Evaluating the utility of DNA methylation signatures in tissue and biofluids for lung adenocarcinoma brain metastasis prediction and non-invasive detection.

2030 Background: Brain metastases (BM) are common and arise in 30% of lung adenocarcinoma (LUAD) patients. Patients with LUAD that develop BM have significantly poorer outcomes, with a 10-16 month median overall survival. Unfortunately, current clinical practice for BM prediction is limited and so BM are typically detected after they develop and grow to cause neurological symptoms. Once BM are detected, currently neurosurgical tumor biopsies are performed to enable BM diagnosis via neuropathological evaluation. The aims of this study were to develop DNA methylation-based models that predict LUAD BM and non-invasively detect BM in blood to enable early diagnosis and treatment. Methods: DNA methylomes were acquired from 402 tumor tissue and plasma samples in a cohort of 346 LUAD and BM patients. Machine learning models were built using DNA methylation signatures that stratify BM risk in tissue and detect BM in plasma. Models were evaluated in independent validation datasets. A predictive nomogram was developed using the BM prediction model together with clinical factors to provide composite patient-specific scores reflecting BM risk. Results: The methylation-based BM predictor accurately stratified BM risk in a univariable Cox model using validation set data (HR = 5.65, 95%CI 1.85–17.2, p = 0.0023). Model utility was independent of the predictive value of clinical factors in a multivariable Cox model using validation set data (Table 1: HR = 8.92, 95%CI 1.97–40.5, p = 0.0046). The 5-year model accuracy was 0.81 and significantly higher than a similarly built cancer stage-based model (0.65), demonstrating utility over current practice. The combinatorial clinical-methylomic predictive nomogram had enhanced utility with an accuracy of 0.82 univariable Cox HR of 17.2 (95%CI 4.13–71.3, p < 0.0001), demonstrating comprehensive patient-specificity. The plasma-based model accurately classified BM from gliomas and lymphomas (AUROC=0.80), as typical clinical differential diagnoses, in validation set data. The models were validated further in additional external data. Conclusions: DNA methylation-based modeling of BM can accurately predict LUAD patients at risk for BM development and can non-invasively detect BM that develop. Future treatment approaches may tailor initial LUAD treatment and ongoing cancer surveillance to a patient’s BM risk, allowing for the potential to prevent and treat BM early. DNA methylation-based BM prediction is independent of clinical factors in a multivariable Cox proportional hazards model. Variable HR 95% CI p Methylome risk score 8.92 1.97–40.5 0.005 Age Years 0.96 0.92–1.02 0.177 Smoking Pack-years 0.99 0.95–1.03 0.496 EGFR Mutant vs wildtype 0.92 0.25–3.34 0.895 T T2 vs T1 1.58 0.41–6.04 0.505 T3−4 vs T1 1.49 0.28–7.98 0.642 N N1 vs N0 1.05 0.31–3.58 0.943 N2−3 vs N0 1.00 0.27–3.69 0.995 M M1 vs M0 145 12.2–1730 <0.001