Co-optation of the regenerative role of type 2 alveolar cells in distal lung repair in mice by terminal airway epithelial cells in humans

Abstract

The lung is endowed with extensive regenerative capacity. The principal cell of the lung is the type 1 alveolar epithelial (AT1) cell, which mediates gas exchange. Mouse models implicated surfactant-producing type 2 alveolar epithelial (AT2) cells as facultative stem cells that regenerate AT1 cells after injury through a transitional KRT8⁺ intermediate. Larger mammals, however, possess terminal and respiratory bronchioles that are lined by alveoli and by epithelial cells that are absent in mice. Here we show, using human pluripotent stem cell-derived lung organoids and comparative computational analysis, a prime role for terminal and respiratory bronchiole cells in AT1 regeneration without AT2 intermediate. Furthermore, cells similar to aberrant basaloid cells, profibrotic elements that accumulate in pulmonary fibrosis, are physiological intermediates in a more rapidly committing trajectory from terminal and respiratory bronchioles to AT1 cells marked by expression of KRT17, are regulated by Hippo and TGFβ signaling, and are transcriptionally distinct from mouse AT2-derived KRT8⁺ transitional cells. Our findings indicate that terminal and respiratory bronchioles in humans have to a large extent co-opted the regenerative and pathogenic functions of AT2 cells in mice. Efforts to enhance or correct human lung regeneration should therefore focus on facultative airway-derived alveolar progenitors using human models.

INTRODUCTION

The distal lung is endowed with remarkable regenerative capacity that relies on facultative stem cell populations. Most of our insights into human lung regeneration are based on extrapolation from genetic lineage tracing studies in mice, where the surfactant-producing type 2 alveolar epithelial (AT2) cells are the regenerative hubs of the mouse alveolus. AT2 cells, or a subset thereof, can self-renew and convert into AT1 cells, the flat alveolar cells that mediate gas exchange.^1-4 This transition occurs predominantly through a KRT8⁺ population, called ‘pre-AT1 transitional state’,⁵ ‘damage-associated transitional progenitor’,⁶ ‘transitional’,⁷ or ‘alveolar differentiation intermediate’ (ADI) cells⁸ (we will denote these cells here as ‘ADI cells’). ADI cells express senescence and extracellular matrix (ECM) markers and are deemed profibrotic.⁷ AT2 cells can also regenerate from subsets of airway secretory cells (SCs) after severe injury,^6,9-11 and from rare stem cells at the bronchioalveolar junction.^12,13

How these findings translate to human lung regeneration is unclear. AT2 cells derived from human pluripotent stem cells (hPSCs)¹⁴ have been shown to differentiate into AT1-like cells after inhibition of the Hippo pathway.¹⁵ In contrast to mice, where distal airways connect directly to alveolar sacs, lungs in larger mammals contain a zone of terminal and respiratory bronchioles (TRBs), consisting of small airways and alveolar structures that connects the airway system with the alveolar acini.¹⁶ TRBs are lined by cell types (collectively called ‘TRBCs’ here) that share expression of SCGB3A2 with SCs and of SFTPB with AT2 cells.^17,18 A role in regeneration is suggested by the accumulation of various TRBC subsets after lung injury, in chronic obstructive pulmonary disease (COPD) and idiopathic pulmonary fibrosis (IPF).^17-19 Trajectory and organoid studies have yielded conflicting results on their potential, however. Studies using adult lung organoids have suggested that AT2 generate the most distal subset of TRBCs, the AT0 cells, and then give rise to AT1 cells and to other TRBC subsets.¹⁸ Basil et al., on the other hand, indicated based on hPSC-derived airway organoids that TRBC unidirectionally generate AT2 cells.¹⁷ In contrast, we previously showed in hPSC-derived organoids corresponding to TRBCs that TRBCs generate AT1 cells through a pathway that does not involve AT2 cells.²⁰

IPF, a progressive fibrotic disease characterized by aberrant alveolar regeneration, is also characterized by the accumulation of KRT17⁺KRT5⁻ ‘aberrant basaloid cells’ (aBCs).^19,21 aBCs may be profibrotic by virtue of their expression of ECM components and location near fibroblastic foci.^19,22,23 Rare cells of similar phenotype have been reported in normal lungs in one report,¹⁹ suggesting a physiological role. As IPF is widely viewed as a disease caused by profibrotic AT2 dysfunction and as ADI cells and aBCs share expression of senescence and ECM genes, derivation of aBCs from AT2 cells has been suggested.^5,7,8 This notion was supported by studies using precision-cut lung slices²⁴ and hPSC-derived AT2 (iAT2) cells.²⁵ However, our previous studies in organoids corresponding to TRBCs indicated that aBCs can be derived directly from TRBCs, implicating intrinsic TRBC dysfunction in IPF pathogenesis.^20,26

Here, we show that AT2-to-AT1 transition is a much less likely trajectory in humans than in mice. Instead, TRBCs generate AT1 cells in trajectories that include a KRT17⁺ emergency pathway involving aBC-like cells that accumulate as aBCs in IPF. Our observations therefore indicate that TRBCs in humans have largely co-opted the regenerative and pathogenic functions of AT2 cells in mice.

RESULTS

Comparative trajectory analysis of human and mouse lung injury.

We re-analyzed publicly available scRNAseq datasets, subsetted on non-ciliated epithelial cells, from injured lungs in mice and humans (Table 1). We focused on two mouse datasets using bleomycin,^27,28 which causes a transient inflammatory fibrotic response, and one dataset using influenza²⁹ as injury models. As human datasets, we used the IPF/ILD dataset from the Human Lung Cell Atlas (HLCA),³⁰ the IPF dataset of Adams et al.²¹ and Habermann et al.¹⁹ (the latter is also included in the much larger HLCA dataset; both publications were the first to describe aBCs), and a dataset from patients who succumbed to Covid.³¹

In mice (Fig. S1a), partition-based graph abstraction (PAGA)³² placed Krt8⁺ ADI cells between SCs, AT1 and AT2 cells (Fig. 1a). In the influenza model, where injury was presumably more severe, SCs and downstream Krt5⁺Trp63⁺ BC-like cells provided the major input into the ADI subset as evaluated by PAGA (Fig. 1a). These trajectories align with genetic lineage tracing studies.^{3-5,7,8,29,33-35} In the human datasets, TRBCs were present in the distal lung, as previously reported (Fig. S1b-d).^17,18 In the Covid dataset a cluster where a fraction of the cells expressed KRT17 and ITGB6 but not KRT5 was identified, a phenotype consistent with aBCs^19,21 (Fig. S1b,d). We called this fraction ‘aBC-like’. PAGA indicated extensive plasticity among airway cells with connectivity of this aBC-like cluster with TRBCs, SCs and BCs. aBC-like cells in turn showed much closer connectivity to AT1 than to AT2 cells, which appeared as a more isolated population (Fig. 1a). Connectivity with AT1 cells was also supported by the expression of the AT1 marker, SCEL, in the aBC-like cluster (Fig. S1b,d). In the three IPF datasets, AT1 cells clustered among airway subsets (Fig. 1a, Fig. S1c,d), with AT2 cells again as a more isolated population. PAGA placed AT1 cells between AT2 cells and aBCs, without a direct trajectory between the latter (Fig. 1a), suggesting that aBCs are primarily derived from airway subsets, including TRBCs, rather than from AT2 cells (Fig. 1a). All datasets also showed a direct connection from TRBCs to AT1 cells, suggesting that TRBCs can give rise to AT1 cells through at least two trajectories.

Figure 1. In silico comparison of human and mouse lung regeneration.

a. PAGA trajectory analysis from mouse influenza and bleomycin, and human Covid and IPF datasets. b-e. Diffusion pseudotime (DPT) analysis of hypothetical trajectories indicated on top of each panel in mice (b) and human Covid (c,d) and human IPF (e).

We next used diffusion pseudotime (DPT),³⁶ which, by ranking cells according to increasing DPT, informs on the most plausible trajectory. DPT analysis indicated that in the mouse AT2>AT1 conversion is likely, occurring both directly or through the ADI population, with progressively increasing DPT towards AT1 cells along this trajectory (Fig. 1b). In contrast, DPT curves were much more discontinuous along AT2>aBC-like>AT1 and direct AT2>AT1 trajectories in Covid lungs (Fig. 1c), suggesting that AT2>AT1 transition is less likely in humans. Transition from TRBCs to AT1 and AT2 cells, however, yielded DPT curves that were more similar to those for the experimentally proven AT2>ADI>AT1 trajectory in mice (Fig. 1c), although the trajectory towards AT2 cells was more discontinuous than the trajectory towards AT1 cells (Fig. 1c). Derivation of AT1, and possibly to a lesser extent, of AT2 cells from TRBCs is therefore more likely in humans. Although Murthy et al. suggested derivation of TRBCs from AT2 cells,¹⁸ DPT analysis of AT2-to-TRBC conversion yielded entirely disjointed curves, suggesting that this trajectory is unlikely in vivo (Fig. 1d). In all IPF datasets continuity of DPT curves in the trajectories from TRBCs to AT1 cells was largely lost (Fig. 1e). These data are indicative of an AT1 differentiation in IPF.

