below 1%, editing was largely confined to the liver (Fig. 4E, Fig. S7F), and off-target editing was detected only at the Ass1 pseudogene locus (Fig. 4F).

To specifically assess correction rates in the hepatocyte subpopulation functionally relevant to urea cycle activity, we next performed deep sequencing of Ass1^fold transcripts. Editing rates were slightly elevated compared to the values measured by genomic DNA sequencing, reaching 30.6 ± 14.3% correction in neonate-treated mice and 21.6 ± 7.5% in re-dosed juvenile animals (Fig. 4G). In situ sequencing of liver sections further confirmed Ass1^fold correction in zone 1 and 2 hepatocytes (Fig. 4H, Fig. S7G). Moreover, immunohistochemical analysis of ASS1 demonstrated restoration of the characteristic speckled aggregation pattern of mutant ASS1 to a wild-type-like cytosolic distribution in a substantial fraction of hepatocytes (Fig. 4I).

To determine whether molecular correction translated into sustained metabolic and functional rescue, we next analyzed biochemical and behavioral parameters. Blood citrulline concentrations were reduced from 1851± 250 µM in untreated mice, to 330 ± 103 µM in neonate-treated animals, 350 ± 74 µM in re-dosed juveniles (Fig. 5A), and to 612 ± 359 µM in single-dosed juveniles (Fig. S7H). Likewise, plasma ammonia concentrations normalized from 820 ± 138 µM in untreated mice, to wild-type ranges in neonates and re-dosed juveniles, (233 ± 33 µM, 237 ± 62 µM respectively; Fig. 5B), and were reduced to 533 ± 180 µM in single-dosed juveniles (Fig. S7I). Arginine concentrations were restored to wild-type values across all treated groups (Fig. S7J). Consistent with these biochemical improvements, in all treated groups the ureagenesis assay demonstrated complete restoration of hepatic ammonia clearance (Fig. 5C, Fig. S7K), and behavioral testing showed full correction of stereotypic jumping and marble-burying deficits (Fig. 5, D and E; Fig. S7, L and M).

Figure 5. RNA-LNP-mediated prime editing corrects biochemical and behavioral phenotypes in Ass1^fold mice.

(A and B) Blood citrulline (A) and plasma ammonia (B) concentrations in wild type animals (grey, n=10; n=8), PBS-injected Ass1^fold animals (orange, n=10; n=5), and Ass1^fold animals injected with RNA-LNP as neonates (green, n=5; n=5) or juveniles (blue, n=8; n=8). (C) Urea enrichment rates in the ureagenesis assay in wild type mice (grey, n=12), and Ass1^fold mice injected with PBS (orange, n=12), with RNA-LNP as neonates, (green, n=5) or with RNA-LNP as juveniles (blue, n=8). (D) Number of stereotypic jumps per minute during the open field test in wild type mice (grey, n=22), and in Ass1^fold mice injected with PBS (orange, n=15), with RNA-LNP as neonates (green, n=5) or with RNA-LNP as juveniles (blue, n=8). (E) Number of uncovered marbles during the marble burying test in wild type mice (grey, n=15) and in Ass1^fold mice injected with PBS (orange, n=11) or with RNA-LNP as neonates (green, n=5) or with RNA-LNP as juveniles (blue, n=8). Ureagenesis and behavioral experiments were done with 12-week-old mice, citrulline and ammonia concentrations were measured at experimental endpoints. Data show means ± SD. Statistical significance was assessed by Brown-Forsythe and Welch ANOVA with Dunnett’s T3 multiple-comparisons test. Only significant comparisons are shown: * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001

Lastly, we assessed potential immune activation and inflammatory responses after systemic RNA-LNP delivery and repeated dosing. To this end, we quantified circulating pro-inflammatory cytokines, including interleukin-1α (IL-1α), interferon-γ (IFNγ), tumor necrosis factor-α (TNF-α), and monocyte chemoattractant protein-1 (MCP-1). Although transient elevations of IL-1α, IFNγ, and MCP-1 were observed 6 hours after treatment, cytokine concentrations returned to baseline within 3 days (Fig. S8A). Consistent with these findings, serum alanine aminotransferase (ALT) and aspartate aminotransferase (AST) concentrations did not increase in either neonate- or juvenile-treated animals at the experimental endpoints, suggesting an absence of persistent liver injury (Fig. S8B). Taken together, treatment of Ass1^fold mice with RNA-LNPs encoding PE7 and pegRNA6 enabled correction of the pathogenic mutation in hepatocytes, resulting in the rescue of both metabolic and behavioral phenotypes.

