Download Hi-Res ImageDownload to MS-PowerPointCite This:J. Proteome Res. 2019, 18, 9, 3360-3368

ArticleJuly 18, 2019

Automated Analysis of Large-Scale NMR Data Generates Metabolomic Signatures and Links Them to Candidate Metabolites
Click to copy article linkArticle link copied!

Bita Khalili
Bita Khalili
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Bita Khalili
http://orcid.org/0000-0001-5630-1812
Mattia Tomasoni
Mattia Tomasoni
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Mattia Tomasoni
Mirjam Mattei
Mirjam Mattei
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Mirjam Mattei
Roger Mallol Parera
Roger Mallol Parera
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Roger Mallol Parera
Reyhan Sonmez
Reyhan Sonmez
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Reyhan Sonmez
Daniel Krefl
Daniel Krefl
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Daniel Krefl
Rico Rueedi
Rico Rueedi
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
More by Rico Rueedi
Sven Bergmann*
Sven Bergmann
Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland
Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
Department of Integrative Biomedical Sciences, University of Cape Town, Cape Town 7700, South Africa
*E-mail: [email protected]
More by Sven Bergmann

Open PDF Supporting Information (2)

Journal of Proteome Research

Cite this: J. Proteome Res. 2019, 18, 9, 3360–3368

Click to copy citationCitation copied!

https://doi.org/10.1021/acs.jproteome.9b00295

Published July 18, 2019

Request reuse permissions

Abstract

Click to copy section linkSection link copied!

Identification of metabolites in large-scale ¹H NMR data from human biofluids remains challenging due to the complexity of the spectra and their sensitivity to pH and ionic concentrations. In this work, we tested the capacity of three analysis tools to extract metabolite signatures from 968 NMR profiles of human urine samples. Specifically, we studied sets of covarying features derived from principal component analysis (PCA), the iterative signature algorithm (ISA), and averaged correlation profiles (ACP), a new method we devised inspired by the STOCSY approach. We used our previously developed metabomatching method to match the sets generated by these algorithms to NMR spectra of individual metabolites available in public databases. On the basis of the number and quality of the matches, we concluded that ISA and ACP can robustly identify ten and nine metabolites, respectively, half of which were shared, while PCA did not produce any signatures with robust matches.

Subjects

Keywords

Introduction

Click to copy section linkSection link copied!

Nuclear magnetic resonance (NMR) spectroscopy is a powerful technique for metabolomic profiling. NMR spectroscopy does not consume the sample and has high accuracy and reproducibility. Single proton NMR spectroscopy (¹H NMR) can be used to generate one-dimensional spectra of biofluids at high throughput and low cost, facilitating the generation of large sets of spectral data.

A first step in NMR spectral analysis is usually to identify the main metabolites giving rise to a given spectrum or set of spectra. This is nontrivial, since human biofluids typically contain a large number of individual metabolites and their corresponding peak positions may overlap and are often affected by the pH, ionic strength, and overall protein content of the fluid.

For small sets of samples, expert analysis is therefore still the most accurate means for metabolite identification, yet for large sets this approach is costly, time-consuming, and potentially less reproducible. As a result, various methods have been suggested to assist or fully automate metabolite identification.

In their landmark paper on statistical total correlation spectroscopy (STOCSY), Cloarec et al. showed that analyzing the correlation patterns between features across a sizable collection of ¹H NMR spectra has great potential for metabolite identification. (1) This is because features corresponding to the same molecule (or molecules whose concentrations covary) tend to be significantly correlated in large data sets. Analyzing data from 612 mouse urine samples, they observed that the correlation matrix exhibited correlated peaks of features characteristic of valeramide, glucose, hippurate, 2-oxoglutarate, 3-hydroxyphenylpropionate, citrate, as well as methylamine, dimethylamine, and trimethylamine.

Subsequent variations of STOCSY attempted to make clusters of NMR peaks to simplify the interpretation of the information stored in the correlation matrix from STOCSY analysis. Recoupled-STOCSY (R-STOCSY) employs a variable size bucketing method to reduce the dimensionality of NMR data and statistical recoupling of variables (SRV) to identify correlations between distant clusters. (2) Iterative-STOCSY (I-STOCSY) aims at separating the intermetabolite connections from intrametabolite connection by recursively applying STOCSY analysis first from a selected driver peak and then for all peaks correlating with the driver peak above a specific threshold. (3) Subset optimization by reference matching (STORM) selects subsets of ¹H NMR spectra that contain specific spectroscopic signatures of biomarkers differentiating between different human populations. (4)

Once metabolite identification has been achieved, the next challenge is to quantify metabolite concentrations. This process works robustly only for a relatively small set of metabolites, and requires expert refinement when using publicly available quantification tools such as BATMAN, (5) FOCUS, (6) BAYESIL, (7) ASICS, (8) AQuA, (9) or rDolphin. (10) This is unsatisfactory in light of the fact that a sizable number of metabolites have been identified in human biofluids. For example, the latest version of the Human Metabolome Database (HMDB 4.0) (11) includes more than 1500 metabolites with ¹H NMR spectra, 179 of which have been identified in urine (12) and 67 in serum. (13,14)

There are several reasons why it remains difficult to perform fully automated quantification from large-scale NMR data for the vast majority of metabolites. First, human biofluids contain a large number of individual metabolites whose concentrations vary across several orders of magnitude. This makes it difficult to disentangle the contributions of metabolites with low concentrations, in particular when their NMR features are not unique. Second, the exact feature positions depend on the biofluid and may have been different when acquiring reference spectra. Third, while the number of reference spectra continues to grow, reference databases are certainly not yet exhaustive.

In two recent studies, we demonstrated that the limitations of targeted NMR metabolomics can be addressed by linking metabolites to external variables. (15,16) Specifically, in the context of genome-wide association studies (GWAS) applied to metabolomics (known as mGWAS) the aim is to associate metabolites with genotypic variants. We observed that the effect of a genetic variant on the concentration of a metabolite often translates into associations with all or many features of the metabolite NMR spectrum. The set of association scores with all measured features provides a pseudospectrum across the full range of ppm covered by the ¹H NMR spectra. The challenge is then to identify the metabolite underlying the most significant associations. To this end we developed the analysis tool metabomatching, which takes as input a pseudospectrum and a collection of reference spectra for individual metabolites found, for example, in HMDB. (11) Our previous work showed that metabomatching works well to prioritize the most likely metabolite candidates for pseudospectra derived from metabolome feature association with genotypes. (15−17)

In the present work, we tested the metabomatching methodology for identifying metabolites that vary across large collection of samples without the need for any external variables associated with this variation. We investigated three methods to identify covarying spectral features within large-scale NMR data: principal component analysis (PCA), the iterative signature algorithm (ISA), and averaged correlation profiles (ACP) inspired by the STOCSY approach. For each method, we devised a principled way for processing their output into pseudospectra. In addition, we extended metabomatching to process the respective outputs of the methods, and implemented a permutation-based robustness test to assess the quality of the matches. This allowed us to compare the matches across different methods, and assess the consistency or complementarity of the methods. We incorporated our analysis for unsupervised generation of metabolomic signatures from large-scale NMR data and integration with metabomatching (including further documentation) into the metabomodules software tool, which is publicly available at https://github.com/BergmannLab/metabomodules-docker.

Methods

Click to copy section linkSection link copied!

Preprocessing

For this study, we used 968 ¹H NMR spectra acquired from urine samples from the CoLaus cohort. (15) The samples are referenced to the TSP signal, phase-corrected, and baseline-corrected.

We used the FOCUS (6) tool to align and bin the spectra to a resolution of about 0.02 ppm, and a correspondingly large number of 687 peaks. To obtain this resolution, we set the downsampling frequency parameter window.fs to 1 (no downsampling), the sliding window length for spectral segmentation window.length to 0.03 ppm, the minimum peak width parameter peak.DFL to 0.02 ppm, and the peak sample frequency parameter peak.pS to 0.2, keeping the rest of the parameters at their default values.

To normalize the data, we log-transformed, standardized across features (thereby normalizing the concentration of each sample), then standardized across samples (thereby making intensities comparable).

Confounding

In order to allow for the identification of metabolites that may be hidden by confounding, we additionally generated a data set of residuals, created by regressing out the confounders from the feature metabolome. The main confounding factors of the NMR data that we investigated here are age, sex, serum creatinine, and urine creatinine. (15,16)

Metabomatching

Our original metabomatching method was designed to match the NMR spectra of individual metabolites recorded in a database with pseudospectra from the association between metabolome features and an external variable, typically a SNP genotype. For a metabolite m, metabomatching computes the sum

(1)

where F_m is the set of N_f features that fall within a neighborhood of any peak of m according to the database, and z_f denotes the significance value for feature f, and is given by z_f = β̂_f/

, where β̂_f and

refer to the point estimates of the effect size and its standard error, respectively. Under the null hypothesis of normally and independently distributed z_f, the sum s follows a χ²-distribution with N_f degrees of freedom, and metabomatching defines the score for m as the negative logarithm of the nominal p-value for the sum. This score is then used to rank all tested metabolites as metabolites with more similar NMR spectra to a given pseudospectrum achieve higher scores.

In addition to pseudospectra from regression analysis, provided as columns headed by beta, se, and p, we extended metabomatching to accept pseudospectra produced by PCA, ACP, and ISA as columns headed by pca, cr, and isa respectively. For ACP pseudospectra, metabomatching translates a correlation c to a z-score with the Fisher transformation z = λ arctanh(c). For independent features,

produces z-scores with unit standard deviation, where N is the number of samples across which the correlations are computed. However, since proximal features are usually not independent, metabomatching allows for a user-provided estimate for λ (obtained from the pairwise feature–feature correlation matrix), or re-estimates λ from the given correlations. For ISA and PCA pseudospectra, metabomatching standardizes the loadings or module scores.

We used the plus/minus mode of metabomatching since features are z-scored and have positive and negative signs. This allows for detecting metabolites corresponding either to the negative or positive features (see metabomatching documentation at https://github.com/rrueedi/metabomatching for more details).

We also introduced a measure of the quality of a match, which allows to compare matches between different pseudospectra. This adjusted score is obtained by reshuffling the pseudospectrum, and defining a heuristic p-value by the number N_p of all N_r reshuffled pseudospectra that produce a metabomatching score (for any reference spectrum) higher than the metabomatching score of the input pseudospectrum with the highest ranked reference spectrum. This p-value is defined as (N_p + 1)/(N_r + 1), and the adjusted score as −log(p). We used N_r = 9999, which sets the upper limit for the adjusted score to 4.

For the reshuffling to be consistent with the structure of NMR spectra, metabomatching identifies cut points that separate the pseudospectrum into peak-preserving clusters of features and only reshuffles these clusters. The cut points are obtained as follows. Let f_i be the positions on the chemical shift axis of the metabolome features, sorted such that f_i < f_i+1, and C the set of cut points. First, metabomatching populates C with features bordering a gap, that is a region absent from the spectrum and larger than δ_gap (i.e., f_i > f_i–1 + δ_gap), with δ_gap = 0.3 ppm as default value. Next it sorts the remaining features by their corresponding absolute-valued z-scores. Starting from the feature with the lowest absolute z-score it adds features to C provided they have a distance greater than δ_min to any features already assigned to C and an absolute z-score below a threshold z_min. Default values are 0.04 ppm for δ_min and the standard deviation of all absolute z-scores across the features of a given pseudospectrum for z_min.

ACP: Averaged Correlation Profile

ACP is a greedy approach to generate a list L of feature pairs and their corresponding correlation profiles c as input for metabomatching: (1) We compute and sort all pairwise correlations C_ij between features f_i and f_j separated by at least 0.1 ppm. (2) Starting with the feature pair P = (i,j) with the highest correlation, we successively add feature pairs to L unless there is already a feature pair in L whose features are within 0.1 ppm of f_i and f_j, respectively. (3) For each feature pair in L, we define an averaged correlation profile as the average of the correlation profiles of f_i and f_j: c_k = (C_ik + C_jk)/2. The correlation profiles of strongly correlated features are similar, consequently their average is similar to both of them. Crucially, the average does not contain an element equal to 1, as c_i = c_j = (1 + C_ij)/2 < 1 given that C_ij < 1 in real data. For our analysis, we limit L to 179 averaged correlation profiles, 179 being the number of spectra in UMDB, the reference database on which metabomatching will run.

As an alternative approach, we tried agglomerative clustering of features, iteratively joining features (or sets of features) whose correlation was above a threshold C_min. At each step, we averaged joined (sets of) features into a metafeature and recomputed its correlation to all remaining (meta)features. We then built a correlation profile for each feature cluster by averaging the correlations profiles of the component features.

ISA: Iterative Signature Algorithm

ISA is a biclustering method first developed for modular analysis of gene expression data. (18,19) ISA uses a heuristic iterative procedure starting with random features to refine modules, consisting of self-consistent subsets of features and samples. Each module is defined for a set of two thresholds, determining how extreme the features and samples are allowed to be. Importantly, scanning through an array of thresholds usually identifies a set of modules (or module families) that is smaller than the number of samples or features.

We first ran the ISA algorithm to generate modules from the NMR data using the default values for the parameters except the following: (1) we changed both row and column thresholds from the default values {1, 2, 3} to {1, 2, 3, 4, 5, 6} to produce modules containing fewer rows or columns that are more likely to represent single metabolites, (2) we increased the number of seeds from 100 to 250, and (3) we lowered the correlation threshold below which ISA considers two modules equal to one another from 0.95 to 0.50 to favor diversity in modules.

We allowed feature scores to be either positive or negative, while sample scores were always positive. This means that modules can include features which have on average higher or lower intensities in the selected samples than for the remaining samples. Modules which include such a mixture are likely to represent (at least) two metabolites whose concentrations are inversely related to each other (like a substrate and its product).

This procedure generated 216 modules. To select the 179 modules as an input for metabomatching, we sorted them by the size of their basin of attraction. To measure the basin size, we ran ISA a second time with the same parameters, but on 10 000 seeds, turning off the sweeping option, and keeping all converged modules (by setting the purge option to false). We then assigned a run 2 module to the basin of the run 1 module with which it had the highest correlation (provided that correlation be greater than 0.5). We assumed that the run 1 modules attractor basin size is approximately equal to the count of modules from run 2 which were assigned to them in the previous step. Finally, we passed to metabomatching the 179 modules from run 1 with largest basins.

PCA: Principal Component Analysis

We used the sklearn library for Python (2.7) to compute all principal components of the preprocessed data. Specifically, we used the decomposition.PCA object, with n_components set to 687 and svd_solver set to full.

Identification of Metabolites

After running metabomatching on the pseudospectra generated by ISA, ACP and PCA, we applied a filtering algorithm to select only the most robust matches among all pseudospectra. The filtering passes only those pseudospectra which achieve an adjusted score above 2 with their top metabolite match (ensuring the pseudospectra finds a reasonable match by metabomatching) and have at least one peak with z-score above 4 (ensuring there is a strong signal). Note that multiple pseudospectra can match with the same metabolite NMR spectrum.

Out of the 179 pseudospectra from ACP, 10 pseudospectra matching different metabolites passed the filtering. Among these, only for one pseudospectrum, i.e., 3.87 and 3.75, the top match to hydroxypropionate (Figure S24) did not look convincing because of slightly worse matches to mannitol and arabitol (Figure S24).

