Search

The biopharmaceuticals developed under Personalised Medicine (PM) are the most promising medical treatments of this century, yet their commercialisation on large scale remains sub-optimal. The manufacturing and delivery of advanced therapy medicinal products created for individual patients rather than population groups are highly affected by worsened bottlenecks, such as low global demand, low shelf-lives, or increased product fragility. The current pharmaceutical supply chain configurations are optimised for mass delivery and, hence, they lead to long waiting times and high costs per patient in the delivery of PM, where each product corresponds to one exact patient. This thesis argues for the immediate necessity of new decision-making frameworks adaptable to the requirements of PM. We compare the bottlenecks encountered in the new personalised medical products to the most common mature supply chains of the healthcare and pharmaceutical industries, highlighting their dissimilarities. We make an initial theoretical contribution by uncovering the need for a new supply chain in light of the rapid clinical developments of personalised medical products. To address some of these challenges, we focus the rest of the thesis on the strategical level of the supply chain. Considering the expectations of the real-world, we propose several multi-objective mathematical models and solution methods for large scale facility location problems. Our formulations follow both centralised and decentralised networks and aim to find optimal locations for multiple types of interdependent facilities commonly met in the PM supply chain. The models and algorithms proposed are validated using data corresponding to products with current market approval and an estimation of global demand. The results presented throughout the thesis are consistent with the current motivations and proposed directions of how the PM supply chain should look. We show that the benefits brought by considering PM as standalone and not part of the regulations of the more traditional pharmaceuticals overcome the disadvantages. The development of specialised decision support tools can lead to smaller costs for biopharmaceutical companies, lower delivery times, and overall better global coverage. This research was in collaboration with an industrial partner, Biopharm Services, who has contributed to validating the mathematical formulations with respect to their practicality for biopharmaceutical companies and have provided part of the data used.

Personalised Medicine (PM) is touted as a medical revolution where medical treatment and diagnosis is tailored to the individual patient so that it is optimal, safe, and exactly appropriate. Each chapter in this dissertation deals with the conceptual, methodological, epistemic, or ethical issues of personalising medicine that influence our ability to reliably predict, diagnose, and treat disease for individual patients. I argue that PM in its current form is insufficient in several ways. My dissertation contributes to the literature by providing arguments for why and how we should reconceive of PM. In Chapter 1, I describe and analyse two conceptions of PM: broad-based and biological. I reject the popular claim that biological conceptions of PM are not truly personalised because they do not capture holistic aspects of personhood. Instead, I argue that the real problem for genome-based conceptions is that genomics is frequently imprecise and explanatorily not robust about underlying causes of disease. In Chapter 2, I illustrate that some theorists combine their defence of the use of race in medicine with an appeal to the value-free ideal. Against this view, I contend that contextual value judgements are important in assessing the use of race in medicine because evidence on the epistemic usefulness of racial categories in medicine can be transiently underdetermined and inductively risky. In Chapter 3, I interrogate the controversial use of racial categories in equations to predict kidney function and show that equations that include race in their estimations are inadequate at accurately predicting kidney function. In Chapter 4, I assess the underexplored reliability of what I call the 'stratification strategy'. This strategy requires that clinicians make therapeutic predictions about individual patients based on evidence of commonly shared molecular biomarker status among patient subgroups. I argue that in many instances of its use, this strategy has low reliability because biomarker evidence lacks credibility. Lastly, in Chapter 5, I draw attention to African Traditional Medicine as a broad-based form of PM. I consider the rationale to decolonise medicine as it applies to African Traditional Medicine in South Africa. I argue that interpreting this rationale as many do-that African Traditional Medicine is epistemically equivalent to Western mainstream medicine-implies an untenable wholesale medical relativism.

Colorectal cancer (CRC) is a multifaceted and heterogeneous disease. The molecular profiling of CRC has made it possible to detect tumours with unique biological features. Gene expression-based signatures have been at the forefront of cancer research in providing valuable prognostic and predictive information. However, these published signatures rarely make any clinical impact. This thesis utilises in silico methods to interrogate a pre-existing gene signature and evaluate the underlying CRC biology defined by the stratification based on the gene signatures. Additionally, accessing the published gene signatures, a new classification system was developed that defines pathway-driven tumours for patient stratification and tested its prognostic value. A transcriptional breast cancer (BC) originated DNA Damage Immune Response (DDIR) assay, revealed overlapping proinflammatory-like tumour biology in CRC, but with no clinical benefit from oxaliplatin (Chapter 3). Utilising transcriptional gene signatures from the gene set database, pathway-level unsupervised class discovery and classifier development from KRAS mutant CRC revealed three groups, named KM1, KM2 and KM3 (Chapter 4). The biological and prognostic evaluation revealed that KM1 were MYC driven cell-cycle activated, tubular, LGR5+ stem-like tumours with good prognosis, KM2 were stromal/immune rich, of serrated origin and regenerative stem-like biology with diverse pathway activation, including TGF-β activation, EMT and immune responses, and an intermediate prognosis. A novel finding of this classification was KM3, with large repression of pathways such as cell cycle, EMT, KRAS signalling activity, but elevation in neurotransmitter- and enteroendocrine-like biology, highly differentiated tumours, the potential inclusion of epigenetic dysregulation in maintaining stem plasticity, and a clinically worse prognosis (Chapter 5). In summary, this thesis provides a new pathway-driven classifier for CRC that includes a novel, previously uncharacterised subset of CRC subtype as KM3 with clinically poor outcome, which provides a unique biological phenotype that needs further research to translate into the clinic.

