CRISPR-engineered human lung organoids with a biomolecular condensate reporter enable mechanistic toxicity monitoring

Abstract

Understanding how chemical stress perturbs human lung physiology requires models that capture dynamic molecular responses in real time. Here, we established a CRISPR/Cas9-engineered human induced pluripotent stem cell (hiPSC)-derived lung organoid expressing endogenous G3BP1–mCherry, enabling live, non-destructive visualization of stress granule (SG) formation under toxicant exposure. The organoids recapitulated airway and alveolar epithelial diversity and displayed lamellar body-like ultrastructures, indicating advanced maturation. Time-lapse imaging revealed rapid and reversible SG dynamics across chemically distinct stressors, while cytotoxicity assays showed that these organoids are significantly more sensitive than conventional 2D or cancer-derived lung models. Importantly, SG dynamics were linked to exposure duration–dependent changes in epithelial barrier integrity, indicating that SG formation precedes overt epithelial injury and serves as an early indicator of toxicant-induced cellular stress. Integration with high-content screening enabled quantitative, image-based analysis of cellular stress phenotypes, greatly enhancing throughput and mechanistic insight, thereby provided next-generation New Approach Methodologies for lung toxicity assessment. Together, this hiPSC-derived lung organoid SG reporter platform links early molecular stress adaptation to tissue-level responses, offering a predictive and mechanistically informative framework for human-relevant lung toxicity evaluation.

Graphical abstract

Highlights

• •CRISPR-engineered human lung organoids express endogenous G3BP1–mCherry reporter.
• •Live imaging enables rapid and reversible visualization of stress granules.
• •Stress granule dynamics act as an early and sensitive biomarker of chemical stress.
• •Lung organoids show higher chemical sensitivity than immortalized and cancer cell models.
• •Reporter enables non-destructive, real-time monitoring of lung toxicant responses.

1. Introduction

Household and industrial agents, including disinfectants, surfactants, and preservatives are extensively used in daily life, resulting in continuous human exposure through air, water, and various consumer products. Such exposures have been implicated in severe pulmonary disorders characterized by epithelial injury, fibrosis, and chronic respiratory dysfunction [1,2]. The nationwide outbreak of humidifier disinfectant-related lung disease in Korea, caused by polyhexamethylene guanidine (PHMG)-containing products, tragically demonstrated that prolonged inhalation of household chemicals can induce acute respiratory failure and progressive fibrotic damage [3,4]. The global outbreak of COVID-19 has recently driven a surge in the use of disinfectant sprays and hand sanitizers, substantially increasing indoor concentrations of biocidal compounds such as quaternary ammonium derivatives that are associated with respiratory and systemic health risks [5,6]. These observations emphasize the urgent demand for human-relevant in vitro platforms that can predict lung toxicity before irreversible damage occurs.

Conventional lung toxicity assessments primarily measure late-stage outcomes such as reactive oxygen species generation, apoptosis, or loss of cell viability [7,8]. Although these endpoints provide useful information, they predominantly reflect irreversible cellular injury and have limited utility in detecting early stress responses. Moreover, most assays yield static measurements, often failing to capture transient or reversible molecular changes that precede overt cytotoxicity. Addressing this gap requires improved in vitro systems capable of detecting early perturbations that serve as sensitive indicators of toxicity.

Stress granules (SGs) have emerged as early markers of cellular stress, reflecting translational arrest and activation of adaptive pathways [9]. SGs are transient cytoplasmic assemblies of RNA and proteins that form when cells encounter stress conditions, temporarily suppressing global protein translation and protecting untranslated mRNAs to support cellular adaptation and recovery. However, when stress is prolonged or unresolved, cellular damage can accumulate and may progress toward cell injury or death. SGs are dynamic, non-membranous ribonucleoprotein condensates that assemble via liquid–liquid phase separation in the cytoplasm in response to translational inhibition triggered by oxidative stress, chemical insults, or viral infection, contributing to cellular stress adaptation [10,11]. SGs primarily consist of untranslated mRNAs complexed with various RNA-binding proteins, including Ras-GTPase-activating protein-binding protein 1 (G3BP1), TIA-1, and PABP, along with translation initiation factors such as eIF3 and eIF4E [12]. The interplay between these proteins and RNA molecules forms a dynamic network that drives condensate formation and shapes SG composition in a stress- and context-dependent manner. Their rapid and reversible assembly acts as an adaptive defense mechanism to preserve mRNA integrity and maintain cellular homeostasis. As diverse environmental stressors, including heavy metals, heat shock, and chemical toxicants, trigger SG formation [13,14], they represent a sensitive and broadly applicable biomarker of early cellular stress. Among SG components, G3BP1 is a key nucleator of granule assembly and a canonical SG marker [15,16]. Fluorescent tagging of G3BP1 enables visualization of SG dynamics in living cells, providing a non-destructive readout of cellular stress responses [17,18].

Conventional toxicity assessment systems have predominantly relied on immortalized cell lines, which possess inherent structural and functional limitations in modeling the physiological characteristics of normal human tissue [19]. Two-dimensional (2D) monolayer cultures of immortalized cell lines lack the cellular diversity and architectural complexity of the human lung epithelium [20]. These limitations underscore the urgent need for experimental models that more accurately recapitulate human physiology and the organization of lung tissue.

Human pluripotent stem cell-derived lung organoids (hLOs) offer a physiologically relevant alternative, recapitulating the complex cellular organization and diverse functional properties of the human lung epithelium. These organoids self-organize into three-dimensional (3D) structures that recapitulate both airway and alveolar epithelial lineages [21,22]. Unlike immortalized or transformed cell lines, which lack polarity and lineage complexity [23], hLOs mimic the heterogeneity and spatial organization of native lung tissue, making them ideally suited for lung toxicity assessments [24]. Incorporating a fluorescent SG reporter into such organoids enables continuous, real-time monitoring of early stress responses in a human-relevant context. In parallel, high-content screening (HCS) technologies has emerged as a key technology in modern toxicology, enabling automated and multiparametric imaging to capture subtle cellular responses across a broad spectrum of toxicants [25]. Integrating this approach with organoid-based models allows quantitative and reproducible analysis of SG dynamics, combining scalability with precision to provide a powerful tool for predictive toxicology and large-scale chemical risk evaluation.

In this study, we engineered a human induced pluripotent stem cell (hiPSC) line using CRISPR/Cas9-mediated genome editing to insert an mCherry fluorescent tag into the endogenous G3BP1 locus, ensuring accurate and endogenously regulated reporter expression [26]. This iPSC-derived human lung organoid reporter system is capable of real-time visualization of SG formation under toxicant exposure, directly linking translational stress with chemical-induced cytotoxicity. By integrating G3BP1-based SG monitoring with a hLO system, this platform establishes a sensitive and predictive New Approach Methodology (NAM) for evaluating lung toxicity and can be broadly applied to human stem cell-derived tissue models for chemical safety assessment.

