Withnell, Eloise; (2025) Spatial Modelling of the Epithelial-to-Mesenchymal Transition in Cancer Using Geostatistical and Machine Learning Approaches. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Withnell_10210046_Thesis.pdf Download (11MB) | Preview

Abstract

The epithelial-to-mesenchymal transition (EMT) is an important cellular process involved in tumour progression, metastasis, and therapy resistance. However, the influence of the tumour microenvironment (TME) and genomic factors on EMT, and the discrete states within this transition, remains incompletely understood. In this thesis, I develop geostatistical and machine learning methods to analyse spatial transcriptomic data, to understand the spatial relationships of cancer cells undergoing EMT. I present a novel Python package, SpottedPy, which can identify spatial hotspots of gene signatures and cell types and assess their spatial interactions with other hotspots. Using this approach, I identified EMT niches associated with angiogenic and hypoxic regions, surrounded by CAFs and macrophages. EMT hybrid and mesenchymal hotspots followed transformation gradients, becoming increasingly immunosuppressed. Importantly, SpottedPy is a flexible package, which enables users to explore spatial relationships at different scales, from immediate neighbours to larger tissue modules, allowing for new insights into the tumour microenvironment. Building on these spatial insights, I develop a graph neural network and geographically weighted regression framework to quantify the relative contributions of intrinsic genomic changes and extrinsic microenvironmental signals on cell plasticity programmes. The approach strengthens the evidence that targeting the TME is more important for targeting EMT as opposed to targeting genomic factors. It highlights the importance of the TME in inducing both subtle, short-term changes and stable, long term phenotypic change, whereas genomic alterations primarily contribute to more stable, long-term changes. I showed that the mesenchymal phenotype is more deterministic, while hybrid states are less predictable and thus potentially more plastic. Additionally, I found that relationships between EMT states and particular TME populations do vary across different tissue regions, notably with myoepithelial cells. Overall, my work provides an in-depth molecular and spatial characterisation of EMP, while highlighting novel methodological approaches for capturing and measuring cell plasticity. These insights could help inform therapeutic approaches that target the genetic and microenvironmental factors linked to cancer cell plasticity.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVJSONXML

Archive Staff Only

View Item