Glioblastoma, a WHO grade IV primary brain tumour, remains as one of the most aggressive forms of human cancer. Despite intensive research efforts into understanding the key drivers of tumour progression, few therapeutic advances have been made, with the current standard of care (the Stupp protocol) remaining unchanged for 15 years. The overall improvement to glioblastoma survival in the real-world population has been attributed to the use of the Stupp protocol, yet evidence suggests that survival outcomes were already significantly improving in the years prior to the introduction of this standard of care questioning the overall veracity of this claim. Using the Surveillance, Epidemiology and End Results (SEER) registry data we analysed the survival outcomes for real-world glioblastoma patients diagnosed from 2000 – 2016. Our findings show a consistent incremental survival improvement that preceded the introduction of the Stupp protocol and continued to increase at the same rate till 2009, stagnating afterwards. Significantly, however, this survival improvement is short-term for patients, with no survival improvement observed in patients surviving more than 2 years. Additionally, with the exception of complete tumour resection, all treatment modalities did not improve survival beyond 2 years for glioblastoma. These findings highlight the clinical stagnation of glioblastoma treatment and highlight the inability of current treatments to target the underlying causes of tumour progression. Following the introduction of the Stupp protocol attempts to develop new treatment options have universally been disappointing with a close to 0% success rate for over 1000 phase II and above clinical trials. Conflictingly, many of the therapeutic agents tested have shown promising results in preclinical trials. Current preclinical models, however, test therapies against the primary tumour, which does not recapitulate the biology or targets of tumour recurrence. We therefore developed a highly sensitive luciferase-based glioblastoma mouse model capable of single cell detection in mouse tissue. Analysis of mouse brain tissue implanted with luciferase-labelled human glioblastoma U87MG or MU20 tumours revealed the presence of tumour cells ubiquitously spread across the supratentorial regions of the brain, and distally located from the primary tumour. These tumour cells were observed as single cells in U87MG implanted mice and clusters in MU20 glioblastoma cells. Remarkably, U87MG tumours did not exhibit invasive margins and were contained within an expansive growth phenotype, suggesting invasion-independent dissemination. Our model is consistent with reports of glioblastoma as a systemic brain disease and is capable of sensitive detection of disseminated tumour cells, a model of recurrence potential. Furthermore, this model can be utilised to investigate new mechanisms of glioblastoma infiltration. Targeting aberrant angiogenesis in glioblastoma has been the major focus for glioblastoma treatment since the Stupp protocol. Yet after over a decade of basic research and clinical trials, antiangiogenic inhibitors have failed to translate into improved patient outcome. The discovery of abnormalities in the tumour vasculature suggest that there may be alternate mechanisms driving tumour progression. Vasculogenic mimicry has been observed in glioblastoma and presents as a novel aspect of tumour biology, yet the mechanisms and functional relevance of these structures remain unknown. Our study has confirmed the ability of some glioblastoma cell lines to undergo endothelialisation, forming lattice structures similar to endothelial cells when seeded onto Matrigel in vitro. One lattice forming cell line, U87MG, was also found incorporate into the tumour vasculature in an in vivo orthotopic mouse model. This behaviour was found to be regulated by an expanded TGF-beta-ALK1-Smad1/5 signalling pathway. In vivo inhibition of the Smad1/5 signalling pathway via intracranial treatment with Ad-Smad6 resulted in reduced endothelialisation in the tumour vasculature and inhibited whole brain infiltration in the U87MG mouse model. Since U87MG xenograft tumours are non-invasive, these results suggest that endothelialisation may lead to haematogenous dissemination and distal brain infiltration, providing a novel mechanism for glioblastoma progression.

