Emergence of fibrotic pericytes and their transcriptional regulation in pulmonary fibrosis

Abstract

Multiple cell types have been implicated in pathogenesis of pulmonary fibrosis with pericytes emerging as a new focus due to their role in promoting fibrotic remodeling. Fibrotic environment changes normal pericyte functions, but transcriptional programs regulating pericyte transition towards fibrotic state remain unclear. Utilizing single cell RNA-sequencing of human pulmonary fibrotic lungs and mouse genetic models, we identified a unique cluster of fibrosis-associated pericytes which was not present in normal lungs and exhibited a distinctive transcriptional signature indicating transition of normal pericytes to a fibrotic state. FOXF1 was identified as one of the transcription factors decreased in fibrotic pericytes. Pericyte-specific deletion of Foxf1 increased severity of pulmonary fibrosis in bleomycin mouse model as demonstrated by reduced survival, impaired lung functions, increased body weight loss and increased fibrotic remodeling. Pericyte-specific overexpression of Foxf1 attenuated pulmonary fibrosis, reversed fibrotic changes and improved survival outcomes. Based on single cell RNA-sequencing and chromatin immunoprecipitation sequencing, FOXF1 transcriptionally regulates an extensive pericyte signaling network critical for pulmonary fibrosis. In vitro, FOXF1 transcriptionally activated ID3 which inhibited fibroblast activation through decreased secretion of IL8 and CXCL1 by pericytes. Altogether, FOXF1 prevents transition of pericytes to a fibrotic state, suggesting new therapeutic opportunities for treatment of pulmonary fibrosis.

Using single-cell transcriptomics, the authors identified a novel fibrosis-associated pericyte population with reduced FOXF1 expression. FOXF1 loss worsened lung fibrosis, highlighting FOXF1 as a potential therapeutic target in lung fibrosis.

Introduction

Pulmonary fibrosis (PF) is characterized by the aberrant deposition of extracellular matrix (ECM), persistent activation of myofibroblasts, and the disruption of normal alveolar architecture and function. Idiopathic pulmonary fibrosis (IPF) is the progressive form of PF and represents the most common type of interstitial lung disease (ILD). The pathogenesis of IPF is multifactorial, involving a complex interplay of environmental exposures, aging, genetic predisposition, epigenetic modifications, immunologic dysregulation, and gerontologic mechanisms¹. Multiple cell types are involved in IPF pathogeneses, including epithelial cells, endothelial cells, fibroblasts and immune cells. Chronic injury to alveolar epithelial cells plays a central role in the pathogenesis of IPF, followed by an abnormal repair that includes dis-regulated activation of fibroblasts, endothelial cells and recruitment of immune cells^2–4. In addition to the well-documented roles of alveolar epithelial cells, endothelial cells, fibroblasts and immune cells in pulmonary fibrosis, emerging evidence highlights the importance of lung pericytes in regulating pulmonary fibrogenesis^5,6.

Pericytes are the perivascular cells that share a basement membrane with endothelial cells and contribute to vascular integrity. During pulmonary fibrosis, pericytes detach from the endothelium, leading to vascular leakage and contributing to the progression of fibrosis⁷. Pericytes have been shown to undergo phenotypic changed in the fibrotic environment, acquiring myofibroblast-like properties with increased collagen production^8,9. While pericyte to myofibroblast-like changed have been observed across various tissues^10,11, the transcriptional regulators governing this process during lung fibrotic remodeling remain unclear. Given the multifaceted role of pericytes in mediating tissue repair and shaping fibrotic microenvironment, elucidating the mechanisms by which pericytes contribute to fibrosis is essential and holds great promise for uncovering novel therapeutic targets and improving outcomes for patients with pulmonary fibrosis.

Our study aims to elucidate the transcriptional mechanisms by which pericytes regulate pulmonary fibrosis. Through transcriptional profiling of lung pericytes from donors and patients with pulmonary fibrosis, we have identified a subcluster of fibrosis-associated pericytes that is not present in normal lungs. We have also identified FOXF1 as a transcriptional regulator involved in the transition of normal pericytes to fibrosis-associated pericytes that support lung fibrogenesis. FOXF1 (Forkhead box F1) is a transcription factor expressed in lung mesenchyme during lung development^12–15. Mutations in FOXF1 gene locus cause alveolar capillary dysplasia with misalignment of pulmonary veins (ACDMPV), a lethal congenital disorder of neonates and infants with significant fibrotic component^16–19. FOXF1 is known to play a role in embryonic and neonatal angiogenesis by regulating pro-angiogenic genes in endothelial cells^20–22, and has been shown to inhibit lung cancer progression while promoting tumor vessel normalization²³. During lung fibrosis, FOXF1 regulates the reprogramming of normal endothelial cells into fibrosis-associated endothelial cells³ and prevents expansion of myofibroblasts²⁴. While FOXF1 is expressed in lung pericytes, its role in pericytes remains unknown.

In this study, we utilized single-cell RNA sequencing and mouse genetic models to demonstrate that the loss of FOXF1 in pericytes leads to upregulation of pro-fibrotic signaling pathways, including the TGF-β pathway, extracellular matrix organization and actin cytoskeleton rearrangement. FOXF1 loss in pericytes significantly exacerbates the severity of lung fibrosis. Conversely, FOXF1 overexpression in pericytes attenuates pulmonary fibrosis by reducing fibrotic remodeling. These findings underscore the critical role of FOXF1 in modulating lung fibrogenesis and highlight its potential as a therapeutic target for pulmonary fibrosis.

Results

Transcriptome analysis identified a unique sub-cluster of fibrotic pericytes in the lungs of patients with Pulmonary Fibrosis

To characterize the role of lung pericytes in PF, we first compared the transcriptome of lung pericytes from donors and patients with pulmonary fibrosis (PF) by analyzing publicly available scRNA-seq datasets²⁵. In these datasets, the samples from PF patients were stratified into two groups, including with “less fibrosis” and “more fibrosis” based on the differentially affected areas of one lung used to prepare sequencing libraries²⁵. Mesenchymal cells from the dataset were re-clustered and annotated using cell-type-specific marker genes provided by LungMap²⁶ (Fig.1A and Supplementary Fig. 1A). Human pericytes were identified based on the expression of CSPG4, HIGD1B, COX4I2, KCNK3, and LAMC3 and visualized by immunostaining in donor and fibrotic lungs (Fig. 1F and Supplementary Fig. 1A–D). A unique pericyte sub-cluster (Peri2) was detected that was present exclusively in both groups of PF lungs but was absent in donor lungs (Fig. 1B and Supplementary Fig. 1E–G). Peri2 pericyte sub-cluster is characterized by a reduced expression of pericyte-specific markers and an increased expression of fibrosis-associated markers, including COL1A1, COL3A1, and ACTA2, as well as the pericyte activation marker RGS5 (Fig.1C). In contrast, Peri1 pericyte cluster was present in both donors and PF patients (Fig. 1B and Supplementary Fig. 1E–G). To further elucidate the differences between Peri1 and Peri2 pericyte clusters, we performed KEGG pathway analysis. Unlike Peri1 subcluster, which was enriched in pathways associated with cell adhesion and metabolic activity, Peri2 pericytes showed enrichment in fibrosis-associated pathways, including TGF-beta signaling, extracellular matrix (ECM) deposition, and regulation of the actin cytoskeleton (Fig. 1D). Pseudotime analysis predicted a transition from Peri1 to Peri2, with colors representing inferred pseudotime values from early-stage (dark blue) to late-stage (yellow) (Fig. 1E), suggesting that pericytes undergo a trans-differentiation process under fibrotic conditions, with cell fate shifting toward a fibrotic state. Co-localization studies demonstrated the existence of NG2⁺aSMA⁺ cells in the fibrotic lungs, supporting scRNAseq data (Fig. 1F). The lack of MYH11 staining indicated that these cells were not smooth muscle cells (Supplementary Fig. S1D). Ligand-receptor interaction analysis using the CellChat computational method²⁷ shows the crosstalk between pericytes and other lung cell types, especially alveolar fibroblasts and endothelial cells (Fig. 2A). Based on this analysis, Peri1 govern a “homeostatic support function” by being the dominant source of Angiopoietin 1 (ANGPT1), a critical homeostatic growth factor that promotes lung vascular stability, reduces capillary leakage, and is anti-inflammatory during lung injury (Fig. 2B–D). In contrast, Peri2 govern a “pathogenic activation function”, transitioning toward a reactive, pro-fibrotic state evidenced by increased Fibronectin 1 (FN1), Thrombospondin (THBS), NOTCH and Collagen signaling network, which collectively promote myofibroblast differentiation, excessive collagen deposition and vessels remodeling (Fig. 2B-C). Thus, a unique cluster of fibrotic Per2 pericytes was identified in human PF lungs that was not present in donor lungs, and that had a distinctive transcriptional signature indicating transition of normal pericytes to a distinct fibrotic state.

Fig. 1. Unique subcluster of fibrosis-associated pericytes is identified in lungs of patients with pulmonary fibrosis.