2. Materials and methods

2.1. Cell culture

This study was reviewed and approved by the Public Institutional Review Board of the Ministry of Health and Welfare, Republic of Korea (P01-202208-02-010). The hiPSC line CMC-hiPSC-009 was obtained from National Stem Cell Bank, originally established by the Catholic University of Korea. hiPSCs were maintained on hESC-qualified Matrigel (354277, Corning Inc., Corning, NY, USA)-coated plates (Thermo Fisher Scientific, Waltham, MA, USA) in mTeSR™1 medium (85850, STEMCELL Technologies, Vancouver, Canada) at 37 °C in a humidified 5% CO₂ incubator. The culture was replaced daily, and cells were passaged every 3–4 days. A549 human lung epithelial carcinoma cells, purchased from the American Type Culture Collection (Manassas, VA, USA), were cultured in Dulbecco's modified eagle medium, high glucose (LM001-05, WELGENE, Gyeongsan, Korea) supplemented with 10% fetal bovine serum (16000044, Gibco, Waltham, MA, USA), 1% penicillin/streptomycin (15040122, Gibco), 1 mM sodium pyruvate (11360070, Gibco), and 1 × GlutaMAX (35050061, Gibco). Cells were cultured at 37 °C in a humidified 5% CO₂ incubator. The patient-derived lung cancer organoid (LCO) line SNU-2689-CO (Korean Cell Line Bank [KCLB] no. 02689-CO) was established from human lung adenocarcinoma tissues and obtained from the KCLB (Seoul National University, Seoul, Korea). LCOs were embedded in reduced growth factor basement membrane extract (3533-010-02, R&D systems, Minneapolis, MN, ULA) and maintained in the organoid culture medium supplied by KCLB following the manufacturer's protocol. Cultures were passaged every 10–14 days.

2.2. Generation of G3BP1–mCherry knock-in hiPSCs

Single guide RNAs (sgRNAs) targeting the G3BP1 locus near the stop codon were designed using the CHOPCHOP online tool. The sgRNA sequence used in this study (E12sg-1) was identical to that described previously [27]. Candidate sgRNAs were cloned into the Cas9 expression vector pX330-U6-Chimeric_BB-CBh-hSpCas9. The cleavage efficiency of each sgRNAs was validated by a T7 endonuclease I assay following the manufacturer's instructions (M0302, New England Biolabs, Ipswich, MA, USA). For donor plasmid, the mCherry containing vector was newly generated for this study, following the targeting strategy described earlier [27]. The donor backbone was derived from the promoterless pDsRed-Express2-1 vector (Clontech/Takara Bio, Shiga, Japan), in which the DsRed cassette was replaced with the mCherry sequence flanked by left and right homology arms corresponding to the G3BP1 locus. The homology arms were amplified by PCR using genomic DNA extracted from the CMC-hiPSC-009 line and ligated using T4 DNA ligase (M0202, New England Biolabs). Human iPSCs were co-transfected with the Cas9-sgRNA plasmid and the donor plasmid using Lipofectamine Stem Transfection Reagent (STEM00015, Invitrogen, Carlsbad, CA, USA) according to the manufacturer's protocol. After several days of recovery, mCherry-positive cells were isolated by fluorescence-activated cell sorting (FACSAria Fusion, BD Biosciences, San Jose, CA, USA). Single-cell-derived colonies were subsequently expanded and screened to identify correctly targeted clones.

2.3. Genotyping PCR

Genomic DNA was extracted from G3BP1–mCherry knock-in hiPSCs using the G-DEX™ IIc Genomic DNA Extraction Kit (17231, iNtRON Biotechnology, Seongnam, Korea) following the manufacturer's instructions. Genotyping PCR was performed with KOD FX Neo DNA polymerase (KFX-201, Toyobo, Osaka, Japan) on a SimpliAmp Thermal Cycler (Applied Biosystems, Thermo Fisher Scientific). PCR products were analyzed by agarose gel electrophoresis. Primer sets used for genotyping were G3BP1_genotype F and R, are listed in Table S1.

2.4. Embryoid body differentiation

Single-cell suspensions of G3BP1–mCherry knock-in hiPSCs were seeded at a density of 2 × 10³ cells per well in Ultra-Low Attachment 96-well plates (7007, Corning Inc.) using TeSR™-E6 medium (05946, STEMCELL Technologies). After centrifugation at 150×g for 3 min, the plates were placed in a humidified incubator maintained at 37 °C and 5% CO₂. Embryoid bodies were maintained under these conditions for 6 days prior to subsequent experiments.

2.5. Teratoma formation assay

G3BP1–mCherry knock-in hiPSCs were injected into immunodeficient NOG mice (NOD.Cg-Prkdc^scid Il2rg^tm1Sug/JicKoat, female, 7 weeks old) obtained from Koatech (Pyeongtaek, Korea). Mice were maintained for 11 weeks to allow teratoma formation, after which tumors were excised and processed for histological examination using hematoxylin and eosin staining. All experimental protocols involving animals were reviewed and approved by the Institutional Animal Care and Use Committee of Chungnam National University (202304A-CNU-059).

2.6. Lung organoid differentiation and maturation

Differentiation of the CMC-hiPSC-009 line into sequential stages of definitive endoderm (DE), anterior foregut endoderm (AFE), lung progenitor (LP), and 3D organoids was carried out according to a previously reported protocol [28] with slight modifications. Stage-specific cues were applied at each step: the DE stage was induced using the STEMdiff™ Definitive Endoderm Kit (05110, STEMCELL Technologies); AFE induction was achieved with 100 ng/mL Noggin, 10 μM SB431542, and 1 μM IWP-4, while 10 ng/mL BMP4, 10 ng/mL FGF7, 10 ng/mL FGF10, 3 μM CHIR99021, and 50 nM all-trans retinoic acid promoted further commitment toward the lung progenitor and lung organoid differentiation. From approximately day 30, organoid maturation was initiated and sustained in LO media supplemented with DCIF10 or DCIF10A formulations, which contained 50 nM dexamethasone, 0.1 mM cAMP, 0.1 mM IBMX, and 10 ng/mL FGF10, with or without 1 μM 5-aza-dC (referred to as DCIF10 and DCIF10A, respectively). Organoids were cultured in DCIF10 medium for 5 days, followed by cultured 10 days in DCIF10A medium. For suspension culture, hLOs were dissociated into single cells using TrypLE™ Express (12604021, Gibco) for 10 min at 37 °C and seeded into AggreWell™400 plates (34411, STEMCELL Technologies) at a density of 300–500 cells per microwell for approximately 7 days. Organoid-based assays, including immunofluorescence and quantitative PCR, were independently repeated using organoids derived from at least three separate differentiation batches, and representative results are shown.

2.7. Transmission electron microscopy

Samples were fixed with 2.5% glutaraldehyde in PBS and imaged using a transmission electron microscope (HT7800, Hitachi, Tokyo, Japan) operated at 80 kV.