Rare monogenic diseases affect around 300 millions worldwide at any given time. Although genetically and clinically heterogenous, rare diseases are frequently caused by pathogenic mutation in a single gene; in particular, missense and in-frame insertion and/or deletion (indel) variants. Differentiating the few truly pathogenic alterations from the background of functionally neutral variants is a challenge which has resulted in a translational bottleneck. Largely due to the absence of segregation and functional data, the majority of reported variants-of-interest remain clinically defined as variants of unknown significance (VUS). Since experimental validation of all variants is impractical, computational prediction tools have been designed to attempt to prioritise variants for downstream analysis. While accurate variant prioritisation can lead to fewer VUS and a higher rate of diagnosis, this has yet to reach clinical standard. The aim of this study was to i) design a protein-specific prediction model to improve missense variant prioritisation and ii) use integrative structural analysis to differentiate pathogenic from benign in-frame indel variants. Using a protein-specific approach to variant analysis, structural and clinical data were combined using machine learning to differentiate pathogenic from benign missense variants in a subset of genes. The resulting prediction model, ProSper (protein-specific variant interpreter), outperformed a number of widely-used or recently developed missense prediction tools for the majority of genes studied. For a group of in-frame indel variants that were classified as VUS in a lab, structural interpretation was used to show variant clustering and accurately differentiate pathogenic variants in a family of genes. Variant prioritisation and interpretation using these approaches in clinically relevant situations can improve rare disease diagnosis and patient management. In addition, structural analysis at the atomic-level can be used to assess a variant's structural impact and provide data on the molecular mechanism of disease. This research adds to the body of evidence showing improved accuracy in variant prediction and prioritisation in groups of genes using a protein-specific approach compared to general predictors. Incorporating integrative structural analysis into the diagnostic framework can accelerate variant interpretation and minimise the rate of VUS leading to correct and timely diagnosis in more patients.

Improving healthcare systems mandates a shift towards personalised and preventive medicine, especially in the management of cardiovascular diseases, the world leading cause of death. The anatomy and function of the heart can be accurately and non-invasively assessed using cardiac magnetic resonance. Nevertheless, it is not yet fully understood how their interplay modulates disease outcomes. In this regard, this Thesis (1) reviews the crucial role of computational modelling towards precision cardiology, by building the 'digital twin' of a patient; (2) develops a fully automated pipeline to construct 3D computational models of human left ventricle from 2D cardiac magnetic images, using state-of-art deep learning approaches; and (3) proposes an enhanced structural and functional analysis to disentangle how adverse outcomes are modulated in cardiovascular disease. This is showcased in acute myocardial infarction, the deadliest cardiovascular disease not only on first manifestation, but on reoccurrence and secondary adverse outputs. In this scenario, the approach introduced in this Thesis is shown to provide additional prognostic information towards a better post-infarction management. Moreover, it identifies 3D shape and contraction patterns that are related to adverse outcomes, building a more profound understanding of remodelling after infarction. Furthermore, the automation of the analysis removes the burden of manual annotations and intra- and inter-observer variability, as confounders of prognosis. All this argues for the clinical impact of this Thesis. Ultimately, the Thesis evidences the potential of 3D computational models for cardiovascular disease management, towards personalised and preventive care.

Triple negative breast cancer (TNBC) is a highly heterogeneous disease that accounts for around 15% of all breast cancers but is responsible for a disproportionately high rate of mortality. Characterised by lack of expression of the hormone receptors estrogen receptor α (ERα), progesterone receptor (PR) and devoid of amplification of the human epidermal growth factor receptor (HER2), these cancers cannot be treated with targeted therapies such as Tamoxifen and Herceptin which are often used in breast cancer. Thus, the current standard of care (SoC) for these patients is cytotoxic chemotherapy (often DNA damaging agents +/- anti-microtubule agents) with some patients responding well while others derive no clinical benefit. However, all patients experience unwanted and unpleasant side effects regardless of response. There are currently no biomarkers to predict which patients will respond to SoC and no targeted treatment approaches for those who do not respond. Subsequently, less than 30% of these patients will survive beyond 5 years. An in-house, microarray-based gene expression analysis has identified a gene signature that has been shown to predict outcome in TNBC following SoC treatment. This project aims to build on findings from this analysis by identifying novel biomarkers and therapeutic targets that can be exploited to allow a personalised approach to the treatment of TNBC. Firstly, a hypothesis-led approach will characterise the role of NR3C1, which encodes the glucocorticoid receptor (GR) in TNBC as a biomarker and a therapeutic target. Secondly, a discovery-led approach will be taken by conducting a high throughput loss-of-function screen in a panel of TNBC cell lines to identify functional genes identified by the in-house analysis. We have shown that high expression of GR at a gene and protein level is associated with an improved response to anthracycline-based chemotherapy. This relationship is unique to TNBC, with no association seen in other breast cancer subtypes. Additionally, high GR expression is indicative of patients who will respond poorly to taxane-based treatments, in accordance with previous publications. These results highlight how GR could be used as a predictive biomarker to discern the most effective chemotherapy option for TNBC patients at an early stage, forgoing the need for ineffective cytotoxic treatments. This will have the consequence of improving survival and reducing treatment-associated side effects. The functional role of GR has also been interrogated using in vitro models of TNBC. We have shown that GR may be targeted to improve the response to anthracycline- and taxane-based chemotherapies in the luminal androgen receptor (LAR) subtype of TNBC. LAR TNBC accounts for 12% of all TNBCs and is characterised by androgen receptor (AR) driven growth and highlevels of chemoresistance. GR silencing using siRNA led to increased resistance to FEM (5- fluorouracil, epirubicin, mitomycin C) and docetaxel treatment in the LAR TNBC cell line, MDA-MB-453. Moreover, stimulation of GR using a synthetic agonist, dexamethasone, increased the sensitivity of this cell line to FEM chemotherapy. The GR-regulated effects on response to chemotherapy could be due to regulation of chemosensitivity mediators such as NF-κB or by the production of ROS in response to GR signalling. Given that these results were confined to the AR-expressing, LAR subtype of TNBC, it is likely that interactions between GR and AR are implicated in this process. An siRNA screen guided by genes identified from the in-house bioinformatic analysis was conducted to identify genes that regulate viability, chemosensitivity and migration of TNBC cell lines. These genes could be exploited for use as targeted treatments for TNBC. Several genes were identified which regulated the survival and response to chemotherapy of cell lines spanning multiple molecular subtypes of TNBC. One gene of interest, COPZ1 was taken forward for further characterisation. COPZ1 encodes a subunit of the coatomer complex which forms vesicles responsible for retrograde trafficking of cellular proteins from the Golgi to the endoplasmic reticulum (ER) as well as regulating autophagy and endosome maturation. COPZ1 silencing induces ER stress and cell death in multiple TNBC cell lines while having no effect on the normal breast cell line, 184A1. These effects can be attributed to cancer-specific silencing of the COPZ1 isoform, COPZ2 and can be counteracted by overexpression of COPZ2. Overexpression of COPZ1 leads to increased migration of the TNBC cell line, MDA-MB-468 highlighting its potential role in metastasis. The small molecule inhibitor of retrograde trafficking, brefeldin A led to dose dependent cytotoxicity and inhibition of migration in TNBC cells. These results highlight the potential for COPZ1 and/or retrograde trafficking to be targeted in the treatment of TNBC.

