Acute myeloid leukaemia (AML) and ovarian cancer (OVC) are two difficult to treat cancers. AML is often treatable however minimal residual disease (MRD) endures such that many patients who achieve remission eventually relapse and succumb to the disease. OVC affects approximately 7000 women in the U.K. every year. It can occur at any age but is most common after menopause. Diagnosis at an early stage of disease greatly improves the chances of survival however, patients tend to be diagnosed in the later stages of disease when treatment is often less effective. Immunotherapy has the potential to reduce MRD and delay or prevent relapse. In order for immunotherapy to work, tumour antigens need to be identified and characterised so they can be effectively targeted. Personalised treatments require the identification of biomarkers, for disease detection and confirmation, as well as to provide an indication of best treatment and the prediction of survival. PASD1 has been found to be frequently expressed in haematological malignancies and I wanted to determine if there was a correlation between the presence of antigen-specific T cells in the periphery of patients with AML and PASD1 protein expression in the leukaemic cells. The expression of other leukaemia antigens were concurrently examined as comparators. I performed RT-PCR on nine antigens and immunocytochemistry on PASD1 in 18 samples from AML patients. I found a correlation between PASD1 expression in AML samples and the presence of PASD1-specific T cells as detected on the pMHC array. OVC lacks suitable targets for immunotherapy with few CTAs having been identified. I examined the expression of SSX2IP and the CTAs PASD1 and SSX2 in OVC. I compared the protein expression of these known tumour antigens to the “gold standard” biomarker for the diagnosis of OVC, CA125 and two other proteins known to be promising in the diagnosis of OVC, HE4 and WT1. I analysed commercially available paraffin-embedded OVC multiple tissue arrays (MTAs) containing 191 samples, predominantly stage I (n= 166), II (n= 15) and III (n= 6) OVC as well as healthy donor (n= 8) and normal adjacent tissues (n= 8). Scoring was performed in a single blinded fashion. I found SSX2A to be expressed at a score level of 3 with a frequency (37/191) that exceeded that of CA125 (14/191), HE4 (14/191), WT1 (1//191) or PASD1 (0/191). To confirm this expression I used two additional commercially-available antibodies that recognise the region common to SSX2A and B, and an antibody specific for SSX2A. Using SSX2 peptides, I blocked the immunolabelling of SSX2 in SSX2-positive cell lines showing that the immunolabelling of SSX2 and SSX2A was specific. I demonstrated that the expression of SSX2 and specifically SSX2A was reproducible and restricted to ovarian cancer with little or no expression in endometrial tissues, or diseased or inflamed endometrial tissue. In summary, these studies demonstrated that PASD1 expression in leukaemia cells correlated with the presence of PASD1-specific T cells in the periphery of presentation AML patients. I have shown that PASD1 specific-T cells are present in AML patients at diagnosis and that immunotherapy targeting PASD1 could be used to break tolerance and clear residual leukaemia cells during first remission. Analysis of the expression of three antigens in OVC, identified the specific expression of SSX2, in particular SSX2A in OVC but not healthy or diseased endometrial tissues. The expression of SSX2A was more frequent and more specific to OVC, than HE4 and WT1, and more frequent at higher intensity, especially in early stage OVC, than CA125. SSX2 and explicitly SSX2A requires further investigation to determine whether the high level of background at score 2 can be reduced with better blocking of non-specific sites. This may require the use of different SSX2 antibodies or an improved staining protocol.