A UMAP visualization of lung mesenchymal cells from 49 donors and 67 patients with pulmonary fibrosis (PF) using the scRNA-seq dataset GSE227136. B UMAP projections of pericyte sub-clusters in the Donor group, PF patients with less fibrosis and PF patients with more fibrosis. A Unique subcluster of fibrotic pericytes (Peri2) was found only in the lungs of PF patients. C Peri2 pericytes display increased expression of fibrosis-associated markers, but decreased expression of pericyte-specific markers. D KEGG pathways enrichment analysis of Peri1 and Peri2 pericyte sub-clusters. Peri1 is enriched in pathways related to cell adhesion and metabolic activity. Peri2 is enriched in pathways related to TGF-beta signaling and collagen deposition. KEGG pathway enrichment was performed using DAVID based on a modified Fisher’s exact test (EASE score). E Pseudotime analysis predicted a transition from Peri1 to Peri2. UMAP depicts the differentiation trajectory of single cells. Each dot represents an individual cell, with colors (blue, pink) distinguishing Peri1 and Peri2 pericytes. Pericytes were colored according to its inferred pseudotime value from early-stage (dark blue) to late-stage (yellow). F H&E and immunostaining for NG2 (magenta) and α-SMA (green) showed that pericytes acquire fibrotic characteristics in PF lung. DAPI was used for nuclear staining. Bar = 100 μm. Staining was performed on tissue sections from 5 independent human samples.

Fig. 2. Peri1 and Peri2 exhibit functionally distinct homeostatic and pathogenic signaling profiles in the lung niche.

A Cord diagram was performed on combined single-cell RNA-sequencing datasets from human donor and IPF lung samples, integrating endothelial and mesenchymal cell populations. Edge thickness represents interaction frequency or strength among different cell types, respectively. B Differential information flow analysis comparing signaling activity between Peri1 and Peri2 in their interactions with other cell types. Positive values (pink) denote pathways enriched in Peri1, while negative values (green) denote pathways enriched in Peri2. C Heatmaps representing the hierarchical communication probability for ANGPT1, ANGPT2, NOTCH, FN1 and THBS signaling pathways. D Schematic diagram of the top intercellular pathways of pericyte subtypes with endothelial cells and alveolar fibroblasts. Created in BioRender. Gao, W. (2026) https://BioRender.com/qxqiyoh.

The existence of the unique fibrotic subcluster of pericytes was confirmed in the bleomycin-treated mouse lungs

To determine whether the unique subcluster of fibrotic pericytes is also present in the pre-clinical murine model of PF, we analyzed publicly available scRNA-seq datasets of the bleomycin-induced PF mouse model (GSE132771)²⁸. Mouse mural cells were sub-clustered and pericytes were identified based on the expression of Cspg4, Higd1b, and Pdgfrb (Fig. 3A and Supplementary Fig. 2A). Consistent with human data, the mouse Peri2 subcluster of pericytes was found in bleomycin-treated fibrotic lungs and was absent in the untreated normal murine lungs (Fig. 3B). Similar to human data, the mouse fibrotic Peri2 pericytes were characterized by increased expression of Rgs5 (Fig. 3C), decreased expression of pericyte markers (Higd1b, Cspg4 and Cox4i2) and increased expression of fibrosis-associated markers Col1a1, Col3a1, and Acta2 (Fig. 3D). KEGG pathway analysis showed that Peri2 pericytes were enriched in fibrosis-associated signaling pathways, including TGF-beta signaling, extracellular matrix and actin cytoskeleton regulation, whereas Peri1 was enriched in pathways related to adhesion and metabolism (Fig. 3E). To make sure that the mouse data is not limited to one dataset, we have also analyzed the different publicly available scRNA-seq datasets of bleomycin-induced lung fibrosis (GSE210341)². Using the second data set, we verified the existence of the unique fibrotic cluster of Peri2 pericytes in PF lungs but not in the normal mouse lungs (Supplementary Fig. 2B). We have also verified that Peri2 pericytes had increased expression of fibrosis-associated markers and were enriched in fibrosis-associated signaling pathways, including TGF-beta signaling (Supplementary Fig. 2C, D). Pseudotime analysis predicted transition from Peri1 to Peri2 (Supplementary Fig. 2E), a finding consistent with human PF data (Fig. 1E). Also, consistent with human data, the CellChat analysis identified Peri1 as being the dominant source of Angiopoietin 1 (ANGPT1), while Peri2 decreasing ANGPT1 and increasing ANGPT2 to destabilize endothelial cells. In addition, Peri2 are identified as a source of pro-fibrotic TGFß, FGF and IGF signaling (Supplementary Fig. 3A–C), promoting myofibroblast differentiation, excessive collagen deposition and vessels remodeling (Supplementary Fig. 3D). Altogether, the pre-clinical murine model of PF confirmed the presence of the unique fibrotic cluster of Peri2 pericytes in PF lungs but not in the normal mouse lungs. Similar to human data, mouse Peri2 pericytes demonstrated a distinctive gene signature indicating the transitional state of pericytes toward fibrotic phenotype.

Fig. 3. Unique subcluster of fibrosis-associated pericytes is identified in lungs of bleomycin-treated mice.

A UMAP visualization of mural cells from GSE132771. B UMAP projections of pericyte sub-clusters per treatment group revealed a distinct pericyte subcluster (Peri2) in bleomycin-treated but not in the control lungs. C, D Peri2 exhibits fibrosis-associated features, characterized by decreased expression of canonical pericyte markers (Cspg4, Cox4i2, Higd1b) and increased expression of fibrotic markers (Acta2, Col1a1, Col3a1), along with a pericyte activation marker, Rgs5. E KEGG pathway enrichment analysis of Peri1 and Peri2 based on scRNAseq. Peri1 is enriched in pathways related to cell adhesion and metabolism, whereas Peri2 is enriched in pathways associated with TGF-beta signaling, actin cytoskeleton regulation and extracellular matrix (ECM) deposition. KEGG pathway enrichment was performed using DAVID based on a modified Fisher’s exact test (EASE score). F Schematic timeline of tamoxifen and bleomycin administration to induce pulmonary fibrosis and tdTomato expression. G Quantification of tdTomato⁺ cells as a percentage of live cells by FACS analysis. Ctrl, n = 3 mouse lungs; Bleo, n = 4 mouse lungs. **p = 0.0039. H, I Immunostaining of tdTomato⁺ pericytes in NG2-CreER; Rosa26-LSL-tdTomato mice shows reduced pericyte marker (NG2) and increased fibrosis marker (α-SMA) following bleomycin treatment. NG2 and α-SMA expression in tdTomato⁺ cells were quantified across random fields. α-SMA: ctrl (n = 5), bleo (n = 5); NG2: ctrl (n = 6), bleo (n = 5). Scale bar = 25 µm. *** p = 0.0002. J Relative mRNA expression of Acta2, Col1a1, and Col3a1 in tdTomato⁺ pericytes FACS-sorted from control and bleomycin-treated mice. Col3a1: ctrl (n = 5), bleo (n = 5), **p = 0.0079; Col1a1: ctrl (n = 4), bleo (n = 4), *p = 0.0349; Acta2: ctrl (n = 5), bleo (n = 4), ***p = 0.0004. Data are presented as mean ± SEM. For two group comparisons, T-test (two-tailed) analyses and Mann–Whitney test were performed. *, p < 0.05, **, p < 0.01, ***, p < 0.001, ****, p < 0.0001.

Lineage-tracing reveals the transition of normal pericytes to a fibrotic phenotype in bleomycin-treated mouse lungs

Since scRNAseq and pseudotime analysis predicted that lung pericytes undergo a trans-differentiation process under fibrotic conditions with cell fate shifting toward a fibrotic state, we next used a lineage tracing approach to determine the fate of lung pericytes during their transition from normal to fibrotic phenotype. The NG2-CreER; Rosa26-LSL-tdTomato mice (abbreviated as NG2; LSL-tdT) were treated with Tamoxifen to label pericytes (Supplementary Fig. 4A, B). Efficient recombination was detected in approximately 81.6% of NG2⁺ cells following Tamoxifen treatment (Supplementary Fig. 4C). FACS analysis and immunostaining demonstrated the presence of tdTomato-positive cells in the mouse lung (Supplementary Fig. 4D, E). To assess the fate of NG2⁺ cells during fibrotic remodeling, Tamoxifen-treated NG2; LSL-tdT mice received two intratracheal administrations of bleomycin, and lungs were harvested on Day 21 (Fig. 3F and Supplementary Fig. 2F). FACS analysis revealed an increase in the percentage of tdTomato⁺ cells after bleomycin treatment compared to untreated control lungs (Fig. 3G). Co-immunostaining for NG2 and αSMA showed that the percentage of αSMA⁺/tdTomato⁺ cells increased, while the percentage of NG2⁺ / tdTomato⁺ cells decreased after bleomycin treatment (Fig. 3H, I). Using FACS-sorted tdTomato⁺ cells, we next demonstrated that mRNA levels of fibrosis-associated Col1a1, Col3a1, Acta2 genes were increased in tdTomato⁺ cells from bleomycin-treated lungs compared to untreated controls (Fig. 3J). These findings are consistent with the increased fibrotic markers found in fibrotic pericytes from human and mouse scRNA-seq analysis (Figs. 1C, 3D). Thus, using the lineage-tracing approach, we have shown directly that during fibrotic lung remodeling, normal pericytes undergo transition to a fibrotic state, characterized by the increase in fibrosis-associated markers Col1a1, Col3a1, and Acta2.