Non-Alcoholic Fatty Liver Disease (NAFLD) is a chronic and dangerous condition which has grown in prevalence over recent decades due to increasing rates of obesity, affecting roughly a quarter of adults globally. This thesis develops a useable NAFLD definition which can be applied to large retrospective data sets consisting of medical records for selected cohorts. The morbidity and mortality associated with NAFLD is investigated, especially related to extrahepatic cancer. This thesis also aims to identify genetic modifiers of NAFLD risk in Scottish and South Indian populations. Data from three retrospective Scottish cohorts with electronic health records (EHRs) were analysed in the current thesis. These were the GoDARTS, SHARE and Tayside and Fife Diabetics cohorts. Genotypic data was available for a number of patients in GoDARTS and SHARE. Data from the Dr Mohan’s Diabetes Speciality Clinic (DMDSC) were also analysed, which consisted of clinical measurements from clinic visits, and genotypic data. An accurate and practical NAFLD definition based on two raised ALT levels was developed, which had a sensitivity of 97.4% in the GoDARTS cohort. This definition was used for subsequent NAFLD analyses. Patients with NAFLD experienced significantly more hospital admissions, and earlier death than those without NAFLD. We found NAFLD is associated with increased risk of cancer incidence, and cancer death, which accounts for a large portion of the excess morbidity and mortality seen in NAFLD patients. GWAS analyses revealed that PNPLA3 rs738409 is a major genetic risk factor in both Scottish and South Indian populations. Candidate gene studies revealed variants associated with endothelin function and with GLP1 receptors had significant effects on NAFLD. This thesis applies an accurate and novel NAFLD definition to retrospective cohorts with EHRs, and found increased morbidity and mortality in individuals with this phenotype. A large proportion of this is explained by increased cancer incidence and cancer death seen in these individuals. A number of genetic risk factors for NAFLD are described, including novel loci in endothelin and GLP1R related genes.

There is a growing interest in personalised medicine where individual heterogeneity is incorporated into decision-making and treatments are tailored to individual patients or patient subgroups in order to provide better healthcare. The National Health Service Blood and Transplant (NHSBT) in England aims to move towards a more personalised service and the National Institute for Health Research Blood and Transplant Research Unit (NIHR BTRU) call has mandated research to "identify, characterise and exploit biomarkers in personalising donation strategies to maximise donor health and the blood supply". The work presented in this thesis was motivated by a large-scale, UK-based blood donation trial called INTERVAL. In INTERVAL, male donors were randomly assigned to 12-week, 10-week, and 8-week inter-donation intervals, and female donors to 16-week, 14-week, and 12-week inter-donation intervals. The outcomes of this trial include the amount of blood collected (primary), the number of low haemoglobin deferrals, and donor's quality of life. The INTERVAL trial has collected a wealth of information on individual donor characteristics, enabling us to explore (i) whether different inter-donation intervals should be recommended for donors with different characteristics (by age, blood measurements, etc.), and (ii) donor stratification schemes, for example, how to partition donors into those who have the capacity to give blood more frequently than the general donor population and those who tend to be deferred more often due to safety concerns than the average donors. One of the main statistical challenges arising from the development of personalised donation strategies using the data from the INTERVAL trial is that there are three (ordered) randomised groups for each gender in this trial, while the majority of existing statistical approaches developed in the personalised medicine context can only handle two randomised groups and thus are not directly applicable to the INTERVAL data. This thesis aims to address issues related to this added methodological complexity. We hope that the methodologies developed in this thesis can not only help us better analyse the INTERVAL data but also facilitate the analysis of other multi-arm trials in a wider range of medical applications in addition to blood donation. We begin by summarising methods that can be used to estimate the optimal individualised treatment rule (ITR) in multi-arm trials and comparing their performance in large-scale trials via simulation studies in Chapter 2. We also apply these methods to the data from male donors in the INTERVAL trial to estimate the optimal personalised donation strategies under three different objectives: (i) maximise the total units of blood collected by the blood service, (ii) minimise the low haemoglobin deferral rates, and (iii) maximise a utility score that "discounts" the total units of blood collected by the incidences of low haemoglobin deferrals. The three inter-donation intervals in the INTERVAL trial exhibit a natural ordering, and applying the ITR estimation methods that ignore the ordinality may result in suboptimal decisions. We are thus motivated to propose a method that effectively incorporates information on the ordinality of randomised groups to identify the optimal ITR in the ordinal-arm setting in Chapter 3. We further develop variable selection methods under the proposed framework to handle situations with noise covariates that are irrelevant for decision-making. Through simulation studies and an application to the data from a target donor population ("much-in-demand but vulnerable") in the INTERVAL trial, we demonstrate that the proposed method has superior performance over methods that ignore the ordinality. In Chapter 4, we switch focus to donor (or "patient" in a more general sense) stratification in multi-arm trials and develop a novel method for stratifying subjects with heterogeneous intervention responses and covariate profiles into more homogeneous subgroups using Bayesian clustering techniques. The "imputed" potential outcomes under different randomised groups are linked to subjects' baseline characteristics nonparametrically through cluster (subgroup) memberships. We examine the performance of our proposed method via simulation studies and we illustrate the utility of the method by applying it to the INTERVAL data to stratify donors based on their capacity to donate.

