The platform relies on patient-derived organoids, three-dimensional clusters of tumor cells grown in a supportive gel that recapitulate key architectural and molecular features of the original cancer. Organoids have generated enormous enthusiasm in precision oncology because they allow clinicians to screen multiple drugs against a living surrogate of a patient’s tumor before committing that patient to a regimen. Yet the conventional route to building them, which begins with surgically resected or biopsied tissue, carries substantial drawbacks. Tissue acquisition is invasive, often requires a procedure that may not be clinically justified, and yields samples with low tumor cellularity. The resulting cultures can be contaminated with stromal and immune cells that dilute the tumor-specific signal, and establishment rates for pancreatic cancer organoids have historically been frustratingly low.

The Yonsei University team, led by researchers from the Division of Gastroenterology in collaboration with the Departments of Pathology and Hepatobiliary and Pancreatic Surgery at Severance Hospital, took a different route entirely. Rather than solid tissue, they started with malignant effusions, the pleural fluid that accumulates around the lungs and the ascitic fluid that pools in the abdomen of patients with advanced pancreatic ductal adenocarcinoma. These fluids are collected routinely for symptom management through minimally invasive drainage procedures, meaning that the raw material for organoid culture is essentially a clinical byproduct. Because the fluid already contains free-floating tumor cells shed from metastatic deposits, the researchers reasoned that it could serve as a rich, relatively pure starting inoculum.

Their reasoning proved correct. Fluid-derived organoids, or FDOs, established from these effusions grew faster than organoids generated from matched tissue samples, showed a higher establishment success rate, and carried markedly less non-tumor contamination. The comparison was not simply a matter of convenience. The team performed extensive quality control to demonstrate that FDOs faithfully mirror the biology of the parental tumors. Histopathological examination of hematoxylin and eosin stained sections showed that the organoids retained the glandular architecture characteristic of pancreatic ductal adenocarcinoma. Immunostaining for cytokeratin 7, an epithelial marker expressed in pancreatic ductal cells, confirmed ductal origin. Critically, mutation analysis confirmed that the organoids carried the same KRAS driver mutations as the original tumors. Since activating mutations in KRAS, most commonly at codon 12, occur in the vast majority of pancreatic cancers and anchor much of the field’s targeted drug development, this genetic concordance is essential for the model to have any translational value.

To characterize organoid morphology and drug response in fine detail without destructive processing, the researchers turned to holotomography, a label-free imaging technique that uses coherent light to reconstruct three-dimensional refractive index maps of living cells. This allowed quantitative measurement of cellular and organoid morphology and of how the structures changed in response to drug exposure, complementing conventional viability assays.

One of the most clinically significant demonstrations involved MRTX1133, a selective inhibitor of the KRAS G12D mutant protein. KRAS G12D is among the most common KRAS variants in pancreatic cancer, and MRTX1133 has emerged as a preclinical benchmark for direct KRAS targeting in this tumor type. In the study, FDOs harboring the KRAS G12D mutation showed marked sensitivity to the inhibitor, confirming that the fluid-derived platform can reproduce the drug-response behavior expected of a genetically defined tumor. The result establishes a proof of concept that FDOs can serve as a rapid and scalable test bed for emerging targeted agents, potentially shortening the path from genetic diagnosis to an individualized treatment decision.

The second major contribution of the study goes beyond the platform itself and into the molecular roots of chemotherapy failure. Gemcitabine, a nucleoside analog that has anchored pancreatic cancer chemotherapy for years, frequently stops working as tumors evolve resistance. To understand why, the team performed transcriptomic profiling, comparing gene expression in FDOs that responded to chemotherapy with expression in those that did not. Gene set enrichment and differential expression analysis converged on three genes that were consistently upregulated in the resistant cultures: CEMIP, which encodes cell migration inducing hyaluronidase 1; CALB2, which encodes calbindin 2, also known as the heart and neural crest derivatives expressed protein; and LY6D, a member of the lymphocyte antigen 6 family of glycosylphosphatidylinositol-anchored cell surface proteins.

Expression alone does not prove causation, so the researchers moved to functional validation. When they manipulated the activity of these genes in pancreatic cancer cell lines, the results were unambiguous: elevated CEMIP, CALB2, and LY6D suppressed apoptosis, the programmed cell death pathway that gemcitabine is designed to trigger, and thereby conferred resistance to the drug. CEMIP in particular has been previously implicated in hyaluronic acid metabolism and epithelial-mesenchymal transition, processes that pancreatic tumors exploit to remodel their microenvironment and escape cytotoxic stress. The new findings place all three genes squarely in the mechanistic chain linking cellular stress to survival.