2.8. Quantitative PCR

Total RNA was extracted from cells using TRIzol reagent (15596026, Ambion, Austin, TX, USA), and 1 μg of RNA was reverse-transcribed into cDNA using the GoScript™ Reverse Transcriptase (A2801, Promega, Madison, WI, USA). Reverse transcription was performed at 25 °C for 5 min (annealing), 42 °C for 60 min (extension), and 70 °C for 15 min (enzyme inactivation). Quantitative PCR was carried out using GoTaq® quantitative PCR (qPCR) Master Mix (A6002, Promega) on a CFX Duet Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA). Amplification was carried out with an initial denaturation at 95 °C for 10 min, followed by 40 cycles of denaturation at 95 °C for 15 s and annealing/extension at 60 °C for 1 min. Relative mRNA expression levels were calculated using the 2^-ΔΔCt method, with normalized to GAPDH. Primer sequences used for qPCR are listed in Table S2.

2.9. Western blotting

Protein lysates were prepared using RIPA lysis buffer (#89900, Thermo Fisher Scientific) supplemented with a protease inhibitor cocktail (P3100-001, GenDEPOT, Katy, TX, USA) and a phosphatase inhibitor cocktail (P3200-001, GenDEPOT). Protein concentration was determined using the Pierce™ BCA Protein Assay Kit (A55865, Thermo Fisher Scientific). Equal protein samples were electrophoresed on 10% SDS-PAGE gels and transferred onto PVDF membranes. After blocking in 5% skim milk for 1 h at room temperature, membranes were incubated with primary antibodies at 4 °C overnight and with horseradish peroxidase-conjugated secondary antibodies for 1 h at room temperature. Protein bands were detected using Amersham™ ECL Detection Reagents (RPN3004, Cytiva, Marlborough, MA, USA) and imaged with a LuminoGraph I system (ATTO, Tokyo, Japan). The primary and secondary antibodies employed in this study are summarized in Table S3.

2.10. Immunofluorescence staining

Cells were fixed with 4% formaldehyde for 1 h at 4 °C, followed by permeabilization using 0.1% Triton X-100 in PBS for 30 min at room temperature. Subsequently, cells were blocked for 1 h in PBS containing 2% bovine serum albumin and 0.1% Triton X-100. Primary antibodies, diluted in blocking buffer, were then applied and incubated overnight at 4 °C. After washing with PBS, cells were incubated with fluorophore-conjugated secondary antibodies for 2 h at room temperature in the dark. Nuclei were counterstained with Hoechst 33342 (H3570, Thermo Fisher Scientific). Fluorescence images were acquired using a Cytation C10 imaging system and analyzed with Gen5 software (Agilent Technologies, Santa Clara, CA, USA). Primary and secondary antibodies used for immunofluorescence are listed in Table S3.

2.11. Single-cell RNA sequencing and data processing

Single-cell suspensions from differentiated hLOs were prepared and processed using the Chromium GEM-X Single Cell 3′ Reagent Kit v4 (10x Genomics, Pleasanton, CA, USA; User Guide CG000732). Libraries were quality-checked with a Bioanalyzer (Agilent Technologies) and sequenced on an Illumina NovaSeq 6000 (28/10/10/90 bp, paired-end). Raw data were converted to FASTQ format using BCL Convert v4.3.6 (Illumina), and mapped to the GRCh38-2024-A reference genome with Cell Ranger v9.0.1 (10x Genomics) to generate gene–cell matrices. Data analysis was performed using Scanpy v1.11.3. Low-quality cells were filtered based on mitochondrial and ribosomal transcript content, and doublets were excluded using Scrublet v0.2.3. Epithelial cells were selected based on expression of epithelial (EPCAM, KRT8, and KRT18) and non-epithelial (PTPRC, PECAM1, and COL1A1) markers. After filtering, 6050 cells were retained for downstream analysis. Counts were normalized, log-transformed, and processed by PCA, UMAP, and Leiden clustering. Cell type annotation was performed based on cluster-specific marker expression via Wilcoxon rank-sum test. Seven populations were identified: Basal, KRT5^-/KRT17⁺ transitional, Secretory, Goblet, AEC, Proliferating, and Neuroendocrine (NE) cells. For developmental benchmarking, pseudobulk profiles were compared with public single-cell datasets (LungMAP: LMEX0000001624, LMEX0000003690) using PCA and hierarchical clustering. A detailed summary of all software is provided in Table S4.

2.12. Chemicals

The following chemicals were used in this study: Sodium arsenite (NaAsO2; CAS No. 7784-46-5; Honeywell Fluka, Charlotte, NC, USA), benzalkonium chloride (BAC; CAS No. 63449-41-2; obtained from a collaborating laboratory), 1,2-benzisothiazolone (BIT; CAS No. 2634-33-5; Sigma-Aldrich, St. Louis, MO, USA), Didecyldimethylammonium chloride (DDAC; CAS No. 7173-51-5; Sigma-Aldrich), polyhexamethylene guanidine (PHMG-p; CAS No. 89697-78-9; obtained from a collaborating laboratory), 2-Octyl-4-isothiazolin-3-one (OIT; CAS No. 26530-20-1; Supelco, Merck kGaA, Darmstadt, Germany), 4,4′-Methylenebis(phenyl isocyanate) (4,4′-MDI; CAS No. 101-68-8; Sigma-Aldrich), and acrylonitrile (CAS No. 107-13-1; Sigma-Aldrich) were used in this study. Stock solutions were prepared in sterile distilled water or dimethyl sulfoxide, followed by dilution in culture medium immediately prior to use. Final concentrations and exposure durations are described in the respective experimental section.

2.13. Cell viability assay

Cell viability was evaluated using the CellTiter-Glo® Luminescent Cell Viability Assay (G7571, Promega) for 2D cultures and the CellTiter-Glo® 3D Cell Viability Assay (G9682, Promega) for 3D cultures, according to the manufacturer's protocols. For assay optimization and to minimize cell number variability, organoid samples were dissociated into single cells and embedded in Matrigel droplets prior to viability measurements. Briefly, cells or organoids were incubated with an equal volume of the CellTiter-Glo® reagent in opaque-walled 96-well plates (33696, SPL Life Sciences, Pocheon, Korea), followed by shaking to induce lysis and incubation at room temperature for signal stabilization. Luminescence was measured using a GloMax® Discover Microplate Reader (Promega). Background luminescence from wells containing medium only was subtracted, and cell viability was expressed as percentage relative to vehicle-treated controls (set as 100%).