Out of the 179 pseudospectra from ISA, 19 pseudospectra matching different metabolites passed the filtering. Among these, we discarded 9 for one or more of the following three reasons: (1) There are several metabolites which all achieve high adjusted scores (Figures S25 and S26). (2) There is at least one strong peak in the pseudospectrum that does not match with any of the spectra of the best matching metabolites (see Figures S27–S32). This may happen for pseudospectra with a large number of peaks. These pseudospectra are not necessarily biologically irrelevant and might carry a signature for two or more related metabolites. (3) The pseudospectrum passed the filtering, yet did not appear to have sufficiently strong signals at the peak positions of its putative matching metabolites (Figures S33 and S34).

We analyzed all 687 pseudospectra from PCA and observed an elevation in metabomatching scores for the last principal components (all between components #505–687, which jointly explain only 1% of variation; Figures S1A, S35–S39). However, since these components explain almost no variation in the metabolome and the last 9 principal components matched the same metabolite, hippurate (Figure S1D), we investigated whether these matches rely on numerical instability. Indeed, when we removed one feature from all feature pairs that correlated above 0.95 (i.e., 8 features from 34 feature pairs all belonging to hippurate multiplets regions 7.54–7.56, 7.63–7.65, 7.83–7.84, and 3.96 ppm), and ran metabomatching on all pseudospectra generated from the principal components of the remaining features, only five passed the filtering (Figure S1F). However, none of these seemed convincing when applying the same curation as for the ISA pseudospectra.

Pseudoquantification of Metabolite Concentrations

We use the term pseudoquantification as this approach should not be confused with traditional quantification approach which sometimes rely on experiments that target a specific metabolite and often require the use of proprietary software operated by an expert.

We perform our pseudoquantification by using discrete integration to estimate relative metabolite concentrations, according to

(2)

where K is the number of multiplets, H_k the number of protons in multiplet k, [l_k, r_k] the range of multiplet k, s_j the chemical shift of feature j, I_ij the intensity of this feature in individual i, and Δ_j the width of the bin at s_j. For example, for hippurate K = 4, H = [2, 2, 1, 2], l = m – 0.025, r = m + 0.025, where m = [3.98, 7.54, 7.65, 7.84]. We then evaluated this relative concentration first using the peak positions from the reference spectrum as listed in UMDB, and second using the peak positions as suggested by the pseudospectrum found by the modular approach.

To perform the pseudoquantification based on the pseudospectra of the modules, we defined a set of multiplet positions to use in eq 2 for each module that robustly identified a metabolite. This set is composed of all the chemical shifts from the module of interest with z-scores above 3 and within 0.025 ppm of the multiplet positions of the matching metabolite in reference database. It includes all relevant peaks from the metabolite detected by the modular analysis.

Analysis and Results

Click to copy section linkSection link copied!

In this work, we show that metabomatching can be used with pseudospectra capturing the internal structure of large-scale NMR data rather than their correlation with external variables. Specifically, our premise is that in sizable sample collections there is sufficient power for methods identifying coherent features that may point to the same metabolite.

We used three methods for identifying weighted sets of covarying spectral features from large-scale NMR data that can be used as input for metabomatching (see Methods for more details and Figure 1 for an illustration of the workflow).

Figure 1. Workflow for unsupervised analysis of large-scale NMR data. Raw ¹H NMR data are normalized then aligned. These processed profiles are used as input for the averaged correlation profile (ACP), iterative signature algorithm (ISA), and principal component analysis (PCA) methods, which output correlation profiles, module scores, and PCA loadings, respectively. These outputs constitute possible pseudospectra for metabomatching, which identifies the most plausible candidate metabolites underlying the coherent feature variations.

Correlation-Based Pseudospectra

Our first approach to select covarying features was to use the correlations between features across all samples. We faced two challenges: First, using the correlations of a given feature with all features as a pseudospectrum would break the scoring algorithm in metabomatching, due to self-correlation of features that result in infinite z-scores. Second, as generating only a limited number of pseudospectra was desired ranking the input sets was necessary to select only the most relevant ones.

To address these challenges, we devised an algorithm that ranks all pairs of sufficiently distant features and computes averaged correlation profiles (ACP) for pairs of highly correlated features. Strictly speaking, ACPs are not the correlations but they are the average correlation profiles, with the premise that for highly correlated feature pairs the correlations to other features tend to be similar, while none of the averages equals to one. Within metabomatching these ACPs are then translated into z-scores using the Fisher z-transformation (see Methods for more details). We also tried hierarchical clustering to define pseudospectra from multiple highly similar features (see Methods), but this approach did not work as well.

Iterative Signature Algorithm

ISA has been designed for the unsupervised identification of coherent subsets in large-scale data. (18,19) Specifically, coherence between features is not defined by total correlation across all samples, but rather by a subset of samples for which a set of features takes more extreme values than for the rest of the samples. Such a joint set of features and samples is called a module. In order to obtain a pseudospectrum for each module we averaged each feature’s score (whether part of the module or not) across the samples assigned to the module. These averages were then transformed into z-scores. By definition the features of the module have the most extreme z-scores, yet other features that were just below the threshold may also have a sizable contribution. We then used these z-scores as input for metabomatching.

Principal Component Analysis

We also used PCA to compute the loadings of all features onto the eigenvectors of the sample–sample correlation matrix across all features. These eigenvectors (or eigensamples) characterize independent axes of variation in sample space. The corresponding eigenvalues reflect the fraction of total variation explained by each eigenvector. It was not clear a priori which principal components might characterize variation due to single metabolites. We therefore applied metabomatching to all of them. Specifically, we generated pseudospectra by standardizing the loadings corresponding to each eigensample (see Methods for more details).

Many Pseudospectra Defined by the ACP Method and ISA Match to Urine Metabolites

We observed a trend of elevated metabomatching scores for pseudospectra corresponding to principal components with small eigenvalues (starting from component #505), jointly explaining only 1% of variation (Figure S1A). The last nine principal components matched to hippurate, but disappeared when running PCA on the metabolome stripped of features that are highly correlated to other features (see Methods; Figures S1D and S1F). Additionally, the adjusted scores of all potential hits decreased significantly for the stripped metabolome (Figure S1E). We therefore concluded that PCA is not well-suited for generating robust metabolite signatures.

In contrast, our ACP method and ISA resulted in a sizable number of pseudospectra for which metabomatching produced robust matches to urine metabolites (see Figure 2 and Methods for details). Specifically, both ACP and ISA identified feature sets pointing to glucose, citrate, ethanol, hippurate, and P-hydroxyphenylacetate (Figures S2–S11). Glucose and hippurate were among the metabolites identified by Cloarec et al. with the correlation matrix of mouse urine NMR data. (1)

Figure 2. Urine metabolites that were robustly matched by metabomatching to pseudospectra derived from average correlation profiles (ACP, green), the iterative signature algorithm (ISA, blue), or both methods (black).

P-hydroxyphenylacetate shares an aromatic ring with 3-hydroxyphenylpropionate, another metabolite highlighted in the original STOCSY paper. (1) Both compounds are part of phenylalanine metabolism and occur as products of bacterial degradation of aromatic compounds. In human urine high concentrations of these compounds may reflect an overgrown Clostridium species in gut microbiota, which has been associated with autism spectrum disorders. (20)

In healthy humans, urine glucose should be low, but concentrations may be elevated due to diabetes or chronic kidney disease (CKD), conditions which are prevalent in the CoLaus population.

Citrate is an additive commonly used by the food industry and it is also synthesized as an intermediate product in the tricarboxylic acid cycle, a central pathway that releases stored energy from fat, proteins, and carbohydrates. Low urinary citrate is associated with CKD and kidney stone formation.

There are a number of metabolites that matched to pseudospectra generated only by one of the two methods. With ISA, we found modules matching 3-aminoisobutyrate, an end product of nucleic acid metabolism that has been considered a potential biochemical marker for cancer (21) (Figure S12); creatinine, a breakdown product of creatine, whose high and stable concentration in urine is often used for normalization (Figure S13); lactose (Figure S14); and lactate, the bacterial breakdown product of lactose (Figure S15). Conversely, ACP produced correlation profiles matching to taurine (Figure S16), an organic compound widely distributed in animal tissues and a major constituent of bile; creatine (Figure S14); oxoglutarate (α-ketoglutarate), an important biological compound produced by deamination of glutamate, and an intermediate in the Krebs cycle (Figure S18); and 3-hydroxyisovalerate, a byproduct of valine, leucine, and isoleucine degradation and a marker for biotin deficiency (22) (Figure S19). These compounds are all common urine metabolites that can exist in high concentrations.

Correcting features for significant covariates, both methods also found set of features matching carnitine, which owes its name to its high concentration in meat (see Methods for more details). While it is produced in both animal and plant cells, this may explain why we could detect its signature in human urine samples.

Metabolite Concentration Pseudoquantification with NMR Features of Matched Pseudospectra

We next investigated whether the sets of weighted NMR features generated by ISA or the ACP method can not only be used to identify metabolites, but also facilitate their pseudoquantification. This pseudoquantification approach aims to estimate the relative concentration of the metabolites in untargeted ¹H NMR of urine samples. We performed our pseudoquantification by computing the area under the peak of each multiplet in the metabolite spectrum using discrete integration, dividing it by the number of protons associated with the multiplet and then averaging scaled areas over all multiplets in the metabolite spectrum (see Methods for more details). We hypothesized that the leading features selected by our algorithms for a certain metabolite may be better suited for pseudoquantification than the full reference spectra from public databases such as UMDB. There are two possible reasons for this. First, the exact feature positions extracted from the data by ISA or the ACP method may be more accurate, even if, by design, they fall within the margin of the matching window of reference spectrum features. Second, for metabolites with several peaks, only a subset might have been picked up by these algorithms. Indeed, both ISA and the ACP method may leave out peaks that did not contribute coherently to the peak set since their signal was too noisy (e.g., due to overlap with those from other metabolites).

We performed our pseudoquantification method (using eq 2 in Methods) to estimate concentrations of glucose and ethanol, for which relevant phenotypes were available in the cohort. For urine glucose, the phenotype was fasting blood glucose. For urine ethanol, relevant biomarkers included serum asialotransferrin and disialotransferrin, which combined are known as carbohydrate-deficient transferrin (CDT), a biomarker for heavy alcohol use. Furthermore, self-reported alcohol consumption was available. These biomarkers were measured in a different biofluid (i.e., blood), which was collected on the same day as the urine sample. We argue that detecting significant correlations between our estimated metabolite concentrations and these biomarkers provides a proof of concept that our pseudoquantification is reliable, and comparing correlations between different pseudoquantification approaches provides a means to evaluate them.

The ¹H NMR spectra of glucose has nine multiplets. Including all these multiplets chemical shifts from the UMDB database (Table 1) to perform pseudoquantification, we obtain a correlation of 0.46 (with a 95% confidence interval (CI) of [0.41, 0.52]) between the estimated concentration of urine glucose and fasting blood glucose. This correlation increases to 0.50 [0.44, 0.55] if the subset of seven multiplets from ISA module #16 (Figure 3A) is used for pseudoquantification (see Methods for details). From the 179 ACP pseudospectra, two robustly matched glucose, one from averaging the correlation profiles from feature pair 3.48 and 5.24 and another from averaging correlation profiles from 3.89 and 5.24 (Figure 3B,C), each capturing four out of nine multiplets of glucose (Table 1). Using the subset of peaks from 3.48 and 5.24 ACP pseudospectra we obtain a correlation of 0.48 [0.43, 0.54] between glucose pseudoquantification and fasting blood glucose while using the subset of peaks from 3.89 and 5.24 ACP pseudosspectra a correlation of 0.44 [0.38, 0.49] is obtained. Combining the peak subsets from both pseudospectra to a subset of 6 peaks did not improve the correlation beyond 0.48 [0.43, 0.54] (see Table 1 and Methods for more details on peak sets used for pseudoquantification).

Table 1. Correlation between Pseudoquantification and Measured Biomarkers of Glucose and Ethanol

urine metabolite	feature source	multiplet positions (ppm)	related biomarker	correlation [95% CI]
glucose	UMDB	3.23, 3.40, 3.46	serum	0.46
		3.52, 3.73, 3.82	glucose	[0.41, 0.52]
		3.88, 4.63, 5.22
glucose	ACP: 3.48 and 5.24	3.40, 3.48, 4.65	serum	0.48
		5.24	glucose	[0.43, 0.54]
glucose	ACP: 3.89 and 5.24	3.82, 3.89, 4.65	serum	0.44
		5.24	glucose	[0.38, 0.49]
glucose	ISA: module #16	3.25, 3.41, 3.48	serum	0.50
		3.50, 3.89, 4.65	glucose	[0.44, 0.55]
		5.24
ethanol	UMDB	1.17, 3.65	serum	0.29
			CDT	[0.23, 0.35]
ethanol	ACP: 1.18 and 3.67	1.18, 3.67	serum	0.16
			CDT	[0.10, 0.22]
ethanol	ISA: module #57	1.18, 3.67	serum	0.16
			CDT	[0.10, 0.22]
EtG	Nicholas et al. (23)	1.24, 3.30, 3.52	serum	0.36
		3.71, 3.99, 4.48	CDT	[0.30, 0.42]
EtG	ISA: module #240	1.24, 3.52, 4.47	serum	0.46
			CDT	[0.40, 051]

Figure 3. Pseudospectra from ACP and ISA algorithms matching glucose, ethanol, and EtG. Each plot shows the pseudospectrum in blue in the upper half and the reference spectrum from UMDB in black and in the lower half. Dark blue indicates chemical shifts and their ±0.025 ppm vicinity that were used for pseudoquantification.

The ¹H NMR spectra of ethanol has only two multiplets (as well as one singlet from the hydroxyl group, which can not be discerned in water solutions like urine). Using the reference spectrum from UMDB to perform pseudoquantification, we obtained a relatively low correlation of 0.29 [0.23, 0.35] between the estimated concentration of urine ethanol and CDT levels in serum (Table 1, see Table S1 for other alcohol markers). The ACP method produced one pseudospectra, 1.18 and 3.67, and ISA produced two modules (#57 and #240) that metabomatching matched to ethanol (Figures S6, S7, and S22). The positions of the ACP peak set (i.e., 1.18 and 3.67) were identical to those of ISA module #57 and were more similar to the ethanol spectrum (achieving a higher adjusted score in metabomatching) than those of ISA module #240 (Figure 3D–F). Nevertheless, pseudoquantification for the ACP pseudospectrum and ISA module #57 yielded a lower correlation of 0.16 [0.10, 0.22] with CDT levels than the UMDB reference peaks (0.29 [0.23, 0.35]) (Table 1). This is due to a high correlation of CDT with the features at 1.145–1.155 ppm, which are within a 0.025 ppm neighborhood of the UMDB ethanol peak at 1.17 ppm but not within the same neighborhood of the 1.18 ethanol peak from ACP and ISA module #57 (Figure S20). Yet, these peaks at 1.145–1.155 ppm are unlikely to correspond to ethanol, since their correlation with the other ethanol peak at 3.67 ppm is much weaker than the correlation between the 1.18 and 3.67 ppm peaks. Instead, they may belong to a different metabolite whose concentration is correlated with CDT (Figure S21). In contrast, summing up the intensities over all the features of ISA module #240 with a z-score above 3 as a pseudoquantification measure (in the absence of any multiplet information), we obtained a correlation of 0.51 [0.46, 0.57] with the CDT measurements.

