High content profiling in oesophageal adenocarcinoma to inform drug discovery and patient stratification

Abstract

Oesophageal adenocarcinoma (OAC) is a highly heterogeneous disease, dominated by copy number alterations and large-scale genomic rearrangements. Such characteristics have hampered both therapeutic target discovery and the clinical success of targeted therapies, contributing to its status as an area of unmet therapeutic need. Phenotypic drug discovery describes the screening and selection of hit or lead compounds based on quantifiable phenotypic endpoints from cellbased assays or model organisms without prior knowledge of the drug target.

Thus, this may prove to be a beneficial strategy for complex, heterogeneous diseases where target biology is poorly understood and where modern, target directed drug discovery strategies have made little impact on patient care, as exemplified by OAC.

In the chapters that follow, I describe the development and validation of a high content profiling assay and machine learning pipelines to identify repurposing opportunities in OAC. Following on from this I focus on one group of compounds which show high potency and selectivity for targeting OAC cell lines relative to tissue-matched controls. I further explore the mechanism of action of these compounds and potential patient stratification hypotheses to support clinical translation.

In the first chapter I present an in-depth study of compound induced cell morphology in the context of a heterogeneous panel of eight oesophageal cell lines. To achieve this I profiled a small number of reference compounds and built a library of phenotypic fingerprints for each mechanistic class. I then interrogated the phenotypic fingerprints using a variety of supervised and unsupervised machine learning techniques. This work demonstrates that compound induced morphologies are reproducible across cell lines, despite underlying morphological heterogeneity, allowing machine learning classifiers to be applied to previously unseen cell lines. The second chapter describes the application of the high content assay to a comprehensive small-molecule screen encompassing 19,555 small molecules, eight cell lines, 512x384 well plates, 3.9 million images and 36 TB of data for the identification of new drug discovery and drug repurposing opportunities in OAC. Following on from this, in the third chapter, I focus on one particular group of compounds identified as highly selective for OAC and explore their mechanism of action. In the final chapter, through the integration of transcriptomic pathway analyses we gain insight into drug selectively and the basis for future biomarkerbased clinical trials in OAC.