Search

INTRODUCTIONSince the beginning of the SARS-CoV-2 pandemic in December 2019 multiple metabolomics studies have proposed predictive biomarkers of infection severity and outcome. Whilst some trends have emerged, the findings remain intangible and uninformative when it comes to new patients.OBJECTIVESIn this study, we accurately quantitate a subset of compounds in patient serum that were found predictive of severity and outcome.METHODSA targeted LC-MS method was used in 46 control and 95 acute COVID-19 patient samples to quantitate the selected metabolites. These compounds included tryptophan and its degradation products kynurenine and kynurenic acid (reflective of immune response), butyrylcarnitine and its isomer (reflective of energy metabolism) and finally 3’,4’-didehydro-3’-deoxycytidine, a deoxycytidine analogue, (reflective of host viral defence response). We subsequently examine changes in those markers by disease severity and outcome relative to those of control patients’ levels.RESULTS & CONCLUSIONFinally, we demonstrate the added value of the kynurenic acid / tryptophan ratio for severity and outcome prediction and highlight the viral detection potential of ddhC.

Introduction The diagnosis of COVID-19 is normally based on the qualitative detection of viral nucleic acid sequences. Properties of the host response are not measured but are key in determining outcome. Although metabolic profiles are well suited to capture host state, most metabolomics studies are either underpowered, measure only a restricted subset of metabolites, compare infected individuals against uninfected control cohorts that are not suitably matched, or do not provide a compact predictive model. Objectives Here we provide a well-powered, untargeted metabolomics assessment of 120 COVID-19 patient samples acquired at hospital admission. The study aims to predict the patient’s infection severity (i.e., mild or severe) and potential outcome (i.e., discharged or deceased). Methods High resolution untargeted UHPLC-MS/MS analysis was performed on patient serum using both positive and negative ionization modes. A subset of 20 intermediary metabolites predictive of severity or outcome were selected based on univariate statistical significance and a multiple predictor Bayesian logistic regression model was created. Results The predictors were selected for their relevant biological function and include deoxycytidine and ureidopropionate (indirectly reflecting viral load), kynurenine (reflecting host inflammatory response), and multiple short chain acylcarnitines (energy metabolism) among others. Currently, this approach predicts outcome and severity with a Monte Carlo cross validated area under the ROC curve of 0.792 (SD 0.09) and 0.793 (SD 0.08), respectively. A blind validation study on an additional 90 patients predicted outcome and severity at ROC AUC of 0.83 (CI 0.74–0.91) and 0.76 (CI 0.67–0.86). Conclusion Prognostic tests based on the markers discussed in this paper could allow improvement in the planning of COVID-19 patient treatment.

The ‘inverse problem’ of mass spectrometric molecular identification (‘given a mass spectrum, calculate/predict the 2D structure of the molecule whence it came’) is largely unsolved, and is especially acute in metabolomics where many small molecules remain unidentified. This is largely because the number of experimentally available electrospray mass spectra of small molecules is quite limited. However, the forward problem (‘calculate a small molecule’s likely fragmentation and hence at least some of its mass spectrum from its structure alone’) is much more tractable, because the strengths of different chemical bonds are roughly known. This kind of molecular identification problem may be cast as a language translation problem in which the source language is a list of high-resolution mass spectral peaks and the ‘translation’ a representation (for instance in SMILES) of the molecule. It is thus suitable for attack using the deep neural networks known as transformers. We here present MassGenie, a method that uses a transformer-based deep neural network, trained on ~6 million chemical structures with augmented SMILES encoding and their paired molecular fragments as generated in silico, explicitly including the protonated molecular ion. This architecture (containing some 400 million elements) is used to predict the structure of a molecule from the various fragments that may be expected to be observed when some of its bonds are broken. Despite being given essentially no detailed nor explicit rules about molecular fragmentation methods, isotope patterns, rearrangements, neutral losses, and the like, MassGenie learns the effective properties of the mass spectral fragment and valency space, and can generate candidate molecular structures that are very close or identical to those of the ‘true’ molecules. We also use VAE-Sim, a previously published variational autoencoder, to generate candidate molecules that are ‘similar’ to the top hit. In addition to using the ‘top hits’ directly, we can produce a rank order of these by ‘round-tripping’ candidate molecules and comparing them with the true molecules, where known. As a proof of principle, we confine ourselves to positive electrospray mass spectra from molecules with a molecular mass of 500Da or lower, including those in the last CASMI challenge (for which the results are known), getting 49/93 (53%) precisely correct. The transformer method, applied here for the first time to mass spectral interpretation, works extremely effectively both for mass spectra generated in silico and on experimentally obtained mass spectra from pure compounds. It seems to act as a Las Vegas algorithm, in that it either gives the correct answer or simply states that it cannot find one. The ability to create and to ‘learn’ millions of fragmentation patterns in silico, and therefrom generate candidate structures (that do not have to be in existing libraries) directly, thus opens up entirely the field of de novo small molecule structure prediction from experimental mass spectra.