To better understand why module #240 correlates more strongly to the alcohol consumption biomarker while being a worse match to ethanol than module #57 (Figure 3), we studied whether any of its features point to other compounds related to ethanol metabolism. Indeed, we found that this module contains three features, at 1.26, 3.52, and 4.47 ppm, that individually correlate more strongly to CDT (0.40 [0.34, 0.45], 0.29 [0.23, 0.35], 0.33 [0.27, 0.39], respectively) than the features mapping to ethanol. Interestingly, these features appeared to be close to those of ethyl glucuronide (EtG), a direct product of ethanol nonoxidative metabolism by conjugation with uridine diphosphate (UDP)-glucuronic acid, which had previously been detected in ¹H NMR spectra of liver extracts (23) and more recently in human urine of alcohol drinkers. (24) To confirm EtG as a possible match for ISA module #240, we added its features as extracted from Nicholas et al. (23) to the metabomatching library manually, since EtG had no entry in UMDB. We observed that ethanol and EtG spectra together provided a better match to ISA module #240 than ethanol alone (Figure S22). Interestingly, the distance between the two peaks corresponding to the doublet at 4.48 ppm is about 0.0126 ppm, corresponding to a coupling of 8.8 Hz (for a 700 MHz spectrometer) consistent with the coupling of 8 Hz reported in Nicholas et al. (23) (see Figure S23 and Supporting Informationfor more details).

Performing the pseudoquantification of EtG using the peak set of 6 reference positions extracted from Nicholas et al. (23), we obtained a correlation of 0.36 [0.30, 0.42] with CDT levels; performing the pseudoquantification with the 3 feature subset from module #240, we obtained a correlation of 0.46 [0.40, 0.51] (Table 1). This indicates that EtG pseudoquantification correlates better with CDT than ethanol, which is in agreement with the fact that EtG is detectable in urine for a longer time window (2–5 days) than ethanol (12–24 h), and CDT is a marker for heavy alcohol use (at least five drinks a day over a period of 2 weeks before giving the sample (25)). Remarkably, the pseudoquantification facilitated by the three features of module #240 correlates even more strongly with CDT than the full set of EtG reference features, presumably because these features have the best signal-to-noise ratio and optimal position for our data. They may therefore constitute a promising urine biomarker for heavy alcohol consumption. Indeed, while the correlation between EtG pseudoquantification and CDT measure increases to 0.59 [0.46, 0.72] when focusing on subjects who have self-reported heavy drinking, pseudoquantification of module #240 gives rise to a slightly higher correlation of 0.61 [0.48, 0.74].

Conclusions and Discussion

Click to copy section linkSection link copied!

In this work, we implemented and tested new methodologies for analyzing large-scale ¹H NMR spectroscopy data. Building on previous ideas to use the correlation structure of such data to generate metabolomic signatures, we investigated three complementary methods for generating such signatures and benchmarked the methods in terms of how many of their signatures matched with reference spectra in public databases. By design, these approaches will only identify metabolites with at least two distinct peaks, and therefore complement peak-picking identification approaches, which tend to focus on single peak metabolites.

We found that average correlation profiles (ACP) of highly correlated feature pairs, a method inspired by STOCSY, as well as the iterative signature algorithm (ISA) identified ten and nine metabolites, respectively, five of which overlapped. In contrast, principal component analysis (PCA) did not generate any pseudospectra with robust metabomatching, likely because leading components explain variation driven by many metabolites.

While ACP is designed to pick up individual metabolites with at least two (nonproximal) features in their spectrum (or those of metabolite pairs whose concentrations are coupled), ISA is able to generate modules where many features exhibit coherent variation, yet potentially only over a subset of samples. We believe that this may be particularly useful when integrating data from a heterogeneous set of samples (e.g., including those from diseased or medicated subpopulations).

One interesting property of our modular approach is that the feature sets identified by ACP or ISA do not need to match perfectly with those of the reference spectrum of the corresponding compound. Indeed, the two ACP signatures matching glucose each only cover four and jointly six of the nine glucose peaks, while the ISA module with the best match to glucose includes seven of its peaks. Adding the “missing” peaks in our pseudoquantification slightly reduced the correlation with serum glucose, indicating there is a marginal improvement in the pseudoquantification using ppm positions only from the multiplets found by our algorithms rather than the database. Further work will be needed to substantiate this observation.

Another interesting aspect of our approach is that modular feature sets may match multiple compounds. Our current implementation of metabomatching allows simultaneous identification of up to two compounds. Indeed, our finding that ISA picked up a module whose signature mapped well to ethanol and its specific metabolic product ethyl glucuronide demonstrated the potential power of ISA to identify metabolite pairs within the same pathway. Moreover, the strong correlation of this module with the alcohol abuse marker CDT was likely driven by the fact that ISA can extract context specific covariance, which in this case is strongest in samples with particularly high alcohol consumption. This module also highlighted that using the relevant chemical shifts found by the module rather than all shifts from the reference database can lead to more accurate pseudoquantification of the underlying metabolites, due to different contribution of shifts specific to the experimental conditions in the complex urine spectra.

Extending metabomatching beyond compound pairs is challenging due to the large number of possible trios and higher order combinations, but could be feasible in future work, for example by using metabolic pathway information to limit the number of relevant metabolite combinations to test.

A critical element of our analysis was to transform the signatures generated by the different methods into a universal format (i.e., z-scores) as input for our metabomatching tool that we previously developed for the analysis of feature signatures generated by regression on external variables. Indeed, being able to query both internal and external signatures of large-scale NMR data against a reference data set of known spectra from individual metabolites is pivotal for exploring new methods dissecting the auto- and cross-correlation structure for integrative analyses.

In conclusion, we believe that our study using fewer than 1000 samples gives ample evidence for the potential of automated analysis of large-scale NMR data, and that increased sample sizes are likely to result in further identifications and more accurate pseudoquantifications of individual metabolites. To this end, our analysis software, metabomodules, is made publicly available on GitHub https://github.com/BergmannLab/metabomodules-docker.

Supporting Information

Click to copy section linkSection link copied!

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jproteome.9b00295.

Supporting figures and table (PDF)

Terms & Conditions

Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.

Author Information

Click to copy section linkSection link copied!

Corresponding Author
- Sven Bergmann - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland; Department of Integrative Biomedical Sciences, University of Cape Town, Cape Town 7700, South Africa; Email: [email protected]
Authors
- Bita Khalili - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland; http://orcid.org/0000-0001-5630-1812
- Mattia Tomasoni - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
- Mirjam Mattei - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
- Roger Mallol Parera - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
- Reyhan Sonmez - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
- Daniel Krefl - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
- Rico Rueedi - Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland; Swiss Institute of Bioinformatics, 1015 Lausanne, Switzerland
Author Contributions
B.K., M.T., and M.M. are co-first authors. R.R. and S.B. are co-last authors. S.B. and R.R. designed the study. The manuscript was written by B.K. and S.B. The correlation-based methods was implemented and applied by B.K. The modular analysis with ISA was performed by M.M. and R.R. PCA was run by M.T., D.K., and B.K. All authors discussed the results and implications, and contributed to the manuscript.
Notes
The authors declare no competing financial interest.
Our analysis software, metabomodules, is made publicly available on GitHub: https://github.com/BergmannLab/metabomodules-docker.

Acknowledgments

Click to copy section linkSection link copied!

This work was supported by the Swiss National Science Foundation (grant FN 310030_152724/1) and the NIH (grant R03 CA211815).

References

Click to copy section linkSection link copied!

This article references 25 other publications.