Glioma is the most common intracranial malignant cancer despite the rarity. Glioblastoma (GBM, grade IV) has very dismal outcomes where fewer than 20% of patients survive beyond 5 years. Low Grade Glioma (LGG, grade II), on the other hand, frequently occurs in a younger population with a variable outcome that depends on their individual genetic alterations. A tissue-based biopsy requires an invasive operation for every glioma patient and it is the gold standard diagnosis. In monitoring the patients for recurrence, MRI has the evitable limitation that it could not differentiate pseudo-progression after irradiation treatment for glioma patients. Therefore, a non-invasive biomarker, in addition to tissue biopsy and MRI, is urgently needed to provide more precise cancer care and management with glioma patients. In fact, cancer cells including glioma cells, release microRNA (miRNA) into the microenvironment and peripheral circulation, which allows physicians and scientists the ability to detect the disease status by a blood test. In addition, microRNA (miRNA) displays accuracy in diagnosing and monitoring other cancer patients by analysis of miRNA abundance. To investigate the diagnostic capacity of circulating miRNA in glioma, a meta-analysis of previous relevant studies was performed. Circulating miRNA provided an over 90% accuracy of diagnosing glioma patients from healthy controls. A bioinformatic analysis revealed seven IDH1 mutation-associated miRNAs and these miRNA signatures were down-regulated in IDH1 mutant glioblastoma due to the hypermethylation in their promoter areas. Our proof-of-concept cohort recruited 91 glioma patients and 17 healthy controls and new circulating miRNAs were identified with remarkable diagnostic capacity. However, no circulating miRNA showed an association with IDH1 mutation in our cohort. As most circulating miRNAs were packaged and delivered by extracellular vesicles (EVs), serial cell experiments were conducted to examine the function of these miRNA-enriched EVs. Glioma Stem Cells-derived EVs significantly promoted glioma cell proliferation, cell migration and radiation resistance. This effect was largely caused by the degradation of PTEN and activation of AKT in the recipient cells through miRNAs. Collectively, these works established circulating miRNA as a clinical biomarker where it provides a complementary tool for glioma management. These results also formed the basis of future research into the role of EVs in glioma and highlighted the potential therapeutic target of EVs to glioma.

Glioblastoma is the most common and aggressive form of malignant glioma. Currently, despite treatment with surgery followed by radiotherapy and the chemotherapeutic agent temozolomide (TMZ), mean patient survival time is approximately 12 months and the 5-year survival rate is close to 0%. A key factor for the dismal prognosis is tumour recurrence post-treatment which is largely due to: 1) the infiltrative nature of glioblastoma rendering complete resection impossible and 2) glioblastoma cell resistance to radio-chemotherapy. In this thesis we aimed to investigate the cellular mechanisms of receptor tyrosine kinases in conferring resistance to therapy. We first performed a literature search and found that almost all studies that advocated for the utility of targeting RTKs in overcoming treatment resistance did not employ both therapeutic agents comprising standard therapy – radiotherapy and TMZ. We next generated an in vitro glioblastoma resistant model via short-term treatment with radiotherapy and TMZ and found that these cells had down-regulated RTK activity in addition to down-regulated protein and gene expression of the commonly altered and studied epidermal growth factor receptor (EGFR) and MET receptor. After generating an in vitro glioblastoma recurrent model via long-term treatment we demonstrated that the surviving sub-population of cells also displayed down-regulated EGFR and MET expression compared to treatment naive cells. Furthermore, we also showed that the resistant cell population already pre-exists within the parental population which suggests the possibility of pre-emptively targeting the inherently resistant population. Interestingly, we also observed differential microRNA expression in radiotherapy- and TMZ-treated cells and, specifically, found that miR-221 confers resistance to glioblastoma cells and is capable of down-regulating EGFR expression. We validated this relationship in a human cohort of 105 primary and 36 recurrent glioblastoma patients, showing a significant inverse relationship between miR-221 and EGFR. Consistently, we showed that high miR-221 and low-EGFR expression at recurrence is associated with a poorer prognosis. Lastly, we investigated the relevance of epithelial to mesenchymal transition markers after observing that migration rates were maintained in resistant cells despite low EGFR and MET. Both N-Cadherin and CD44 were found to be highly expressed in treatment-resistant cells and the down-regulation of AKT activity with wortmannin led to reduced levels of EMT markers, suggesting that AKT is a regulator of key EMT transcription factors that are specific to N-Cadherin and CD44. The thesis gains it significance by providing an explanation to the failure of RTK inhibitors in the glioblastoma clinic by suggesting that standard radio-chemotherapy down-regulates RTK activity and expression, thereby diminishing any theorised benefit of targeting RTKs. Furthermore, the thesis advocates for microRNAs to be crucial regulators of therapy resistance, potential biomarkers and targetable molecules for the clinic.

