A fibroblast-centric network drives cold fibrosis in the tumor microenvironment of lung squamous cell carcinoma

Abstract

The tumor microenvironment (TME) of chronic inflammation-associated cancers (CIACs) is shaped by cycles of injury and maladaptive repair, yet the principles organizing fibrotic stroma in these tumors remain unclear. Here, we applied the concept of hot versus cold fibrosis, originally credentialed in non-cancerous fibrosis of heart and kidney, to lung squamous cell carcinoma (LUSC), a prototypical CIAC. Single-cell transcriptomics of matched tumor and adjacent-normal tissue from 16 treatment-naive LUSC patients identified a cold fibrotic architecture in the LUSC TME: cancer-associated fibroblasts (CAFs) expanded and adopted myofibroblast and stress-response states, while macrophages were depleted. This macrophage-poor, CAF-rich stroma was maintained by CAF autocrine growth factor loops, including TIMP1, INHBA, TGFB1, and GMFB. In parallel, the immune compartment exhibited a hot tumor phenotype with abundant T and B cells, forming spatially distinct but molecularly engaged networks with CAFs. CAF gene programs typifying cold fibrosis in LUSC were conserved in other CIACs, including esophageal and gastric adenocarcinomas. These results redefine desmoplastic regions of tumors through the lens of a non-cancer fibrosis model, demonstrating that conserved stromal circuits constitute therapeutic vulnerabilities in CIACs.

Figure 1.. An atlas of the LUSC TME reveals a cold fibrosis-hot tumor phenotype.

A. Representative hematoxylin and eosin (H&E) stained images of matched adjacent-normal lung tissue (left) and lung squamous cell carcinoma (LUSC) tumor tissue (right) from the same patient. Scale bar, 0.5mm. B. Uniform manifold approximation and projection (UMAP) embedding of 197,232 cells and delineation of 13 cell types including Macrophage, T cells, Epithelial Cells (including cancer epithelial cells), Natural Killer (NK) cells, Mast cells, Fibroblasts, Endothelial Cells, B Cells, Dendritic Cells (DC), Granulocytes, Plasmacytoid Dendritic Cells (pDC), Smooth Muscle cells, and NK/Innate Lymphoid Cells (ILC). Individual cells are colored by cell type. C. Dot plot showing the expression of marker genes across major cell types in the LUSC tumor and adjacent-normal lung microenvironment. Each row represents a cell type, and each column corresponds to a selected marker gene. The size of each dot indicates the percentage of cells within a given cell type expressing the gene (Pct. Exp), while the color scale represents the average expression level (Avg. Exp), from low (gray) to high (red). D.UMAP visualization of single-cell transcriptomes from adjacent-normal lung tissue (top; n = 106,676 cells) and LUSC tumor tissue (bottom; n = 90,556 cells). Individual cells are colored by cell type. E. Comparison of cell type abundances between adjacent-normal and LUSC tumor tissues on a per-patient basis (n = 14 patients with paired tumor and adjacent-normal samples). Each plot shows the percentage of a given cell type out of total cells in matched adjacent-normal and tumor samples, with paired lines representing individual patients. Statistical significance was calculated using a paired two-tailed Student’s t-test. F. Quantification of differentially expressed genes across major cell types between tumor (T) and adjacent-normal (N) lung tissue. Each point represents a cell type, plotted by the number of genes upregulated (y-axis) versus downregulated (x-axis) in the tumor relative to adjacent-normal tissue, both on a log₁₀ scale.

Figure 2.. LUSC, associated epithelial cells acquire distinct archetypes characterized by enhanced translation, inflammation, and immune interaction.

A. UMAP embedding of all cells, with epithelial cells highlighted in red and all other cell types shown in gray. B. UMAP plot of epithelial cells only, following subclustering using Seurat (19 principal components, resolution = 0.1), identifying 13 transcriptionally distinct epithelial subpopulations (clusters 0-12). C. Same UMAP as in (B), now split by condition, showing epithelial cells derived from adjacent-normal (left, n = 16,835) and tumor (right, n = 16,087) tissues. D. Pareto Task Inference (ParTI) was used to characterize the continuum of gene expression programs within epithelial cells across tumor and adjacent-normal tissues. Cells were projected into gene expression space and approximated by a tetrahedron (p < 10⁻⁵), indicating four transcriptional archetypes. Cells are projected on the first three principal components. E. Bar plot showing the top enriched Gene Ontology (GO) terms for each epithelial archetype (1-4) identified by ParTI analysis. GO terms are ranked by statistical significance (−log₁₀ p-value). F. Examples of top marker genes associated with each epithelial archetype (Table S4). The expression of TP63 (archetype 1), MUC5B (archetype 2), SFTPA2 (archetype 3), and TEKT1 (archetype 4) is shown. Each gene’s expression is overlaid onto the tetrahedral projection of epithelial cells from ParTI analysis, where cells are colored by normalized mRNA expression (blue: low, red: high). G. Projection of adjacent-normal (left, blue) and tumor-derived (right, red) epithelial cells onto the tetrahedral archetype structure identified by ParTI. Each dot represents a single cell, positioned in gene expression space relative to the four identified archetypes. H. Enrichment of tumor-derived epithelial cells near each archetype, plotted as a function of Euclidean distance from the archetype vertex. I. Gene signature enrichment scores for each of the four epithelial archetypes (1-4) were calculated and projected onto a UMAP of epithelial cells. Cells are colored by signature score (purple to dark blue; linear scale).

