A self-assembled human lung model captures IPF-like fibrosis and responds to pirfenidone

What the work claims

This is a primary method-and-mechanism paper. Its central claim is that a staged, growth-factor-directed differentiation of human induced pluripotent stem cells (hiPSCs) can generate a multi-lineage lung model in a single Transwell insert without genetic editing or exogenous cell supplementation, and that this model is sufficiently humanized to reproduce key cellular and transcriptomic features of idiopathic pulmonary fibrosis (IPF) when stimulated with transforming growth factor-beta 1 (TGF-beta1).¹ IPF is a progressive scarring disease of the alveolar lung with limited therapies and poor survival; existing organoid models typically capture only one epithelial lineage and rely on added fibroblasts or immune cells to approximate the diseased microenvironment.

The authors report that day-27 models contain proximal ductal epithelium, distal alveolar cells including a small alveolar type II (AT2) population, fibroblasts, endothelial cells and CD68+ macrophage-like cells. Single-cell RNA sequencing identified 13 transcriptional subclusters from 21,132 high-quality cells. TGF-beta1 treatment at 10 ng/mL for 72 hours induced epithelial barrier disruption, ferroptosis-like mitochondrial ultrastructural changes, extracellular-matrix remodeling and elevated secretion of IPF-associated proteins. RNA-seq returned 169 differentially expressed genes, and bidirectional gene-set enrichment analysis showed the model's upregulated genes were significantly enriched in three independent human IPF cohorts. Pirfenidone at 1 mg/mL reduced collagen I, fibronectin and alpha-smooth muscle actin signals and partially restored the AT2 marker pro-SPC.

How it works

The protocol runs for 27 days and is built from four sequential medium switches. hiPSCs are first driven to definitive endoderm with Activin A, BMP4, CHIR99021 and Y-27632 for 4 days; then to anterior foregut endoderm with Noggin and SB431542 for 3 days; then to lung progenitors with FGF-7, FGF-10, BMP4, retinoic acid, EGF and FGF-2 for 10 days; and finally to differentiated lung tissue with FGF-7, FGF-10, dexamethasone, 8-Br-cAMP, IBMX and ITS for another 10 days. The cells are cultured on 0.4 micrometer pore Transwell inserts, which provides an air-liquid interface-like compartmentalization without requiring an explicit air-lift step in the methods.

Characterization combines morphology, ultrastructure and transcriptomics. Hematoxylin and eosin staining showed a biphasic structure: a compact basal pseudostratified epithelial layer with duct-like lumina, and a loose upper stroma. Scanning electron microscopy revealed branched ridges, microvilli-covered spherical cells resembling AT2 cells, flattened areas suggestive of transitional alveolar cells, occasional ciliated cells, spindle-shaped stromal cells and an interwoven fibrous network. Transmission electron microscopy confirmed tight junctions, microvilli, cilia and, after TGF-beta1, disrupted junctions and mitochondria with reduced volume, increased electron density and loss of cristae, a ferroptosis-like appearance.

The single-cell atlas assigns the 13 subclusters to proximal ductal cells (about 41.9% of cells), distal alveolar lineages (about 27.7%), fibroblasts (2.31%), endothelial cells (1.15%) and proliferative epithelial subsets. Notably, no mature ciliated, club or basal airway subclusters were detected; proximal differentiation appeared to stall at a ductal-like stage. After TGF-beta1, quantitative PCR showed significant upregulation of COL1A1, COL3A1, FN1, ACTA2 and CDH2, while classic epithelial-to-mesenchymal transition markers such as VIM, SNAI1 and SNAI2 did not change significantly. The authors interpret this as fibroblast-to-myofibroblast transition rather than epithelial-to-mesenchymal transition. Supernatant ELISAs detected elevated SP-A, SP-D, MMP-7, MMP-1, CCL18, CXCL13, IL-6, IL-8, TNF-alpha and osteopontin.

Transcriptomic validation against GEO datasets GSE17978, GSE47460 and GSE213001 used a bidirectional strategy: IPF-upregulated genes from the patient cohorts were enriched in the model, and model-upregulated genes were enriched in the patient cohorts, with normalized enrichment scores of 1.80, 1.46 and 1.97 respectively. Protein-protein interaction analysis placed COL1A1, FN1 and integrins at the center of the fibrosis network. Pirfenidone co-treatment reduced these readouts without affecting control cultures.

Where a skeptic should push

The strongest assumption is that endogenous co-development in a dish recapitulates the cellular interactions that matter in the human lung. The model does contain multiple lineages, but the proportions are highly skewed: fibroblasts and endothelial cells together make up only about 3.5% of sequenced cells, and genuine AT2 cells are only 0.64%. Macrophage-like cells were detected by immunofluorescence on dissociated cells and by ultrastructure, yet no independent immune-cell cluster survived single-cell sequencing, which the authors attribute to low abundance and detection limits. A fibrotic microenvironment in vivo depends on robust fibroblast and immune populations; whether the sparse mesenchymal and immune representation here is enough to model paracrine signaling over longer timescales is unclear.

