Six-gene signature links triptolide biology to pancreatic cancer risk stratification

Pancreatic ductal adenocarcinoma lacks effective tools for sorting patients into groups that respond differently to therapy, and a computational study published in Medical Oncology now proposes a prognostic model built around genes computationally predicted to interact with triptolide.

The researchers combined computational target predictions with transcriptomic data from The Cancer Genome Atlas and the International Cancer Genome Consortium, identifying differentially expressed triptolide-associated genes. Using Cox regression and SHAP-based feature ranking, they distilled the signal into a six-gene prognostic signature comprising AADAC, BTNL9, ADGRF1, AQP2, F5 and H2AC6. Each patient’s expression pattern of these genes yields a risk score that divides the cohorts into high-risk and low-risk groups with different survival behavior.

Predictive performance was quantified with time-dependent area under the curve values. In the training set, the signature achieved 1-, 3- and 5-year AUC values of 0.759, 0.748 and 0.741. In the validation cohort the corresponding values were 0.626, 0.677 and 0.703, a pattern the authors describe as consistent performance, though the absolute values are more modest than in the training data.

Beyond survival prediction, the study mapped what distinguishes high-risk tumors biologically. Patients in the high-risk group displayed what the authors call a classic immune-cold microenvironment, with enrichment of M0 macrophages and upregulated expression of the immune checkpoint genes BTLA, PDCD1 and TIGIT. TIDE analysis further indicated that high-risk tumors carry an immune-excluded phenotype, marked by elevated Exclusion and MDSC scores.

The in silico drug-sensitivity profiling produced a pattern with potential treatment implications. High-risk patients showed higher predicted sensitivity to paclitaxel and erlotinib, while low-risk patients appeared more responsive to axitinib and metformin. The analysis is computational and does not establish clinical efficacy for any regimen.

To connect the signature back to triptolide itself, the team performed molecular docking between the compound and the proteins encoded by all six genes. The simulations confirmed stable, high-affinity binding in every case, which the authors present as a biophysical basis for the model. They frame this as computational evidence: the constituent genes were predicted in silico to interact with triptolide, and the docking results provide a hypothesis that future laboratory work can test, not proof of a causal drug-gene relationship in tumors.

Single-cell analysis added a cellular dimension to the findings. The six genes are primarily expressed in endothelia, ductal cells and acinar cells.

Protein-level validation came from immunohistochemistry. Staining confirmed enhanced protein levels of AADAC, AQP2, ADGRF1 and H2AC6 in pancreatic ductal adenocarcinoma lesions, supporting the transcriptomic signal for four of the six signature genes at the protein level.

For now, the six-gene signature stands as a prognostic model for risk stratification in pancreatic ductal adenocarcinoma, developed computationally across two cohorts and supported by protein-level staining. The authors position the triptolide connection as a testable hypothesis for future mechanistic exploration.

Subject of Research: A six-gene prognostic model for risk stratification and treatment guidance in pancreatic ductal adenocarcinoma based on triptolide-associated genes.

Article Title: A triptolide-associated prognostic model for personalized treatment stratification in pancreatic ductal adenocarcinoma

Article References: Xie, H., Deng, M., He, H., Liang, J., & Li, J. (2026). A triptolide-associated prognostic model for personalized treatment stratification in pancreatic ductal adenocarcinoma. Medical Oncology, 43(12), Article 364. https://doi.org/10.1007/s12032-026-03430-3

Image Credits: AI Generated

DOI: 10.1007/s12032-026-03430-3

Keywords: pancreatic ductal adenocarcinoma, triptolide, prognostic model, gene signature, TCGA, ICGC, immune microenvironment, drug sensitivity, molecular docking, SHAP, Cox regression, Medical Oncology