Metabolomics-driven discoveries of biological samples remain hampered by the grand challenge of metabolite annotation and identification. Only few metabolites have an annotated spectrum in spectral libraries; hence, searching only for exact library matches generally returns a few hits. An attractive alternative is searching for so-called analogues as a starting point for structural annotations; analogues are library molecules which are not exact matches but display a high chemical similarity. However, current analogue search implementations are not yet very reliable and relatively slow. Here, we present MS2Query, a machine learning-based tool that integrates mass spectral embedding-based chemical similarity predictors (Spec2Vec and MS2Deepscore) as well as detected precursor masses to rank potential analogues and exact matches. Benchmarking MS2Query on reference mass spectra and experimental case studies demonstrate improved reliability and scalability. Thereby, MS2Query offers exciting opportunities to further increase the annotation rate of metabolomics profiles of complex metabolite mixtures and to discover new biology.

Large-scale metabolic profiling requires the development of novel economical high-throughput analytical methods to facilitate characterization of systemic metabolic variation in population phenotypes. We report a fit-for-purpose direct infusion nanoelectrospray high-resolution mass spectrometry (DI-nESI-HRMS) method with time-of-flight detection for rapid targeted parallel analysis of over 40 urinary metabolites. The newly developed 2 min infusion method requires 10 000 24 h urine samples from the INTERMAP study in 12 weeks and >2200 spot urine samples from the ARIC study in <3 weeks with the required sensitivity and accuracy. We illustrate the application of the technique by characterizing the differences in metabolic phenotypes of the USA and Japanese population from the INTERMAP study.

Results from observational studies regarding associations between fish (including shellfish) intake and cardiovascular disease risk factors, including blood pressure (BP) and BMI, are inconsistent. To investigate associations of fish consumption and associated urinary metabolites with BP and BMI in free-living populations. We used cross-sectional data from the International Study of Macro-/Micronutrients and Blood Pressure (INTERMAP), including 4680 men and women (40–59 y) from Japan, China, the United Kingdom, and United States. Dietary intakes were assessed by four 24-h dietary recalls and BP from 8 measurements. Urinary metabolites (2 timed 24-h urinary samples) associated with fish intake acquired from NMR spectroscopy were identified. Linear models were used to estimate BP and BMI differences across categories of intake and per 2 SD higher intake of fish and its biomarkers. No significant associations were observed between fish intake and BP. There was a direct association with fish intake and BMI in the Japanese population sample (P trend = 0.03; fully adjusted model). In Japan, trimethylamine-N-oxide (TMAO) and taurine, respectively, demonstrated area under the receiver operating characteristic curve (AUC) values of 0.81 and 0.78 in discriminating high against low fish intake, whereas homarine (a metabolite found in shellfish muscle) demonstrated an AUC of 0.80 for high/nonshellfish intake. Direct associations were observed between urinary TMAO and BMI for all regions except Japan (P < 0.0001) and in Western populations between TMAO and BP (diastolic blood pressure: mean difference 1.28; 95% CI: 0.55, 2.02 mmHg; P = 0.0006, systolic blood pressure: mean difference 1.67; 95% CI: 0.60, 2.73 mmHg; P = 0.002). Urinary TMAO showed a stronger association with fish intake in the Japanese compared with the Western population sample. Urinary TMAO was directly associated with BP in the Western but not the Japanese population sample. Associations between fish intake and its biomarkers and downstream associations with BP/BMI appear to be context specific. INTERMAP is registered at www.clinicaltrials.gov as NCT00005271.

Liquid chromatography–mass spectrometry (LC-MS) is a powerful and widely used technique for measuring the abundance of chemical species in living systems. Its sensitivity, analytical specificity, and direct applicability to biofluids and tissue extracts impart great promise for the discovery and mechanistic characterization of biomarker panels for disease detection, health monitoring, patient stratification, and treatment personalization. Global metabolic profiling applications yield complex data sets consisting of multiple feature measurements for each chemical species observed. While this multiplicity can be useful in deriving enhanced analytical specificity and chemical identities from LC-MS data, data set inflation and quantitative imprecision among related features is problematic for statistical analyses and interpretation. This Perspective provides a critical evaluation of global profiling data fidelity with respect to measurement linearity and the quantitative response variation observed among components of the spectra. These elements of data quality are widely overlooked in untargeted metabolomics yet essential for the generation of data that accurately reflect the metabolome. Advanced feature filtering informed by linear range estimation and analyte response factor assessment is advocated as an attainable means of controlling LC-MS data quality in global profiling studies and exemplified herein at both the feature and data set level.

Genome-wide association studies (GWASs) have identified genetic loci associated with the risk of Alzheimer’s disease (AD), but the molecular mechanisms by which they confer risk are largely unknown. We conducted a metabolome-wide association study (MWAS) of AD-associated loci from GWASs using untargeted metabolic profiling (metabolomics) by ultraperformance liquid chromatography–mass spectrometry (UPLC-MS). We identified an association of lactosylceramides (LacCer) with AD-related single-nucleotide polymorphisms (SNPs) in ABCA7 ( P = 5.0 × 10 ⁻⁵ to 1.3 × 10 ⁻⁴⁴ ). We showed that plasma LacCer concentrations are associated with cognitive performance and genetically modified levels of LacCer are associated with AD risk. We then showed that concentrations of sphingomyelins, ceramides, and hexosylceramides were altered in brain tissue from Abca7 knockout mice, compared with wild type (WT) ( P = 0.049–1.4 × 10 ⁻⁵ ), but not in a mouse model of amyloidosis. Furthermore, activation of microglia increases intracellular concentrations of hexosylceramides in part through induction in the expression of sphingosine kinase, an enzyme with a high control coefficient for sphingolipid and ceramide synthesis. Our work suggests that the risk for AD arising from functional variations in ABCA7 is mediated at least in part through ceramides. Modulation of their metabolism or downstream signaling may offer new therapeutic opportunities for AD.

Integration of multiple datasets can greatly enhance bioanalytical studies, for example, by increasing power to discover and validate biomarkers. In liquid chromatography–mass spectrometry (LC–MS) metabolomics, it is especially hard to combine untargeted datasets since the majority of metabolomic features are not annotated and thus cannot be matched by chemical identity. Typically, the information available for each feature is retention time (RT), mass-to-charge ratio (m/z), and feature intensity (FI). Pairs of features from the same metabolite in separate datasets can exhibit small but significant differences, making matching very challenging. Current methods to address this issue are too simple or rely on assumptions that cannot be met in all cases. We present a method to find feature correspondence between two similar LC–MS metabolomics experiments or batches using only the features’ RT, m/z, and FI. We demonstrate the method on both real and synthetic datasets, using six orthogonal validation strategies to gauge the matching quality. In our main example, 4953 features were uniquely matched, of which 585 (96.8%) of 604 manually annotated features were correct. In a second example, 2324 features could be uniquely matched, with 79 (90.8%) out of 87 annotated features correctly matched. Most of the missed annotated matches are between features that behave very differently from modeled inter-dataset shifts of RT, MZ, and FI. In a third example with simulated data with 4755 features per dataset, 99.6% of the matches were correct. Finally, the results of matching three other dataset pairs using our method are compared with a published alternative method, metabCombiner, showing the advantages of our approach. The method can be applied using M2S (Match 2 Sets), a free, open-source MATLAB toolbox, available at https://github.com/rjdossan/M2S.