Andrea Cerratoa,b, Alessandra Biancolilloc, Aldo Laganàa,b, A. Lleo De Naldad,Enrico Taglionia,b, Anna Laura Capriottia,b

a Department of Chemistry, Sapienza University of Rome, Piazzale Aldo Moro 5, 00185, Rome, Italy;
b Biostructures and Biosystems National Institute (INBB), Viale delle Medaglie d'Oro, 305, 00136, Rome, Italy;
c Department of Physical and Chemical Sciences, University of L'Aquila, Via Vetoio, 67100, L'Aquila, Coppito, Italy
d Department of Biomedical Sciences, Humanitas University, 20089, Milan, Italy

Cholangiocarcinoma (CCA) is a deadly cancer of the biliary epithelium whose etiopathogenesis remains largely unknown. CCA is commonly asymptomatic in the early stages, and no clinical molecular markers for early diagnosis or management are known. They are frequently diagnosed in advanced phases when the disease is found disseminated. Moreover, CCA is a highly chemoresistant tumor, and pharmacological therapies are generally unsuccessful, with a 5-year survival rate persisting below 10% since the 1980s. Cancer cells are highly metabolically active, both cause and consequence of their pathogenesis, but this may also represent an opportunity for diagnosis and treatment. The comprehensive analysis of the metabolome in body fluids is emerging as a diagnostic strategy that could be associated with the progression of the disease. In the present study, a combined approach of unbiased multi-omics to develop an accurate predictive model for identifying non-invasive biomarkers for personalized management of CCA patients. Serum samples of metastatic and resected CCA patients as well as healthy controls were analyzed by mass spectrometry (MS)-based metabolomics and lipidomics following the guidelines of the Metabolomics Quality Assurance & Quality Control Consortium (mQACC) for data acquisition and pre-processing. 

Partial least square discriminant analysis (PLS-DA) classification with repeated double cross-validation and permutation tests was performed to provide rigorously validated models that prevent overfitting. Variable of importance (VIP) analysis allowed the selection of the metabolites of interest, which were then identified by inspection of the MS and MS/MS spectra. Exogenous metabolites were then removed from the data matrix which was again subject to PLS-DA analysis. The iteration process was repeated six times for the removal of the exogenous compounds selected by VIP. The final metabolomics model furnished correct classification rates higher than 99% and highlighted the role of short peptides, hexanoylcarnitine, and inosine. Moreover, the lipidomics model furnished correct classification rates of 91% for CCA patients and 83% for healthy controls, suggesting dysregulations in the metabolism of phosphatidylethanolamines and ethanolamine plasmalogens connected to the insurgence of CCA.

Keywords: metabolomics; lipidomics; chemometrics; bile duct cancer