1143 - A Blood-based DNA Methylation Signature for Non-invasive Diagnosis and Prediction of Clinical Outcomes in Patients with Glioblastoma Enrolled on the Phase II Randomized VERTU Trial
Presenter(s)
E. J. Vaios1, C. Wimberly2, H. W. Sim3, E. Barnes3, E. S. Koh4, S. Yip3, M. Hall3, G. Herrgott5, H. Noushmehr6, K. Ayasoufi2, K. Batich2, D. M. Ashley2, S. R. Floyd1, H. A. Shih7, E. P. Sulman1, J. Simes3, Z. J. Reitman1, K. Walsh8, B. Hu9, and M. Khasraw2; 1Department of Radiation Oncology, Duke University School of Medicine, Durham, NC, 2Department of Neurosurgery, Duke University School of Medicine, Durham, NC, 3NHMRC Clinical Trials Centre, University of Sydney, Sydney, NSW, Australia, 4University of New South Wales, Sydney, Australia, 5Department of Neurosurgery, Omics Laboratory, Hermelin Brain Tumor Center, Detroit, MI, 6Henry Ford Health System, Detroit, MI, 7Department of Radiation Oncology, Mass General Brigham Cancer Institute, Boston, MA, 8National Institute of Environmental Health Sciences, Durham, NC, 9Department of Biostatistics and Bioinformatics, Duke University School of Medicine, Durham, NC
Purpose/Objective(s):
A barrier to understanding glioblastoma tumor biology and developing more effective therapies is the dependence on surgically-derived tissue for diagnosis and assessment of treatment response. Liquid biopsy approaches relying on somatic mutation detection in peripheral blood are limited by the low fraction of circulating tumor DNA (ctDNA), due to restricted shedding of ctDNA across the blood-brain barrier. Given that brain tumors exhibit discrete DNA methylation patterns and that methylation profiling of whole blood captures systemic immune-mediated responses and low-level ctDNA, we hypothesized that glioblastoma-specific DNA methylation patterns are detectable in peripheral blood and can be integrated in a machine learning classifier to non-invasively diagnose glioblastoma and predict outcomes.Materials/Methods:
DNA methylation quantification using the Illumina Infinium MethylationEPIC array was completed in paired tumor tissue and whole blood samples collected prior to radiotherapy from participants with newly diagnosed MGMT unmethylated glioblastoma on the VERTU trial (ACTRN12615000407594). Overlapping differentially methylated CpGs (FDR < 0.05) in tumor and whole blood were identified using epigenome-wide association studies (EWAS) in comparison with external controls (NCBI GEO: GSE183656, GSE280465). A support vector machine model was trained using the VERTU blood EWAS methylation data, and model performance was tested using peripheral blood DNA methylation data from two external validation cohorts of patients with glioblastoma and other primary brain tumors (Mendeley Data: 10.17632/zrc982rvjm.2) or no history of malignancy (NCBI GEO: GSE286313).Results:
97 CpGs with robust effect sizes (fold-change >1.05 or <0.95) were differentially methylated (FDR < 0.05) and significantly overlapping (?²=139.66, df=1, p<2.2×10?¹6) in paired whole blood and tumor tissue from 86 VERTU trial participants, with enrichment of glioma-, cell cycle-, proteasome-, and mismatch repair-related pathways in blood. 42 CpGs (43%) overlapped with known islands (n=19), shelves (n=15), and shores (n=10), and were associated with genes implicated in tumorigenesis, neural development, cell cycling, DNA damage repair, cell growth, and immune activation. A methylation risk score was trained and externally validated to distinguish glioblastoma from non-glioblastoma blood samples (ROC AUC=0.87, 78% accuracy, 86% sensitivity, 75% specificity, F1 score = 0.68). The methylation signature was detectable in external serum samples, and risk scores were associated with survival (HR, 6.12 [95% CI, 1.08-34.81]; P = .04).Conclusion:
Glioblastoma-specific DNA methylation signatures are detectable in peripheral whole blood and serum, and correlate with survival outcomes. This study provides a novel approach for the use of blood-based DNA methylation profiling as a non-invasive diagnostic and prognostic tool for glioblastoma, and warrants prospective evaluation.