2.14. High-content live-cell imaging

For live-cell imaging, hiPSCs were seeded into 96-well imaging plates (Agilent Technologies) and cultured for approximately two days until colonies formed, after which cells were exposed to 500 μM sodium arsenite, 30 μg/mL BAC, or 60 μM BIT. Pre-formed hLOs from AggreWell plates were transferred into 384-well imaging plates (Agilent Technologies) and immediately subjected to the same chemical treatments. Kinetic imaging was performed for 1 h at 3 min intervals using a Cytation C10 imaging system (Agilent BioTek) equipped with a temperature- and CO₂-controlled chamber, with mCherry fluorescence monitored in the TRITC channel. SG formation was quantified via automated spot counting using the Spot Count function in Gen5 (v3.17; Agilent BioTek). For selected washout experiments, the treatment medium was removed and organoids were washed three times with DPBS before fresh culture medium was added, followed by continued live-cell imaging.

2.15. Statistical analysis

All statistical analyses were carried out using GraphPad Prism (v10.2.2; GraphPad Software, San Diego, CA, USA). Quantitative data are shown as mean ± standard deviation (SD). Statistical differences between two groups were analyzed by an unpaired two-tailed t-test, and multiple group comparisons were evaluated using one-way analysis of variance (ANOVA) with Tukey's test. A significance threshold of p < 0.05 was applied, and IC₅₀ values were determined by nonlinear regression using a variable-slope dose–response model (log[inhibitor] vs. normalized response). For comparative visualization, each IC₅₀ value was normalized to the mean IC₅₀ of the corresponding compound across all cell models, and the resulting ratios were transformed to log₂ scale to obtain relative log₂ IC₅₀ values. Heatmaps were generated using a color scale ranging from −2 to +2, representing up to a fourfold deviation from the mean IC₅₀, with values beyond this range clipped for clarity.

3. Results

3.1. CRISPR/Cas9-mediated generation and validation of endogenous G3BP1–mCherry knock-in hiPSCs

To establish a physiologically relevant reporter system for monitoring SG dynamics, we generated a hiPSC line harboring an endogenous G3BP1–mCherry knock-in. The knock-in cassette was designed to fuse mCherry in-frame with the C-terminus of G3BP1, preserving native regulation and enabling fluorescence-based visualization (Fig. 1A). Among five candidate sgRNAs, the exon 12-targeting sgRNA with the highest cleavage efficiency (T7E1 assay) was selected for homology-directed repair-mediated integration (Fig. S1).

Fig. 1.

Generation and validation of the G3BP1–mCherry knock-in hiPSC line. (A) Schematic overview of the CRISPR/Cas9-mediated knock-in strategy to fuse mCherry to the C-terminus of endogenous G3BP1. (B) PCR amplification of genomic DNA showing approximately 1.4 kb product in knock-in cells. M, DNA molecular weight marker. (C) Sanger sequencing of the knock-in junction confirming the in-frame fusion of mCherry to the G3BP1 coding region and the presence of a silent PAM-disrupting mutation introduced to prevent Cas9 re-cleavage (highlighted). A detailed sequence alignment is shown in Fig. S2. (D) Western blot analysis using an anti-mCherry antibody showing the G3BP1–mCherry fusion protein (∼96 kDa) in knock-in hiPSCs; no signal was observed in wild-type (WT) cells. β-actin served as a loading control. (E) Representative fluorescence images showing mCherry-labeled G3BP1 in untreated and 500 μM sodium arsenite-treated hiPSCs. The reporter visualizes stress granule formation under oxidative stress. Nuclei were stained with Hoechst 33342 (gray). Scale bars, 50 μm.

Following co-transfection with Cas9/sgRNA and donor plasmids, mCherry-positive cells were isolated by fluorescence-activated cell sorting and subsequently expanded from single clones. Polymerase chain reaction analysis confirmed successful integration, yielding approximately 1.4 kb product corresponding to the targeted G3BP1–mCherry allele (Fig. 1B). Sanger sequencing verified precise in-frame insertion and confirmed the protective point mutation introduced to prevent Cas9 re-cleavage (Fig. 1C and S2). Western blotting using an anti-mCherry antibody detected a single ∼96 kDa fusion band exclusively in knock-in cells, confirming expression of the endogenously regulated G3BP1–mCherry protein (Fig. 1D). To evaluate reporter functionality, knock-in hiPSCs were exposed to sodium arsenite (500 μM), a canonical SG inducer [29]. Upon treatment, diffuse cytoplasmic mCherry signals rapidly condensed into distinct puncta (Fig. 1E), indicating that the fusion protein retained its SG-forming capacity. BAC (30 μg/mL) and BIT (70 μM) also induced SG formation (Fig. S3), confirming reporter functionality across chemically diverse stressors. These results validate the molecular and functional integrity of the G3BP1–mCherry reporter prior to organoid differentiation.

Next, we assessed whether genome editing affected the fundamental properties of the hiPSCs. Karyotype analysis revealed a normal male karyotype (46, XY), indicating that CRISPR/Cas9-mediated knock-in did not compromise genomic stability (Fig. 2A). Immunofluorescence staining confirmed robust expression of pluripotency markers OCT3/4, SSEA4, and NANOG (Fig. 2B), consistent with maintenance of the undifferentiated state [30]. The knock-in hiPSCs retained tri-lineage differentiation potential, as shown by the induction of ectodermal (PAX6, Nestin), mesodermal (Desmin, TBXT), and endodermal (SOX17, GATA4) markers during embryoid body differentiation, with corresponding protein expression (Fig. 2C) [31]. Teratoma formation in vivo further demonstrated differentiation into derivatives of all three germ layers, including neural tissue and neural tube (ectoderm), cartilage and muscle (mesoderm), and gut-like epithelium (endoderm) (Fig. 2D).

Fig. 2.

Karyotype and pluripotency validation of the G3BP1–mCherry knock-in hiPSC line. (A) Representative G-banding karyotype image of G3BP1–mCherry knock-in hiPSCs showing a normal male karyotype (46, XY). (B) Immunofluorescence staining of pluripotency markers OCT3/4, SSEA4, and NANOG in G3BP1–mCherry knock-in hiPSCs. Nuclei were counterstained with Hoechst 33342. Scale bars, 50 μm. (C) In vitro embryoid body (EB) differentiation demonstrating tri-lineage potential of knock-in hiPSCs. Quantitative PCR analysis showing expression of lineage-specific markers for ectoderm (PAX6, Nestin), mesoderm (Desmin, TBXT), and endoderm (SOX17, GATA4). Data are presented as mean ± SD. Statistical analysis was performed using an unpaired two-tailed t-test (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). Representative immunofluorescence images of EBs showing expression of Nestin (ectoderm), α-SMA (mesoderm), and SOX17 (endoderm). Nuclei were counterstained with Hoechst 33342. Scale bars, 200 μm (bright-field) and 50 μm (immunofluorescence). (D) Teratoma formation assay demonstrating in vivo differentiation into the three germ layers. Representative hematoxylin and eosin (H&E)-stained sections display ectoderm (N, neural tissue; NT, neural tube), mesoderm (C, cartilage; M, muscle), and endoderm (G, gut-like epithelium). Scale bars, 200 μm.

Collectively, these findings establish a genetically stable, pluripotent G3BP1–mCherry knock-in hiPSC line that faithfully reports SG formation, providing a robust foundation for downstream lung organoid differentiation and application in predictive toxicology.

