Tiny Cell Structures Called Migrasomes May Predict Gastric Cancer Outcomes

Gastric cancer remains one of the world’s most lethal malignancies, and clinicians have long struggled to explain why patients with seemingly similar tumors can follow dramatically different courses. A new study published in Cancer Cell International by researchers at Xiangya Hospital of Central South University offers a striking new angle on that problem, one that hinges on a recently discovered cellular structure known as the migrasome. By weaving together single-cell genomics, computed tomography radiomics, spatial transcriptomics, machine learning, and laboratory validation, the team reports that fibroblast-associated migrasome programs in the tumor microenvironment are tightly linked to gastric cancer progression and may provide a powerful new framework for predicting patient risk. The work centers on a tetraspanin protein called TSPAN4, which the researchers show drives cancer cell proliferation through the PI3K/AKT signaling pathway, positioning it as a potential drug target.

Migrasomes are membrane-bound vesicles that form behind cells as they migrate, leaving behind a trail of signaling packages that can influence neighboring cells. Although first characterized only in the last decade, these organelles have increasingly been implicated in cancer biology, where migrating tumor and stromal cells may use them to reshape their surroundings. What has remained murky, however, is precisely which cells in a tumor produce migrasomes, how they communicate with immune and structural cells, and whether migrasome activity carries clinically meaningful information about prognosis. The new study set out to answer those questions for gastric cancer, an aggressive disease characterized by dense and complex interactions between malignant cells and the surrounding tumor microenvironment.

The investigators began with single-cell RNA sequencing of fourteen gastric cancer specimens using the 10x Genomics platform. Unsupervised clustering resolved nine transcriptionally distinct cell clusters, which the team annotated into eight major cellular compartments spanning malignant epithelial cells, immune populations, and stromal cells. To quantify migrasome activity, the researchers scored each cell against a curated set of migrasome-associated genes. Fibroblasts stood out immediately: they displayed the highest migrasome scores of any compartment and also achieved the highest cell-type identification accuracy in the Augur computational framework, with an area under the curve of 0.805. This dual prominence suggested that fibroblasts are not merely present in gastric tumors but are the primary migrasome-producing population whose behavior changes across the disease landscape.

Digging deeper, the team identified eleven fibroblast-specific migrasome-associated genes, abbreviated FSMAGs, that defined this signature. Pseudotime trajectory analysis, which orders cells along a computational timeline of differentiation, revealed five distinct fibroblast differentiation states within the tumors. As fibroblasts progressed along this trajectory, the expression of the chemokine CXCL12 and the tetraspanin TSPAN4 rose steadily, while the related gene TSPAN9 remained comparatively stable across all states. The divergence of TSPAN4 from its family member hinted that it plays a specialized, state-dependent role rather than serving as a generic structural component, a hypothesis the researchers went on to test functionally.

Because tumors are communication hubs as much as collections of cells, the team used CellChat, a computational tool that infers ligand-receptor signaling between cell populations, to map how high-migrasome-score fibroblasts interact with their neighbors. The analysis predicted substantially stronger communication between these fibroblasts and both macrophages and dendritic cells than between fibroblasts and other populations. Among the inferred signaling routes, the macrophage migration inhibitory factor pathway, in which MIF engages the receptors CD74 and CXCR4, emerged as a leading candidate mechanism for fibroblast-to-dendritic-cell communication. This finding suggests that migrasome-active fibroblasts may actively sculpt the immune landscape of gastric tumors, potentially dampening or redirecting antitumor immune responses in ways that influence clinical outcomes.

The study’s most clinically ambitious component was the construction of a prognostic model that bridges the microscopic and macroscopic worlds. Using 101 machine learning algorithm combinations applied to bulk RNA sequencing data from 385 patients in The Cancer Genome Atlas training cohort, the researchers built what they call the Radiomics-Derived Migrasome-associated Gene Signature, or Rad_MGsig. The model integrates five migrasome-associated genes with nine radiomic features extracted from computed tomography scans of forty-six patients. Radiomic features capture quantitative patterns in medical images, such as texture and heterogeneity, that the human eye cannot reliably assess. By fusing imaging data with gene-expression programs, Rad_MGsig aims to translate what happens at the cellular level into information available from routine clinical imaging.