Together, our analysis suggests that human aBC-like cells are primarily airway-derived AT1-biased progenitors, but confirms the importance of the AT2>ADI>AT1 trajectory in alveolar repair in mice. The inferred AT1 differentiation defect from TRBCs in IPF is consistent with our previously reported studies in human lung organoids.²⁰

A KRT17⁺ injury-associated AT1 differentiation pathway.

We next used TRBCs generated from hPSC-derived lung organoids (LOs) to examine the veracity of the inferred trajectories. We previously reported the generation of TRBCs from hPSCs (both embryonic and induced pluripotent cells) and their differentiation into AT1 cells (Fig. 2a).²⁰ Starting from LOs^37,38 expandable spheres, called ‘induced respiratory airway progenitors’ (iRAPs), are generated that match TRBC populations as defined by Murthy et al.¹⁸ A second stage corresponds to transitional AT2 cells as described by Habermann et al.,¹⁹ where all cells are SCGB3A2⁺ while expressing varying amounts of AT1 and AT2 markers. Finally, plating in 2D in the presence of TGFβ inhibition (TGFβi) and BMP4 generates induced AT1 (iAT1) cells.

Figure 2. KRT17⁺ AT1 lineage from iRAPs.

a. Schematic representation of iRAP and iAT1 differentiation protocol from LOs. b. expression of TRBC markers (n=5), collagens (n=5), senescence (n=3), and AT1 (n=6) markers in purified KRT17-eGFP iRAPs normalized to eGFP−. *p<0.05, **p< 0.01, ****p<0.0001. unpaired two-sided Student’s t-test. c. Schematic of generation of transitional AT2 cells. d-f. eGFP geometric MFI normalized to eGFP⁻ (d), percent HT1-56⁺ (e) and HT2-280⁺ (f) cells among transitional AT2 cells generated from KRT17-eGFP⁺ and KRT17-eGFP⁻ iRAPs. **p < 0.01, ***p < 0.001, ****p < 0.0001, unpaired two-sided Student’s t-test, n=6. g. mRNA expression of AT1 markers (left) and AT2 markers (right) in sorted and unsorted transitional AT2 cells. *p<0.05, **p<0.01 and ****p<0.0001, two-way ANOVA with Tukey’s multiple comparison test, n=3 for total and eGFP⁻, n=2 for eGFP⁺. h. Schematic of generation of iAT1 cells. i. Representative flow cytometry histogram (left) and quantification of eGFP geometric MFI (right) normalized to eGFP⁻ in iAT1 cells derived from KRT17-eGFP⁻ and KRT17-eGFP⁺ iRAPs. *p<0.05, unpaired two-sided Student’s t-test, n=6. j. KRT17 mRNA expression in iAT1 cells derived from KRT17-eGFP⁻ and KRT17-eGFP⁺ iRAPs. *p < 0.05, unpaired two-sided Student’s t-test, n=3. k. Representative flow cytometry histogram (left) and quantification of HT1-56 geometric MFI (right) normalized to eGFP⁻ in iAT1 cells derived from KRT17-eGFP⁻ and KRT17-eGFP⁺ iRAPs. **p<0.01, unpaired two-sided Student’s t-test, n=6. l. mRNA expression of AT1 markers in in iAT1 cells derived from KRT17-eGFP⁻ and KRT17-eGFP⁺ iRAPs. *p<0.05, and ***p<0.001, two-way ANOVA with Šidák multiple comparison test, n=6. m. Fraction of KRT17⁺ AT1 and AT2 cells in normal and injured lungs from the datasets shown in Fig. 1 (two-way ANOVA).

n. IF for KRT17 in human lungs with diffuse alveolar damage.

To investigate the potential of TRBCs to generate aBC-like cells, we generated a KRT17-GFP reporter line (Fig. S2a). When iRAPs were differentiated from this reporter line, they contained a KRT17-eGFP⁺ population (Fig. S2b) that showed overlapping KRT17 immunofluorescence (IF) and eGFP staining by microscopy (Fig. S2c). KRT17-eGFP⁺ cells were quiescent (Fig. S2d), and showed reduced expression of the TRBC markers, SFTPB and SFTPC, as well as increased expression not only of the aBC markers, COL1A1, COL3A1, CDKN2A, CDKN2B and CCL2 but also of the AT1 markers, CAV1 and CLIC5 (Fig. 2b). Bulk RNAseq comparing KRT17-eGFP⁺ and KRT17-eGFP⁻ iRAPs yielded 71 significantly differentially expressed genes that included aBC markers (CCL2 and CDKN2B) and AT1 markers (SCEL and CLIC5) (Fig. S2e). Pathway analysis showed intermediate filaments, keratinization and skin development as significantly induced in the KRT17-eGFP⁺ fraction (Fig. S2f). The combination of aBC and AT1 markers within the KRT17-eGFP⁺ fraction suggested a close relationship between aBC-like and AT1 cells, a relationship that was also inferred computationally (Fig. 1).

We therefore examined the AT1 potential of KRT17⁺ iRAPs in vitro. Purified KRT17-eGFP⁺ and KRT17-eGFP⁻ iRAPs were cultured in transitional AT2 and subsequently in iAT1 conditions. At the transitional AT2 stage (Fig. 2c), flow cytometry (Fig. S2g) showed that cells generated from KRT17-eGFP⁺ iRAPs expressed more KRT17-eGFP than those derived from KRT17-eGFP⁻ iRAPs (Fig. 2d). Furthermore, transitional AT2 cells generated from KRT17-eGFP⁺ iRAPs were enriched in HT1-56⁺ (a human AT1-specific surface marker)³⁹ cells (Fig. 2e, Fig. S2g) and depleted of HT2-280⁺ (a human AT2-specific surface marker)⁴⁰ cells (Fig. 2f, Fig. S2g). Consistent with these observations, they expressed more mRNAs encoding AT1 markers and fewer mRNAs encoding the AT2 markers than those derived from KRT17-eGFP⁻ iRAPs (Fig. 2g). Transitional AT2 cells derived from KRT17-eGFP⁺ iRAPs therefore appeared AT1-primed.

iAT1 cells (Fig. 2h) generated from KRT17-eGFP⁺ iRAPs expressed more KRT17-eGFP (Fig. 2i) and KRT17 mRNA (Fig. 2j) compared to those generated from KRT17-eGFP⁻ iRAPs, but expressed less HT1-56 (Fig. 2k) and lower levels of mRNAs encoding AT1 markers (Fig. 2l). KRT17-eGFP⁻ iRAPs also generated iAT1 cells expressing KRT17, though at lower levels than KRT17-eGFP⁺ iRAPs (Fig. 2i). KRT17-eGFP⁻ AT2-primed transitional cells can therefore transition to KRT17⁺ iAT1 cells as well, at least in vitro, albeit less efficiently.

To verify existence of KRT17⁺ AT1 cells in vivo, we analyzed the datasets in Fig. 1. KRT17⁺ AT1, but not KRT17⁺ AT2 cells, were elevated in Covid and IPF datasets and were virtually absent in control datasets (Fig. 2m). KRT5, which is co-expressed with KRT17 in airways BCs but not in aBCs or aBC-like cells, was nearly undetectable in AT1 and AT2 cells. IF revealed locoregional presence of KRT17⁺ AT1 cells in severely injured areas of lungs from patients with acute respiratory distress syndrome, but absence in normal controls (Fig. 2n). These findings indicate the existence of an injury-associated KRT17⁺ AT1 differentiation pathway.

We next examined whether KRT17-eGFP⁺ iRAPs were primed for more rapid generation of AT1 cells. We previously showed that optimal iAT1 generation requires an intervening ‘transitional AT2 stage’, where cells express SCGB3A2 together with AT1 and AT2 markers.²⁰ When iRAPs were plated directly in 2D iAT1 conditions (direct pathway) (Fig. 3a), the resulting iAT1 cells indeed expressed lower amounts of mRNAs encoding AT1 markers (Fig. 3b) compared to iAT1 cells generated through the stepwise pathway, but appeared more pure as evaluated by flow cytometric analysis of HT1-56 staining (Fig. 3c) and expressed more KRT17-eGFP (Fig. 3d). Omitting the transitional AT2 stage therefore favors the KRT17⁺ AT1 lineage. We next plated purified KRT17-eGFP⁺ and KRT17-eGFP⁻ iRAPs directly in iAT1 conditions (Fig. 3e). iAT1 cells directly generated from KRT17-eGFP⁺ iRAPs expressed more KRT17-eGFP⁺ (Fig. 3f) and KRT17 mRNA (Fig. 3g) than those directly generated from KRT17-eGFP⁻ iRAPs. Within the direct pathway, iAT1 cells derived from KRT17-eGFP⁺ iRAPs expressed more AGER and CLIC5 than those generated from KRT17-eGFP⁻ iRAPs, although expression of two other AT1 markers, CAV1 and RTKN2, was similar (Fig. 3h). The direct pathway therefore not only favors a KRT17⁺ fate, but KRT17⁺ iRAPs also engage the direct pathway more efficiently.

