NOVEL IN-SILICO APPROACHES TO IDENTIFY DRIVER MUTATIONS IN CANCER GENOMICS DATA

Abstract

The cancer genomics era has now witnessed multiple examples of the clinicalbenefits of targeting mutated oncogenes with selective inhibitors. Manypatients who would have been too frail to withstand chemotherapy, andtherefore have no treatment options, have been given additional months oreven years of good quality life with these therapies. These successes havedriven the search for oncogenes in all tumour subtypes and next generationsequencing (NGS) has been the main tool used for this task. However despitethe sequencing of hundreds of thousands of cancer samples with NGS,knowledge is still lacking regarding driver genes for a significant proportion ofcancer subtypes. The overarching theme of this thesis is the development ofnovel in silico approaches to enhance the detection of novel driver mutations.The first study uses discrepancies in NGS mutation calling between differentinstitutes to identify consistent areas of the exome that are poorly sequenced.These areas are predominantly GC-rich causing a problem for cancers, suchas lung cancer, with a mutational signature preferentially affecting GC-richregions. As proof of principle, a PAK4 mutation in a GC-rich region previouslymissed by two large sequencing studies is shown to be activating upon theERK pathway. Analysis of NGS read coverage demonstrates that inadequateNGS of GC-rich regions is prevalent in older projects, potentially obscuringdriver mutation detection.The remainder of the work focuses on using novel filtering algorithms to cutthrough the mutational noise caused by the hundreds of inconsequentialpassenger mutations present in each NGS dataset. Linking structural andfunctional data of critical kinase motifs I developed a specific loss-of-function(LOF) kinase mutation caller. Analysing pan-cancer somatic mutational datawith this algorithm revealed MKK7 as a novel tumour suppressor in gastriccancer that was subsequently validated biochemically. The final piece of workapplies the LOF algorithm to the germline data of a rare neuroendocrineproliferative condition called DIPNECH. This identified a previously unreportedSNV of CDK8 in one patient. Reanalysing pan cancer somatic mutational datareveals similar predicted LOF mutations of CDK8 in small cell lung cancer,also a neuroendocrine malignancy. Additional algorithms were used to filterthe DIPNECH data to provide the first genomic insight into this rare disease,highlighting potential genetic links with other neuroendocrine malignancies.Overall the approaches presented in this thesis offer additional solutions toenhance the detection of driver genes from NGS datasets. Most importantlythis work demonstrates the power of using functional considerations to filterdriver mutations from the mutational noise. In the future I aim to develop thisapproach to assist gain-of-function (GOF) kinase mutation identification andwrite similar algorithms for other protein families.

Date of Award	1 Aug 2017
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Paul Lorigan (Supervisor) & John Brognard (Supervisor)