Histological signatures map anti-fibrotic factors in mouse and human lungs

Abstract

Fibrosis, the replacement of healthy tissue with collagen-rich matrix, can occur following injury in almost every organ^1,2. Mouse lungs follow a stereotyped sequence of fibrogenesis-to-resolution after bleomycin injury³, and we reasoned that profiling post-injury histological stages could uncover pro-fibrotic versus anti-fibrotic features with functional value for human fibrosis. Here we quantified spatiotemporally resolved matrix transformations for integration with multi-omic data. First, we charted stepwise trajectories of matrix aberration versus resolution, derived from a high-dimensional set of histological fibre features, that denoted a reversible transition in uniform-to-disordered histological architecture. Single-cell sequencing along these trajectories identified temporally enriched 'ECM-secreting' (Csmd1-expressing) and 'pro-resolving' (Cd248-expressing) fibroblasts at the respective post-injury stages. Visium-based spatial analysis further suggested divergent matrix architectures and spatial-transcriptional neighbourhoods by fibroblast subtype, identifying distinct fibrotic versus non-fibrotic biomolecular milieu. Critically, pro-resolving fibroblast instillation helped to ameliorate fibrosis in vivo. Furthermore, the fibroblast neighbourhood-associated factors SERPINE2 and PI16 functionally modulated human lung fibrosis ex vivo. Spatial phenotyping of idiopathic pulmonary fibrosis at protein level additionally uncovered analogous fibroblast subtypes and neighbourhoods in human disease. Collectively, these findings establish an atlas of pro- and anti-fibrotic factors that underlie lung matrix architecture and implicate fibroblast-associated biological features in modulating fibrotic progression versus resolution.

Extended Data Fig. 1 ∣. Alveolar cells and fibroblasts exhibit differential signaling and enrichment during fibrosis and post-fibrotic resolution.

(A) Specific ligand-receptor interactions between fibroblasts and either alveolar type II (ATII) or type I (ATI) cells. Fibroblast-ATII interactions at PID 14 are inferred to be largely similar to fibroblast-ATI interactions at PID 35 (red box). (B) Timepoint-specific differences in alveolar cells primarily consist of an established ATII-ATI transitional signature (Clu/Krt8/Krt18) enriched in ATI cells at PID 35. (C) Pdgfra, Col1a1, and Col3a1 expression by fibroblast subtypes. ECM-secreting fibroblasts (red boxes) exhibit notably higher average expression of Col1a1 and Col3a1 compared to other fibroblast subtypes. (D-F) Timepoint-dependent enrichment (D), top 3 differentially expressed genes (E), and top 10 gene ontology terms (F) of fibroblast subtypes. Red boxes indicate the relevant fibroblast subtype. P-values calculated based on the two-tailed Fisher’s exact test in enrichR.

Extended Data Fig. 2 ∣. Pro-resolving fibroblast delivery enhances in vivo resolution of fibrosis.

(A) Delivery of freshly FACS-sorted fibroblasts to mice for investigation of functional impact. Created with BioRender. Guo, J. https://BioRender.com/l13t441 (2025). (B) Impact of fibroblast subtype enrichment during peak fibrosis on downstream collagen content (left; n = 3 mice for post-injury day [PID] 21 and all baseline timepoints; n = 4 mice for all other treatments and timepoints) (Lin−: p = 0.294/0.438/0.750/0.688; Pro-Resolving: p = 0.043/0.049/0.696/0.688; ECM-Secreting: p = 0.600/0.873/0.696/0.688; for PID 28/35/42/49, respectively, vs. baseline), matrix architecture (middle; n = 5 mice) (Lin−: p = 0.035/0.004/0.027/0.851; Pro-Resolving: p < 0.0001/0.0001/0.0001/p = 0.0004; ECM-Secreting: p = 0.484/0.806/0.243/0.383; for PID 28/35/42/49, respectively, vs. baseline), and immature collagen staining (right; n = 5 mice) (Lin−: p = 0.125/0.636/0.502/0.913; Pro-Resolving: p = 0.005/0.179/0.306/0.710; ECM-Secreting: p = 0.009/0.001/0.706/0.036; for PID 28/35/42/49, respectively, vs. baseline). Pro-resolving fibroblasts help accelerate post-fibrotic resolution when enriched at PID 21. * indicates statistically significant difference from baseline at same timepoint (p < 0.05). (C) Immunofluorescence of instilled tdTomato+ fibroblasts after PID 21 indicates that both ECM-secreting and pro-resolving fibroblasts integrate within the interstitium, largely within less fibrotic areas. Orange arrows indicate subtype-positive instilled cells, while blue arrows indicate subtype-negative instilled cells. Results are representative of n = 3 independent experiments. (D) FACS quantification suggests that instilled fibroblasts proliferate from post-delivery day 1 to 7 (Pro-Resolving: p = 0.010/0.010; ECM-Secreting: p = 0.002/0.048; for post-delivery day 7/14, respectively, vs. day 1), with ECM-secreting fibroblasts losing their original phenotype during the post-fibrotic phase (Pro-Resolving: p = 0.355/0.140; ECM-Secreting: p < 0.0001/0.0001; for post-delivery day 7/14, respectively, vs. day 1) (n = 3 mice for post-delivery day 1 and 14; n = 4 mice for post-delivery day 7). * indicates statistically significant difference from post-delivery day 1 (p < 0.05). For panels B and D, p-values are based on ANOVA with post-hoc testing by Fisher’s least significant difference and Benjamini-Hochberg correction (all tests two-tailed), while all values are displayed as mean ± standard deviation.

