Defining the molecular profile of oral tongue squamous cell carcinomas and its impact on patient outcome

Abstract

Amongst all head and neck squamous cell carcinomas (HNSCC), oral tongue carcinomas (OTSCC) have the worst prognosis for early stage disease. The current staging system is unable to consistently identify patients with high risk disease. In one of the largest comprehensively annotated OTSCC cohorts to date, this thesis examines a combination of literature identified candidate biomarkers, and also sought to determine if a novel prognostic molecular signature could be identified.

Given the high frequency of reported CDKN2A alterations in HNSCC, a comprehensive analysis of the differential mechanisms of CDKN2A alteration was performed. Promoter methylation status, mutation status, copy number variation and protein expression were assessed. The majority of samples (95%) did not demonstrate p16 over-expression assessed by immunohistochemistry. Although disruption of CDKN2A was found to be a frequent event arising from a variety of mechanisms, no correlation between CDKN2A alteration and clinicopathological features was found.

A panel of loci frequently reported to be hypermethylated in HNSCC was investigated within the OTSCC cohort, with three quantitative methodologies that assessed DNA methylation. In contrast to the literature, these loci were not commonly methylated. Findings were confirmed in an external cohort of HNSCC samples from The Cancer Genome Atlas (TCGA) that had methylation levels quantified with a fourth, orthogonal methodology. The use of non-quantitative methodology in the literature was the likely cause for the overestimation of significant methylation events, with this study highlighting the need for the cautious interpretation of this literature.

Utilising a genome-scale wide methylation platform, a prognostic methylation signature was sought for the OTSCC cohort. Given the absence of consensus on data analysis, comprehensive bioinformatics analyses were performed utilising multiple contemporary R software library packages, to enable a thorough examination of the data with published algorithms used for pre-processing and downstream analysis. However, methylation assessed over greater than 450,000 CpG dinucleotides did not reveal a differentially methylated group of samples, and was not informative for clinicopathological variables. Furthermore, despite increasing the total number of samples to include the TCGA OTSCC dataset, a prognostic methylation signature was not identified in any cohort.

Targeted mutational profiling of the cohort was also performed. A disproportionately large number of variant calls were identified on the initial processing of samples. Validation of a subset of variant calls with orthogonal methodology was able to confirm the presence of only 12/50 (24%) selected mutations. Despite pre-analytical quality assurance assessments, the replicated analysis of samples and the use of stringent filtering criteria, the presence of artefactual variant calls masked the identification of true mutations. The likely source of the large number of artefactual variant calls was from the PCR based amplification of artefact in DNA extracted from formalin-fixed, paraffin-embedded tissue.

This thesis emphasises the importance and the impact of the choice of methodology on the successful identification of clinically relevant biomarkers. Within the limitations of current understanding and the size of the cohort examined, it also suggests that both CDKN2A alteration and DNA hypermethylation in isolation are not prognostically informative biomarkers for OTSCC. Further research is required into the prognostic value of other molecular alterations and the combined impact of simultaneous aberrations.