Fig. 3. Direct and stepwise AT1 differentiation pathways.

a. Schematic of direct differentiation pathway. b. mRNA expression of AT1 markers in stepwise versus direct iAT1 cells. *p < 0.05, **p < 0.01, and ****p < .0001, two-way ANOVA with Šídák's multiple comparisons test, n=5 (stepwise) and n=6 (direct). c. Representative flow cytometry histogram (left) and quantification of HT1-56 geometric MFI (right) normalized to stepwise iAT1 cells. ****p < 0.0001, unpaired two-sided Student’s t-test, n=6. d. Representative flow cytometry histogram (left) and quantification of eGFP geometric MFI (right) normalized to stepwise iAT1 cells. *p < 0.05, unpaired two-sided Student’s t-test, n=6. e. Schematic of experimental design. f. Representative flow cytometry histogram (left) and quantification of eGFP geometric MFI (right) normalized to eGFP⁻ sorted direct iAT1 cells. *p < 0.05, unpaired two-sided Student’s t-test, n=3. g. mRNA expression of KRT17 in sorted direct iAT1 cells. *p < 0.05. unpaired two-sided Student’s t-test, n=3. h. mRNA expression of AT1 markers in sorted direct iAT1 cells. *p < 0.05, unpaired two-sided t-test, n=6. i. Feature plots for AT1, AT2 and TRBC markers as well as KRT17 in hashed scRNAseq analysis of iAT1 cells generated through direct and stepwise pathways. j. Corresponding dotplot. k. Feature plots for top differentially regulated genes in the direct pathway in hashed scRNAseq analysis of iAT1 cells generated through direct and stepwise pathways. l. Corresponding dotplot. m. Feature plots for AT1, AT2 and TRBC markers as well as KRT17 in immature AT1 cells from the Covid dataset. n. Corresponding dotplot. o. Top differentially regulated genes in the direct pathway in immature AT1 cells from the Covid dataset. p. Corresponding dotplot.

We compared iAT1 cells generated through either pathway by hashed scRNAseq. iAT1 cells generated through the direct pathway expressed more SCEL and KRT17, but less RTKN2, CAV1 and AGER than cells generated through the stepwise pathway, consistent with RT-qPCR data (Fig. 3i,j). A subset of cells generated through the stepwise pathway co-expressed AT1 markers with SFTPB and SCGB3A2, a finding aligned with derivation and slower commitment from TRBCs over transitional AT2 cells to AT1 cells (Fig. 3i,j). The most upregulated pathways encoded by the 200 top-upregulated genes in directly generated iAT1 cells included cell adhesion, wound healing and TGFβ production, pathways consistent with an injury response (Fig. S3a). The top-upregulated genes in the direct pathway (Fig. 3k,l) included ANKRD1 and AREG, both established canonical direct targets of Hippo signaling,^41,42 suggesting that Hippo signaling may drive the direct pathway.

To verify the existence of the direct pathway in human lungs, we examined expression of the differentially expressed genes between iAT1 cells derived from the direct and stepwise pathway in AT1 cells in Covid samples (too few AT1 cells were present in the IPF datasets for meaningful analysis). The AT1 subset contained three clusters, each highly expressing either CAV1, RTKN2 or AGER (clusters, 2, 5 and 6, respectively), suggesting heterogeneity among AT1 cells with respect to expression of canonical AT1 markers (Fig. S3b). A small cluster (3) of contaminating SCs (SCGB3A2⁺SCGB1A1⁺SFTPB⁻) was also detected. Three additional clusters (0, 1, 4) expressed SCEL, variable levels of SFTPB and lower levels of the aforementioned AT1 markers (Fig. S3b), suggesting immaturity. Subsetting this likely immature AT1 fraction revealed three clusters (Fig. S3b). A first cluster contained cells expressing SFTPB, SCGB3A2 and SFTPC as well as AT1 markers, a gene expression signature consistent with transitional AT2 cells (Fig. 3m,n).^19,20 A second, closely positioned cluster did not express SCGB3A2 or SFTPC but expressed more SFTPB, CAV1, AGER and RTKN2 than a third cluster, which expressed more KRT17 and SCEL. These data indicate that the second cluster corresponds to the stepwise pathway (~stepwise), while the third cluster corresponds to the direct pathway (~direct) (Fig. 3m,n). Indeed, all genes upregulated in the direct pathway in vitro were also elevated in the ~direct cluster relative to the ~stepwise cluster in vivo (Fig. 3o,p). A similar dichotomy among immature AT1 cells can therefore be discerned in vitro and in vivo. Furthermore the top-differentially regulated genes in the direct pathway were also enriched in aBCs in IPF datasets and in aBC-like cells in the Covid dataset (Fig. S3c), indicating close association of aBCs with the direct pathway.

Collectively, functional in vitro studies combined with computational analysis and imaging of human injured lungs indicate the existence of an injury-associated, likely Hippo-driven AT1 differentiation pathway through aBC-like cells from TRBCs that at least transiently expresses KRT17, is more prone to differentiation through the direct differentiation pathway, indicative of more rapid commitment, and yields AT1 cells expressing lower levels of canonical AT1 markers except for SCEL, suggesting generation of a specific AT1 subset.

aBC-like cells are driven by Hippo and TGFβ.

To gain mechanistic insight into the dual nature of KRT17-eGFP⁺ iRAPs, which share features of IPF-associated aBCs and of AT1 progenitors, we exposed iRAPs to TGFβ, which drives fibrosis, and to Hippo inhibition, which drives AT1 regeneration and directly regulates expression of the two top-induced genes in the direct pathway, ANKRD1 and AREG.^41,42

Hippo signaling phosphorylates the kinases, LATS1 and LATS2, which in turn phosphorylate the homologues YAP and TAZ, resulting in cytoplasmic retention. Inhibition of LATS induces nuclear translocation of YAP and TAZ where they associate with the transcriptional regulators, TEAD1-4, which play an essential role in AT1 regeneration after injury.^43,44,44-47 IF revealed that ~60% of nuclear (n)YAP⁺ iRAPs expressed KRT17-eGFP⁺, whereas on only ~10% of nYAP⁻ iRAPs did (Fig. 4a,b). nYAP is therefore largely associated with KRT17-eGFP⁺ iRAPs. In the presence of a small molecule LATS inhibitor (LATSi)⁴⁸ virtually all cells became nYAP⁺, whereas TGFβ1 abrogated nYAP expression (Fig. 4a,c). Both LATSi (Fig. 4a,d) and TGFβ1 (Fig. 4a,e) increased the fraction of KRT17-eGFP⁺ cells, however. Conversely, small molecule inhibition of the interaction between YAP and TEAD (TEADi) (Fig. 4d) as well as TGFβi (Fig. 4e) reduced the KRT17-eGFP⁺ fraction.

Figure 4. Regulation of KRT17 by TGFβ and Hippo signaling.

a. IF representative of 3 biological replicates of KRT17-eGFP iRAPs in vehicle (top), LATSi and TGFβ1 (bottom) stained for markers on top of the panel. b. Fraction of nYap⁺ and nYAP⁻ cells in vehicle condition that express KRT17. **p < 0.01, unpaired two-sided Student’s t-test, n=3. c. Fraction of nYAP⁺ cells after addition of LATSi or TGFβ1. ****p < 0.0001, one-way ANOVA with Tukey’s multiple comparison test, n=3. d. Representative flow cytometry histogram (left) and quantification of fraction of KRT17⁺ cells (right) in the presence of LATSi or TEADi. *p < 0.05, ****p < 0.0001, one-way ANOVA with Holm-Šídák's multiple comparisons test, n=5. e. Representative flow cytometry histograms (left) and quantification of fraction of KRT17⁺ cells (right) in the presence of TGFβ1 or TGFβi. *p < 0.05, **p < 0.01, one-way ANOVA with Dunnett’s multiple comparison test, n=6. f. mRNA expression of AT1 markers in LATSi and vehicle treated iRAPs. **p < 0.01, ***p < 0.001, ****p < 0.0001, one-way ANOVA with Tukey’s multiple comparison test, n=6 (KRT17) and n=3 (remaining). g. mRNA expression of aBC markers in LATSi and vehicle treated iRAPs. *p < 0.05, **p < 0.01, one-way ANOVA with Tukey’s multiple comparison test, n=6 (COL1A1 and COL3A1) and n=3 (remaining). h. mRNA expression of TRBC markers in LATSi treated iRAPs over multiple passages. *p < 0.05, **p < 0.01, ***p < 0.001,****p < 0.0001, two-way ANOVA with Dunnett’s multiple comparison test, n=3. i. mRNA expression of AT1 markers in stepwise iAT1 cells from vehicle and LATSi treated cells. *p < 0.05, unpaired two-sided Student's t-test, n=3 (left). Quantification of cell numbers retrieved from seeding 20K cells per condition. *p < 0.01, unpaired two-sided Student's t-test, n=3 (right). j. mRNA expression of AT1 markers in direct iAT1 cells from vehicle and LATSi treated cells. Two-way ANOVA with Šidák multiple comparison test, n=6 (left). Quantification of cell numbers retrieved from seeding 20K cells per condition. **p < 0.01, unpaired two-sided Student's t-test, n=6 (right). k. mRNA expression of aBC markers in TGFβ1 and TGFβi treated cells. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, two-way ANOVA with Dunnett’s multiple comparison test, n=6.