Glioblastoma Multiforme is the most common adult primary brain tumour, with median survival rates of around 15 months. The aggressive and heterogeneous nature of this disease means that despite multimodal treatment incorporating maximal surgical resection with radiotherapy and chemotherapy, the tumour inevitably recurs. Second-line therapeutic options are limited and so to address the need for novel options at recurrence, this project aimed to combine phenotypic screening with drug repurposing in patient-derived glioma stem-like cells to rationally select second-line chemotherapeutics based on the sensitivity of the patient's tumour cells. The first part of this project aimed to validate a proof-of-principle high content screen in vivo to elucidate whether cell culture format had an effect on drug-sensitivity. We determined through the evaluation of a sub-set of drugs that suggested three-dimensional culture of glioma stem-like cells as neurospheres correlated better with our in vivo results when compared with traditional two-dimensional culture. The second part of this project then built on the learnings taken from the high content screen and developed a reproducible, 384-well multicellular spheroid high throughput screen that identified actinomycin-D to be a potent cytotoxic agent in all of our patient-derived lines. We subsequently validated this effect in a patient-derived orthotopic xenograft model of recurrent glioblastoma, demonstrating that this drug may also have the potential to specifically target the stem-cell population of tumours through downregulation of SOX2. Thirdly, based on literature that suggested a class of amphiphilic cationic drugs, also known as Functional Inhibitors of Acid Sphingomyelinase (FIASMA), could specifically target cancer cells we incorporated loratadine, a typical FIASMA, into our high throughput screen. We demonstrate that loratadine induces glioma cell-specific death through lysosomal membrane permeabilisation, as well as validating the in vitro synergistic anti-tumour activity of afatinib and loratadine in our recurrent glioblastoma patient-derived orthotopic xenograft. Overall, this project has established a workflow that streamlines patient tumour processing, giving high quality drug sensitivity data on patient-derived cancer stem cells in a clinically actionable timeframe. The ability to predict synergistic combinations of chemotherapeutics in a short window of time is vital in glioblastoma, as many patients are found to have an increased resistance to therapy at recurrence. The screen developed in this project quickly enables the stratification of patients by phenotypic response. Through future screening of a larger cohort of patients to enrich the data set, we hope to strengthen its prognostic power and optimise this platform into an invaluable tool that can help clinicians as well as improving the outlook for many patients.

The tumour microenvironment comprises the non-cancerous cells present in the tumour mass (fibroblasts, endothelial, and immune cells), as well as signalling molecules and extracellular matrix. Tumour growth, invasion, metastasis, and response to therapy are influenced by the tumour microenvironment. Therefore, characterising the cellular and molecular components of the tumour microenvironment, and understanding how they influence tumour progression, represent a crucial aim for the success of cancer therapies. High-grade serous ovarian cancer provides an excellent opportunity to systematically study the tumour microenvironment due to its clinical presentation of advanced disseminated disease and debulking surgery being standard of care. This thesis first presents a case report of a long-term survivor (>10 years) of metastatic high-grade serous ovarian cancer who exhibited concomitant regression/progression of the metastatic lesions (5 samples). We found that progressing metastases were characterized by immune cell exclusion, whereas regressing metastases were infiltrated by CD8+ and CD4+ T cells. Through a T cell - neoepitope challenge assay we demonstrated that pre- dicted neoepitopes were recognised by the CD8+ T cells obtained from blood drawn from the patient, suggesting that regressing tumours were subjected to immune attack. Immune excluded tumours presented a higher expression of immunosuppressive Wnt signalling, while infiltrated tumours showed a higher expression of the T cell chemoattractant CXCL9 and evidence of immunoediting. These findings suggest that multiple distinct tumour immune microenvironments can co-exist within a single individual and may explain in part the hetero- geneous fates of metastatic lesions often observed in the clinic post-therapy. Second, this thesis explores the prevalence of intra-patient tumour microenvironment het- erogeneity in high-grade serous ovarian cancer at diagnosis (38 samples from 8 patients), as well as the effect of chemotherapy on the tumour microenvironment (80 paired samples from 40 patients). Whole transcriptome analysis and image-based quantification of T cells from treatment-naive tumours revealed highly prevalent variability in immune signalling and distinct immune microenvironments co-existing within the same individuals at diagnosis. ConsensusTME, a method that generates consensus immune and stromal cell gene signatures by intersecting state-of-the-art deconvolution methods that predict immune cell populations using bulk RNA data was developed. ConsensusTME improved accuracy and sensitivity of T cell and leukocyte deconvolutions in ovarian cancer samples. As previously observed in the case report, Wnt signalling expression positively correlated with immune cell exclusion. To evaluate the effect of chemotherapy on the tumour microenvironment, we compared site-matched and site-unmatched tumours before and after neoadjuvant chemotherapy. Site- matched samples showed increased cytotoxic immune activation and oligoclonal expansion of T cells after chemotherapy, unlike site-unmatched samples where heterogeneity could not be accounted for. In addition, low levels of immune activation pre-chemotherapy were found to be correlated with immune activation upon chemotherapy treatment. These results cor- roborate that the tumour-immune interface in advanced high-grade serous ovarian cancer is intrinsically heterogeneous, and that chemotherapy induces an immunogenic effect mediated by cytotoxic cells. Finally, the different deconvolution methods were benchmarked along with ConsensusTME in a pan-cancer setting by comparing deconvolution scores to DNA-based purity scores, leukocyte methylation data, and tumour infiltrating lymphocyte counts from image analysis. In so far as it has been benchmarked, unlike the other methods, ConsensusTME performs consistently among the top three methods across cancer-related benchmarks. Additionally, ConsensusTME provides a dynamic and evolvable framework that can integrate newer de- convolution tools and benchmark their performance against itself, thus generating an ever updated version. Overall, this thesis presents a systematic characterisation of the tumour microenvironment of high grade serous ovarian cancer in treatment-naive and chemotherapy treated samples, and puts forward the development of an integrative computational method for the systematic analysis of the tumour microenvironment of different tumour types using bulk RNA data.

Our arsenal of weapons to fight against bacterial infections is weakening: bacteria are gaining resistance to the common antibiotics, while industries are struggling to develop new effective ones. To avoid triggering de-novo antibiotic resistance, we need the right antibiotic for the specific bacteria, at a dose adapted to the patient genetics. Genes driving the degradation of antibiotics have indeed known genetic variants that can dramatically affect the kinetics of antibiotic metabolism from one patient to another. This could lead to treatment failure, excessive side effects or emergence of resistance. I first investigated the clinical relevance of the vancomycin-rifampicin combination to treat Methicillin-Resistant Staphylococcus aureus infections (Chapter 3). I showed in various experimental settings that these two antibiotics may promote an environment prone for antibiotic resistance. Their interaction might be unstable in vitro because of environmental factors, one could wonder how the host environment might generate such instability. I then explored how interactions between antibiotics and host xenobiotic genetics could influence antibiotic concentrations, potentially triggering increased treatment failure, side-effects and antibiotic resistance in patients carrying particular variants. In silico, I estimated the effects of genetic variants of the Cytochrome P450 3A4 gene to its enzyme, and, as they are unequally distributed in the world, their global relevance (Chapter 4). In vivo, I focused on the Carboxylesterase 2 gene and I found two of its variants, rs11075646 and rs8192925, capable of significantly altering the degradation of various drugs, including rifampicin and mycophenolate mofetil. A clinical study was designed, to explore possible correlations between genotype for these variants and treatment response in patients (Chapter 5). Altogether, this body of work highlights the prescribing importance of considering not only the strain in bacterial infections, but also the genetics of the human host. This raises a need to make sure the right antibiotics are used in practices, at doses adapted to the patients. As part of personalised medicine, checking their genotype for these biomarkers could tailor their therapy, improving recovery while avoiding antibiotic resistance.

