Tailored sampling approaches to capture cancer evolution in human tumour tissue
Loading...
Embargo End Date
2027-04-08
ICR Authors
Authors
Hanley, B
Document Type
Thesis or Dissertation
Date
2025-04-08
Date Accepted
Abstract
Intratumour heterogeneity (ITH) is a pervasive feature of solid cancers. This thesis examines how existing and novel sampling methodologies capture cancer evolution emphasising two forms of sampling; a modified form of random sampling termed representative sampling (RepSamp), and machine-learning augmented multi-regional profiling called TILAgg. There are four main results sections.Section 1 explores RepSamp as a tool to capture genomic ITH in breast cancer in the clinic. VAULT trial endpoints are reported. The landscape of RepSeq in breast cancer is described. The benefit of RepSamp in identifying evolving mechanisms of therapy resistance is highlighted. RepSamp for phenotypic biomarkers is introduced. Section 2 utilises RepSamp to decipher tumour evolution. A bespoke in silico sampling model is generated. Simulated and real-world tumour data are contrasted. Pheno-phylogenies are reconstructed using RepSamp of dividing cells. Evidence of selection in certain genes highlights the ability to identify existing and novel driver genes. The role of RepSamp in the identification of metastasis-seeding subclones is explored.Section 3 focuses on iTME, particularly in quantifying lymphocytic infiltration and a novel classifier is generated (TILAgg). This is validated using bulk sequencing approaches and is strongly predictive of patient outcomes. Section 4 presents an iteration of TILAgg using spatial transcriptomic profiling of 1000 genes applied to multi-regional melanoma TRACERx samples which are used to train a deep learning classifier. In parallel, a deep learning model (TILAgg2.0) is trained on data from TILAgg1.0 annotations and lymphocyte annotations in addition to spatial transcriptomics, to classify H&E slides with high throughput and automation. I conclude that sampling is an essential component of cancer molecular profiling but must be tailored correctly by context. RepSamp can better select a sample of cancer cells for molecular profiling. Deep learning can extend the inferences of molecular assays to larger sample areas by classifying H&E image features. .
Citation
2025
DOI
Source Title
Publisher
Institute of Cancer Research (University Of London)