Genetic Correction of the Most Common Mutation Causing Primary Hyperoxaluria Restores Enzyme Localization and Oxalate Metabolism

Primary hyperoxalurias are autosomal recessive inborn errors of metabolism that lead to the accumulation of oxalate in various tissues. Primary hyperoxaluria type 1 (PH1, [MIM: 259900]) is the most common of primary hyperoxalurias with an incidence rate of 1 in 120000 live births in Europe [1]. PH1 results from pathogenic mutations in AGXT (MIM: 604285), which encodes liver‐specific peroxisomal alanine‐glyoxylate aminotransferase (AGT, [EC 2.6.1.44]). The AGT monomer forms a pyridoxal‐phosphate‐dependent homodimer, which catalyzes the transamination of glyoxylate to glycine in the liver hepatocytes [2] (Figure 1A). In PH1, the lack of catalytically active AGT in liver peroxisomes leads to the accumulation of glyoxylate that is converted into oxalate by lactate dehydrogenase (EC 1.1.1.27) [3]. The accumulating oxalate reacts with calcium and forms insoluble calcium oxalate crystals in the renal tubules. As the renal damage decreases the filtration rate, the kidneys cannot excrete the accumulating oxalate, which eventually saturates in the plasma and leads to systemic oxalosis [1, 4]. Until recently, PH1 treatment has comprised intensive hyperhydration, crystallization inhibitors, and pyridoxine supplementation. After kidney failure, a combined or sequential liver and kidney transplantation is the preferred treatment [5]. Novel therapeutical approaches disrupt the oxalate metabolism by gene‐targeted knockdown [6].

The most common pathogenic AGXT variant (c.508G>A [GenBank: NM_000030.3]; [Gly170Arg]; rs121908529) is found in 24%–37% of PH1 patients, is associated with significant catalytic activity, and leads to peroxisome to mitochondrion mistargeting [7, 8]. The variant is pathogenic when it is associated with the minor allele (AGT‐Mi) haplotype (c.32C>T [GenBank: NM_000030.3]; [p.Pro11Leu]; rs34116584). As a monomer, the AGT‐Mi N‐terminus adopts a conformation of an amphiphilic α‐helix able to act as a mitochondrial targeting sequence [9, 10, 11]. The healthy AGT dimerizes rapidly and irreversibly after synthesis, making it incompatible with mitochondrial import [12]. The Gly170Arg variant potentially acts synergistically with AGT‐Mi slowing down dimerization, thus keeping the mitochondrial targeting sequence unmasked [7, 13]. However, either Gly170Arg or AGT‐Mi variants alone do not promote significant enzyme mistargeting [12, 14].

CRISPR‐Cas9 base editors are precise genome editing tools that can be directed to correct transition changes in a narrow genomic locus. The adenine and cytosine base editors have collectively the potential to reverse most pathogenic point mutations [15]. Adenine base editors [16] (ABE) catalyze a base pair transition from A‐T to G‐C in the activity window and cleave the non‐edited DNA strand without leading to double‐stranded breaks, thus reducing the risk of genomic rearrangement, insertions, and deletions [17]. The base editors are being tested for their therapeutic potential in several clinical trials (ClinicalTrials.gov↗: NCT05398029, NCT05885464, and NCT05456880). Lipid nanoparticles (LNPs) facilitate mRNA transport into target tissues with low immunogenic effects compared to alternative delivery methods, such as viral vectors [18]. LNPs can be directed to different tissues by adjusting the LNP composition [19], but by default, they accumulate in the liver [20].

We hypothesized that AGXT mutations could be corrected by gene editing and proceeded to model PH1 with a genetic background of the Gly170Arg‐ and AGT‐Mi variants in vitro. Additionally, we tested the potential of base editors to correct the pathogenic mutation in differentiated hepatocyte‐like cells by an in vivo‐compatible LNP delivery.