1
Cloarec, O.; Dumas, M.-E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gauguier, D.; Lindon, J. C.; Holmes, E. Statistical Total Correlation Spectroscopy: An Exploratory Approach for Latent Biomarker Identification from Metabolic 1H NMR Data Sets. Anal. Chem. 2005, 77 (5), 1282– 1289, DOI: 10.1021/ac048630x
Google Scholar
1
Statistical Total Correlation Spectroscopy: An Exploratory Approach for Latent Biomarker Identification from Metabolic 1H NMR Data Sets
Cloarec, Olivier; Dumas, Marc-Emmanuel; Craig, Andrew; Barton, Richard H.; Trygg, Johan; Hudson, Jane; Blancher, Christine; Gauguier, Dominique; Lindon, John C.; Holmes, Elaine; Nicholson, Jeremy
Analytical Chemistry (2005), 77 (5), 1282-1289CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
We describe here the implementation of the statistical total correlation spectroscopy (STOCSY) anal. method for aiding the identification of potential biomarker mols. in metabonomic studies based on NMR spectroscopic data. STOCSY takes advantage of the multicollinearity of the intensity variables in a set of spectra (in this case 1H NMR spectra) to generate a pseudo-two-dimensional NMR spectrum that displays the correlation among the intensities of the various peaks across the whole sample. This method is not limited to the usual connectivities that are deducible from more std. two-dimensional NMR spectroscopic methods, such as TOCSY. Moreover, two or more mols. involved in the same pathway can also present high intermol. correlations because of biol. covariance or can even be anticorrelated. This combination of STOCSY with supervised pattern recognition and particularly orthogonal projection on latent structure-discriminant anal. (O-PLS-DA) offers a new powerful framework for anal. of metabonomic data. In a first step O-PLS-DA exts. the part of NMR spectra related to discrimination. This information is then cross-combined with the STOCSY results to help identify the mols. responsible for the metabolic variation. To illustrate the applicability of the method, it has been applied to 1H NMR spectra of urine from a metabonomic study of a model of insulin resistance based on the administration of a carbohydrate diet to three different mice strains (C57BL/6Oxjr, BALB/cOxjr, and 129S6/SvEvOxjr) in which a series of metabolites of biol. importance can be conclusively assigned and identified by use of the STOCSY approach.
2
Blaise, B. J.; Navratil, V.; Domange, C.; Shintu, L.; Dumas, M.-E.; Elena-Herrmann, B.; Emsley, L.; Toulhoat, P. Two-Dimensional Statistical Recoupling for the Identification of Perturbed Metabolic Networks from NMR Spectroscopy. J. Proteome Res. 2010, 9 (9), 4513– 4520, DOI: 10.1021/pr1002615
Google Scholar
2
Two-Dimensional Statistical Recoupling for the Identification of Perturbed Metabolic Networks from NMR Spectroscopy
Blaise, Benjamin J.; Navratil, Vincent; Domange, Celine; Shintu, Laetitia; Dumas, Marc-Emmanuel; Elena-Herrmann, Benedicte; Emsley, Lyndon; Toulhoat, Pierre
Journal of Proteome Research (2010), 9 (9), 4513-4520CODEN: JPROBS; ISSN:1535-3893. (American Chemical Society)
The development of Statistical Total Correlation Spectroscopy (STOCSY), a representation of the autocorrelation matrix of a spectral data set as a 2D pseudospectrum, has allowed more reliable assignment of one- and two-dimensional NMR spectra acquired from the complex mixts. that are usually used in metabolomics/metabonomics studies, thus, improving precise identification of candidate biomarkers contained in metabolic signatures computed by multivariate statistical anal. However, the correlations obtained cannot always be interpreted in terms of connectivities between metabolites. In this study, the authors combine statistical recoupling of variables (SRV) and STOCSY to identify perturbed metabolite systems. The resulting Recoupled-STOCSY (R-STOCSY) method provides a 2D correlation landscape based on the SRV clusters representing phys., chem., and biol. entities. This enables the identification of correlations between distant clusters and extends the recoupling scheme of SRV, which was previously limited to the assocn. of neighboring clusters. This allows the recovery of only meaningful correlations between metabolic signals and significantly enhances the interpretation of STOCSY. The method is validated through the measurement of the distances between the metabolites involved in these correlations, within the whole metabolic network, which shows that the av. shortest path length is significantly shorter for the correlations detected in this new way compared to metabolite couples randomly selected from within the entire KEGG metabolic network. This enables the identification without any a priori knowledge of the perturbed metabolic network. The R-STOCSY completes the recoupling procedure between distant clusters, further reducing the high dimensionality of metabolomics/metabonomics data set and finally allows the identification of composite biomarkers, highlighting disruption of particular metabolic pathways within a global metabolic network. This allows the perturbed metabolic network to be extd. through NMR based metabolomics/metabonomics in an automated, and statistical manner.
3
Sands, C. J.; Coen, M.; Ebbels, T. M. D.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Data-Driven Approach for Metabolite Relationship Recovery in Biological 1H NMR Data Sets Using Iterative Statistical Total Correlation Spectroscopy. Anal. Chem. 2011, 83 (6), 2075– 2082, DOI: 10.1021/ac102870u
Google Scholar
3
Data-Driven Approach for Metabolite Relationship Recovery in Biological 1H NMR Data Sets Using Iterative Statistical Total Correlation Spectroscopy
Sands, Caroline J.; Coen, Muireann; Ebbels, Timothy M. D.; Holmes, Elaine; Lindon, John C.; Nicholson, Jeremy K.
Analytical Chemistry (Washington, DC, United States) (2011), 83 (6), 2075-2082CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
Statistical total correlation spectroscopy (STOCSY) is a well-established and valuable method in the elucidation of both inter- and intrametabolite correlations in NMR metabonomic data sets. Here, the STOCSY approach is extended in a novel Iterative-STOCSY (I-STOCSY) tool in which correlations are calcd. initially from a driver peak of interest and subsequently for all peaks identified as correlating with a correlation coeff. greater than a set threshold. Consequently, in a single automated run, the majority of information contained in multiple STOCSY calcns. from all peaks recursively correlated to the original user defined driver peak of interest are recovered. In addn., highly correlating peaks are clustered into putative structurally related sets, and the results are presented in a fully interactive plot where each set is represented by a node; node-to-node connections are plotted alongside corresponding spectral data colored by the strength of connection, thus allowing the intuitive exploration of both inter- and intrametabolite connections. The I-STOCSY approach has been here applied to a 1H NMR data set of 24 h postdose aq. liver exts. from rats treated with the model hepatotoxin galactosamine and has been shown both to recover the previously deduced major metabolic effects of treatment and to generate new hypotheses even on this well-studied model system. I-STOCSY, thus, represents a significant advance in correlation based anal. and visualization, providing insight into inter- and intrametabolite relationships following metabolic perturbations.
4
Posma, J. M.; Garcia-Perez, I.; De Iorio, M.; Lindon, J. C.; Elliott, P.; Holmes, E.; Ebbels, T. M. D.; Nicholson, J. K. Subset Optimization by Reference Matching (STORM): An Optimized Statistical Approach for Recovery of Metabolic Biomarker Structural Information from 1H NMR Spectra of Biofluids. Anal. Chem. 2012, 84 (24), 10694– 10701, DOI: 10.1021/ac302360v
Google Scholar
4
Subset Optimization by Reference Matching (STORM): An Optimized Statistical Approach for Recovery of Metabolic Biomarker Structural Information from 1H NMR Spectra of Biofluids
Posma, Joram M.; Garcia-Perez, Isabel; De Iorio, Maria; Lindon, John C.; Elliott, Paul; Holmes, Elaine; Ebbels, Timothy M. D.; Nicholson, Jeremy K.
Analytical Chemistry (Washington, DC, United States) (2012), 84 (24), 10694-10701CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
We describe a new multivariate statistical approach to recover metabolite structure information from multiple 1H NMR spectra in population sample sets. Subset optimization by ref. matching (STORM) was developed to select subsets of 1H NMR spectra that contain specific spectroscopic signatures of biomarkers differentiating between different human populations. STORM aims to improve the visualization of structural correlations in spectroscopic data by using these reduced spectral subsets contg. smaller nos. of samples than the no. of variables (n « p). We have used statistical shrinkage to limit the no. of false pos. assocns. and to simplify the overall interpretation of the autocorrelation matrix. The STORM approach has been applied to findings from an ongoing human metabolome-wide assocn. study on body mass index to identify a biomarker metabolite present in a subset of the population. Moreover, we have shown how STORM improves the visualization of more abundant NMR peaks compared to a previously published method (statistical total correlation spectroscopy, STOCSY). STORM is a useful new tool for biomarker discovery in the 'omic' sciences that has widespread applicability. It can be applied to any type of data, provided that there is interpretable correlation among variables, and can also be applied to data with more than one dimension (e.g., 2D NMR spectra).
5
Hao, J.; Astle, W.; De Iorio, M.; Ebbels, T. M. D. BATMAN—an R Package for the Automated Quantification of Metabolites from Nuclear Magnetic Resonance Spectra Using a Bayesian Model. Bioinformatics 2012, 28 (15), 2088– 2090, DOI: 10.1093/bioinformatics/bts308
Google Scholar
5
BATMAN-an R package for the automated quantification of metabolites from nuclear magnetic resonance spectra using a Bayesian model
Hao, Jie; Astle, William; De Iorio, Maria; Ebbels, Timothy M. D.
Bioinformatics (2012), 28 (15), 2088-2090CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Motivation: NMR (NMR) spectra are widely used in metabolomics to obtain metabolite profiles in complex biol. mixts. Common methods used to assign and est. concns. of metabolites involve either an expert manual peak fitting or extra pre-processing steps, such as peak alignment and binning. Peak fitting is very time consuming and is subject to human error. Conversely, alignment and binning can introduce artifacts and limit immediate biol. interpretation of models. Results: We present the Bayesian automated metabolite analyzer for NMR spectra (BATMAN), an R package that deconvolutes peaks from one-dimensional NMR spectra, automatically assigns them to specific metabolites from a target list and obtains concn. ests. The Bayesian model incorporates information on characteristic peak patterns of metabolites and is able to account for shifts in the position of peaks commonly seen in NMR spectra of biol. samples. It applies a Markov chain Monte Carlo algorithm to sample from a joint posterior distribution of the model parameters and obtains concn. ests. with reduced error compared with conventional numerical integration and comparable to manual deconvolution by experienced spectroscopists. Availability and implementation: http://www1.imperial.ac.uk/medicine/people/t.ebbels/ Contact: [email protected].
6
Alonso, A.; Rodríguez, M. A.; Vinaixa, M.; Tortosa, R.; Correig, X.; Julià, A.; Marsal, S. Focus: A Robust Workflow for One-Dimensional NMR Spectral Analysis. Anal. Chem. 2014, 86 (2), 1160– 1169, DOI: 10.1021/ac403110u
Google Scholar
6
Focus: A Robust Workflow for One-Dimensional NMR Spectral Analysis
Alonso, Arnald; Rodriguez, Miguel A.; Vinaixa, Maria; Tortosa, Raul; Correig, Xavier; Julia, Antonio; Marsal, Sara
Analytical Chemistry (Washington, DC, United States) (2014), 86 (2), 1160-1169CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
One-dimensional 1H NMR represents one of the most commonly used anal. techniques in metabolomic studies. The increase in the no. of samples analyzed as well as the tech. improvements involving instrumentation and spectral acquisition demand increasingly accurate and efficient high-throughput data processing workflows. We present FOCUS, an integrated and innovative methodol. that provides a complete data anal. workflow for one-dimensional NMR-based metabolomics. This tool will allow users to easily obtain a NMR peak feature matrix ready for chemometric anal. as well as metabolite identification scores for each peak that greatly simplify the biol. interpretation of the results. The algorithm development has been focused on solving the crit. difficulties that appear at each data processing step and that can dramatically affect the quality of the results. As well as method integration, simplicity has been one of the main objectives in FOCUS development, requiring very little user input to perform accurate peak alignment, peak picking, and metabolite identification. The new spectral alignment algorithm, RUNAS, allows peak alignment with no need of a ref. spectrum, and therefore, it reduces the bias introduced by other alignment approaches. Spectral alignment has been tested against previous methodologies obtaining substantial improvements in the case of moderate or highly unaligned spectra. Metabolite identification has also been significantly improved, using the positional and correlation peak patterns in contrast to a ref. metabolite panel. Furthermore, the complete workflow has been tested using NMR data sets from 60 human urine samples and 120 aq. liver exts., reaching a successful identification of 42 metabolites from the two data sets. The open-source software implementation of this methodol. is available at http://www.urr.cat/FOCUS.
7
Ravanbakhsh, S.; Liu, P.; Bjorndahl, T. C.; Mandal, R.; Grant, J. R.; Wilson, M.; Eisner, R.; Sinelnikov, I.; Hu, X.; Luchinat, C. Accurate, Fully-Automated NMR Spectral Profiling for Metabolomics. PLoS One 2015, 10 (5), e0124219 DOI: 10.1371/journal.pone.0124219
Google Scholar
7
Accurate, fully-automated NMR spectral profiling for metabolomics
Ravanbakhsh, Siamak; Liu, Philip; Bjorndahl, Trent C.; Mandal, Rupasri; Grant, Jason R.; Wilson, Michael; Eisner, Roman; Sinelnikov, Igor; Hu, Xiaoyu; Luchinat, Claudio; Greiner, Russell; Wishart, David S.
PLoS One (2015), 10 (5), e0124219/1-e0124219/15CODEN: POLNCL; ISSN:1932-6203. (Public Library of Science)
Many diseases cause significant changes to the concns. of small mols. (a.k.a. metabolites) that appear in a person's biofluids, which means such diseases can often be readily detected from a person's "metabolic profile"-i.e., the list of concns. of those metabolites. This information can be extd. from a biofluids NMR (NMR) spectrum. However, due to its complexity, NMR spectral profiling has remained manual, resulting in slow, expensive and error-prone procedures that have hindered clin. and industrial adoption of metabolomics via NMR. This paper presents a system, BAYESIL, which can quickly, accurately, and autonomously produce a person's metabolic profile. Given a 1D 1H NMR spectrum of a complex biofluid (specifically serum or cerebrospinal fluid), BAYESIL can automatically det. the metabolic profile. This requires first performing several spectral processing steps, then matching the resulting spectrum against a ref. compd. library, which contains the "signatures" of each relevant metabolite. BAYESIL views spectral matching as an inference problem within a probabilistic graphical model that rapidly approximates the most probable metabolic profile. Our extensive studies on a diverse set of complex mixts. including real biol. samples (serum and CSF), defined mixts. and realistic computer generated spectra; involving > 50 compds., show that BAYESIL can autonomously find the concn. of NMR-detectable metabolites accurately (∼ 90% correct identification and ∼ 10% quantification error), in less than 5 min on a single CPU. These results demonstrate that BAYESIL is the first fully-automatic publicly-accessible system that provides quant. NMR spectral profiling effectively-with an accuracy on these biofluids that meets or exceeds the performance of trained experts. We anticipate this tool will usher in high-throughput metabolomics and enable a wealth of new applications of NMR in clin. settings.
8
Tardivel, P. J. C.; Canlet, C.; Lefort, G.; Tremblay-Franco, M.; Debrauwer, L.; Concordet, D.; Servien, R. ASICS: An Automatic Method for Identification and Quantification of Metabolites in Complex 1D 1H NMR Spectra. Metabolomics 2017, 13 (10), 109, DOI: 10.1007/s11306-017-1244-5
Google Scholar
There is no corresponding record for this reference.
9
Röhnisch, H. E.; Eriksson, J.; Müllner, E.; Agback, P.; Sandström, C.; Moazzami, A. A. AQuA: An Automated Quantification Algorithm for High-Throughput NMR-Based Metabolomics and Its Application in Human Plasma. Anal. Chem. 2018, 90 (3), 2095– 2102, DOI: 10.1021/acs.analchem.7b04324
Google Scholar
9
AQuA: An Automated Quantification Algorithm for High-Throughput NMR-Based Metabolomics and Its Application in Human Plasma
Rohnisch Hanna E; Eriksson Jan; Mullner Elisabeth; Agback Peter; Sandstrom Corine; Moazzami Ali A
Analytical chemistry (2018), 90 (3), 2095-2102 ISSN:.
A key limiting step for high-throughput NMR-based metabolomics is the lack of rapid and accurate tools for absolute quantification of many metabolites. We developed, implemented, and evaluated an algorithm, AQuA (Automated Quantification Algorithm), for targeted metabolite quantification from complex (1)H NMR spectra. AQuA operates based on spectral data extracted from a library consisting of one standard calibration spectrum for each metabolite. It uses one preselected NMR signal per metabolite for determining absolute concentrations and does so by effectively accounting for interferences caused by other metabolites. AQuA was implemented and evaluated using experimental NMR spectra from human plasma. The accuracy of AQuA was tested and confirmed in comparison with a manual spectral fitting approach using the ChenomX software, in which 61 out of 67 metabolites quantified in 30 human plasma spectra showed a goodness-of-fit (r(2)) close to or exceeding 0.9 between the two approaches. In addition, three quality indicators generated by AQuA, namely, occurrence, interference, and positional deviation, were studied. These quality indicators permit evaluation of the results each time the algorithm is operated. The efficiency was tested and confirmed by implementing AQuA for quantification of 67 metabolites in a large data set comprising 1342 experimental spectra from human plasma, in which the whole computation took less than 1 s.
10
Cañueto, D.; Gómez, J.; Salek, R. M.; Correig, X.; Cañellas, N. rDolphin: A GUI R Package for Proficient Automatic Profiling of 1D 1H-NMR Spectra of Study Datasets. Metabolomics 2018, 14 (3), 24, DOI: 10.1007/s11306-018-1319-y
Google Scholar
10
rDolphin: a GUI R package for proficient automatic profiling of 1D (1)H-NMR spectra of study datasets
Canueto Daniel; Correig Xavier; Canellas Nicolau; Gomez Josep; Salek Reza M; Correig Xavier; Canellas Nicolau
Metabolomics : Official journal of the Metabolomic Society (2018), 14 (3), 24 ISSN:.
INTRODUCTION: Adoption of automatic profiling tools for (1)H-NMR-based metabolomic studies still lags behind other approaches in the absence of the flexibility and interactivity necessary to adapt to the properties of study data sets of complex matrices. OBJECTIVES: To provide an open source tool that fully integrates these needs and enables the reproducibility of the profiling process. METHODS: rDolphin incorporates novel techniques to optimize exploratory analysis, metabolite identification, and validation of profiling output quality. RESULTS: The information and quality achieved in two public datasets of complex matrices are maximized. CONCLUSION: rDolphin is an open-source R package ( http://github.com/danielcanueto/rDolphin ) able to provide the best balance between accuracy, reproducibility and ease of use.
11
Wishart, D. S.; Feunang, Y. D.; Marcu, A.; Guo, A. C.; Liang, K.; Vázquez-Fresno, R.; Sajed, T.; Johnson, D.; Li, C.; Karu, N. HMDB 4.0: The Human Metabolome Database for 2018. Nucleic Acids Res. 2018, 46 (D1), D608– D617, DOI: 10.1093/nar/gkx1089
Google Scholar
11
HMDB 4.0: the human metabolome database for 2018
Wishart, David S.; Feunang, Yannick Djoumbou; Marcu, Ana; Guo, An Chi; Liang, Kevin; Vazquez-Fresno, Rosa; Sajed, Tanvir; Johnson, Daniel; Li, Carin; Karu, Naama; Sayeeda, Zinat; Lo, Elvis; Assempour, Nazanin; Berjanskii, Mark; Singhal, Sandeep; Arndt, David; Liang, Yonjie; Badran, Hasan; Grant, Jason; Serra-Cayuela, Arnau; Liu, Yifeng; Mandal, Rupa; Neveu, Vanessa; Pon, Allison; Knox, Craig; Wilson, Michael; Manach, Claudine; Scalbert, Augustin
Nucleic Acids Research (2018), 46 (D1), D608-D617CODEN: NARHAD; ISSN:1362-4962. (Oxford University Press)
A review. The Human Metabolome Database or HMDB (www. hmdb.ca) is a web-enabled metabolomic database contg. comprehensive information about human metabolites along with their biol. roles, physiol. concns., disease assocns., chem. reactions, metabolic pathways, and ref. spectra. First described in 2007, the HMDB is now considered the std. metabolomic resource for human metabolic studies. Over the past decade the HMDB has continued to grow and evolve in response to emerging needs for metabolomics researchers and continuing changes in web stds. This year's update, HMDB 4.0, represents the most significant upgrade to the database in its history. For instance, the no. of fully annotated metabolites has increased by nearly threefold, the no. of exptl. spectra has grown by almost fourfold and the no. of illustrated metabolic pathways has grown by a factor of almost 60. Significant improvements have also been made to the HMDB's chem. taxonomy, chem. ontol., spectral viewing, and spectral/text searching tools. A great deal of brand new data has also been added to HMDB 4.0. This includes large quantities of predicted MS/MS and GC-MS ref. spectral data as well as predicted (physiol. feasible) metabolite structures to facilitate novel metabolite identification. Addnl. information on metabolite-SNP interactions and the influence of drugs on metabolite levels (pharmaco-metabolomics) has also been added. Many other important improvements in the content, the interface, and the performance of the HMDB website have been made and these should greatly enhance its ease of use and its potential applications in nutrition, biochem., clin. chem., clin. genetics, medicine, and metabolomics science.
12
Bouatra, S.; Aziat, F.; Mandal, R.; Guo, A. C.; Wilson, M. R.; Knox, C.; Bjorndahl, T. C.; Krishnamurthy, R.; Saleem, F.; Liu, P. The Human Urine Metabolome. PLoS One 2013, 8 (9), e73076 DOI: 10.1371/journal.pone.0073076
Google Scholar
12
The human urine metabolome
Bouatra, Souhaila; Aziat, Farid; Mandal, Rupasri; Guo, An Chi; Wilson, Michael R.; Knox, Craig; Bjorndahl, Trent C.; Krishnamurthy, Ramanarayan; Saleem, Fozia; Liu, Philip; Dame, Zerihun T.; Poelzer, Jenna; Huynh, Jessica; Yallou, Faizath S.; Psychogios, Nick; Dong, Edison; Bogumil, Ralf; Roehring, Cornelia; Wishart, David S.
PLoS One (2013), 8 (9), e73076CODEN: POLNCL; ISSN:1932-6203. (Public Library of Science)
A review. Urine has long been a "favored" biofluid among metabolomics researchers. It is sterile, easy-to-obtain in large vols., largely free from interfering proteins or lipids and chem. complex. However, this chem. complexity has also made urine a particularly difficult substrate to fully understand. As a biol. waste material, urine typically contains metabolic breakdown products from a wide range of foods, drinks, drugs, environmental contaminants, endogenous waste metabolites and bacterial byproducts. Many of these compds. are poorly characterized and poorly understood. In an effort to improve our understanding of this biofluid we have undertaken a comprehensive, quant., metabolome-wide characterization of human urine. This involved both computer-aided literature mining and comprehensive, quant. exptl. assessment/validation. The exptl. portion employed NMR spectroscopy, gas chromatog. mass spectrometry (GC-MS), direct flow injection mass spectrometry (DFI/LC-MS/MS), inductively coupled plasma mass spectrometry (ICP-MS) and high performance liq. chromatog. (HPLC) expts. performed on multiple human urine samples. This multi-platform metabolomic anal. allowed us to identify 445 and quantify 378 unique urine metabolites or metabolite species. The different anal. platforms were able to identify (quantify) a total of: 209 (209) by NMR, 179 (85) by GC-MS, 127 (127) by DFI/LC-MS/MS, 40 (40) by ICP-MS and 10 (10) by HPLC. Our use of multiple metabolomics platforms and technologies allowed us to identify several previously unknown urine metabolites and to substantially enhance the level of metabolome coverage. It also allowed us to critically assess the relative strengths and weaknesses of different platforms or technologies. The literature review led to the identification and annotation of another 2206 urinary compds. and was used to help guide the subsequent exptl. studies. An online database contg. the complete set of 2651 confirmed human urine metabolite species, their structures (3079 in total), concns., related literature refs. and links to their known disease assocns. are freely available at online.
