A Life or Death Decision: Quantifying the Balance Between Cell Survival and Death in Skin Development

Abstract

This thesis describes a collection of experimental methods and mathematical techniques that were developed to analyse intracellular signalling processes for adherent cells within their native tissue. Human epidermis was selected as a model system, and interfollicular keratinocytes were studied along the spatiotemporal differentiation gradient that occurs across the width of the epidermis. Several experimental modalities were considered. Immunouorescence labelling with confocal microscopy was shown to produce quantitative image data inherently rich in quantitative spatial information, as required for a quantitative analysis. The resulting data had a field of view that was large enough to capture the entire epidermis in cross- section, while retaining sufficient resolution to identify major sub-cellular compartments. A set of proteins and phospho-proteins related to the canonical ERK- MAPK intracellular signalling cascade were labelled within human epidermis. The image data were sampled to produce a quantitative description of protein abundance within the cytoplasm, nucleus and plasma membrane. A spatial transformation was applied, converting the signal intensity data from a two-dimensional spatial domain, to a one-dimensional tissue-normalised distance that explicitly defined the discrete epidermal layers. These data were analysed within a statistical framework, and the strength of pairwise statistical association was evaluated using Pearson correlation and Mutual Information. A null hypothesis was proposed, that the observed statistical associations are independent of spatial conditioning of the data. A null distribution associated with this hypothesis was constructed, and a threshold was derived by applying a confidence interval over this null distribution, identifying relationships that had a relatively strong statistical association when their abundance data were conditioned upon the spatial information. Selected relationships were investigated in further detail and properties of their association were elucidated. Several inter-compartment relationships could be interpreted as translocation events, and features of these associations were examined with respect to known signalling behaviours. Finally, a collection of published network inference techniques were applied to examine the data within a multinomial framework developed for large signalling networks. The solution stability of these methods was investigated, and each technique was evaluated by its ability to recapitulate a literature derived reference network.