Modelling the effect of vascular architecture on the outcome of cancer therapies

Thesis

Tumour angiogenesis leads to the formation of abnormal blood vessels that are structurally and spatially heterogeneous. Consequently, poor blood perfusion and increased hypoxia impair a tumour's response to cancer therapies. Strategies to normalise vasculature have yielded mixed results, preventing their widespread clinical deployment. In this thesis, we use mathematical and computational models to determine the influence of a tumour’s vascular architecture on its response to treatment.

We first identify vascular architectural features that correlate with enhanced perfusion following radiotherapy in human tumour xenografts. Our findings reveal that vascular networks with thinner vessels and higher proportions of angiogenic sprouts exhibit the largest increases in perfusion after irradiation. Moreover, we identify cases in which perfusion increases due to the rerouting of blood as a consequence of radiotherapy-induced pruning, underscoring the potential of a single dose to induce a window of normalisation in which successive doses can be more effective.

To formalise the link between perfusion and oxygenation, we evaluate and refine mathematical rules to model the distribution of haematocrit using data from microfluidic networks. We use these rules to simulate oxygen transport in synthetic vascular networks and evaluate architectural metrics on their ability to predict the extent of hypoxia. Our investigation finds that network connectivity is a robust characteristic for inferring tumour oxygenation and, as a result, the oxygen-modulated response to radiotherapy. We also demonstrate that a normalisation window, in which adjuvant treatments are likely to be more effective, can be induced by biophysical factors, rather than solely by biochemical factors.

We conclude this work by assembling and demonstrating a computational pipeline to detect, analyse, and simulate real-world vascular networks from microscopy images of murine tumours. Using this pipeline, we find evidence to support our inferences from synthetic networks. With further experimental validation, our findings would lay the foundation for new translational research to improve a tumour’s sensitivity to cancer therapies.

Authors: Narain, V

• DOI: 10.5287/ora-r8rj2v7qg
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: digital pathology; mathematical biology; blood flow; spatial statistics; simulations; oncology; vasculature; machine learning; networks; hypoxia
• Subjects: Mathematical models; Computational biology; Cancer