Extended Data Fig. 3 ∣. In situ cell proportions and spatial interactions vary by post-injury stage.

(A) Cell proportions at PID 14 and 35, normalized to baseline. (B-C) −Log10 normalized P values of differential interactions in fibrotic lungs (B) and differential interactions between fibrotic and post-fibrotic lungs (C). For panels B and C, p-values were calculated by an unpaired, two-tailed t-test.

Extended Data Fig. 4 ∣. Alveolar cells and fibroblasts strongly influence tissue-level cellular organization in fibrotic and resolving lungs.

(A) Eigenvector centrality scores at PID 14 and 35, representing influence of individual cell types on tissue-level cellular organization. ATI cells, ATII cells, and fibroblasts constitute the highest-scoring nodes at both stages. (B-C) −Log10 normalized P values of differential fibroblast-alveolar cell interactions in fibrotic lungs (B) and differential fibroblast-alveolar cell interactions between fibrotic and post-fibrotic lungs (C). For panels B and C, p-values were calculated by an unpaired, two-tailed t-test.

Extended Data Fig. 5 ∣. Fibroblast subtypes are inferred to reside in cellular niches that are either relatively immune/endothelium-adjacent or fibroblast-adjacent.

(A) Identification of cellular niches associated with each fibroblast subtype. ECM-secreting and pro-resolving fibroblasts appeared to occupy distinct niches that are comparatively immune/endothelium-adjacent (Spatial Niche A, top box) or fibroblast-adjacent (Spatial Niche B, bottom box), respectively. (B) Fibroblast subtype-associated signaling pathways in CellChat. ECM-secreting fibroblasts are involved in pro-fibrotic pathways such as Collagen and Spp1 (red box), while pro-resolving fibroblasts are involved in other pathways such as Galectin and Tenascin (blue box).

Extended Data Fig. 6 ∣. ECM-secreting and pro-resolving fibroblasts are enriched in regions of fibrotic and non-fibrotic histological architecture, respectively.

(A) Mapping of matrix architecture onto Visium to identify spatially enriched cellular/transcriptional features. (B) Correlation of localized cell-cell interactions with matrix architecture and stained matrix quantity in all bleomycin-treated lungs. Top 5 positively and negatively correlated interactions are shown, suggesting that regions of fibrotic architecture are predominantly associated with communities of ATI cells, fibroblasts, and macrophages (box). (C) Fibroblast subtypes enriched in regions of differential matrix pseudotime and matrix quantity, demonstrating that ECM-secreting and pro-resolving fibroblasts are localized in regions of high and low matrix pseudotime, respectively (boxes). Conversely, ECM-secreting and baseline alveolar fibroblasts are the most enriched subtypes in regions of high and low matrix quantity, respectively, suggesting that quantification of matrix architecture may elucidate unique biological features not apparent based on analysis of matrix quantity alone.

