Abstract
BACKGROUND AND OBJECTIVE: Multidrug-resistant (MDR) bacterial infections are a leading cause of sepsis-related death. A rapid method to identify patients with MDR infections upon hospital admission is urgently needed. This study aimed to characterize the distinct plasma metabolomic signatures associated with MDR gram-positive (G+) and gram-negative (G-) sepsis and to develop predictive models for rapid, risk stratification during the initial clinical encounter. METHODS: Two independent cohorts of septic patients were recruited, with 198 subjects (117 MDR and 81 susceptible) in the discovery cohort, and 198 patients (95 MDR and 103 susceptible) in the validation cohort. Plasma metabolomic profiling was performed using liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS). Multiple machine learning algorithms were employed to identify differential metabolomic signatures and to construct and validate multi-metabolite models for the early identification of MDR bacteria. RESULTS: Distinct metabolomic signatures were identified for both MDR G- and G+ infections. MDR G- sepsis showed significant elevations in metabolites related to host inflammatory responses, such as histamine, alongside decreased levels of gut microbiota-derived metabolites, including cholic acid and benzoic acid, indicating profound host-microbe dysregulation. Conversely, MDR G+ sepsis was characterized by alterations in energy and amino acid metabolism, notably elevated 2-hydroxyglutarate, a marker of mitochondrial stress. An 8-metabolite model for MDR G- infection achieved excellent discrimination in both the discovery (AUROC = 0.885, 95% CI: 0.787-0.982) and validation (AUROC = 0.878, 95% CI: 0.782-0.951) cohorts. The model for MDR G+ infection demonstrated good predictive performance (AUROC = 0.763 and 0.715 in discovery and validation, respectively). CONCLUSION: This study identifies robust and distinct plasma metabolomic signatures that differentiate MDR from antibiotic-susceptible sepsis. These findings support the development of rapid, metabolomics-based testing using admission plasma to risk-stratify patients. This approach could guide early, stewardship-aligned antimicrobial decisions while conventional culture results are pending, potentially improving clinical outcomes.