3.2. Differentiation of G3BP1–mCherry hiPSCs into lung organoids and optimization of maturation conditions

Stepwise differentiation of the G3BP1–mCherry hiPSC line into DE, AFE, LP, and organoid stages was performed as outlined in Fig. 3A. Each stage showed morphological transitions, including epithelial sheet (AFE) and bud tip–like (LP) structures. Upon embedding LPs in Matrigel droplets, cells expanded into 3D branching structures that developed into spherical lung organoids (Fig. 3B; see also Fig. S4A for the full-time course). Successful lineage progression was confirmed by the upregulation of canonical DE markers, CXCR4, CER, FOXA2, and SOX17, at the transcript level (qPCR), together with protein expression of EpCAM, CXCR4, c-KIT, and SOX17 (flow cytometry) (Fig. S4B and C) [32]. Differentiation toward lung progenitors was further validated by CPM expression (Fig. S4D and E) [33], confirming efficient commitment to the pulmonary epithelial lineage prior to 3D organoid formation. Western blot analysis detected the G3BP1–mCherry fusion protein in differentiated organoids but not in wild-type hLOs, indicating stable maintenance of the reporter allele throughout differentiation (Fig. 3C).

Fig. 3.

Differentiation of G3BP1–mCherry hiPSCs into lung organoids and optimization of maturation conditions. (A) Schematic overview of the stepwise differentiation of G3BP1-mCherry hiPSCs into lung organoids, showing the key stages and timeline of the protocol. (B) Representative brightfield images of cells at different stages of differentiation. Scale bars, 100 μm. Representative images from day 0, 2, 4, 13, and 30 are shown, and the full-time course (Day 0–30) is provided in Fig. S4A. (C) Western blot analysis using an anti-mCherry antibody showing the G3BP1–mCherry fusion protein (∼96 kDa) detected in knock-in hLOs, but absent in wild-type (WT) hLOs. β-actin served as a loading control. (D) Transmission electron microscopy (TEM) images of mature hLOs showing lamellar body-like organelles (LB) within the cytoplasm. Scale bars, 200 nm. (E) Quantitative PCR analysis of lung lineage marker gene expression, including airway markers for basal (TP63), goblet (MUC5AC), club (SCGB1A1), and ciliated (FOXJ1) cells, as well as alveolar markers for type I (HOPX, AGER) and type II (SFTPB, SFTPC) cells. Data are presented as mean ± SD. Statistical analysis was performed using an unpaired two-tailed t-test (****p < 0.0001). (F) Immunofluorescence images showing representative airway and alveolar markers, including SCGB3A2 (green, club cells), AQP5 (green, alveolar type I cells), MUC1 (green, alveolar type II cells), and SFTPB (green, alveolar type II cells). Nuclei were counterstained with Hoechst 33342 (blue). Scale bars, 20 μm.

To promote epithelial maturation, two maturation media were compared: DCIF10 (dexamethasone, cAMP, IBMX, and FGF10) and its modified form, DCIF10A, supplemented with the DNA demethylating agent 5-aza-2′-deoxycytidine (5-aza-dC). The DCIF10 cocktail, classically used to enhance alveolar maturation and surfactant secretion in vitro, served as the baseline condition, while FGF10 supported distal epithelial growth and branching morphogenesis [[34], [35], [36]]. Based on prior evidence that 5-aza-dC can derepress AQP5 expression [37], DCIF10 and DCIF10A were first compared under droplet culture conditions. Although 5-aza-dC modestly upregulated several lung lineage markers, including FOXJ1, alveolar-associated genes—particularly alveolar type Ⅱ (AT2) markers—showed minimal change (Fig. S5). Given the limited effect in droplet culture, we examined whether culture format influenced differentiation outcomes. Transitioning to suspension culture markedly increased expression of multiple lineage markers, including alveolar type Ⅰ (AT1)-associated genes, indicating that culture format strongly affects epithelial maturation (Fig. S6). Under suspension conditions, 5-aza-dC supplementation (DCIF10A) further enhanced expression of lung-specific genes, demonstrating that epigenetic modulation synergizes with suspension culture to promote epithelial maturation (Fig. S7).

Transmission electron microscopy of hLOs differentiated under optimized DCIF10A suspension culture revealed lamellar body-like organelles within the cytoplasm (Fig. 3D). Concordantly, qPCR showed robust induction of both airway and alveolar epithelial markers, including basal (TP63), goblet (MUC5AC), club (SCGB1A1), ciliated (FOXJ1), AT1 (HOPX, AGER), and AT2 (SFTPB, SFTPC) cell markers (Fig. 3E). Immunofluorescence confirmed the presence of these cell types, showing KRT5⁺ basal, MUC5AC⁺ goblet, SCGB3A2⁺ club, acetylated tubulin⁺ ciliated, AQP5⁺, PDPN⁺, or HOPX⁺ AT1, and HTII-280⁺ or SFTPB⁺ AT2 cells distributed throughout mature organoids (Fig. 3F and S8).

Collectively, these results demonstrate that DCIF10A in suspension culture reproducibly enhances lung epithelial maturation and supports the coexistence of diverse airway and alveolar cell types, establishing this condition as the optimized maturation protocol for subsequent toxicological applications.

3.3. Single-cell transcriptomic analysis identifies diverse airway and alveolar epithelial cell populations in human lung organoids

To define the epithelial composition of hLOs, epithelial and non-epithelial cells were first separated using canonical marker panels and per-cell module scoring, resulting in a clear bifurcation and retention of epithelial populations for downstream analysis (Fig. S9). Single-cell transcriptomic profiling identified multiple transcriptionally distinct epithelial clusters, Basal, KRT5^-/KRT17⁺, Secretory, Goblet, AEC, Proliferating, and NE cells (Fig. 4A). Secretory cells constituted the majority of the epithelial compartment (54.0%), followed by Basal (19.9%), Goblet (13.5%), and AEC (9.4%) populations (Fig. 4B). UMAP embedding revealed a continuum of epithelial states, with airway-like and alveolar-like lineages coexisting within the epithelial compartment. A proliferating cluster localized at the interface between major lineages, reflecting dynamic turnover and transitional potential. Marker gene expression supported the annotation of each cluster (Fig. 4C and D). Basal cells expressed KRT5, KRT17, and TP63, while Secretory cells were characterized by DPP4, CEACAM6, and TSPAN8 expression. Goblet cells exhibited TFF1, TFF3, and AGR2, representing a secretory trajectory. The KRT5^-/KRT17⁺ population showed loss of KRT5 with retained KRT17 expression alongside goblet-associated markers. AEC clusters expressed alveolar markers including CAV1, HOPX, and ABCA3, while NE cells displayed CHGA and ASCL1, and SCG2 expression. The proliferating cluster was enriched for MKI67 and TOP2A, indicative of active cycling. Consistent with the single-cell transcriptomic analysis, qPCR validation of representative epithelial lineage markers further supported epithelial cell-type annotation in hLOs (Fig. S10). Developmental maturity was further assessed using pseudobulk principal-component analysis (PCA) of aggregated epithelial transcriptomes benchmarked against fetal and adult lung references (Fig. 4E and F). hLOs clustered proximal to the fetal group along PC1, which accounted for 99.7% of the variance, indicating transcriptional similarity to late-fetal or early-postnatal lung. Pseudobulk heatmap analysis confirmed that hLO expression patterns aligned closely with fetal profiles, whereas adult-enriched transcripts were underrepresented.