While both LATSi and TGFβ signaling induced KRT17, the nature of the KRT17-eGFP⁺ cells differed. LATSi enhanced cellular expansion (Fig. S4a), increased expression of AT1 markers (Fig. 4f), decreased expression of some aBC markers (COL1A1, CDKN2A, CCL2) (Fig. 4g), and, over several passages, decreased expression of TRBC markers (Fig. 4h). LATSi changed iRAP morphology from individual spheres to a ‘bunch of grapes’ (Fig. S4b). Removing LATSi changed morphology back to individual spheres (Fig. S4b), reduced KRT17-eGFP and HT1-56 expression (Fig. S4c,d), and restored AT1 and TRBC marker expression over several passages (Fig. S4e). We next examined the effect of LATSi on the direct and the indirect pathways. After plating in direct iAT1 conditions, LATSi-treated iRAPs generated more iAT1 cells with similar expression of all AT1 marker mRNAs compared to control iRAPs (Fig. 4i). Conversely, after differentiation through the stepwise pathway, LATSi-treated iRAPs yielded fewer iAT1 cells with reduced AT1 marker expression compared to vehicle-treated iRAPs (Fig. 4j). Hippo inhibition therefore reversibly drives KRT17⁺ AT1 differentiation preferentially through the more rapid direct pathway from TRBCs.

In contrast to LATSi, TGFβ1 halted proliferation (Fig. S4f). TGFβ1 increased, while TGFβi decreased, expression of canonical aBC markers (Fig. 4k). TGFβ1 also reduced expression of TRBC markers (Fig. S4g). Furthermore, TGFB1 and TGFBR1 mRNAs were elevated in KRT17-eGFP⁺ cells (Fig. S4h), while the amount of TGFβ1 was higher in the supernatant of purified KRT17-eGFP⁺ than KRT17-eGFP⁻ cells (Fig. S4i). Both endogenous and exogenous TGFβ therefore induce KRT17-eGFP⁺ aBC-like cells. Replating TGFβ1-induced KRT17-eGFP⁺ cells in iRAP conditions, both the presence or absence of TGFβi, reduced expression of KRT17-eGFP (Fig. S4j) and of aBC markers (Fig. S4k) while restoring expression of TRBC markers (Fig. S4l). TGFβ therefore reversibly induces proliferation arrest and expression of canonical aBC markers in iRAPs.

As, at variance with our findings, AT1 cells^1,2 and aBCs^7,24,25 are assumed to be descended from AT2 cells, we examined the effect of TGFβ and LATSi in AT2 cells. We procured human iAT2 cells, generated by repeated isolation and expansion of tdTomato⁺ SFTPC-reporter expressing cells in the same conditions as iRAPs.^14,49 While expressing elevated levels of SFTPC, iAT2 cells expressed similar amounts of SCGB3A2 mRNA (Fig. S5a) and protein (Fig. S5b) as iRAPs. When differentiated into iAT1 cells either according to our protocol (BMP4 and TGFβi)²⁰ or to the protocol published by the same group (DCI and LATSi),¹⁵ the cells maintained elevated SFTPC (Fig. S5c) and tdTomato (Fig. S5d) expression. Similarly, TGFβ1 upregulated aBC genes in iAT2 cells (Fig. S5e), but only slightly downregulated TRBC genes in those compared to iRAPs (Fig. S5f). TGFβ1 downregulated HT2-280 expression (Fig. S5g,h), but did not downregulate TdTomato (Fig. S5i). iAT2 cells are therefore more similar to iRAPs than to bona fide AT2 cells, but may be epigenetically dysregulated because of repeated reporter selection. A determination of whether AT2 cells can generate aBCs can therefore not be made. Our computational analysis (Fig. 1), however, suggests that derivation of aBCs from AT2 cells is less likely.

Cross-species comparison aBCs, aBC-like and ADI cells.

To gain insight into the molecular differences between TGF-β1 and LATS-induced KRT17-eGFP⁺ cells, we performed hashed scRNAseq analysis (Fig. 5a). TGFβ1-treated cells showed increased expression of aBC markers, while LATSi-treated cells expressed more AT1 markers. The TRBC markers, SCGB3A2, SFTPB and SFTPC, were reduced in both conditions (Fig. 5b). These findings confirm RT-qPCR data (Fig. 4f,g,h,k, Fig. S4g). Although the top-200 TGFβ induced genes included the ECM components, FN1 and COL1A1, and the senescence marker, CDKN2A, regulation of axon guidance, cell migration and motility were the top-induced processes (Fig. S6a). In LATSi-treated cells, on the other hand, a non-overlapping set of upregulated genes included the AT1 markers, AGER and CLIC5, with epithelial development, wound healing and cell-substrate adhesion as top-induced processes (Fig. S6a). Live imaging confirmed increased motility of control TGFβ1-induced KRT17-eGFP⁺ cells, while LATSi-induced cells were immobile (Supplementary Movies 1-3).

Figure 5. TGFβ and LATSi scores.

a. UMAP of hash-tagged scRNAseq analysis of iRAPs colored by treatment (vehicle, TGFβ1 or LATSi). b. Dot plot of TRBC, AT1 and aBC markers in vehicle, TGFβ or LATSi-treated iRAPs. c. Expression of TGFβ and LATSi signature genes in all annotated epithelial cells types the LungMap IPF dataset with dendrogram. d. Dot plot of TGFβ1 and LATSi scores in WT and HPS1^−/− iRAPs. e. UMAP feature plots of integrated scRNAseq analysis of WT and HPS1^−/− iRAPS for TGFβ and LATSi scores and for KRT17. f. TGFβ and LATSi score all annotated epithelial cell types identified in datasets analyzed in Fig. 1. g. DPT graph with TGFβ and LATSi scores within aBC-like cells, aBCs and ADI cells in datasets analyzed in Fig. 1.

We extracted TGFβ and LATSi gene signatures comprising the 50 top-upregulated genes in TGFβ1 and LATSi-treated iRAPs, respectively (Fig. S6b). Based on these signatures, we determined TGFβ and LATSi scores. We verified both scores in integrated single cell transcriptomic analysis of WT iRAPs and iRAPs mutant for the Hermansky-Pudlak Syndrome 1 (HPS1) gene, which causes early-onset IPF with high penetrance.^50,51 We previously showed accumulation of aBCs and a defect in AT1 differentiation in HPS1^−/− iRAPs.²⁰ Not only the TGFβ score, but also the LATSi score was upregulated in HPS1^−/− compared to WT iRAPs (Fig. 5c,d, Fig. S6c). The data suggest that aBCs are in fact characterized by both elevated TGFβ and LATSi scores. To verify this finding in vivo, we examined expression of both signatures in the LungMap atlas.⁵² In normal lungs, the LATSi signature was associated with AT1 cells, as expected, whereas the TGFβ signature was not clearly associated with any specific population (Fig. S6d). In the LungMap IPF dataset, on the other hand, the TGFβ signature was selectively and highly expressed in aBCs (Fig. 5e). aBCs, however, clustered next to AT1 cells with respect to expression of LATSi signature genes (Fig. 5e). We then examined the TGFβ and LATSi scores in the lung injury datasets described in Fig. 1 (Fig. 5f). Within each dataset, aBC-like cells (Covid) and aBCs (IPF) showed the highest TGFβ score, as well as elevated LATSi score, with the latter the highest in AT1 cells in most datasets. These observations demonstrate that elevated TGFβ and LATSi scores are a feature of aBCs in vivo as well. Mouse ADI cells also showed enrichment in both scores, although the TGFβ score appeared less specific for ADI cells in mice than for aBCs or aBC-like cells in humans (Fig. 5f). Plotting TGFβ and LATSi scores in aBCs, aBC-like and ADI cells ordered according to DPT showed mostly out-of-phase waxing and waning of both scores (Fig. 5g), explaining why the aBC-like and aBC clusters show enrichment of both scores. We conclude that aBCs in HPS1^−/− iRAPs and IPF lungs, aBC-like cells in Covid lungs, as well as to some extent mouse ADI cells, all share elevated TGFβ and more mildly elevated LATSi scores.

We next used these scores to define aBC-like populations in control datasets and in a ferret injury model, where KRT17 is not annotated. A small population with the highest TGFβ score within each dataset as well as elevated LATSi score, which is highest in AT1 cells, was present in all human (Fig. 6a), but not in any mouse control datasets (not shown). In lungs of bleomycin-treated ferrets,⁵³ which are endowed with TRBs,^16,53 a TRBC population could be identified based on expression of SC markers and SFTPB. In contrast to human lungs, however, SCGB1A1, not SCGB3A2, was the predominant SC marker in SFTPB⁺ cells in the distal lung (Fig. 6b), suggesting reversal of the expression patterns of SCGB1A1 and SCGB3A2 in TRBCs or misannotation in the ferret genome. A distinct TP63⁻KRT5⁻ aBC-like population with the highest TGFβ score and ranking second behind AT1 cells for the LATSi score was identified. PAGA mapped this population in between TRBCs and AT1 and AT2 cells, a position similar to that of ADI cells in mouse injury models (Fig. 6b). DPT analysis (Fig. 6b) furthermore suggests that AT2>AT1 transition is more likely than TRBC>AT1 transition. The highest DPT on any trajectory towards AT1 cells that included aBC-like cells was reached in aBC-like cells, suggesting a differentiation block (Fig. 6c). Lung regeneration in ferrets therefore shares features with both humans and mice.

