Detection of tuberculosis-associated compounds from human skin by GCxGC-TOFMS

Abstract

Tuberculosis (TB) remains a global health concern. This study aimed to investigate the potential of human skin volatile organic compounds (VOCs) as prospective biomarkers for TB diagnosis. It employed a non-invasive approach using a wearable silicone rubber band for VOC sampling, comprehensive gas chromatography - time of flight mass spectrometry (GCxGC-TOFMS), and chemometric techniques. Both targeted and untargeted biochemical screening was utilized to explore biochemical differences between healthy individuals and those with TB infection. Results confirmed a correlation between compounds found in this study, and those reported for TB from other biofluids. In a comparison to known TB-associated compounds from other biofluids our analysis established the presence of 27 of these compounds emanating from human skin. Additionally, 16 previously unreported compounds were found as potential biomarkers. The diagnostic ability of the VOCs selected by statistical methods was investigated using predictive modelling techniques. Artificial neural network multi-layered perceptron (ANN) yielded two compounds, 1H-indene, 2,3 dihydro-1,1,3-trimethyl-3-phenyl; and heptane-3-ethyl-2-methyl, as the most discriminatory, and could differentiate between TB-positive (n = 15) and TB-negative (n = 23) individuals with an area under the receiver operating characteristic curve (AUROC) of 92 %, a sensitivity of 100 % and a specificity of 94 % for six targeted features. For untargeted analysis, ANN assigned 3-methylhexane as the most discriminatory between TB-positive and TB- negative individuals. An AUROC of 98.5 %, a sensitivity of 83 %, and a specificity of 88 % were obtained for 16 untargeted features as chosen by high performance variable selection. The obtained values compare highly favourable to alternative diagnostic methods such as breath analysis and GeneXpert. Consequently, human skin VOCs hold considerable potential as a TB diagnostic screening test.

Keywords: GCxGC-TOFMS; Human skin volatiles; Non-invasive wearable sampler; Predictive modelling; Tuberculosis biomarkers.