Figure 3.. LUSC fibroblasts acquire two distinct archetypes are acquired, associated with a myofibroblast phenotype and stress response.

A. UMAP embedding of all fibroblasts, with six transcriptionally distinct subpopulations identified by Seurat subclustering (24 principal components, resolution = 0.2; clusters 0–5). B. Same UMAP as in (A), now colored by tissue of origin, showing fibroblasts derived from adjacent-normal lung (blue, n = 1,290) and LUSC tumor tissue (red, n = 3,939). C. Pareto Task Inference (ParTI) was used to characterize the continuum of gene expression programs within fibroblasts from adjacent-normal and tumor samples. Cells were projected into gene expression space and approximated by a tetrahedron (p = 0.03), indicative of four transcriptional archetypes. Cells are projected onto the first three principal components. D. Examples of top marker genes associated with each fibroblast archetype (Table S6). The expression of ADH1B (archetype 1), DNAJA1 (archetype 2), CTHRC1 (archetype 3), and RPS12 (archetype 4) is shown. Each gene’s expression is overlaid onto the tetrahedral projection of fibroblasts from ParTI analysis, where cells are colored by normalized mRNA expression (blue: low, red: high). E. Bar plot showing the top enriched Gene Ontology (GO) terms for each fibroblast archetype (1-4) identified by ParTI analysis. GO terms are ranked by statistical significance (−log₁₀ p-value). F. Enrichment of tumor-derived fibroblasts near each archetype, plotted as a function of Euclidean distance from the archetype vertex. G. left- UMAP embedding of single-cell RNA-seq data from esophageal adenocarcinoma (EAC) samples, with cells colored by tissue origin (normal: blue; tumor: orange). Fibroblasts are outlined. Right-Violin plots show gene signature scores for each fibroblast archetype (1-4) in normal versus tumor-derived fibroblasts, per patient. Each dot represents a biological replicate. Each plot displays the paired Wilcoxon p-value and the corresponding fold-change (tumor vs. normal) for the respective fibroblast archetype. H. Left, UMAP embedding of single-cell RNA-seq data from gastric adenocarcinoma (GAC), with cells colored by tissue origin (normal: blue; tumor: orange). The fibroblast cluster is outlined. Right, Violin plots show gene signature scores for each fibroblast archetype (1-4) in normal versus tumor-derived fibroblasts, per patient. Each dot represents a biological replicate. Each plot displays the paired Wilcoxon p-value and the corresponding fold-change (tumor vs. normal) for the respective fibroblast archetype.

Figure 4.. Network of growth factor interactions in tumor and adjacent-normal LUSC.

A. Network of inferred growth factor-mediated interactions in adjacent-normal (left, blue) and tumor (right, red) tissue, as predicted by NicheNet. Nodes represent major cell types and are sized according to their relative abundance in the dataset (% of total cells). Directed edges represent ligand-receptor-mediated communication between sender and receiver cell types. Edge thickness reflects the relative contribution of each interaction to the total signaling within the network (% of total predicted communication). B. Fraction of outgoing growth factor interactions contributed by each cell type in adjacent-normal (left, blue) and LUSC tumor (right, red) samples. Bars represent the proportion of total outgoing signals attributed to each cell type, sorted in descending order. The overlaid line shows the cumulative contribution across cell types, with horizontal dashed lines marking the 90% cumulative thresholds. C. Fold-change network visualization comparing growth factor-mediated interactions in LUSC tumor versus adjacent-normal tissue. Nodes represent cell types and are scaled by the fold change in relative abundance (tumor/adjacent-normal). Directed edges represent ligand-receptor interactions, with edge thickness proportional to the fold change in interaction strength (tumor/adjacent-normal). Only interactions contributing to the top 90% of total network communication are shown. D. Scatter plot showing the fold change in incoming (y-axis) versus outgoing (x-axis) signaling for each cell type. E. Inferred ligand contributions to cell-cell interactions within the fold-change network of LUSC tumor versus adjacent-normal tissue. Arrows above each column represent directional communication between the sender (top node) and receiver (bottom node) cell types, including epithelial cells (E), fibroblasts (F), T cells (T), and macrophages (M). Each dot indicates a predicted ligand mediating the interaction, as inferred by NicheNet. F. Expression of TIMP1, INHBA, TGFB1, and GMFB in fibroblasts from tumor (red) and adjacent-normal (blue) tissues. G. Expression of NENF, HDGF, and BMP7 in epithelial cells from tumor and adjacent-normal tissues. These ligands are predicted to mediate epithelial-to-fibroblast signaling.