The raw material for the new models is malignant ascites, the fluid that accumulates in the peritoneal cavity when cancer cells seed the abdominal lining. Ascites is clinically accessible through routine paracentesis, and it is rich in viable tumor cells shed from peritoneal deposits, making it an attractive biopsy surrogate for patients who are too advanced for surgery. Conventional two-dimensional cell cultures, however, strip away the architecture and cell-cell interactions that define how tumors behave, and patient-derived xenografts take months to establish in mice, far too slow to inform a first-line treatment decision. The researchers set out to bridge that gap with extrusion-based three-dimensional bioprinting, a technique that deposits living cells suspended in a soft hydrogel in precise, layered patterns.

The bioink at the heart of the system combines two photo-crosslinkable materials: gelatin methacryloyl, known as GelMA, and hyaluronic acid methacryloyl, or HAMA. GelMA is derived from gelatin and carries methacrylate groups that polymerize under light, giving the printed construct mechanical stability while its cell-adhesive motifs, inherited from native collagen, allow tumor cells to attach, migrate and organize. HAMA contributes a hydrated, hyaluronic-acid-rich matrix that mimics components of the tumor microenvironment and helps regulate stiffness and porosity. A photoinitiator called lithium phenyl-2,4,6-trimethylbenzoylphosphinate triggers crosslinking under visible light during printing, so the structure solidifies gently enough to keep cells alive. Ascites-derived cells were mixed directly into this hydrogel blend, and the resulting ink was printed into small constructs in a process the authors report took less than six hours from sample processing to finished model.

Speed matters because the clinical window for treatment decisions in metastatic gastric cancer is narrow. The printed constructs proved remarkably self-organizing: within seven to ten days in culture, the embedded malignant cells reconstituted organized glandular-like structures or solid tumor architectures, echoing the histological patterns of the parent tumors. The models remained viable for at least two weeks, a sufficient window to run full dose-response experiments. In 13 of 14 patient cases, or 92.9 percent, the team successfully established a bioprinted model, a success rate that compares favorably with organoid cultures and far exceeds the yield of xenograft approaches for this disease stage.

Establishing that the models faithfully represent each patient’s cancer was the central analytical challenge. The researchers performed immunohistochemistry on the printed constructs and found preservation of core gastric cancer markers, including the intestinal lineage transcription factor CDX2 and the carcinoembryonic antigen CEA. More therapeutically important, the expression patterns of three clinically actionable biomarkers, HER2, PD-L1 and Claudin 18.2, were generally consistent between the bioprinted models and the parental tumors. These are the targets that determine eligibility for HER2-directed antibodies, immune checkpoint inhibitors and Claudin 18.2-directed therapies, so their retention in the model is a prerequisite for any claim that the platform can guide treatment selection.

Genomic fidelity was assessed with whole-exome sequencing. The team compared somatic single-nucleotide variants detected in the bioprinted models with those in the parental ascites samples and found that key genomic features were preserved. To quantify this, they correlated the variant allele frequencies of mutations shared between each model and its parent sample, obtaining a median Pearson correlation coefficient of 0.817, with values ranging from 0.679 to 0.921. Variant allele frequency reflects the proportion of sequencing reads carrying a mutation and serves as a proxy for the cellular composition of a sample, so a high correlation indicates that the printed construct recapitulates not just which mutations are present but in what proportions. For a model built from a heterogeneous fluid containing tumor cells, immune cells and stromal elements, that degree of concordance is a strong indicator of faithful representation.

With fidelity established, the platform was put to its intended use: drug sensitivity testing. The researchers exposed each patient’s model to the standard chemotherapeutic backbone for gastric cancer, including oxaliplatin, fluorouracil and paclitaxel, and measured viability across a range of concentrations. Dose-response curves were fitted with a four-parameter logistic model, and the area under the curve, or AUC, was calculated as a summary metric of drug response, with lower values indicating greater sensitivity. The results revealed marked inter-patient heterogeneity: models from different patients responded very differently to the same agents, mirroring the unpredictable treatment responses seen in the clinic and underscoring why one-size-fits-all regimens so often fail in peritoneal disease.

To translate continuous dose-response data into a clinically interpretable verdict, the team needed a decision boundary. They derived an exploratory normalized AUC threshold of 0.675 using the Jenks natural breaks method, a statistical classification technique that identifies natural groupings within a data distribution by minimizing within-class variance and maximizing between-class variance. Applying this threshold, each model’s response to each drug was classified as sensitive or resistant. When the researchers compared these in vitro classifications with the clinical outcomes of the corresponding patients, the results were generally consistent, suggesting that the printed models captured enough of each tumor’s drug biology to predict, at least in an exploratory sense, whether a given agent would help or fail.

The authors are careful to frame the threshold as exploratory rather than validated. A single-arm study of 14 patients cannot establish predictive accuracy with the statistical rigor required for clinical deployment, and the concordance between in vitro classification and clinical outcome will need to be confirmed in larger, prospective cohorts with predefined endpoints such as progression-free survival and overall survival. Questions also remain about how the ascites microenvironment, including immune and stromal cells that may or may not survive the printing process, influences drug responses, and whether the platform can be extended to targeted agents and immunotherapies beyond cytotoxic chemotherapy. The open-access study was conducted under approval from the Medical Ethics Committee of Peking Union Medical College Hospital with written informed consent from all participants, and the authors declare no competing interests.

Even with those caveats, the work represents a meaningful advance in translational cancer modeling. It demonstrates that a clinically obtainable fluid, processed within hours into a bioprinted three-dimensional construct, can reproduce the histological, biomarker and genomic identity of a patient’s gastric cancer and yield drug-response data on a timescale compatible with real treatment decisions. If validated at scale, ascites-derived bioprinted models could give oncologists caring for patients with peritoneal metastasis something they currently lack: a personalized, functional readout of which chemotherapy is likely to work before the first dose is given. The study was supported by the Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences, the National High Level Hospital Clinical Research Funding, the Peking Union Medical College Hospital Talent Cultivation Program and the Ningbo Major Research and Development Plan Project.

Subject of Research: 3D bioprinted patient-derived ascites models for drug sensitivity testing in gastric cancer with peritoneal metastasis

Article Title: 3D bioprinted ascites-derived models for gastric cancer patients with peritoneal metastasis

Article References: Hua, Y., Du, L., Sun, H., Sun, M., Pang, M., Jiang, S., Zhang, K., Lu, Y., Mao, Y., Sun, Z., Ge, Y., Nie, M., Wang, C., Wang, X., Yao, G., Yang, D., Bai, C., Yang, H., & Zhao, L. (2026). 3D bioprinted ascites-derived models for gastric cancer patients with peritoneal metastasis. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08953-x

Image Credits: AI Generated

DOI: 10.1186/s12967-026-08953-x

Keywords: 3D bioprinting, gastric cancer, peritoneal metastasis, malignant ascites, drug sensitivity testing, bioink, GelMA, HAMA, whole-exome sequencing, patient-derived models, precision oncology, hydrogels