Introduction: Recent ex vivo genomic analyses demonstrate the existence of several molecular subtypes of glioblastoma, though the factors influencing the formation of these subtypes remain unclear. We hypothesized different brain regions may give rise to brain tumors that differ in their molecular profiles and clinical characteristics.
Methods: We developed a new method—Voxel-based Neuroanatomic Genomic Mapping (VNGM)—to identify distinct brain regions of tumor formation and characterize region-specific molecular features associated with each cluster of tumor formation. Briefly, N=576 glioblastoma patients underwent MR imaging; data were registered to a common stereotactic coordinate space. Radiographic tumor margins were identified and voxel-level annotations of tumor formation were made in each patient—i.e., for each measured voxel, patients were dichotomously coded according to whether they did or did not have tumor involvement measured. A voxel matrix was constructed containing comprehensive voxel-wise data on patterns of tumor formation across all 576 individuals. Hierarchical clustering of the matrix allowed individuals to cluster on the basis of shared common voxels of tumor involvement. Integrating genomic analysis with neuroanatomic clustering, region-specific molecular features associated with each cluster of tumor formation were characterized.
Results: VNGM identified three distinct brain regions of glioblastoma formation. Molecular subtype was influenced by brain region of tumor formation (P=0.0002). Region 1 was more likely to develop proneural/neural tumors (OR: 6.8, P<0.0001), whereas Region 2 developed mesenchymal/classical tumors (OR: 4.0, P<0.0001). Pathway analysis revealed Region 2 to be enriched in NF-kappa B, innate stromal immune responses, response to wounding, extracellular matrix formation and coagulation. Region 1 was characterized by synaptic transmission, with upregulated glutamate receptor signaling and calcium channel activity, and fetal neurodevelopmental gene expression pathways. These distinct brain regions also differed in resulting tumor histopathology, necrosis, invasion, patient age at onset and overall survival.
Conclusions: VNGN detected otherwise covert relationships among grossly distinct-appearing tumors, identifying three distinct brain regions of tumor formation. This integrated radiogenomic technique holds promise for uncovering region-specific influences underlying the development of molecular tumor subtypes.
Patient Care: A greater understanding of the factors that determine molecular subtypes and distinct cancer biology between patients will help clinicians better manage heterogeneous populations of glioblastoma patients in a patient-specific manner. This era of “precision medicine” will involve integrated decision models, with the goals of surgery (aggressive gross total resection vs cytoreduction) and the selection of adjuvant targeted therapies being guided by the region-specific features of the patient’s tumor.
Learning Objectives: 1) Population-level voxel-wise radiographic analyses
2) Integrated radiogenomic analyses and molecular characterizations
3) Clinical significance of region-specific difference in tumor subtypes for clinical decision-making and targeted therapeutics