High grade serous ovarian cancer (HGSOC) and basal like breast cancer (BLBC) are genomically unstable and aggressive cancers that frequently co-occur and share common molecular features. Of these molecular characteristics are P53 inactivation occurring in almost all cases of HGSOC and BLBC, BRCA1/2 inactivation reported in more than 50% of both cancers and CCNE1 amplification reported in up to 30% and 8% of HGSOC and BLBC respectively. Both HGSOC and BLBC are currently grouped in many clinical trials to test new drugs or drug combination, for instance, PARP inhibitors in the context of BRCA1/BRCA2 mutation. We asked whether deregulated cyclin E1 (CCNE1 amplification and/or its encoding protein, cyclin E1, overexpression) is an additional biomarker that can potentially be used to group patients with both diseases for therapeutic purposes. We studied two well characterised cohorts of 262 HGSOC and 222 familial breast cancer (BLBC enriched) samples of formalin fixed paraffin embedded sections. HGSOC and the BLBC enriched cohort were from patients enrolled in the Australian ovarian (AOCS) and the Kathleen Cuningham Foundation Consortium for research into Familial Breast cancer (KConFab) respectively. Using automated tissue based assay and an in situ hybridization probe that spans 19q12 locus harbouring CCNE1, we assessed the level of CCNE1 amplification. We also assessed the expression of cyclin E1 and a cyclin E1 degradation associated protein FBXW7 and a cyclin E1 deubiquitinase, USP28, by immunohistochemistry, as possible drivers of high cyclin E1 expression in amplified and non-amplified cyclin E1hi subsets. We also assessed the expression of URI1 in our HGSOC cohort. URI1 is a protein encoded by the URI1 gene which co-localise with CCNE1 on 19q12 locus. In HGSOC, we identified seemingly two separate subsets of cyclin E1hi tumors that have different pathological and biological characteristics as well as different clinical outcomes. These are the amplified/cyclin E1hi group that had amplification and high expression of cyclin E1, low expression of FBXW7, higher genomic instability, intact BRCA1/2 and worse outcome. The other is the non-amplified/cyclin E1hi tumors that typically had high expression of cyclin E1 in the absence of amplification, high USP28 expression, lower genomic instability, more prevalent BRCA1/2 loss and more favorable outcome compared to the amplified group. Next we assessed cyclin E1 deregulation in the overlapping groups BRCA1 mutant breast cancer and BLBC. Both subtypes had significantly higher expression of cyclin E1 and amplification compared to other breast cancer types. However, the intensity of cyclin E1 expression and level of 19q12 amplification were lower in BLBC compared to those observed in HGSOC. Moreover, in BRCA1 mutant breast cancer and BLBC patients, only high expression of cyclin E1 was associated with lower overall survival while amplification did not seem to impact outcome. These observations were further supported by our meta-analysis that included our cohorts as well as other published datasets. In the meta-analysis, both CCNE1 amplification and cyclin E1 expression were found to be adverse prognostic factors in HGSOC while only high expression was associated with worse survival in BLBC patients. In fact, both amplified and non-amplified cyclin E1hi BLBC subsets shared almost all cyclin E1 deregulation associated features as well as many features with the non-amplified cyclin E1hi HGSOC subset. Of these are the prevalence of high expression of the cyclin E1 deubiquitinase, USP28, BRCA inactivation, the lower genomic instability and cyclin E1hi linked adverse outcome. In order to provide better therapeutic options for cyclin E1hi BRCA1 mutant breast cancer/BLBC patients, we sought to further assess mechanisms behind the co-occurrence of cyclin E1 overexpression and BRCA1 inactivation. Using the KConFab cohort we have found that BRCA1 loss correlated with decreased phosphorylation of cyclin E1, on Threonine 62, assessed by immunohistochemistry. We also showed by in vitro analysis that BRCA1 loss in cell lines led to cell cycle specific stabilisation of cyclin E1 by reducing cyclin E1 T62 phosphorylation. Conversely, BRCA1 overexpression increased T62 phosphorylation. Overexpression of cyclin E1 with an inactivated T62 site, to mimic loss of phosphorylation, increased cyclin E1 stability and resistance to Paclitaxel. These findings suggest that BRCA1 regulates cyclin E1 stability in breast cancer cells via regulating T62 phosphorylation. We next assessed a combination therapy that target cyclin E1 and BRCA1 inactivation using