Metabolic preference assay for rapid diagnosis of bloodstream infections

The study addresses the urgent need for faster diagnostics for bloodstream infections (BSIs), where delays in administering effective antimicrobials markedly increase mortality, particularly in septic shock. Conventional workflows require 2–5 days for pathogen identification (ID) and antimicrobial susceptibility testing (AST), leading to frequent empiric use of broad-spectrum antibiotics and contributing to antimicrobial resistance. Existing improvements (e.g., direct MALDI-TOF) and DNA-based panels expedite ID but generally do not provide comprehensive phenotypic AST, and genetic tests can miss unknown or non-genetic resistance mechanisms. The authors propose a metabolomics-based diagnostic strategy that leverages species-specific metabolic boundary fluxes to both identify pathogens and rapidly infer antimicrobial susceptibility, potentially integrating into current clinical workflows and reducing time-to-result.

The paper surveys current clinical diagnostics: culture-based ID via MALDI-TOF after subculture and growth-based AST on automated platforms (e.g., VITEK 2), which are constrained by microbial growth rates. DNA/multiplex PCR platforms (e.g., FilmArray) can rapidly identify organisms and some resistance genes directly from positive blood cultures but often still require culture for AST and miss phenotypic resistance or novel mechanisms. Prior metabolomics efforts largely sought host or pathogen biomarkers in patient samples, challenged by host metabolic variability and limited for AST. Classical biochemical identification tests assess single metabolic traits. The authors position metabolomics as a sensitive, high-throughput alternative compatible with clinical mass spectrometry, capable of capturing multiple metabolic phenotypes simultaneously for both species ID and phenotypic AST.

Discovery of species biomarkers: The authors measured metabolic boundary fluxes of seven common BSI pathogens (Candida albicans, Klebsiella pneumoniae, Escherichia coli, Pseudomonas aeruginosa, Staphylococcus aureus, Enterococcus faecalis, Streptococcus pneumoniae) under controlled conditions. Three independent experiments were performed; for each species, clinical isolates (three per species, two for P. aeruginosa) were incubated in triplicate in Mueller Hinton broth with 10% human blood (MHb), seeded at 0.5 McFarland (OD ~0.15; ~7.5×10^7 CFU/mL). Culture supernatants were sampled at 0 h and 4 h and analyzed by LC–MS on a Thermo Q Exactive HF. Untargeted features (4,362) were processed with MAVEN; candidate biomarkers were filtered by statistical significance (Bonferroni-corrected p < 1.15×10^-5), fold change (≥4-fold versus MHb), and minimum absolute intensity (>20,000 intensity units). This yielded 53 putative markers, further clustered/curated to define 210 biomarker features using retention time, adduct/fragment m/z relationships, and covariance across replicates. Parent ions were assigned using in-house tools, and putative metabolite IDs were made via MMCD and HMDB. Selected assignments were confirmed by matching MS/MS spectra and chromatographic retention times to commercial standards and by spiking experiments showing concentration-dependent signal increases.

Validation of species biomarkers: An independent validation cohort of 596 clinical isolates (one per patient) encompassing the seven target taxa was analyzed by the same MPA workflow. One-way ANOVA with Bonferroni correction assessed biomarker significance. Demographic associations (age, sex) were tested and found non-significant. Quantitation utilized calibration standards and interspersed QC samples across batches; concentrations were computed from calibration curves and RMSE calculated per metabolite.

Metabolic inhibition assay (MIA) for AST: For rapid phenotypic AST, isolates were incubated for 4 h in MHb with and without antibiotics at CLSI breakpoint concentrations. Inocula were prepared to ~10% of a 2.5 McFarland, yielding OD600 ~0.015. Supernatant metabolomes were profiled using the same LC–MS approach. The principle is that susceptible strains show inhibition or alteration of signature metabolite production/consumption in the presence of an effective drug, whereas resistant strains maintain their metabolic patterns. Multiple antibiotic classes were evaluated: for C. albicans (azoles, polyenes, pyrimidines), and for bacteria both bactericidal and bacteriostatic agents (penicillins, cephalosporins, carbapenems, glycopeptides, aminoglycosides, fluoroquinolones, tetracyclines, macrolides, etc.). Initial concordance with traditional growth-based AST was assessed on a small set (2–3 isolates per species).

Performance validation of MIA: A larger cohort (n=246 isolates: E. coli 50, S. aureus 63, K. pneumoniae 35, S. pneumoniae 49, E. faecalis 23, E. faecium 24) was profiled with a rapid HILIC LC–MS method (5-minute runtime) to maximize throughput. Isolates were challenged at antibiotic breakpoint concentrations. The dataset was split into a training set (n=80) to select robust AST markers per species and derive metabolite inhibition thresholds, and a test set (n=166) to validate predictions against VITEK 2 calls. Where possible, shared markers were used across species (e.g., glucose consumption for S. aureus and Enterococcus; succinate for Enterobacterales; nicotinate for S. pneumoniae). ROC analysis quantified performance.

Head-to-head workflow timing comparison: Three independent runs compared the MPA/MIA workflow to VITEK 2 starting from positive blood culture bottles seeded with S. aureus and E. coli. Following bottle positivity, aliquots were processed in parallel. For MPA/MIA, a positive growth control (no antibiotic) enabled rapid species ID, then only relevant antibiotic conditions were analyzed to reduce MS time. LC–MS used a 5-minute HILIC method with real-time analysis in MAVEN. Total times to report ID+AST were recorded.

