Investigating intratumour heterogeneity analysis methods and their application in GBM

Glioblastoma (GBM) is an incurable cancer with a median survival of 15 months. Despite debulking surgery, cancer cells are inevitably left behind in the surrounding brain, with a minority able to resist subsequent chemoradiotherapy and eventually form a recurrent tumour. This resistance is likely influenced by the cells’ genotypes, which show high variability (intratumour heterogeneity), as a result of tumour evolution. Characterising changes in the genetic architecture of tumours through therapy, may allow us to understand the effect that different mutations and pathways have on cell survival, and potentially identify novel targets for counteracting resistance in GBM. Such analyses involve detection of mutations from bulk tumour samples, and then delineating them into individual genetically distinct ‘subclones’, through subclonal deconvolution. This is a complex process, with no reliable guidelines for the best pipelines to use. I therefore developed methods to allow simulation and in silico sequencing of genomes from realistically complex, artificial tumour samples, so that I could benchmark such pipelines. This revealed that no tested pipelines, using single bulk samples, showed a high level of accuracy, though mutation calling with Mutect2 and FACETS, followed by subclonal deconvolution with Ccube, showed the best results. I then used alternative approaches with the largest longitudinal GBM dataset investigated to date. I found that evidence of strong subclonal selection is absent in many samples, and not associated with therapy. Nonetheless, this does not negate the possibility of smaller, or less frequent, pockets of altered fitness. Using pathway analysis combined with variants that are informative of tumour progression, I identified processes that may confer increased resistance, or sensitisation to therapy, and which warrant further investigation. Lastly, I apply subclonal deconvolution to investigate mouse-specific evolution in GBM patient-derived orthotopic xenografts and found no clear evidence to suggest these models are unsuitable for investigations relevant to humans.