The clinical implications of the three-gene signature were reinforced by outcome data. In analyses of patient cohorts, high expression of the CEMIP, CALB2, and LY6D signature correlated with worse progression-free survival and worse overall survival, indicating that the same genes that protect organoids from gemcitabine in a dish are associated with poorer outcomes in patients. This dual role, as both a mechanistic driver and a prognostic marker, is what gives the finding its translational weight. A test measuring the three-gene signature could in principle identify patients unlikely to benefit from standard chemotherapy, steering them toward alternative regimens or clinical trials of targeted and resistance-overcoming strategies. The genes themselves also represent candidate therapeutic targets, since interfering with their activity might restore sensitivity to apoptosis-inducing drugs.

The work also carries broader implications for how organoid models are built across oncology. Effusions are not unique to pancreatic cancer; malignant pleural and peritoneal effusions arise in ovarian, gastric, lung, and breast cancers, among others. A methodology that converts a routine drainage procedure into a high-fidelity drug-screening platform within days rather than weeks could be adapted widely, particularly for patients with advanced disease for whom tissue biopsy is impractical or unsafe. The scalability of the approach addresses one of the persistent bottlenecks of precision oncology: the sheer logistics of generating a personalized model quickly enough for it to influence a treatment decision made under time pressure.

The study was conducted under ethical approval from the Institutional Review Board of Yonsei University with written informed consent from all patients, and it was supported by grants from the National Research Foundation of Korea and the Korea Health Technology R&D Project through the Korea Health Industry Development Institute. The research article was published as an accepted, citable open-access version carrying a permanent digital object identifier, with the final version of record to follow.

Taken together, the findings advance pancreatic cancer research on two fronts simultaneously. They provide a minimally invasive, rapid, and genetically faithful organoid platform derived from malignant effusions, validated against a state-of-the-art KRAS targeted inhibitor. And they expose a concrete molecular mechanism of chemotherapy resistance, distilled into a three-gene signature with demonstrated prognostic power. For a disease in which treatment options remain scarce and clinical timelines are unforgiving, tools that accelerate both drug selection and biomarker discovery are welcome indeed. The next steps, which the researchers and the field more broadly will be watching closely, involve prospective validation of the gene signature in larger patient cohorts and exploration of whether targeting CEMIP, CALB2, or LY6D can resensitize resistant tumors to gemcitabine and other cytotoxic agents.

Pancreatic ductal adenocarcinoma, the most common form of pancreatic cancer, is often described as a genetically complex and highly heterogeneous disease. Heterogeneity means that cancer cells within the same tumour may carry different molecular programs, use different nutrients, interact differently with surrounding tissues and respond differently to chemotherapy or targeted drugs. Two broad transcriptional identities have attracted particular attention. Classical tumour cells generally retain features associated with more differentiated pancreatic epithelial cells and can display gene programs linked to secretory or epithelial functions. Basal-like cells, in contrast, are typically associated with a more aggressive state, altered cellular architecture, enhanced stress responses and poorer clinical outcomes. These categories are not rigid boxes; cancer cells can shift between states as their environment changes. Reproducing that flexibility outside the body has been one of the central difficulties in pancreatic cancer research.

Organoids are three-dimensional cell cultures that grow from tumour tissue and self-organise into structures that preserve some features of the original cancer. Unlike conventional two-dimensional cell lines, organoids can maintain cell-to-cell contacts, three-dimensional architecture and, in many cases, a portion of the genetic and phenotypic diversity found in a patient’s tumour. However, organoids are also shaped by the conditions in which they are grown. Oxygen is one of the most influential variables. Tumours are not uniformly supplied with oxygen: blood vessels are unevenly distributed, and rapidly dividing cells can consume oxygen faster than it can be delivered. This creates local oxygen gradients, with relatively oxygen-rich regions existing beside hypoxic zones. Such gradients can alter metabolism, activate stress pathways and influence which genes cancer cells express. The study’s dual-oxygen strategy addresses this environmental factor directly rather than treating oxygen as a fixed background condition.

