PeakDetective: A Semisupervised Deep Learning-Based Approach for Peak Curation in Untargeted Metabolomics.

Peak-detection algorithms currently used to process untargeted metabolomics data were designed to maximize sensitivity at the sacrifice of selectively. Peak lists returned by conventional software tools therefore contain a high density of artifacts that do not represent real chemical analytes, which, in turn, hinder downstream analyses. Although some innovative approaches to remove artifacts have recently been introduced, they involve extensive user intervention due to the diversity of peak shapes present within and across metabolomics data sets. To address this bottleneck in metabolomics data processing, we developed a semisupervised deep learning-based approach, PeakDetective, for classification of detected peaks as artifacts or true peaks. Our approach utilizes two techniques for artifact removal. First, an unsupervised autoencoder is used to extract a low-dimensional, latent representation of each peak. Second, a classifier is trained with active learning to discriminate between artifacts and true peaks. Through active learning, the classifier is trained with less than 100 user-labeled peaks in a matter of minutes. Given the speed of its training, PeakDetective can be rapidly tailored to specific LC/MS methods and sample types to maximize performance on each type of data set. In addition to curation, the trained models can also be utilized for peak detection to immediately detect peaks with both high sensitivity and selectivity. We validated PeakDetective on five diverse LC/MS data sets, where PeakDetective showed greater accuracy compared to current approaches. When applied to a SARS-CoV-2 data set, PeakDetective enabled more statistically significant metabolites to be detected. PeakDetective is open source and available as a Python package at https://github.com/pattilab/PeakDetective.

Using mass spectrometry imaging to map fluxes quantitatively in the tumor ecosystem.

Tumors are comprised of a multitude of cell types spanning different microenvironments. Mass spectrometry imaging (MSI) has the potential to identify metabolic patterns within the tumor ecosystem and surrounding tissues, but conventional workflows have not yet fully integrated the breadth of experimental techniques in metabolomics. Here, we combine MSI, stable isotope labeling, and a spatial variant of Isotopologue Spectral Analysis to map distributions of metabolite abundances, nutrient contributions, and metabolic turnover fluxes across the brains of mice harboring GL261 glioma, a widely used model for glioblastoma. When integrated with MSI, the combination of ion mobility, desorption electrospray ionization, and matrix assisted laser desorption ionization reveals alterations in multiple anabolic pathways. De novo fatty acid synthesis flux is increased by approximately 3-fold in glioma relative to surrounding healthy tissue. Fatty acid elongation flux is elevated even higher at 8-fold relative to surrounding healthy tissue and highlights the importance of elongase activity in glioma.

Cytosolic and mitochondrial NADPH fluxes are independently regulated.

Although nicotinamide adenine dinucleotide phosphate (NADPH) is produced and consumed in both the cytosol and mitochondria, the relationship between NADPH fluxes in each compartment has been difficult to assess due to technological limitations. Here we introduce an approach to resolve cytosolic and mitochondrial NADPH fluxes that relies on tracing deuterium from glucose to metabolites of proline biosynthesis localized to either the cytosol or mitochondria. We introduced NADPH challenges in either the cytosol or mitochondria of cells by using isocitrate dehydrogenase mutations, administering chemotherapeutics or with genetically encoded NADPH oxidase. We found that cytosolic challenges influenced NADPH fluxes in the cytosol but not NADPH fluxes in mitochondria, and vice versa. This work highlights the value of using proline labeling as a reporter system to study compartmentalized metabolism and reveals that NADPH homeostasis in the cytosolic and mitochondrial locations of a cell are independently regulated, with no evidence for NADPH shuttle activity.

A metabolomic signature of the APOE2 allele.