Cervical cancer still represents the fourth most common cause of cancer deaths in women worldwide. Human papilloma virus (HPV) infection plays a role in cervical carcinoma initiation, but other genomic changes are needed for pre-malignant abnormalities to fully develop to cancer. This often happens through genomic instability caused by the virus oncoproteins. Several integrative genomic analysis studies have found that one of the most common imbalances in cervical squamous cell carcinoma (SCC) is copy number gain and amplification of chromosome 5p. In this region, copy number gain of the OSMR gene was found to correlate significantly with adverse outcome independent of the tumour stage (p=0.046). Furthermore, this copy number gain correlated with Oncostatin M receptor (OSMR) overexpression and sensitised these cells to Oncostatin M (OSM) leading to increased Signal transducer and activator of transcription 3 (STAT3) phosphorylation, cell migration, invasion and proangiogenic signalling. The aim of this PhD project was to study the role of OSMR overexpression in the SCC tumour microenvironment (TME) and tumour growth in vivo and to study the role of hypoxia inducible factor driven hypoxia signalling in OSMR overexpressing SCC cells and their tumour microenvironment. OSMR overexpression was found to sensitise tumour cells to induce Hypoxia inducible factor 1a and 2a (HIF1a, HIF2a) signalling in normoxic conditions, to promote pro-angiogenic signalling. Furthermore, hypoxic conditions were found to enhance OSM signalling in OSMR overexpressing cells leading to increased expression of markers of epithelial to mesenchymal transition, angiogenesis and migration. In the SCC tumour microenvironment, OSMR overexpression was found to sensitise tumour cells to OSM secreted from macrophages and other immune cells leading to improved tumour growth, angiogenesis and STAT3 activation at the tumour site. Removal of OSMR from either tumour cells or tumour microenvironment led to reduced tumour growth and angiogenesis, along with increased tumour necrosis. I conclude that OSMR overexpression is an important driver of SCC tumour progression and malignancy via STAT3- and HIF-driven signalling and removal of it from either tumour cells or tumour microenvironment drastically hampers tumour growth in vivo. Based on the results of this study, OSMR blockade is a potential novel therapeutic option in advanced SCC.

Tumour tissue is characterised by fluctuating oxygen concentrations, decreased nutrient supply, and acidic pH. Angiogenic signalling pathways that drive a certain metabolic 'configuration' may give endothelial cells a selective advantage in the tumour environment. Previously it has been shown that glycolysis drives proliferation, migration and tip cell formation during sprouting of endothelial cells (De Bock, Georgiadou et al. 2013), and is increased by VEGFA. DLL4-NOTCH has been shown to limit angiogenesis and slow down proliferation of endothelial cells, and promote stalk cell formation during angiogenic sprouting, leading to sprout elongation. DLL4-NOTCH is implicated in tumour angiogenesis, and its overexpression is a potential mechanism of resistance to anti-VEGFA therapy (Li, Sainson et al. 2011). This thesis aimed at investigating the effect of the DLL4-NOTCH signalling pathway on endothelial metabolism and its implications in angiogenesis. Firstly, it was found that DLL4-NOTCH decreases the glycolytic rate and mitochondrial respiratory parameters in endothelial cells. When given exogenous fatty acids, DLL4-NOTCH activation caused increased fatty acid uptake, storage and oxidation. This shows that the induction of DLL4-NOTCH signalling results in increased fatty acid utilisation. Secondly, this research identified fatty acid oxidation as a target metabolism pathway for angiogenic therapy. More specifically, inhibition of fatty acid oxidation decreased proliferation of endothelial cells, decreased sprout elongation in the sprouting assay, and decreased sprouting from the axial vein in the zebrafish model. ATP production was not affected. Therefore, it was hypothesised that DLL4-NOTCH activation promotes and maintains the stalk cell phenotype through an increase of fatty acid oxidation, thereby promoting biomass production for endothelial cell proliferation and growth during angiogenic sprout elongation. Thirdly, a key fatty acid metabolism gene, fatty acid binding protein 4 (FABP4), was identified, that is positively regulated by NOTCH at its promoter region. FABP4 is a candidate for mediating increased fatty acid flux in endothelial cells in response to DLL4-NOTCH. This study shows that FABP4 is induced by VEGFA in a manner dependent on DLL4-NOTCH, and the insulin-responsive transcription factor FOXO1 was required for FABP4 expression in response to DLL4-NOTCH. FABP4 is pro-angiogenic and implicated in tumour angiogenesis in ovarian cancer omental metastasis. Taken together, this study shows for the first time that DLL4-NOTCH signalling increases FABP4 induction, contributing to a key pro-angiogenic pathway, and also fatty acid utilisation in endothelial cells, and thereby contributes to the formation of blood vessels.

