Fibroblasts That Show Tumors’ IDs May Be Shapeshifters, Not Fixed Subtypes

A class of fibroblast that lives inside gastrointestinal tumors and carries molecular identity badges has puzzled cancer biologists for years: in pancreatic cancer these cells appear to help tumors suppress the immune system, while in gastric cancer their abundance tracks with earlier-stage disease, longer survival, and better responses to immunotherapy. A new perspective article published in Frontiers in Oncology argues that this contradiction may be an artifact of how the cells are defined, and proposes that antigen-presenting cancer-associated fibroblasts, or apCAFs, are not a stable cell subtype at all but a reversible, context-dependent state whose immune function depends on properties that most studies never measure.

CAF plasticity itself is well established, they note; what they add is a way of evaluating apCAFs along three independent attributes—cellular origin, MHC class II-restricted antigen presentation, and co-stimulatory capacity—rather than a single label, and a set of predictions with explicit conditions under which the framework would fail. Crucially, they acknowledge that their central model has not been validated in apCAFs and is inferred from fibroblast studies that never examined these cells.

apCAFs were first identified through single-cell RNA sequencing of mouse and human pancreatic ductal adenocarcinoma by Elyada and colleagues. These cells express MHC class II molecules and CD74, the molecular machinery of antigen presentation, but lack the canonical co-stimulatory molecules CD40, CD80, and CD86 that professional antigen-presenting cells use to fully activate T cells. Functional studies showed that apCAFs activate CD4+ T cells in an antigen-dependent manner and promote the differentiation of regulatory T cells, the immune cells that dampen antitumor responses. In mouse pancreatic cancer, a proportion of apCAFs arise from mesothelial cells through a transition regulated by the signaling molecules IL-1 and TGF-β; an antibody against the mesothelial marker MSLN inhibited this process, reduced regulatory T-cell induction, and decreased tumor burden in an orthotopic model.

Gastric cancer tells a strikingly different story. Song and colleagues identified an apCAF population there marked by high CD74 and HLA class II expression, and found that greater infiltration of these cells was associated with earlier-stage disease, longer overall and recurrence-free survival, and improved responses to anti-PD-1-based treatment. The authors of the perspective argue that this apparent paradox partly reflects asymmetric evidence rather than proven biological difference. In pancreatic cancer, regulatory T-cell induction has been confirmed in antigen-specific co-culture, and anti-MSLN therapy has been shown to reduce both Treg induction and in vivo tumor burden. In gastric cancer, by contrast, apCAF abundance has only been correlated with T-cell infiltration, staging, survival, and checkpoint blockade response, without any measurement of co-stimulatory molecules or interference experiments. A demonstrated function and a statistical association, they caution, describe different levels of evidence and cannot by themselves establish a biological contrast.

The perspective draws on work outside cancer to explain how an antigen-presenting fibroblast could be tolerogenic rather than stimulatory. Dubrot and colleagues showed that lymph node stromal cells, including fibroblastic reticular cells, acquire peptide–MHC II complexes from dendritic cells while expressing low levels of co-stimulatory molecules; their interaction with antigen-specific CD4+ T cells impaired T-cell survival, proliferation, and recall responses, producing antigen-specific tolerance. Similar effects have been reported for other nonprofessional MHC class II-positive cells, including certain epithelial, endothelial, and stromal cells that fail to efficiently prime naïve CD4+ T cells and can instead promote peripheral tolerance. By this logic, the opposing effects of apCAFs in pancreas and stomach might reflect differences in co-stimulatory capacity and local accessory signals rather than distinct fibroblast lineages.

Technical artifacts may also inflate the paradox. Droplet-based single-cell RNA sequencing is vulnerable to ambient RNA contamination and to doublets—two cells captured in one droplet—that generate hybrid expression profiles and distort cell-type assignment. Bioinformatic tools such as SoupX and DoubletFinder can reduce these artifacts. Fibroblasts may also acquire ready-made MHC class II molecules on their surfaces from professional antigen-presenting cells, so protein detection alone does not prove endogenous expression. The authors stress that these caveats do not challenge the existence of apCAFs, which have been validated in pancreatic cancer by protein-level analyses, antigen-presentation assays, and functional T-cell experiments. But they do mean that MHC class II or CD74 signatures alone cannot establish lineage, function, or immune outcome.

