A noninvasive and highly efficient epigenetic predictive model for efficacy of the GOLP regimen in patients with intrahepatic cholangiocarcinoma

Abstract

The GOLP regimen (Gemcitabine, Oxaliplatin, Lenvatinib and anti-PD1 antibody) has been a promising first-line treatment for advanced intrahepatic cholangiocarcinoma (iCCA). A noninvasive tool to predict the response to GOLP regimen is needed for clinical practice. In this study, 188 cell-free DNA samples were collected before each cycle and after the third cycle (P0 to P3) from 47 iCCA patients receiving GOLP regimen. Genome-wide 5-hydroxymethylcytosine (5hmC) profiles of samples were generated by 5hmC-Seal. Tumor response was assessed per Response Evaluation Criteria in Solid Tumors 1.1. Differential and functional analyses revealed that cell proliferation and malignancies associated pathways exhibited higher 5hmC level in poor responders at P1. Immune response related pathways presented higher 5hmC level in good responders at P2. Patients were split into training and validation cohorts by simple randomization at a ratio of 2:1 and a 5-features weighted predictive (wp-) model showing an area under the curve of 0.967 in the validation samples was constructed by machine learning. The model-derived wp-scores showed an opposite trend between good responders and poor responders from P0 to P3. In conclusion, our study identified novel epigenetic modifications and pathways across treatment process to reflect response to the GOLP regimen for iCCA patients. We developed a highly sensitive and specific 5hmC-based 5-features predictive model for GOLP treatment, holding the promise as a noninvasive tool for precision care of iCCA patients.