Cross-comparison of human and mouse datasets identified evolutionary conserved transcription factors associated with Peri1 to Peri2 transition

To identify differentially expressed transcription factors associated with the transition of normal pericytes (Peri1) to fibrotic (Peri2) pericytes, we used human PF scRNAseq datasets to compare the transcriptome of Peri2 and Peri1 pericytes. Upregulated transcription factors in Peri2 pericytes included PRRX1, ATF3, MYC, ARID5B, FOSB, and BACH1, whereas downregulated transcription factors in Peri2 pericytes included FOXF1, ID1, TBX3, DUSP4, and HEYL (Fig. 4A). Next, we used mouse bleomycin PF scRNA-seq datasets to compare the transcriptome of Peri2 and Peri1 pericytes. Upregulated transcription factors in murine fibrotic Peri2 pericytes were Prrx2, Scx, Bnc2, Sox9 and Egr2, while downregulated transcription factors were Foxf1, Id1, Tbx3, and Heyl (Fig. 4A). Cross-comparison of human and mouse differentially expressed transcription factors in normal and fibrotic lungs identified FOXF1, ID1, TBX3, and HEYL as being consistently downregulated in fibrotic Peri2 pericytes of both human and mouse fibrotic lungs. Since the loss of function mutations in FOXF1 are associated with human ACDMPV (Alveolar Capillary Dysplasia with Misalignment of Pulmonary Veins), a severe congenital lung disorder with a known fibrotic component²¹, we focused our next studies on the role of FOXF1 during the transition of Peri1 to Peri2 pericytes.

Fig. 4. FOXF1 expression is reduced in fibrotic pericytes within PF human and bleomycin-treated mouse lungs.

A Volcano plot shows differentially expressed genes in Peri2 versus Peri1 in human and mouse lungs using scRNA-seq datasets. The x-axis indicates log₂ fold change, and the y-axis represents –log₁₀ P-value. Differential expression analysis between groups was performed with a Wilcoxon rank-sum test. Genes with statistically significant changes and substantial fold differences are highlighted in red. Selected upregulated and downregulated genes are annotated. B In human scRNA-seq datasets, FOXF1 mRNA decreased in Peri2 pericytes compared to Peri1. Comparison significance was tested using a two-sided Wilcoxon rank-sum test. **** p = 2.2e-06. C Dot plot showing the expression of FOXF1 in different pulmonary mesenchymal cell types. D, E Immunofluorescence staining for α-SMA (green), NG2 (magenta), and FOXF1 (cyan) in lung sections from PF patients showed decreased FOXF1 levels in pericytes located within fibrotic foci. DAPI was used to counterstain nuclei. Scale bar = 20 µm. Quantification of FOXF1⁺/NG2⁺ cells was performed across random fields (Donor, n = 8; IPF, n = 5). F In mouse scRNA-seq datasets, Foxf1 mRNA was decreased in Peri2 compared to Peri1 lung pericytes. Statistical analysis was performed using a two-sided Wilcoxon rank-sum test.****p = 2.4e-10. G Split UMAP projections of pericyte sub-clusters within mural cells from control and bleomycin-treated mice. The lower cluster corresponds to smooth muscle cells. The upper cluster corresponds to pericytes. Foxf1 expression is high in Peri1 but is decreased in Peri2. H Foxf1 mRNA was decreased in FACS-sorted NG2⁺ pericytes from bleomycin-treated lungs. Ctrl, n = 3; Bleo, n = 5. ***p = 0.0001. I Immunostaining for α-SMA (green), NG2 (magenta), and Foxf1 (cyan) in lung sections from bleomycin-treated mice demonstrated reduced Foxf1 expression in NG2⁺ pericytes within fibrotic regions. DAPI was used for nuclear staining. Quantification of Foxf1⁺ pericytes was conducted in five random fields. Scale bar = 20 µm. J Reduced Foxf1 expression was observed in tdTomato⁺ pericytes within fibrotic areas of lungs from bleomycin-treated NG2-CreER; Rosa26-LSL-tdTomato mice. Quantification of Foxf1⁺ tdTomato⁺ cells was conducted across random fields (ctrl, n = 5; bleo, n = 6). K Foxf1 mRNA was decreased in tdTomato⁺ pericytes FACS-sorted from bleomycin-treated mouse lungs. Ctrl, n = 6; Bleo, n = 8. ***p = 0.0003. Data are presented as mean ± SEM. ***, p < 0.001, ****, p < 0.0001. For two group comparisons, T test (two-tailed) analyses were performed.

FOXF1 is decreased in pericytes of human PF lungs

To analyze further the differential expression of FOXF1 in normal versus fibrotic pericytes, the FeaturePlot, DotPlot and ViolinPlot analyses were used to demonstrate that FOXF1 was highly expressed in Peri1 pericytes but was dramatically decreased in fibrotic Peri2 pericytes (Fig. 4B, C and Supplementary Fig. 5A). Moreover, the severity of PF further decreased FOXF1 levels in Peri2 pericytes, with a significant reduction observed in patients with more advanced fibrosis compared to those with less fibrosis (Supplementary Fig. 5B). Immunostaining for FOXF1, NG2, and αSMA showed high levels of FOXF1 protein in pericytes of donor lungs, but nearly undetectable levels of FOXF1 protein in pericytes within fibrotic αSMA-positive regions in PF lungs (Fig. 4D). FOXF1 protein was detected in approximately 60% of pericytes from donor lungs, but only 10% pericytes from patients with PF were positive for FOXF1 (Fig. 4E). Thus, both mRNA and protein of FOXF1 are significantly decreased in lung pericytes of patients with pulmonary fibrosis.

Foxf1 is decreased in pericytes of bleomycin-treated mouse lungs

Next, we determined whether a bleomycin mouse model of PF recapitulated the results obtained from human patients. Based on scRNAseq analysis of mouse lungs from control and bleomycin-treated mice, Foxf1 mRNA was decreased in fibrotic Peri2 pericytes compared to normal Peri1 pericytes (Fig. 4F, G), which is consistent with the data from human donor and PF lungs (Fig. 4B). The decrease in Foxf1 mRNA in bleomycin-treated pericytes was confirmed by qRT-PCR using FACS-sorted CD31^-/CD45^-/NG2⁺ lung pericytes from control and bleomycin-treated mice (Fig.4H). Next, immunostaining for FOXF1, NG2, and αSMA, was used to show that FOXF1 protein was highly expressed in the lung pericytes of control lungs but was nearly undetectable in the pericytes within fibrotic lesions of bleomycin-treated mice (Fig. 4I). FOXF1 protein was detected in approximately 60% of pericytes from control lungs, but only 20% of pericytes from bleomycin-treated lungs were positive for FOXF1 (Fig. 4I, left panel). Next, FACS analysis was used to quantify the number of FOXF1-positive pericytes, demonstrating that the number of FOXF1⁺/NG2⁺ pericytes is reduced in the lungs of bleomycin-treated mice (Supplementary Fig. 6A–C). In addition, decreased expression of FOXF1 was verified at both mRNA and protein levels in tdTomato-labeled pericytes using immunofluorescence staining and qRT-PCR of FACS-sorted tdTomato-lineage traced pericytes (Fig. 4J, K). Thus, Foxf1 mRNA and protein levels were decreased in pericytes of bleomycin-treated fibrotic lungs, which is consistent with our findings in human lung pericytes from the lungs of PF patients.

Pericyte-specific deletion of Foxf1 increases the severity of pulmonary fibrosis in pre-clinical animal models

To determine whether FOXF1 in pericytes is important during fibrotic remodeling, we generated mice with pericyte-specific deletion of the Foxf1 by breeding NG2-CreER transgenic mice with Foxf1-floxed mice (Supplementary Fig. 7A, B). Tamoxifen treatment decreased Foxf1 expression as shown by qRT-PCR using FACS-sorted CD31^-/CD45^-/NG2⁺ pericytes (Supplementary Fig. 7C). Tamoxifen-treated NG2-CreER; Foxf1 ^f/f mice (abbreviated as periFoxf1^KO) had normal lung histology and function, as assessed by H&E staining, immunostaining for NG2 and CD31, and FlexiVent lung function tests (Supplementary Fig. 7D, E). To induce fibrosis, Tamoxifen-treated periFoxf1^KO mice had received two doses of bleomycin and lungs were harvested at days 14 and 21 post-bleomycin injury (Supplementary Fig. 8A). Compared to controls, bleomycin-treated periFoxf1^KO mice experienced reduced survival rates, profound body weight loss, and decreased lung functions (Fig. 5A–C and Supplementary Fig. 8B). Hydroxyproline was increased in bleomycin-treated periFoxf1^KO mice compared to controls, indicating increased collagen depositions (Fig. 5D). Histological examination revealed progressive fibrotic remodeling in bleomycin-treated periFoxf1^KO mice, with severity intensifying over time, as indicated by H&E, Sirius Red/Fast Green, Masson’s trichrome staining, and Ashcroft score (Fig. 5E, F). Immunostaining for αSMA showed time-dependent collagen deposition in bleomycin-treated periFoxf1^KO mice (Fig. 5G). FACS-sorted pericytes from bleomycin-treated periFoxf1^KO lungs exhibited increased pro-fibrotic Acta2, Col1a1, and Col3a1 mRNAs compared to control lungs (Fig. 5H–J). Thus, these findings demonstrate that the loss of Foxf1 in pericytes exacerbates bleomycin-induced pulmonary fibrosis.

Fig. 5. Deletion of Foxf1 in pericytes increased severity of pulmonary fibrosis.

