Identifying image biomarkers in the spatial evolution of liver cancer

BACKGROUND While many cancers arise through the acquisition of specific oncogenic mutations, a wide variety of non-genetic (or epigenetic) alterations and aberrant transcriptional programs ultimately determine cancer cell phenotypes and tumour histomorphology.

We recently discovered DNA lesion segregation, which provides one explanation for multiallelic and combinatorial heterogeneity, and occurs most frequently in hepatobiliary cancers.

Hepatocellular carcinoma (HCC, the most common primary liver cancer) is a heterogenous, poor-prognosis disease which is a major research priority as a cancer of unmet need. Despite significant progress establishing molecular subgroups of HCC, clinical outcomes are heterogenous.

In this fellowship, I test the hypothesis that patients with liver cancer can be effectively stratified using genomic pathology approaches to identify clinically relevant image biomarkers, refine current morphomolecular classifications, and ultimately improve patient care. AIMS Aim 1: Identify prognostic and predictive image biomarkers in HCC.

Aim 2: Characterise the genomic and transcriptional spatial heterogeneity in HCC.

METHODS I have access to tissue and clinicopathological details for 354 patients undergoing liver transplantation for HCC at Cambridge Transplant Centre, and collaborative access to cohorts in two other UK Transplant Centres.

We will use digitised histopathology whole slide images in orthogonal machine learning approaches, (i) identifying objects (e.g. tumour/cells/nuclei/mitoses) to generate prognostic models; and (ii) deep learning using existing histopathological classification methods to determine morphological correlates of molecular/clinical data.

This approach will identify clinically relevant image biomarkers that are currently ‘invisible’ to histopathologists.

We will analyse whole-genome sequencing of tumour samples to extract mutational signatures and drivers, map intra-/inter-tumour genomic heterogeneity, and identity molecular candidates associated with HCC aetiology/recurrence.

Additionally, sequencing will allow the novel application of our recent discoveries of lesion segregation and multiallelism to solid human tumours. We will deploy spatial transcriptomics to map in situ gene expression at near single-cell granularity.

Integrating high-depth spatial transcriptomics with high-resolution whole slide imaging will allow us to find dysregulated pathways within/between tumours and characterise the tumour microenvironment.

OUTCOMES In this Fellowship, I expect to successfully identify prognostic image biomarkers and use integrative models to provide human-intuitive explainability.

All data will made freely available to the scientific community, and similar approaches could be applied across tumour types. This translational project has potential to affect significant positive change for patient management.

Specifically, by making use of assays routinely performed within NHS diagnostic tissue pathways (histopathology and genome sequencing), prognostic image biomarkers could be immediately implemented into prospective studies at minimal cost.