Correction of recurrent human ASS1 mutations

CTLN1 is genetically heterogeneous, with over 100 pathogenic ASS1 mutations reported in patients (36). Given the versatility of prime editing in correcting nearly all types of small nucleotide variants, we next evaluated whether other disease-causing ASS1 mutations could be edited with efficiencies comparable to the T389I variant in Ass1^fold mice.

We selected 15 recurrent and clinically relevant ASS1 variants, which collectively account for most of reported alleles (36). These mutations include the human pathogenic T389I variant modeled in Ass1^fold mice, as well as ASS1^G390R, the most common variant found in patients (Fig. 6A). We then used PRIDICT2.0 for in silico testing of different pegRNA designs for each mutation, including silent bystander substitutions. The three top-scoring pegRNAs per allele were selected (Fig. S9) and experimentally tested in K562 reporter cells harboring all 15 pathogenic alleles as well as the murine Ass1^fold allele. Co-transfection of PE7 and the respective pegRNAs showed 6 out of the 15 tested mutations, including the most common variant ASS1^G390R, could be corrected with similar or higher efficiency (>20%) than the Ass1^fold mutation (Fig. 6B), with all but three alleles reaching correction efficiencies exceeding 10%. In addition, indel frequencies remained below 0.4% at all loci. These findings suggest that prime editing could achieve therapeutically relevant correction efficiencies across a broad spectrum of pathogenic ASS1 variants, highlighting its potential for the treatment of CTLN1.

Figure 6. Correction of recurrent human pathogenic ASS1 mutations.

(A) Location of the targeted recurrent pathogenic mutations in the human ASS1 gene. Yellow boxes illustrate 5’ and 3’ UTRs, blue boxes illustrate coding exons. (B) pegRNA efficiencies were in silico tested with PRIDICT2.0 and the three top-performing designs per mutation were experimentally tested in a reporter K562 cell line harboring the mutation (see fig. S9). Prime editing and indel rates are shown for the most active pegRNA design per mutation. Data show means ± SD for n=3 replicates.

Discussion

In this study, we demonstrated durable correction of a CTLN1 mouse model using in vivo prime editing. Delivery of the PE7 prime editor and an optimized pegRNA via AAV or RNA-LNP resulted in efficient correction of the pathogenic Ass1^fold allele in mouse hepatocytes in vivo, leading to normalization of survival, body weight gain, and plasma ammonia and citrulline concentrations, restoration of ureagenesis, and reversal of CTLN1-associated behavioral abnormalities. Gene correction was highly specific, with low rates of indel generation at the target site and off-target editing limited to a non-expressed Ass1 pseudogene.

A major advantage of prime editing over CRISPR-Cas9 nuclease-based strategies is that it enables precise editing without DSB formation and without relying on homology-directed repair (20, 37). Accordingly, we observed Ass1^fold correction not only in newborn mice but also in juvenile animals in which most hepatocytes are quiescent and do not divide within the 2-3 day time window of PE expression after RNA-LNP delivery. This observation is relevant for clinical translation, because hepatocyte turnover in pediatric patients is substantially slower than in neonatal mice, where the liver mass doubles within the first 5-7 days after birth (38) compared to about 6-7 months in humans (39). In addition, avoiding DSBs markedly reduces the risk of deleterious indel mutations. This is a particularly important consideration in CTLN1, where many patients carry at least one hypomorphic ASS1 allele with residual enzyme function that could be lost through imprecise editing (36).

Base editing is another DSB-free genome editing strategy that has been successfully applied in the livers of mice and macaques (31, 40), and more recently in a patient diagnosed with severe carbamoyl-phosphate synthetase 1 (CPS1) deficiency (41). However, although base editing is often more efficient than prime editing, it is largely restricted to the correction of transition point mutations. In addition, base editors require a PAM at a fixed distance from the target nucleotide, and editing is often accompanied by bystander mutations. In contrast, the programmable RT template in prime editing enables correction of various classes of mutations, including small insertions and deletions, without introducing bystander mutations.

