Identification of a biomarker panel for improvement of prostate cancer diagnosis by volatile metabolic profiling of urine

Abstract

Background: The lack of sensitive and specific biomarkers for the early detection of prostate cancer (PCa) is a major hurdle to improve patient management.

Methods: A metabolomics approach based on GC-MS was used to investigate the performance of volatile organic compounds (VOCs) in general and, more specifically, volatile carbonyl compounds (VCCs) present in urine as potential markers for PCa detection.

Results: Results showed that PCa patients (n = 40) can be differentiated from cancer-free subjects (n = 42) based on their urinary volatile profile in both VOCs and VCCs models, unveiling significant differences in the levels of several metabolites. The models constructed were further validated using an external validation set (n = 18 PCa and n = 18 controls) to evaluate sensitivity, specificity and accuracy of the urinary volatile profile to discriminate PCa from controls. The VOCs model disclosed 78% sensitivity, 94% specificity and 86% accuracy, whereas the VCCs model achieved the same sensitivity, a specificity of 100% and an accuracy of 89%. Our findings unveil a panel of 6 volatile compounds significantly altered in PCa patients' urine samples that was able to identify PCa, with a sensitivity of 89%, specificity of 83%, and accuracy of 86%.

Conclusions: It is disclosed a biomarker panel with potential to be used as a non-invasive diagnostic tool for PCa.

Fig. 1

a PLS-DA scores scatter plot (Pareto scaling; 2 components) obtained for VOCs training model of PCa patients (n = 40, squares) vs. cancer-free controls (n = 42, circles), after variable selection; b Assessment of the diagnostic performance of the PLS-DA model obtained for VOCs using the training set (AUC = 0.975; sensitivity = 92%; specificity = 100%) and the external set (AUC = 0.898; sensitivity = 78%; specificity = 94%) through ROC analysis; c PLS-DA scores scatter plot (Pareto scaling; 2 components) obtained for VCCs training model of PCa patients (n = 40, squares) vs. cancer-free controls (n = 40, circles), after variable selection; d Assessment of the diagnostic performance of the PLS-DA model obtained for VCCs using the training set (AUC = 0.878; sensitivity = 71%; specificity = 97%) and the external set (AUC = 0.944; sensitivity = 78%; specificity = 100%) through ROC analysis

Fig. 2

Description, % of variation and assessment of the diagnostic performance of the 6-biomarker panel using the training (AUC = 0.856; sensitivity = 72%; specificity = 96%) and the external (AUC = 0.904; sensitivity = 89%; specificity = 83%) sets through ROC analysis

Fig. 3

Heatmap with the Spearman’s correlations among the 30 identified and putatively identified metabolites significantly altered. C1: 2-hexanone; C2: hexanal; C3: 2-methylcyclopentan-1-one; C4: 4-methylhexan-3-one; C5: 5-methylheptan-2-one; C6: 4-methyldec-1-ene; C7: 3,7,7-trimethylbicyclo[4.1.0] hept-3-ene; C8: 2,6-dimethyl-6-hepten-2-ol; C9: 3-methyl-6-(propan-2-ylidene)cyclohex-1-ene; C10: 4,6-dimethylheptan-2-one; C11: 3,7-dimethylocta-1,6-dien-3-ol; C12: 3,4-dimethylcyclohex-3-ene-1-carbaldehyde; C13: 1-methyl-4-propan-2-ylcyclohex-2-en-1-ol; C14: terpinen-4-ol; C15: 2,5-dimethylbenzaldehyde; C16: 2-hydroxy-2-methyl-1-phenylpropan-1-one; C17: dihydroedulan IA; C18: 5-methyl-2-(propan-2-yl)cyclohexyl acetate; C19: 2,6,6,10-tetramethyl-1-oxaspiro[4.5]dec-9-ene; C20: 4,5,910-dehydroisolongifolene; C21: 2-butanone; C22: 2-pentanone; C23: cyclohexanone; C24: hexadecane; C25: phenylacetaldehyde; C26: 3-phenylpropionaldehyde; C27: 2-butenal; C28: decanal; C29: glyoxal; C30: methylglyoxal