Subtle changes in microbial populations that inhabit different areas of the human body - known as microbiomes or microbiota - can contribute to disease development, and restoring these imbalances may provide a cure. Localised and systemic diseases such as Inflammatory Bowel Disease (IBD) and depression have been linked with alterations to microbiota across the human body. Our understanding of how both diseases develop contains significant gaps, and the microbiome - described by some as our "second genome" - offers a compelling new area for knowledge discovery. This thesis aimed to advance the field of microbiome research and is an account of the work conducted whilst investigating the human gut and oral microbiome for links with IBD and depression. In this thesis, a hybrid model and aggregating ensemble feature selection (EFS) approach are applied to microbiome census data gathered from subjects with IBD. Microbial ecology techniques are applied to identify alterations to the oral microbiome in depressed subjects, and a multimodal Computational Intelligence (CI) classification paradigm known as a Super Self-Organising Map (sSOM) is applied to predict depression from a saliva sample. Finally, a rough set characterisation approach was developed and applied to gut and oral microbiome census data in depressed subjects to avoid destructive data normalisation and to enable knowledge discovery. The outcomes from the development of the hybrid model and aggregating EFS approach include the accurate non-invasive prediction of IBD, and the identification of novel and robust alterations to the gut microbiome in an adult cohort of IBD patients. The result provides a potential alternative to invasive colonoscopy, improve the time to diagnosis and treatment of IBD, and delivers new insights into the aetiology of IBD. The investigation of the oral microbiome identified novel alterations in depressed subjects for the first time. The changes to the structure and composition of the oral microbiome were significant enough to enable the accurate prediction of depression from a saliva sample. The results contribute to the microbiome-gutbrain axis theory by associating alterations to the oral microbiome with depression for the first time, and offer an alternative to subjective criteria for diagnosing depression, which currently relies on patient self-report and clinical judgement. The rough set microbiome characterisation approach replicated existing results and identified previously undescribed alterations to the gut microbiome in depressed subjects. The results provide an alternative approach to destructive normalisation techniques that are often applied to microbiome census data (identifying an optimal approach is an open research question), and contribute to our understanding of the microbiome-gut-brain axis, which could lead to psychobiotic treatments of depression in the future.

Technological advancements in mass spectrometry allowing quantification of almost complete proteomes make proteomics a key platform for generating unique functional molecular data. Furthermore, the integrative analysis of genomic and proteomic data, termed proteogenomics, has emerged as a new field revealing insights into gene expression regulation, cell signalling, and disease processes. However, the lack of software tools for high-throughput integration and unbiased modification and variant detection hinder efforts for large-scale proteogenomics studies. The main objectives of this work are to address these issues by developing and applying new software tools and data analysis methods. Firstly, I address mapping of peptide sequences to reference genomes. I introduce a novel tool for high-throughput mapping and highlight its unique features facilitating quantitative and post-translational modification mapping alongside accounting for amino acid substitutions. The performance is benchmarked. Furthermore, I offer an additional tool that permits generation of web accessible hubs of genome wide mappings. To enable unbiased identification of post-translational modifications and amino acid substitutions for high resolution mass spectrometry data, I present algorithmic updates the mass tolerant blind spectrum comparison tool ’MS SMiV’. I demonstrate the applicability of the changes by benchmarking against a published mass tolerant database search of a high resolution tandem mass spectrometry dataset. I then present the application of ‘MS SMiV’ on a panel of 50 colorectal cancer cell lines. I show that the adaption of ‘MS SMiV’ outperforms traditional sequence database based identification of single amino acid variants. Furthermore, I highlight the utility of mass tolerant spectrum matching in combination with isobaric labelled quantitative proteomics in distinguishing between post-translational modifications and amino acid variants of similar mass. In the last part of this work I integrate both tools with a high-throughput proteogenomic identification pipeline and apply it to a pilot study of chondrocytes derived from 12 osteoarthritic individuals. I show the value of this approach in identifying variation between individuals and molecular levels and highlight them with individual examples. I show that multi-plexed proteogenomics can be used to infer genotypes of individuals.

Advances in next-generation sequencing have accelerated the rate at which novel gene fusions are discovered. The discovery of gene fusions such as EML4-ALK in lung cancer and BCR-ABL1 in chronic myeloid leukaemia have already led to changes in clinical care. However, important questions remain about the role of gene fusions in promoting oncogenic phenotypes and their relevance in drug response. In this study, I combine RNA sequencing, CRISPR/Cas9 screens and high-throughput drug sensitivity data in a panel of 1,011 human cancer cell lines across 42 tissue types to examine the occurrence and functional relevance of gene fusions in cancer. Fusions were called using three algorithms and filtered to reveal 8,354 fusion events with a validation rate of 70%. Cell lines exhibit known fusions in their corresponding tissue types as well as a large number of putative passenger events and fusion recurrence across tissue types correlates with that found in patient samples. The panel of 1,011 cell lines has previously undergone high-throughput drug screening of 409 drug compounds. I implemented a systematic analysis to identify associations between fusion occurrence and drug response. It reliably recapitulates known associations (e.g. BCR-ABL1 and sensitivity to ABL inhibitors). However, the number of novel findings is low, likely due to the low numbers of recurrent fusions, a lack of prior knowledge of novel gene fusions as well as a narrow range of drug targets. Next, I developed a computational approach using whole-genome CRISPR/Cas9 screening data for 339 cell lines. It utilises CRISPR/Cas9 data on a guide-level to systematically evaluate essentiality of novel gene fusions. My analysis predicts essentiality of known gene fusions with high accuracy and provides evidence for the oncogenic relevance of novel gene fusions. A gene fusions in YAP1-MAML2 represents a particularly interesting finding showing functionality across multiple distinct cancer types. Altogether, in my thesis, I demonstrate that innovative computational approaches leveraging new datasets can enable us to elucidate the functionality of rare gene fusions in human cancer. These types of discoveries may aid in the development of targeted therapies and supports the use of clinical basket trials to capture cancer events across multiple tissue types.