Lactate Dehydrogenase-A (LDH-A) is up-regulated in a broad array of cancers and is associated with poor prognosis. Involved in the hypoxic response, LDH-A is a HIF-1 target and is responsible for the enzymatic reduction of pyruvate to lactate. This is important for several reasons, chiefly (1) the regeneration of NAD+ which feeds back into earlier glycolytic stages and (2) the depletion of intracellular pyruvate concentrations. High intracellular pyruvate is known to inhibit HDACs and is associated with increased apoptosis. LDH-A is also known to be controlled by oncogenes such as c-Myc suggesting an oncogenic role. Studies have shown that the knock-out of LDH-A reduces proliferation and tumourgenicity, and stimulates the mitochondria. This thesis therefore had three aims: firstly, to validate LDH-A inhibition and elucidate its full nature in terms of the implications for tumour survival; secondly, to ascertain the role of LDH-B in order to determine whether selectivity towards LDH-A would be a necessary feature of any small molecule; lastly, to recapitulate siRNA mediated LDH-A inhibition with small molecule inhibitors that had the potential for clinical application. The thesis examined both clinical data and a broad panel of cultured cancer cell types in order to select appropriate model in which to validate siRNA mediated inhibition of LDH-A and LDH-B. After it was demonstrated that LDH-A inhibition reduced the growth of cultured cells, a range of techniques were used to quantify this reduced growth in terms of cell death and changes in metabolism. Further to this, literature studies had proposed a role for LDH-B in maintaining lactate fuelled tumour growth; however, this thesis shows that in the cell lines studied, lactate-fuelled tumour growth was an LDH-A dependent phenomenon. Finally, a high throughput assay system was designed and validated and a library of small molecules was selected, synthesized, and screened in order to identify selective inhibitors of LDH-A.

Intracellular redox potential (IRP) is a measure of how oxidising or reducing the environment is within a cell. It is a function of numerous factors including redox couples, antioxidant enzymes and reactive oxygen species. Disruption of the tightly regulated redox status has been linked to the initiation and progression of cancer. However, there is very limited knowledge about the quantitative nature of the redox potential and pH gradients that exist in cancer tumour models. Multicellular tumour spheroids (MTS) are three-dimensional cell cultures that possess their own microenvironments, similar to those found in tumours. From the necrotic core to the outer proliferating layer there exist gradients of oxygen, lactate, pH and drug penetration. Tumours also have inadequate vasculature resulting in a state of hypoxia. Hypoxia is a key player in metabolic dysregulation but can also provide cells with resistance against cancer treatments, particularly chemotherapy and radiation therapy. The primary hypoxia regulators are HIFs (Hypoxia Inducible Factors) which under low O2 conditions bind a hypoxia response element, inhibiting oxidative phosphorylation and upregulating glycolysis which has two significant implications: the first is an increase in levels of NADPH/NADH, the main electron donors found in cells which impacts the redox state, whilst the second is a decrease in intracellular pH (pHi) because of increased lactate production. Thus, redox state and intracellular pHi can be used as indicators of metabolic changes within 3D cultures and provide insight into cellular response to therapy. Surface-Enhanced Raman Spectroscopy (SERS) provides a real-time, high resolution method of measuring pHi and IRP in cell culture. It allows for quick and potentially portable analysis of MTS, providing a new platform for monitoring response to drugs and therapy in an unobtrusive manner. Redox and pH-active probes functionalised to Au nanoshells were readily taken up by prostate cancer cell lines and predominantly found to localise in the cytosol. These probes were characterised by density functional theory and spectroelectrochemistry, and their in vitro behaviour modelled by the chemical induction of oxidative and reductive stress. Next, targeting nanosensors to different zones of the MTS allowed for spatial quantification of redox state and pHi throughout the structure and the ability to map the effects of drug treatments on MTS redox biology. The magnitude of the potential gradient can be quantified as free energy (ΔG) and used as a measurement of MTS viability. Treatment of PC3 MTS with staurosporine, an apoptosis inducer, was accompanied by a decrease in free energy gradients over time, whereas treatment of MTS with cisplatin, a drug to which they are resistant, showed an increase in viability indicating a compensatory mechanism and hence resistance. Finally, using this technique the effects of ionising radiation on IRP and pHi in the tumour model was explored. Following exposure to a range of doses of x-ray radiation, as well as single and multi-fractionated regimes, IRP and pHi were measured and MTS viability assessed. Increased radiation dosage diminished the potential gradient across the MTS and decreased viability. Similarly, fractionation of a single large dose was found to enhance MTS death. This novel SERS approach therefore has the potential to not only be used as a mode of drug screening and tool for drug development, but also for pre-clinical characterisation of tumours enabling clinicians to optimise radiation regimes in a patient-specific manner.

