Identification and Annotation of Lipid Species in Metabolomics Studies Need Improvement

To the Editor: Lipids, which play important roles in health and disease, encompass a diverse group of biomolecules with a wide range of structural attributes and chemical properties. Individual species are typically identified and quantified with mass spectrometry (MS)-based1 lipidomics platforms using dedicated sample preparation routines, and synthetic lipid standards enabling gold-standard absolute lipid quantification. Lipids can also be profiled with untargeted metabolomics analysis, in which no synthetic lipid standards are applied and lipid abundance is inferred solely from arbitrary intensity values. Regardless of the analytical approach, the reporting of new lipid biomarkers should be validated by use of “good biomarker practices” that unequivocally demonstrate the validity and fidelity of biomarkers. Here, we question the lipid data presented in a recent study by Drogan et al. (1). In particular, we question the identities of the reported lipid biomarkers and the lack of validation thereof. Drogan et al. applied ultraperformance liquid chromatography coupled with an Orbitrap mass spectrometer to record spectral features of metabolites in deproteinized serum samples in both positive and negative ion mode. Identification of spectral features was performed by matching recorded m/z values, within 5-ppm mass tolerance, to putative metabolites and lipid molecules with the software tool PUTMEDID_LCMS. Drogan et al. (1) reported an association between type 2 diabetes and 6 spectral features matching various lipid molecules (Table 1). Surprisingly, the authors provided limited information on identities, abundances, and chromatographic properties, which in our opinion is inadequate for reporting and validating biomarkers. We examined the 6 spectral features and their respective lipid identities. Alarmingly, we discovered that only 1, m/z 786.564 (RT 914), could be correctly matched to the identities reported (Table 1). No information was provided as to how the other 5 features were matched to the short-listed lipid identities, which clearly have m/z values outside the 5-ppm mass tolerance. We also noted that 2 spectral features have odd-numbered m/z values: m/z 525.338 (RT 707) matched to lysophosphatidylcholine(dm16:0) and m/z 811.609 (RT 957) matched to various isobaric phosphatidylcholine (PC) and phosphatidylethanolamine (PE) species. These lipid identities are impossible because the elemental composition of lysoPC, PC, and PE molecules yield even-numbered m/z values. Given that the 6 spectral features are matched primarily to PC and PE lipids, which readily ionize in both positive and negative ion mode, it is highly disconcerting that none of the reported lipid biomarkers show a clear type 2 diabetes association in both polarities. Assessment of spectral features and lipid species reported to be associated with type 2 diabetes by Drogan et al. (1).a We attempted to match reported m/z values to lipid species in the LIPID MAPS Structure Database (http://www.lipidmaps.org/data/structure/index.html). We corrected the inappropriate lipid nomenclature used by Drogan et al. to the more correct sum composition nomenclature [Liebisch et al. (2)]. Bold text indicates matches of the spectral features with lipid species retrieved from the LIPID MAPS database or reported by Drogan et al. Data or information in the 4 columns below correspond to details reported by Drogan et al. Lipid species retrieved from the LIPID MAPS database using a tolerance of 5 ppm. Lipid species according to accepted annotation [Liebisch et al. (2)]. Lipid species denoted with dm (dimethyl) by Drogan et al. are interpreted as lipid molecules featuring an odd number of methylene groups. For example, LysoPC(dm16:0) corresponds to the odd-numbered lipid species LPC 15:0. Calculated m/z values based on the most common adduct ions observed for ultraperformance LC-MS analysis by use of water and methanol containing 0.1% formic acid. Isobaric to the formate adduct of PE O-37:3, dimethyl PE (DMPE) O-36:3, monomethyl PE (MMPE) O-35:3, and deprotonated phosphatidylserine (PS) O-37:2. Isobaric to the formate adduct of PE O-41:3, DMPE O-40:3, MMPE O-39:3, and deprotonated PS O-41:2. Assessment of spectral features and lipid species reported to be associated with type 2 diabetes by Drogan et al. (1).a We attempted to match reported m/z values to lipid species in the LIPID MAPS Structure Database (http://www.lipidmaps.org/data/structure/index.html). We corrected the inappropriate lipid nomenclature used by Drogan et al. to the more correct sum composition nomenclature [Liebisch et al. (2)]. Bold text indicates matches of the spectral features with lipid species retrieved from the LIPID MAPS database or reported by Drogan et al. Data or information in the 4 columns below correspond to details reported by Drogan et al. Lipid species retrieved from the LIPID MAPS database using a tolerance of 5 ppm. Lipid species according to accepted annotation [Liebisch et al. (2)]. Lipid species denoted with dm (dimethyl) by Drogan et al. are interpreted as lipid molecules featuring an odd number of methylene groups. For example, LysoPC(dm16:0) corresponds to the odd-numbered lipid species LPC 15:0. Calculated m/z values based on the most common adduct ions observed for ultraperformance LC-MS analysis by use of water and methanol containing 0.1% formic acid. Isobaric to the formate adduct of PE O-37:3, dimethyl PE (DMPE) O-36:3, monomethyl PE (MMPE) O-35:3, and deprotonated phosphatidylserine (PS) O-37:2. Isobaric to the formate adduct of PE O-41:3, DMPE O-40:3, MMPE O-39:3, and deprotonated PS O-41:2. While assessing the reported lipid biomarkers, we further noted use of inappropriate lipid nomenclature. The annotation applied by Drogan et al. assumes complete knowledge about the molecular structures of identified lipid molecules, but their methodology does not support such extrapolations. For example, identification of the lipid species PC(O-16:0/18:3) (Table 1) requires tandem MS and chromatographic behavior identical to a synthetically analogous standard. We argue that Drogan et al. should use the more appropriate sum composition nomenclature that denotes only the lipid class and total number of carbon atoms and double bonds in fatty acid moieties (e.g., PC O-34:3) (2) (Table 1). To independently establish whether the 6 reported spectral features could be assigned to any lipid molecules, we performed a search in the LIPID MAPS database (http://www.lipidmaps.org/) (Table 1). We could match only 2 of the spectral features reported, m/z 786.565 (RT 914) and m/z 842.628 (RT 951), to the formate adducts of PC O-34:3 and PC O-38:3, respectively. Notably, LIPID MAPS profiling of human plasma (3) detected only PC O-38:3 but no PC O-34:3. In our own data (4), we have detected PC O-34:3, but at very low concentrations. Here, we raise our concern about the validity of the 6 lipid biomarkers reported by Drogan et al. (1). We highlight their improper naming of these putative lipid molecules and provide a more correct annotation. We argue that reporting of lipid biomarkers, especially by untargeted metabolomics analysis, should adhere to good biomarker practices, with reporting standards that allow the scientific community to unequivocally assess the validity and fidelity of candidate (lipid) biomarkers. Such practices and reporting standards should, at a minimum, include combinations of tandem MS in both positive and negative ion mode, if possible, and chromatographic retention times of the candidate (lipid) biomarkers matched against identical synthetic standards. Moreover, the absolute amounts of (lipid) biomarkers should be quantified by use of dedicated assays that support gold-standard absolute quantification providing accurate measurements of differences between study groups. We encourage editors and reviewers to demand more rigorous standards for validation of (lipid) biomarkers (5) and also to require appropriate use of lipid nomenclature (2). mass spectrometry phosphatidylcholine phosphatidylethanolamine.