Identification of volatile biomarkers in exhaled breath by polythiophene solid phase microextraction fiber for disease diagnosis using GC-MS

Abstract

The diagnosis of diseases through monitoring of volatile organic compounds (VOCs) in exhaled breath (EB) holds great potential for clinical applications. However, a standardized method for VOC analysis in EB yet to be proposed. The present study presents an untargeted method for screening and identifying potential volatile biomarkers in EB by a lab-made solid phase microextraction (SPME) fiber. A polythiophene-based SPME fiber was produced by an electrochemical method and VOC sampling was performed under dynamic and controlled conditions. Following the sampling step, the adsorbed VOCs on the SPME fiber were analyzed using gas chromatography-mass spectrometry (GC-MS). The VOCs in EB were screened by the MS detector in selected ion monitoring (SIM) mode within the mass/charge (m/z) range of 13-94 values. Potential biomarkers among all detected VOCs in each subject's EB sample were identified through machine learning algorithms, employing a comparative analysis of distinctive retention times (RT) and peak areas between the lung cancer (LC) and control groups in two stages. In the initial stage of the study, the areas of all peaks observed in the SIM-GC-MS chromatograms of 25 LC and 51 control group subjects were integrated, and the resulting retention times and peak areas were recorded for subsequent analysis to identify potential biomarkers. A total of 1.346 distinct compounds were detected among the 76 subjects in this step, and statistical analysis using the LightGBM algorithm revealed the potential biomarkers for LC diagnosis. The PTh-SPME fibre successfully identified four novel cancer biomarkers in breath matrix: 4-heptenal, 4-methyl-1-octene, 1,2,3,4-tetrahydro-5,8-dimethyl-1-octylnaphthalene and tetrahydro-2-(2,5-undecadiynyloxy)-2H-pyran. In the second step of the study, the efficacy of the top ten selected biomarkers was evaluated in a cohort of 166 subjects, including 70 individuals with LC and 96 in the control group. The model achieved accuracy, area under the curve (AUC), and F Score values of 0.818, 0.816, and 0.817, respectively. The test model correctly predicted 27 out of 33 subjects between LC and control groups.