13
Psychogios, N.; Hau, D. D.; Peng, J.; Guo, A. C.; Mandal, R.; Bouatra, S.; Sinelnikov, I.; Krishnamurthy, R.; Eisner, R.; Gautam, B. The Human Serum Metabolome. PLoS One 2011, 6 (2), e16957 DOI: 10.1371/journal.pone.0016957
Google Scholar
13
The human serum metabolome
Psychogios, Nikolaos; Hau, David D.; Peng, Jun; Guo, An. Chi; Mandal, Rupasri; Bouatra, Souhaila; Sinelnikov, Igor; Krishnamurthy, Ramanarayan; Eisner, Roman; Gautam, Bijaya; Young, Nelson; Xia, Jianguo; Knox, Craig; Dong, Edison; Huang, Paul; Hollander, Zsuzsanna; Pedersen, Theresa L.; Smith, Steven R.; Bamforth, Fiona; Greiner, Russ; McManus, Bruce; Newman, John W.; Goodfriend, Theodore; Wishart, David S.
PLoS One (2011), 6 (2), e16957CODEN: POLNCL; ISSN:1932-6203. (Public Library of Science)
Continuing improvements in anal. technol. along with an increased interest in performing comprehensive, quant. metabolic profiling, is leading to increased interest pressures within the metabolomics community to develop centralized metabolite ref. resources for certain clin. important biofluids, such as cerebrospinal fluid, urine and blood. As part of an ongoing effort to systematically characterize the human metabolome through the Human Metabolome Project, we have undertaken the task of characterizing the human serum metabolome. In doing so, we have combined targeted and non-targeted NMR, GC-MS and LC-MS methods with computer-aided literature mining to identify and quantify a comprehensive, if not absolutely complete, set of metabolites commonly detected and quantified (with today's technol.) in the human serum metabolome. Our use of multiple metabolomics platforms and technologies allowed us to substantially enhance the level of metabolome coverage while critically assessing the relative strengths and weaknesses of these platforms or technologies. Tables contg. the complete set of 4229 confirmed and highly probable human serum compds., their concns., related literature refs. and links to their known disease assocns. are freely available online.
14
Nagana Gowda, G. A.; Gowda, Y. N.; Raftery, D. Expanding the Limits of Human Blood Metabolite Quantitation Using NMR Spectroscopy. Anal. Chem. 2015, 87 (1), 706– 715, DOI: 10.1021/ac503651e
Google Scholar
14
Expanding the Limits of Human Blood Metabolite Quantitation Using NMR Spectroscopy
Nagana Gowda, G. A.; Gowda, Yashas N.; Raftery, Daniel
Analytical Chemistry (Washington, DC, United States) (2015), 87 (1), 706-715CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
A current challenge in metabolomics is the reliable quantitation of many metabolites. Limited resoln. and sensitivity combined with the challenges assocd. with unknown metabolite identification have restricted both the no. and the quant. accuracy of blood metabolites. Focused on alleviating this bottleneck in NMR-based metabolomics, studies of pooled human serum combining an array of 1D/2D NMR expts. at 800 MHz, database searches, and spiking with authentic compds. enabled the identification of 67 blood metabolites. Many of these (∼1/3) are new compared with those reported previously as a part of the Human Serum Metabolome Database. In addn., considering both the high reproducibility and quant. nature of NMR as well as the sensitivity of NMR chem. shifts to altered sample conditions, exptl. protocols and comprehensive peak annotations are provided here as a guide for identification and quantitation of the new pool of blood metabolites for routine applications. Further, studies focused on the evaluation of quantitation using org. solvents revealed a surprisingly poor performance for protein pptn. using acetonitrile. One-third of the detected metabolites were attenuated by 10-67% compared with methanol pptn. at the same solvent-to-serum ratio of 2:1 (vol./vol.). Nearly 2/3 of the metabolites were further attenuated by up to 65% upon increasing the acetonitrile-to-serum ratio to 4:1 (vol./vol.). These results, combined with the newly established identity for many unknown metabolites in the NMR spectrum, offer new avenues for human serum/plasma-based metabolomics. Further, the ability to quant. evaluate nearly 70 blood metabolites that represent numerous classes, including amino acids, org. acids, carbohydrates, and heterocyclic compds., using a simple and highly reproducible anal. method such as NMR may potentially guide the evaluation of samples for anal. using mass spectrometry.
15
Rueedi, R.; Ledda, M.; Nicholls, A. W.; Salek, R. M.; Marques-Vidal, P.; Morya, E.; Sameshima, K.; Montoliu, I.; Da Silva, L.; Collino, S. Genome-Wide Association Study of Metabolic Traits Reveals Novel Gene-Metabolite-Disease Links. PLoS Genet. 2014, 10 (2), e1004132 DOI: 10.1371/journal.pgen.1004132
Google Scholar
15
Genome-wide association study of metabolic traits reveals novel gene-metabolite-disease links
Rueedi, Rico; Ledda, Mirko; Nicholls, Andrew W.; Salek, Reza M.; Marques-Vidal, Pedro; Morya, Edgard; Sameshima, Koichi; Montoliu, Ivan; Da Silva, Laeticia; Collino, Sebastiano; Martin, Francois-Pierre; Rezzi, Serge; Steinbeck, Christoph; Waterworth, Dawn M.; Waeber, Gerard; Vollenweider, Peter; Beckmann, Jacques S.; Le Coutre, Johannes; Mooser, Vincent; Bergmann, Sven; Genick, Ulrich K.; Kutalik, Zoltan
PLoS Genetics (2014), 10 (2), e1004132/1-e1004132/10, 10 pp.CODEN: PGLEB5; ISSN:1553-7404. (Public Library of Science)
Metabolic traits are mol. phenotypes that can drive clin. phenotypes and may predict disease progression. Here, we report results from a metabolome- and genome-wide assocn. study on 1H-NMR urine metabolic profiles. The study was conducted within an untargeted approach, employing a novel method for compd. identification. From our discovery cohort of 835 Caucasian individuals who participated in the CoLaus study, we identified 139 suggestively significant (P<5 × 10-8) and independent assocns. between single nucleotide polymorphisms (SNP) and metabolome features. Fifty-six of these assocns. replicated in the TasteSensomics cohort, comprising 601 individuals from S~ao Paulo of vastly diverse ethnic background. They correspond to eleven gene-metabolite assocns., six of which had been previously identified in the urine metabolome and three in the serum metabolome. Our key novel findings are the assocns. of two SNPs with NMR spectral signatures pointing to fucose (rs492602, P = 6.9 × 10-44) and lysine (rs8101881, P = 1.2 × 10-33), resp. Fine-mapping of the first locus pinpointed the FUT2 gene, which encodes a fucosyltransferase enzyme and has previously been assocd. with Crohn's disease. This implicates fucose as a potential prognostic disease marker, for which there is already published evidence from a mouse model. The second SNP lies within the SLC7A9 gene, rare mutations of which have been linked to severe kidney damage. The replication of previous assocns. and our new discoveries demonstrate the potential of untargeted metabolomics GWAS to robustly identify mol. disease markers.
16
Rueedi, R.; Mallol, R.; Raffler, J.; Lamparter, D.; Friedrich, N.; Vollenweider, P.; Waeber, G.; Kastenmüller, G.; Kutalik, Z.; Bergmann, S. Metabomatching: Using Genetic Association to Identify Metabolites in Proton NMR Spectroscopy. PLoS Comput. Biol. 2017, 13 (12), e1005839 DOI: 10.1371/journal.pcbi.1005839
Google Scholar
16
Metabomatching: using genetic association to identify metabolites in proton NMR spectroscopy
Rueedi, Rico; Mallol, Roger; Raffler, Johannes; Lamparter, David; Friedrich, Nele; Vollenweider, Peter; Waeber, Gerard; Kastenmuller, Gabi; Kutalik, Zoltan; Bergmann, Sven
PLoS Computational Biology (2017), 13 (12), e1005839/1-e1005839/17CODEN: PCBLBG; ISSN:1553-7358. (Public Library of Science)
A metabolome-wide genome-wide assocn. study aims to discover the effects of genetic variants on metabolome phenotypes. Most mGWASes use as phenotypes concns. of limited sets of metabolites that can be identified and quantified from spectral information. In contrast, in an untargeted mGWAS both identification and quantification are forgone and, instead, all measured metabolome features are tested for assocn. with genetic variants. While the untargeted approach does not discard data that may have eluded identification, the interpretation of assocd. features remains a challenge. To address this issue, we developed metabomatching to identify the metabolites underlying significant assocns. obsd. in untargeted mGWASes on proton NMR metabolome data. Metabomatching capitalizes on genetic spiking, the concept that because metabolome features assocd. with a genetic variant tend to correspond to the peaks of the NMR spectrum of the underlying metabolite, genetic assocn. can allow for identification. Applied to the untargeted mGWASes in the SHIP and CoLaus cohorts and using 180 ref. NMR spectra of the urine metabolome database, metabomatching successfully identified the underlying metabolite in 14 of 19, and 8 of 9 assocns., resp. The accuracy and efficiency of our method make it a strong contender for facilitating or complementing metabolomics analyses in large cohorts, where the availability of genetic, or other data, enables our approach, but targeted quantification is limited.
17
Raffler, J.; Friedrich, N.; Arnold, M.; Kacprowski, T.; Rueedi, R.; Altmaier, E.; Bergmann, S.; Budde, K.; Gieger, C.; Homuth, G. Genome-Wide Association Study with Targeted and Non-Targeted NMR Metabolomics Identifies 15 Novel Loci of Urinary Human Metabolic Individuality. PLoS Genet. 2015, 11 (9), e1005487 DOI: 10.1371/journal.pgen.1005487
Google Scholar
17
Genome-wide association study with targeted and non-targeted nmr metabolomics identifies 15 novel loci of urinary human metabolic individuality
Raffler, Johannes; Friedrich, Nele; Arnold, Matthias; Kacprowski, Tim; Rueedi, Rico; Altmaier, Elisabeth; Bergmann, Sven; Budde, Kathrin; Gieger, Christian; Homuth, Georg; Pietzner, Maik; Roemisch-Margl, Werner; Strauch, Konstantin; Voelzke, Henry; Waldenberger, Melanie; Wallaschofski, Henri; Nauck, Matthias; Voelker, Uwe; Kastenmueller, Gabi; Suhre, Karsten
PLoS Genetics (2015), 11 (9), e1005487/1-e1005487/28CODEN: PGLEB5; ISSN:1553-7404. (Public Library of Science)
Genome-wide assocn. studies with metabolic traits (mGWAS) uncovered many genetic variants that influence human metab. These genetically influenced metabotypes (GIMs) contribute to our metabolic individuality, our capacity to respond to environmental challenges, and our susceptibility to specific diseases. While metabolic homeostasis in blood is a well investigated topic in large mGWAS with over 150 known loci, metabolic detoxification through urinary excretion has only been addressed by few small mGWAS with only 11 assocd. loci so far. Here we report the largest mGWAS to date, combining targeted and non-targeted 1H NMR anal. of urine samples from 3,861 participants of the SHIP-0 cohort and 1,691 subjects of the KORA F4 cohort. We identified and replicated 22 loci with significant assocns. with urinary traits, 15 of which are new (HIBCH, CPS1, AGXT, XYLB, TKT, ETNPPL, SLC6A19, DMGDH, SLC36A2, GLDC, SLC6A13, ACSM3, SLC5A11, PNMT, SLC13A3). Two-thirds of the urinary loci also have a metabolite assocn. in blood. For all but one of the 6 loci where significant assocns. target the same metabolite in blood and urine, the genetic effects have the same direction in both fluids. In contrast, for the SLC5A11 locus, we found increased levels of myo-inositol in urine whereas mGWAS in blood reported decreased levels for the same genetic variant. This might indicate less effective re-absorption of myo-inositol in the kidneys of carriers. In summary, our study more than doubles the no. of known loci that influence urinary phenotypes. It thus allows novel insights into the relationship between blood homeostasis and its regulation through excretion. The newly discovered loci also include variants previously linked to chronic kidney disease (CPS1, SLC6A13), pulmonary hypertension (CPS1), and ischemic stroke (XYLB). By establishing connections from gene to disease via metabolic traits our results provide novel hypotheses about mol. mechanisms involved in the etiol. of diseases.
18
Ihmels, J.; Bergmann, S.; Barkai, N. Defining Transcription Modules Using Large-Scale Gene Expression Data. Bioinformatics 2004, 20 (13), 1993– 2003, DOI: 10.1093/bioinformatics/bth166
Google Scholar
18
Defining transcription modules using large-scale gene expression data
Ihmels, Jan; Bergmann, Sven; Barkai, Naama
Bioinformatics (2004), 20 (13), 1993-2003CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Large-scale gene expression data comprising a variety of cellular conditions hold the promise of a global view on the transcription program. While conventional clustering algorithms have been successfully applied to smaller datasets, the utility of many algorithms for the anal. of large-scale data is limited by their inability to capture combinatorial and condition-specific co-regulation. In addn., there is an increasing need to integrate the rapidly accumulating body of other high-throughput biol. data with the expression anal. In a previous work, the signature algorithm was introduced, which overcomes the problems of conventional clustering and allows for intuitive integration of addnl. biol. data. However, this approach is constrained by the comprehensiveness of relevant external data and its lacking ability to capture hierarchical modularity. A novel method has been presented for the anal. of large-scale expression data, which assigns genes into context-dependent and potentially overlapping regulatory units. Authors introduce the notion of a transcription module as a self-consistent regulatory unit consisting of a set of co-regulated genes as well as the exptl. conditions that induce their co-regulation. Self-consistency is defined by a rigorous math. criterion. Authors propose an efficient algorithm to identify such modules, which is based on the iterative application of the signature algorithm. A threshold parameter that dets. the resoln. of the modular decompn. is introduced. The method is applied systematically to over 1000 expression profiles of the yeast Saccharomyces cerevisiae, and the results are presented using two complementary visualization schemes developed. The av. biol. coherence, as measured by the conservation of putative cis-regulatory motifs between four related yeast species, is higher for transcription modules than for clusters identified by other methods applied to the same dataset. This method is related to singular value decompn. (SVD) and to the pairwise av. linkage clustering algorithm. It extends SVD by filtering out noise in the expression data and offering variable resoln. to reveal hierarchical organization. It furthermore has the advantage over both methods of capturing overlapping modules in the presence of combinatorial regulation.
19
Bergmann, S.; Ihmels, J.; Barkai, N. Iterative Signature Algorithm for the Analysis of Large-Scale Gene Expression Data. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2003, 67 (3), 031902 DOI: 10.1103/PhysRevE.67.031902
Google Scholar
There is no corresponding record for this reference.
20
Xiong, X.; Liu, D.; Wang, Y.; Zeng, T.; Peng, Y. Urinary 3-(3-Hydroxyphenyl)-3-Hydroxypropionic Acid, 3-Hydroxyphenylacetic Acid, and 3-Hydroxyhippuric Acid Are Elevated in Children with Autism Spectrum Disorders. BioMed Res. Int. 2016, 2016, 9485412, DOI: 10.1155/2016/9485412
Google Scholar
20
Urinary 3-(3-Hydroxyphenyl)-3-hydroxypropionic Acid, 3-Hydroxyphenylacetic Acid, and 3-Hydroxyhippuric Acid Are Elevated in Children with Autism Spectrum Disorders
Xiong Xiyue; Liu Dan; Wang Yichao; Zeng Ting; Peng Ying
BioMed research international (2016), 2016 (), 9485412 ISSN:.
Autism spectrum disorders (ASDs) are a group of mental illnesses highly correlated with gut microbiota. Recent studies have shown that some abnormal aromatic metabolites in autism patients are presumably derived from overgrown Clostridium species in gut, which may be used for diagnostic purposes. In this paper, a GC/MS based metabolomic approach was utilized to seek similar biomarkers by analyzing the urinary information in 62 ASDs patients compared with 62 non-ASDs controls in China, aged 1.5-7. Three compounds identified as 3-(3-hydroxyphenyl)-3-hydroxypropionic acid (HPHPA), 3-hydroxyphenylacetic acid (3HPA), and 3-hydroxyhippuric acid (3HHA) were found in higher concentrations in autistic children than in the controls (p < 0.001). After oral vancomycin treatment, urinary excretion of HPHPA (p < 0.001), 3HPA (p < 0.005), and 3HHA (p < 0.001) decreased markedly, which indicated that these compounds may also be from gut Clostridium species. The sensitivity and specificity of HPHPA, 3HPA, and 3HHA were evaluated by receiver-operating characteristic (ROC) analysis. The specificity of each compound for ASDs was very high (>96%). After two-regression analysis, the optimal area under the curve (AUC, 0.962), sensitivity (90.3%), and specificity (98.4%) were obtained by ROC curve of Prediction probability based on the three metabolites. These findings demonstrate that the measurements of the three compounds are strong predictors of ASDs and support the potential clinical utility for identifying a subgroup of ASDs subjects.
21
Nielsen, H. R.; Killmann, S. A. Urinary Excretion of Beta-Aminoisobutyrate and Pseudouridine in Acute and Chronic Myeloid Leukemia. J. Natl. Cancer Inst. 1983, 71 (5), 887– 891
Google Scholar
21
Urinary excretion of β-aminoisobutyrate and pseudouridine in acute and chronic myeloid leukemia
Nielsen, Henrik R.; Killmann, Sven Aage
JNCI, Journal of the National Cancer Institute (1983), 71 (5), 887-91CODEN: JJIND8; ISSN:0198-0157.
β-Aminoisobutyrate (β-AIB) and pseudouridine (ψ-Urd) urinary excretion was investigated in abnormal hematopoietic conditions including 26 patients with acute myeloid leukemia (AML) and chronic myeloid leukemia (CML) and was compared to that of 25 healthy controls. β-AIB excretion in CML was directly correlated to the leukocyte count, the indicator of tumor cell mass. β-AIB excretion was elevated in 27 and 75% of untreated AML and CML cases, resp. Marrow blast cell content tended to correlate pos. with ψ-Urd excretion in AML. ψ-Urd excretion was elevated in 82 and 87% of untreated AML and CML, resp. Turnover of hematopoietic cells seemed to be a determinant for β-AIB excretion, including higher cell turnover in CML patients compared to that in AML patients and in controls. With cytostatic treatment, excretion levels of β-AIB and/or ψ-Urd decreased after a transient rise.
22
Ziegler, E. E., Ed.; Present Knowledge in Nutrition; Filer, L. J. J., Engl. Ed.; International Life Sciences Inst.: Washington, D.C., 1996.
Google Scholar
There is no corresponding record for this reference.
23
Nicholas, P. C.; Kim, D.; Crews, F. T.; Macdonald, J. M. Proton Nuclear Magnetic Resonance Spectroscopic Determination of Ethanol-Induced Formation of Ethyl Glucuronide in Liver. Anal. Biochem. 2006, 358 (2), 185– 191, DOI: 10.1016/j.ab.2006.08.033
Google Scholar
23
Proton nuclear magnetic resonance spectroscopic determination of ethanol-induced formation of ethyl glucuronide in liver
Nicholas, Peter C.; Kim, Daniel; Crews, Fulton T.; Macdonald, Jeffrey M.
Analytical Biochemistry (2006), 358 (2), 185-191CODEN: ANBCA2; ISSN:0003-2697. (Elsevier)
Et glucuronide (ethyl-β-D-6-glucosiduronic acid, EtG), a unique metabolite of EtOH, has received much recent attention as a sensitive and specific biol. marker of EtOH consumption. Formed in the liver via conjugation of EtOH with activated glucuronate, EtG remains detectable in serum, plasma, and hair for days after EtOH abuse. Thus far, gas chromatog.-mass spectrometry and enzyme-linked immunosorbent assays were developed to detect trace quantities of EtG for forensic purposes, but reports of the NMR properties of EtG were scarce. Herein the authors present the first report of EtG detn. using proton NMR spectroscopy. The authors collected 700-MHz proton spectra of liver exts. from rats treated with a 4-day binge EtOH protocol (av. EtOH dose: 8.6 g/kg/day). An unexpected signal (triplet, 1.24 ppm) appeared in EtOH-treated liver exts. but not in control samples; based on chem. shift and multiplicity, the authors suspected EtG. The authors obsd. quant. hydrolysis of the unknown species to EtOH while incubating the samples with β-glucuronidase, confirming that the Me protons of EtG were responsible for the triplet at 1.24 ppm. This study demonstrates that proton NMR spectroscopy is capable of detecting EtG and that future NMR-based metabolomic studies may encounter this metabolite of EtOH.
24
Kim, S.; Lee, M.; Yoon, D.; Lee, D.-K.; Choi, H.-J.; Kim, S. 1D Proton NMR Spectroscopic Determination of Ethanol and Ethyl Glucuronide in Human Urine. Bull. Korean Chem. Soc. 2013, 34 (8), 2413– 2418, DOI: 10.5012/bkcs.2013.34.8.2413
Google Scholar
24
1D proton NMR spectroscopic determination in ethanol and ethyl glucuronide in human urine
Kim, Siwon; Lee, Minji; Yoon, Dahye; Lee, Dong-kye; Choi, Hye-Jin; Kim, Suhkmann
Bulletin of the Korean Chemical Society (2013), 34 (8), 2413-2418CODEN: BKCSDE; ISSN:0253-2964. (Korean Chemical Society)
Forensic and legal medicine require reliable data to indicate excessive alc. consumption. Ethanol is oxidatively metabolized to acetate by alc. dehydrogenase and non-oxidatively metabolized to Et glucuronide (EtG), Et sulfate (EtS), phosphatidylethanol, or fatty acid Et esters (FAEE). Oxidative metab. is too rapid to provide biomarkers for the detection of ethanol ingestion. However, the nonoxidative metabolite EtG is a useful biomarker because it is stable, non-volatile, water sol., highly sensitive, and is detected in body fluid, hair, and tissues. EtG anal. methods such as mass spectroscopy, chromatog., or ELISA techniques are currently in use. We suggest that NMR (NMR) spectroscopy could be used to monitor ethanol intake. As with current conventional methods, NMR spectroscopy doesn't require complicated pretreatments or sample sepn. This method has the advantages of short acquisition time, simple sample prepn., reproducibility, and accuracy. In addn., all proton-contg. compds. can be detected. In this study, we performed 1H NMR analyses of urine to monitor the ethanol and EtG. Urinary samples were collected over time from 5 male volunteers. We confirmed that ethanol and EtG signals could be detected with NMR spectroscopy. Ethanol signals increased immediately upon alc. intake, but decreased sharply over time. In contrast, EtG signal increased and reached a max. about 9 h later, after which the EtG signal decreased gradually and remained detectable after 20-25 h. Based on these results, we suggest that 1H NMR spectroscopy may be used to identify ethanol non-oxidative metabolites without the need for sample pretreatment.
25
Solomons, H. D. Carbohydrate Deficient Transferrin and Alcoholism. GERMS 2012, 2 (2), 75– 78, DOI: 10.11599/germs.2012.1015
Google Scholar
25
Carbohydrate deficient transferrin and alcoholism
Solomons, Hilary Denis
GERMS (2012), 2 (2), 75-78CODEN: GERMCF; ISSN:2248-2997. (European Academy of HIV/AIDS and Infectious Diseases)
A review. Alc. abuse is an important public health problem, with major implications in patients with a preexisting liver pathol. of viral origin. Hepatitis C, for example, is one of the diseases in which alc. consumption can lead to the transition from a fairly benign outline to a potentially life-threatening liver disease. Alc. abuse is usually identified on the basis of clin. judgment, alcoholism related questionnaires, lab. tests and, more recently, biomarkers. Also on this list of tests, carbohydrate deficient transferrin (CDT) is widely available and useful for detg. recent alc. consumption, particularly when corroborated with elevation of other liver-assocd. enrymes. Clinicians should be aware of the indications and limitations of this test in order to better evaluate alc. consumption in their patients.

