Spencer, Charlotte; (2024) Defining Adaptive Phenotypes in Kidney Cancer using Computational Pathology. Doctoral thesis (Ph.D), UCL (University College London).

Text Spencer_PhDThesis.pdf - Submitted Version Access restricted to UCL open access staff until 1 September 2025. Download (17MB)

Abstract

ccRCC (clear cell Renal Cell Carcinoma) is common and can be deadly. Clinical management is complicated by varied and unpredictable clinical outcomes, leaving patients at risk of over or under treatment. The diverse clinical outcomes can be predicted from the genetic features of the tumour. Conserved patterns of genetic alterations (including PBRM1 mutation, BAP1 mutation, loss of 9p and loss of 14q) drive the evolutionary trajectory of the tumour and so dictate the tempo of disease progression. However, translating this understanding to clinical workflows remains costly and is not logistically feasible. In this thesis I aim to define the link between routine H&E histology (the cornerstone of cancer diagnostics), and the genetic events responsible for disease pathogenesis to provide a critical link in understanding biological determinants of tumour behaviour. Furthermore, by developing artificial intelligence centred on H&E histology I create reproducible, scalable computational pathology tools to study tumour biology in large phase 3 cohorts and build foundations of future computational pathology biomarkers. I show that PBRM1 and BAP1 mutations, already recognised as driving distinct, prognostically relevant tumour evolutionary trajectories, are associated with distinct histological hallmarks centred in vascular network structure. I demonstrate the plastic nature of this histological imprint where both tumour cell and vascular features change with the acquisition of somatic copy number alterations. Finally, I illustrate that histology reflects global, prognostically relevant genetic markers of tumour evolution and that histology can track clonal evolution within individual tumours. Critically, such features can be robustly extracted from routine H&E WSIs with many of my pipelines incorporating high-throughput, generalisable computational pathology approaches. This work provides a mechanism to translate our knowledge of the molecular determinants of patient outcome into the clinical setting without changes to the routine of the laboratories. It represents foundational work for developing H&E based computational pathology biomarkers that incorporate complex molecular features to offer groundbreaking diagnostics to optimise patient management.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item