A Single-Cell Compendium of Intratumoural Heterogeneity

Tumour microenvironments (TMEs) are complex entities comprised of distinct malignant, immune, and stromal cell populations driving cancer progression. Recent advances in single cell transcriptomic technologies have enabled the study of intratumoural heterogeneity (ITH) through whole transcriptional profiling of cell types and states comprising TMEs, however, the complexity associated with ITH presents computational challenges for the characterization of cell phenotypes. In this thesis, I develop and apply novel methods to characterize ITH within malignant and non-malignant cells of the TME and evaluate downstream clinical implications. First, I develop a machine learning cell type classification framework, named single cell annotation of tumour microenvironments in pan-cancer settings (scATOMIC). Using a curated reference of over 300,000 cells, I train scATOMIC, a stepwise, hierarchical annotation tool for both malignant and non-malignant cells across 19 common cancer types. Through extensive validation of scATOMIC’s accuracy in over 225 tumour biopsies, I demonstrate superior classification performance in the context of TMEs compared to other methods. I highlight potential clinical uses in scATOMIC’s modular design through accurate subclassification of breast cancer cells, and its application to predict the origin of metastatic tumours. Next, I apply scATOMIC to analyze over 1,000 tumours, creating a comprehensive pan-cancer atlas that includes over 3.6 million cells, all with standardized cell annotations. I characterize unique gene expression programs representing different phenotypic cell states across malignant cells within the atlas. Through deconvolution of clinical datasets, I validate prognostic survival signatures in 11 cancer types. I explore the spatial colocalization of cell states through high resolution spatial transcriptomics, revealing a link between aggressive transcriptional cell states and histological growth patterns. Finally, using a drug repurposing screen, I nominate candidate drugs for selective elimination of aggressive cancer cell states. In summary, this thesis contributes novel methods and applications to enhance the characterization of the cells underlying ITH and their clinical implications. My research illuminates rare phenotypic populations associated with metastatic disease and poor survival as novel candidates for precision oncology.