Cited By

Click to copy section linkSection link copied!

Explore this article's citation statements on scite.ai

This article is cited by 5 publications.

Reyhan Sönmez Flitman, Bita Khalili, Zoltan Kutalik, Rico Rueedi, Anneke Brümmer, Sven Bergmann. Untargeted Metabolome- and Transcriptome-Wide Association Study Suggests Causal Genes Modulating Metabolite Concentrations in Urine. Journal of Proteome Research 2021, 20 (11) , 5103-5114. https://doi.org/10.1021/acs.jproteome.1c00585
Michael T. Judge, Timothy M. D. Ebbels. Problems, principles and progress in computational annotation of NMR metabolomics data. Metabolomics 2022, 18 (12) https://doi.org/10.1007/s11306-022-01962-z
Carmelo Corsaro, Sebastiano Vasi, Fortunato Neri, Angela Maria Mezzasalma, Giulia Neri, Enza Fazio. NMR in Metabolomics: From Conventional Statistics to Machine Learning and Neural Network Approaches. Applied Sciences 2022, 12 (6) , 2824. https://doi.org/10.3390/app12062824
Olimpia Masetti, Angela Sorbo, Luigi Nisini. NMR Tracing of Food Geographical Origin: The Impact of Seasonality, Cultivar and Production Year on Data Analysis. Separations 2021, 8 (12) , 230. https://doi.org/10.3390/separations8120230
David Joseph, Sujeesh Sukumaran, Kousik Chandra, Shivanand M. Pudakalakatti, Abhinav Dubey, Amrinder Singh, Hanudatta S. Atreya. Rapid nuclear magnetic resonance data acquisition with improved resolution and sensitivity for high‐throughput metabolomic analysis. Magnetic Resonance in Chemistry 2021, 59 (3) , 300-314. https://doi.org/10.1002/mrc.5106

Download PDF

Get e-Alerts

Journal of Proteome Research

Cite this: J. Proteome Res. 2019, 18, 9, 3360–3368

Click to copy citationCitation copied!

https://doi.org/10.1021/acs.jproteome.9b00295

Published July 18, 2019

Request reuse permissions

Article Views

Altmetric

Citations

Learn about these metrics

Article Views are the COUNTER-compliant sum of full text article downloads since November 2008 (both PDF and HTML) across all institutions and individuals. These metrics are regularly updated to reflect usage leading up to the last few days.

Citations are the number of other articles citing this article, calculated by Crossref and updated daily. Find more information about Crossref citation counts.

The Altmetric Attention Score is a quantitative measure of the attention that a research article has received online. Clicking on the donut icon will load a page at altmetric.com with additional details about the score and the social media presence for the given article. Find more information on the Altmetric Attention Score and how the score is calculated.

