Hidden Fibroblast State at the Pancreatic Tumor Frontier Predicts Poor Survival and Points to Existing Drugs

Pancreatic ductal adenocarcinoma remains one of the most lethal cancers in the world, with a five-year survival rate that stubbornly sits below 13 percent. Part of the reason is that these tumors are wrapped in a dense, fibrotic shell of connective tissue — the desmoplastic stroma — which can account for up to 70 percent of the tumor volume. For years, scientists have catalogued the individual cell types inside this hostile terrain, but a new study published in Advanced Biotechnology argues that the real story lies not in any single cell type, but in the multicellular neighborhoods they form together. By combining single-cell RNA sequencing with high-resolution spatial transcriptomics, a team led by researchers at Sun Yat-sen University has mapped the architectural logic of pancreatic tumors and uncovered a previously unrecognized fibroblast state that sits squarely at the tumor’s invasive frontier — and whose molecular fingerprint independently predicts worse patient survival.

The research team profiled paired tumor and margin tissue specimens from five treatment-naive pancreatic cancer patients who underwent curative surgery. The margin was defined as a one-millimeter-wide band at the tumor-stroma interface, while tumor tissue was sampled at least five millimeters from that boundary. Single-cell RNA sequencing of 61,234 high-quality cells revealed twelve major lineages, with T and natural killer cells making up 43.4 percent of the cellular landscape and cancer-associated fibroblasts accounting for 10.5 percent. Spatial transcriptomics, performed on a platform with two-micrometer spots that offer subcellular resolution exceeding the widely used 10x Visium system, added 30,430 spatially barcoded measurement points, allowing the researchers to reconstruct where each cell population actually resides within intact tissue rather than inferring it from dissociated suspensions.

Within the fibroblast compartment, sub-clustering revealed four distinct subpopulations: canonical myofibroblastic CAFs, extracellular-matrix-producing CAFs, complement-secreting CAFs, and a novel population the investigators named IF-myCAF, short for invasive-frontier myofibroblastic cancer-associated fibroblast. What makes IF-myCAF unusual is its bipartite transcriptional identity. It simultaneously expresses contractile genes such as ACTA2 and TAGLN — the hallmark of myofibroblasts — and matrix genes including COL11A1, THBS2, CTHRC1, and POSTN, which are normally associated with the ECM-producing subtype. Neither canonical myCAFs nor ECM-CAFs display this dual program. Pseudotime trajectory analysis using Monocle3 placed IF-myCAF at the terminus of the myofibroblastic differentiation branch, suggesting it represents a mature, functionally specialized endpoint of fibroblast differentiation rather than a transient intermediate state.

The spatial story proved even more compelling. Gradient analysis along the tumor interior-to-invasive-front axis showed that the proportion of IF-myCAFs peaked at low malignant ductal abundance — near the tumor’s advancing edge — and progressively declined toward the tumor core in three of four spatial samples. Nearest-neighbor distance analysis confirmed that IF-myCAFs were the cell type in closest physical proximity to malignant ductal cells across independent samples. Immunofluorescence co-staining for ACTA2 and COL11A1, the defining markers of this population, independently confirmed its enrichment at the invasive front. Crucially, the finding was not a quirk of the discovery cohort: when the team analyzed eight independent pancreatic cancer single-cell datasets encompassing 159 samples and 67,959 quality-filtered fibroblasts, the IF-myCAF-corresponding population emerged as the most prevalent fibroblast subtype, accounting for 48.4 percent of all CAFs across datasets generated on multiple sequencing platforms.

With the cell states mapped, the researchers moved to a higher level of organization. Using a pipeline that combined spatial neighborhood smoothing, super-spot aggregation, and consensus clustering across three algorithms, they identified five recurrent cellular communities — spatially defined multicellular ecosystems with conserved compositions and gene expression programs. These included an acinar-parenchymal community representing residual healthy pancreatic tissue, an immune-stromal community rich in B and T cells, a fibrotic-stromal community dominated by matrix-producing cells and macrophages, a stromal-remodeling community, and — most strikingly — the invasive-desmoplastic community, or ID-CC, which localized to the tumor-stroma interface. The ID-CC showed a 5.8-fold enrichment of IF-myCAFs compared with all other communities, alongside 3.1-fold enrichment of malignant ductal cells, and was consistently detected in all four spatially profiled samples. Permutation testing confirmed that these communities represented genuine spatially contiguous units rather than random co-occurrence of cell types.

