Patient-Specific In Vivo Gene Editing to Treat a Rare Genetic Disease

SUMMARY

Base editors can correct disease-causing genetic variants. We developed a customized base-editing therapy, starting immediately after a neonate was diagnosed with severe carbamoyl phosphate synthetase 1 deficiency, a disease with an estimated 50% mortality in early infancy. After regulatory approval, he received two infusions of the therapy, at approximately seven and eight months of age. In the 7 weeks after initial treatment, he tolerated increased dietary protein and nitrogen scavenger reduction despite viral illnesses. There were no serious adverse events. Longer follow-up is required to assess safety and efficacy. (Funded by the U.S. National Institutes of Health and others.)

Programmable clustered regularly interspaced short palindromic repeats (CRISPR) gene-editing technologies¹ have matured into therapeutic approaches that are improving the lives of patients with a handful of diseases such as sickle cell disease, beta-thalassemia, and hereditary angioedema.^2–4 Precise, corrective CRISPR technologies—namely base editing, which can effect cytosine-to-thymine changes (cytosine base editing⁵) or adenine-to-guanine changes (adenine base editing⁶), and prime editing,⁷ which can produce any single-nucleotide change or small insertion or deletion—can potentially address >90% of pathogenic variants in genetic diseases that, while individually rare, collectively affect hundreds of millions of people worldwide.⁸ However, drug development efforts have largely focused on recurrent variants in a few relatively common genetic diseases due to the extensive resources needed to develop and bring to market any given therapy.⁹

We developed a workflow for the rapid development of customized, corrective gene-editing therapies for patients with ultra-rare or unique “N-of-1” variants (Fig. 1). More specifically, we developed a base-editing therapy, delivered in vivo to hepatocytes via lipid nanoparticles, for a single patient diagnosed at birth with neonatal-onset carbamoyl phosphate synthetase 1 (CPS1) deficiency, an ultra-rare inborn error of metabolism impacting the urea cycle (incidence of 1:1,300,000¹⁰) with an estimated 50% mortality in early infancy.¹¹ Liver transplantation provides a functional urea cycle and improves outcomes.^12,13 However, infants often suffer hyperammonemic crises and irreversible neurologic injury before growing large enough to undergo transplantation.^14–16 We administered the customized therapy to the patient twice, at approximately 7 and 8 months of age, with the goal of providing protection against hyperammonemia.

Figure 1.

Timeline from Birth to Second Treatment with K-abe.

METHODS

Studies for Investigational New Drug Application

Full descriptions of cellular studies, animal studies, and off-target assessments are included in the Supplementary Appendix. Mouse and nonhuman primate studies were approved by institutional animal care and use committees at the University of Pennsylvania and AmplifyBio, respectively. The patient’s genome-sequencing data and his father’s blood-derived genomic DNA were obtained under a human-subjects research protocol approved by the institutional review board (IRB) of the University of California, Berkeley.

Clinical Study

Because the therapy (kayjayguran abengcemeran, or k-abe) was dosed as part of clinical care under a single-patient expanded-access Investigational New Drug (IND) application, the clinical protocol was reviewed through alternative procedures by the Children’s Hospital of Philadelphia (CHOP) IRB. We obtained U.S. Food and Drug Administration (FDA) authorization to obtain concurrence by the IRB chairperson who approved the study. The patient’s parents provided written informed consent. Clinical activities were overseen by a multidisciplinary oversight committee comprising physicians from the CHOP metabolism, hepatology, immunology, gene therapy, and medical ethics services. The investigators vouch for the accuracy and completeness of the data and for the fidelity of the study to the protocol, available with the full text of this article at NEJM.org.

RESULTS

Clinical Presentation

The patient developed symptoms of CPS1 deficiency including lethargy and respiratory distress within the first 48 hours of life. Measurement of blood ammonia revealed a level greater than 1,000 μmol/L (normal 9–33 μmol/L). He was promptly started on continuous renal replacement therapy (CRRT). Plasma amino-acid profiling demonstrated critically elevated glutamine, undetectable citrulline, and a normal level of urine orotic acid suggestive of a proximal urea cycle defect. Rapid targeted analysis of the patient’s genome identified two severe, truncating CPS1 variants, c.1003C>T (p.Gln335Ter, referred to as Q335X, on the paternal allele) and c.2140G>T (p.Glu714Ter or E714X, on the maternal allele). The Q335X variant is absent in the Genome Aggregation Database but has been reported in a case of neonatal-onset CPS1 deficiency.¹⁷

