IntroductionCell and gene therapies have rapidly advanced as transformative modalities for inherited disorders that were once considered incurable1,2. In vivo gene therapy using adeno-associated virus (AAV) vectors has achieved durable gene delivery and functional benefit across several genetic diseases3, while ex vivo approaches have traditionally centered on hematopoietic stem cells owing to their robust regenerative capacity and the feasibility of autologous transplantation following gene correction or transgene integration4. The extension of ex vivo strategies beyond hematopoietic disorders to tissue-specific genetic diseases is exemplified by the recent US Food and Drug Association (FDA) approvals of Zevaskyn (Prademagene Zamikeracel), an autologous epidermal cell therapy delivering intact COL7A1 for recessive dystrophic epidermolysis bullosa, and Encelto (Revakinagene Taroretcel), an allogeneic retinal pigment epithelial cell therapy secreting recombinant human ciliary neurotrophic factor for idiopathic macular telangiectasia type 2 (refs. 5,6). The advent of safe and efficient gene editing of patient-derived induced pluripotent stem (iPS) cells not only facilitates the generation of isogenic disease models but also provides a renewable source of autologous cells for therapeutic transplantation7,8.Duchenne muscular dystrophy (DMD), an X-linked disorder caused by mutations in the gene DMD that abolish dystrophin expression, is characterized by progressive muscle degeneration, loss of ambulation and premature mortality9. Although FDA-approved antisense oligonucleotide therapies targeting exon skipping are available, they achieve only limited dystrophin restoration; accordingly, cell-based strategies using patient-derived iPS cell-based myogenic progenitors have been extensively investigated as alternative therapeutic approaches10. Established directed differentiation protocols can generate skeletal myogenic cells from human pluripotent stem cells11,12, and complementary teratoma-based approaches have shown that in vivo differentiation can produce functional myogenic progenitors from mouse and human pluripotent stem cells13,14. Clinical-grade production of iPS cell-derived myogenic progenitors has recently been achieved, yielding the product MyoPAXon, which aims to provide an off-the-shelf cell therapy for muscular dystrophies and has successfully been used to restore dystrophin expression in dystrophic mouse and nonhuman primate models15. Beyond these approaches, in vivo AAV-based delivery of mini-dystrophin16 and precise genome-editing strategies—including exon skipping17,18, base editing19,20 and prime editing21—have expanded the therapeutic toolbox for correcting the wide spectrum of pathogenic variants observed in patients with DMD22.Despite the promise of these therapies, their translational evaluation remains constrained. The mdx mouse model, which carries a nonsense mutation in the Dmd gene23, and its subsequent variants24 have been widely used for both mechanistic studies and translational research. However, murine Dmd differs substantially from its human counterpart in sequence, exon–intron organization and splicing regulation, and the phenotypes resulting from its alteration are relatively mild. Consequently, these models are inadequate for directly modeling many patient-specific mutations24. To address this limitation, diverse humanized DMD mouse models have been generated by replacing murine exons with human sequences, thereby allowing genome-editing tools designed for human mutations to be tested in vivo20,25,26,27. While these models provide proof of concept for editing at authentic human sites, each line corresponds to a specific mutation and thus captures only a narrow fraction of the mutational spectrum. Moreover, generating individual humanized mouse models for each clinically relevant mutation is a highly laborious and time-consuming process, limiting their scalability and translational relevance.Here, we leveraged teratomas derived from the iPS cells of patients with DMD (DMD-teratomas), which recapitulate the absence of dystrophin expression, as a patient-derived humanized preclinical model. Using this platform, we demonstrate that both autologous ex vivo therapy with gene-corrected iPS cell-derived myogenic progenitors and in vivo adenine base editor (ABE) mRNA delivery via lipid nanoparticles (LNPs) can restore dystrophin expression within the teratoma. These findings establish DMD-teratomas as a versatile and mutation-inclusive system for evaluating and comparing cell- and gene-based therapeutic strategies for DMD.Material and methodsCell culture and iPS cell transfection/electroporationhiPS cell lines (NM004006.2 c.433 C>T DMD iPS cell; National Institutes of Health) were maintained in iPS cell-brew MACS (Miltenyi Biotec, #130-104-368) with 0.1% gentamycin (Gibco, #15750-060) on a Matrigel (Corning, #354277)-coated cell culture dish at 37 °C and humidified to 5% in a CO2 incubator. On transfer, hiPS cells were rinsed with Dulbecco’s phosphate-buffered saline and detached enzymatically with Dispase (Life Technologies). Y27632 (10 μM; PeproTech #1293823) was used for 1 day after the attachment if needed. According to the standardized protocol for