The thesis provides a comprehensive overview of key issues affecting practice, policy and patients, in current efforts for colorectal cancer (CRC) disease control. The global burden of CRC is expected to increase by 60% to more than 2.2 million new cases and 1.1 million deaths by 2030. CRC incidence and mortality rates vary up to 10-fold worldwide, which is thought to reflect variation in lifestyles, especially diet. Better primary prevention, and more effective early detection, in screening and surveillance, are needed to reduce the number of patients with CRC in future1. The risk factors for CRC development include genetic, behavioural, environmental and socio-economic factors. Changes to surveillance, which offer non-invasive testing and provide primary prevention interventions represent promising opportunities to improve outcomes and personalise care in those at risk of CRC. By systematic review of the literature, I highlight the gaps in comparative effectiveness analyses of post-polypectomy surveillance. Using micro-simulation methods I assess the role of non-invasive, faecal immunochemical testing in surveillance programmes, to optimise post-polypectomy surveillance programmes, and in an accompanying sub-study, I explore the value of adding an adjunct diet and lifestyle intervention. The acceptability of such revisions is exposed to patient preference evaluation by discrete choice experiment methods. These preferences are accompanied by evidence generated from the prospective evaluation of the health literacy, numeracy, sedentary behaviour levels, body mass index (BMI) and information provision about cancer risk factors, to highlight the potential opportunities for personalisation and optimisation of surveillance. Additional analysis examines the optimisation of a screening programme facing colonoscopy constraints, highlighting the attendant potential to reduce costs and save lives within current capacity.

Patients discharged from the ICU will commonly be placed in intermediary care, such as the step-down ward, where the nurse-to-patient ratio is reduced (compared to that of the ICU). Although most of these patients will continue to recover and stabilise, a significant portion will suffer cardiac arrest and/or other clinical emergencies, and readmission into intensive care. Upon readmission, the risk of mortality is significantly higher than that of the general ICU population. Evidence suggests that early detection of deterioration may prevent or alleviate the severity of clinical emergencies. Notable shortcomings of current practices are that they (i) involve manual calculation of risk scores, (ii) depend on heuristic decision criteria, (iii) ignore time-series dynamics of physiological measurements, and (iv) lack patient-specificity. Gaussian process regression (GPR) models are proposed as a principled, probabilistic method to address the clinical need to continuously monitor patient vital-sign time-series with the flexibility to address the aforementioned weaknesses of current methods. The proposed GPR models focus on the robust forecasting of patient vital-sign time-series and early detection of patient deterioration. The primary contributions of this thesis describe how: 1. Probabilistic models may be used to identify artefactual measurements from continuously-acquired vital-sign monitoring devices. 2. GP covariance functions may be constructed and regularised for robust modelling, suitable for both patient-cohorts and personalised care. 3. GPR-based methods may quantify erratic physiological time-series and provide useful advanced warning of deterioration events. Each of the above contributions use the time-series correlation of vital-sign measurements for advantageous clinical inference.

Personalised medicine describes the practice of tailoring medical care to the individual characteristics of each patient. Fundamental to this practice is the identification of markers associated with differential treatment response. Such markers can be identified through the assessment of treatment effect modification using statistical methods. Randomised controlled trials provide the optimal setting for evaluating differential response to treatment. Due to restrictions regarding sample size, study length and ethics, observational studies are more appropriate in many circumstances, particularly for the identification of markers associated with adverse side-effects and long term response to treatments. However, the analysis of observational data raises some additional challenges. The overall aim of this thesis was to develop statistical methodology for the analysis of observational data, specifically primary care databases, to identify and evaluate markers associated with differential treatment response. Three aspects of the assessment of treatment effect modification in an observational setting were addressed. The first aspect related to the assessment of treatment effect modification on the additive measurement scale which corresponds to a comparison of absolute treatment effects across patient subgroups. Various ways in which this can be assessed in an observational setting were reviewed and a novel measure, the ratio of absolute effects, which can be calculated from certain multiplicative regression models, was proposed. The second aspect regarded the confounding adjustment and it was investigated how the presence of interactions between the moderator and confounders on both treatment receipt and outcome can bias estimates of treatment effect modification if unaccounted for using Monte Carlo simulations. It was determined that the presence of bias differed across different confounding adjustment methods and, in the majority of settings, the bias was reduced when the interactions between the moderator and confounders were accounted for in the confounding adjustment model. Thirdly, it has been proposed that patient data in observational studies be organised into and analysed as series of nested nonrandomised trials. This thesis extended this study design to evaluate predictive markers of differential treatment response and explored the benefits of this methodology for this purpose. It was suggested how absolute treatment effect estimates can be estimated and compared across patient subgroups in this setting. A dataset comprising primary care medical records of adults with rheumatoid arthritis was used throughout this thesis. Interest lay in the identification of characteristics predictive of the onset of type II diabetes associated with steroid (glucocorticoid) therapy. The analysis in this thesis suggested older age may be associated with a higher risk of steroid-associated type II diabetes, but this warrants further investigation. Overall, this thesis demonstrates how observational studies can be analysed such that accurate and meaningful conclusions are made within personalised medicine research.

