Translating tumor epigenetic subtyping into methylome network-based prognostic models in early-stage NSCLC: results from the prospective MOBIT study

DNA methylation (DNAm) is a key epigenetic feature contributing to tumorigenesis with coordinated methylation changes across the genome potentially reflecting tumor biology and disease aggressiveness. There is a clinical need for identification of patients with high risk of short time to recurrence (TTR) and overall survival (OS) post-surgery in an early-stage Non-Small Cell Lung Cancer (NSCLC). The aim of this study was to translate tumor epigenetic subtyping and methylome network approach into DNAm-based prognostic models for post-surgical risk stratification in early-stage NSCLC. We have investigated 252 genome-wide methylomes derived from Next Generation Sequencing (NGS) in the Polish MOBIT prospective study cohort of 126 NSCLC patients. Epigenetic subtyping was performed in adenocarcinoma (AC) and squamous cell carcinoma (SCC) tumors methylome profiles using a hierarchical clustering and Monte Carlo simulations. We have developed an approach of elastic net-based machine learning survival modelling informed by weighted correlation network analysis (WGCNA) of cancer methylomes. Obtained models were cross-validated and subjected to further survival and biomarker selection analyses. Epitypes were characterized by tumor immune microenvironment (TIME) using RNA sequencing (NGS) based deconvolution. Five epitypes of AC and SCC with different TIMEs were detected. In SCC, epitype 1a was associated with the significantly worse OS compared to epitype 3b (p = 0.0018). We developed a model for 5-year post-surgery OS in SCC (AUC = 0.762; p = 0.007) that included sex, TNM staging, and a single epitype 1a-related and network-based DNAm biomarker independently associated with survival. In AC, using network strategy we identified DNAm biomarker significantly associated with shorter TTR in a time-frame of 5 years post-surgery recurrence (p = 0.0011). A joint recurrence AC model combining single DNAm locus with SUVmax, reached an AUC of 0.8857 (p = 0.0035), compared to AUC = 0.71 for SUVmax only (p = 0.0111). Epigenetic subtyping and methylome network analysis can be translated into prognostic models in NSCLC. Developed risk stratification models incorporating single-locus DNAm biomarkers with promising performance for 5-year mortality prediction in SCC and 5-year recurrence prediction in AC. DNAm signatures carry prognostic value beyond established clinical variables, suggesting potential utility for decision support tool for post-surgical risk stratification and improvement of individualized therapy in early-stage NSCLC.