Functional analysis underscored why this particular neighborhood matters. Among all five communities, the ID-CC exhibited the highest activity in invasion and epithelial-to-mesenchymal transition programs as well as extracellular matrix remodeling, with enriched gene sets pointing to cell-substrate adhesion, TGF-beta response, and collagen fibril assembly. When the team scored the top fifty marker genes of each community in bulk tumor transcriptomes from 177 patients in The Cancer Genome Atlas pancreatic cohort, only the ID-CC signature carried prognostic weight: patients with high ID-CC scores had significantly worse overall survival, with a hazard ratio of 1.94 that remained significant after adjusting for stage and age in multivariate Cox regression. None of the other four community signatures reached prognostic significance, indicating that the adverse outcome is specifically tied to the transcriptional program of the IF-myCAF-enriched invasive niche rather than to stromal abundance in general.

Signaling analysis revealed how this niche is wired. CellChat analysis of more than 30,000 ligand-receptor interactions identified 4,009 pairs involving IF-myCAFs, establishing them as the dominant sender of extracellular-matrix-related signals within the ID-CC. As senders, IF-myCAFs were dominated by the collagen signaling axis, followed by laminin, fibronectin, and thrombospondin pathways. As receivers, they predominantly sensed collagen signals originating from ECM-CAFs, suggesting a paracrine feedforward loop that sustains fibroblast activation and matrix production. Spatially, IF-myCAFs co-localized with macrophages, consistent with inferred signaling through collagen-CD44 and fibronectin-CD44 interactions, while occupying adjacent but distinct domains relative to malignant ductal cells — a geometry consistent with juxtacrine signaling across the tumor-stroma interface.

The most translational aspect of the study came next. The team selected four IF-myCAF-derived ligand-receptor axes — THBS2-CD47, CTHRC1-FZD8, PPIA-BSG/CD147, and MIF-CD74 — based on parallel evidence channels including communication probability, in silico knockout validation, and core-gene status. A fifth candidate pair, COL11A1-ITGA1, was excluded after AlphaFold3-Multimer structure prediction revealed a large, shallow binding interface exceeding 2,000 square angstroms with no well-defined pocket for small molecules. For the four remaining complexes, the researchers predicted protein structures with AlphaFold3, refined the interfaces with LightDock docking, and virtually screened a library of 3,000 FDA-approved drugs. Five compounds achieved consensus mean binding affinities of -11.0 kcal/mol or better across all four targets: conivaptan, lonafarnib, rucaparib, desloratadine, and ketotifen. Orthogonal validation — including high-precision re-docking, comparison against an empirical null distribution of 99 randomly selected drugs, and decoy-pocket controls — showed the five candidates ranked in the top 1.1 percent of the null distribution at every target.

The authors are careful to frame these docking results as computational hypotheses rather than validated therapeutics. Notably, the five compounds also bound serum albumin’s hydrophobic cavities with comparable affinity, consistent with their lipophilic pharmacokinetics, and the team acknowledges that functional validation in experimental systems is required before any clinical relevance can be claimed. Still, the broader literature-curated analysis identified fourteen additional drug-gene interactions involving ID-CC hub genes, including agents already approved or in advanced trials — losartan and pirfenidone targeting collagen synthesis, cyclosporine A against cyclophilin A, the anti-CD47 antibody magrolimab in Phase III testing, and the anti-CD147 antibody metuximab approved in China.

Beyond the drug candidates, the study carries a conceptual punch that may finally help resolve a long-standing controversy in pancreatic cancer biology. Mouse studies have variously shown that depleting myofibroblastic fibroblasts accelerates tumor progression and immunosuppression, or that these same cells promote invasion and therapeutic resistance. The IF-myCAF findings suggest the contradiction dissolves once spatial context is considered: the same broad stromal lineage can assume opposing roles depending on the multicellular community it inhabits. An ECM-producing fibroblast embedded in benign parenchyma participates in tissue maintenance, while its counterpart recruited into the invasive-desmoplastic community sustains pro-invasive signaling. The authors caution that the community architecture itself was derived from only four spatially profiled patients and must be validated in larger independent cohorts. Even so, the work marks a decisive shift in how pancreatic tumors should be read — not as a catalogue of cell types, but as an ecosystem of recurring multicellular neighborhoods, some of which may prove to be the most actionable targets this devastating disease has yet offered.

Subject of Research: Spatial multi-omics characterization of cancer-associated fibroblast subpopulations and cellular communities in pancreatic ductal adenocarcinoma

Article Title: Spatial multi-omics uncovers an invasive-desmoplastic cellular community with prognostic and therapeutic implications in pancreatic ductal adenocarcinoma

Article References: Spatial multi-omics uncovers an invasive-desmoplastic cellular community with prognostic and therapeutic implications in pancreatic ductal adenocarcinoma. (n.d.). https://doi.org/10.1007/s44307-026-00141-8

Image Credits: AI Generated

DOI: 10.1007/s44307-026-00141-8

Keywords: pancreatic cancer, spatial transcriptomics, single-cell RNA sequencing, cancer-associated fibroblasts, tumor microenvironment, desmoplastic stroma, cellular communities, drug repurposing, AlphaFold3, prognostic biomarker, extracellular matrix, tumor invasion