Glioblastoma (GBM) is the most common, most deadly adult brain cancer and while treatment is standardised and aggressive, almost 100% of tumours recur. Primary GBM tumours that undergo standard treatment recur at the same site and exhibit a universal dysregulation in expression of genes associated with a specific chromatin remodelling protein. This project's hypothesis is that epigenetic switch (non-genetic influence on gene expression) facilitates the inevitable treatment resistance of GBM tumours and that stratifying patients according to the way their tumour will switch will ultimately determine a treatment course which will more effectively kill the tumour.
Explaining the science
The techniques used for the objectives of this project are detailed below:
- Use existing RNAseq data for paired pre- and post-standard treatment glioblastoma (GBM) tumours as training and test datasets to develop a gene-expression based biomarker that predicts the direction of an epigenetic switch observed universally in these patients during standard treatment.
Techniques: RNAseq data processing, differential gene expression analysis, machine learning approaches
- Apply this biomarker to data from the cancer genome atlas (~500 primary GBM tumours) as well as other publicly available primary GBM datasets to investigate the prevalence of responder subtype
Techniques: Data mining, machine learning approaches
- Investigate the association or correlation of responder subtype to biological (e.g. known driver mutations, existing subtypes, signatures of stemness and cell-cycle within the expression profiles) and clinical (e.g. age, sex, progression-free and overall survival) features of these primary patients/tumours.
Techniques: Survival analysis, statistical modelling
The project aims to develop a gene-expression based biomarker that predicts the standard response to treatment for primary glioblastoma (GBM) profiles and use this to investigate the biological and clinical relevance of these subtypes.
A classifier that can predict responder subtype from primary GBM will be useful in the NHS if it can be shown that it a) provides more accurate information with regards prognosis, or b) indicates specific treatments may work better than others.
The project developed a stratification classifier with neural networks using cross-validation. The accuracy of this was variable and will benefit from additional test data, which is currently being acquired. The optimisation work was able to highlight several genes and pathways that appeared to have the most predictive power, so these are being followed up with regards to mechanistic links. The researchers were also able to show that specific immune cell signatures were linked with one type of responder subtype over another, which is an additional avenue of follow up.