Abstract/Summary

An important question in cancer evolution concerns which traits make a cell likely to successfully metastasise. Through a combination of experimental evolution and computer vision a series of mathematical models have been developed throughout this thesis to investigate the individual signal processing behaviour of cancer cells during dispersal. In Chapter 2 a convolutional neural network is used to demonstrate how the morphology of individual cells can be automatically segmented within phase contrast time-lapse videos. The segmented morphologies are then used in Chapter 3 to explore the idea of signal processing mediated dispersal to reveal a density-dependent phenotype only seen in cells selected for distant site colonisation. Specifically, the model shows that the rate of morphological change is positively correlated with the speed of migration when the local cell density is high. However, when the local cell density is low the opposite relationship is displayed: the rate of morphological change decreases with an increase in migration speed. Chapter 4 then builds upon the results of Chapter 3 to develops two temporally dependent morphological model that quantify short term temporal changes in dispersal dynamics at both a population and single cell level. The temporally dependent models reveal that in fact a subset of cells in all of the experimental populations can adopt similar complex behaviour. However, the populations differ in their behavioural demography as well as the frequency at which a given behaviour is adopted through time. Finally, Chapter 5 employs a similar temporally resolved approach to investigate the interaction between the broader cancer cell population and a small subset of cancer cells known as poly-aneuploid cancer cells. In summary, this thesis harnesses the power of mature mathematical techniques to investigate novel and emergent characteristics of metastatic dispersal in a quantitative and statistically robust manner.