A, B Bleomycin-treated periFoxf1^KO mice exhibited reduced survival and increased body weight loss compared to control groups (PBS-treated periFoxf1^KO and bleomycin-treated Foxf1 ^f/f mice). Body weight was monitored over time (untreated control, n = 6 mice; bleo control, n=8 mice; bleo periFoxf1^KO, n = 6 mice, **p=0.003), and survival was assessed up to 21 days post-bleomycin (n = 12 mice per group). C Lung function was impaired in bleomycin-treated periFoxf1^KO mice compared to controls, as measured by flexiVent. IC, inspiratory capacity, n = 6 mice, ** p = 0.009. Rrs, resistance, n = 6 mice, *p = 0.0364. D Lung collagen depositions were increased in periFoxf1^KO mice after bleomycin treatment, as measured by hydroxyproline assay. Untreated control, n = 3; Bleo control, n = 4; Bleo periFoxf1^KO, n = 6. **p = 0.002 & 0.0011.***p = 0.0001. E, F Exacerbated lung fibrosis in periFoxf1^KO lungs at days 14 and 21 post-bleomycin is shown using Sirius Red/Fast Green, Masson’s Trichrome staining and Ashcroft score. Ashcroft score was measured in six random fields in each group. Scale bar = 25 µm. **p = 0.0027. G Immunofluorescence staining for α-SMA (green) showed progressive and severe fibrosis in periFoxf1^KO lungs after bleomycin treatment. Nuclei were counterstained with DAPI (blue). Scale bar = 25 µm. H–J Pericyte-specific deletion of Foxf1 resulted in increased mRNA expression of fibrotic markers (Col1a1, Col3a1, and Acta2) in NG2⁺ pericytes, as assessed by qRT-PCR. RNA was isolated from flow-sorted pericytes. Acta2: Day 14(untreated ctrl, n = 6; bleo ctrl: n = 4, bleo periFoxf1^KO: n = 4, *p = 0.0279, **p = 0.0015, ***p = 0.0003); Day21(untreated ctrl, n = 6; bleo ctrl: n = 3, bleo periFoxf1^KO: n = 3, *p = 0.0145 & 0.0147). Col3a1: Day 14(untreated ctrl, n = 4; bleo ctrl: n = 5, bleo periFoxf1^KO: n = 5, ** p = 0.0055 & 0.0085); Day21 (untreated ctrl, n = 6; bleo ctrl: n = 4, bleo periFoxf1^KO: n = 5, *p = 0.0216, **p = 0.0040). Col1a1: Day 14(untreated ctrl, n = 5; bleo ctrl: n = 6, bleo periFoxf1^KO: n = 6, *p = 0.0212, **p = 0.0010 & 0.0063); Day21(untreated ctrl, n = 4; bleo ctrl: n = 6, bleo periFoxf1^KO: n = 6). Data are presented as mean ± SEM. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. T test (two-tailed) analyses were performed.

To verify that the protective role of FOXF1 in pericytes is not limited to one animal model, we have also used the model of silica-induced pulmonary fibrosis in mice and the ex vivo mouse precision-cut lung slices (PCLS) treated with fibrosis-induced cocktail (Supplementary Fig. 10). In both models, deletion of FOXF1 in pericytes increases the severity of lung fibrosis as quantified by increased collagen depositions (Supplementary Fig. 10A–J). Collectively, all three pre-clinical models of pulmonary fibrosis demonstrate that the loss of FOXF1 in pericytes exacerbates pulmonary fibrosis.

Overexpression of Foxf1 in pericytes before bleomycin injury prevents pulmonary fibrosis

Since deletion of Foxf1 in pericytes exacerbates pulmonary fibrosis, we sought to investigate whether overexpression of Foxf1 in pericytes could mitigate the progression of lung fibrosis. To test this, we generated a pericyte-specific Foxf1 overexpression mouse model, NG2-CreER; Rosa26-LSL-rtTA; TetO7-HA-mFoxf1 (abbreviated as periFoxf1^OE) (Supplementary Fig. 11A). Upon tamoxifen treatment, Cre-mediated recombination removes the LoxP-floxed stop codon, leading to the expression of rtTA in pericytes. In the presence of doxycycline(Dox), rtTA activates the TetO7-CMV promoter, which in turn drives the expression of the mouse Foxf1 transgene. Consequently, the combined administration of tamoxifen and doxycycline results in increased Foxf1 expression specifically in pericytes (Supplementary Fig. 11A). In the absence of bleomycin treatment, the combined administration of tamoxifen and doxycycline did not alter lung histology or the distribution of pericytes and endothelial cells (Supplementary Fig. 11B).

To induce pulmonary fibrosis, Tam- and Dox-treated periFoxf1^OE were given two weekly IT administrations of bleomycin (Fig. 6A). Mice were harvested on day 21 following the first dose of bleomycin. Efficient overexpression of Foxf1 was shown by qRT-PCR using FACS-sorted lung pericytes (Fig. 6B). In the absence of bleomycin injury, Foxf1 in periFoxf1^OE pericytes was approximately six-fold higher than in controls. While bleomycin treatment decreased Foxf1 mRNA in pericytes in both groups, Foxf1 expression levels were higher in periFoxf1^OE pericytes (Fig. 6B). On day 21, bleomycin-treated periFoxf1^OE mice showed remarkable improvements compared to controls, as evidenced by increased oxygen saturation in the peripheral blood (Fig. 6C) and decreased collagen depositions in the lung tissue based on quantitative hydroxyproline assay (Fig. 6D). Fibrotic lung remodeling was attenuated in bleomycin-treated periFoxf1^OE mice, as shown by Ashcroft score assessment, Sirius Red/Fast Green and trichrome staining (Fig. 6E, F). In addition, FACS-sorted lung pericytes from bleomycin-treated periFoxf1^OE lungs demonstrated lower expression of fibrosis-associated Col1a1, Col3a1, and Acta2 mRNAs compared to lung pericytes from bleomycin-treated control mice (Fig. 6G). Altogether, overexpression of Foxf1 in pericytes prior to bleomycin-induced injury attenuates pulmonary fibrosis and inhibits the transition of pericytes to the fibrotic state.

Fig. 6. Pericyte-specific overexpression of Foxf1 prior to bleomycin-induced injury attenuates pulmonary fibrosis.

A Schematic diagram of the experimental timeline for tamoxifen, doxycycline, and bleomycin administration in periFoxf1^OE mice. B qRT-PCR showed efficient expression of Foxf1 in FACS-sorted NG2⁺ pericytes after doxycycline administration (untreated control, n = 4 mice; untreated periFoxf1^OE, n = 3 mice; bleo control, n = 5 mice; bleo periFoxf1^KO, n = 6 mice,*p = 0.0419 & 0.0135 &0.0140, **p = 0.0070). C Oxygen saturation was improved in Dox-treated periFoxf1^OE mice following bleomycin treatment, compared to controls (Dox-untreated periFoxf1^OE). n = 5 per group. **p = 0.0042. D Collagen content, assessed via hydroxyproline assay, was reduced in periFoxf1^OE lungs after bleomycin treatment (untreated control, n = 3 mice; untreated periFoxf1^OE, n = 4 mice; bleo control, n = 4 mice; bleo periFoxf1^KO, n = 5 mice). E Ashcroft score revealed attenuated fibrosis in periFoxf1^OE mice after bleomycin treatment. Ashcroft score was measured across random fields in each group (control: n = 6, periFoxf1^OE, n = 7, ***p = 0.0001). F Reduced fibrotic remodeling periFoxf1^OE mice shown by H&E, Sirius Red/Fast Green, and Masson’s Trichrome staining. Scale bar = 50 µm. Staining was performed on tissue sections from 5 independent mouse samples. G Pro-fibrotic Col1a1, Col3a1, and Acta2 mRNAs were decreased in FACS-sorted pericytes from periFoxf1^OE mice. *p = 0.0448 & 0.0159 & 0.0391. Untreated control: Col1a1, n = 6; Col3a1, n = 6; Acta2, n = 5.Untreated periFoxf1^OE: Col1a1, n = 4; Col3a1, n = 4; Acta2, n = 4.Bleo control: Col1a1, n = 4; Col3a1, n = 5; Acta2, n = 4. Bleo periFoxf1^OE: Col1a1, n = 3; Col3a1, n = 4; Acta2, n = 3. Data are presented as mean ± SEM. *, p < 0.05, **, p < 0.01, ***, p < 0.001, ****, p < 0.0001. T test (two-tailed) analyses and Mann–Whitney test were performed.

Overexpression of Foxf1 in pericytes during already established fibrosis decreases progression of pulmonary fibrosis

Next, we tested whether increasing Foxf1 in pericytes during already established fibrosis will improve the outcomes. Control and periFoxf1^OE mice were treated with Tamoxifen at days 8–12 after bleomycin treatment (Fig. 7A) when mice already had significant collagen depositions³. Pericyte-specific overexpression of Foxf1 during already established fibrosis was sufficient to alleviate pulmonary fibrosis and improve lung functions, as demonstrated by decreased body weight loss in periFoxf1^OE mice (Fig. 7B) and increased oxygen saturation in the peripheral blood (Fig. 7C). Fibrotic lung remodeling was also decreased as demonstrated by decreased hydroxyproline levels, improved Ashcroft score, as well as Sirius Red/Fast Green, and trichrome staining (Fig. 7D–F). FACS-sorted pericytes from periFoxf1^OE lungs had decreased Col1a1, Col3a1, and Acta2 mRNAs compared to controls (Fig. 7G). Thus, pericyte-specific overexpression of Foxf1 during already established lung fibrosis attenuates fibrotic lung remodeling, improves lung function, and delays the transition of pericytes to the fibrotic state.