Abstract
High Resolution Image
Download MS PowerPoint Slide
Figure 1
Figure 1. Workflow for unsupervised analysis of large-scale NMR data. Raw ¹H NMR data are normalized then aligned. These processed profiles are used as input for the averaged correlation profile (ACP), iterative signature algorithm (ISA), and principal component analysis (PCA) methods, which output correlation profiles, module scores, and PCA loadings, respectively. These outputs constitute possible pseudospectra for metabomatching, which identifies the most plausible candidate metabolites underlying the coherent feature variations.
High Resolution Image
Download MS PowerPoint Slide
Figure 2
Figure 2. Urine metabolites that were robustly matched by metabomatching to pseudospectra derived from average correlation profiles (ACP, green), the iterative signature algorithm (ISA, blue), or both methods (black).
High Resolution Image
Download MS PowerPoint Slide
Figure 3
Figure 3. Pseudospectra from ACP and ISA algorithms matching glucose, ethanol, and EtG. Each plot shows the pseudospectrum in blue in the upper half and the reference spectrum from UMDB in black and in the lower half. Dark blue indicates chemical shifts and their ±0.025 ppm vicinity that were used for pseudoquantification.
High Resolution Image
Download MS PowerPoint Slide
References
This article references 25 other publications.
1. 1
  Cloarec, O.; Dumas, M.-E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gauguier, D.; Lindon, J. C.; Holmes, E. Statistical Total Correlation Spectroscopy: An Exploratory Approach for Latent Biomarker Identification from Metabolic 1H NMR Data Sets. Anal. Chem. 2005, 77 (5), 1282– 1289, DOI: 10.1021/ac048630x
  1
  Statistical Total Correlation Spectroscopy: An Exploratory Approach for Latent Biomarker Identification from Metabolic 1H NMR Data Sets
  Cloarec, Olivier; Dumas, Marc-Emmanuel; Craig, Andrew; Barton, Richard H.; Trygg, Johan; Hudson, Jane; Blancher, Christine; Gauguier, Dominique; Lindon, John C.; Holmes, Elaine; Nicholson, Jeremy
  Analytical Chemistry (2005), 77 (5), 1282-1289CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
  We describe here the implementation of the statistical total correlation spectroscopy (STOCSY) anal. method for aiding the identification of potential biomarker mols. in metabonomic studies based on NMR spectroscopic data. STOCSY takes advantage of the multicollinearity of the intensity variables in a set of spectra (in this case 1H NMR spectra) to generate a pseudo-two-dimensional NMR spectrum that displays the correlation among the intensities of the various peaks across the whole sample. This method is not limited to the usual connectivities that are deducible from more std. two-dimensional NMR spectroscopic methods, such as TOCSY. Moreover, two or more mols. involved in the same pathway can also present high intermol. correlations because of biol. covariance or can even be anticorrelated. This combination of STOCSY with supervised pattern recognition and particularly orthogonal projection on latent structure-discriminant anal. (O-PLS-DA) offers a new powerful framework for anal. of metabonomic data. In a first step O-PLS-DA exts. the part of NMR spectra related to discrimination. This information is then cross-combined with the STOCSY results to help identify the mols. responsible for the metabolic variation. To illustrate the applicability of the method, it has been applied to 1H NMR spectra of urine from a metabonomic study of a model of insulin resistance based on the administration of a carbohydrate diet to three different mice strains (C57BL/6Oxjr, BALB/cOxjr, and 129S6/SvEvOxjr) in which a series of metabolites of biol. importance can be conclusively assigned and identified by use of the STOCSY approach.
2. 2
  Blaise, B. J.; Navratil, V.; Domange, C.; Shintu, L.; Dumas, M.-E.; Elena-Herrmann, B.; Emsley, L.; Toulhoat, P. Two-Dimensional Statistical Recoupling for the Identification of Perturbed Metabolic Networks from NMR Spectroscopy. J. Proteome Res. 2010, 9 (9), 4513– 4520, DOI: 10.1021/pr1002615
  2
  Two-Dimensional Statistical Recoupling for the Identification of Perturbed Metabolic Networks from NMR Spectroscopy
  Blaise, Benjamin J.; Navratil, Vincent; Domange, Celine; Shintu, Laetitia; Dumas, Marc-Emmanuel; Elena-Herrmann, Benedicte; Emsley, Lyndon; Toulhoat, Pierre
  Journal of Proteome Research (2010), 9 (9), 4513-4520CODEN: JPROBS; ISSN:1535-3893. (American Chemical Society)
  The development of Statistical Total Correlation Spectroscopy (STOCSY), a representation of the autocorrelation matrix of a spectral data set as a 2D pseudospectrum, has allowed more reliable assignment of one- and two-dimensional NMR spectra acquired from the complex mixts. that are usually used in metabolomics/metabonomics studies, thus, improving precise identification of candidate biomarkers contained in metabolic signatures computed by multivariate statistical anal. However, the correlations obtained cannot always be interpreted in terms of connectivities between metabolites. In this study, the authors combine statistical recoupling of variables (SRV) and STOCSY to identify perturbed metabolite systems. The resulting Recoupled-STOCSY (R-STOCSY) method provides a 2D correlation landscape based on the SRV clusters representing phys., chem., and biol. entities. This enables the identification of correlations between distant clusters and extends the recoupling scheme of SRV, which was previously limited to the assocn. of neighboring clusters. This allows the recovery of only meaningful correlations between metabolic signals and significantly enhances the interpretation of STOCSY. The method is validated through the measurement of the distances between the metabolites involved in these correlations, within the whole metabolic network, which shows that the av. shortest path length is significantly shorter for the correlations detected in this new way compared to metabolite couples randomly selected from within the entire KEGG metabolic network. This enables the identification without any a priori knowledge of the perturbed metabolic network. The R-STOCSY completes the recoupling procedure between distant clusters, further reducing the high dimensionality of metabolomics/metabonomics data set and finally allows the identification of composite biomarkers, highlighting disruption of particular metabolic pathways within a global metabolic network. This allows the perturbed metabolic network to be extd. through NMR based metabolomics/metabonomics in an automated, and statistical manner.
3. 3
  Sands, C. J.; Coen, M.; Ebbels, T. M. D.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Data-Driven Approach for Metabolite Relationship Recovery in Biological 1H NMR Data Sets Using Iterative Statistical Total Correlation Spectroscopy. Anal. Chem. 2011, 83 (6), 2075– 2082, DOI: 10.1021/ac102870u
  3
  Data-Driven Approach for Metabolite Relationship Recovery in Biological 1H NMR Data Sets Using Iterative Statistical Total Correlation Spectroscopy
  Sands, Caroline J.; Coen, Muireann; Ebbels, Timothy M. D.; Holmes, Elaine; Lindon, John C.; Nicholson, Jeremy K.
  Analytical Chemistry (Washington, DC, United States) (2011), 83 (6), 2075-2082CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
  Statistical total correlation spectroscopy (STOCSY) is a well-established and valuable method in the elucidation of both inter- and intrametabolite correlations in NMR metabonomic data sets. Here, the STOCSY approach is extended in a novel Iterative-STOCSY (I-STOCSY) tool in which correlations are calcd. initially from a driver peak of interest and subsequently for all peaks identified as correlating with a correlation coeff. greater than a set threshold. Consequently, in a single automated run, the majority of information contained in multiple STOCSY calcns. from all peaks recursively correlated to the original user defined driver peak of interest are recovered. In addn., highly correlating peaks are clustered into putative structurally related sets, and the results are presented in a fully interactive plot where each set is represented by a node; node-to-node connections are plotted alongside corresponding spectral data colored by the strength of connection, thus allowing the intuitive exploration of both inter- and intrametabolite connections. The I-STOCSY approach has been here applied to a 1H NMR data set of 24 h postdose aq. liver exts. from rats treated with the model hepatotoxin galactosamine and has been shown both to recover the previously deduced major metabolic effects of treatment and to generate new hypotheses even on this well-studied model system. I-STOCSY, thus, represents a significant advance in correlation based anal. and visualization, providing insight into inter- and intrametabolite relationships following metabolic perturbations.
4. 4
  Posma, J. M.; Garcia-Perez, I.; De Iorio, M.; Lindon, J. C.; Elliott, P.; Holmes, E.; Ebbels, T. M. D.; Nicholson, J. K. Subset Optimization by Reference Matching (STORM): An Optimized Statistical Approach for Recovery of Metabolic Biomarker Structural Information from 1H NMR Spectra of Biofluids. Anal. Chem. 2012, 84 (24), 10694– 10701, DOI: 10.1021/ac302360v
  4
  Subset Optimization by Reference Matching (STORM): An Optimized Statistical Approach for Recovery of Metabolic Biomarker Structural Information from 1H NMR Spectra of Biofluids
  Posma, Joram M.; Garcia-Perez, Isabel; De Iorio, Maria; Lindon, John C.; Elliott, Paul; Holmes, Elaine; Ebbels, Timothy M. D.; Nicholson, Jeremy K.
  Analytical Chemistry (Washington, DC, United States) (2012), 84 (24), 10694-10701CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
  We describe a new multivariate statistical approach to recover metabolite structure information from multiple 1H NMR spectra in population sample sets. Subset optimization by ref. matching (STORM) was developed to select subsets of 1H NMR spectra that contain specific spectroscopic signatures of biomarkers differentiating between different human populations. STORM aims to improve the visualization of structural correlations in spectroscopic data by using these reduced spectral subsets contg. smaller nos. of samples than the no. of variables (n « p). We have used statistical shrinkage to limit the no. of false pos. assocns. and to simplify the overall interpretation of the autocorrelation matrix. The STORM approach has been applied to findings from an ongoing human metabolome-wide assocn. study on body mass index to identify a biomarker metabolite present in a subset of the population. Moreover, we have shown how STORM improves the visualization of more abundant NMR peaks compared to a previously published method (statistical total correlation spectroscopy, STOCSY). STORM is a useful new tool for biomarker discovery in the 'omic' sciences that has widespread applicability. It can be applied to any type of data, provided that there is interpretable correlation among variables, and can also be applied to data with more than one dimension (e.g., 2D NMR spectra).
5. 5
  Hao, J.; Astle, W.; De Iorio, M.; Ebbels, T. M. D. BATMAN—an R Package for the Automated Quantification of Metabolites from Nuclear Magnetic Resonance Spectra Using a Bayesian Model. Bioinformatics 2012, 28 (15), 2088– 2090, DOI: 10.1093/bioinformatics/bts308
  5
  BATMAN-an R package for the automated quantification of metabolites from nuclear magnetic resonance spectra using a Bayesian model
  Hao, Jie; Astle, William; De Iorio, Maria; Ebbels, Timothy M. D.
  Bioinformatics (2012), 28 (15), 2088-2090CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Motivation: NMR (NMR) spectra are widely used in metabolomics to obtain metabolite profiles in complex biol. mixts. Common methods used to assign and est. concns. of metabolites involve either an expert manual peak fitting or extra pre-processing steps, such as peak alignment and binning. Peak fitting is very time consuming and is subject to human error. Conversely, alignment and binning can introduce artifacts and limit immediate biol. interpretation of models. Results: We present the Bayesian automated metabolite analyzer for NMR spectra (BATMAN), an R package that deconvolutes peaks from one-dimensional NMR spectra, automatically assigns them to specific metabolites from a target list and obtains concn. ests. The Bayesian model incorporates information on characteristic peak patterns of metabolites and is able to account for shifts in the position of peaks commonly seen in NMR spectra of biol. samples. It applies a Markov chain Monte Carlo algorithm to sample from a joint posterior distribution of the model parameters and obtains concn. ests. with reduced error compared with conventional numerical integration and comparable to manual deconvolution by experienced spectroscopists. Availability and implementation: http://www1.imperial.ac.uk/medicine/people/t.ebbels/ Contact: [email protected].
6. 6
  Alonso, A.; Rodríguez, M. A.; Vinaixa, M.; Tortosa, R.; Correig, X.; Julià, A.; Marsal, S. Focus: A Robust Workflow for One-Dimensional NMR Spectral Analysis. Anal. Chem. 2014, 86 (2), 1160– 1169, DOI: 10.1021/ac403110u
  6
  Focus: A Robust Workflow for One-Dimensional NMR Spectral Analysis
  Alonso, Arnald; Rodriguez, Miguel A.; Vinaixa, Maria; Tortosa, Raul; Correig, Xavier; Julia, Antonio; Marsal, Sara
  Analytical Chemistry (Washington, DC, United States) (2014), 86 (2), 1160-1169CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
  One-dimensional 1H NMR represents one of the most commonly used anal. techniques in metabolomic studies. The increase in the no. of samples analyzed as well as the tech. improvements involving instrumentation and spectral acquisition demand increasingly accurate and efficient high-throughput data processing workflows. We present FOCUS, an integrated and innovative methodol. that provides a complete data anal. workflow for one-dimensional NMR-based metabolomics. This tool will allow users to easily obtain a NMR peak feature matrix ready for chemometric anal. as well as metabolite identification scores for each peak that greatly simplify the biol. interpretation of the results. The algorithm development has been focused on solving the crit. difficulties that appear at each data processing step and that can dramatically affect the quality of the results. As well as method integration, simplicity has been one of the main objectives in FOCUS development, requiring very little user input to perform accurate peak alignment, peak picking, and metabolite identification. The new spectral alignment algorithm, RUNAS, allows peak alignment with no need of a ref. spectrum, and therefore, it reduces the bias introduced by other alignment approaches. Spectral alignment has been tested against previous methodologies obtaining substantial improvements in the case of moderate or highly unaligned spectra. Metabolite identification has also been significantly improved, using the positional and correlation peak patterns in contrast to a ref. metabolite panel. Furthermore, the complete workflow has been tested using NMR data sets from 60 human urine samples and 120 aq. liver exts., reaching a successful identification of 42 metabolites from the two data sets. The open-source software implementation of this methodol. is available at http://www.urr.cat/FOCUS.
7. 7
  Ravanbakhsh, S.; Liu, P.; Bjorndahl, T. C.; Mandal, R.; Grant, J. R.; Wilson, M.; Eisner, R.; Sinelnikov, I.; Hu, X.; Luchinat, C. Accurate, Fully-Automated NMR Spectral Profiling for Metabolomics. PLoS One 2015, 10 (5), e0124219 DOI: 10.1371/journal.pone.0124219
  7
  Accurate, fully-automated NMR spectral profiling for metabolomics
  Ravanbakhsh, Siamak; Liu, Philip; Bjorndahl, Trent C.; Mandal, Rupasri; Grant, Jason R.; Wilson, Michael; Eisner, Roman; Sinelnikov, Igor; Hu, Xiaoyu; Luchinat, Claudio; Greiner, Russell; Wishart, David S.
  PLoS One (2015), 10 (5), e0124219/1-e0124219/15CODEN: POLNCL; ISSN:1932-6203. (Public Library of Science)
  Many diseases cause significant changes to the concns. of small mols. (a.k.a. metabolites) that appear in a person's biofluids, which means such diseases can often be readily detected from a person's "metabolic profile"-i.e., the list of concns. of those metabolites. This information can be extd. from a biofluids NMR (NMR) spectrum. However, due to its complexity, NMR spectral profiling has remained manual, resulting in slow, expensive and error-prone procedures that have hindered clin. and industrial adoption of metabolomics via NMR. This paper presents a system, BAYESIL, which can quickly, accurately, and autonomously produce a person's metabolic profile. Given a 1D 1H NMR spectrum of a complex biofluid (specifically serum or cerebrospinal fluid), BAYESIL can automatically det. the metabolic profile. This requires first performing several spectral processing steps, then matching the resulting spectrum against a ref. compd. library, which contains the "signatures" of each relevant metabolite. BAYESIL views spectral matching as an inference problem within a probabilistic graphical model that rapidly approximates the most probable metabolic profile. Our extensive studies on a diverse set of complex mixts. including real biol. samples (serum and CSF), defined mixts. and realistic computer generated spectra; involving > 50 compds., show that BAYESIL can autonomously find the concn. of NMR-detectable metabolites accurately (∼ 90% correct identification and ∼ 10% quantification error), in less than 5 min on a single CPU. These results demonstrate that BAYESIL is the first fully-automatic publicly-accessible system that provides quant. NMR spectral profiling effectively-with an accuracy on these biofluids that meets or exceeds the performance of trained experts. We anticipate this tool will usher in high-throughput metabolomics and enable a wealth of new applications of NMR in clin. settings.
8. 8
  Tardivel, P. J. C.; Canlet, C.; Lefort, G.; Tremblay-Franco, M.; Debrauwer, L.; Concordet, D.; Servien, R. ASICS: An Automatic Method for Identification and Quantification of Metabolites in Complex 1D 1H NMR Spectra. Metabolomics 2017, 13 (10), 109, DOI: 10.1007/s11306-017-1244-5
  There is no corresponding record for this reference.
