CMET-32. DNA METHYLATION ALTERATIONS IN LUNG ADENOCARCINOMAS THAT DEVELOP BRAIN METASTASES

Abstract

Abstract INTRODUCTION The development of brain metastases from primary cancer profoundly impacts patient prognosis. Metastases are the most common adult brain tumor with up to one quarter of lung cancers developing metastases and median overall survival after metastasis being one year. Clinical factors do not reliably predict brain metastasis development and over 90% are identified after symptoms develop. DNA methylation signatures predict outcomes in other cancers and so identifying signatures that predict metastasis development may allow for treatment strategies that prevent development in high risk patients. METHODS Whole genome DNA-methylation profiling was undertaken on N=124 lung adenocarcinoma patients after bisulfite conversion of DNA from formalin-fixed paraffin-embedded tissue. In a randomly selected 70% training cohort, the most differentially methylated CpG sites between patients developing and not developing brain metastases were identified with p< 0.05. A generalized boosted regression model built on these selected features output brain metastasis risk scores for patients in the independent 30% testing cohort. RESULTS Brain metastases developed in 49/124 (39.5%) of patients and 2.3K CpG sites were significantly differentially methylated between patients developing and not developing metastases. Methylation-based brain metastasis risk scores predicted time to brain metastasis in a univariate cox regression model (HR=3.2, 95% CI 1.1–9.4, p=0.03). A corresponding area under the receiver operating characteristic curve at 52 months was 0.64. A multivariate cox analysis including tumor size and nodal status, representing the non-metastatic components of cancer stage, identified methylation score as the only independent predictor of brain metastasis (HR=4.3, 95%CI 1.1–17, p=0.038). CONCLUSIONS DNA methylation signatures in lung adenocarcinoma predict brain metastasis development independent of stage components, which classically predict patient outcome in cancer. Future work developing a comprehensive nomogram utilizing methylation scores together with other clinical factors to determine patient specific risk values may aid in treatment decisions and patient prognosis counselling.