Fig. 7. Pericyte-specific overexpression of Foxf1 after bleomycin-induced injury reduces pulmonary fibrosis severity.

A Schematic diagram of the experimental timeline for tamoxifen, doxycycline, and bleomycin administration in periFoxf1^OE mice. Doxycycline was administered after bleomycin injury. B Bleomycin-treated periFoxf1^OE mice exhibited significantly less body weight loss compared to control mice (Dox-untreated periFoxf1^OE). Control, n = 3 mice. PeriFoxf1^OE,n = 4 mice. *p = 0.0428. C Oxygen saturation was improved in periFoxf1^OE mice compared to controls. n = 3 mice per group. D–F Pericyte-specific Foxf1 overexpression after bleomycin injury attenuated fibrotic remodeling, as demonstrated by H&E, Sirius Red/Fast Green, and Masson’s Trichrome staining. Scale bar = 50 µm. Fibrosis severity was evaluated using the Ashcroft score, which was measured across random fields in each group (control: n = 8, periFoxf1^OE, n = 7,***p = 0.0002), and collagen content was quantified by hydroxyproline assay (untreated control, n = 3 mice; untreated periFoxf1^OE,n = 4 mice; bleo control, n = 6 mice; bleo periFoxf1^OE,n = 4 mice, ***p = 0.0002). G Decreased Col1a1, Col3a1, and Acta2 mRNAs in FACS-sorted periFoxf1^OE pericytes. *p = 0.0495&0.0312, **p = 0.0099. Untreated control: Col1a1, n = 6; Col3a1, n = 6; Acta2, n = 5.Untreated periFoxf1^OE: Col1a1, n = 4; Col3a1, n = 4; Acta2, n = 4.Bleo control: Col1a1, n = 3; Col3a1, n = 4; Acta2, n = 3. Bleo periFoxf1^OE: Col1a1, n = 3; Col3a1, n = 4; Acta2, n = 4. Data are presented as mean ± SEM. *, p < 0.05, **, p < 0.01, ***, p < 0.001. T test (two-tailed) analyses were performed.

FOXF1 regulates a network of pericyte genes critical to the progression of pulmonary fibrosis and transition of pericytes to the fibrotic state

To investigate the transcriptional targets regulated by FOXF1 in pericytes during pulmonary fibrosis, we performed scRNA-seq comparing lungs from bleomycin-treated control and periFoxf1^KO mice. Pericytes were annotated using mouse-specific pericyte marker genes (Cspg4, Higd1b, Cox4i2, and Pdgfrβ) as defined by LungMap²⁶ (Fig. 8A). The efficiency of Foxf1 knockdown was confirmed by comparing Foxf1 mRNA in pericytes from bleomycin-treated control and periFoxf1^KO mice (Fig. 8B). Consistent with the data from human and mouse fibrotic lungs, we identified two pericyte sub-clusters, Peri1 and Peri2 (Fig. 8C). Peri1 was characterized by high expression of canonical pericyte markers (Cspg4, Higd1b, Cox4i2, Pdgfrβ), whereas Peri2 exhibited decreased expression of pericyte-specific markers and an increased expression of fibrosis-associated genes (Acta2, Col1a1, Col3a1), as well as the pericyte activation marker RGS5 (Supplementary Fig. 12).

Fig. 8. Foxf1 regulates pulmonary fibrosis through downstream transcriptional targets Id3, Aplp2 and Acvrl1 in pericytes.

A UMAP visualization of single-cell RNA-seq data from mesenchymal cells of bleomycin-treated wild-type and periFoxf1^KO mice. Pericyte populations were annotated based on the expression of specific markers (Cox4i2, Cspg4, Higd1b, and Pdgfrβ). B Foxf1 knockout efficiency in pericytes from periFoxf1^KOmice was confirmed via violin plot (VlnPlot) analysis of scRNA-seq data. Single-cell RNA-seq was performed using pooled bleo control (n = 4) and bleo periFoxf1^KO (n = 4) mouse lungs. ****p < 2e-16. Statistical analysis was performed using a two-sided Wilcoxon rank-sum test. Center line indicates the median; box limits indicate the interquartile range (25th–75th percentiles); whiskers indicate 1.5 × IQR. C Two pericyte subclusters were identified in both bleomycin-treated control and bleomycin-treated periFoxf1^KO mice. D Differentially expressed genes (DEGs) in Peri2 from bleomycin treated periFoxf1^KO mice compared to bleomycin-treated control mice. Differential expression analysis between groups was performed with a Wilcoxon rank-sum test. E Positive regulators of the TGF-β signaling pathway (e.g., Tgfbr1, Tgfbr2) were upregulated, while the negative regulator Smad7 was downregulated after Foxf1depletion. F DEGs were identified in Peri2 between bleomycin-treated control and periFoxf1^KO mice. Similarly, DEGs were determined by comparing FOXF1^low Peri2 pericytes from patients with pulmonary fibrosis to FOXF1^hi Peri1 pericytes from healthy donors. Overlapping DEGs from both the mouse and human datasets revealed a subset of genes consistently regulated under fibrotic conditions in both models, suggesting they may represent conserved FOXF1 targets. Pathway enrichment analysis (KEGG and GO Biological Process) of the overlapping DEGs indicated significant enrichment in cytoskeleton regulation, TGF-β signaling, extracellular matrix (ECM) organization, and cell differentiation pathways. G These candidate genes were screened against Foxf1 ChIP-seq data from mice lung tissue. Among them, Id3, Aplp2, Acvrl1 was identified as a direct Foxf1 target, exhibiting both a Foxf1 binding peak and an H3K4me3 activation/H3K27me3 repression mark. H mRNA expression levels of Id3 and Aplp2 were significantly increased, Acvrl1 was significantly decreased in pericytes from periFoxf1^OE mice, Foxf1 overexpression occurred before bleomycin-induced injury. *p = 0.0446&0.0460, **p = 0. 0048.Control: Aplp2, n = 5; Id3, n = 3; Acvrl1, n = 4 (lungs). PeriFoxf1^OE: Aplp2, n = 5; Id3, n = 5; Acvrl1, n = 4 (lungs). I mRNA expression levels of Id3 and Aplp2 were significantly increased, Acvrl1 was significantly decreased in pericytes from periFoxf1^OE mice, Foxf1 overexpression occurred after bleomycin-induced injury. **p = 0.0011, ***p = 0. 0001.Control: Aplp2, n = 8; Id3, n = 7; Acvrl1, n = 3 (lungs). PeriFoxf1^OE: Aplp2, n = 8; Id3, n = 7; Acvrl1, n = 4 (lungs). Data are presented as mean ± SEM. *, p < 0.05, **, p < 0.01, ***, p < 0.001, ****, p < 0.0001. T test (two-tailed) analyses were performed.

To identify genes regulated by FOXF1, we first determined differentially expressed genes (DEGs) in Peri2 from bleomycin-treated periFoxf1^KO mice compared to Peri2 from bleomycin-treated control mice (Fig. 8D). Several positive regulators of the TGF-β signaling pathway, including Tgfbr1, Tgfbr2, Tgfbr3, Acvrl1 were upregulated in the periFoxf1^KO pericytes, while the negative regulator Smad7 was downregulated (Fig. 8E). Next, for human fibrotic lungs, where pericyte-specific FOXF1 knockout is not feasible, we instead compared FOXF1^low fibrotic pericytes from patients with PF to FOXF1^hi pericytes from donors. This allowed us to capture genes associated with reduced FOXF1 expression in the human fibrotic lung. We then cross-compared DEGs from mouse and human datasets to identify overlapped genes consistently regulated under fibrotic conditions in both human and mouse lungs, potentially reflecting conserved FOXF1 targets (Fig. 8F). The overlapped DEGs were subjected to KEGG pathway and Gene Ontology (GO) analysis, demonstrating enrichment in pro-fibrotic signaling pathways, especially TGF-β signaling (Fig. 8F and Supplementary Fig. 13A).

Direct transcriptional targets of FOXF1 in lung pericytes

To determine which of these conserved DEGs are direct transcriptional targets of FOXF1, we used a publicly available Foxf1 ChIP-seq dataset of mouse lung tissue to identify FOXF binding sites. Focusing on genes involved in the regulation of pro-fibrotic pericyte transition, we prioritized five candidate genes, Id3, Id1, Aplp2, Fkbp1a, and Acvrl1, based on strong FOXF1-binding signals in their promoter regions. FOXF1-binding sites in Id3, Id1, Aplp2, Fkbp1a promoter regions coincided with the H3K4me3 binding, indicating active gene transcription and absence of repressive H3K27me3 marks (Fig. 8G and Supplementary Fig. 13B). For Acvrl1, FOXF1 binding was associated with suppressive H3K27me3 marks and lack of active H3K4me3 (Fig. 8G). Thus, based on ChIP-seq data, FOXF1 activates transcription of Id3, Id1, Aplp2, Fkbp1a, but inhibits transcription of Acvrl1.