Analytical platform: Chromatography used a Thermo Vanquish UPLC with HILIC; ionization by ESI; MS on a Thermo Q Exactive HF operating predominantly in negative ion full-scan mode (50–750 m/z) with high resolving power and AGC settings. MS/MS was used for metabolite confirmation. Calibration and QC were incorporated into each 96-sample batch. Computational/statistical analyses employed R (ANOVA with Bonferroni correction; Tukey-Kramer post hoc), and in-house software (Supplementary Software). Ethics approval and informed consent were noted for sample use.

• Species differentiation: Patterns of consumption/production across 210 curated biomarkers robustly differentiated the seven common BSI pathogens. A concise panel of seven metabolites (Arabitol, Urocanate, Succinate, Xanthine, Mevalonate, N1,N2-diacetylspermine [formate adduct], Lactate) was sufficient to act as binary predictors for species in the discovery dataset and reproduced species-specific clustering in validation.
• Statistical robustness: In the validation cohort (596 isolates), 203 of 210 biomarkers remained significant by one-way ANOVA with stringent Bonferroni correction (p < 1.15×10^-5). Selected key biomarkers showed highly significant species associations (p ~ 1.6×10^-16 to 1×10^-29). No associations were found between biomarker signals and patient demographics (age, sex).
• Quantitative precision: Calibration/QC analyses across 945 total runs (including 864 clinical samples) yielded RMSEs ranging from ~8.9 to 30.3 (arbitrary concentration units), while species-linked biomarker fold changes relative to control were large (64–215×), indicating that measurement error was small relative to biological signal. Addition of 10% blood from different donors (n=20) had negligible effect on profiles.
• Rapid phenotypic AST (MIA): Metabolic inhibition patterns after 4 h reliably distinguished susceptible vs resistant strains across antibiotic classes. In a small multi-species set, MIA matched growth-based AST in 96% of profiles; many ID biomarkers also served as AST markers (e.g., arabitol for C. albicans, succinate for E. coli/K. pneumoniae, N1,N2-diacetylspermine for E. faecalis, anthranilate for P. aeruginosa, lactate for S. pneumoniae), with exceptions (mannitol unreliable for S. aureus resistance, replaced by an m/z 240.069 feature).
• Cohort AST validation: In the 246-isolate study with training/test split, ROC AUCs for metabolite-based susceptibility predictions ranged from 0.91 to 1. Using thresholds from training, test-set agreement with VITEK 2 was 95.2% overall; species-specific agreements included ~97.6% for S. aureus and ~90.3% for K. pneumoniae. Major error rate was 3.1% and very major error rate 1.7%.
• Time-to-result: In head-to-head comparisons from positive blood culture bottles, the MPA/MIA workflow reduced total testing time by an average of 24.3 h for S. aureus and 22.4 h for E. coli (approximately 2.2–2.3× faster overall). Post-flagging ID+AST steps were 4.7–5.0× faster than the conventional pipeline.

The findings demonstrate that ex vivo metabolic boundary fluxes provide robust, species-specific phenotypes for rapid pathogen identification and can be leveraged to infer antimicrobial susceptibility within hours. Metabolism is a sensitive, rapid reporter of cell physiology, enabling phenotypic AST far faster than growth-based methods and complementary to genetic assays that may miss unknown mechanisms. The MPA/MIA workflow can integrate ID and AST on a single LC–MS platform, requires minimal handling from positive blood culture bottles, and aligns with clinical laboratory practices given the availability of medical device-certified mass spectrometers. While classical biochemical tests examine single traits, this multiplexed metabolomics approach expands coverage to multiple metabolites simultaneously, offering broader discriminatory power. The approach holds promise for expanding to additional, less prevalent BSI pathogens and for diverse infection types. Faster diagnostics support timely targeted therapy and antibiotic stewardship, potentially reducing morbidity, mortality, and the selection pressure for resistance.

The study introduces a metabolomics-based diagnostic workflow that: (i) accurately identifies common BSI pathogens using species-specific metabolic signatures, (ii) rapidly determines antibiotic susceptibility phenotypes via a 4-hour metabolic inhibition assay with high concordance to standard AST, and (iii) significantly shortens diagnostic turnaround time compared with current clinical methods. These advances could enable earlier, more precise antimicrobial therapy and improved stewardship. Future work should extend biomarker panels to less common pathogens, refine cross-species AST markers, and further streamline high-throughput LC–MS implementations for routine clinical use.

• Scope of organisms: The primary validation focused on seven prevalent BSI pathogens; expansion to less common species is needed for broader coverage.
• Strain-level resolution: Metabolite-based assays have limited ability to differentiate closely related species or strains, as noted by the authors.
• Biomarker universality: Not all ID biomarkers reliably indicate resistance across antibiotics (e.g., mannitol for S. aureus), necessitating alternative markers per species/drug.
• Resistance profiles: Some cohorts lacked highly resistant isolates for certain species (e.g., E. faecalis largely susceptible; E. faecium added for resistant Enterococcus), and MRSA strains were not included.
• Instrumentation and standardization: Implementation requires LC–MS platforms, standardized methods, and quality controls, which may vary across laboratories and impact transferability until clinically validated.