Multi-omics modelling to identify actionable targets in epithelial ovarian cancer

Abstract

Constraint-based modelling is a powerful technology, which can reconstitute the metabolism of an organism through the integration of omics data with a genome-scale model (GEM) and provide a framework for hypothesis-generating, in silico genetic engineering studies. Facilitated by the complete mapping of the human genome and high-throughput omics technologies, constraint-based modelling has been extended to human studies. Altered metabolism is an emerging hallmark of cancer, therefore, this disease represents an ideal case study for the optimisation of novel constraint-based modelling methods. Despite there being recognised metabolic signatures and vulnerabilities within ovarian cancer, this cancer type has been under-represented by constraint-based modelling projects. Furthermore, clinical phenotypes limiting cancer treatment, such as distinct histological subtypes, some of which exhibit strong chemoresistance, provide an incentive for the constraint-based modelling of ovarian cancer. Here, a novel growth-directed, multi-omics integration algorithm was developed and used to study ovarian cancer. Optimisation of this algorithm highlighted the need for the standardisation of media conditions specified across models, and to address this, an RNAseq analysis and growth studies were performed to understand the biological impact of changing media. To validate model growth predictions, a single genetic target, triosephosphate isomerase 1 (TPI1), was identified by gene knockout simulations and explored experimentally. Importantly, the flux prediction profiles predicted by the in silico knockout of TPI1 were validated using a model which had been constrained using gene expression data derived from an experimental knockdown. Experimental work confirmed that the knockdown of TPI1 inhibits cell proliferation and RNAseq analysis predicted roles in apoptosis, highlighting this enzyme as a potentially actionable target within ovarian cancer. Lastly, my integration algorithm was used to generate personalised models representing patient-derived ovarian cancer samples. This application suggested four distinct metabolic signatures within high-grade serous ovarian cancer, which future studies could associate with metadata to explain clinical phenotypes.

Date of Award	31 Dec 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Stephen Taylor (Supervisor), Jean-Marc Schwartz (Supervisor) & Paul Townsend (Supervisor)