A fragmentomic tumor score algorithm for enhanced minimal residual disease detection and prognostic prediction in bile duct cancer.

Abstract

e16287 Background: Bile duct cancer (BDC) remains a challenging malignancy with high recurrence rates after surgery, thus minimal residual disease (MRD) detection is critical for predicting outcomes and guiding postoperative management. Methods: This study prospectively enrolled 79 patients with BDC (intrahepatic, n = 19; perihilar, n = 15; distal, n = 42; mixed, n = 3) undergoing curative-intent surgeries (30 recurrence and 49 non-recurrence), of whom pre-surgery, 1-, 3-, and 6-month post surgery plasma cell-free DNA (cfDNA), surgical tumor tissue, and adjacent normal tissue samples were collected. Plasma cfDNA samples of 54 non-cancer individuals were additionally collected. We developed a comprehensive fragmentomic tumor score (FTS) algorithm for cfDNA MRD assessment, which integrated a copy number variant (CNV) score through whole-genome sequencing performed on non-BDC and pre-surgery BDC cfDNA, a methylation (Methy) score based on applying fragmentomics-based methylation analysis on BDC-specific differentially methylated regions, and a fragmentomic transcription start site (TSS) score representing the TSS coverage of differentially expressed genes identified by RNA-sequencing applied to tumor and adjacent tissues. FTS performance in MRD detection was evaluated, and its utility in impending progression prediction was investigated in post-surgery cfDNA samples. Results: The CNV, Methy, and TSS scores achieved an area under the curve (AUC) of 0.904, 0.779, and 0.891, respectively, when distinguishing cancer from non-cancer samples. The combined FTS demonstrated superior performance, achieving an AUC of 0.950 with 77.8% sensitivity at 98.0% specificity. Furthermore, patients with MRD-positive 1-month post-surgery cfDNA samples, whose FTS were over the threshold fixed at 98.0% specificity, had inferior disease-free survival [DFS; median, 7.89 months vs. not reached (NR); hazard ratio (HR), 7.64; 95% confidence interval (CI), 2.85–20.45; p < 0.001] and overall survival (OS; median, 13.3 months vs. NR; HR: 9.62; 95% CI, 2.21–41.79; p = 0.003) than those with MRD-negative samples. Similar trends towards worse DFS in MRD-positive patients were observed at 3- (median, 9.99 months vs. NR; HR: 7.59; 95% CI, 1.54–37.37; p = 0.013), and 6-month (median, 11.5 months vs. NR; HR: 12.76; 95% CI, 1.25–130.00; p = 0.032) after surgery. Conclusions: A novel MRD assessment algorithm was established by integrating fragmentomic CNV, methylation, and transcription features, offering a promising tool for monitoring disease progression and prognostic prediction.