Abstract A014: Protein-informed gene methylation signatures to predict treatment response and outcomes in pancreatic ductal adenocarcinoma

Abstract Pancreatic ductal adenocarcinoma (PDA) is a deadly cancer with dismal outcomes. We need new biomarkers to optimize the limited treatment combinations used in current clinical practice. Many machine-learning models have been presented based on mutational or transcriptome data claiming to predict outcomes. Still, they do not provide actionable targets (genes) that could improve treatment response. Moreover, DNA-methylation changes that influence gene/protein expression have been explored for this purpose. From the literature, we curated a 122-gene panel (January 1970 to April 2024) constructed on the proteins with preclinical (on cell lines or mouse models) or clinical (tissue-based studies) evidence suggesting their individual roles in influencing response to drugs often used in managing PDA such as gemcitabine (Gem), nab-paclitaxel, oxaliplatin, 5-fluorouracil, irinotecan, and cisplatin. We hypothesized that methylation changes in these genes are surrogates for gene/protein expression and tested their clinical significance in The Cancer Genome Atlas (TCGA) PDA population (n=184). The goal was to develop methylation signatures composed of specific cytosines followed by guanine residue (CpG) sites that aid in identifying patients at high risk of treatment failure and poor outcomes. We used elastic net multivariate analysis to calculate survival and plot Kaplan Meier plots based on our gene panel. The alpha (α) used for this script was calculated based on the lowest mean error and the minimum lambda (λ) value accordingly for each iteration (100 iterations to calculate the best values for α and λ). Our analysis yielded 28 methylation signatures that could identify high-risk PDA patients with poor overall survival (OS). The number of CpG sites in these signatures ranged from 1 to 1933. The differences in OS for these signatures were better than those available in the literature. The best signature had 18 CpG sites (17 months (m) vs. not evaluable) followed by a 53-CpG site signature (16 m vs. 67m) and a 6-CpG site signature (17m vs. 67m). A signature with the highest CpG sites (1933) was not better than those with smaller ones (16m vs. 21m). Multiple signatures had 1 CpG sites from one gene. Early growth response 1 (EGR1) coding the KCNH2 gene has featured in most signatures. One CpG site in this gene had a clinically significant prognostic value (15m vs. 30m). Other frequently featured genes were CCDC85A, EGFR, SST, and HMGA1. The main drawbacks for this study were that Gem-related genes (60) dominated the panel, and those with Gem treatment dominated the patient population. This panel should be further refined to make it more relevant to current clinical practice. We provide preliminary evidence suggesting the clinical significance of our panel and lay the foundation for future studies that would enable us to personalize prognostication and treatment selection in PDA patients. Our work can provide novel therapeutic targets that, alone or in combination with chemotherapy, can improve the outcomes for PDA. Citation Format: Ashish Manne, Harshitha Puram, Deepak Sherpally, Sravan Jeepalyam, Upender Manne. Protein-informed gene methylation signatures to predict treatment response and outcomes in pancreatic ductal adenocarcinoma abstract. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pancreatic Cancer Research; 2024 Sep 15-18; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2024;84 (17 Suppl₂): Abstract nr A014.