CDK2 and PARP inhibitors. CDK inhibitors are suggested to induce DNA damage and therefore we hypothesised that CDK2 inhibition would enhance sensitivity of BRCA1 deficient cells to PARP inhibition. Our finding is that CDK2 inhibition induced DNA damage and synergised with the PARP inhibitor Rucaparib in BRCA1 mutated cell lines. Combination treatment of xenograft are in progress but the preliminary data is supportive of our hypothesis. Our results propose a new therapeutic strategy for BRCA1-mutant breast cancer/BLBC by combining CDK2 and PARP inhibitors to enhance synthetic lethality. As this group shares similarities with non-amplified cyclin E1hi HGSOC subset, we suggest that this combination is likely to be effective in the comparable HGSOC subset.

“Knowledge is power. Information is liberating.” Kofi Atta Annan (1998). Modern postoperative recovery assessment has progressed from that which was focused purely on the physiological restitution in the immediate postoperative period to one that is multidimensional, individualised, dichotomised and provided in real time. The Postoperative Quality of Recovery Score (PostopQRS) is an extensively validated multidimensional recovery assessment tool that has been developed de-novo for assessment of recovery in postoperative patients, and has been widely adopted as an outcome measure in observational trials. It is unique in its ability to provide individualised recovery information to each patient, and in its assessment of a patient’s postoperative recovey in relation to their own unique preoperative baseline performance. Through its digital interface, it has the potential to provide individualized, contemporaneous recovery data to each patient, thus keeping patients informed of their own recovery throughout their postoperative journey, and thus potentially improving their ultimate postoperative outcome. Prior to this thesis, the PostopQRS’ cognitive domain, whilst based on widely accepted neurocognitive tests, was yet to demonstrate clinical face validity in its assessment of cognitive recovery, and was currently unable to assess cognitive recovery in patients who score low on preoperative cognitive baseline testing. Simmilarly, the PostopQRS had been widely adopted as an outcome measure in observational trials but was yet to be implemented in interventional studies. Furthermore, the utility of the PostopQRS in providing real-time recovery assessment was yet to be fully explored. The EchoNOF-I pilot study demonstrated the clinical utility of the PostopQRS as an outcome measure in the interventional trial setting, and demonstrated feasibility with providing fractured neck of femur surgery patients with focused point of care ultrasound. This study highlighted the need for there to be a validated method with which to measure cognitive recovery in patients with low cognitive baseline PostopQRS scoring. Head-to-head comparison of patient cognitive performance on both the PostopQRS and formal neurocognitive test battery was performed, both at preoperative baseline and during postoperative recovery. This demonstrated clinical face validity in defining patients as having low, as opposed to normal, PostopQRS baseline performance, and demonstrated face validity in the proposed method with which to score cognitive recovery in thse patients with low cognitive baseline scoring. The RTR-I pilot study was ground-breaking as it was the first study in which the PostopQRS was used as the intervention itself, and which demonstrated the clinical feasibility of the PostopQRS in providing patients with individualised real-time recovery information throughout their recovery journey. This study has the potential to revolutionise recovery assessment from one that has been traditionally research focused, to one that has direct clinical applications that may ultimately improve patient outcomes postoperatively. Work from this thesis has formed the basis for three additional multicentre randomised control trials (RTR-II, ECHONOF –II and ECHONOF –III), one MACH (Translational Research Projects 2019), one ANZCA (ANZCA Projects 2019) and two NHMRC grant applications (Project Grants 2019, MRFF Keeping Australians Out of Hospital 2019). The modified scoring system with which to measure cognitive recovery in patients with low cognitive baseline PostopQRS scoring has been adopted into the PostopQRS.com website. What is truly exciting is the future post-doctoral exploration of the potential for the PostopQRS to provide individualised real-time recovery information to both patient and health care provider, and through this, ultimately improve patient outcomes.