With the goal of identifying metabolites that significantly correlate with the protective e2 allele of the apolipoprotein E (APOE) gene, we established a consortium of five studies of healthy aging and extreme human longevity with 3545 participants. This consortium includes the New England Centenarian Study, the Baltimore Longitudinal Study of Aging, the Arivale study, the Longevity Genes Project/LonGenity studies, and the Long Life Family Study. We analyzed the association between APOE genotype groups E2 (e2e2 and e2e3 genotypes, N = 544), E3 (e3e3 genotypes, N = 2299), and E4 (e3e4 and e4e4 genotypes, N = 702) with metabolite profiles in the five studies and used fixed effect meta-analysis to aggregate the results. Our meta-analysis identified a signature of 19 metabolites that are significantly associated with the E2 genotype group at FDR < 10%. The group includes 10 glycerolipids and 4 glycerophospholipids that were all higher in E2 carriers compared to E3, with fold change ranging from 1.08 to 1.25. The organic acid 6-hydroxyindole sulfate, previously linked to changes in gut microbiome that were reflective of healthy aging and longevity, was also higher in E2 carriers compared to E3 carriers. Three sterol lipids and one sphingolipid species were significantly lower in carriers of the E2 genotype group. For some of these metabolites, the effect of the E2 genotype opposed the age effect. No metabolites reached a statistically significant association with the E4 group. This work confirms and expands previous results connecting the APOE gene to lipid regulation and suggests new links between the e2 allele, lipid metabolism, aging, and the gut-brain axis.

An Untargeted Metabolomics Workflow that Scales to Thousands of Samples for Population-Based Studies.

The success of precision medicine relies upon collecting data from many individuals at the population level. Although advancing technologies have made such large-scale studies increasingly feasible in some disciplines such as genomics, the standard workflows currently implemented in untargeted metabolomics were developed for small sample numbers and are limited by the processing of liquid chromatography/mass spectrometry data. Here we present an untargeted metabolomics workflow that is designed to support large-scale projects with thousands of biospecimens. Our strategy is to first evaluate a reference sample created by pooling aliquots of biospecimens from the cohort. The reference sample captures the chemical complexity of the biological matrix in a small number of analytical runs, which can subsequently be processed with conventional software such as XCMS. Although this generates thousands of so-called features, most do not correspond to unique compounds from the samples and can be filtered with established informatics tools. The features remaining represent a comprehensive set of biologically relevant reference chemicals that can then be extracted from the entire cohort's raw data on the basis of m/z values and retention times by using Skyline. To demonstrate applicability to large cohorts, we evaluated >2000 human plasma samples with our workflow. We focused our analysis on 360 identified compounds, but we also profiled >3000 unknowns from the plasma samples. As part of our workflow, we tested 14 different computational approaches for batch correction and found that a random forest-based approach outperformed the others. The corrected data revealed distinct profiles that were associated with the geographic location of participants.

Saturation of the mitochondrial NADH shuttles drives aerobic glycolysis in proliferating cells.

Proliferating cells exhibit a metabolic phenotype known as "aerobic glycolysis," which is characterized by an elevated rate of glucose fermentation to lactate irrespective of oxygen availability. Although several theories have been proposed, a rationalization for why proliferating cells seemingly waste glucose carbon by excreting it as lactate remains elusive. Using the NCI-60 cell lines, we determined that lactate excretion is strongly correlated with the activity of mitochondrial NADH shuttles, but not proliferation. Quantifying the fluxes of the malate-aspartate shuttle (MAS), the glycerol 3-phosphate shuttle (G3PS), and lactate dehydrogenase under various conditions demonstrated that proliferating cells primarily transform glucose to lactate when glycolysis outpaces the mitochondrial NADH shuttles. Increasing mitochondrial NADH shuttle fluxes decreased glucose fermentation but did not reduce the proliferation rate. Our results reveal that glucose fermentation, a hallmark of cancer, is a secondary consequence of MAS and G3PS saturation rather than a unique metabolic driver of cellular proliferation.

A Workflow to Perform Targeted Metabolomics at the Untargeted Scale on a Triple Quadrupole Mass Spectrometer.