The ‘inverse problem’ of mass spectrometric molecular identification (‘given a mass spectrum, calculate the molecule whence it came’) is largely unsolved, and is especially acute in metabolomics where many small molecules remain unidentified. This is largely because the number of experimentally available electrospray mass spectra of small molecules is quite limited. However, the forward problem (‘calculate a small molecule’s likely fragmentation and hence at least some of its mass spectrum from its structure alone’) is much more tractable, because the strengths of different chemical bonds are roughly known. This kind of molecular identification problem may be cast as a language translation problem in which the source language is a list of high-resolution mass spectral peaks and the ‘translation’ a representation (for instance in SMILES) of the molecule. It is thus suitable for attack using the deep neural networks known as transformers. We here present MassGenie, a method that uses a transformer-based deep neural network, trained on ~6 million chemical structures with augmented SMILES encoding and their paired molecular fragments as generatedin silico, explicitly including the protonated molecular ion. This architecture (containing some 400 million elements) is used to predict the structure of a molecule from the various fragments that may be expected to be observed when some of its bonds are broken. Despite being given essentially no detailed nor explicit rules about molecular fragmentation methods, isotope patterns, rearrangements, neutral losses, and the like, MassGenie learns the effective properties of the mass spectral fragment and valency space, and can generate candidate molecular structures that are very close or identical to those of the ‘true’ molecules. We also use VAE-Sim, a previously published variational autoencoder, to generate candidate molecules that are ‘similar’ to the top hit. In addition to using the ‘top hits’ directly, we can produce a rank order of these by ‘round-tripping’ candidate molecules and comparing them with the true molecules, where known. As a proof of principle, we confine ourselves to positive electrospray mass spectra from molecules with a molecular mass of 500Da or lower. The transformer method, applied here for the first time to mass spectral interpretation, works extremely effectively both for mass spectra generatedin silicoand on experimentally obtained mass spectra from pure compounds. The ability to create and to ‘learn’ millions of fragmentation patternsin silico, and therefrom generate candidate structures (that do not have to be in existing libraries)directly, thus opens up entirely the field ofde novosmall molecule structure prediction from experimental mass spectra.

Introduction It is widely but erroneously believed that drugs get into cells by passing through the phospholipid bilayer portion of the plasma and other membranes. Much evidence shows, however, that this is not the case, and that drugs cross biomembranes by hitchhiking on transporters for other natural molecules to which these drugs are structurally similar. Untargeted metabolomics can provide a method for determining the differential uptake of such metabolites. Objectives Blood serum contains many thousands of molecules and provides a convenient source of biologically relevant metabolites. Our objective was to detect and identify metabolites present in serum, but to also establish a method capable of measure their uptake and secretion by different cell lines. Methods We develop an untargeted LC-MS/MS method to detect a broad range of compounds present in human serum. We apply this to the analysis of the time course of the uptake and secretion of metabolites in serum by several human cell lines, by analysing changes in the serum that represents the extracellular phase (the ‘exometabolome’ or metabolic footprint). Results Our method measures some 4000–5000 metabolic features in both positive and negative electrospray ionisation modes. We show that the metabolic footprints of different cell lines differ greatly from each other. Conclusion Our new, 15-min untargeted metabolome method allows for the robust and convenient measurement of differences in the uptake of serum compounds by cell lines following incubation in serum. This will enable future research to study these differences in multiple cell lines that will relate this to transporter expression, thereby advancing our knowledge of transporter substrates, both natural and xenobiotic compounds.