Although in this study we first evaluated AAV-mediated PE7 delivery, which resulted in high efficiencies and achieved a phenotypic cure after a single AAV administration, this approach comes with important translational limitations. In particular, prolonged editor expression from AAV raises concerns about cumulative off-target activity over time, and adaptive immune responses against AAV capsids limit clinical applicability in patients with pre-existing immunity and preclude efficient re-dosing (42). We therefore explored RNA-LNP-mediated PE7 delivery, because LNPs are less prone to inducing durable adaptive immune responses against the delivery vehicle, thereby reducing concerns related to pre-existing immunity and enabling repeat administration if required. In addition, mRNA expression is inherently transient and thus restricts the temporal window of prime editor activity. Importantly, despite this transient expression, RNA-LNP-mediated delivery was sufficient to durably restore ureagenesis and biochemical markers back to wild-type levels in Ass1^fold mice. This contrasts previous AAV-mediated ASS1 cDNA replacement approaches, where the therapeutic efficacy declined within six months due to episomal vector loss during hepatocyte turnover (9).

We also demonstrated that prime editing could efficiently correct a broad range of recurrent pathogenic ASS1 mutations in a cellular reporter system, including the most common variant, ASS1^G390R. Our pegRNA design pipeline integrates in silico prediction via PRIDICT with functional validation in MMR proficient K562 cells, which closely mimic editing outcomes in hepatocytes (25). pegRNAs encoding silent bystander mutations often outperformed those correcting only the pathogenic variant, likely due to their capacity to evade MMR. These results underscore the importance of systematically screening a broad set of pegRNA designs to identify the most efficient and clinically relevant candidates.

This study has several limitations that are important for the interpretation and clinical translation of our findings. First, all in vivo data were generated in the murine model, and editing efficiencies, dosing requirements, and immune responses may not fully translate to human patients, particularly given differences in hepatocyte turnover and liver growth dynamics. Second, the prime-editing efficiency achieved with RNA-LNP delivery in juvenile animals was relatively low (3.5% correction in hepatocytes after a single dose), necessitating repeat dosing to achieve editing rates sufficient for a full phenotypic rescue (13.2% correction after three doses of RNA-LNP). As discussed above, juvenile mice likely represent a more clinically relevant model for genome editing in postnatal pediatric patients than neonatal animals with rapidly proliferating hepatocytes (43). Accordingly, the editing efficiencies observed here suggest that multiple administrations of RNA-LNP may be required to achieve therapeutic correction in patients with CTLN1. Lastly, an additional limitation for the clinical translation of prime editing is that reagents must be individually designed and validated for each distinct pathogenic CTLN1 mutation, increasing regulatory complexity and costs compared with mutation-agnostic gene replacement strategies.

These limitations highlight the need for further optimization of prime editing strategies and clinical development frameworks. To improve the efficiency of in vivo prime editing, several approaches could be explored. For example, PE delivery could be enhanced by employing GalNAc-functionalized LNPs, which bind the asialoglycoprotein receptor (ASGPR) - the most abundant hepatocyte surface receptor (44, 45) - thereby increasing cellular uptake of LNP-encapsulated PE mRNA and pegRNA. In addition, in a previous study we demonstrated that a dual delivery strategy, in which the PE was delivered via mRNA-LNP whereas the pegRNA was encoded on an AAV, could substantially increase editing efficiency, likely by increasing the amount of functional pegRNA throughout the entire window of prime editor expression (31). However, although this approach maintains temporally restricted prime editor activity, pre-dosing patients with an AAV prior to mRNA-LNP administration would introduce additional complexity and potential barriers, such as pre-existing immunity to AAV capsids. A more attractive alternative would therefore be to enhance the availability of functional pegRNA through further chemical optimization of synthetic, LNP-delivered pegRNAs, for example by incorporating additional stabilizing modifications to reduce rapid intracellular degradation (46). The regulatory complexity associated with mutation-specific prime-editing approaches may also be mitigated by emerging clinical and regulatory frameworks. These have gained increasing attention following the recent treatment of an infant with CPS1 deficiency (41), which demonstrated the feasibility of individualized, mutation-specific genome editing in a clinical setting. One such approach is FDA-supported umbrella trial concepts, which aim to streamline development by allowing multiple mutation-specific therapies to be evaluated within a single clinical and regulatory framework, rather than requiring fully independent trials for each targeted mutation (47).

In summary, our study demonstrates that in vivo RNA-LNP-mediated prime editing in the liver enables precise and durable correction of the Ass1^fold mouse model for CTLN1. With continued optimization of editing efficiency and delivery, this platform may hold promise as a potentially curative therapy for individuals with CTLN1 and other genetic liver diseases.