We aimed to model PH1 in vitro, specifically the effects of the Gly170Arg variant, and to correct the cell phenotype using base editors. We started by culturing fibroblasts from a skin biopsy of a compound heterozygous subject with variants c.508G>A (Gly170Arg) with minor allele c.32C>T (Pro11Leu), and c.673_676delAAGG. Next, we genetically corrected the Gly170Arg variant by electroporating the fibroblasts with in vitro transcribed ABEmax [30] mRNA and the Gly170Arg variant targeting single guide RNA. Sanger sequencing analysis of the editing outcomes with the EditR software [24] revealed that the ABEmax produced an average on‐target A‐T to G‐C editing efficiency of 98% without any unspecific editing along the protospacer sequence, validating the predicted potential of ABE to correct this variant (Figure 1B,C). To have a deeper insight into the editing outcomes, we sequenced the on‐target amplicons with Oxford Nanopore Technology (ONT) sequencing and analyzed the reads with CRISPResso2 software [25]. After editing, 86.7% of the reads that aligned to the reference amplicon showed a G base in position 6 encoding the healthy Gly170 variant without additional bystander edits in the protospacer, 6.1% of the reads showed an A base encoding the pathogenic Arg170 variant without bystanders, 5.6% a deletion, and 1.5% the healthy Gly170 with synonymous bystander edits that do not lead to additional amino acid changes (Figure 1D). To assess potential RNA‐guide‐directed off‐target base editing across the genome in the edited cells, we ONT‐sequenced amplicons from the top six off‐target loci (Table 1) predicted by three independent off‐target tools. The mean A>G, C>T, or total transition substitution rates across all sites were not higher for the edited samples in comparison to the non‐treated samples (Figure 1E). Additionally, we studied the on‐ and off‐target editing profile of a more processive ABE variant ABE8e [31] (Figure S1A–C), which produced equally high on‐target editing, but also significant editing in the nearby bystander bases.

Next, we generated corrected iPSC lines by simultaneously reprogramming and editing the patient fibroblasts with ABEmax, as previously described [21]. To assess the quality of the corrected cell lines and to screen out lines with potential off‐target edits, we sequenced 13 top off‐target loci (Table 1) predicted by three independent software programs. The Sanger sequencing data from these sites showed no A‐T to G‐C editing above a 1% threshold in the 54 adenines in the 13 potential protospacer regions where the guide could theoretically bind (Table S2). All patient‐derived iPSC lines expressed canonical pluripotency markers OCT4, NANOG, TRA1‐60, and SOX2 in immunostainings (Figure S1D).

To model the disease phenotype and the effects of ABE‐mediated Gly170Arg editing on the phenotype in vitro, we differentiated the iPSC lines into hepatocyte‐like cells using established protocols [26, 32] as shown previously [22]. At the end of the 20‐day differentiation protocol, the differentiated lines showed a robust expression of hepatic marker mRNA, such as ALB, HNF1A, AFP, and APOA2 in comparison to the iPSC stage (Figures 1F and S2). The hepatic marker expression in the individual cell lines was comparable to the HepG2 hepatocellular carcinoma line, but did not reach the level of primary hepatocytes in ALB expression. Pluripotency marker SOX2 expression decreased in the differentiated lines compared to the iPSC stage. AGXT expression increased significantly in all differentiated lines from the iPSC stage. The patient‐derived lines carrying one copy of the 673_676delAAGG allele showed approximately half of the AGXT expression levels of the healthy control line with two wild‐type AGXT alleles. All hepatocyte‐like cell lines showed an accumulation of hepatic markers AGT, HNF4A, and AFP in immunostainings (Figures 1G, 2A,B, and S2).

As the Gly170Arg variant has been shown to lead to AGT enzyme mislocalization, we determined the enzyme localization both in the non‐corrected and corrected hepatocyte‐like cell lines. After co‐staining the cells with anti‐mitochondrial and anti‐AGT labels, we observed AGT in non‐corrected cell lines colocalizing with mitochondria, whereas the corrected and control lines showed a different pattern and no co‐localization with the mitochondrial marker (Figures 2A and S2). Conversely, the immunostaining with anti‐peroxisomal and anti‐AGT labels showed co‐localization in the corrected and control lines, while the non‐corrected lines showed a different pattern (Figures 2B and S2). Quantitative analysis confirmed the pattern of mitochondrial enzyme localization in the non‐corrected lines and peroxisomal enzyme localization in the corrected and control lines (Figure 2C,D).

