DNA Methylome-based Prediction of Lung Adenocarcinoma Brain Metastasis Development and Diagnostic Plasma Biomarkers

Introduction: Brain metastases (BM) occur in 10-35% of cancer patients, most commonly from lung adenocarcinoma (LUAD), and carry a poor overall survival of 10-16 months. Currently, BM cannot be reliably predicted using clinical factors or genomic alterations, leading to most being detected after they develop and grow to cause neurological deficits. Neurosurgical biopsies are required for neuropathological diagnosis. Accordingly, this work aimed to identify DNA methylation biomarkers predicting BM development to guide personalized treatment decisions as well as liquid diagnostic biomarkers to avoid surgical risks in candidates for upfront stereotactic radiotherapy.

Methods: DNA methylomes were obtained a total of 402 tumor and plasma samples from 346 LUAD and/or BM patients. Models using tissue DNA methylation signatures to predict BM development and using plasma methylomes to non-invasively identify BM from representative differential diagnoses (glioma and lymphoma) were developed in 60 and 80% discovery sets, respectively, and tested in independent validation sets. A combinatorial nomogram using the BM predictive model results together with cancer staging variables was built to provide individual patient composite BM risk values.

Results: The BM predictive methylome-based model accurately identified BM risk independent of clinical factors in validation data (Multivariable Cox: HR=8.92, 95% CI: 1.97–40.5, p=0.0046) and more significantly than clinical factors (5-year AUC: 0.81 vs. 0.65, respectively). The combinatorial nomogram showed enhanced BM prediction (Multivariable Cox: HR=17.2, 95% CI: 4.13–71.3, p< 0.0001; AUROC=0.82) in validation data. Immune pathways differentially methylated at gene promoters were observed in LUAD that develop BM. The plasma-based BM classifiers accurately distinguished BM from differential diagnoses in validation data (AUROC=0.80, 95% CI: 0.68–0.93). All models underwent further independent external validation with publicly-available data.

Conclusion : Overall, this work used DNA methylation signatures to reliably predict and non-invasively identify brain metastases, moving treatment towards a personalized molecular medicine approach that may allow for their future prevention and early treatment.