A comprehensive molecular and phenotypic analysis of congenital and childhood cataractRachel L. Gillespie; The University of Manchester, Doctor of Philosophy, 2015Congenital and childhood cataract (CCC) is estimated to affect 3.5-6 per 10,000 children under 16 years in developed countries - a major cause of lifelong visual impairment. It is estimated that 25-50% of CCC cases are caused by genetic mutations. CCC demonstrates extreme heterogeneity with more than 100 associated genes, and may occur as an isolated anomaly of the eye (non-syndromic) or as a manifestation of a multisystem condition (syndromic). Limitations of conventional sequencing technologies have precluded precise genetic diagnosis and limited understanding of the epidemiological basis of the condition. Next generation sequencing (NGS) technologies have revolutionised the approach to the study of human disease. The aim of this research was to conduct a comprehensive molecular and phenotypic analysis of CCC using NGS.A disease-targeted NGS assay was designed to screen, in parallel, 115 genes associated with all forms of CCC. DNA from 36 patients, randomly selected from the study cohort, underwent cataract-targeted NGS. Putative cataract-causing variants were identified in 75% of individuals; 85% of non-syndromic patients and 63% of syndromic CCC patients. Cataract-targeted NGS was able to efficiently delineate disease sub-type and in some cases identified rare syndromic forms of the condition. These findings were envisaged to alter care and management of CCC patients demonstrating the potential clinical utility of the test. In a subset of cases, NGS identified CCC was a manifestation of an inborn error of metabolism. A number of these conditions were eligible for preventative treatment emphasizing the importance of early diagnosis. A strategic approach to the identification of novel recessive causes of CCC was also undertaken. Affected children from seven consanguineous families underwent pre-screening by cataract-targeted NGS to delineate those with mutations in known genes. Mutation negative patients underwent autozygosity-guided whole exome sequencing (WES) analysis. This strategic approach to disease gene discovery led to the identification three novel cataract-causing candidate genes, TRIM8, CCDC13 and GRWD1. It also led to the association of EIF2B2, known to cause adult-onset leukoencephalopathy with vanishing white matter (VWM) disease featuring pre-senile onset cataract, with childhood-onset cataract. This work demonstrated that cataract-targeted NGS offers an efficient and unbiased means of pre-screening, however, causation is difficult to assign to novel disease genes in the absence of experimental evidence. Correspondingly, in vitro analysis of a missense variant in HMX1 demonstrated the deleterious effect of the mutation on protein function. This work confirmed HMX1 as the cause of a rare oculoauricular phenotype and expanded the class of disease-causing mutations in this gene. In conclusion, this study has demonstrated that NGS is effective in the study of CCC and has provided a platform for future studies in to the genetic aetiology of the condition, as well as the molecular mechanisms underlying lens transparency and human development. The work adds to the increasing body of evidence that augurs an era of personalised genomic medicine in ophthalmology that will foresee improved patient outcomes attributable to the implementation of a stratified approach to medicine.

Pharmacogenetics (PGx) and personalised medicine are new disciplines that, gathering the existing knowledge about the genetic and phenotypic factors that underpin drug response, aim to deliver more targeted therapies that avoid the existing problems of adverse drug reactions or lack of drug efficacy. PGx and personalised medicine imply a shift in the way drugs are prescribed, as they require introducing diagnostic tools and implementing pre-screening mechanisms that assess patients' susceptibility to new or existing drugs. The direct benefit is an improvement in drug safety and/or efficacy. However, neither pharmacogenetics nor personalised medicine, are widely used in clinical practice. Both technologies face a number of controversies that hamper their widespread use in clinical practice. This thesis investigates the scientific; technological; social; economic; regulatory and ethical implications of PGx and personalised medicine, to understand the enablers and barriers that drive the process of technology diffusion in three conditions: autoimmune diseases, acute lymphoblastic leukaemia and non-small cell lung cancer.The thesis uses concepts of the sociology of science and a qualitative approach, to explore the arguments for and against the use of the technology by different actors (pharmaceutical and biotechnology companies, researchers, clinicians, regulators and patient organisations). The core of this analysis lies in the understanding of how, diagnostic testing (TPMT testing in the case of autoimmune diseases, acute lymphoblastic leukaemia, and EGFR testing in the case of non-small-cell lung cancer) may affect the existing drug development and service delivery mechanisms, with a particular focus on the user-producer interactions and feedback mechanisms that underpin diffusion of medical innovations and technological change in medicine.The thesis concludes by identifying gaps in knowledge and common issues among TPMT and EGFR testing, which might be used, in the future, to inform policy on how to improve PGx service delivery through a public Health System such as the NHS.

The heterogenous tumour microenvironment (TME) has been well established in the progression of cancer and patient outcome. In particular, the success of anticancer therapy is influenced by various constituents and intercommunicating networks within the TME. Drug development research often focuses on targeting cancer cells alone without considering all the cellular and non-cellular components of the TME, which plays a crucial role in chemoresistance and disease relapse. Thus, there is a necessity to understand the individual features of the TME in order to deliver a more personalised therapy that improves the rate of successful outcomes while reducing the recurrence of the disease. However, the mechanisms of acquired drug resistance and pharmacokinetics within the tumour still remains elusive. Currently, there is a lack of high-throughput methods suitable to study the functional effects of anticancer drugs on tumour samples where the interplay between cancer cells and the TME remains intact. This thesis embodies the work in the development of the ALTEN (Alginate-based Tissue Engineering) platform, a biomimetic hydrogel system for rapid functional testing of anticancer drugs in explanted tumours. The alginate hydrogels provides a 3D scaffold that resembles the native extracellular matrix (ECM) that preserve the original characteristics of the tumour to permit detailed assessment of the TME and the complex molecular interactions between cell species. Here, we use high-resolution single-cell RNA-seq technologies to analyse the molecular effects of anticancer treatments within the tumour to study the mechanisms of acquired drug resistance. The combination of ALTEN and scRNA-seq technology enable high-throughput and high-resolution screening of tumour explants exposed to cytotoxic agents and immunomodulators. The impact of drug therapies within an intact TME could be evaluated to 1) determine treatment efficacy and 2) characterise drug resistant cancer populations. In summary, the ALTEN platform provides an innovative avenue allowing engineering of patient tumour explants to investigate the efficacy of therapies in a high-throughput manner and provide preclinical information on optimising treatments. The integration of both the high-fidelity drug testing capacity of ALTEN and the high-resolution molecular phenotyping scRNA-seq aims to assist oncologists in providing personalised therapies to improve patient outcome.