Head and neck squamous carcinomas (HNSCCs) are a diverse group of squamous cell cancers in the upper aerodigestive tract. Whilst tobacco, alcohol, and the human papilloma virus (HPV) have been implicated in HNSCC oncogenesis, there is an incomplete understanding of how these factors impact upon molecular characteristics. In this study, we aim to better understand the genomic characteristics of HNSCC and the role of these risk factors in tumour development. Data for 528 patients from the Cancer Genome Atlas (TCGA) was used for in silico analyses to delineate the role of HPV in oral cavity cancers, where evidence for viral-induced oncogenesis remains contentious. A clear mutational signature for viral oncogenesis was identified in oral cavity carcinomas, as well as corresponding signatures in gene expression and DNA methylation, indicative of a role for HPV in the formation of oral cavity cancers. TCGA patients with and without tobacco and alcohol risk factors - non-smokers/non-drinkers (NSND) – were evaluated. Unfortunately, the proportion of NSND patients in the TCGA was insufficient for adequate analysis. Targeted next-generation sequencing for 69 genes as well as 4 HPV subtypes was therefore carried out on 186 patients recruited from our hospital. As frozen tissue was not available for all patients, a workflow to sequence and analyse formalin fixed paraffin embedded (FFPE) samples was developed. Differential mutation of 5 different genes was identified in the NSND group, more than for any other clinicopathologic variable. A high rate of mutations in the extra-cellular domain of NOTCH1 was noted in the mutational data. The role of these putatively inactivating NOTCH1 mutations in HNSCC remains uncertain. NOTCH1 knockouts were induced in HNSCC cell lines using the CRISPR/Cas9 gene editing system to assess the role of these mutations. No phenotypical differences could be identified using in vitro assays. Transcriptomic analysis of the edited cell lines identified alterations in gene expression in cellular adhesion pathways, consistent with epithelial-mesenchymal transition and the known activity of the Notch pathway. Overall, it appears that inactivating mutations in NOTCH1 may result in a paradoxical activation of the Notch pathway. In conclusion, we have established that differing exposures to risk factors for HSNCC result in molecular differences. These differences require validation and in-depth exploration. HPV has also been implicated in the development of oral cavity carcinomas, and novel treatment protocols for HPV-positive oropharyngeal cancers should also be considered for patients with these cancers. The prominence of the Notch pathway in HNSCC is notable, and the apparent paradoxical role of NOTCH1 mutations on pathway function deserves further follow up.

