BACKGROUND: The primary treatment of GBM is surgical resection of the contrast enhancing mass, while regions of non-enhancing tumor are left behind, and inevitably gives rise to recurrence. Thus, there is therapeutic significance to understanding the cellular and molecular features of the non-enhancing margins of GBM. Advances in comparative genetic and molecular analyses have uncovered patterns of gene expression and genetic alterations in GBM, and led to the recognition of clinically significant tumor subtypes. However, most previous studies have been based on analysis of tissue that was removed during surgery without documenting the radiographic location. This has precluded the systematic study of intra-tumoral heterogeneity, and its correlation with radiographic regions. We propose a method of sequential image guided biopsies, allowing for specimen localization, intra-tumoral comparative analysis, and the systematic study of infiltrated tissue beyond the borders of contrast enhancement. METHODS: In 77 patients with GBM, multiple stereotactic-guided biopsies were obtained from radiographically distinct regions prior to resection. The radiographic localization of each biopsy was mapped using the 3D intraoperative neuro-navigation system (BrainLab). Sampled regions included areas of contrast enhancement (CE) and adjacent non-enhancing FLAIR+ (NEF+). Samples from each region were divided into pieces processed for RNA-seq and histological analyses. RESULTS: There were significant differences in the cellular density and cellular composition between the CE and NEF+ samples. The CE samples had higher cellularity than NEF+ samples, and were more likely to contain the histological hallmarks of GBM, including vascular proliferation and pseudopalisading necrosis. In contrast, the NEF+ regions showed the histological features of diffusely infiltrating glioma with neoplastic glial cells intermingled with non-neoplastic and reactive cells. RNA-seq showed that the CE and NEF+ regions also had markedly different expression patterns. Comparisons between our image guided biopsies and the TCGA data set showed that the CE regions were enriched for genes associated with the mesenchymal, classical and proneural subtypes while the NEF+ region were enriched for genes of the neural subtype. Further analysis revealed that region-specific and subtype-specific differences in expression patterns were predominantly due to differences in the cellular composition. CONCLUSIONS: Cellular composition is a major determinant of the expression patterns seen at the non-enhancing margins of GBM. The different GBM subtypes show distinct expression patterns that relate the contrast enhancing centers to the non-enhancing margins of the tumors. Understanding these patterns provides a means to infer the molecular and cellular features of glioma tissue left behind after surgery. SECONDARY CATEGORY: Tumor Biology.