Since FOXF1 deficiency decreased Id3, Id1, Aplp2, and Fkbp1a and increased Acvrl1 mRNAs in both mouse Foxf1^KO and human FOXF1^low fibrotic pericytes (Supplementary Fig. 13C), we next determined whether Foxf1 overexpression could rescue these changes. Using FACS-sorted lung pericytes from bleomycin-treated periFoxf1^OE and control mice, we have shown that pericyte-specific overexpression of Foxf1 increased expression of Id3 and Aplp2, and decreased expression of Acvrl1 compared to bleomycin-treated controls, regardless of whether Foxf1 was overexpressed prior to bleomycin injury (Fig. 8H) or after (Fig. 8I). Id1 and Fkbp1a mRNAs were not changed after Foxf1 overexpression, indicating that these genes are not the direct targets of Foxf1 (Supplementary Fig. 13D, E). These findings indicate that Foxf1 directly activates Id3 and Aplp2, while represses Acvrl1 in pericytes.

To further assess the FOXF1-dependent roles of ID3, APLP2 and ACVRL1 in pericyte functions, we have restored ID3 or APLP2 expression in the FOXF1-deficient human lung pericytes using expression plasmids in vitro (Fig. 9A). We have also inhibited ACVRL1 expression in pericytes using siACVRL1 (Fig. 9A). Restoration of either ID3 or APLP2, or depletion of ACVRL1 in FOXF1-deficient pericytes decreased fibrosis-associated ACTA2, COL1A1, and COL3A1 mRNAs, indicating that each gene individually contributed to the fibrotic phenotype of FOXF1-deficient pericytes (Fig. 9B). Since FOXF1 expression in endothelial cells is inhibited by hypoxia^3,29, we have also tested if hypoxia suppresses FOXF1 in pericytes. Treatment of human lung pericytes with either 1% oxygen or cobalt chloride in vitro decreased FOXF1 mRNA, as shown by qRT-PCR (Supplementary Fig. 14A, B), suggesting that hypoxic conditions in fibrotic lungs can contribute to the inhibition of FOXF1 expression in pericytes.

Fig. 9. Restoring expression of ID3, APLP2 or suppressing ACVRL1 expression in FOXF1-deificient pericytes rescues fibrotic phenotype of lung fibroblasts.

A FOXF1 siRNA and with ID3^OE plasmid or APLP2^OE plasmid orlACVRL1 siRNA, followed by treatment with human TGFβ. B ACTA2, COL1A1 and COL3A1 mRNAs were decreased after rescuing the expression of ID3, APLP2 or suppressing ACVRL1 in the FOXF1-deficient pericytes. n = 5 biological replicates per group. ACTA2, *p = 0.0401, ***p = 0.0002. COL1A1, ***p = 0.0004, 0.0001. COL3A1, **p = 0.0082, 0.0022, 0.0054, 0.0075. ***p = 0.0004, 0.0006, 0.0007. C, D Human lung fibroblasts were cultured with conditioned medium from FOXF1^KD or FOXF1^KD + ID3 OE plasmid or FOXF1^KD + APLP2 OE plasmid or FOXF1^KD + ACVRL1 siRNA pericytes. Restoration of ID3 or APLP2, or inhibition of ACVRL1, suppressed fibroblast migration in a wound-healing assay (n = 4). Scale bar = 250 µm. **p = 0.0032&0.0085, ***p = 0.0009. E ID3 or APLP2 restoration, or ACVRL1 inhibition suppress fibroblast proliferation, n = 6 biological replicates per group. F, G Conditioned medium from FOXF1^KD and FOXF1^KD + ID3^OE pericytes was analyzed using a Proteome Profiler Human Cytokine Array. Increasing ID3 expression in FOXF1-deficient pericytes led to decreased secretion of IL-8 and CXCL1 (pixel density was quantified from duplicate spots for each analyte using two membranes). ***p = 0.0001. H The efficiency of ID3/APLP2 overexpression and the efficiency of ACVRL1 knockdown were verified by qRT-PCR. **p = 0.0024, ***p = 0.0003, 0.0002. n = 5 biological replicates per group. Data are presented as mean ± SEM. *, p < 0.05, **, p < 0.01, ***, p < 0.001; ****, p < 0.0001. T test (two-tailed) analyses were performed.

FOXF1-deficient pericytes increase fibroblast activation in vitro

Since lung fibroblasts are responsible for collagen deposition in pulmonary fibrosis, and CellChat analysis predicted the extensive crosstalk between lung pericytes and fibroblasts (Fig. 2 and Supplementary Fig. 3), we assessed the contribution of FOXF1-deficient lung pericytes to the activation of fibroblasts in vitro. Human lung fibroblasts were cultured in the presence of conditioned medium (CM) from control and FOXF1-deficient human lung pericytes (Supplementary Fig. 9A, B). CM from FOXF1-deficient lung pericytes increased migration, proliferation and expression of pro-fibrotic CTHRC1, COL1A1 and ACTA2 mRNAs in lung fibroblasts compared to fibroblasts cultured in CM from control pericytes (Supplementary Fig. 9C–F). Thus, FOXF1-deficient pericytes promoted a pro-fibrotic phenotype in lung fibroblasts in vitro.

We next tested whether pericyte-derived ID3, APLP2 and ACVRL1 function downstream of FOXF1 to activate lung fibroblasts. Human lung fibroblasts were cultured with conditioned medium from FOXF1-deficient lung pericytes with or without restored expression of ID3 or APLP2, or from FOXF1-deficient pericytes after ACVRL1 depletion (Fig. 9C–E). Migration and proliferation assays demonstrated that ID3, APLP2 and ACVRL1 in pericytes independently contribute to the regulation of pro-fibrotic behavior in lung fibroblasts (Fig. 9C–E). We also assessed the changes in pericytes’ conditioned medium after ID3 restoration using a human cytokine array. Restoration of ID3 in the FOXF1-deficient human lung pericytes decreased the secretion of pro-fibrotic mediators, such as IL-8 and CXCL1 (Fig. 9F–H). Thus, ID3 is one of the FOXF1-dependent regulators of pericyte homeostatic functions, acting in part through regulation of pro-fibrotic cytokines.

Altogether, these findings demonstrate that FOXF1 directly regulates a network of pericyte genes critical to the progression of pulmonary fibrosis and transition of pericytes to the fibrotic state.

Discussion

Pulmonary fibrosis is a complex, multicellular process involving well-documented contributions from epithelial cells³⁰, endothelial cells^3,31, fibroblasts³² and immune cells³³. These cell types play critical roles in injury-induced aberrant activation of fibroblasts, fibrotic remodeling of the interstitium and inflammatory cell recruitment, all of which drive the progression of pulmonary fibrosis. Recent studies have identified several unique cell populations associated with pulmonary fibrosis, including Sfrp1⁺ transitional fibroblasts implicated in matrix production³⁴, distinct epithelial and mesenchymal cell types³⁵, and a novel population of profibrotic alveolar macrophages³⁶. Despite these advancements, the role of pericytes in pulmonary fibrosis has often been underestimated.

In several organ tissues, including the heart and kidney, emerging evidence suggests that pericytes can serve as a source of myofibroblasts^37,38. Their potential to adopt a fibrotic phenotype has drawn attention to their possible involvement in fibrotic diseases. Recent studies have identified a specialized pericyte subtype in pulmonary capillaries during hypoxia-induced pulmonary hypertension³⁹, however, its role in fibrotic lung remodeling is unclear. In the present study, we used scRNA-sequencing to identify a distinct pericyte population in the lung tissue of PF patients. These pericytes were exclusively observed in the lungs of PF patients and bleomycin-treated mice but were absent in normal donor lungs and untreated mice. This fibrosis-associated pericyte population exhibited a unique gene signature, including elevated expression of collagens and pro-fibrotic markers, suggesting their contribution to the promotion of a fibrotic microenvironment and fibrotic remodeling. In fibrotic lungs, differentiating fibrosis-associated pericytes from smooth muscle cells is challenging since both cell populations can acquire overlapping mesenchymal phenotypes and express common markers, including as α-SMA (ACTA2), TAGLN, and PDGFRβ during injury-induced remodeling. In addition, activated pericytes may transition toward myofibroblast-like states, while vascular smooth muscle cells can also change their transcriptional programs and cell state. Thus, the precise cell type identification dependent on combining lineage tracing, spatial localization, and multi-marker analysis rather than single-marker staining alone. In our current studies, we have used a lineage-tracing approach and pseudotime analysis of mouse and human scRNAseq datasets to demonstrate that Peri2 cells can derive from Peri1 cells. A limitation of our study is that Peri2 cells can potentially derive from smooth muscle cells, and this cannot be distinguished by immunostaining.

The significant contribution here is that we have identified a cluster of fibrotic pericytes (Peri2) that is uniquely present in fibrotic but not in normal lungs. We have also identified a transcriptional signature of fibrotic pericytes, and an evolutionary conserved signature of differentially expressed transcription factors in fibrotic pericytes from patients with pulmonary fibrosis and bleomycin-treated mice. Several of them have been already shown to be important in PF, including ID1, DUSP4, and HEYL^40–42. Id1-deficient mice exhibited increased susceptibility to bleomycin-induced lung injury⁴⁰. DUSP4 has been reported to attenuate cardiac fibrosis by suppressing the TGF-β/p38 signaling pathway⁴¹. Inhibition of NOTCH3 signaling in a HEYL-dependent manner has been found to mitigate meniscus fibrosis⁴². The anti-fibrotic roles of these transcription factors align with their observed downregulation in fibrotic pericytes in our study. Conversely, upregulated transcription factors in fibrotic pulmonary pericytes include PRRX1, FOSB, BNC2, and SOX9. These factors are recognized for their pro-fibrotic functions in promoting extracellular matrix (ECM) gene expression and myofibroblast differentiation^43–46, consistent with their upregulation in fibrotic pericytes. Among the most significantly downregulated transcription factors in pericytes, FOXF1 emerged as a key candidate, with its downregulation observed in both PF patients and bleomycin-treated mice. Consequently, we used mouse genetic models to focus on elucidating the role of FOXF1 in fibrosis-associated changes in pericytes.