The patient was weaned from CRRT and transitioned to chronic therapy including nitrogen scavenger medication (glycerol phenylbutyrate), citrulline replacement (200 mg/kg/day, unchanged during his clinical course), and a protein-restricted diet (given as a 1:1 mix of natural protein and essential amino acid formula). He had the expected infantile “honeymoon” period¹⁴ from days of life 50–100, after which his biochemical status worsened, necessitating further reduction of protein intake and uptitration of glycerol phenylbutyrate to manage elevated ammonia and glutamine levels. Each hyperammonemic episode incurred a risk of permanent neurologic damage and death. Given his disease severity, he was listed for liver transplantation at 5 months despite the acknowledged risks at such a young age.

Optimization of Base-Editing Therapy

Reliable assessment of base editing of the CPS1 Q335X variant ideally would use human hepatocytes with the variant, which were not available. As a proxy, we used the cultured human HuH-7 cell line. We synthesized a cassette harboring a 100-bp human genomic segment spanning the CPS1 Q335X variant, as well as 100-bp segments spanning the patient’s other CPS1 variant and two reference variants in PAH to serve as positive controls (Fig. S2A). We transduced HuH-7 cells with a lentiviral vector containing the cassette, thus inserting it into the genome. This was completed 1 month after the patient’s birth.

To develop a patient-specific, bespoke editing solution, we screened various adenine base editors (ABEs) with guide RNAs (gRNAs) tiling the site of the Q335X variant in the lentivirus-transduced HuH-7 cells (Fig. S2B–D, Fig. S3). We identified an ABE with a preference for NGC protospacer-adjacent motifs, termed NGC-ABE8e-V106W, and a gRNA with the target Q335X adenine in the eighth position of its protospacer sequence as the optimal solution; although there was bystander editing of adenines besides the target adenine, all such edits were synonymous (Figs. S4–S7). This was completed 2 months after the patient’s birth. We named the version of the gRNA used in the lipid nanoparticle therapy (Table S1) “kayjayguran”, the mRNA used in the therapy to express the ABE (Fig. S1, Table S2) “abengcemeran”, and the therapy “k-abe” for short.

Preclinical Studies

Following initial regulatory review by the FDA, we manufactured a toxicology batch of k-abe and undertook a limited safety study in cynomolgus monkeys to characterize single-dose toxicity of the lipid nanoparticle therapy. A total RNA dose of 1.5 milligrams per kilogram of body weight (mg/kg) was administered intravenously. There were no clinical signs, and there were transient elevations in alanine aminotransferase (ALT) and aspartate aminotransferase (AST) to several times the upper limit of normal, consistent with prior studies^18,19 (Fig. S12). Two weeks following treatment, plasma levels of lipid excipients had fallen more than 99.5% from peak levels, supporting re-dosing at intervals greater than 2 weeks (Fig. S13). This study in cynomolgus monkeys, completed 5 months after the patient’s birth, supported 0.1 mg/kg as a potentially safe initial clinical dose for the patient.

Immediately upon learning of the patient’s genetic diagnosis, we started generating mouse models to assess the in vivo editing efficiency of k-abe. To maximize the chance of success, we used established CRISPR reagents in mouse zygotes to insert a cassette harboring a 100-bp human genomic segment spanning the CPS1 Q335X variant into the Rosa26 “safe harbor” locus (the same cassette used for the lentivirus-transduced HuH-7 cells) (Fig. 2A). We performed a limited dose-response study with a small number of Rosa26-Q335X mice, 5 months after the birth of the patient, using the toxicology batch of k-abe and observing up to 42% whole-liver corrective editing, along with the expected synonymous bystander editing (Fig. 2B, Figs. S10–S11A). Editing was evident at the lowest dose, 0.1 mg/kg, supporting that dose as the initial clinical dose for the patient. Subsequent validation of in vivo corrective editing using a second mouse model, with the Q335X variant introduced into the endogenous mouse Cps1 locus, is described in the Supplementary Appendix (Fig. S9, Fig. S11B).

Figure 2. Preclinical Studies.

