Tumour heterogeneity poses significant challenges in the development and implementation of personalised cancer treatments, and the optimisation of sampling strategies for diagnostic assays. Expanding our understanding of cancer proteomic heterogeneity is essential to improving cancer treatment (Dagogo-Jack et al, 2018).Ovarian cancer is currently the most lethal gynaecological malignancy, of which high-grade serous ovarian cancer (HGSOC) is the most common subtype. Studies of HGSOC have reported spatial heterogeneity at a genomic level (Schwarz et al, 2015), but this had yet to be explored at a proteomic level. In this study, we explored within-patient and between-patient heterogeneity in eleven HGSOC patients using sequential window acquisition of all theoretical fragment ion spectra mass spectrometry (SWATH-MS).We analysed matched samples taken from primary ovarian tumours and omental metastasis. The MS data acquisition was performed in duplicate, with technical replicates being run on independent 6600 triple TOF instruments. The dataset consisted of 1,354 SWATH-MS runs, the majority representing 454 sections of cryopreserved cancerous tissue. A spectral library was created from 145 LC MS/MS runs in DDA mode. The raw wiff files were searched with ProteinPilot against the Uniprot database, to generate nine different libraries. The libraries were filtered at 1% peptide FDR. A merged library, used for the SWATH-MS runs, contained 76,209 peptides. The SWATH data was filtered at 1% PSM FDR, and 67,775 peptides were found in more than ten SWATH-MS runs.
Our analysis shows that the variation between patients is greater than the local variation within a tumour. The local variation within a tumour can, to some extent, be explained by histological content variation. We show that patient specific proteomic signatures can distinguish between samples from different patients. Our large dataset also highlights the effects of confounding variables, such as differences between mass spectrometers, and sample preparation methods. These confounding variables often cause differences in the proteins identified as having non-zero intensities, rather than causing relative differences in intensities. This has important consequences for the interpretation of proteomic analysis of large cohorts.