Background: Research into Colorectal cancer (CRC) require maintenance of clinical cancer databases with complex datasets. These are resource intensive, region specific, and compromised by reporting bias [1]. Administrative data are routinely captured for each hospital admission and may serve as an alternative source for populating databases. However, the accuracy of administrative data has not been fully explored and may vary by data item. The aims of this study included identifying a cohort of new CRC patients from administrative data, measuring its accuracy, and deriving coding algorithms to improve the accuracy of diagnoses, procedures and short-term outcomes. There has been much debate that major surgery, in particular for cancer patients, should be concentrated in tertiary centres, based on the premise that high volume centres achieve better outcomes. In this study, we investigated two hypotheses: that the majority of complex colorectal cancer resections are performed in major city hospitals and that the short-term outcomes are better in CSSANZ (Colorectal Surgical Society of Australia and New Zealand) hospitals. Large Inpatient administrative databases are a common source used to identify comorbidities recorded with International Classification of Disease (ICD) diagnostic codes. These data sources may be used to assess the effect of baseline comorbidity status on surgical care outcomes. In this study, we hypothesized that the ASA PS (American Society of Anaesthesiologists physical status) classification can predict short-term outcomes after a colorectal cancer resection when compared to the Elixhauser comorbidity index (ECI). Methods: A retrospective study was conducted to identify all new colorectal cancer resections at The Royal Melbourne Hospital from 1st of January 2008 to 31st of December 2013, using administrative data. Code combinations and algorithms were used to improve the accuracy of administrative data. These algorithms were utilized to identify an accurate cohort of colorectal cancer resection cases from the Victorian Admitted Episodes Dataset (VAED), between July 2008 to June 2013. The short-term outcomes and workloads were compared in public hospitals across the state of Victoria. The algorithms constructed were also utilised to identify an accurate cohort of CRC resection cases from Dr Foster Global Comparators Victorian dataset. ASA PS classification scores were identified from these cases. Multiple linear regression models were constructed to study the association between comorbidity indices and short-term outcomes. Results: It is possible to use administrative data to identify new colorectal cancer patients who have had a surgical resection, using specific coding algorithms. Administrative data has an accuracy of 80-100% for most data fields, and this accuracy can be improved using coding algorithms. An accurate cohort of colorectal cancer resection cases was identified from the VAED dataset. Seventy-three percent of CRC resections in the state were performed in metropolitan city hospitals. There was no significant difference in LOS (length of stay), mortality and reoperation rates between CSSANZ and non-CSSANZ hospitals. This study demonstrates that administrative data is both cost-effective and informative. The ASA PS model was indeed shown to be a strong predictor of the primary outcome: length of stay (LOS). The significant predictors of LOS were emergency operations, rectal cancer resections, ASA3 and patients age. The Elixhauser model was a better predictor than the ASA PS model. However, the full model adjusted for both the ECI and ASA PS grade was the best predictor of outcome. The study indeed showed the ability of the ASA PS classification to identify short-term clinical outcomes. Conclusion: These studies make the possibility of a Victorian CRC registry containing all surgical CRC patients a real possibility. Such a registry would enable outcomes research across the whole state with the possibility of data linkage to international administrative data sets.

Patients with defects in the mismatch repair pathway, driven by either MSH2 or MSH6 loss, experience a significant increase in the incidence of prostate cancer, while germline mismatch repair defects in either MLH1 or PMS2, exhibit no such increase. This PhD project demonstrates that androgen-receptor activation, a known driver of prostate cancer, can disrupt the MSH2 gene in prostate cancer model systems through the induction of structural variations. Prostate tumours from two contrasting risk cohorts were screened to confirm loss of MSH2 protein expression in a small number of patients. Surprisingly, it was also found that a small but significant fraction of high-risk cases exhibited reduced expression of MSH2 without complete loss. Stratifying a large independent TCGA prostate cancer cohort for MSH2 expression levels revealed that patients whose tumours exhibited either complete loss or aberrant levels of MSH2 had significantly worse survival outcomes and accelerated clinical progression. In contrast, aberrant MSH2 levels had no impact on clinical survival in colorectal cancer. This PhD project also demonstrates that reduced expression of MSH2 can be explained by androgen-induced microRNA regulatory mechanisms. Here, it is demonstrated that miR-21 and miR-141 may both target the MSH2 gene leading to reduced MSH2 protein staining and both microRNAs were seen to be upregulated in prostate cancer patients with reduced MSH2 levels. Interestingly, this thesis also shows that miR-21 and miR-141 are both regulated by androgens, implicating this mechanism as a second androgen driven method of MSH2 downregulation. Importantly this PhD project also found that aberrant MSH2 expression in prostate tumours does not induce the same enhanced immune cell mobilisation seen in colorectal tumours suggesting that the prostate is an immune privileged site. This is contrary to the findings of other studies and may warrant a re-evaluation of whether MSH2 deficient prostate cancers are likely to benefit from immunotherapies. To further investigate if the prostate tumour microenvironment is indeed in an immunosuppressive state, a detailed investigation of the transcriptomic profile of the cells of the tumour microenvironment was also conducted. This resulted in not only the discovery of immunosuppressive signatures in tumour infiltrating T-cells but also significant transcriptomic alterations in other cancer associated pathways such as osteogenesis, cell migration, epithelial mesenchymal transition (EMT), hormone signalling and angiogenesis throughout the tumour microenvironment compared to the cellular make-up of benign prostatic tissue. The data presented in this thesis constitute a significant contribution to the current understanding of how defects in the mismatch repair gene MSH2 may affect prostate cancer severity. Additionally, these studies demonstrate multiple mechanisms through which the prostate tumour microenvironment may enhance prostate cancer progression.