Figure 6. TGFβ and LATSi scores in homeostasis and across injury models.

a. UMAP plots colored by cell type, dot plot of TGFβ and LATSi scores and select markers, and PAGA trajectory analysis in control datasets of Covid and IPF datasets analyzed in Fig. 1. b. UMAP plot colored by cell type, dot plot of TGFβ and LATSi scores and select markers, feature plots of TGFβ and LATSi scores and PAGA trajectory analysis for ferret bleomycin dataset from Wu et al. (2025). c. DPT analysis of the dataset in panel (b). d. Heatmap of TGFβ signature gene expression in aBCs, aBC-like and ADI cells, scaled to 1 across analyzed datasets, with dendrogram.

Finally, as the TGFβ score was the most defining for aBCs, aBC-like and ADI cells, we compared these subsets with respect to this score, with inclusion of WT and HPS1^−/− iRAPs, by clustering based on identically scaled raw expression values of TGFβ signature genes (Fig. 6d). IPF aBCs and HPS1^−/− iRAPs clustered together, further validating HSPS1^−/− iRAPs as an aBC model. aBC-like cells in human Covid and the ferret bleomycin model and mouse ADI cells formed separate clades. Human and ferret aBC-like cells are therefore transcriptionally similar to each other, but distinct from ADI cells, and both differ from aBCs in IPF, which have the most uniform expression of the TGFβ score.

DISCUSSION

We show here that the regenerative trajectories in human and mouse lungs are incompletely conserved. While the major role of AT2 cells in alveolar regeneration in mice is undisputed, the regenerative role of AT2 cells in humans is less clear, as computational analysis indicates that after lung injury TRBCs, which possess AT1 potential in vitro, are more likely AT1 progenitors than AT2 cells. AT2 cells may therefore be a more autonomous population that can migrate into injured areas,⁵⁴ although a role in alveolar maintenance cannot be excluded. TRBCs may also have AT2 potential, although we have not been able to generate bona fide AT2 cells from iRAPs in vitro. We note that iAT2 cells^14,49 appear more similar to iRAPs and therefore to TRBCs than to AT2 cells in our hands. Computational analysis indicates that a trajectory suggested by Murthy et al.,¹⁸ AT2 cells to TRBCs, is unlikely, however. Although ferrets are endowed with TRBCs, computational inference suggests that AT2>AT1 transition is, similar to the mouse, the most plausible trajectory. Furthermore, aBC-like cells in the ferret bleomycin model show similar lineage relations as ADI cells in the mouse influenza and bleomycin models as evaluated by PAGA. They cluster, however, with human aBC-like cells with respect to the TGFβ signature. Ferrets therefore share features with mice and humans with respect to the response to lung injury. Collectively, our findings indicate that TRBCs have progressively co-opted the regenerative functions of AT2 cells as they arose during evolution, thus providing a parsimonious explanation for the discrepancy in lung regeneration between mice and humans, with ferrets sharing features of both. It is tempting to speculate that the evolution of regenerative dynamics in the distal lung was an adaptation to the evolution of lung size in larger mammals.

Our study identifies aBCs and aBC-like cells, defined by elevated TGFβ and LATSi scores, as physiological AT1 precursors derived from airway cells. While we demonstrated derivation from TRBCs using the iRAP model, PAGA also suggests a connection with proximal airway cells, especially in IPF, that merits further exploration. aBCs and aBC-like cells transition between elevated LATSi or TGFβ signatures over pseudotime, a feature shared with mouse ADI cells. As the TGFβ signature confers migratory capacity whereas Hippo inhibition drives AT1 differentiation, migration and differentiation appear temporally mutually exclusive. In line with this idea, AT1 differentiation from iRAPs requires inhibition of endogenous TGFβ signaling.²⁰ After injury, a fraction of AT1 cells expresses KRT17, consistent with transit through a KRT17⁺ aBC-like cell in this trajectory. While not exclusive to KRT17⁺ cells, the more rapid direct AT1 pathway favors and is favored by KRT17⁺ cells and yields AT1 cells with higher SCEL expression, but lower expression of other AT1 markers. This dichotomy, identified functionally in the iRAP model, was computationally validated in human lung injury. The dominance of the KRT17 pathway in vitro is likely a reflection of the fact that cytokine-supported organoids recapitulate injury rather than homeostasis. It is possible that KRT17⁺ AT1 cells will differentiate into KRT17⁻ AT1 cells. Alternatively, they may ultimately be replaced by mature, canonical AT1 cells generated through the KRT17⁻ pathway. In this interpretation, the KRT17⁺ trajectory represents a rapidly committing emergency AT1 regeneration pathway. Consistent with this notion and with the role of Hippo signaling in AT1 regeneration in mice,⁴³ Hippo signaling preferentially drives this pathway in humans. The existence of an emergency AT1 differentiation pathway is reminiscent of emergency megakaryopoiesis in the hematopoietic system, where, especially in conditions of hematopoietic stress, megakaryocytes are derived directly from hematopoietic stem cells instead of the passing through a stepwise differentiation pathway, and generate hyperreactive platelets.^55-58

Faithfully recapitulating IPF in mouse models has remained elusive.^59-62 Our observations support the validity of iRAPs to model key aspects of IPF, including aBC generation and impaired AT1 differentiation, but call into question the validity of the widely used mouse bleomycin model to study IPF. Mouse ADI cells differ from human aBCs and aBC-like cells in terms of transcriptomic profile, confirming a previous report using comparative spatial transcriptomics,²³ and in terms of lineage relation with AT2 cells, with ADI cells primarily descended from AT2 cells whereas our observations collectively indicate that derivation of aBCs from AT2 cells in humans is unlikely. The ferret bleomycin model⁵³ is more similar to the mouse model in terms of lineage relations of mouse ADI and ferret aBC-like cells, but more similar to human Covid with respect to the TGFβ signature in aBC-like cells. The ferret bleomycin model therefore recapitulates some aspects of human lung injury but not IPF.

The concept that TRBCs constitute an important source of alveolar repair in humans has important implications for our understanding of IPF as it indicates a major role for abnormalities in TRBCs, while not excluding a contribution of defects in AT2 cells, in pathogenesis. This idea is supported by the fact that progression of DPT in TRBCs towards AT1 cells was abnormal in IPF, that loss of AT1 cells²² and morphological changes in small airways^63-65 are the earliest changes in the fibrotic process, that HPS1^−/− iRAPs, which do not contain mature AT2 cells, spontaneously generate aBCs while AT1 differentiation was impaired,²⁰ and that aBCs are connected to airway cells and not to AT2 cells in PAGA analysis of IPF lungs and, finally, that HPS1-mutant LOs show spontaneous fibrosis despite absence of mature AT2 cells.^37,66 A critical role of TRBCs would also explain why recapitulating IPF in mice is challenging, given the absence of TRBCs in rodents.^16-18

In conclusion, the cellular mechanisms involved in alveolar regeneration in humans are distinct and more complex than in mice. This study validates hPSC-derived LOs and iRAPs, which fit the definition of ‘new approach methodologies' (NAMs),⁶⁷ as reliable cell-based models to dissect the unique features of normal and aberrant lung regeneration in humans and will be invaluable in efforts that focus on facultative airway-derived alveolar progenitors to enhance or correct lung regeneration.

Methods

Human lung samples

Deidentified human lung samples were provided by the Herbert Irving Comprehensive Cancer Center Molecular Pathology Shared Resource Tissue Bank core (Institutional Review Board AAAB2667). Normal controls were pathologically normal tissue from pneumothorax patient samples. Acute lung injury samples had a pathological diagnosis of diffuse alveolar damage.

Human pluripotent stem cells (hPSCs)

Maintenance.

Rockefeller University ES cell line 2 (passage 24–32) were maintained on mouse embryonic fibroblasts (MEFs) plated at 22,500 cells per cm². Cells were cultured in human ES (hES) cell maintenance medium (DMEM/F12 (Thermo Fisher Scientific), 20% knockout serum (Stem Cell Technologies), 0.1 mM β-mercaptoethanol (Sigma-Aldrich), 0.2% Primocin (InvivoGen), 20 ng ml⁻¹ FGF2 (R&D Systems/Biotechne) and 1% Glutamax (Thermo Fisher Scientific)), which was changed daily. Human ES cells were passaged every 3–4 days with Accutase (Innovative Cell Technologies) at least two times before differentiation, washed and replated at a dilution of 1:24. Cultures were maintained in a humidified 5% CO2 atmosphere at 37 °C. Lines were karyotyped and verified for Mycoplasma contamination using PCR (InVivoGen) every 6 months.

