Uncovering the long tail of oesophageal cancer driver genes for patient stratification


Session type:


Lorena Benedetti1,Thanos Mourikis1,Elisabeth Foxall1,Damjan Temelkovski1,Joel Nulsen1,Juliane Perner2,Matteo Cereda3,Christopher Yau4,Rebecca Fitzgerald2,Michael Howell1,Paola Scaffidi1,Francesca Ciccarelli1
1The Francis Crick Institute, London, UK,2University of Cambridge, Cambridge, UK,3Italian Institute for Genomic Medicine, Torino, Italy,4University of Birmingham, Birmingham, UK



The incidence of oesophageal adenocarcinoma (OAC) in the UK is among the highest in the world and the five-year survival rate is below 20%. Its heterogeneous landscape, with high mutation and copy number burden where few cancer driver genes recur across patients, limits the choices of targeted therapies


We developed a new method based on machine learning that, unlike most available methods, identifies altered cancer driver genes in each individual patient. Once cancer genes were identified in each patient, we clustered patients by recurrently perturbed biological processes to allow patient stratification. Finally, we experimentally validated representative genes in pre-cancer and cancer models.


We applied our method to OACs from the Oesophageal Cancer Clinical and Molecular Stratification (OCCAMS) Consortium, a UK-wide initiative of tissue collection aiming at characterising OAC molecular landscape, stratifying patients and identifying new therapeutic targets. We identified cancer driver genes in each of the analysed patients, most of which are rare or patient-specific. However, they converge towards the perturbation of well-known biological processes related to cancer, including intracellular signalling, cell cycle regulation, proteasome activity and Toll-like receptor signalling. Interestingly, these genes are also altered in pre-cancer Barrett’s oesophagus, suggesting early clonal alteration. We mimicked the alteration of these genes in OAC lines and Barrett’s models, observing a growth promoting phenotype. Moreover, reverting their alteration led to cell death, pointing at dependencies that can be exploited in therapy.


Our machine learning approach allows the identification of cancer driver genes in individual patients independently of their alteration recurrence. This is particularly suitable for cancers with highly unstable mutational landscapes, like OAC, where recurrent alterations are rare. Our study proposes a new stratification of OAC patients and unravels vulnerabilities exploitable in therapy (Mourikis, et al. 2019 Nature communications, in press).