Cancer continues to pose a major public health and economic burden worldwide. Whilst early diagnosis is frequently associated with improved treatment outcomes, vague initial symptoms means that many cancers are detected at a late, often incurable, state. There remains therefore an important unmet clinical need to develop accurate, acceptable, and affordable methods for cancer early detection.

The objective of this thesis was to develop a high throughput and standardised methodology for the deep profiling of urinary volatile organic compounds (VOCs), and to apply the developed methods for the discovery of urinary VOC biomarkers to non-invasively diagnose gastrointestinal cancers.

Towards this objective the optimal process and conditions of urine sample preparation were determined. Two urinary VOC extraction techniques, solid-phase microextraction (SPME) and HiSorb sorptive extraction, were compared and optimised for urinary VOC analysis. Following method development, HiSorb outperformed SPME for urinary VOC analysis, providing broader VOC detection, reduced batch effects, and high throughput reliable automated sampling. A standardised workflow using HiSorb was therefore developed for urinary VOC analysis to minimise the methodological heterogeneity and maximise the extraction of urinary VOCs. Three gas chromatography mass spectrometry (GC-MS) approaches, including conventional quadrupole-based thermal desorption GC-MS (TD-GC-MS) with a mid-polar column as a benchmark method, TD-GC-time of flight (TOF)-MS (non-polar column), and TD-GC-TOF-MS (polar column), were evaluated with a comprehensive method development being conducted to determine the optimal assay for urinary VOC analysis. Among the eight optimised assays, the best one was able to reliably identify 167 VOCs in urine of healthy subjects. Quality control (QC) measures, including the analysis of pooled QC urine samples, the usage of internal standards, and instrumental quality control, were evaluated and implemented. Removal of non-reproducible, non-linear, and contamination-related VOCs, data normalisation, and urine density correction during data pre-processing were also assessed for quality control purpose. These quality control measures were shown to significantly improve the dataset quality and enable the monitoring and correction of analytical variabilities, ensuring the delivery of reliable results. Subsequently, urine samples from a cohort consisting of pancreatic adenocarcinoma patients (n = 28) and control patients (n = 33) were analysed using the optimised methodology. Univariate analysis identified 20 (ChromSpace/Gavin 3 data pre-processing pipeline) and 28 (MSHub/GNPS pipeline) candidate VOC biomarkers for the classification of cancer against control, 8 of which were identified by both pipelines, all with false discovery rate (FDR) corrected p value < 0.25. The majority of the candidate biomarkers identified by the optimal non-polar assay were highly expressed in the cancer group, while the compounds identified by the optimal polar assay were suppressed in cancer. Multivariate partial least squares-discriminant analysis (PLS-DA) showed weak differentiation between groups based on their global VOC profile. Receiver operating characteristic (ROC) curves were plotted using the candidate biomarkers whose adjusted p value < 0.05 (5-(2-thienoyl)butyric acid; butanal, 3-methyl-; furan, 2-methoxy-, and; cyclohexanone, 4R-acetamido-2,3-cis-epoxy-) and < 0.1 (5-(2-thienoyl)butyric acid; butanal, 3-methyl-; furan, 2-methoxy-; cyclohexanone, 4R-acetamido-2,3-cis-epoxy-; 2-butanone, 3-methyl-; 2-vinylfuran; 2,6-di-tert-butyl-4-methyl-phenol; 2-butanol, 2-methyl-; 2-cyclopenten-1-one, 2,3,4-trimethyl-; didodecyl phosphate, and; 2(3h)-furanone‚ 5-heptyldihydro-), with the obtained area under curves at 0.914 and 0.942, respectively, indicating an excellent diagnostic performance of the models.

This thesis provides standardised high throughput and reliable HiSorb-TD-GC-MS methods for the analysis of urinary VOCs. Preliminary analysis of urine from pancreatic cancer patients supports its application in future large scale clinical trials. Clinical implications and future research directions of these findings are discussed.