Previous studies have demonstrated that FOXF1 has anti-fibrotic functions in fibroblasts and endothelial cells^24,47. For instance, FOXF1-dependent CDH2-CD11 switching in fibroblasts is essential for their transition to a pro-fibrotic phenotype²⁴. FOXF1 has been shown to alleviate silica-induced pulmonary fibrosis by inhibiting the TGF-β/SMAD2/3 pathway, decreasing myofibroblast differentiation and ECM accumulation⁴⁷. FOXF1 has also been implicated in regulating endothelial cell reprogramming into a pro-fibrotic phenotype³. In addition to lung endothelial cells and lung fibroblasts, FOXF1 is highly expressed in normal lung pericytes. Our results demonstrate that loss of FOXF1 in pericytes exacerbates pulmonary fibrosis, whereas pericyte-specific FOXF1 restoration mitigates fibrotic progression. Together with the previously published studies, our findings indicate that FOXF1 acts as an anti-fibrotic factor across multiple cell types involved in pulmonary fibrosis, underscoring its potential as a therapeutic target in human IPF.

In the current work, we have also identified the network of FOXF1- regulated pericyte genes important for pulmonary fibrosis. Cross-comparison of human and mouse DEGs in fibrotic pericytes and KEGG/GO pathway analysis⁴⁸ demonstrated that FOXF1-depleted pericytes were characterized by increased pro-fibrotic signaling pathways, including TGF-β signaling, extracellular matrix organization, actin cytoskeleton regulation, and cell differentiation. TGF-β signaling is known to promote mesenchymal cell differentiation, including myofibroblast differentiation, during the progression of pulmonary fibrosis^49,50. Regulation of the actin cytoskeleton pathways are intricately linked to TGF-β signaling, contributing to pro-fibrotic phenotype and extracellular matrix (ECM) deposition⁵¹. Using ChIP-seq analysis and in vivo Foxf1 overexpression experiments, we identified several direct transcriptional targets of FOXF1 in pericytes. ID3 and APLP2 are known for their anti-fibrotic activity and were downregulated in the fibrotic FOXF1-deficient pericytes. ID3, a member of the inhibitor of DNA binding (Id) family of transcription factors, has been shown to suppress COL1A2 promoter activity in TGF-β1-stimulated pulmonary myofibroblasts when overexpressed⁵². In this study, we found that Id3 was significantly downregulated in mouse Foxf1^ko fibrotic pericytes and human FOXF1^low fibrotic pericytes. APLP2, a type I transmembrane glycoprotein, inhibits TGF-β signaling and regulates ECM-receptor interactions. It inhibits TGF-β signaling by promoting degradation of the TGFBR2⁵³ and facilitates cell adhesion to fibronectin and collagen IV⁵⁴. Acvrl1, was found to be upregulated in Foxf1^KO group. This gene encodes a type I cell surface receptor that belongs to the TGF-β superfamily of ligands⁵⁵. Our co-culture experiments with FOXF1-deficient pericytes were consistent with the hypothesis that ID3, APLP2 and ACVRL1 contribute to pericyte re-programming and increased fibroblast activation downstream of FOXF1.

In summary, FOXF1 exhibits an anti-fibrotic function in lung pericytes during pulmonary fibrosis. Loss of FOXF1exacerbates the severity of pulmonary fibrosis, characterized by increased collagen deposition, tissue remodeling, loss of lung function and increased mortality. FOXF1 is important in maintaining pericyte homeostasis and suppressing the pericyte transition to fibrotic phenotype. Further investigation into the precise molecular pathways downstream of FOXF1 could unveil promising therapeutic targets for the treatment of pulmonary fibrosis, offering new avenues for intervention in this devastating disease.

Methods

Ethical statement

The use of human lung tissue was approved by the Institutional Review Board (IRB) of the University of Arizona. Donor and human idiopathic pulmonary fibrosis (IPF) lung tissue specimens were obtained from the tissue repository and explanted lungs at the University of Cincinnati Medical Center and St. Joseph’s Hospital in Arizona under the protocols approved by the Institutional Review Board (IRB) of the University of Cincinnati and St. Joseph’s Hospital Medical Center. Written informed consent was obtained from all participants. All samples were provided in a de-identified manner. All animal care and experimental protocols were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Arizona. Lung tissue sections were from both female and male human patients. For mouse experiments, both males and females were included in equal numbers.

Transgenic mice

Both male and female 8–10-week-old mice were used in this study. Homozygous Foxf1^f/f mice were crossed with heterozygous NG2-CreER mice to generate NG2-CreER; Foxf1^f/f mice (abbreviated as periFoxf1^KO), enabling pericyte-specific depletion of Foxf1 upon Tamoxifen administration. Rosa26-LSL-tdTomato mice were crossed with NG2-CreER mice to generate NG2-CreER; Rosa26-LSL-tdTomato mice (abbreviated as NG2; LSL-tdT). To induce Cre recombinase activity, Tamoxifen (Sigma-Aldrich, St. Louis, MO, USA) were administered intraperitoneally (I.P.) into mice at a dose of 80 µg/g body weight per day for 5 consecutive days. To achieve pericyte-specific overexpression of Foxf1, TetO7-HA-mFoxf1³ was crossed with NG2-CreER mice and Rosa26-LSL-rtTA mice (abbreviated as periFoxf1OE). For Foxf1 overexpression in pericytes, Tamoxifen (Sigma-Aldrich, St. Louis, MO, USA) were administered intraperitoneally (80 µg/g body weight per day) for 5 consecutive days, followed by maintenance on doxycycline-containing chow (Inotiv, Indianapolis, IN, USA) for the duration of the experiment. Mice are maintained in a specific pathogen–free animal facility under controlled environmental conditions (12 h light/dark cycle, temperature 22 ± 2 °C, relative humidity 40–60%) with free access to standard chow and water.

Bleomycin-induces and silica-induced pulmonary fibrosis models

To induce pulmonary fibrosis, mice were administered bleomycin sulfate (EMD Biosciences, San Diego, CA, USA) intratracheally (I.T.) at doses of 1.5 µ/kg or 2 µ/kg once per week for 2 weeks⁵⁶. Mice were harvested on day 14 or day 21. Control mice received phosphate-buffered saline (PBS) intratracheally once per week for 2 weeks. Control groups included Foxf1^f/f mice receiving tamoxifen/PBS and Foxf1^f/f mice receiving tamoxifen/bleomycin treatments. For the silica-induced fibrosis mouse model, silica particles (Sigma-Aldrich, St. Louis, MO, USA) were washed in 1 N HCl at 100 °C for 2 h, rinsed thoroughly with sterile water, autoclaved for 1 h, then dried and suspended in PBS. The silica suspension was sonicated before administration. Mice were administered silica intratracheally (I.T.) at a dose of 200 mg/kg once and harvested on day 21.

Precision-cut lung slices (PCLS) generation

Lung lobes were inflated and embedded in 2% low-melting point agarose (Invitrogen, Waltham, Massachusetts, USA) in DMEM/F12 (Gibco, Billings, Montana, USA), and then sliced (300 μm thickness) using a vibrating microtome (Precisionary Instruments, Ashland, Massachusetts, USA). After slicing, PCLS were cultivated in DMEM/F12 supplemented with 10% FBS,2% Antibiotic-Antimycotic (Gibco, Billings, Montana, USA). PCLS was treated with a cocktail consisting of 5 ng/ml TGFβ (Sino Biological, Beijing, China), 10 ng/ml PDGF-AB (Gibco, Billings, Montana, USA), 10 ng/ml TNFα (Proteintech, Rosemont, Illinois, USA), 5 µM LPA (Cayman Chemical, Ann Arbor, Michigan, 48108) for 10 days to induce ex vivo fibrosis. Medium was changed 3 times.

Respiratory mechanics and oxygen saturation

Mice were anesthetized on day 14 or day 21 following the initial bleomycin treatment. Oxygen saturation (SpO₂) was measured using the MouseOX Plus pulse oximeter (STARR Life Sciences, Oakmont, PA, USA). A collar sensor was utilized to obtain signals from the carotid arteries of the mice. Data acquisition and analysis were performed using the MouseOX Plus 1.6.x software. In vivo respiratory mechanics were assessed using the FlexiVent system (Scireq, Montreal, QC, Canada). The collected data was analyzed with FlexiVent software version 7.5.