Androgen deprivation therapy is the mainstay of treatment for advanced prostate cancer. Although castration results in tumour regression and symptom relief in the majority of cases, the effects are short lived, with most patients demonstrating castration resistant progression within 1-3 years. Numerous molecular mechanisms have been implicated in the development of this resistance to treatment, including maintenance of intraprostatic testosterone levels and changes in the expression of and/or alternative splicing of the androgen receptor. Many of these observations however have been made in cell lines and/or xenograft models, which do not necessarily recapitulate the complex microenvironment in which clinical resistance develops. Whole Genome Sequencing (WGS) and RNASeq has been performed on samples from a cohort of patients who received neo-adjuvant androgen deprivation therapy prior to prostatectomy, and had varying responses, to elucidate the genomic drivers of castration resistant disease. WGS analysis has shown no commonality in neither Single Nucleotide Variations or Copy Number Variations in each response group, nor a distinct change between the pre and post-treatment samples indicating that there is not a selection for a resistant sub clone as a result of androgen deprivation treatment. Analysis of RNA-Seq data has shown no overexpression of the androgen receptor, nor the presence of ARV7. It has however shown a striking difference between treated and untreated samples, displaying a phenotypic change which has occurred as a response to treatment; including an epithelial to mesenchymal transition signature and basal cell signature; as well as an up-regulation in expression of FGF7 and FGF2. Cell line validation assays have indicated that these may be playing a role in treatment persistence in a castrate environment.

Colorectal cancer (CRC) is the third most common cancer worldwide, affecting over 15,000 individuals in Australia each year. While CRC is often detected at a stage where resection of the primary tumour is possible, approximately 50% will relapse and die from metastatic disease. Current practice to determine clinical management and prognosis is determined by tumour depth (T) and lymph node stage (N) and the extent of cancer spread at diagnosis (M) (TNM staging). However, clinical outcomes of patients with the same TNM staging can be heterogeneous. While adjuvant 5-flurorouracil (5-FU) based chemotherapy is offered to the majority of patients with stage III CRC it is only offered for high risk stage II CRC and many will relapse to 5-FU as a first line therapy. Therefore, there is a need to identify markers to better predict prognosis and better stratify patients with stage II/III CRC for treatment regimes. Currently, the use of biomarkers in prognostication for the management of CRC is still not common practice with only a few markers used in the clinic. The mutational landscape of CRC has revealed large numbers of mutated genes; however, it is not known which are drivers that contribute to carcinogenesis and whether such mutations provide prognostic information. This thesis aims to (I) identify clinically relevant markers and patient subgroups of CRC to better predict prognosis and better stratify patients with stage II/III CRC for treatment regimes, (II) to investigate the potential for a MACROD2 deletion as a novel driver of CRC tumourigenesis and (III) to elucidate the impact of a MACROD2 gene deletion on DNA repair and chromosomal instability in CRC.