Generation of LOs.

Human LOs were generated as described previously. Briefly, MEFs were depleted by passaging 5–7 × 10⁶ ES cells onto Matrigel (Corning) coated 10-cm dish. Cells were maintained in hES cell medium in a humidified 5% CO2 atmosphere at 37 °C. After 24 h, cells were detached with 0.05% Trypsin/EDTA (Thermo Fisher Scientific) and distributed to six-well low-attachment plates containing primitive streak/embryoid body (EB) medium (10 μM Y-27632 (Tocris/Biotehne) and 3 ng ml⁻¹ BMP4 (R&D/Biotechne)) to allow EB formation. EBs were fed every day with fresh endoderm induction medium (10 μM Y-27632, 0.5 ng ml⁻¹ BMP4, 2.5 ng ml⁻¹ FGF2 and 100 ng ml⁻¹ Activin A (R&D/Biotechne)) and maintained in a humidified 5% CO2/5% O2 atmosphere at 37 °C. Endoderm yield was determined by dissociating EBs and evaluating CXCR4 and c-KIT (Biolegend) co-expression by flow cytometry on day 4. Cells used in all experiments had >90% endoderm yield and were plated on 0.2% fibronectin-coated wells (R&D/Biotechne) at a density of 80,000 cells per cm². Cells were incubated in anteriorization medium 1 (100 ng ml⁻¹ Noggin (R&D/Biotechne) and 10 μM SB-431542 (Tocris)) for 24 h, followed by anteriorization medium 2 (10 μM SB-431542 (Tocris) and 1 μM IWP-2 (Tocris)) for another 24 h. At the end of anterior foregut endoderm induction, cells were switched to ventralization/branching medium (3 μM CHIR-99021 (Tocris), 10 ng ml⁻¹ FGF10 (R&D/Biotechne), 10 ng ml⁻¹ recombinant human (rh)KGF (R&D/Biotechne), 10 ng ml⁻¹ BMP4 and 50 nM all-trans-retinoic acid (Tocris)) for 48 h and 3D clump formation was observed. The adherent clumps were detached by gentle pipetting and transferred to a low-attachment plate, where they folded into lung bud organoids (LBOs) at days 10–12. The branching medium was changed every other day until days 20–25 and LBOs were embedded in 100% Matrigel in 24-well transwell (BD Falcon) inserts. The branching medium was added after the Matrigel solidified and changed every 2–3 days. Embedded LOs can be maintained for more than 6 months.

Generation of KRT17-eGFP reporter line.

Alt-R^™ technology (IDT) was used to design crRNAs and perform nucleofection. crRNA (100 μM) targeting the sequence 5'-GTCCACCAGACCACCCGCTG-3' was combined at equal molar amounts (200 pmols) with Alt-R^™ CRISPR-Cas9 tracrRNA-ATTO^™ 550 (200 μM) and annealed at 95°C for 5 min in a thermocycler (Bio-Rad). The annealed duplex was then combined with Alt-R^™ S.p. Cas9 Nuclease V3 (IDT) at a 3:1 molar ratio (duplex:Cas9) and incubated at room temperature for 10–20 min to form the ribonucleoprotein (RNP) complex. RUES2 Cells (passage 22) were maintained on MEF-coated 6-wells until reaching ~80% confluency. The cells were then collected with Accutase, washed, and resuspended in PBS. 1 x 10⁶ cells were transferred to a new tube, pelleted, and resuspended in 25 μL of P3 primary cell solution supplemented with supplement 1, as described by the manufacturer (Lonza, P3 Primary Cell 4D-Nucleofector Kit). To this, 1.25 μL of the RNP complex was added, as well as Alt-R^™ Cas9 Electroporation Enhancer at a final concentration of 1.75 μM (IDT), and 150 pmol of HDR donor Oligo (IDT, Supplementary Data Table 2). Cells were then transferred to cuvettes provided in P3 Primary Cell 4D-Nucleofector Kit, and nucleofection was performed on an Amaxa 4D-nucleofector using hESC preset settings. Cells were replated onto MEF-coated 6-well plates in hES maintenance medium supplemented with 1.38 μM Alt-R^™ HDR Enhancer V2 (IDT). After 12–24 h, medium was replaced with unmodified hES maintenance medium. After an additional 24 h, cells were collected with Accutase, stained for human CD90 (BioLegend), and single-cell sorted into MEF-coated 96-well plates (2–3 plates per sample) using a Sony MA900 cell sorter. Plates were centrifuged briefly at 100g and returned to a normoxic incubator for 2 weeks for colony formation, with medium changed every other day. Clones were collected and genomic DNA was isolated for genotyping by PCR using primers KRT17 F1 InGene (5'-CCCCAGACTTACTCTTTATGCC-3') and KRT17 Insert R1 (5'-GGAGGATGGTAATATCTTGGGG-3'). PCR products were submitted to Genewiz for Sanger sequencing.

Generation of Induced Respiratory Airway progenitors (iRAPs).

Human lung organoids (LOs)were first generated as described in Matkovic Leko et al. (2023). Matrigel-embedded LOs are ready for iRAP generation on day 42 of development. The medium was removed from the transwell and 1 ml of 2 mg ml⁻¹ dispase (Corning) was added to release LOs from the Matrigel for 30–45 min in a normoxic incubator. The LOs were transferred to a 15-ml conical tube and washed with wash media to neutralize dispase and then centrifuged at 200g for 5 min. The pellet was incubated with 1 ml of 0.05% trypsin–EDTA in a normoxic incubator for 10–12 min with occasional pipetting with a P1000 pipette tip. Single-cell dissociation was verified using a bright-field microscope. If a single-cell suspension was not obtained after 12 min, cells were washed with wash medium and incubated for an additional 5 min with 0.05% trypsin–EDTA. Cells were counted using a hemocytometer and 400 cells per μl of undiluted Matrigel were plated in a well of a 12-well non-tissue culture plate. The plate was placed in a normoxic incubator for 30 min until the Matrigel polymerized and 1 ml of CK-DCI (3 μM CHIR-99021, 10 ng ml⁻¹ rhKGF, 50 ng ml⁻¹ dexamethasone (Thermo Fisher Scientific), 0.1 mM 8-bromoadenosine 3′,5′-cyclic monophosphate sodium salt (Tocris) and 0.1 mM IBMX (Sigma-Aldrich)) medium was gently added using a 10-ml serological pipette. The medium was changed every 3 days. After 2–3 weeks, an iRAP culture was established that could be maintained for at least seven passages.

Generation of transitional AT2 cells.

Seven to ten days after passaging, small iRAP spheres formed. CKDCI medium was then replaced with DCISB medium (50 ng ml⁻¹ dexamethasone (Thermo Fisher Scientific), 0.1 mM 8-bromoadenosine 3′,5′-cyclic monophosphate sodium salt (Tocris), 0.1 mM IBMX (Sigma-Aldrich) and 10 μM SB-431542 (Tocris)) for 10–14 days.

Generation of iAT1 cells.

24-well tissue culture treated plates were coated with 1:30 Matrigel and left at 4 °C overnight. For the stepwise differentiation pathway, mature transitional AT2-like cells were used. For the direct differentiation pathway, mature iRAPs were used. In both cases, cells were released from Matrigel using dispase, washed and dissociated with 0.05% trypsin–EDTA, as described above. Matrigel was aspirated from the plate when ready to seed with cells. Then, 2 × 10⁵ cells were resuspended in 0.5 ml of SB–BMP4 (10 μM SB-431542 (Tocris), 10 ng ml⁻¹ BMP4 (Tocris) and 10 μM Y-27632 (Tocris)) and added into one well. The cells were cultured for 14-21 days in a normoxic incubator at 37 °C.

iRAP treatments (TGFβ, TGFβi, LATSi, TEADi).

For TGFβ1 treatment, recombinant human TGFβ1 (10 ng/ml; R&D Systems, Cat. No. 7754-BH) dissolved in 4 mM HCl was added to CKDCI medium. Cells were maintained in this condition for 5 days. Cells were then collected for analysis, sorted and analyzed, or sorted and replated in CK-DCI ± 10 μM SB-431542 for reversibility experiments. For LATSi treatment, TRULI (10 μM; MedChem Express, Cat. No. HY-138489) dissolved in DMSO was added to CKDCI and maintained for 7–10 days. Cells were then either passaged in CKDCI + LATSi to maintain treatment, or cells were passaged into CKDCI for reversibility experiments. For TEADi treatment, K-975 (10 μM; MedChem Express, Cat. No. HY-138565) dissolved in DMSO was added to CKDCI and maintained for 7–10 days before analysis.For TGFβi, SB-431542 (10 μM; Tocris, Cat. No. 16-141-0) dissolved in DMSO was added to CK-DCI and maintained for 10 days, or added immediately upon replating of TGFβ1-treated sorted cells for reversibility experiments. Vehicle was DMSO at equal volumes as added to treatment conditions unless stated otherwise. For all conditions, treatment was initiated after sphere formation (7–10 days after passaging) and medium was refreshed every 2-3 days.

iAT2 cell culture.

iAT2 cells were purchased from the Center for Regenerative Medicine (CREM) iPS core (Boston University, Boston, MA). Cells were received as mature organoids in suspension Matrigel in media. Upon receipt, Matrigel was pelleted by centrifugation (400g), digested with dispase, washed, and trypsinized as described above to make single cell suspensions. Cells were then seeded in undiluted Matrigel at 400 cells/μl on non tissue-culture treated 12 well plates and placed in a 37 °C incubator for 30 minutes to solidify. CKDCI supplemented with 10 μM Y-27632 (Tocris) was added for the first 2 days after replating, followed by CKDCI alone, with media changed every 2-3 days. Stepwise differentiation into iAT1 cells was done as described above. Differentiation using Burgess et. al. (2024) protocol was done with 7 days of CKDCI followed by 12 days of DCI-LATSi. TGFβ treatment was done the same as done for iRAPs as described above.