Immunofluorescent staining and histological staining

Lung paraffin sections were prepared for histological staining, including hematoxylin and eosin (H&E), Sirius Red/Fast Green, immunohistochemistry, and Masson’s Trichrome staining⁵⁷. Lung tissue sections were from both female and male patients. For mouse experiments, both males and females were included in equal numbers. Both frozen and paraffin lung sections were used for immunofluorescent staining. For paraffin sections, mouse lung tissues were perfused with phosphate-buffered saline (PBS) and inflated with 10% formalin. The tissues were fixed in 10% formalin and embedded in paraffin. For frozen sections, lung tissues were perfused with PBS, inflated with a 1:1 mixture of 30% sucrose and OCT, and fixed in 4% paraformaldehyde for 1 h. Tissues were then transferred to 30% sucrose overnight before embedding in OCT. For immunofluorescent staining, lung sections were rehydrated in PBS (two washes, 5 min each) and permeabilized with PBST. After blocking with 3% bovine serum albumin (BSA) in PBST for 1 h, primary antibodies were applied (Supplementary Table T1). Slides were incubated overnight at 4 °C and Alexa Fluor-conjugated secondary antibodies (Alexa Fluor 488, 594, or 647) and DAPI (Thermo Fisher Scientific, Waltham, MA, USA) were added. Images were acquired using the EVOS Imaging System, the Zeiss Axio Imager M2 upright widefield fluorescent microscope, and the Zeiss LSM710 laser scanning confocal microscope. For histological staining, paraffin slides were deparaffinized and rehydrated with xylene and a graded ethanol series of decreasing concentrations. Staining was performed following the protocols for H&E (ScyTek Laboratories, Logan, UT, USA), Sirius Red/Fast Green (Chondrex, Woodinville, WA, USA), and Masson’s Trichrome Staining Kit (ScyTek Laboratories, Logan, UT, USA).

Collagen quantification and fibrosis severity assessment

Quantification of lung collagen contents were performed using the hydroxyproline assay (Cell Biolabs, San Diego, CA, USA) following the manufacturer’s protocols. The severity of fibrosis was assessed by the Ashcroft score⁵⁸. The score was assigned from 0(no fibrosis) to 8(fibrous obliteration) based on the sections observed under an EVOS brightfield microscope.

qRT-PCR

RNA was extracted from FACS-sorted lung pericytes, including NG2⁺ cells from periFoxf1KO and periFoxf1OE mice or tdTomato⁺ cells from NG2; LSL-tdT mice, using the Quick-RNA Microprep Kit (Zymo Research, Irvine, CA, USA). The RNA was then reverse transcribed into cDNA using either the High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, Waltham, MA, USA) or the iScript™ cDNA Synthesis Kit (Bio-Rad, Hercules, CA, USA). Quantitative real-time PCR (qRT-PCR) was performed on a QuantStudio 3 system using Taqman probes or SYBR Green probes, as detailed in the supplementary table (Supplementary Table T2). GAPDH and 18S were used as internal controls.

Flow cytometry and flow sorting

Flow cytometry experiments on mouse lung tissue were performed as previously described^59,60, using the antibodies listed in the supplementary table(Supplementary Table T3). Stained cells were analyzed with an Attune NxT Flow Cytometer, provided by the Flow Cytometry Core Facility at the University of Arizona. Pericytes were sorted using a BD FACS Symphony S6, available through the Flow Cytometry Core Facility at Arizona State University. Data analysis was conducted using FlowJo software (version 10.8.1).

10x Genomics single cell RNA-seq and data analysis

The publicly available human lung scRNA-seq datasets from GSE227136 was used that included lung tissue samples from 116 individuals, including 67 patients with pulmonary fibrosis (PF) and 49 unaffected donors²⁵. The publicly available murine lung scRNA-seq datasets from GSE210341² and GSE132771²⁸ were used. The datasets include lung tissue samples collected at day 0 (non-bleomycin-treated), 7, 14, and 21(after bleomycin treatment).

Single-cell RNAseq libraries were generated from day 14 bleomycin-treated control and periFoxf1KO mice using the Chromium Next GEM Single Cell 3’ Reagent Kits v3.1 (Dual Index) (10x Genomics, Pleasanton, CA, USA) according to the manufacturer’s protocol by the Genomics Core at Arizona State University. Three or four lungs in each group were combined to prepare sequencing libraries. To enrich for pericytes, hematopoietic cells (CD45⁺) were depleted from the cell suspensions using FACS-sorting. Approximately 11,000 cells were loaded for each sample, with a targeted cell recovery estimate of 7000 cells. The libraries were sequenced with NovaSeq X at a sequencing depth of 3.25 billion reads per sample provided by the TGen-Translational Genomics Research Institute. Cellranger mkfastq was used to extract sequence reads in FASTQ format. The cellranger count with mm10 reference genome (as provided by 10X) and the Cell Ranger software (v8.0.0, 10X Genomics) were used to generate gene-count matrices from the FASTQ files. Reads were processed by counting Unique Molecular Identifiers (UMI). Cells with unique feature counts less than 500 and more than 8000 were excluded from the analysis. Low-quality cells were removed based on the fraction of mitochondrial genes ( > 20%). Data filtering, normalization criteria and principal component analysis are described in the previous paper²¹.

To further ensure that clustering would not be impacted by batch effects or biological variability associated with mouse age or oxygen conditions, we performed the data integration method implemented by the Seurat software (version 4.4.0 in 4.4.2 environment) for SCTransform-normalized data. The data were log-normalized with a scale factor of 10⁴. Variable features were determined by the FindVariableGenes function. FindIntegrationAnchors and IntegrateData functions were used to integrate datasets. Next, highly variable features from the integrated matrix were used as input for linear dimensionality reduction via principal component analysis (PCA). Principal components (PCs) with standard deviation > 3.5 were used as input for unsupervised clustering with a resolution set to 0.2, using the nonlinear UMAP down dimensional algorithm.

For differential analysis of gene expression, pericytes were defined using the Idents function in the Seurat package. Next, the FindMarkers function was used with the logFC threshold set to 0.25 to identify genes downregulated or upregulated between samples. For molecular pathway analysis, the functional annotation clustering was performed using DAVID Bioinformatics Resources Ver.6.8. The KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway enrichment analysis was used to produce the annotation data. Functional clusters with Benjamini-corrected P- value  < 0.05 were considered statistically significant.

Cell lines, wound healing and proteome profiler human cytokine array

Human lung fibroblast (MRC5) and human lung pericyte cell lines, along with the corresponding culture medium kit, were commercially obtained. A cocktail of siRNAs was used to knockdown FOXF1 or ACVRL1 in human lung pericytes using Dharmafect transfection reagent according to the manufacturer’s protocol (Dharmacon, Lafayette, Colorado, USA). A cocktail of non-targeting siRNAs was used as a control. To create the human ID3 and APLP2 expression vectors, the vector backbone was amplified using Primestar Max polymerase (Takara, Kusatsu, Shiga, Japan) with primers in the supplementary table (Supplementary Table T4). Human pulmonary artery endothelial cells (HPAECs) RNA was extracted using the Quick-RNA micro prep kit (Zymo Research, Irvine, CA, USA) following the manufacturer’s instructions. cDNA synthesis was completed using Induro Reverse Transcriptase (New England Biolabs, Ipswich, Massachusetts, USA) following the manufacturer’s instructions. ID3 and APLP2 cDNA sequences were amplified with homology to the vector using Primestar Max polymerase and primers in the supplementary table (Supplementary Table T4). The vector and insert were purified using the Nucleospin Gel and Clean up kit (Takara, Kusatsu, Shiga, Japan) following the manufacturer’s instructions. The purified vector and inserted were incubated with Infusion snap assembly master mix following the manufacturer’s instructions (Takara, Kusatsu, Shiga, Japan). ID3 and APLP2 mixes were then transformed into chemically competent Stellar cells (Takara, Kusatsu, Shiga, Japan). Cell migration was evaluated using a wound healing assay, with the gap created by 2-well Culture-Inserts (Ibidi, Fitchburg, WI, USA). Proteome profiling was performed using the Proteome Profiler Human Cytokine Array Kit (R&D Systems, Minneapolis, MN, USA) following the manufacturer’s instructions.

Statistical analysis

Statistical analysis and graphical data presentation were performed using GraphPad Prism 10 (GraphPad Software, Inc., San Diego, CA, USA). Normally distributed variables between control and experimental groups were compared using a two-tailed Student’s t test, while non-normally distributed variables were analyzed with the Mann-Whitney U test. Data are presented as mean ± SEM (standard error of the mean), and a P- value of less than 0.05 was considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

W.G. and T.V.K. conceived and designed the study. W.G., Y.W.L., Z.D., A.A., X.X., D.L., A.A.S., and N.M. performed the experiments. W.G., Y.W.L., N.E.B., and E.L. performed the sc-RNA-seq and RNA-seq bioinformatics analysis. W.G. and T.V.K. analyzed data. W.G., V.V.K., and T.V.K. interpreted the data. J.D., A.S., R.B., and V.K.K. provided critical sample and intellectual discussions. W.G. and T.V.K. wrote the paper. All authors discussed the data. T.V.K. approved the submission of the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the NIH grants R01 HL132849 (T.V.K.), R01 HL158659 (T.V.K.), R01 HL141174 (V.V.K.), R01 HL149631 (V.V.K.), R01 HL152973 (V.V.K./T.V.K.).

Data availability

Single-cell RNA-seq datasets of donor and PF lungs were retrieved from the GEO database (accession number GSE227136 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= GSE227136). Single-cell RNA-seq datasets of bleomycin-treated and control untreated mouse lungs were retrieved from the GEO database (accession number GSE210341 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE210341 and GSE132771 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE132771). Single-cell RNA-seq datasets generated in this study of bleomycin-treated control and periFoxf1^KO mice is also deposited in the GEO database (accession number GSE331327 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE331327) Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74910-3.