To determine the metabolic state in the iPSC‐derived hepatocyte‐like cells, we used liquid chromatography‐mass spectrometry to evaluate the levels of the disease‐specific metabolites oxalate and pyridoxine in the cells and culture media after a 48‐h incubation (Figure 3A–D). The fresh cell culture media contained pyridoxine, which is a precursor of a co‐factor for AGT and significantly improves the function of the Gly170Arg variant [33]. To challenge the function of AGT, we reduced the concentration of pyridoxine by 50% and added glyoxylate precursor ethylene glycol [34] in the final media. Oxalate, the main diagnostic metabolite in PH1, was significantly elevated in the non‐corrected cell lines in comparison to the corrected and healthy control lines in both intracellular (Figure 3A) and media (Figure 3B) samples. The oxalate level in the corrected cell line remained moderately higher than in the control line both intracellularly and in the media. The non‐corrected lines had the highest, the corrected line the second highest, and the control line the lowest pyridoxine concentration in both the intracellular (Figure 3C) and the media (Figure 3D) samples after a 48‐h incubation.

After showing the efficient ABE‐mediated genetic correction of the AGXT Gly170Arg variant and the resulting restoration of the cell phenotype, we wanted to model potential hepatocyte editing in our in vitro model. We decided to employ a highly efficient ABE variant ABE8e [31] as mRNA and our Gly170Arg variant targeting single guide RNA encapsulated in LNP formulations using a strategy that we have optimized earlier to edit fibroblasts [22]. We added the formulations to the hepatocyte‐like cell culture media for 24 h on day 19 of the differentiation. In this reaction, we observed dose‐dependent on‐target editing reaching an average of 5% with the highest dose of 2500 ng of total RNA used (Figure 3E).

In this study, we analyzed the effects of the most common PH1‐causing pathogenic AGXT variant using a relevant patient‐derived hepatocyte‐like cell model. We studied the effects of the precise gene correction on the cell phenotype, focusing specifically on enzyme localization and metabolism. In addition, we assessed the potential of base editors to correct this pathogenic variant with both in vitro‐ and in vivo–compatible delivery methods.

The achieved editing results corroborate the potential of ABE to correct the targeted variant efficiently and precisely. Utilizing electroporation and in vitro‐transcribed ABEmax mRNA, we reached a high on‐target mean editing efficiency of 98% without any proximal bystander editing in the protospacer region, confirmed by EditR analysis of the Sanger sequencing data. The ABE8e variant performed almost as efficiently as ABEmax on the target mutation, but produced more bystander edits. ONT sequencing analysis confirmed the successful on‐target editing, showing that the vast majority of the reads encoded for the healthy allele after the editing reaction. Analysis of the top in silico predicted off‐target loci showed no off‐target substitutions above the baseline of the non‐treated samples on average. In the edited iPSC clones, the in silico predicted off‐target sites 3 and 11 showed 1% editing in one position along the guide. We estimate that this editing is a background in the Sanger sequencing data. As the off‐target analyzed corrected iPSC lines were monoclonal in origin, a true off‐target editing event would show around 50% or 100% editing, respectively, if one or both alleles were edited. To model potential in vivo editing of the liver hepatocytes, we used the more processive ABE8e [31] mRNA encapsulated within LNPs to target the differentiating hepatocyte‐like cells. In this experiment, we reached a maximum of 5% mean on‐target editing without any bystander base changes. LNPs have been shown to preferentially accumulate in the liver and transfect hepatocytes through low‐density lipoprotein receptor (LDL‐R)‐mediated endocytosis [20]. However, our cell model does not fully replicate the authentic in vivo environment, including continuous perfusion of the liver and circulating apolipoprotein E, which facilitates LNP delivery to hepatocytes via LDL‐R [35], and fully mature primary hepatocytes presenting LDL‐R for LNP binding. Previous preclinical studies have shown high LNP‐mediated in vivo CRISPR editing of the liver in murine and non‐human primate models [36, 37, 38]. Additionally, a recent report already established amelioration of a severe metabolic phenotype in a carbamoyl‐phosphate synthetase 1 deficiency patient through LNP‐mediated base editing of the liver [39]. Hence, we assume that our editing system would yield substantially higher editing in vivo than we reached with the hepatocyte‐like cell model. This hypothesis could be tested in animal models. However, even modest editing efficiency establishes proof‐of‐concept that grants further investigation.