electroporation of human pluripotent stem cells28, gene delivery with the electroporator (NEPA-21, NEPAGENE), iPS cells were rinsed with DPBS and detached with Accutase (#561527, BD Bios-ciences). Cells were resuspended and diluted to 1 × 106 cells per 100 μl with Opti-MEM (#31985070, Gibco) after three rounds of Opti-MEM washing (#31985070, Gibco). A total of 3 μg of the singel guide (sg)RNA or prime editing guide (peg)RNA vector and 2 μg of the ABE vectors were added to the cell solution. The electroporation was performed with NEPA-21 electroporator with 175 V and 2.5 ms of poring pulse as described. Dissociated cells after electroporation were plated into a culture dish with 10 μM of Y27632.Teratoma formationHuman iPS (hiPS) cells were collected at ~70–80% confluency using Dispase (Life Technologies), washed and resuspended in iPS cell-brew MACS. Approximately 1 × 106 iPS cells were injected into the testis of 4-week-old Balb-c/nu mice under isoflurane anesthesia. For experiments involving gene-edited cells, both mutant iPS (M-iPS) cells and edited iPS (E-iPS) cells were transplanted in parallel. Mice were monitored weekly for tumor growth. After 6–12 weeks, teratomas were excised, fixed in 4% paraformaldehyde, dehydrated, embedded in paraffin and sectioned at 7 μm. Hematoxylin and eosin staining was performed to confirm the presence of derivatives of all three germ layers. For molecular analyses, fresh teratoma tissue was snap-frozen in liquid nitrogen for protein or nucleic acid extraction. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Seoul National University (institutional review board (IRB) no. 2310/001-027) and were performed in accordance with institutional and national guidelines for animal welfare.CM differentiationDMD-iPS cells were maintained on matrigel-coated plates in iPS cell-brew MACS medium and differentiated into CMs following a chemically defined monolayer protocol. In brief, cells at ~90% confluence were treated with 2–3 µM CHIR99021 (Biogems) for 48 h to activate Wnt signaling, followed by 1–2 µM Wnt C59 (Selleckchem) for 48 h to promote cardiac specification. The culture medium (Roswell Park Memorial Institute medium (RPMI) 1640 supplemented with B27 minus insulin) was subsequently replaced every 2 days until spontaneous contractile CMs appeared at day 10–12. Beating clusters were metabolically purified using lactate-based selection, and the resulting CMs were maintained in RPMI 1640/B27 (Gibco) medium for downstream analyses.Myotube differentiationBoth M-iPS and E-iPS cells were dissociated into single cells and seeded on Matrigel-coated plates, followed by induction of paraxial mesoderm using 3 µM CHIR99021 and 0.5 µM LDN193189 (Stemgent) in serum-free DMEM/F12 medium supplemented with ITS and FGF2 (Peprotech). Cultures were maintained for 30–40 days with sequential media changes to promote myogenic commitment and fusion. After subculturing in Skeletal Muscle Growth Medium (Lonza), secondary differentiation was induced by switching to N2- or horse serum-based medium up to day 90, leading to the formation of multinucleated, striated myotubes expressing MyHC, MYOG and dystrophin. The resulting cultures contained both mature myofibers and Pax7+ satellite-like progenitors suitable for downstream analysis.Myo-progenitor engraftmentUsing the above differentiation protocol, E-iPS cells were differentiated up to day 12 to obtain myo-progenitor-stage cells. The resulting myo-progenitors were dissociated with TrypLE Express, and approximately 1 × 105 cells were injected into M-teratomas using a 23-gauge needle. Following transplantation, teratoma-bearing mice were maintained for ~4 weeks to allow engraftment and proliferation of the transplanted cells. Subsequently, teratomas were surgically excised and collected for downstream analyses.Teratoma transplantationTeratomas were excised and washed thoroughly in sterile PBS, then cut into approximately 1-cm3 pieces. Each fragment was transplanted subcutaneously into BALB/c nude (OrientBio) mice under sterile conditions. Following implantation, the surgical sites were closed appropriately to ensure recovery and prevent infection.IVT mRNA synthesisDNA templates for in vitro transcription (IVT) were generated from the plasmid pCMV-SpRY-ABE8e (Addgene, #185671) by PCR amplification using PrimeSTAR HS DNA Polymerase (Takara), followed by linearization. A 120-nucleotide poly A-tail was encoded in the template using a primer containing a poly T sequence during PCR. The PCR products were purified with the AccuPrep PCR/Gel Purification Kit (Bioneer) and verified by electrophoresis on a 1% agarose gel in 1× TBE buffer. smRNA was synthesized using the MEGAscript T7 Transcription Kit (Thermo Fisher Scientific), with CleanCap AG (TriLink) included in the reaction and UTP substituted with N1-methyl-pseudouridine-5′-triphosphate (N1-Me-ΨUTP). Transcription reactions were carried out at 37 °C for 2 h, followed by DNase treatment. The resulting mRNA was purified using the HiGene DNA/RNA Purification Mini Column (Biofact). Transcript integrity was confirmed by electrophoresis on a 1% denaturing agarose gel in 1× 3-(N-morpholino)propanesulfonic acid (MOPS) buffer.Preparation and characterization of LNPsLNPs were prepared using a