Harmonizing human plasma metabolite annotation with Plasma Benchmark

The human plasma metabolome has been extensively characterized: version 5.0 of the Human Metabolome Database (HMDB)¹ currently encompasses 37,229 entries for metabolites reported in human blood. Although the analytical coverage of modern LC–MS platforms enables the detection and identification of 1,000–2,000 plasma metabolites, the number of well-known and regularly detected plasma metabolites is considerably smaller: reference values of 144 plasma metabolites in a healthy population² and 588 lipids in the National Institute of Standards and Technology’s (NIST) reference plasma sample³ have been reported previously. The plasma metabolome also exhibits vast interindividual and intraindividual variability, which is explained by variation in the microbiome, dietary habits and genetics⁴. Challenges in consistently reporting the human plasma metabolome arise from interlaboratory variation in non-targeted LC–MS methodologies and varying practices and capabilities in the annotation of the metabolites themselves⁵.

To construct Plasma Benchmark, three participating laboratories analysed the same in-house pooled plasma and the NIST1950 human reference plasma⁶ in four analytical modes, including reversed-phase and hydrophilic interaction chromatography in the positive and negative ionization modes followed by data matrix generation in MS-DIAL⁷. We applied a set of inclusion criteria based on signal-to-noise ratio, relative standard deviation, sample-to-blank ratio, and acquisition of MS/MS data and predicted molecular formula to narrow the number of detected molecular features down to 639 robust molecular features that most probably represent actual plasma metabolites. Nearly 88% of the detected molecular features did not fulfil the signal-to-noise ratio and sample-to-blank ratio criteria, probably due to inherent noise and contaminants in LC–MS data but also because of high variability in the detection of many molecular features across laboratories⁸ (Fig. 1b). Manual curation of the robust molecular features resulted in 288 unique metabolites, whereas the rest of the molecular features were classified as redundant molecular features and in-source fragments. The inclusion of four analytical modes was crucial for an extensive coverage of metabolites because most were detected in only one mode (Fig. 1b).