In this study, we showed that the correction of the Gly170Arg variant alone in the genetic background of AGT‐Mi leads to the rescue of AGT localization from mitochondria to peroxisomes. Since there is strong evidence that the root cause of pathogenesis caused by the variant is due to enzyme mislocalization [14], the genetically corrected cell line was expected to exhibit a rescue of the metabolic pattern. Indeed, the corrected cell line showed a significant decrease in oxalate compared to the non‐corrected counterparts, although not fully matching the level of the control. Several factors could explain this observation. First, the healthy control line has two fully functional wild‐type alleles, whereas the genetically corrected lines have one functional corrected allele and the c.673_676delAAGG variant, which results in a premature termination codon. qPCR of the hepatocyte‐like cells showed around 50% reduction of AGXT mRNA levels in the patient‐derived cell lines in comparison to the healthy control line (Figure 1E), which suggests that the deletion‐containing transcript might be subjected to nonsense‐mediated decay. This dose‐dependent reduction in enzyme activity could explain the difference in oxalate accumulation between the corrected and control lines. In addition, the cell model might not be robust enough to replicate the full metabolic pathway with all relevant enzymes fully present. The variation in differentiation and the resulting maturity of cells might lead to artificial differences in enzyme expressions and metabolite accumulation between cell lines and differentiation batches. Finally, our differentiation medium contains a significant pyridoxine supplement (Figure 3C), which could not be completely removed without an apparent decrease in cell viability. Pyridoxine, a precursor to the AGT co‐enzyme pyridoxal‐5′‐phosphate (PLP), has been shown to significantly improve the function of the 170Arg variant [33, 40], and the supplementation might lead to a decreased difference in the accumulation of oxalate between the 170Arg and 170Gly variants. Pyridoxine was supplemented in media in supraphysiological concentrations; normally, it is not measurable in human plasma [41, 42]. However, we cannot directly deduce the level of the active form, membrane‐impermeable PLP [43], based on the measured amount of pyridoxine in the differentiated cells and media. Taken together, rescuing AGT peroxisomal localization by gene editing ameliorates oxalate metabolism.

For PH1 management, hyperhydration and crystallization inhibitors are used as supportive treatments [5, 44]. Additionally, pyridoxine can be supplemented to responsive subjects who have mistargeting variants. Until recently, the only curative treatment has been liver transplantation or combined liver and kidney transplantation. Two recently developed novel therapeutics are based on RNA interference (RNAi) of mRNA encoding enzymes in the glyoxylate pathway. Lumasiran targets HAO1 (MIM: 605023) encoding hydroxyacid oxidase (EC 1.1.3.15) and leads to a high accumulation of soluble glycolate and depletion of oxalate, and Nedosiran targets LDHA (MIM: 150000) encoding lactate dehydrogenase and prevents the final conversion of glyoxylate to oxalate in the hepatocyte cytosol (Figure 1A) [6]. Recent preclinical work targeting HAO1 [45, 46, 47], and LDHA [48] with adeno‐associated virus (AAV) or LNP‐delivered CRISPR‐Cas9 in PH1 mouse and rat models presented a significant decrease in urinary oxalate accumulation. The CRISPR gene knockout strategy could deliver a permanent amelioration of the toxic oxalate accumulation with a single dose. However, potential off‐target effects of double‐stranded DNA breaks inducing CRISPR/Cas9 nucleases need to be carefully assessed for therapeutic applications, and the long‐term effects of elevated glycolate or glyoxylate need to be monitored. Strategies to treat PH1 through delivery of the human AGXT complementary DNA with viral vectors have been effective in restoring the metabolic phenotype in a mouse model [33] and AGT expression in a cell model of PH1 [49]. Recently, a human AGT mRNA‐Lipopolyplex therapy recovered functional enzyme expression and significantly reduced urinary oxalate in a rat model of PH1 [50]. As the in vitro differentiation protocols improve, autologous transplantations of genetically corrected iPSC‐derived hepatocyte‐like cells could provide a treatment for PH1 [49, 51]. The function and safety of these in vitro cultured cells need to be carefully assessed before reintroduction to patients; however, methods are being developed to make the cells safe for transplantation [52]. Our strategy to treat PH1 is based on CRISPR base editor mRNA and single guide RNA encapsulated within LNPs for liver targeting. This base editing approach corrects the mutant variant into a healthy one, a change that would be permanent and inherited by the daughter cells of proliferating liver hepatocytes. Furthermore, gene correction with base editing allows physiological regulation of expression, an advantage compared to the gene addition approaches. Unlike with the RNAi or mRNA replacement strategies, continuous redosing would not be necessary. However, compared to the viral vectors, LNPs are potentially less immunogenic, and redosing could be applied to target a higher proportion of edited hepatocytes. HAO1‐ and LDHA‐targeting, and AGXT gene or mRNA addition therapies could theoretically be applied to all PH1 subjects regardless of the AGXT variant they carry, whereas only 24%–37% of PH1 subjects carry the Gly170Arg variant [6]. Nevertheless, the base editing strategy could potentially be applied to most of the PH1 mutations. Data retrieved from the gnomAD database show that out of all AGXT variant alleles, which are annotated "Pathogenic", "Likely pathogenic" or "Pathogenic/Likely pathogenic", 74% are transition variant alleles that have the potential to be corrected with either ABE (69% of all alleles) or cytosine base editor (5% of all alleles) (Table 2). With just minor changes, it could be possible to target transition variant alleles other than Gly170Arg, switching only the single guide RNA, and when necessary, the type of base editor used. Additionally, the AGT‐Mi remains an alternative target for ABE. Several pathogenic variants associated with the mistargeting of AGT, including Gly170Arg, Ile244Thr, Phe152Ile, and Gly41Arg, are pathogenic only in the genetic background of the AGT‐Mi [53]. Thus, editing the AGT‐Mi into the major allele could rescue the phenotype of these four pathogenic variant alleles, collectively representing 58% of all pathogenic variant alleles, and the same genetic treatment could help most of the PH1 patients.