Panel A shows how a single-strand DNA oligonucleotide cassette harboring the CPS1 Q335X variant sequence and three other variants was inserted into the endogenous mouse Rosa26 locus in mouse zygotes using CRISPR-Cas9 to introduce a double-strand break in the Rosa26 locus, followed by homology-directed repair with the cassette. Panel B shows whole-liver corrective adenine base editing of the CPS1 Q335X variant in Rosa26-Q335X mice. Following a single treatment with the toxicology batch of k-abe at the indicated dose, each mouse had multiple liver samples, distributed throughout the liver, collected upon necropsy (N = 8 samples per mouse in juvenile mice treated at 1–2 months of age, with necropsy several days after treatment) and assessed with next-generation sequencing of the Rosa26-Q335X cassette. Across the dose groups, there was ≤1% indel mutagenesis at the target site. Panel C shows corrective adenine base editing of the CPS1 Q335X variant in lentivirus-transduced HuH-7 cells treated with k-abe. Editing was determined three days following treatment at the stated dose (concentration after dilution into cell medium). The best-fit agonist response curve with variable slope (4 parameters) and EC₅₀ and EC₉₀ values were calculated with GraphPad Prism. Panel D shows the evaluation of a high-priority subset of nominated off-target sites for any adenine-to-guanine editing via individual targeted amplicon sequencing in untreated cells versus cells 3 days following treatment with k-abe, at 1000 nanograms per milliliter of media, of four types: Q335X lentivirus-transduced HuH-7 cells and primary human hepatocytes (PHHs) from donor ICH (13 months of age, male), donor PDV (7 weeks of age, male), and donor YEQ (6 months of age, male). Out of 21 high-priority nominated off-target sites, data for only the 16 sites for which sequencing was successful are shown here.

Upon production of the clinical batch of k-abe, 5 months after the birth of the patient, we performed a dose-response potency assessment in lentivirus-transduced HuH-7 cells (Fig. 2C, Fig. S8). To assess off-target editing, we performed ONE-seq¹⁸ and CHANGE-seq-BE,²⁰ using recombinant NGC-ABE8e-V106W protein and kayjayguran, and a modified GUIDE-seq²¹ method, using a nuclease version of the editor (Figs. S14–S18). The ONE-seq assay used a synthetic library designed with the patient’s genome as the reference genome. The CHANGE-seq-BE assay used genomic DNA obtained from the patient’s father, who carried the Q335X variant (we were unable to obtain enough genomic DNA from the patient). We prioritized the on-target CPS1 site and 21 nominated off-target sites for verification with individual targeted amplicon sequencing (Fig. 2D, Tables S3–S6). We exposed lentivirus-transduced HuH-7 cells and primary human hepatocytes (PHHs) from three donors to a supersaturating dose of k-abe. Low-level synonymous bystander editing was evident at the endogenous wild-type CPS1 genomic site in all four cell lots, consistent with the gRNA (kayjayguran) having a 1-base mismatch to the wild-type sequence (the HuH-7 cells retained endogenous wild-type CPS1 alleles in addition to the transduced CPS1 Q335X variant sequence). Minimal off-target editing was observed only at an intronic site in ATP7B in HuH-7 cells, though not in the three PHH lots. ATP7B encodes a copper transporter and was judged not to represent biological risk because loss of function has not been linked to carcinogenesis.²² Subsequent analysis of a larger set of nominated off-target sites detected no off-target editing in treated PHHs, as described in the Supplementary Appendix (Fig. S19).

Treatment of the Patient

A single-patient expanded-access IND application was submitted to the FDA when the patient was 6 months of age and approved 1 week later. Out of concern for his being cross-reactive immunologic material (CRIM)-negative and potentially mounting an immune response to full-length CPS1 protein, he was started on prophylactic immunosuppression with sirolimus and tacrolimus. This steroid-sparing regimen was selected because corticosteroids can trigger hyperammonemia in patients with CPS1 deficiency. On day of life 208, the patient received an intravenous infusion of k-abe, comprising a total RNA dose of 0.1 mg/kg. After treatment, it was possible to liberalize his dietary protein intake (as he was born at 35 weeks gestation, at times his prescribed protein goal was above the chronological recommended daily allowance) (Fig. 3A–C). He recovered from a viral respiratory infection without experiencing an illness-associated hyperammonemic crisis; however, he received intravenous fluids as is standard during illness and was on a protein-free diet for one day (day of life 225). We were unable to wean glycerol phenylbutyrate (reduced from 10.1 to 8.1 mL/m²/day but restored to the original dose due to rising glutamine levels).

