A machine learning–based radiomics approach to predict MUC4 mutation and response to immunotherapy in cholangiocarcinoma.

Abstract

e16203 Background: The immunogenicity of MUC4 made it a tumor antigen for immunotherapeutic interventions in cholangiocarcinoma (CCA). Identifying CCA patients with elevated MUC4 expression would be crucial for MUC4-targeted immunotherapies and the prediction response to immunotherapy. Methods: In this retrospective study, 55 CCA patients who received first-line treatment with pembrolizumab in combination with gemcitabine plus cisplatin (GP) chemotherapy were enrolled. Enrolled patients underwent surgery before, and their postoperative specimens were subjected to whole exome sequencing and proteomic sequencing, and matched magnetic resonance imaging (MRI) data were collected. We employed a random forest classifier based on radiomic features to predict the status of key genes mutation including MUC4 and tested it on an internal and an external validation CCA cohort. Odds ratio of MUC4 expression for objective response to pembrolizumab plus GP was tested with logistic regression analysis. Association of MUC4 expression with progression-free survival (PFS) and overall survival (OS) was analyzed with a Cox proportional hazard regression. Results: A radiomics nomogram that incorporates radiomics signature and MUC4 mutation showed good calibration and discrimination in the training cohort (AUC = 0.762), internal validation cohort (AUC = 0.812) and external validation cohort (AUC = 0.801). The radiomics nomogram was an independent preoperative predictor of overall and recurrence-free survival. High expression of MUC4 portended significantly lower overall and recurrence-free survival than decreased MUC4 expression levels (P < 0.001). Findings from cell experiments revealed targeting MUC4 with bosutinib showed inhibitory effects on cell proliferation and promoted apoptosis. Additionally, the combination of bosutinib with a PD-1 inhibitor increased the inhibitory effect. Investigating the feasibility of translating these findings into clinical trials for patients with CCA is an important next step. Conclusions: The integration of MUC4-based immunotherapy and incorporating radiomics-based biomarkers could significantly advance personalized medicine and contribute to more effective treatment strategies for CCA patients.