The thousands of features commonly observed when performing untargeted metabolomics with quadrupole time-of-flight (QTOF) and Orbitrap mass spectrometers often correspond to only a few hundred unique metabolites of biological origin, which is in the range of what can be assayed in a single targeted metabolomics experiment by using a triple quadrupole (QqQ) mass spectrometer. A major benefit of performing targeted metabolomics with QqQ mass spectrometry is the affordability of the instruments relative to high-resolution QTOF and Orbitrap platforms. Optimizing targeted methods to profile hundreds of metabolites on a QqQ mass spectrometer, however, has historically been limited by the availability of authentic standards, particularly for "unknowns" that have yet to be structurally identified. Here, we report a strategy to develop multiple reaction monitoring (MRM) methods for QqQ instruments on the basis of high-resolution spectra, thereby enabling us to use data from untargeted metabolomics to design targeted experiments without the need for authentic standards. We demonstrate that using high-resolution fragmentation data alone to design MRM methods results in the same quantitative performance as when methods are optimized by measuring authentic standards on QqQ instruments, as is conventionally done. The approach was validated by showing that Orbitrap ID-X data can be used to establish MRM methods on a Thermo TSQ Altis and two Agilent QqQs for hundreds of metabolites, including unknowns, without a dependence on standards. Finally, we highlight an application where metabolite profiling was performed on an ID-X and a QqQ by using the strategy introduced here, with both data sets yielding the same result. The described approach therefore allows us to use QqQ instruments, which are often associated with targeted metabolomics, to profile knowns and unknowns at a comprehensive scale that is typical of untargeted metabolomics.

DecoID improves identification rates in metabolomics through database-assisted MS/MS deconvolution.

Chimeric MS/MS spectra contain fragments from multiple precursor ions and therefore hinder compound identification in metabolomics. Historically, deconvolution of these chimeric spectra has been challenging and relied on specific experimental methods that introduce variation in the ratios of precursor ions between multiple tandem mass spectrometry (MS/MS) scans. DecoID provides a complementary, method-independent approach where database spectra are computationally mixed to match an experimentally acquired spectrum by using LASSO regression. We validated that DecoID increases the number of identified metabolites in MS/MS datasets from both data-independent and data-dependent acquisition without increasing the false discovery rate. We applied DecoID to publicly available data from the MetaboLights repository and to data from human plasma, where DecoID increased the number of identified metabolites from data-dependent acquisition data by over 30% compared to direct spectral matching. DecoID is compatible with any user-defined MS/MS database and provides automated searching for some of the largest MS/MS databases currently available.

Longitudinal metabolomics of human plasma reveals prognostic markers of COVID-19 disease severity.

There is an urgent need to identify which COVID-19 patients will develop life-threatening illness so that medical resources can be optimally allocated and rapid treatment can be administered early in the disease course, when clinical management is most effective. To aid in the prognostic classification of disease severity, we perform untargeted metabolomics on plasma from 339 patients, with samples collected at six longitudinal time points. Using the temporal metabolic profiles and machine learning, we build a predictive model of disease severity. We discover that a panel of metabolites measured at the time of study entry successfully determines disease severity. Through analysis of longitudinal samples, we confirm that most of these markers are directly related to disease progression and that their levels return to baseline upon disease recovery. Finally, we validate that these metabolites are also altered in a hamster model of COVID-19.

Isotope tracing in adult zebrafish reveals alanine cycling between melanoma and liver.

The cell-intrinsic nature of tumor metabolism has become increasingly well characterized. The impact that tumors have on systemic metabolism, however, has received less attention. Here, we used adult zebrafish harboring BRAFV600E-driven melanoma to study the effect of cancer on distant tissues. By applying metabolomics and isotope tracing, we found that melanoma consume ~15 times more glucose than other tissues measured. Despite this burden, circulating glucose levels were maintained in disease animals by a tumor-liver alanine cycle. Excretion of glucose-derived alanine from tumors provided a source of carbon for hepatic gluconeogenesis and allowed tumors to remove excess nitrogen from branched-chain amino acid catabolism, which we found to be activated in zebrafish and human melanoma. Pharmacological inhibition of the tumor-liver alanine cycle in zebrafish reduced tumor burden. Our findings underscore the significance of metabolic crosstalk between tumors and distant tissues and establish the adult zebrafish as an attractive model to study such processes.

Targeting unique biological signals on the fly to improve MS/MS coverage and identification efficiency in metabolomics.