Fig. 4.

Single-cell transcriptomic profiling of epithelial populations in human lung organoids. (A) UMAP visualization of epithelial cells derived from hLOs showing distinct clusters corresponding to basal, KRT5^-/KRT17⁺, secretory, goblet, AEC-like, proliferating, and neuroendocrine (NE) populations. (B) Relative proportions of epithelial cell types in hLOs. Secretory cells were most abundant (54.0%), followed by basal (19.9%), goblet (13.5%), AEC-like (9.4%), proliferating (2.1%), KRT5^-/KRT17⁺ (0.9%), and NE (0.1%) cells. (C) Dot plot showing expression of representative marker genes used for cluster annotation. Dot size indicates the fraction of cells expressing each gene, and color intensity represents mean expression level. (D) Heatmap of marker gene expression across cell types. Up to 50 cells were randomly sampled per cell type for balanced visualization. (E) PCA based on aggregated epithelial transcriptomes from organoid, fetal, and adult lung datasets summarize group-level variation and positions the organoid transcriptome relative to fetal and adult references. (F) Pseudobulk heatmap of high-variance genes (top 25 by group-wise expression change) shared across all groups provides a scaled comparative view of organoid similarity and maturation tendency.

3.4. Real-time visualization of SG dynamics in G3BP1–mCherry hLOs

To evaluate whether the G3BP1–mCherry reporter enables real-time tracking of SG dynamics in a human-relevant lung model, we performed live imaging of hLOs following sodium arsenite exposure. As a prototypical SG inducer, arsenite triggered a dose- and time-dependent SG response. At 12.5 and 25 μM, SG formation was minimal, whereas ≥50 μM progressively increased cytoplasmic G3BP1–mCherry puncta over a 1 h (Fig. S11A). High-content image analysis confirmed that positive correlation between arsenite concentration, exposure duration, and SG number (Fig. S11B). Live imaging of hLOs exposed to 500 μM arsenite revealed rapid condensation of diffuse cytoplasmic G3BP1–mCherry into discrete puncta within 10–20 min, followed by gradual disassembly, indicating transient and reversible SG formation (Fig. 5A and Video S1). Quantitative analysis showed that SG numbers peaked at 30 min and declined thereafter, consistent with adaptive stress resolution. Together, these results provide quantitative mapping between toxicant dose and exposure duration and SG kinetics in hLOs, revealing a concentration-dependent acceleration of SG onset followed by a transient peak and subsequent recovery.

Fig. 5.

Real-time live-cell imaging of stress granule dynamics in G3BP1–mCherry hLOs following toxicant exposure. (A) Representative fluorescence images showing the distribution of G3BP1–mCherry into cytoplasmic puncta in hLOs after 1 h of exposure to sodium aresnite (500 μM). Magnified insets highlight G3BP1-positive stress granules (SGs). The accompanying time-course plot (right) depicts changes in total SG counts over 60 min, recorded at 3 min intervals, showing rapid SG assembly within 20–30 min followed by gradual reduction. (B, C) Live-cell imaging of SG formation in G3BP1–mCherry hLOs after 1 h exposure to benzalkonium chloride (BAC; 30 μg/mL) or 1,2-benzisothiazolin-3-one (BIT; 70 μM). Magnified views show discrete G3BP1-positive granules (white arrows). Scale bars, 20 μm.

To determine whether SG induction was specific to oxidative stress or generalized across other chemical classes, hLOs were exposed to benzalkonium chloride (BAC; 30 μg/mL) or 1,2-benzisothiazolin-3-one (BIT; 70 μM), representing membrane-active and electrophilic biocides, respectively. Both toxicants induced cytoplasmic SG formation within 1 h (Fig. 5B and C), demonstrating that G3BP1-mCherry hLOs can detect chemically diverse insults. These results establish G3BP1–mCherry hLOs as a robust system for real-time, quantitative visualization of SG dynamics under toxicant exposure, providing a sensitive, non-destructive platform for human-relevant lung toxicity assessment.

3.5. Differential toxicant sensitivity across lung organoid and conventional models

Having established real-time SG monitoring in hLOs, we next compared their overall toxicant sensitivity with conventional lung models to assess their predictive value for chemical toxicity testing. Cell viability was measured after 24 h exposure to a panel of environmentally relevant compounds representing distinct functional categories (Fig. 6). The panel included disinfectants and biocides—a quaternary ammonium compound (DDAC), an isothiazolinone preservative (OIT), and a polyguanidine-based disinfectant (PHMG-p)—as well as industrial isocyanate (4,4′-MDI) and volatile toxicant (acrylonitrile). These agents activate diverse cellular stress pathways, including oxidative, endoplasmic reticulum, and mitochondrial stress, that ultimately lead to cytotoxicity [[38], [39], [40], [41], [42]]. Across all tested chemicals, G3BP1–mCherry hLOs exhibited markedly lower IC₅₀ values than A549 cells, indicating significantly higher susceptibility to chemical-induced cytotoxicity (Fig. 6A and B). This trend was consistent across chemical classes, supporting the notion that 3D organoid models display greater physiological responsiveness and sensitivity than conventional immortalized 2D cell lines [43,44]. The overall pattern was visualized in a heatmap (Fig. 6C), where hLOs showed a distinct blue shift corresponding to lower relative log₂(IC₅₀) values compared with A549 cells. These findings establish hLOs as a more sensitive and physiologically relevant platform for chemical toxicity evaluation.

Fig. 6.

Dose–response profiling of representative toxicant groups in A549 and G3BP1–mCherry hLO. Cell viability was measured after 24 h exposure to five representative toxicants in two in vitro lung models: A549 (2D monolayer) and G3BP1–mCherry knock-in human lung organoids (hLOs). (A) Dose–response curves for the disinfectant/biocide group— didecyldimethylammonium chloride (DDAC), 2-octyl-4-isothiazolin-3-one (OIT), and polyhexamethylene guanidine (PHMG-p). (B) Dose–response curves for industrial chemicals—4,4′-MDI and acrylonitrile. Curves were fitted using nonlinear regression (log[inhibitor] vs. normalized response, variable-slope model) in GraphPad Prism. Data are presented as mean ± SD. (C) Heatmap displaying relative log₂(IC₅₀) values comparing A549 and hLO models. Each IC₅₀ value was normalized to the mean IC₅₀ of the respective compound. The color scale represents deviations from the mean, with blue indicating higher sensitivity and red indicating lower sensitivity. The scale ranges from −2 to +2, corresponding to up to a fourfold deviation from the mean IC₅₀; values outside this range were clipped for clarity.

