In this study, we aimed to identify potential metabolic biomarkers that can improve the diagnostic accuracy of sepsis. Sixty-six patients including 30 septic and 36 nonsepsis patients from an intensive care unit were recruited. The global plasma and urine metabolomic profiles were determined by ultraperformance liquid chromatography coupled with a quadrupole time-of-flight mass spectrometry-based methodology. The risk factors, including both traditional physiological indicators and metabolic biomarkers, were investigated by binary logistic regression analysis and used to build a least absolute shrinkage and selection operator (Lasso) regression model to evaluate the ability of diagnosis. Fifty-five metabolites in plasma and 11 metabolites in urine were identified through orthogonal projections to latent structures discriminant analysis (OPLS-DA). Among them, ten (PE (20:4(5Z, 8Z, 11Z, 14Z)/P-18:0), harderoporphyrinogen, chloropanaxydiol, (Z)-2-octenal, N1,N8-diacetylspermidine, 1-nitroheptane, venoterpine, α-CEHC, LysoPE (20:0/0:0), corticrocin) metabolites were identified as risk factors. The Lasso regression model incorporating these ten metabolic biomarkers and five traditional physiological indicators displayed better differentiation than the traditional model, represented by the elevated area under receiver operating characteristic curve (AUROC) from 96.80 to 100.0%. Furthermore, patients with septic shock presented a significantly lower level of PE-Cer (d16:1(4E)/19:0). This study suggests that metabolomic profiling could be an effective tool for sepsis diagnosis.
Keywords: UPLC−Q-TOF/MS; diagnosis; metabolomics; sepsis.