TGFβ1 ELISA.

The Invitrogen Human TGFβ1 ELISA kit (Catalog no. BMS249-4) was used. KRT17-eGFP− and KRT17-eGFP⁺ cells were sorted and replated at 400 cells/μl of Matrigel in a 25 μl droplet in a 24 well non-tissue treated plate. After 7 days, media was conditioned by leaving 400 μl CKDCI on spheres for 3 days. Conditioned medium was diluted 1:10 in Assay Buffer, acid-activated by addition of 20 μl 1N HCl with incubation for 1 hour at room temperature, and neutralized with 20 μl 1N NaOH. The ELISA was then performed according to manufacturer’s instructions using the provided microwell strips, biotin-conjugate, streptavidin-HRP, TMB substrate, and stop solution. Absorbance was measured at 450 nm as the primary wave length and TGFβ1 concentration was determined based on a 7-point standard curve.

RT–qPCR.

Total RNA was extracted according to the manufacturer’s instructions using the Direct-zol RNA microprep (Zymo Research) and 500 ng of total RNA was reverse-transcribed using the qScript XLT complementary DNA (cDNA) SuperMix (Quantabio). Technical triplicates of a 15-μl reaction (for use in Applied Biosystems QuantStudio7 384-well System) were prepared with 3 μl of diluted cDNA and run for 40 cycles. Primers used are listed in Supplementary Data Table 3. Relative gene expression was calculated on the basis of the average cycle (C_t) value, normalized to GAPDH as the internal control and reported as the fold change (2^−ΔΔCt).

Flow cytometry and cell sorting

iRAPs embedded in Matrigel were released by incubating with dispase for 30–60 min, then washed and dissociated into single cells with 0.05% Trypsin/EDTA in a normoxic incubator for 10–12 min with occasional pipetting with a P1000 tip. The single-cell suspension was stained in polystyrene round-bottom 12 × 75-mm tubes (BD Falcon). Primary HT1-56 (Terrace Biotech; 1:200 dilution) and HT2-280 (Terrace Biotech; 1:200 dilution) antibodies were incubated for 1 h at room temperature. Cells were washed two times with fluorescence-activated cell sorting (FACS) buffer (PBS and 1% BSA) and centrifuged for 5 min at 420g. Fluorochrome-labeled secondary antibody (Alexa Fluor 488 (A21042) or Alexa Fluor 594 (A21044) goat anti-mouse IgM, Thermo Fisher Scientific) were diluted in FACS buffer at 1:200 and added for 45 min in the dark. Cells were washed twice and resuspended in FACS buffer for flow cytometric analysis on a Novocyte Quanteon (Agilent). Cell sorting was performed on a Sony MA900 (Sony).

EdU incorporation.

Cells were incubated with 15 μM EdU (5-ethynyl-2′-deoxyuridine) for 20 h before fixation and permeabilization using reagents provided in the kit. Incorporated EdU was detected using the Alexa Fluor 647 picolyl azide provided and analyzed by flow cytometry.

Histology and immunofluorescence (IF) staining.

For analysis of cultured cells, a cell scraper was used to carefully detach the Matrigel droplet with embedded iRAPs from the bottom of the 12-well plate. The droplet was then transferred to an OCT histology mold and embedded in Tissue-Tek OCT (Sakura Finetek). Frozen samples were cut using a cryotome at the thickness of 10–12 μm, collected on adhesion microscope slides, air-dried, fixed in 4% paraformaldehyde and then washed twice for 5 min in 50 mM glycine to inactivate PFA, followed by washing in PBS. Samples were permeabilized for 10 min in 0.2% PBST (PBS + 0.2% Triton X-100 (Thermo Fisher Scientific), blocked by incubating in PBS containing 5% donkey serum and then incubated overnight in primary antibody (Supplementary Data Table 4) in 2% donkey serum. On the next day, samples were washed three times in PBS and 1% donkey serum and incubated with secondary antibodies (1:200; Supplementary Data Table 4) for 1h at room temperature. Nuclei were stained with DAPI (Thermo Fisher Scientific). Sections were mounted with Mounting Reagent (Agilent) on coverslips. Samples were imaged using a Leica TCS SP8 Stellaris Laser scanning confocal microscope and Leica DMi8 inverted phase-contrast microscope (Leica Microsystems).

For IF signal intensity analysis, Fiji (ImageJ 1.54p) was used. For each analysis, 4 regions of interest (ROIs) from at least 3 biological replicates were analyzed, with at least 100 cells per ROI. Thresholds were applied uniformly across all analyzed images. Integrated density per channel of interest was divided by the number of cells, found using a DAPI-based mask. For YAP+ nuclei analysis, the DAPI mask was applied to the YAP channel and only YAP staining that overlapped with DAPI staining was counted.

Live Imaging

iRAPs were plated in Matrigel domes at 400 cells/μl onto 12 well non-tissue culture treated plates and maintained for 15 days in CKDCI. Cells were then switched to media containing CKDCI+vehicle, CKDCI+LATSi, or CKDCI+TGFβ1 as described above. Cells were returned to 37 °C normoxic incubator conditions for 24 hours. Before live imaging on the Axio Observer Z1 (Zeiss), the station was warmed to 37 °C and set to 5% CO2. Plates were then imaged individually. Images were taken using the 10x objective every 30 minutes for 96 hours. Images were process in ImageJ (Fiji), removing out of focus or overexposed frames, and exported as a .AVI files.

RNA-seq and data analysis

Total RNA was extracted from sorted KRT17-eGFP− and KRT17-eGFP⁺ CKDCI iRAPs using the Direct-zol RNA microprep kit (Zymo) according to the manufacturer’s instructions. RNA quality was evaluated by an Agilent 2100 BioAnalyzer. RNA-seq libraries were prepared and sequenced by the JP Sulzberger Columbia Genome Center High-Throughput Screening Center at Columbia University. Paired-end sequencing was performed on a NovaSeq 6000 platform. RNA-seq read quality assessment was performed using fastQC version 0.12.0. Processed reads were mapped against the human genome (GRCh38.p14) using Gencode version 44. The number of reads per gene was counted using Salmon version 1.10.12 providing the count matrix. Differential gene expression was analyzed using the DESeq2 version 1.30.1 R package. DEGs were identified for our study by setting a cutoff of P value <1 × 10⁻¹⁰ and fold change > ±2. GO analysis was performed using the DEGs with enrichplot package version 1.10.2, using genome-wide annotation for human database version 3.12.0.

Single-cell cDNA library preparation and scRNA-seq

Direct and Stepwise scRNA sequencing: iAT1 cells from both pathways were prepared as described above, and cultured in the same iAT1 condition media for 2 weeks. Cells were then treated with trypsin for 8 minutes, collected and resuspended in 0.1%BSA. Cells were then stained with TotalSeq-B anti-human hashtags 1 and 2 for 3 minutes on a rotator. Cells were then washed in PBS 5 times and equal numbers of cells were combined. The viability of single cells was assessed using Trypan blue staining and debris-free suspensions of >80% viability were deemed suitable for scRNA-seq. Cells were then submitted for scRNA sequencing.

TGFβ and LATSi treated iRAPs: Samples were cultured in CKDCI for 2 weeks before being treated with DMSO, TGFβ, or LATSi at concentrations described above. Samples were then released from Matrigel using dispase and dissociated into single cells using trypsin for 10 minutes. Samples were then multiplexed using the same protocol described above.

scRNA-seq was performed using the Chromium platform (10x Genomics) with the 3′ gene expression V3 kit, using an input of ~10,000 cells. Briefly, gel bead-in emulsions (GEMs) were generated on the sample chip in the Chromium controller. Barcoded cDNA was extracted from the GEMs by post-GEM RT cleanup and amplified for 12 cycles. Amplified cDNA was fragmented and subjected to end repair, poly(A) tailing, adaptor ligation and 10x-specific sample indexing following the manufacturer’s protocol. Libraries were quantified using Bioanalyzer (Agilent) and QuBit (Thermo Fisher Scientific) analysis and were sequenced in paired-end mode on a NovaSeq instrument (Illumina) targeting a depth of 50,000–100,000 reads per cell.

scRNA-seq computational analysis

Sequencing data were aligned and quantified using the Cell Ranger single-cell software suite (version 6.1.2, 10x Genomics) against the provided GRCh38 (Ensembl 98) human reference genome. Publicly available were downloaded from the GEO database or from the HLCA database (Supplementary Data Table 1).

