A pre-treatment cfDNA methylation signature linking prognosis and drug-resistance biology in pancreatic ductal adenocarcinoma

Pre-treatment risk-stratification of pancreatic ductal adenocarcinoma (PDAC) can individualize response monitoring and streamline trial triage. We refined the previously reported treatment resistance-associated, literature-curated protein-informed cell-free DNA (cfDNA) methylation signature for baseline prognostication. Notably, the panel includes genes whose protein products are implicated in resistance to commonly used PDAC therapies. The goal was to identify high-risk PDAC populations before administration of the current standard of care first-line chemotherapy (FLC) combinations, such as FOLFIRINOX (FFX) or gemcitabine and nab-paclitaxel (G-NP). The study included PDAC patients treated at The Ohio State University between 2010 and 2022 who had plasma samples available before initiation of first-line chemotherapy. Targeted enzymatic methylation sequencing was performed for cfDNA methylation profiling. In our 45-patient retrospective cohort the distribution was as follows, Stage at diagnosis (StD): 18 early stage (ES = borderline/resectable), 13 metastatic (met), and 14 locally advanced (LA); Ultimately, 18 had surgery (SR) (9 upfront surgery (UpS) and 9 had neoadjuvant therapy (NAT)) and 27 had palliative therapy (PT, 2 LA had surgery and 2 ES progressed to met); FLC in PT-group, 10 FFX, 15 GA, 2 others; Most of the patients who received NAT (7/9) or adjuvant (AT) after UpS (5/9) received FFX. A fixed 16-gene signature (cfMeth-OS16) was scored per patient (continuous z-score) and dichotomized at the study median into high-risk (HR) and low-risk (LR) groups; overall survival (OS) was modeled with Cox regression, discrimination summarized by Harrell's C-index, and survival compared by log-rank. cfMeth-OS16 separated the population significantly (HR vs. LR: 11 vs. 39 months (m); hazard ratio (HR) of 7; C-index of 0.8, p-value -<0.01). Adding StD improved the cfMeth-OS's performance (7 vs. 39m; 12; 0.87; <0.01) while FLC (FFX vs. G-NP vs. other) in any setting (PT, NAT, or AT in UpS) did not substantially improve it (10 vs. 39m, 12, 0.83; <0.01). A comprehensive model (cfMeth-OS + FLC + age + gender + RS (yes/no) + radiation received (yes/no)) achieved a higher C-index with a large, adjusted HR (7 vs. 39; 34; 0.89, <0.01). Compared to our previous signature derived from a cohort that included some post-treatment samples, cfMeth-OS showed modestly increased discrimination and extended long-survival estimates (vs. 10 vs. 33m; 8.7; 0.78). cfMeth-OS16 provides independent risk stratification for PDAC beyond FLC selection and routine clinical factors. The study's limitations include its modest sample size, single-center retrospective design, and heterogeneity in treatments and stages, which may constrain generalizability. Prospective, multi-institutional validation will be critical to confirm the robustness and potential predictive value of cfMeth-OS16