Abstract
Liquid chromatography-high resolution mass spectrometry (LC-MS) has emerged as one of the most widely used platforms for untargeted metabolomics due to its unparalleled sensitivity and metabolite coverage. Despite its prevalence of use, the proportion of true metabolites identified in a given experiment compared to background contaminants and ionization-generated artefacts remains poorly understood. Salt clusters are well documented artefacts of electrospray ionization MS, recognized by their characteristically high mass defects (for this work simply generalized as the decimal numbers after the nominal mass). Exploiting this property, we developed a method to identify and remove salt clusters from LC-MS-based human metabolomics data using mass defect filtering. By comparing the complete set of endogenous metabolites in the human metabolome database to actual plasma, urine and stool samples, we demonstrate that up to 28.5 % of detected features are likely salt clusters. These clusters occur irrespective of ionization mode, column type, sweep gas and sample type, but can be easily removed post-acquisition using a set of R functions presented here. Our mass defect filter removes unwanted noise from LC-MS metabolomics datasets, while retaining true metabolites, and requires only a list of m/z and retention time values. Reducing the number of features prior to statistical analyses will result in more accurate multivariate modeling and differential feature selection, as well as decreased reporting of unknowns that often constitute the largest proportion of human metabolomics data.Electronic supplementary materialThe online version of this article (doi:10.1186/s13321-016-0156-0) contains supplementary material, which is available to authorized users.
Highlights
Liquid chromatography-high resolution mass spectrometry (LC-MS) has emerged as one of the most widely used platforms for untargeted metabolomics due to its unparalleled sensitivity and metabolite coverage
Untargeted metabolomics has a wide array of applications, from biomarker discovery, to elucidating disease mechanisms, and characterizing the function of microbial communities
Upon further investigation, we noted that a large proportion of significant features were not endogenous metabolites, but rather salt clusters composed of different combinations of potassium and/or sodium, with chloride and/or formate anions
Summary
Given the ubiquity of salt clusters in LC-MS data [1–4], and their predictable mass defect, we developed a method to identify and remove salt cluster artefacts from untargeted LC-MS data using mass defect filtering This comprised performing a linear regression of compounds with the highest mass defect in the hmdb (Fig. 2), modelling CnHn+2 alkanes, which represent the theoretical maxima mass defect for compounds containing only carbon, hydrogen, oxygen, and/or nitrogen. The metabolites of interest contained multiply-charged peptides, which were not modelled by the hmdb dataset (Fig. 2) Some of these peptides occupied the same mass defect space as the salt clusters, and would be removed from the analysis by our original, ‘mass defect only’ method (Fig. 3a, b). To determine if the proportion of salt clusters could be reduced instrumentally, and if they occurred in other biological matrices, we ran a series of tests comparing the effect of sweep gas and column type (reverse phase or HILIC) on salt cluster formation in a set of three
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