From these observations the authors build their speculative two-layer model. A permissive epigenetic layer, defined by chromatin accessibility at the CIITA locus and at co-stimulatory loci such as CD80, CD86, and CD40, would determine which functional programs a given apCAF can engage at all. An instructive signaling layer, driven by cues such as interferon-gamma, IL-1, and TGF-β, would then select specific phenotypes within that accessible range. Under this model, stable silencing of co-stimulatory loci could uncouple MHC class II induction from effective helper T-cell activation, favoring immune tolerance, while accessible loci might confer susceptibility to functional reprogramming toward a state that supports CD4+ T-cell help and CD8+ T-cell infiltration. The authors are explicit that chromatin accessibility at these loci has never been analyzed in apCAFs and that no reprogramming experiments in these cells have been reported; if accessibility proves indistinguishable between tolerogenic and immunostimulatory apCAFs, the model is simply wrong.

The framework also addresses what therapy-induced shifts in fibroblast composition can and cannot show. TGF-β blockade reduces myofibroblastic CAFs and is accompanied by the emergence of interferon-licensed CAFs, increased CD8+ T-cell infiltration, and reduced immune exclusion. In high-grade serous ovarian cancer, platinum- and taxane-based chemotherapy decreases ANTXR1-positive ECM-myCAFs and relatively enriches ANTXR1-negative inflammatory CAFs, while residual ECM-myCAFs continue to suppress CD8+ T-cell cytotoxicity through a YAP1-dependent mechanism. But cross-sectional single-cell datasets cannot distinguish selective survival of pre-existing cell states from genuine conversion of individual cells, and the authors argue that only lineage tracing, clonal barcoding, and paired pre- and post-treatment single-cell transcriptomic or ATAC-sequencing data applied directly to apCAFs could settle the question. Until then, they write, therapeutic reprogramming of apCAFs should be considered a testable hypothesis supported by indirect evidence.

The perspective lays out three falsifiable predictions. First, immune-activating and tolerogenic apCAF populations should differ more in chromatin accessibility at CD80, CD86, CD40, and CIITA than in MHC class II expression itself; if accessibility is indistinguishable between gastric and pancreatic apCAFs, the model fails. Second, accessibility at these loci should increase transiently after cytotoxic or stroma-directed therapy, before subtype proportions restabilize; if accessibility merely tracks cell proportions with no lead interval, there is no therapeutic window to exploit. Third, pairing an epigenetic modifier with a differentiation cue during that window should increase antigen-specific CD4 help and CD8 infiltration more than either intervention alone or the same combination given at baseline; if timing is irrelevant, the window concept adds nothing. The authors also flag two risks: broad epigenetic agents act on tumor cells and lymphocytes as well as fibroblasts, so any benefit would require fibroblast-restricted genetic controls before being attributed to stromal reprogramming, and enhancing presentation of self-derived antigen by stromal cells carries a plausible autoimmune cost, particularly in the gut, where tolerance to luminal antigens is actively maintained.

Gastrointestinal cancers, the authors argue, are the practical testbed because longitudinal sampling is feasible there. Serial endoscopic biopsies before and during neoadjuvant treatment, paired multi-omic profiling before and after chemotherapy or immunochemotherapy in gastric cancer, and patient-derived tumor organoids co-cultured with matched fibroblasts in pancreatic and colorectal cancer all provide platforms for capturing transient states and testing candidate interventions. Noninvasive monitoring cannot currently define apCAF status: FAP-targeted PET imaging reflects overall stromal matrix activity, and since apCAFs are heterogeneous within the FAP-positive compartment and usually FAP-low, such imaging could at best guide sampling timing, and no circulating apCAF biomarkers have been verified. If the framework holds up, the authors suggest, CAF-directed therapy could move beyond broad depletion—which has produced inconsistent benefits and, in some pancreatic models, increased tumor aggressiveness—toward temporally timed, state-informed functional reprogramming that preserves fibroblasts while shifting them out of tolerogenic configurations. For now, that remains a hypothesis with defined fault conditions, not a demonstrated cell-fate transition.

Subject of Research: Antigen-presenting cancer-associated fibroblasts and their divergent immune roles in gastrointestinal cancers

Article Title: Subtype or state? Rethinking the apCAF paradox and fibroblast plasticity in gastrointestinal cancers

Article References: Yu, Z., Hu, D., & An, Z. (2026). Subtype or state? Rethinking the apCAF paradox and fibroblast plasticity in gastrointestinal cancers. Frontiers in Oncology, 16, Article 2000409. https://doi.org/10.3389/fonc.2026.2000409

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.2000409

Keywords: cancer-associated fibroblasts, apCAFs, tumor microenvironment, MHC class II, co-stimulation, pancreatic cancer, gastric cancer, epigenetic reprogramming, single-cell sequencing, immunotherapy, fibroblast plasticity, regulatory T cells