Data were analyzed using Scanpy 1.11.5, SCVI 1.4.0, Pandas 2.3.3, Seaborn 0.13.2, Numpy 2.3.5. Data were loaded into AnnData objects, and filtered based on minimum number of cells where a genes expressed, and minimum genes per cell. Doublet exclusion was performed using SCVI. Cells were next filtered based on percent ribosomal and mitochondrial count and outliers in the number of genes per cells were excluded at the 98^th percentile. Individual samples were the concatenated and integrated using SCVI. which yields SCVI-normalized count data. The HLCA datasets were already integrated. Neighbors, UMAP and Leiden clustering were next calculate using the X-scVI layer. Data were then subsetted on epithelial cells expressing canonical lung markers and not expressing markers for mesenchymal, hematopoietic and vascular cells. After subsetting neighbors, UMAP and Leiden clusters were each time recalculated. The map function implemented in Scanpy was used to assigned cell names to clusters using a mapping dictionary. Cell identity was assigned based on canonical markers. Next, PAGA analysis as implemented in Scanpy was performed with resolution of 0.03 throughout. Data were subsequently reclustered using PAGA initiation. For diffusion pseudotime plotting, a root cluster was each set with the lowest diffusion pseudotime where a root cell was randomly chosen.

Statistical analysis

Statistical analysis was performed using an unpaired two-tailed, Student’s t-test, one-way analysis of variance (ANOVA) or two-way ANOVA with multiple-testing correction where appropriate using Prism 9. Results are shown as the mean ± s.d. unless specified otherwise. P values < 0.05 were considered statistically significant. The n value refers to biologically independent replicates.

Supplementary Material

Supplement 1

Fig. S1. Cell type assignment in single cell transcriptomic datasets. a-c. Representative example of clustering and cell type assignment, PAGA trajectory analysis, and feature plots of selected markers used for cell type assignment for human mouse influenza (a), Covid (b), and IPF (HLCA, Sikkema et al., 2023) (c) datasets. d. Dot plot for selected marker expression in all datasets analyzed in Fig. 1. e. Diffusion pseudotime curves modeling transition of AT2 and AT1 cells to TRBCs in the Covid dataset.

Fig. S2. KRT17 reporter and identification of a KRT17⁺ AT1 lineage. a. Schematic of generation of KRT17-eGFP reporter. b. Representative flow plot of eGFP expression in iRAPs derived from KRT17-eGFP reporter hESC line. c. IF representative of 6 biological replicates of KRT17-eGFP reporter iRAP (top) and control fibroblasts (bottom) showing colocalization of labelled markers. d. Representative flow cytometry plots (left) and quantification of EdU⁺ cells (right). *p < 0.05, unpaired two-sided Student’s t-test, n=3. e. STRINGDB interaction network of the 71 differentially expressed genes between KRT17-eGFP− and KRT17-eGFP⁺ iRAPs as measured by bulk RNAseq (n=3). f. Gene ontology pathway analysis of upregulated genes in KRT17-eGFP⁺ iRAPs. g. Representative flow cytometry plot of expression of HT1-56 and HT2-280 in transitional AT2 cells generated from KRT17-eGFP⁻ and KRT17-eGFP⁺ iRAPs.

Fig. S3. Direct and stepwise AT1 differentiation pathways. a. Upregulated Gene Ontology biological process pathways in iAT1 cells generated in the direct compared to the stepwise pathway. b. Subsetting strategy in AT1 cells in the Covid dataset. c. Dotplots for expression of top differentially regulated genes in the direct pathway in vitro in IPF and Covid datasets showing consistent enrichment in aBCs and aBC-like cells.

Fig. S4. Reversibility of LATSi and TGFβ effects in iRAPs. a. Cumulative cell number in vehicle and LATSi treated iRAPs over 10 passages. ****p<0.0001, unpaired two-sided Student’s t-test, n=3. b. Representative brightfield images of iRAPs in vehicle, LATSi and LATSi->vehicle reversal conditions. c. Quantification of percent KRT17-eGFP⁺ cells in LATSi and reversal conditions. ****p<0.0001, one-way ANOVA with Tukey’s multiple comparison test, n=5 (vehicle), n=4 (LATSi and LATSi->Veh. P0), n=2 (LATSi->veh. P1), and n=3 (LATSi->veh. P2). d. Percent HT1-56⁺ cells in LATSi and reversal conditions. *p<0.05, **p<0.01, ***p<0.001, one-way ANOVA with Tukey’s multiple comparison test, n=5 (vehicle), n=4 (LATSi and LATSi->veh. P0), n=2 (LATSi->veh. P1), and n=3 (LATSi->veh. P2). e. mRNA expression of AT1 and TRBC markers in LATSi and reversal conditions. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, two-way ANOVA with Tukey’s multiple comparison test, n=4 (vehicle), n=4 (LATSi and LATSi->veh. P0), n=2 (LATSi->veh. P1), and n=3 (LATSi->veh. P2). f. Cell number in TGFβ1-treated iRAPs over 7 days, n=3. g. mRNA expression of TRBC markers in TGFβ1 and vehicle treated iRAPs. *p<0.05, two-way ANOVA with Šidák multiple comparison test, n=3. h. mRNA expression of TGFB1 and TGFBR1 in sorted KRT17-eGFP⁻ and KRT17-eGFP⁺ cells. *p<0.05, **p<0.01, two-way ANOVA with Šidák multiple comparison test, n=3. i. TGFβ1 protein concentration in conditioned media as measured by ELISA from sorted and replated cells, either total , KRT17-eGFP⁻ and KRT17-eGFP⁺ cells. *p<0.05, **p<0.01, one-way ANOVA with Tukey’s multiple comparison test, n=4 (total and KRT17⁺) and n=5 (KRT17−). j. Percent KRT17-eGFP⁺ cells in TGFβ1-treated cells after sorting, and those same cells replated in either vehicle or TGFβi conditions. ****p<0.0001, one-way ANOVA with Dunnett’s multiple comparison test, n=6. k. mRNA expression of aBC markers in TGFβ1-treated, and TGFβ1-treated, sorted and replated conditions. **p<0.01, ***p<0.001, ****p<0.0001, two-way ANOVA with Tukey’s multiple comparison test, n=3. l. mRNA expression of lung markers in TGFβ1-treated, and TGFβ1-treated, sorted and replated conditions. **p<0.01, ****p<0.0001, two-way ANOVA with Tukey’s multiple comparison test, n=3.

Fig. S5. Comparison with iAT2 cells. a. mRNA expression of TRBC markers and SFTPC in iRAPs and SPC-iAT2 cells. **p < 0.01, two-way ANOVA with Šidák multiple comparison test, n=5 (iRAPs) and n=4 (SPC-iAT2). b. Representative IF of 3 biological replicates of iRAPs (top) and SPC-iAT2 cells (bottom) stained for labelled markers. Scale bars = 50 μm. c. Schematic of stepwise differentiation pathway highlighting iAT1 SB-BMP stage (left) with mRNA expression of iAT1 markers and SFTPC in iRAPs and SPC-iAT2 cells differentiated in SB-BMP4 or DCI-LATSi (right). **p < 0.01, ****p < 0.0001, two-way ANOVA with Tukey’s multiple comparison test, n=3. d. Representative brightfield and fluorescence images of 3 biological replicates of iRAP- and SPC-iAT2-derived iAT1 cells. e. mRNA expression of aBC markers in iRAPs and SPC-iAT2 cells treated with vehicle or with TGFβ1. *p<0.05, **p<0.01, ****p<0.0001, two-way ANOVA with Tukey’s multiple comparison test, n=3. f. mRNA expression of TRBC and AT2 markers in iRAPs and SPC-iAT2 cells treated with vehicle or with TGFβ1. *p<0.05, **p<0.01, two-way ANOVA with Tukey’s multiple comparison test, n=3. g-h. Representative flow cytometry plot and quantification of percent HT2-280⁺ cells in iRAPs and SPC-iAT2 cells treated with vehicle or with TGFβ1. ***p<0.001, one-way ANOVA with Tukey’s multiple comparison test, n=3. i. Quantification of percent SPC-tdTomato⁺ cells in SPC-iAT2 cells treated with vehicle or with TGFβ1. Unpaired two-sided Student’s t-test, n=3.

Fig. S6. a. Gene Ontology enrichment analysis of top-200 DEGs in TGFβ1 and LATSi treated cells, respectively. b. Expression of TGFβ and LATSi signature genes in vehicle, TGFβ1 and LATSi-treated iRAPs. c. Expression of TGFβ and LATSi signature genes in WT and HPS1^−/− iRAPs. d. Expression of genes in TGFβ and LATSi scores across all annotated epithelial cell types in the normal human lung dataset in LungMap with hierarchical clustering.

Data Availability

scRNAseq data will be available in the GEO database (GEO Submission (GSE338557) after acceptance. The publicly available datasets analyzed are listed in Supplementary Table 1.