Current Challenges in Glioblastoma: Intratumour Heterogeneity, Residual Disease, and Models to Predict Disease Recurrence

Abstract

Glioblastoma (GB) is the most common primary malignant brain tumor, and despite the availability of chemotherapy and radiotherapy to combat the disease, overall survival remains low with a high incidence of tumor recurrence. Technological advances are continually improving our understanding of the disease, and in particular, our knowledge of clonal evolution, intratumor heterogeneity, and possible reservoirs of residual disease. These may inform how we approach clinical treatment and recurrence in GB. Mathematical modeling (including neural networks) and strategies such as multiple sampling during tumor resection and genetic analysis of circulating cancer cells, may be of great future benefit to help predict the nature of residual disease and resistance to standard and molecular therapies in GB.

Keywords: Bayesian models; GBM; intratumor heterogeneity; neural networks; residual disease.

Figure 1

Intratumour heterogeneity in the evolution of GB. (A) Phylogenetic reconstruction of a GB case based on copy number alterations. (B) It has been hypothesized that TICs reside in the SEZ and can contribute to tumor maintenance (19). These cells (shown in purple) may be potential new targets for molecular therapies in GB. Also shown is the maintenance of the tumor bulk by cells in the SEZ containing the same mutations as described in the phylogenetic tree, and how current therapies target the tumor bulk (shown in red). (C) combines (A,B) to give an overview of tumor evolution in this case across time and in physical location. The case described in all parts of this diagram and all corresponding genetic information was obtained from Sottoriva et al. (10). (B) is adapted from the review by Goffart et al. (20).

Figure 2

Neural networks used in cancer systems biology. (A,B) illustrate the use of dynamic Bayesian networks. (A) describes how events interact in a cyclic causal manner, as displayed in a dynamic Bayesian network (B) that represents all variables at two time points, allowing inference of causal relationships. (C–E) are visualizations of S-systems’ analyses. The data represents breast cancer cells treated with heregulin concurrently with pertuzumab (a HER2 inhibitor). The thickness of the lines indicates the strength of the interaction. (C) represents the first three time points at which data was obtained, (D) the next three overlapping time points, and (E) a further three overlapping time points. These diagrams may assist analysis of interactions in the network, as they appear to show dissociation of HER2 in response to treatment with its inhibitor. All components of this figure, including Bayesian networks and S-systems’ analysis, are adapted from the review by Faratian et al. (31).