Cancers arise from single progenitor cells that acquire mutations, eventually dividing into mixed populations with distinct genotypes. These populations can be estimated by identifying common mutational profiles, using computational techniques applied to sequencing data from tumour tissue samples. Existing methods have largely focused on single nucleotide variants (SNVs), despite growing evidence of the importance of structural variation (SV) as drivers in certain subtypes of cancer. While some approaches use copy-number aberrant SVs, no method has incorporated balanced rearrangements. To address this, I developed a Bayesian inference approach for estimating SV cancer cell fraction called SVclone. I validated SVclone using in silico mixtures of real samples in known proportions and found that clonal deconvolution using SV breakpoints can yield comparable results to SNV-based clustering. I then applied the method to 2,778 whole-genomes across 39 distinct tumour types, uncovering a subclonal copy-number neutral rearrangement phenotype with decreased overall survival. This clinically relevant finding could not have been found using existing methods. To further expand the methodology, and demonstrate its application to low data quality contexts, I developed a novel statistical approach to test for clonal differences in high-variance, formalin-fixed, paraffin-embedded (FFPE) samples. Together with variant curation strategies to minimise FFPE artefact, I applied the approach to longitudinal samples from a cohort of neo-adjuvant treated prostate cancer patients to investigate whether clonal differences can be inferred in highly noisy data. This thesis demonstrates that characterising the evolution of structural variation, particularly balanced rearrangements, results in clinically relevant insights. Identifying the patterns and dynamics of structural variation in the context of tumour evolution will ultimately help improve understanding of common pathways of tumour progression. Through this knowledge, cancers driven by SVs will have clearer prognoses and clinical treatment decisions will ultimately be improved, leading to better patient outcomes.

Glioblastoma multiforme (GBM) represents the most malignant incarnation of glial tumours – a World Health Organisation (WHO) grade IV brain malignancy. GBM is the most common primary brain tumour in adults, accounting for 78% of all malignant central nervous system (CNS) tumours, and affecting 2-3 people per 100,000 in Europe and North America, with an average survival of only 14.6 months. Despite continued research and incremental advances in imaging, surgery, and chemoradiotherapy, patient survival has stagnated in the past decade, with several promising lines of investigation failing to fully deliver on their anticipated translational outcomes. Recent advances in genetic sequencing and computational biology have allowed the simultaneous comparison of large numbers of patient cancer cell genomes and identified several GBM subtypes. It is hoped that such stratification will one day allow clinicians to tailor treatments specific to each GBM subtype as has already happened in cancers like medulloblastoma. However, despite best efforts, GBM remains highly aggressive, infiltrative, and treatment-resistant, rendering it incurable by current treatment modalities. Invasion of tumour cells into normal brain prohibits a surgical cure, while a high cancer stem cell (CSC) component resists treatment with radiation and temozolomide (TMZ) – both of which are more effective against rapidly dividing cells – and relapse remains the rule. Molecular mechanisms underlie GBM’s treatment resistance, and elucidating the key drivers that garner inherent resistance to the quiescent, stem-like fraction of cells that lead to treatment failure therefore presents as an exciting area of research that may uncover new potential drug targets that improve patient outcomes. This study has shown that the proliferation rate of GBM cells is spectral, approximating a positively skewed normal distribution, with highly proliferative cells at one end and quiescent cells at the other. The quiescent cell fraction was subsequently shown to be inherently more resistant to chemoradiotherapy than the proliferative fraction. The quiescent fraction also displayed increased size, complexity, rates of migration and invasion, secretion of extracellular matrix-degrading enzymes, and invadopodia activity than their proliferative counterparts. Similarly, quiescent cells proliferated slower as intracranial tumours but displayed significantly greater invasive properties than a subset of proliferative cells grown in vivo. mRNA expression analysis revealed the genetic signature that underpins the disparity in proliferation rate between quiescent and proliferative cells, and the putative genes that are responsible for the malignant properties identified in both populations. This body of work has uncovered the inherently dichotomous treatment response of quiescent and rapidly dividing GBM cells, as well as the difference in their abilities to migrate and invade. This study has also shed light on the fundamental molecular mechanisms that are at the root of treatment resistance and malignancy in this disease. It is hoped that this expression signature will help to inform future studies and treatments that target these differences and make GBM less of a death sentence and more of manageable, chronic disease.