microfluidic mixing system (Ignite, Precision NanoSystems). A lipid mixture consisting of a proprietary ionizable lipid (undisclosed structure), 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), cholesterol and a polyethylene glycol (PEG)-lipid was dissolved in ethanol according to a defined molar ratio. The ethanol phase was prepared by mixing lipids in ethanol (97.5% v/v) with 10 mM citrate buffer, pH 3.0 (2.5% v/v). ABE mRNA and sgRNA were diluted in an aqueous phase composed of 1× PBS and 10 mM citrate buffer (v/v of 2:1). LNPs were formulated by microfluidic mixing at a total flow rate of 12 ml min−1, with an ethanol-to-aqueous phase ratio of 1:3. The resulting LNPs were diluted in 40 volumes of 1× PBS and concentrated by ultrafiltration using Amicon Ultra-15 centrifugal filters, 10 kDa MWCO (Millipore). The mean particle size and polydispersity index (PDI) of LNPs were determined using a Zetasizer Lab Blue (Malvern Instruments) following dilution to 0.5 μg ml−1 in PBS. RNA encapsulation efficiency was determined with the Quant-iT RiboGreen Assay (Thermo Fisher Scientific), and fluorescence was measured using a SpectraMax iD3 microplate reader (Molecular Devices).NGS sequencingAs previously described29, genomic DNA was isolated from edited and control samples, and target as well as predicted off-target loci were amplified by PCR using locus-specific primers. Amplicon libraries were subjected to paired-end sequencing (2 × 150 base pairs (bp)) on an Illumina MiniSeq platform. Sequencing data were analyzed by alignment to the reference amplicon sequence, and editing frequencies were determined from the proportion of variant reads at the target nucleotide position. Only variants passing predefined thresholds for read depth, supporting read count and variant allele frequency were included in downstream analyses. Background variants detected in untreated controls were excluded. Editing outcomes were evaluated primarily within the expected ABE editing window, while variants located in primer regions, read ends, repetitive sequences or homopolymeric regions were excluded as potential technical artifacts.scRNA sequencingSingle-cell transcriptomic profiling was performed using the 10x Genomics Chromium GEM-X Single Cell 3′ RNA-seq workflow. In brief, dissociated single-cell suspensions were loaded onto the Chromium controller to generate gel beads-in-emulsion (GEMs), in which individual cells were co-encapsulated with barcoded gel beads for cell-specific 10x barcodes and unique molecular identifiers (UMIs). Reverse transcription and cDNA amplification were carried out according to the Chromium Single Cell 3′ v4 (GEM-X) protocol to generate 3′ gene expression libraries, which were subsequently sequenced on an Illumina platform using paired-end reads (read 1: 28-bp cell barcode/UMI; read 2: 90-bp transcript insert; dual 10-bp i7/i5 sample indexes). Base call files (BCL) were converted to FASTQ with BCL Convert, and libraries were processed using Cell Ranger (v9.0.1) for alignment to the GRCh38 reference (2024-A), UMI counting, cell barcode filtering and generation of gene-by-cell count matrices. Cell Ranger ‘count’ was further used to perform dimensionality reduction (principal component analysis), clustering and gene expression quantification for downstream analyses.Primary data processingRaw sequencing BCL files were demultiplexed into FASTQ files using the cellranger mkfastq pipeline. The resulting FASTQ files were processed with Cell Ranger (10x Genomics, v9.0.1) count to align reads to a GRCh38 reference genome and generate a feature-barcode count matrix. This matrix, containing UMI counts per gene for each cell, served as the input for all subsequent analyses. All processing was performed using a consistent version of the Cell Ranger software to ensure reproducibility.Single-cell RNA data analysisSingle-cell RNA sequencing (RNA-seq) datasets from samples were processed using Suerat v5.2.1. Cells with fewer than 1000 UMIs, fewer than 1000 detected genes, more than 20% mitochondrial transcripts or more than 10% ribosomal protein-coding transcripts were removed. Normalization and variance stabilization were performed using an SCTransform-based approach, and the datasets were integrated after selecting 3000 highly variable features. Batch effects were corrected during the integration step. The integrated data were dimensionally reduced with principal component with principal component analysis, and the first 18 principal components were used for constructing a shared nearest-neighbor graph and subsequent clustering. Low-dimensional visualization was performed with UMAP to explore cellular heterogeneity. Cell clusters were annotated into 20 cell types representing the three germ layers on the basis of the marker genes and their match to canonical markers.Statistical analysisThe graphical and quantitative data were presented as mean ± s.d. Statistical significance among the three groups and between groups was determined using one-way or two-way analysis of variance following Tukey’s post hoc test and Student’s t-test, respectively. Statistical analysis was performed with GraphPad Prism 10 software (https://www.graphpad.com/scientific-software/prism/). Significance was assumed for P