Pancreatic cancer has a 5-year survival of only 8%, and persists as the 4th most common cause of cancer-related death in Western societies. A more tailored treatment approach may be beneficial as the current standard-of-care therapies offer only a modest increase in overall patient survival. Recent large-scale genomic studies have revealed that the SRC/JAK/STAT3 signalling pathway is deregulated in up to 35% of pancreatic cancers, and is yet to be systematically examined in this disease. Consequently, we hypothesised that targeting pancreatic tumours with alterations in the SRC/JAK/STAT3 signalling pathway with JAK and SRC inhibitors represents a promising novel therapeutic strategy for this disease. In this thesis, we use well-annotated patient-derived cell-line models (ICGC), along with cell-lines generated from the aggressive KPC mouse model, to show that the combination of selected JAK and SRC inhibitors is synergistic in cell lines characterised by high phospho-STAT3 expression and P53 mutations. Using 3D in vitro models, including organotypic and organoid models, we show that this therapeutic strategy inhibits the invasive and proliferative capacity of tumour cells, disrupts collagen remodelling and extracellular matrix integrity, interferes with paracrine signalling and has strong immunomodulatory effects. Lastly, we examine the in vivo efficacy of dasatinib and ruxolitinib using a syngeneic KPC mouse model, as well as patient-derived pancreatic tumour models, characterised by high phospho-STAT3 expression and P53 mutations. From these studies we demonstrate that the combination of dasatinib and ruxolitinib significantly inhibited tumour progression, improved survival, delayed the development of metastasis and significantly improved response to standard of care chemotherapy. Furthermore, tumours treated with dasatinib and ruxolitinib displayed decreased collagen deposition and remodelling, and altered immune cell infiltration in the syngeneic setting. Our findings demonstrate the potential for tailored therapeutic strategies involving SRC/JAK/STAT3 inhibition in pancreatic cancer, and suggest that therapeutic efficacy may be the result of targeting both tumour cells and the tumour microenvironment, by decreasing fibrosis and overcoming tumour-induced immunosuppression.

The emerging standard of care for patients with inoperable pancreatic cancer is a combination of cytotoxic drugs gemcitabine and Abraxane, but patient response remains moderate. Pancreatic cancer development and metastasis occur in complex settings with reciprocal feedback from micro-environmental cues influencing both disease progression and drug response. Little is known about how sequential dual targeting of tumour tissue tension and vasculature before chemotherapy can impact tumour response. We used intravital imaging to assess how transient manipulation of the tumour tissue, or ‘priming’, using the pharmaceutical Rho-kinase inhibitor Fasudil affects response to chemotherapy. Intravital FRET imaging of a CDK1 biosensor to monitor the efficacy of cytotoxic drugs revealed that priming improves pancreatic cancer response to gemcitabine/Abraxane at both primary and secondary sites. Transient priming also sensitised cells to shear stress and impaired colonisation efficiency and fibrotic niche remodelling within the liver, three important features of cancer spread. Last, we demonstrate a graded response to priming in stratified patient-derived tumours, indicating that fine-tuned tissue manipulation before chemotherapy may offer opportunities in both primary and metastatic targeting of pancreatic cancer.

Pancreas cancer remains a disease with a very poor prognosis. Most patients present with advanced disease and survive only a few months. Distinct phenotypes have been difficult to elucidate and the value of doing so to direct therapy has been minimal because of the limited treatment options available. Through her role in the International Cancer Genome Consortium to sequence the cancer genome, the candidate was involved in deep sequencing of more than 300 cases of pancreas cancer. Cataloguing the pancreas cancer genome revealed that pancreas cancer is a heterogeneous disease with a number of potentially actionable subtypes. We were able to subdivide them into: 1) those with biomarkers that would enrich for response to current therapy for pancreas cancer; 2) those amenable to treatment with currently available targeted therapies with efficacy in other cancers; and 3) those potentially druggable with emerging drugs. Seeing the enormous potential for improving the outcome for pancreas cancer patients, and exploiting her leadership position in the Australasian Gastrointestinal Trials Group, she designed and implemented IMPaCT, a precision medicine clinical trial for metastatic pancreas cancer. The candidate has led a large team of scientists and trialists to apply next generation sequencing technology to patients who present with metastatic pancreas cancer. The trial is open and actively recruiting at 3 sites in NSW and so far more than 90 patients have been screened for enrolment on the study and 22 eligible candidates have been found. Additionally, we identified molecular mechanisms of importance in pancreatic cancer including axon guidance. ROBO/SLIT protein expression levels were examined and analysed in terms of patterns of beta-catenin expression on a cohort of >200 cases of pancreatic cancer. The results highlight the role of the Wnt signalling pathway in pancreas cancer and demonstrate an association with prognosis and also a link with smoking, which may prove to be important as we further elucidate this pathway. Instead of a ‘one size fits all’ approach, this candidate’s work demonstrates that knowledge of tumour biology can lead to a tailored approach that is urgently needed to improve outcomes for patients with pancreas cancer.

Defining phenotypes of prognosis and therapeutic responsiveness is the basis of modern oncology and underpins personalised medicine. Significant improvements in outcomes have been achieved in some cancer types, such as breast and colon, through defining specific phenotypes. Pancreatic cancer (PC) is the fourth leading cause of cancer death in western societies, and as yet has no well-defined phenotypes. This thesis aims to define molecular biomarkers with clinical utility that predict prognosis and therapeutic responsiveness in PC. A systematic approach for biomarker discovery and development based on NCI guideline was applied. First, multiple large independent cohorts of patients with resectable PC were analysed and prognostic factors important to clinical decision making were extensively characterised as the basis for biomarker studies. One key findings demonstrated that surgical margin clearance of more than 1.5mm is important for long-term survival. The results added to our current understanding of microscopic margin involvement definitions and potentially identifies a subgroup of patients who may benefit from aggressive loco-regional therapy such as adjuvant radiotherapy. Aberrant expression of two candidate biomarkers (S100A4 and S100A2 calcium-binding proteins) with potential roles in metastatic processes were identified as independent prognostic factors using these multiple patient cohorts. Differential expression of these two biomarkers categorised patients with resected PC into three distinct prognostic groups. A pre-operative molecular prognostic nomogram was developed using these two biomarkers and tumour size. These markers predicted survival as well as the published post-operative nomogram and can potentially be tested pre-operatively. To implement this strategy, a proof-of-principle study demonstrated that biomarker expression can be reliably assessed in pre-operative EUS-FNAB samples. S100A4 expression was then assessed for its contribution in chemoresistance in clinical cohorts based on previous functional data. S100A4 was identified as a predictive biomarker of adjuvant chemotherapeutic responsiveness in two large cohorts of patients. In addition, the candidate made significant contribution to the Australian Pancreatic Cancer Genome Initiative (APGI), a part of the International Cancer Genome Consortium. The genomic mutational landscape of resected PC using exome sequencing as part of the APGI project was also presented.