Although less immunogenic than viral vectors, base editors pose a risk of an immunogenic effect. In humans, there are preexisting humoral and adaptive immune responses to SpCas9 [54, 55], which is the main component of base editors, including the ABEmax and ABE8e variants used in this study. Off‐target DNA and RNA editing effects remain the other main risk factor with the base editors. The ABE poses an increased risk of RNA editing, since the deaminase component of ABE is evolved from tRNA adenosine deaminase [15]. In this preclinical study, we did not detect clear off‐target editing events in the top loci that are the most likely unspecific binding sites for the guide RNA. However, for clinical applications, the screening for potential off‐target editing sites would have to be extensive and genome‐wide.

Primary hyperoxaluria type 1 is a severe disease often leading to liver and kidney transplantation, a procedure with risks and requiring life‐long immunosuppressive medication. The development of gene therapy options can offer curative treatment for the disease in the future. Here, in this preclinical proof‐of‐concept study, we present a promising gene editing approach to correct AGXT mutations and successfully validate it in vitro.

Conceptualization: T.K, K.W., and M.E.H. Methodology: T.K., S.J., J.J., and E.K. Investigation: T.K., S.J, I.G., J.J., and N.K. Analysis and interpretation of data: T.K., S.J., I.G., J.J., N.K., E.K., V.H., D.B., K.W., and M.E.H; Visualization: T.K. Writing – original draft: T.K. and M.E.H. Writing – review and editing: T.K, S.J., I.G., J.J., N.K., E.K., V.H., D.B., K.W., and M.E.H. Resources: V.H., D.B., K.W., and M.E.H. Supervision: K.W. and M.E.H. Project administration: K.W., and M.E.H. Funding acquisition: K.W. and M.E.H.

The work was funded by Helsinki University Hospital Research Funds, the Academy of Finland Centre of Excellence on Stem Cell Metabolism, the Foundation for Pediatric Research, the Paulo Foundation, the Magnus Ehrnrooth Foundation, the Research Council of Finland (grant 361593), the Novo Nordisk Foundation (NNF24OC0089232), and the Orion Research Foundation sr.

The generation of fibroblast and hiPSC lines from skin biopsies was approved by the Coordinating Ethics Committee of the Helsinki and Uusimaa Hospital District upon informed consent of the donor or their guardians (diary no.: HUS/2754/2019). The study was conducted in accordance with the Declaration of Helsinki.

Informed consent was obtained from the guardians of the subject involved in this study.

The authors declare no conflicts of interest.