Fig. 1 ∣. A reversible transition between uniform and disordered matrix architecture defines fibrotic and post-fibrotic histological trajectories across time.

a, Automated matrix analysis pipeline comprising 294 ultrastructural features. Created with BioRender. Guo, J. https://BioRender.com/l13t441 (2025). b, Minimum spanning tree-based model of architectural progression, including representative images of fibre organization along the identified pseudotime trajectory from root (white dot) to terminus (grey dot). c, Temporal progression of interstitial matrix architecture following bleomycin exposure. Fibrotic lungs exhibited progressive matrix aberration, followed by resolution to baseline architecture (green dashed outline) by PID 42–49. d, Kinetics of matrix progression (left, n = 5 mice) identify PID 14 and 35 as snapshots of active fibrosis and post-fibrotic resolution, respectively (left, n = 5 mice; PID 7–35, P < 0.0001; PID 42, P = 0.191; PID 49, P = 0.498 versus PID 0). Modelled kinetics were supported by quantification of normalized collagen content (middle, n = 3 mice; PID 7, P = 0.502; PID 14, P = 0.177; PID 21, P = 0.028; PID 28, P = 0.112; PID 35, P = 0.177; PID 42, P = 0.415; PID 49, P = 0.502 versus PID 0) and lung radiological density (right; PID 0, n = 6 mice; PID 7, 21–49, n = 4 mice; PID 14, n = 3 mice; PID 21, P < 0.0001; PID 7, P = 0.463; PID 14, P = 0.060; PID 28, P = 0.060; PID 35, P = 0.112; PID 42, P = 0.236; PID 49, P = 0.319 versus PID 0). *P < 0.05 for comparison versus PID 0; analysis of variance (ANOVA) with post hoc test by Fisher’s least significant difference and Benjamini–Hochberg correction (all tests two-tailed). Data are mean ± s.d. HU, Hounsfield units. e, Identification of top ultrastructural features correlated with low and high pseudotime (left), summarized as overall architectural characteristics (right). Equiv., equivalent; max., maximum; min., minimum; skel., skeletonized. All scale bars, 50 μm.

Fig. 2 ∣. Distinct ECM-secreting (Csmd1⁺) and pro-resolving (Cd248⁺) fibroblast subtypes are temporally enriched during fibrosis and post-fibrotic resolution, respectively.

a, Multi-omic sequencing of baseline, fibrotic, and resolving lungs. Created with BioRender. Guo, J. https://BioRender.com/l13t441 (2025). b, UMAP of overall cell phenotypes identified from single-cell transcriptomic profiling of lungs. c, CellChat analysis of differential cell–cell signalling during fibrosis and resolution, based on ligand–receptor interactions. Fibroblasts appear to drive cell–cell signalling with ATII and ATI cells during fibrosis and resolution, respectively (box). d, In silico isolation and clustering of transcriptionally defined fibroblast subtypes. e, Timepoint-dependent enrichment of ECM-secreting (Csmd1⁺) and pro-resolving (Cd248⁺) fibroblast phenotypes at PID 14 and 35, respectively. f,g, Differentially expressed genes (f) and gene ontology (GO) terms (g) expressed by timepoint-enriched fibroblast subtypes. P values calculated based on Fisher’s exact test in enrichR. h, Localization of ECM-secreting and pro-resolving fibroblasts within fibrotic (cell-dense) and non-fibrotic (cell-sparse) regions, respectively, at key timepoints. Green arrows indicate Pdgfra⁺ cells expressing phenotypic markers of interest (Csmd1 and Cd248); blue arrows indicate Pdgfra⁻ cells expressing markers of interest. Scale bars, 20 μm. i, Temporal variation in the relative proportions of Csmd1⁺ versus Cd248⁺ cells among all Pdgfra⁺ fibroblasts (n = 5 mice; PID 21, P < 0.0001; PID 7, P = 0.068; PID 14, P = 0.001; PID 28, P = 0.183; PID 35, P = 0.634; PID 42, P = 0.288; PID 49, P = 0.525 versus PID 0). j, Kinetics of overall Csmd1⁺ versus Cd248⁺ proportions among all Lin⁻ fibroblasts (n = 5 mice). Csmd1⁺: PID 7–35, P < 0.0001; PID 42, P = 0.0005; PID 49, P = 0.102 versus PID 0. Cd248⁺: PID 21–35, P < 0.0001; PID 7, P = 0.281; PID 14, P = 0.0003; PID 42, P = 0.0003; PID 49, P = 0.977 versus PID 0. In i,j, *P < 0.05 for comparison versus PID 0; ANOVA with post hoc test by Fisher’s least significant difference and Benjamini–Hochberg correction (all tests two-tailed). Data are mean ± s.d.