Materials and Methods

Study design

The aim of this study was to establish and compare in vivo prime-editing strategies for the treatment of citrullinemia type I (CTLN1) in the Ass1^fold mouse model. Optimized pegRNA designs were identified in silico using PRIDICT2.0 and experimentally validated in mismatch-repair-proficient reporter cell lines using targeted amplicon sequencing to quantify editing and indel rates. The most active pegRNA was selected for in vivo studies. For animal experiments, homozygous Ass1^fold mice were treated either as neonates or juveniles with AAV9-delivered intein-split PE7 or with RNA-LNPs encoding PE7 and pegRNA. Littermate wild-type and PBS-treated Ass1^fold mice served as controls. Neonatal experiments were performed by treating all pups in a litter, with only homozygous animals included in the analyses. No formal power calculations were performed; sample sizes were based on prior experience with the model and practical considerations of animal availability. Animals were randomly assigned to treatment groups without sex bias. Injected animals were used for targeted deep sequencing of genomic DNA and cDNA, off-target analysis by GUIDE-seq-nominated sites, in situ sequencing, immunofluorescence, biochemical analyses (citrulline, ammonia, ureagenesis), and behavioral testing. No data were excluded from analysis. With the exception of behavioral experiments, investigators were not blinded to group allocation. Biological replicates, animal numbers and group sizes are indicated in the respective figure legends. A detailed table containing the primary data and animal genders is reported in data file S1.

RNA synthesis and LNP encapsulation

mRNA production was performed as previously described (48, 49). In short, the coding sequences of the PEs were PCR amplified with Q5 High-Fidelity DNA Polymerase (NEB) and cloned into the mRNA backbone containing 101 nucleotide-long poly(A) tails the mRNA production plasmid using HiFi DNA Assembly Master Mix (NEB). mRNAs were transcribed to contain 101 nucleotide-long poly(A) tails. For modified nucleoside-containing mRNA, m1Ψ-5′-triphosphate (TriLink) instead of UTP was used to generate was added in the in vitro transcription modified nucleoside-containing mRNA. Capping of the in vitro transcribed mRNAs was performed co-transcriptionally using the trinucleotide cap1 analog known as CleanCap (TriLink). mRNA was purified by cellulose (Sigma-Aldrich) purification as described (50). All mRNAs were analyzed by agarose gel electrophoresis and were stored frozen at −80 °C. All synthetic pegRNAs were ordered from IDT (Table S3).

LNPs were formulated as previously described (51). In short, an ethanolic solution of prepared with 1,2-distearoyl-sn-glycero-3-phosphocholine, cholesterol, a PEG lipid and an ionizable cationic lipid was rapidly mixed together with an acidic aqueous solution containing prime editor mRNA or pegRNA using an automated syringe pump and an in-line mixer. The ionizable lipid belongs to the lipid class described in Table 3 of ref. (52). It is characterized by a tertiary amine headgroup and multiple ester-linked hydrophobic chains, conferring pH-dependent ionization, efficient endosomal escape, and liver-targeted delivery. The resulting LNP formulation was dialyzed overnight against 1x PBS, 0.2-μm sterile filtered, and stored at −80 °C at a concentration of 1 μg/μL of total RNA. Based on Quant-iT Ribogreen Assay (Life Technologies), the encapsulation efficiencies of RNA in the LNP were 94% for the pegRNA and 98% for the PE mRNA (Fig. S7A) as measured by the Quant-iT Ribogreen Assay (Life Technologies). Based on measurements with the Malvern Zetasizer (Malvern Panalytical) the LNP sizes did not exceed 100 nm (Fig. S7B). The polydispersity index (PDI) of the LNP (a measure of particle size uniformity with <0.1 being indicative of a uniform, monodispersed LNP population) was between 0.016-0.082 (Fig. S7C). This uniformity has been validated extensively, verified by cryoEM and single particle analysis. The LNP has a neutral charge at physiological pH, based onthe pKa of the ionizable lipid (as determined by TNS assay) and by the Malvern Zetasizer.

