FiehnLib: mass spectral and retention index libraries for metabolomics based on quadrupole and time-of-flight gas chromatography/mass spectrometry

Abstract

At least two independent parameters are necessary for compound identification in metabolomics. We have compiled 2 212 electron impact mass spectra and retention indices for quadrupole and time-of-flight gas chromatography/mass spectrometry (GC/MS) for over 1000 primary metabolites below 550 Da, covering lipids, amino acids, fatty acids, amines, alcohols, sugars, amino-sugars, sugar alcohols, sugar acids, organic phosphates, hydroxyl acids, aromatics, purines, and sterols as methoximated and trimethylsilylated mass spectra under electron impact ionization. Compounds were selected from different metabolic pathway databases. The structural diversity of the libraries was found to be highly overlapping with metabolites represented in the BioMeta/KEGG pathway database using chemical fingerprints and calculations using Instant-JChem. In total, the FiehnLib libraries comprised 68% more compounds and twice as many spectra with higher spectral diversity than the public Golm Metabolite Database. A range of unique compounds are present in the FiehnLib libraries that are not comprised in the 4345 trimethylsilylated spectra of the commercial NIST05 mass spectral database. The libraries can be used in conjunction with GC/MS software but also support compound identification in the public BinBase metabolomic database that currently comprises 5598 unique mass spectra generated from 19,032 samples covering 279 studies of 47 species (plants, animals, and microorganisms).

Figure 1

Derivatization by methoximation and silylation for GC-MS based metabolomics. Upper panel: Derivatization of estrone yielding two isomers (Z- and E-). Lower panel: Comparison of E,Z-trimethylsilyloxy-estrone methoxime mass spectra. Very small mass spectral differences are observed, but peaks are chromatographically resolved with 2s absolute retention time difference.

Figure 2

Calculation of structure overlaps between chemical libraries is performed by creating substructure fingerprints which can be used in principal components analyses (PCA) or Tanimoto similarity calculations.

Figure 3

Principal component analysis (PCA) of chemical hashed fingerprints between the KEGG metabolite database and FiehnLib. The graph demonstrates a high chemical complexity without forming distinct clusters for KEGG metabolites. Fewer compounds are comprised in the FiehnLib retention index/mass spectral library but show a similar structural diversity and overlap density compared to KEGG structures.

Figure 4

Overlap of the two mass spectral databases FiehnLib and GolmDB using principal component analysis (PCA). The PCA input matrix was based on mass spectral features. Points which overlap have the same peak similarity; the FiehnLib shows a more diverse distribution of mass spectra, while the Golm DB clusters more strongly.

Figure 5

Results of FiehnLib applications in SetupX/BinBase queries across nine studies (846 samples). Upper panel: Unsupervised Principal Component Analysis graphs on 324 identified unique metabolites. For each species/organ combination, median normalized intensities were calculated including all metabolites that were positively detected in minimum 10% of all samples of a class. Mammalian samples are labeled in red, urine in yellow, E. coli in blue and plants and algae in green. Lower panel: Venn diagram on a subset of 251 mammalian metabolites identified in plasma, urine and intestinal effluent metabolome. 36% of these metabolites were identified in all three body fluids such as inositol, alanine, palmitate, uric acid, creatinine, glucose and glycerol-alpha-phosphate. 37% of the compounds were only positively detected in two fluids, and 27% were unique for a specific fluid.