When using liquid chromatography/mass spectrometry (LC/MS) to perform untargeted metabolomics, it is common to detect thousands of features from a biological extract. Although it is impractical to collect non-chimeric MS/MS data for each in a single chromatographic run, this is generally unnecessary because most features do not correspond to unique metabolites of biological relevance. Here we show that relatively simple data-processing strategies that can be applied on the fly during acquisition of data with an Orbitrap ID-X, such as blank subtraction and well-established adduct or isotope calculations, decrease the number of features to target for MS/MS analysis by up to an order of magnitude for various types of biological matrices. We demonstrate that annotating these non-biological contaminants and redundancies in real time during data acquisition enables comprehensive MS/MS data to be acquired on each remaining feature at a single collision energy. To ensure that an appropriate collision energy is applied, we introduce a method using a series of hidden ion-trap scans in an Orbitrap ID-X to find an optimal value for each feature that can then be applied in a subsequent high-resolution Orbitrap scan. Data from 100 metabolite standards indicate that this real-time optimization of collision energies leads to more informative MS/MS patterns compared to using a single fixed collision energy alone. As a benchmark to evaluate the overall workflow, we manually annotated unique biological features by independently subjecting E. coli samples to a credentialing analysis. While credentialing led to a more rigorous reduction in feature number, on-the-fly annotation with blank subtraction on an Orbitrap ID-X did not inappropriately discard unique biological metabolites. Taken together, our results reveal that optimal fragmentation data can be obtained in a single LC/MS/MS run for >90% of the unique biological metabolites in a sample when features are annotated during acquisition and collision energies are selected by using parallel mass spectrometry detection.

Longitudinal Metabolomics of Human Plasma Reveals Robust Prognostic Markers of COVID-19 Disease Severity.

There is an urgent need to identify which COVID-19 patients will develop life-threatening illness so that scarce medical resources can be optimally allocated and rapid treatment can be administered early in the disease course, when clinical management is most effective. To aid in the prognostic classification of disease severity, we performed untargeted metabolomics profiling of 341 patients with plasma samples collected at six longitudinal time points. Using the temporal metabolic profiles and machine learning, we then built a predictive model of disease severity. We determined that the levels of 25 metabolites measured at the time of hospital admission successfully predict future disease severity. Through analysis of longitudinal samples, we confirmed that these prognostic markers are directly related to disease progression and that their levels are restored to baseline upon disease recovery. Finally, we validated that these metabolites are also altered in a hamster model of COVID-19. Our results indicate that metabolic changes associated with COVID-19 severity can be effectively used to stratify patients and inform resource allocation during the pandemic.

Dose-Response Metabolomics To Understand Biochemical Mechanisms and Off-Target Drug Effects with the TOXcms Software.

Small-molecule drugs and toxicants commonly interact with more than a single protein target, each of which may have unique effects on cellular phenotype. Although untargeted metabolomics is often applied to understand the mode of action of these chemicals, simple pairwise comparisons of treated and untreated samples are insufficient to resolve the effects of disrupting two or more independent protein targets. Here, we introduce a workflow for dose-response metabolomics to evaluate chemicals that potentially affect multiple proteins with different potencies. Our approach relies on treating samples with various concentrations of compound prior to analysis with mass spectrometry-based metabolomics. Data are then processed with software we developed called TOXcms, which statistically evaluates dose-response trends for each metabolomic signal according to user-defined tolerances and subsequently groups those that follow the same pattern. Although TOXcms was built upon the XCMS framework, it is compatible with any metabolomic data-processing software. Additionally, to enable correlation of dose responses beyond those that can be measured by metabolomics, TOXcms also accepts data from respirometry, cell death assays, other omic platforms, etc. In this work, we primarily focus on applying dose-response metabolomics to find off-target effects of drugs. Using metformin and etomoxir as examples, we demonstrate that each group of dose-response patterns identified by TOXcms signifies a metabolic response to a different protein target with a unique drug binding affinity. TOXcms is freely available on our laboratory website at http://pattilab.wustl.edu/software/toxcms .

A simplified metabolic network reconstruction to promote understanding and development of flux balance analysis tools.

GEnome-scale Network REconstructions (GENREs) mathematically describe metabolic reactions of an organism or a specific cell type. GENREs can be used with a number of constraint-based reconstruction and analysis (COBRA) methods to make computational predictions on how a system changes in different environments. We created a simplified GENRE (referred to as iSIM) that captures central energy metabolism with nine metabolic reactions to illustrate the use of and promote the understanding of GENREs and constraint-based methods. We demonstrate the simulation of single and double gene deletions, flux variability analysis (FVA), and test a number of metabolic tasks with the GENRE. Code to perform these analyses is provided in Python, R, and MATLAB. Finally, with iSIM as a guide, we demonstrate how inaccuracies in GENREs can limit their use in the interrogation of energy metabolism.