Figure 3. Biochemical Profile Before and After Treatment with K-abe.

Protein intake (panel A) and levels of plasma ammonia (panel B), glutamine (panel C), and alanine aminotransferase (ALT) and aspartate aminotransferase (AST) (panel D) during the patient’s lifetime, including the 7 weeks following k-abe dose 1 (up to day of life 256). The gray bars from left to right indicate periods of rotavirus-positive gastroenteritis prior to treatment, rhinovirus-positive upper respiratory infection after dose 1, and two viral illnesses after dose 2 (gastroenteritis followed by a new rhinovirus/enterovirus infection with associated viral transaminitis). The horizontal dotted lines indicate upper limits of normal laboratory value ranges. Panel E and panel F show pre-treatment and post-treatment plasma ammonia levels and urine orotic acid levels, respectively.

Given the incomplete biochemical correction, and according to the clinical protocol, the patient received a second dose of k-abe, 0.3 mg/kg, 22 days after the first dose. The only adverse event was a coughing episode during the second infusion that resolved with nasal suctioning. Transient ALT and AST elevations occurred a few days following the second dose, with recurrence a few weeks later in the setting of viral illness (Fig. 3D). He tolerated a halving of glycerol phenylbutyrate (5 mL/m²/day) 2 weeks after the second k-abe dose. In the 4 weeks following the second infusion of k-abe, the patient developed two viral infections, each with vomiting and diarrhea. In contrast to a gastroenteritis infection that occurred prior to treatment with k-abe, he recovered from the illnesses without suffering a hyperammonemic crisis and was able to continue his full-protein diet during the illnesses. Blood ammonia levels before the first dose (median 23 μmol/L, interquartile range 14–48 μmol/L), between the first and second doses (9 μmol/L, 9–19 μmol/L), and after the second dose (13 μmol/L, 9–28 μmol/L) support a treatment-related difference (Fig. 3E). CPS1 contributes to orotic acid synthesis, and CPS1 deficiency patients often have low-normal urine orotic acid levels (pre-treatment median 1.7 mmol/mol-Cr, interquartile range 1.6–1.8 mmol/mol-Cr); following the two treatments, levels were often high-normal or above-normal (2.4 mmol/mol-Cr, 2.0–3.0 mmol/mol-Cr; 2.6 mmol/mol-Cr, 2.0–3.6 mmol/mol-Cr) (Fig. 3F). The patient’s weight increased from the 9th percentile prior to the first dose, at day of life 207, to the 26th percentile by day of life 256, the end of the 7-week follow-up period.

DISCUSSION

In this study, we describe a personalized base-editing therapy wholly developed in the 6-month span following a patient’s birth. The patient was able to increase protein intake and tolerate a halving of dose of nitrogen scavenger, despite the “stress tests” presented by consecutive viral infections. The short follow-up is a limitation of this study; longer follow-up is needed to assess the safety and efficacy of k-abe. Liver biopsy to assess for editing was deferred because it posed an unacceptable risk to the infant. The potential for germline editing with k-abe could not be evaluated, although a study of a different lipid nanoparticle gene-editing drug did not detect editing in sperm samples from non-human primates and there was no germline transmission of editing of female mice to offspring.¹⁹

An advantage of lipid nanoparticle therapies is the potential for re-dosing,²³ which is contraindicated with adeno-associated virus (AAV)-delivered therapies, given the immunogenicity of the vector. We opted to start with an initial very low dose to evaluate the safety of this approach in an infant, followed by a moderately higher second dose. In principle, the patient could receive additional and higher doses of k-abe in the future, if needed.

Therapies similar to k-abe could be developed for hundreds of hepatic inborn errors of metabolism. Like antisense oligonucleotides,^24,25 corrective gene editing lends itself to rapid customization for individual patients due to the platform nature of the technology.⁹ Shared components among gene-editing therapies could include the same lipid nanoparticle formulation and mRNA, with the gRNA customized to each patient’s variant.

We assessed k-abe for editing efficiency in mice and for safety in nonhuman primates. Such studies might not be necessary for future patient-specific treatments; perhaps cell-based studies would be sufficient. Although k-abe was developed under emergency conditions for a devastating neonatal-onset metabolic disorder, we anticipate that rapid deployment of patient-specific gene-editing therapies will become routine for many genetic diseases.