Denna rapport syftar till att implementera en metod för att simulera cancerceller och skapa en ökad förståelse för hur Warburgeffekten, vilket är cancercellers användning av anaerob metabolism under aeroba förhållanden, påverkar cancerceller. Detta undersöks genom att simulera i en dator hur syrehalten påverkar andelen anaeroba cancerceller i en tumör och dess spridning. I studien undersöks fem olika syrenivåer. Simuleringen görs med en Cellular Automaton-modell och startar med ett mindre antal cancerceller i mitten av ett 200x200-rutnät, omgivna av friska celler. Cancercellerna och deras beslutsmekanismer modelleras med artificiella neurala nätverk och friska celler med fastställda regler. Cancercellerna kan vid delning muteras och ge upphov till nya beteenden som sedan blir en del av selektionsprocessen. Simuleringarna visar att cancercellerna, oberoende av syrehalten, sprider sig på ett likartat vis. Genom att vissa av cancercellerna övergår från aerob till anaerob metabolism så försurar cancertumören sin omgivning, vilket dödar friska celler. Syrehaltens påverkan på andelen anaeroba celler hos tumören visar sig ha betydelse, men det är främst hos den lägsta syrehalten en markant ökning av andelen anaeroba celler noteras. Noterbart är även att andelen anaeroba celler i den här studien, för alla syrehalter, är avsevärt lägre än de 60 % som påvisats i vissa studier av Warburgeffekten gjorda på levande celler.

At the forefront of cancer treatment development and evaluation, three-dimensional Tumour Spheroid Control Experiments play a pivotal role in the battle against cancer. Conducting and evaluating in vitro experiments are time-consuming processes. This thesis details the development, implementation, and validation of an algorithmic model that classifies spheroids as either controlled or relapsed by assessing the success of their treatments based on criteria rooted in biological insights. The introduction of this model is crucial for biologists to accurately and efficiently predict treatment efficacy in 3D in vitro experiments. The motivation for this research is driven by the need to improve the objectivity and efficiency of treatment outcome evaluations, which have traditionally depended on manual and subjective assessments by biologists. The research involved creating a comprehensive dataset from multiple 60-day in vitro experiments by combining data from various sources, focusing on the growth dynamics of tumour spheroids subjected to different treatment regimens. Through preprocessing and analysis, growth characteristics were extracted and utilized as input features for the model. A feature selection and optimization technique was applied to refine the software model and improve its predictive accuracy. The model is based on a handful of comprehensive criteria, calibrated by employing a grid search mechanism for hyperparameter tuning to optimize accuracy. The validation process, conducted via independent test sets, confirmed the model’s capability to predict treatment outcomes with a high degree of reliability and an accuracy of about 99%. The findings reveal that algorithmic classification models can make a significant contribution to the standardization and automation of treatment efficacy assessment in tumour spheroid experiments. Not only does this approach reduce the potential for human error and variability, but it also provides a scalable and objective means of evaluating treatment outcomes.:1 Introduction 1.1 Background and Motivation 1.2 Biological Background 1.3 Iteration Methodology 1.4 Objective of the Thesis 2 Definition of basic Notation and Concepts 2.1 Time Series Analysis 2.2 Linear Interpolation 2.3 Simple Exponential Smoothing 2.4 Volume of a Spheroid 2.5 Heavyside Function 2.6 Least Squares Method 2.7 Linear Regression 2.8 Exponential Approximation 2.9 Grid Search 2.10 Binary Regression 2.11 Pearson Correlation Coefficient 3 Observation Data 3.1 General Overview 3.1.1 Structure of the Data 3.1.2 Procedure of Data Processing using 3D-Analysis 3.2 Data Engineering 3.2.1 Data Consolidation and Sanitization 3.2.2 Extension and Interpolation 3.2.3 Variance Reduction 4 Model Development 4.1 Modeling of Various Classification-Relevant Aspects 4.1.1 Primary Criteria 4.1.2 Secondary Criteria 4.1.3 Statistical Learning Approaches 4.2 Day of Relapse Estimation 