A second pressure point is the TGF-beta1 induction itself. TGF-beta1 is a generic profibrotic stimulus, and showing that it upregulates collagen and alpha-SMA in a lung model is expected. The paper's real advance is the transcriptomic match to patient IPF cohorts, which is more specific than a generic fibrosis readout. However, the enrichment is driven by upregulated genes; the model may capture a TGF-beta1-dominated slice of IPF biology but miss IPF triggers that are independent of TGF-beta1. The authors explicitly acknowledge that classical epithelial-to-mesenchymal transition markers are not activated, which is consistent with their interpretation but also means the model does not recapitulate every proposed IPF mechanism.

Third, the work uses a single hiPSC line (iCell-DRY0100) from one Chinese donor. Generalizing the differentiation robustness, lineage proportions and drug response to other lines, ethnic backgrounds and genetic backgrounds is an open question. The authors note that experiments were independently repeated at least three times, which addresses technical reproducibility but not genetic diversity. Finally, pirfenidone reversal is a pharmacological positive control, not a novel drug discovery result; it validates the model but does not by itself advance therapeutics.

What a self-assembled multi-lineage lung model changes for organ models and drug screens

For organoid-based drug discovery, the non-obvious implication is that the bottleneck is no longer simply adding more cell types; it is getting the right cell types to emerge from a common progenitor in the correct spatial context. Models that mix separately differentiated epithelial and mesenchymal cells can model some interactions, but they miss the developmental crosstalk that shapes lineage identity and matrix organization. This protocol's ability to generate proximal duct-like, distal alveolar, fibroblast, endothelial and macrophage-like populations from one starting population suggests that early lung progenitors retain broader mesenchymal and endothelial potential than many directed-differentiation recipes exploit. The mechanistic hypothesis offered, dynamic BMP modulation temporally mimicking in vivo lung development, is testable and points to a design principle rather than a black-box cocktail.

The opportunity is a humanized IPF screen in which fibrosis readouts are tied to a transcriptomic signature that has already been validated against patient cohorts. A program could use the bidirectional GSEA framework as an internal standard: a candidate antifibrotic would be expected to push the model's transcriptome away from the IPF-enriched direction. Because the model responds to a clinically approved drug, pirfenidone, it has a built-in assay control. The partial restoration of AT2 markers is particularly useful, because loss of AT2 identity is increasingly viewed as central to IPF progression, so a drug that preserves or restores the alveolar epithelial program can be prioritized even before in vivo testing.

The threat is over-interpretation of lineage completeness. The model is not a miniature lung: proximal differentiation stalls at ductal-like cells, immune representation is minimal, and the endothelial compartment is a thin trace. A screen run on this model could miss drugs that act on mature airway cells, on immune recruitment, or on vascular remodeling. There is also a real risk of false confidence from the patient-cohort enrichment. The enrichment is statistical and correlative; it validates the model's transcriptional direction but does not prove that TGF-beta1 stimulation reproduces the causal drivers of human IPF. A drug that blocks TGF-beta1-induced collagen in this model might still fail in patients whose fibrosis is driven by alveolar senescence, telomere dysfunction or environmental injury rather than TGF-beta1 hyperactivity.

A second, more subtle threat is donor-line dependence. The entire mechanistic and drug-response dataset comes from one hiPSC line. If the line's differentiation trajectory or TGF-beta1 sensitivity is idiosyncratic, the model could generate reproducible but biologically narrow results. The first follow-up experiments any translation program should demand are replication in additional lines, including IPF-patient-derived hiPSCs, and comparison with primary human lung organoids. Until then, the platform is a promising blueprint for a standardized human IPF model, not a finished replacement for patient tissue.

The bottom line

Established with reasonable confidence: a 27-day staged differentiation of hiPSCs can produce a multi-lineage lung model containing proximal duct-like epithelium, distal alveolar epithelium, fibroblasts, endothelial cells and CD68+ macrophage-like cells; TGF-beta1 induces a fibrosis-like phenotype with ECM deposition, myofibroblast activation, epithelial barrier damage and ferroptosis-like mitochondrial changes; and the resulting transcriptome overlaps significantly with three independent human IPF cohorts. Pirfenidone reverses key fibrotic readouts, confirming pharmacological responsiveness. More tentative: that the model captures the causal disease mechanism rather than a TGF-beta1-driven fibrosis caricature; that the sparse fibroblast and immune representation is sufficient for long-term paracrine modeling; and that results generalize beyond the single hiPSC line tested. The claim would be strengthened by replication in multiple hiPSC lines, direct comparison to IPF-patient-derived cells, and demonstration that a TGF-beta1-independent IPF trigger produces a divergent but equally disease-relevant signature. It would be undercut if the model's transcriptomic match to patient IPF collapses when other profibrotic stimuli are tested, or if the lineage proportions vary so widely across lines that the assay is not reproducible.