The performance figures reported for the signature are remarkable. In the training cohort, Rad_MGsig achieved area under the curve values of 0.96, 0.90, and 0.88 for predicting one-year, three-year, and five-year survival, respectively. In an independent validation cohort of 357 patients from the GSE84433 dataset, the values were 0.97, 0.90, and 0.90. Accuracy that remains this high across multiple time points and an external cohort is uncommon for prognostic models in oncology, and it suggests that migrasome biology captures a stable and consequential feature of gastric tumor behavior. If prospectively validated, the approach could help clinicians stratify patients for intensified treatment, closer surveillance, or, conversely, avoid unnecessary toxicity in those at lower risk.

To connect the computational findings to molecular mechanism, the researchers turned to laboratory experiments and structural modeling. Molecular docking simulations predicted that the experimental drug Uprosertib, an AKT inhibitor, binds TSPAN4 with the strongest predicted affinity among screened compounds, at negative 9.0 kilocalories per mole. Consistent with that prediction, Uprosertib inhibited the growth of gastric cancer cells in a dose-dependent manner, with half-maximal inhibitory concentrations of 0.368 micromolar in HGC-27 cells and 0.460 micromolar in AGS cells. Importantly, the drug attenuated the proliferative surge caused by forced TSPAN4 overexpression in AGS cells, indicating that the compound’s effect converges on the TSPAN4 axis. Knockdown and overexpression experiments supported the conclusion that TSPAN4 acts within tumor cells to activate the PI3K/AKT pathway and promote proliferation.

Spatial context provided the final piece of the puzzle. Using multiplex immunofluorescence on tumor tissue, the researchers visualized where TSPAN4-positive signals reside relative to fibroblast markers. They found TSPAN4-positive signals in close proximity to regions stained for Vimentin, a canonical fibroblast marker, specifically at the tumor-stroma interface, the dynamic border where malignant cells meet the supportive stromal tissue. This anatomical colocalization is consistent with a model in which fibroblast-derived migrasome activity and tumor-cell TSPAN4 signaling form a coupled program at the invasive front, where migration, communication, and proliferation converge. Such spatial coupling may explain why migrasome-associated gene signatures carry prognostic weight: they index an active, location-specific dialogue between tumor and stroma rather than an incidental byproduct of cell turnover.

The study, funded by the National Natural Science Foundation of China and the Natural Science Foundation of Hunan Province, was conducted under ethics approvals from Xiangya Hospital and Central South University, with written informed consent from all tissue donors and adherence to ARRIVE guidelines for the xenograft experiments. Its implications extend in several directions. For biomarker developers, Rad_MGsig offers a template for combining single-cell discovery with radiomics to create clinically deployable risk models. For drug developers, TSPAN4’s pro-proliferative role and the predicted binding of Uprosertib suggest a rational target, although the authors note that docking predictions and cell-line results will require substantial follow-up before clinical translation. For cancer biologists, the work adds migrasomes to the growing list of vesicle-mediated communication channels, alongside exosomes and other extracellular vesicles, that shape tumor progression. As migrasome research matures, the gastric cancer data suggest that these trailblazing organelles are more than cellular curiosities; they may be central actors in the tumor microenvironment’s conversation with cancer itself, and their fingerprints in genes and scans could ultimately guide therapy for one of medicine’s most stubborn diseases.

Subject of Research: Fibroblast-associated migrasome programs and TSPAN4 signaling in gastric cancer progression and prognosis

Article Title: Fibroblast-associated migrasome programs define TSPAN4-driven progression and clinical risk stratification in gastric cancer

Article References: Zhang, K., Ren, F., Ge, J., Huang, C., Kang, K., & Wu, Z. (2026). Fibroblast-associated migrasome programs define TSPAN4-driven progression and clinical risk stratification in gastric cancer. Cancer Cell International. https://doi.org/10.1186/s12935-026-04463-4

Image Credits: AI Generated

DOI: 10.1186/s12935-026-04463-4

Keywords: gastric cancer, migrasomes, TSPAN4, fibroblasts, PI3K/AKT pathway, radiomics, single-cell RNA sequencing, spatial transcriptomics, prognostic signature, tumor microenvironment, Uprosertib, machine learning