To explore whether this heightened sensitivity reflected 3D architecture or non-malignant origin, we compared multiple 3D lung models—A549 spheroids, patient-derived LCOs, and hLOs, using three representative toxicants (sodium arsenite, BAC, and BIT) (Fig. 7A). For all compounds, hLOs consistently exhibited the lowest IC₅₀ values (36.84 μM for arsenite, 2.142 μg/mL for BAC, and 70.09 μM for BIT), followed by LCOs and A549 (3D spheroid). Heatmap visualization reinforced this pattern, showing a pronounced blue shift for hLOs and red-to-neutral gradients for LCOs and A549 spheroids (Fig. 7B). These results demonstrate that hLOs exhibit the highest sensitivity among 3D systems, capturing toxicant responses more representative of normal human lung tissue, whereas immortalized cell-derived models display greater resistance.

Fig. 7.

Comparative toxicant sensitivity and stress granule responses in G3BP1–mCherry hLOs and conventional lung models. (A) Cell viability analysis of sodium arsenite (NaAsO₂), benzalkonium chloride (BAC), and 1,2-benzisothiazolin-3-one (BIT) was evaluated in three lung models: A549 (3D spheroid), patient-derived lung cancer organoid (LCO), and G3BP1–mCherry knock-in lung organoid (hLO). IC₅₀ values were determined by nonlinear regression (log[inhibitor] vs. normalized response, variable-slope model) using GraphPad Prism. Data are presented as mean ± SD. (B) Heatmap depicting relative log₂(IC₅₀) values to illustrate model-specific differences in toxicant sensitivity. Each IC₅₀ value was normalized to the mean IC₅₀ of the respective compound. Blue and red represent higher and lower sensitivity relative to the mean, respectively. (C) Immunofluorescence staining showing stress granule (SG) formation in patient-derived LCOs and G3BP1–mCherry hLOs after 1 h exposure to sodium arsenite (0, 50, or 100 μM). G3BP1 (green) indicates SG localization, and nuclei were counterstained with Hoechst 33342 (blue). Representative images at intermediate concentrations (12.5, 25, 75 μM) are shown in Fig. S12. Scale bars, 20 μm. (D) Quantification of SG numbers per cell in patient-derived LCOs and hLOs following sodium arsenite treatment (0–100 μM). Bars represent mean ± SD. Statistical analysis was performed using two-way ANOVA followed by Sidak's multiple comparisons test (ns, not significant; ***p < 0.001; ****p < 0.0001).

To further probe model-specific stress responses, we analyzed SG formation in LCOs and hLOs following sodium arsenite exposure (0–100 μM, 1 h). Immunofluorescence revealed a clear, dose-dependent increase in G3BP1-positive puncta in both models (Fig. 7C). Notably, SGs appeared at lower arsenite concentrations in hLOs than in LCOs, indicating heightened stress sensitivity. Additional intermediate arsenite concentrations further support the lower SG formation threshold in hLOs (Fig. S12). Quantitative analysis showed that although SG numbers were comparable at 12.5 and 25 μM, hLOs exhibited significantly higher SG counts at 50 and 75 μM (Fig. 7D). These results demonstrate that non-malignant hLOs have a lower activation threshold for SG formation, reflecting greater responsiveness to chemical stress compared with tumor-derived LCOs.

3.6. SG dynamics correlate with exposure duration–dependent epithelial barrier integrity

To functionally evaluate epithelial integrity following toxicant exposure and to examine its relationship with SG formation, we analyzed SG dynamics together with epithelial barrier integrity under different sodium arsenite exposure conditions. Real-time live-cell imaging of G3BP1–mCherry hLOs demonstrated rapid and dynamic SG formation following sodium arsenite exposure (500 μM). SG assembly was detectable within 15–30 min of exposure and increased during stress exposure. Upon removal of sodium arsenite, SGs gradually dissolved and returned to a diffuse cytoplasmic distribution, indicating reversible SG dynamics under transient stress conditions (Fig. 8A). To determine whether SG formation is associated with functional epithelial injury, epithelial barrier integrity was evaluated using ZO-1 immunofluorescence following both continuous and transient exposure paradigms. Continuous high-dose exposure (500 μM, 3 h) resulted in marked disruption of junctional ZO-1 organization, indicating loss of epithelial barrier integrity (Fig. 8B). In contrast, short-term exposure (500 μM, 1 h) followed by 24 h recovery largely preserved junctional ZO-1 organization, comparable to untreated controls, suggesting reversible barrier perturbation under transient stress conditions (Fig. 8C). Notably, SG formation was robustly induced during early exposure time points under both conditions, whereas epithelial barrier disruption was predominantly observed only under prolonged high-dose exposure. These findings indicate that SG assembly represents an early and dynamic cellular stress response that precedes overt epithelial barrier injury. These results demonstrate an exposure duration–dependent relationship between SG dynamics and epithelial barrier integrity, supporting the utility of SG imaging as an early and sensitive readout of cellular stress responses in hLOs. Consistent with these findings, additional TEER measurements and time-lapse imaging analyses under continuous and transient exposure conditions further supported exposure duration–dependent differences in epithelial barrier integrity and SG dynamics (Fig. S13).

Fig. 8.

Stress granule dynamics and functional epithelial barrier responses following sodium arsenite exposure in G3BP1–mCherry hLOs. (A) Real-time live-cell imaging of G3BP1–mCherry stress granule dynamics in hLOs following sodium arsenite (NaAsO₂, 500 μM) exposure and washout. Stress granules rapidly formed during exposure and progressively dissolved following stress removal, demonstrating reversible stress granule dynamics. Magnified images highlight representative punctate SG assemblies. Scale bar, 20 μm. (B–C) Immunofluorescence analysis of epithelial junction integrity following NaAsO₂ exposure. Continuous high-dose exposure (500 μM, 3 h) resulted in marked disruption of junctional ZO-1 organization, whereas short-term exposure (1 h) followed by 24 h recovery preserved junctional integrity, indicating reversible epithelial barrier perturbation under transient stress conditions. Nuclei were counterstained with Hoechst 33342 (blue).

Together, these comparative and functional analyses establish G3BP1–mCherry hLOs as a robust and sensitive model for chemical toxicity assessment. Their ability to capture both early SG formation, epithelial functional responses, and cytotoxic thresholds across diverse chemical classes underscores their physiological relevance and predictive utility for evaluating human respiratory toxicity.