Fig. 3 ∣. Fibroblasts undergo an epigenetic transition from mechano-fibrotic to reparative gene accessibility during the switch to post-fibrotic resolution.

a, UMAP of chromatin accessibility-based cell clusters via scATAC-seq, mapped to scRNA-reference. b, Cluster-to-cluster mapping identified a single scATAC-seq cluster as the major fibroblast population. c, Accessibility of characteristic fibroblast genes (Pdgfra, Col1a1, Fn1 and Fbln1) on the scATAC-seq cluster. d, Coverage plots of top differentially accessible peaks with gene annotations for fibroblasts at PID 14 (top) and 35 (bottom). e, Accessible gene ontology terms associated with fibroblasts at PID 14 (for example, mechano-fibrotic signalling; top) and 35 (for example, angiogenesis, cytokine inhibition; bottom). P values based on two-tailed Fisher’s exact test in enrichR. f, Highly enriched DNA motifs for fibroblasts at PID 14 (for example, CCAAT/enhancer-binding proteins, FOS-JUN; top) and PID 35 (for example, HES transcription factors; bottom).

Fig. 4 ∣. Fibroblast spatial organization evolves during the transition from fibrosis to post-fibrotic resolution, including differential cell–cell interactions by subtype.

a, Visium spatial sequencing, followed by integration with scRNA-seq to assign cell type probabilities. Created with BioRender. Guo, J. https://BioRender.com/l13t441 (2025). b, Example visualizations of spatially indexed cell type probabilities. c, Global cell–cell interaction network, showing cell phenotypic co-localizations enriched during fibrosis (red) versus resolution (green). Unpaired, two-tailed t-test. d, Fibroblast- and alveolar cell-mediated interaction networks. Pro-resolving fibroblasts exclusively drive reparative cell–cell interactions in these networks at PID 35, including with ATI cells. Unpaired, two-tailed t-test. e, Alveolar tissue neighbourhoods defined by co-variant transcriptional signatures for histological architecture and ATI-fibroblast subtype interactions. Pro-fibrotic neighbourhoods are associated with markers of pulmonary fibrosis and a subset of ECM-secreting fibroblast genes (for example, Fn1, Spp1, Mmp2 and Serpine2), whereas non-fibrotic neighbourhoods are associated with complement factors, immunoglobulins and specific pro-resolving fibroblast genes (for example, C3, Igkc and Pi16). Dashed boxes indicate two selected extracellular proteins with previously unknown effect on lung fibrosis, which were used for downstream functional testing.

Fig. 5 ∣. PI16 and SERPINE2, which are enriched in non-fibrotic and fibrotic alveolar tissue neighbourhoods, respectively, modulate human lung fibrosis ex vivo.

a, Processing and culture of human PCLS, which maintain the architecture, cellular makeup and biochemical content of human lungs ex vivo. Created with BioRender. Guo, J. https://BioRender.com/l13t441 (2025). b, Pseudotime trajectory of human matrix architecture, including representative polarized picrosirius red images throughout trajectory progression from root (white dot) to terminus (grey dot). Scale bars, 100 μm. c, Identification of top ultrastructural features correlated with low and high matrix pseudotime in human PCLS. Increasing pseudotime was associated with analogous architectural changes to those observed in mouse lungs. d, Differences in pseudotime between consecutively sectioned and paired PCLS (n = 6 patient specimens). SERPINE2 increased matrix pseudotime significantly in non-fibrotic PCLS only (P < 0.05), whereas PI16 decreased pseudotime significantly in fibrotic PCLS only (P < 0.05) based on a paired, two-tailed t-test, suggesting that these proteins selectively induced fibrogenic and anti-fibrotic effects, respectively (box). norm., normalized.

Fig. 6 ∣. Protein-level spatial phenotyping of human IPF.

a, CODEX for in situ spatial profiling. Scale bars, 500 μm. Created with BioRender. Guo, J. https://BioRender.com/l13t441 (2025). b, Cell phenotypes identified across all lung samples, including three overall fibroblast subpopulations with differential protein expression profiles (annotated as PDGFRα⁺, CSMD1⁺ and CD248⁺). c, Fibroblast subtype variability in IPF versus normal lungs. Data are mean ± s.d. (n = 3 healthy lung specimens; n = 6 IPF lung specimens). Unpaired, two-tailed t-test. Scale bars, 200 μm. d, Cellular niches of human fibroblast subpopulations. CSMD1⁺ and CD248⁺ fibroblasts share globally similar niches that are distinct from PDGFRα⁺ fibroblasts. e, Biomolecular neighbourhoods of fibroblasts and alveolar cells, based on extracellular proteins on the CODEX panel. CSMD1⁺ fibroblast neighbourhoods are associated with established fibrogenic factors (for example, SPP1, CTHRC1 and TGFβ1), whereas CD248⁺ fibroblast neighbourhoods exhibit decreased expression of fibrogenic or inflammatory factors.