Animal studies

Animal studies were carried out following protocols authorized by the Kantonales Veterinäramt Zürich and in full accordance with applicable ethical guidelines (licence: 146/2023). C57BL/6J (Strain #:000664) and Ass1^fold (Strain #:006449) mice were housed in a pathogen-free, temperature- and humidity-controlled room on a 12-hour light-dark cycle. at the Institute of Pharmacology and Toxicology of the University of Zurich. Mice were fed a standard laboratory chow (Kliba Nafag no. 3437 with 18.5% crude protein) and genotyped at weaning. Neonate mice were injected with 2.5×10¹³ vg/kg (AAV) or with 3 mg kg⁻¹ RNA-LNP via the temporal vein at P1 for AAVs and at P4 for LNPs. Juvenile mice were retro-orbitally injected at 4 weeks old with 2.5×10¹³ vg/kg (AAV) or with 4 mg kg⁻¹ RNA-LNP. Redosing of LNPs was done 5 days apart. Behavior and metabolic experiments were performed on 12-week-old animals or on 12-month-old animals.

Open field test

Locomotion and stereotypic jumping behavior were assessed in a square arena (40 × 40 × 40 cm, opaque walls) recorded by an overhead camera and analyzed by EthoVision XT. Prior each behavioral test the arena was cleaned with 70% ethanol and the experimental setup was designed as following; Post 60 minute acclimation to the testing room, each mouse was placed in the center of the arena and allowed to explore for 15 min. Primary endpoints included total distance traveled, time spent in the center zone (central 20 × 20 cm) and jump counts.

Marble burying test

Mice were placed individually in standard cages (≈30 × 20 × 15 cm) containing 5 cm of fresh bedding. Twenty glass marbles (15 mm diameter) were evenly arranged in a 4 × 5 grid on the bedding surface (Fig. S4E). After 30 min, mice were removed and the number of marbles buried ≥ 2/3 of their diameter was counted by a blinded observer. Cages were cleaned with 70% ethanol and bedding replaced between subjects. Scientists performing and analyzing behavioral data were blinded during the study.

Ureagenesis quantification

After fasting animals for 2 hours a baseline dried blood sample was collected then ¹⁵NH₄Cl (≥99% isotope enrichment, Cambridge Isotope Laboratories) was administered intraperitoneally at a dose of 4 mmol/kg (53). Additional blood samples were collected at 30- and 60-minutes post-injection, prior to ureagenesis analysis according to a previously published method (54). Samples were stored at −20 °C until analysis. Isotopic ratios were determined as the fraction of urea in the sample labeled with ¹⁵N urea, normalized to the corresponding metabolite pool size, and the area under the curve (AUC) from 0 to 60 minutes was calculated.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 10.6.0. Equal variances were assessed by the Brown-Forsythe test. When variances were unequal, Welch’s analysis of variance (ANOVA) with Dunnett’s T3 multiple comparisons test was used, otherwise ordinary one-way ANOVA with Šídák’s multiple comparisons test was performed. Data are represented as biological replicates and are depicted as means ± standard deviation (s.d.) as indicated in the corresponding figure legends. Likewise, sample sizes and the statistical tests used are described in detail in the respective figure legends. For all analyses, P < 0.05 was considered statistically significant.

Detailed descriptions of the following methods are provided in the Supplementary Materials and Methods: generation of plasmids; cell culture, transfection; hepatocyte isolation; quantification of citrulline and ammonia from blood; generation of reporter cells by PiggyBac transposon; AAV production; histology and immunofluorescence; detection of plasma immune activation and liver damage markers; in situ sequencing; deep sequencing; NGS data analysis; RNA sequencing; GUIDE-seq.

One Sentence Summary.

Prime editing via RNA-LNP corrects multiple pathogenic ASS1 mutations and restores ureagenesis and survival in a mouse model of citrullinemia type I.

Data, code and materials availability

All data associated with this study are present in the paper or supplementary materials. Illumina sequencing data are available in the Sequence Read Archive under the accession number PRJNA1403429. The following plasmids described in this study are available from Addgene: pAT15535_hU6_epegRNA_Ef1a-mCh(#251104), pAT159_AAV_p3_PE7C_hU6_epegRNA (#251105), pAT172_AAV_p3_PE7-N (#251106). The following publicly available codes were used in this study: PRIDICT2.0 (https://github.com/uzh-dqbm-cmi/PRIDICT2), CRISPResso2 (https://github.com/pinellolab/crispresso2), GUIDE-seq (http://github.com/aryeelab/guideseq). All other materials used or generated in this study are commercially available or will be supplied by the corresponding author upon reasonable request, except for the LNP formulations, which will be provided by Acuitas to academic researchers for the purpose of reproducing the results reported in this study, subject to a materials transfer agreement.