4. Discussion

SGs are dynamic, membraneless condensates that assemble in response to cellular stress, functioning as early sentinels of translational arrest. To capture these dynamics under native conditions, we developed a CRISPR/Cas9-engineered hiPSC line expressing G3BP1–mCherry from its endogenous locus. This gene-targeted knock-in strategy maintains endogenous expression of the reporter protein at physiologically relevant levels, avoiding the overexpression artifacts that commonly arise in transient or viral reporter systems [45,46]. This design eliminates non-specific granule formation and enables continuous, non-destructive imaging of authentic stress responses in living human cells. Together, our approach allows live-cell tracking of SG dynamics with high temporal resolution, offering a robust and physiologically faithful platform for predictive toxicity assessment under chemical exposures. This platform provides a complementary, imaging-based readout of early cellular stress responses, enabling rapid screening and prioritization of chemical exposures before overt cytotoxicity becomes apparent. Key limitations include the current lack of immune or vascular components and the inherent variability of hiPSC-derived organoid differentiation, which should be considered when interpreting and applying the assay.

Live imaging revealed that SGs formed rapidly and resolved reversibly in hLOs upon toxicant exposure, confirming that the endogenous reporter faithfully captures dynamic stress responses in real time. By monitoring SG assembly and disassembly, the system provides a quantitative readout of early translational stress responses before the onset of irreversible cytotoxicity—a phase of the response that fixed-cell assays cannot resolve. This early sensitivity addresses a key limitation of conventional toxicity assays, which detect only late-stage cellular damage. Notably, SG induction was observed in both undifferentiated hiPSCs and differentiated hLOs, indicating that the reporter remains functional throughout differentiation and that SG formation is conserved across developmental states.

Epithelial diversification and organization are essential for establishing a human-relevant lung organoid model that enables accurate toxicity assessment. In conventional 2D lung epithelial and organoid models, AT1-associated genes such as AQP5, AGER, and HOPX are often underrepresented [47], limiting physiological relevance. In our culture conditions, suspension culture in DCIF10A medium with 5-aza-dC promoted apical-out polarity and enhanced AT1/AT2 marker expression, likely through epigenetic derepression and improved differentiation [[48], [49], [50]]. Transmission electron microscopy revealed lamellar body-like structures indicative of AT2-like maturation [51]. Single-cell transcriptomics further confirmed the presence of diverse epithelial lineages encompassing basal, secretory, goblet, alveolar, proliferating and NE populations. A KRT5^-/KRT17⁺ population was also identified, characterized by reduced KRT5 expression with retained KRT17 and goblet-associated markers [52], suggesting a transitional state during epithelial differentiation in hLOs. This cellular diversity reflects the physiological complexity of the human lung epithelium [[53], [54], [55]], underscoring its value for human-relevant toxicology. Pseudobulk analysis also showed alignment with fetal lung profiles, consistent with an organotypic yet developmentally immature epithelial state described in previous benchmarking studies of human lung organoid models [[56], [57], [58], [59]]. Together, these results indicate that hLOs recapitulate both the structural and transcriptional heterogeneity of the human lung, providing a physiologically relevant framework for predictive toxicity evaluation. This advanced fidelity overcomes key limitations of traditional cell line models, positioning hLOs as a more physiologically relevant and translationally meaningful alternative to these models for scalable, HCS-based toxicological screening and broader preclinical applications.

The G3BP1–mCherry reporter enabled real-time monitoring of SG formation under toxicant exposure. Sodium arsenite induced robust dose- and time-dependent SG assembly: low concentrations triggered delayed SG formation, whereas higher doses caused earlier and robust assembly. This graded, concentration-dependent onset indicates that SG kinetics reflect stress thresholds, providing a quantitative and sensitive indicator of early translational stress. Mechanistically distinct toxicants such as BAC (a membrane-active biocide) [60] and BIT (an electrophilic preservative) [61], also induced SG formation, indicating that SG assembly represents a conserved response to diverse chemical stressors. For all tested toxicants, hLOs exhibited lower IC₅₀ values and earlier stress onset compared with immortalized (A549) or tumor-derived (patient-derived LCO) models, underscoring their heightened sensitivity to chemical stress. In contrast, immortalized or tumor-derived models showed higher resistance to cytotoxicity, reflecting adaptive and apoptosis resistant phenotypes in transformed epithelium [[62], [63], [64]]. These distinctions emphasize the predictive advantage of the hLO-based SG reporter platform, which distinguishes early and reversible stress responses that remain undetectable in conventional models [65]. Importantly, our data further indicate that SG dynamics are linked to exposure duration–dependent changes in epithelial barrier integrity. While SG assembly was rapidly induced under both transient and continuous stress conditions, epithelial barrier disruption was predominantly observed under prolonged high-dose exposure. These findings suggest that SG formation represents an early adaptive stress response that precedes overt epithelial injury.

From a toxicological perspective, this relationship supports the use of SG imaging as an early, non-destructive biomarker of cellular stress, while complementary functional endpoints, such as epithelial barrier integrity, help distinguish adaptive stress responses from exposure conditions associated with tissue injury or progression toward cell death.

5. Conclusion

Overall, the SG reporter lung organoid platform provides a live-cell, HCS-compatible system for rapid and sensitive detection of translational stress. Integration with HCS technologies extends this system into a quantitative and scalable platform, facilitating high-throughput image-based evaluation of toxicant-induced stress responses with improved reproducibility and predictive value. Leveraging its endogenously expressed, non-destructive, and reversible reporter, the platform further supports repeated or prolonged low-dose exposures that mimic chronic environmental conditions, extending its applicability beyond acute toxicity testing. In addition, integration of early SG imaging with complementary functional endpoints, such as epithelial barrier integrity, further enhances the platform's utility for distinguishing adaptive stress responses from exposure conditions associated with tissue injury. This allows long-term investigation of adaptive or cumulative cellular responses under physiologically relevant conditions. Further studies could broaden its application to diverse environmental and industrial chemicals to define generalizable stress-response signatures and advance predictive models for human lung toxicity. Beyond this, the G3BP1 knock-in hiPSC line retains full pluripotency, allowing the reporter strategy to be extended to other iPSC-derived organ models and supporting broader NAM efforts toward multi-organ evaluation of early stress response.

Ethics approval statement

All procedures involving human induced pluripotent stem cells were approved by the Public Institutional Review Board of the Ministry of Health and Welfare, Republic of Korea (Approval No. P01-202208-02-010). All animal experiments were approved by the Institutional Animal Care and Use Committee of Chungnam National University (Approval No. 202304A-CNU-059).

CRediT authorship contribution statement

Seung-Yeon Kim: Formal analysis, Investigation, Visualization, Writing – original draft. Ji-Won Baek: Data curation, Investigation. Eo Jin Kim: Data curation, Formal analysis, Software, Writing – review & editing. Sathiyaraj Srinivasan: Data curation, Formal analysis, Software, Writing – review & editing. Kee K. Kim: Conceptualization, Methodology, Supervision, Validation, Writing – review & editing. Eun-Mi Kim: Conceptualization, Funding acquisition, Methodology, Project administration, Resources, Supervision, Validation, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the supplementary data to this article:

Data availability

The raw and processed scRNA-seq datasets have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession code GSE310318.