Quantification of cancer biomarkers in urine using volatilomic approach.

Urine analysis is an attractive approach for non-invasive cancer diagnostics. In this study, a procedure for the determination of volatile organic compounds (VOCs) in human urine (acetone, acetonitrile, dimethylsulfide, dimethyl disulfide, dimethyl trisulfide, hexane, benzene, toluene, 2-butanone, 2-pentanone, pentanal) has been described including sample preparation using preconcentration of analytes in sorbent tubes followed by gas chromatography with mass spectrometry (GC-MS). Fractional factorial design and constrained surfaces design were used to optimize preconcentration of VOCs in sorbent tubes. The procedure was validated by analysis of synthetic urine containing VOC standards in the concentration range of 1-5000 ng/mL. Optimized procedure was applied to analyze urine samples of 89 healthy volunteers and 85 patients with cancer of various localizations: 42 patients with lung cancer, 25 - colon, 3 - stomach, 2 - prostate, 2 - esophageal, 2 - pancreas, 2 - kidney, 1 - ovarian, 1 - cervical, 1 - skin, 1 - liver. Concentrations of 2-butanone, 2-pentanone, acetonitrile, and benzene were found different in urine of patients with cancer and healthy individuals. Influence of cancer localization and tumor, nodule, metastasis stage on urine VOC profile was considered. The approach of using ratios of VOCs to the main ones instead of concentrations was considered. A diagnostic model based on significantly different VOC ratios was created to classify healthy individuals and patients with cancer using artificial neural network (ANN). The model was validated during construction by means of 3-fold cross-validation. Average area under receiver operating characteristic (ROC) curve on test dataset was 0.886. Average sensitivity and specificity of the created model were 91 % and 82 %.