4.3 Model Implementation 4.3.1 Combination of Approaches 4.3.2 Implementation in Python 4.4 Model Calibration 4.4.1 Consecutive Growth 4.4.2 Quintupling 4.4.3 Secondary Criteria 4.4.4 Combined Approach 5 Model Testing 5.1 Evaluation Methods 5.1.1 Applying the Model to New Data 5.1.2 Spheroid Control Probability 5.1.3 Kaplan-Meier Survival Analysis 5.1.4 Analysis of Classification Mismatches 5.2 Model Benchmark 5.2.1 Comparison to Human Raters 5.2.2 Comparison to Binary Regression Model 5.3 Robustness 5.3.1 Test using different Segmentation 5.3.2 Feature Reduction 5.3.3 Sensitivity 5.3.4 Calibration Templates 6 Discussion 6.1 Practical Application Opportunities 6.2 Evaluation of the Algorithmic Model 6.3 Limitations 7 Conclusion 7.1 Summary 7.2 Future Research Directions / Dreidimensionale Experimente zur Kontrolle von Tumorsphäroiden sind zentral für die Entwicklung und Evaluierung von Krebstherapien. Die Durchführung und Auswertung von In-vitro-Experimenten ist jedoch zeitaufwendig. Diese Arbeit beschreibt die Entwicklung, Implementierung und Validierung eines algorithmischen Modells zur Einstufung von Sphäroiden als kontrolliert oder rezidivierend. Das Modell bewertet den Behandlungserfolg anhand biologisch fundierter Kriterien. Diese Innovation ist entscheidend für die präzise und effiziente Vorhersage der Wirksamkeit von Behandlungen in 3D-In-vitro-Experimenten und zielt darauf ab, die Objektivität und Effizienz der Beurteilung von Behandlungsergebnissen zu verbessern, die traditionell von manuellen, subjektiven Einschätzungen der Biologen abhängen. Die Forschung umfasste die Erstellung eines umfassenden Datensatzes aus mehreren 60-tägigen In-vitro-Experimenten, bei denen die Wachstumsdynamik von Tumorsphäroiden unter verschiedenen Behandlungsschemata untersucht wurde. Durch Vorverarbeitung und Analyse wurden Wachstumscharakteristika extrahiert und als Eingangsmerkmale für das Modell verwendet. Das Modell basiert auf wenigen umfassenden Kriterien, die mithilfe eines Gittersuchmechanismus zur Abstimmung der Hyperparameter kalibriert wurden, um die Genauigkeit zu optimieren. Der Validierungsprozess bestätigte die Fähigkeit des Modells, Behandlungsergebnisse mit hoher Zuverlässigkeit und einer Genauigkeit von etwa 99 % vorherzusagen. Die Ergebnisse zeigen, dass algorithmische Klassifizierungsmodelle einen wesentlichen Beitrag zur Standardisierung und Automatisierung der Bewertung der Behandlungseffektivität in Tumorsphäroid-Experimenten leisten können. Dieser Ansatz verringert nicht nur das Potenzial für menschliche Fehler und Schwankungen, sondern bietet auch ein skalierbares und objektives Mittel zur Bewertung von Behandlungsergebnissen.:1 Introduction 1.1 Background and Motivation 1.2 Biological Background 1.3 Iteration Methodology 1.4 Objective of the Thesis 2 Definition of basic Notation and Concepts 2.1 Time Series Analysis 2.2 Linear Interpolation 2.3 Simple Exponential Smoothing 2.4 Volume of a Spheroid 2.5 Heavyside Function 2.6 Least Squares Method 2.7 Linear Regression 2.8 Exponential Approximation 2.9 Grid Search 2.10 Binary Regression 2.11 Pearson Correlation Coefficient 3 Observation Data 3.1 General Overview 3.1.1 Structure of the Data 3.1.2 Procedure of Data Processing using 3D-Analysis 3.2 Data Engineering 3.2.1 Data Consolidation and Sanitization 3.2.2 Extension and Interpolation 3.2.3 Variance Reduction 4 Model Development 4.1 Modeling of Various Classification-Relevant Aspects 4.1.1 Primary Criteria 4.1.2 Secondary Criteria 4.1.3 Statistical Learning Approaches 4.2 Day of Relapse Estimation 4.3 Model Implementation 4.3.1 Combination of Approaches 4.3.2 Implementation in Python 4.4 Model Calibration 4.4.1 Consecutive Growth 4.4.2 Quintupling 4.4.3 Secondary Criteria 4.4.4 Combined Approach 5 Model Testing 5.1 Evaluation Methods 5.1.1 Applying the Model to New Data 5.1.2 Spheroid Control Probability 5.1.3 Kaplan-Meier Survival Analysis 5.1.4 Analysis of Classification Mismatches 5.2 Model Benchmark 5.2.1 Comparison to Human Raters 5.2.2 Comparison to Binary Regression Model 5.3 Robustness 5.3.1 Test using different Segmentation 5.3.2 Feature Reduction 5.3.3 Sensitivity 5.3.4 Calibration Templates 6 Discussion 6.1 Practical Application Opportunities 6.2 Evaluation of the Algorithmic Model 6.3 Limitations 7 Conclusion 7.1 Summary 7.2 Future Research Directions