According to the report, pancreatic cancer organoids exposed to distinct oxygen conditions reproduced molecular features corresponding to both basal and classical tumour states. This observation is important because it suggests that the model does not merely preserve a static genetic identity inherited from the tumour sample. Instead, it can reveal how environmental context helps shape cellular behaviour. Oxygen availability can influence the activity of transcription factors that regulate adaptation to low oxygen, including pathways controlled by hypoxia-inducible factors. It can also affect mitochondrial respiration, glycolysis, redox balance and the production of metabolites that serve as signals inside the cell. These changes may, in turn, remodel gene expression and push tumour cells toward distinct phenotypic programs. By generating organoids under two oxygen regimes, the researchers created a controllable way to examine this interaction between the cancer genome and its surroundings.

The study’s central claim was strengthened through spatial transcriptomics, which provides a type of molecular map rather than a simple list of active genes. In conventional RNA sequencing, tissue is often broken apart before analysis, meaning that information about each cell’s original position is largely lost. Spatial transcriptomics retains the physical coordinates of gene expression, allowing researchers to ask whether particular transcriptional programs are concentrated in specific regions or associated with neighbouring cell populations. This distinction matters in pancreatic cancer, where tumour cells, fibroblasts, immune cells, blood vessels and extracellular matrix form tightly interwoven microenvironments. By comparing the organoid-derived signatures with spatially resolved patterns in tumour tissue, the investigators were able to test whether the basal–classical diversity observed in culture corresponded to structures found in real cancers.

The validation does not mean that an organoid is a complete miniature pancreas or a perfect replica of a patient’s tumour. Organoid cultures generally lack the full immune system, blood circulation, nerve supply and complex stromal architecture present in living tissue. Culture media can also select for particular cell populations, while prolonged propagation may gradually favour clones that grow best under laboratory conditions. These limitations are especially relevant when studying oxygen, because real tumours experience constantly changing gradients rather than two neatly separated experimental environments. Even so, the agreement between the dual-oxygen organoid model and spatial transcriptomic patterns provides evidence that oxygen-sensitive tumour states are not simply artefacts of a dish. It indicates that the model captures at least one biologically meaningful layer of pancreatic cancer organisation.

The implications extend beyond classification. If basal and classical states can be influenced by oxygen and other microenvironmental signals, then treatment resistance may emerge not only from permanent mutations but also from reversible changes in cell identity. A tumour cell that appears relatively differentiated under one condition could adopt a more basal-like, stress-tolerant program under another. Such plasticity may help cancer survive chemotherapy, evade immune attack or repopulate a tumour after treatment. Dual-oxygen organoids could allow researchers to expose matched cancer models to drugs while monitoring whether treatment eliminates a particular state, encourages a transition into another state or leaves behind a resistant population. The system may also support studies of combination therapies designed to target both cancer-cell-intrinsic pathways and the environmental signals that maintain aggressive phenotypes.

The work arrives as cancer researchers increasingly move away from the idea that a tumour can be understood from DNA mutations alone. Genomic alterations remain fundamental, but they operate within a living ecosystem in which oxygen, nutrients, mechanical forces and neighbouring cells continuously shape tumour behaviour. By linking three-dimensional culture, controlled oxygen exposure and spatial transcriptomics, the Japanese research team offers a framework for studying that ecosystem with greater precision. The model could help explain why samples carrying similar mutations develop different clinical courses and why a therapy that works in one region of a tumour may fail in another. Further studies will need to determine how stable the induced states are, how closely they predict patient responses and whether immune or stromal components can be incorporated without losing experimental control. For now, the findings highlight a powerful message: in pancreatic cancer, where a cell grows may be almost as important as the mutations it carries.

Subject of Research: Pancreatic cancer heterogeneity, oxygen-dependent tumour states, cancer organoids and spatial transcriptomics

Article Title: Dual-oxygen pancreatic cancer organoids recapitulate basal–classical heterogeneity validated by spatial transcriptomics

Article References: Kumano, K., Nakahashi, H., Shimomura, O. et al. “Dual-oxygen pancreatic cancer organoids recapitulate basal–classical heterogeneity validated by spatial transcriptomics.” British Journal of Cancer (2026). https://doi.org/10.1038/s41416-026-03577-w

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41416-026-03577-w

Keywords: pancreatic cancer, cancer organoids, tumour heterogeneity, basal-like subtype, classical subtype, oxygen tension, hypoxia, spatial transcriptomics, tumour microenvironment, precision oncology