9. 9
  Röhnisch, H. E.; Eriksson, J.; Müllner, E.; Agback, P.; Sandström, C.; Moazzami, A. A. AQuA: An Automated Quantification Algorithm for High-Throughput NMR-Based Metabolomics and Its Application in Human Plasma. Anal. Chem. 2018, 90 (3), 2095– 2102, DOI: 10.1021/acs.analchem.7b04324
  9
  AQuA: An Automated Quantification Algorithm for High-Throughput NMR-Based Metabolomics and Its Application in Human Plasma
  Rohnisch Hanna E; Eriksson Jan; Mullner Elisabeth; Agback Peter; Sandstrom Corine; Moazzami Ali A
  Analytical chemistry (2018), 90 (3), 2095-2102 ISSN:.
  A key limiting step for high-throughput NMR-based metabolomics is the lack of rapid and accurate tools for absolute quantification of many metabolites. We developed, implemented, and evaluated an algorithm, AQuA (Automated Quantification Algorithm), for targeted metabolite quantification from complex (1)H NMR spectra. AQuA operates based on spectral data extracted from a library consisting of one standard calibration spectrum for each metabolite. It uses one preselected NMR signal per metabolite for determining absolute concentrations and does so by effectively accounting for interferences caused by other metabolites. AQuA was implemented and evaluated using experimental NMR spectra from human plasma. The accuracy of AQuA was tested and confirmed in comparison with a manual spectral fitting approach using the ChenomX software, in which 61 out of 67 metabolites quantified in 30 human plasma spectra showed a goodness-of-fit (r(2)) close to or exceeding 0.9 between the two approaches. In addition, three quality indicators generated by AQuA, namely, occurrence, interference, and positional deviation, were studied. These quality indicators permit evaluation of the results each time the algorithm is operated. The efficiency was tested and confirmed by implementing AQuA for quantification of 67 metabolites in a large data set comprising 1342 experimental spectra from human plasma, in which the whole computation took less than 1 s.
10. 10
  Cañueto, D.; Gómez, J.; Salek, R. M.; Correig, X.; Cañellas, N. rDolphin: A GUI R Package for Proficient Automatic Profiling of 1D 1H-NMR Spectra of Study Datasets. Metabolomics 2018, 14 (3), 24, DOI: 10.1007/s11306-018-1319-y
  10
  rDolphin: a GUI R package for proficient automatic profiling of 1D (1)H-NMR spectra of study datasets
  Canueto Daniel; Correig Xavier; Canellas Nicolau; Gomez Josep; Salek Reza M; Correig Xavier; Canellas Nicolau
  Metabolomics : Official journal of the Metabolomic Society (2018), 14 (3), 24 ISSN:.
  INTRODUCTION: Adoption of automatic profiling tools for (1)H-NMR-based metabolomic studies still lags behind other approaches in the absence of the flexibility and interactivity necessary to adapt to the properties of study data sets of complex matrices. OBJECTIVES: To provide an open source tool that fully integrates these needs and enables the reproducibility of the profiling process. METHODS: rDolphin incorporates novel techniques to optimize exploratory analysis, metabolite identification, and validation of profiling output quality. RESULTS: The information and quality achieved in two public datasets of complex matrices are maximized. CONCLUSION: rDolphin is an open-source R package ( http://github.com/danielcanueto/rDolphin ) able to provide the best balance between accuracy, reproducibility and ease of use.
11. 11
  Wishart, D. S.; Feunang, Y. D.; Marcu, A.; Guo, A. C.; Liang, K.; Vázquez-Fresno, R.; Sajed, T.; Johnson, D.; Li, C.; Karu, N. HMDB 4.0: The Human Metabolome Database for 2018. Nucleic Acids Res. 2018, 46 (D1), D608– D617, DOI: 10.1093/nar/gkx1089
  11
  HMDB 4.0: the human metabolome database for 2018
  Wishart, David S.; Feunang, Yannick Djoumbou; Marcu, Ana; Guo, An Chi; Liang, Kevin; Vazquez-Fresno, Rosa; Sajed, Tanvir; Johnson, Daniel; Li, Carin; Karu, Naama; Sayeeda, Zinat; Lo, Elvis; Assempour, Nazanin; Berjanskii, Mark; Singhal, Sandeep; Arndt, David; Liang, Yonjie; Badran, Hasan; Grant, Jason; Serra-Cayuela, Arnau; Liu, Yifeng; Mandal, Rupa; Neveu, Vanessa; Pon, Allison; Knox, Craig; Wilson, Michael; Manach, Claudine; Scalbert, Augustin
  Nucleic Acids Research (2018), 46 (D1), D608-D617CODEN: NARHAD; ISSN:1362-4962. (Oxford University Press)
  A review. The Human Metabolome Database or HMDB (www. hmdb.ca) is a web-enabled metabolomic database contg. comprehensive information about human metabolites along with their biol. roles, physiol. concns., disease assocns., chem. reactions, metabolic pathways, and ref. spectra. First described in 2007, the HMDB is now considered the std. metabolomic resource for human metabolic studies. Over the past decade the HMDB has continued to grow and evolve in response to emerging needs for metabolomics researchers and continuing changes in web stds. This year's update, HMDB 4.0, represents the most significant upgrade to the database in its history. For instance, the no. of fully annotated metabolites has increased by nearly threefold, the no. of exptl. spectra has grown by almost fourfold and the no. of illustrated metabolic pathways has grown by a factor of almost 60. Significant improvements have also been made to the HMDB's chem. taxonomy, chem. ontol., spectral viewing, and spectral/text searching tools. A great deal of brand new data has also been added to HMDB 4.0. This includes large quantities of predicted MS/MS and GC-MS ref. spectral data as well as predicted (physiol. feasible) metabolite structures to facilitate novel metabolite identification. Addnl. information on metabolite-SNP interactions and the influence of drugs on metabolite levels (pharmaco-metabolomics) has also been added. Many other important improvements in the content, the interface, and the performance of the HMDB website have been made and these should greatly enhance its ease of use and its potential applications in nutrition, biochem., clin. chem., clin. genetics, medicine, and metabolomics science.
12. 12
  Bouatra, S.; Aziat, F.; Mandal, R.; Guo, A. C.; Wilson, M. R.; Knox, C.; Bjorndahl, T. C.; Krishnamurthy, R.; Saleem, F.; Liu, P. The Human Urine Metabolome. PLoS One 2013, 8 (9), e73076 DOI: 10.1371/journal.pone.0073076
  12
  The human urine metabolome
  Bouatra, Souhaila; Aziat, Farid; Mandal, Rupasri; Guo, An Chi; Wilson, Michael R.; Knox, Craig; Bjorndahl, Trent C.; Krishnamurthy, Ramanarayan; Saleem, Fozia; Liu, Philip; Dame, Zerihun T.; Poelzer, Jenna; Huynh, Jessica; Yallou, Faizath S.; Psychogios, Nick; Dong, Edison; Bogumil, Ralf; Roehring, Cornelia; Wishart, David S.
  PLoS One (2013), 8 (9), e73076CODEN: POLNCL; ISSN:1932-6203. (Public Library of Science)
  A review. Urine has long been a "favored" biofluid among metabolomics researchers. It is sterile, easy-to-obtain in large vols., largely free from interfering proteins or lipids and chem. complex. However, this chem. complexity has also made urine a particularly difficult substrate to fully understand. As a biol. waste material, urine typically contains metabolic breakdown products from a wide range of foods, drinks, drugs, environmental contaminants, endogenous waste metabolites and bacterial byproducts. Many of these compds. are poorly characterized and poorly understood. In an effort to improve our understanding of this biofluid we have undertaken a comprehensive, quant., metabolome-wide characterization of human urine. This involved both computer-aided literature mining and comprehensive, quant. exptl. assessment/validation. The exptl. portion employed NMR spectroscopy, gas chromatog. mass spectrometry (GC-MS), direct flow injection mass spectrometry (DFI/LC-MS/MS), inductively coupled plasma mass spectrometry (ICP-MS) and high performance liq. chromatog. (HPLC) expts. performed on multiple human urine samples. This multi-platform metabolomic anal. allowed us to identify 445 and quantify 378 unique urine metabolites or metabolite species. The different anal. platforms were able to identify (quantify) a total of: 209 (209) by NMR, 179 (85) by GC-MS, 127 (127) by DFI/LC-MS/MS, 40 (40) by ICP-MS and 10 (10) by HPLC. Our use of multiple metabolomics platforms and technologies allowed us to identify several previously unknown urine metabolites and to substantially enhance the level of metabolome coverage. It also allowed us to critically assess the relative strengths and weaknesses of different platforms or technologies. The literature review led to the identification and annotation of another 2206 urinary compds. and was used to help guide the subsequent exptl. studies. An online database contg. the complete set of 2651 confirmed human urine metabolite species, their structures (3079 in total), concns., related literature refs. and links to their known disease assocns. are freely available at online.
13. 13
  Psychogios, N.; Hau, D. D.; Peng, J.; Guo, A. C.; Mandal, R.; Bouatra, S.; Sinelnikov, I.; Krishnamurthy, R.; Eisner, R.; Gautam, B. The Human Serum Metabolome. PLoS One 2011, 6 (2), e16957 DOI: 10.1371/journal.pone.0016957
  13
  The human serum metabolome
  Psychogios, Nikolaos; Hau, David D.; Peng, Jun; Guo, An. Chi; Mandal, Rupasri; Bouatra, Souhaila; Sinelnikov, Igor; Krishnamurthy, Ramanarayan; Eisner, Roman; Gautam, Bijaya; Young, Nelson; Xia, Jianguo; Knox, Craig; Dong, Edison; Huang, Paul; Hollander, Zsuzsanna; Pedersen, Theresa L.; Smith, Steven R.; Bamforth, Fiona; Greiner, Russ; McManus, Bruce; Newman, John W.; Goodfriend, Theodore; Wishart, David S.
  PLoS One (2011), 6 (2), e16957CODEN: POLNCL; ISSN:1932-6203. (Public Library of Science)
  Continuing improvements in anal. technol. along with an increased interest in performing comprehensive, quant. metabolic profiling, is leading to increased interest pressures within the metabolomics community to develop centralized metabolite ref. resources for certain clin. important biofluids, such as cerebrospinal fluid, urine and blood. As part of an ongoing effort to systematically characterize the human metabolome through the Human Metabolome Project, we have undertaken the task of characterizing the human serum metabolome. In doing so, we have combined targeted and non-targeted NMR, GC-MS and LC-MS methods with computer-aided literature mining to identify and quantify a comprehensive, if not absolutely complete, set of metabolites commonly detected and quantified (with today's technol.) in the human serum metabolome. Our use of multiple metabolomics platforms and technologies allowed us to substantially enhance the level of metabolome coverage while critically assessing the relative strengths and weaknesses of these platforms or technologies. Tables contg. the complete set of 4229 confirmed and highly probable human serum compds., their concns., related literature refs. and links to their known disease assocns. are freely available online.
14. 14
  Nagana Gowda, G. A.; Gowda, Y. N.; Raftery, D. Expanding the Limits of Human Blood Metabolite Quantitation Using NMR Spectroscopy. Anal. Chem. 2015, 87 (1), 706– 715, DOI: 10.1021/ac503651e
  14
  Expanding the Limits of Human Blood Metabolite Quantitation Using NMR Spectroscopy
  Nagana Gowda, G. A.; Gowda, Yashas N.; Raftery, Daniel
  Analytical Chemistry (Washington, DC, United States) (2015), 87 (1), 706-715CODEN: ANCHAM; ISSN:0003-2700. (American Chemical Society)
  A current challenge in metabolomics is the reliable quantitation of many metabolites. Limited resoln. and sensitivity combined with the challenges assocd. with unknown metabolite identification have restricted both the no. and the quant. accuracy of blood metabolites. Focused on alleviating this bottleneck in NMR-based metabolomics, studies of pooled human serum combining an array of 1D/2D NMR expts. at 800 MHz, database searches, and spiking with authentic compds. enabled the identification of 67 blood metabolites. Many of these (∼1/3) are new compared with those reported previously as a part of the Human Serum Metabolome Database. In addn., considering both the high reproducibility and quant. nature of NMR as well as the sensitivity of NMR chem. shifts to altered sample conditions, exptl. protocols and comprehensive peak annotations are provided here as a guide for identification and quantitation of the new pool of blood metabolites for routine applications. Further, studies focused on the evaluation of quantitation using org. solvents revealed a surprisingly poor performance for protein pptn. using acetonitrile. One-third of the detected metabolites were attenuated by 10-67% compared with methanol pptn. at the same solvent-to-serum ratio of 2:1 (vol./vol.). Nearly 2/3 of the metabolites were further attenuated by up to 65% upon increasing the acetonitrile-to-serum ratio to 4:1 (vol./vol.). These results, combined with the newly established identity for many unknown metabolites in the NMR spectrum, offer new avenues for human serum/plasma-based metabolomics. Further, the ability to quant. evaluate nearly 70 blood metabolites that represent numerous classes, including amino acids, org. acids, carbohydrates, and heterocyclic compds., using a simple and highly reproducible anal. method such as NMR may potentially guide the evaluation of samples for anal. using mass spectrometry.
15. 15
  Rueedi, R.; Ledda, M.; Nicholls, A. W.; Salek, R. M.; Marques-Vidal, P.; Morya, E.; Sameshima, K.; Montoliu, I.; Da Silva, L.; Collino, S. Genome-Wide Association Study of Metabolic Traits Reveals Novel Gene-Metabolite-Disease Links. PLoS Genet. 2014, 10 (2), e1004132 DOI: 10.1371/journal.pgen.1004132
  15
  Genome-wide association study of metabolic traits reveals novel gene-metabolite-disease links
  Rueedi, Rico; Ledda, Mirko; Nicholls, Andrew W.; Salek, Reza M.; Marques-Vidal, Pedro; Morya, Edgard; Sameshima, Koichi; Montoliu, Ivan; Da Silva, Laeticia; Collino, Sebastiano; Martin, Francois-Pierre; Rezzi, Serge; Steinbeck, Christoph; Waterworth, Dawn M.; Waeber, Gerard; Vollenweider, Peter; Beckmann, Jacques S.; Le Coutre, Johannes; Mooser, Vincent; Bergmann, Sven; Genick, Ulrich K.; Kutalik, Zoltan
  PLoS Genetics (2014), 10 (2), e1004132/1-e1004132/10, 10 pp.CODEN: PGLEB5; ISSN:1553-7404. (Public Library of Science)
  Metabolic traits are mol. phenotypes that can drive clin. phenotypes and may predict disease progression. Here, we report results from a metabolome- and genome-wide assocn. study on 1H-NMR urine metabolic profiles. The study was conducted within an untargeted approach, employing a novel method for compd. identification. From our discovery cohort of 835 Caucasian individuals who participated in the CoLaus study, we identified 139 suggestively significant (P<5 × 10-8) and independent assocns. between single nucleotide polymorphisms (SNP) and metabolome features. Fifty-six of these assocns. replicated in the TasteSensomics cohort, comprising 601 individuals from S~ao Paulo of vastly diverse ethnic background. They correspond to eleven gene-metabolite assocns., six of which had been previously identified in the urine metabolome and three in the serum metabolome. Our key novel findings are the assocns. of two SNPs with NMR spectral signatures pointing to fucose (rs492602, P = 6.9 × 10-44) and lysine (rs8101881, P = 1.2 × 10-33), resp. Fine-mapping of the first locus pinpointed the FUT2 gene, which encodes a fucosyltransferase enzyme and has previously been assocd. with Crohn's disease. This implicates fucose as a potential prognostic disease marker, for which there is already published evidence from a mouse model. The second SNP lies within the SLC7A9 gene, rare mutations of which have been linked to severe kidney damage. The replication of previous assocns. and our new discoveries demonstrate the potential of untargeted metabolomics GWAS to robustly identify mol. disease markers.
16. 16
  Rueedi, R.; Mallol, R.; Raffler, J.; Lamparter, D.; Friedrich, N.; Vollenweider, P.; Waeber, G.; Kastenmüller, G.; Kutalik, Z.; Bergmann, S. Metabomatching: Using Genetic Association to Identify Metabolites in Proton NMR Spectroscopy. PLoS Comput. Biol. 2017, 13 (12), e1005839 DOI: 10.1371/journal.pcbi.1005839
  16
  Metabomatching: using genetic association to identify metabolites in proton NMR spectroscopy
  Rueedi, Rico; Mallol, Roger; Raffler, Johannes; Lamparter, David; Friedrich, Nele; Vollenweider, Peter; Waeber, Gerard; Kastenmuller, Gabi; Kutalik, Zoltan; Bergmann, Sven
  PLoS Computational Biology (2017), 13 (12), e1005839/1-e1005839/17CODEN: PCBLBG; ISSN:1553-7358. (Public Library of Science)
  A metabolome-wide genome-wide assocn. study aims to discover the effects of genetic variants on metabolome phenotypes. Most mGWASes use as phenotypes concns. of limited sets of metabolites that can be identified and quantified from spectral information. In contrast, in an untargeted mGWAS both identification and quantification are forgone and, instead, all measured metabolome features are tested for assocn. with genetic variants. While the untargeted approach does not discard data that may have eluded identification, the interpretation of assocd. features remains a challenge. To address this issue, we developed metabomatching to identify the metabolites underlying significant assocns. obsd. in untargeted mGWASes on proton NMR metabolome data. Metabomatching capitalizes on genetic spiking, the concept that because metabolome features assocd. with a genetic variant tend to correspond to the peaks of the NMR spectrum of the underlying metabolite, genetic assocn. can allow for identification. Applied to the untargeted mGWASes in the SHIP and CoLaus cohorts and using 180 ref. NMR spectra of the urine metabolome database, metabomatching successfully identified the underlying metabolite in 14 of 19, and 8 of 9 assocns., resp. The accuracy and efficiency of our method make it a strong contender for facilitating or complementing metabolomics analyses in large cohorts, where the availability of genetic, or other data, enables our approach, but targeted quantification is limited.
17. 17
  Raffler, J.; Friedrich, N.; Arnold, M.; Kacprowski, T.; Rueedi, R.; Altmaier, E.; Bergmann, S.; Budde, K.; Gieger, C.; Homuth, G. Genome-Wide Association Study with Targeted and Non-Targeted NMR Metabolomics Identifies 15 Novel Loci of Urinary Human Metabolic Individuality. PLoS Genet. 2015, 11 (9), e1005487 DOI: 10.1371/journal.pgen.1005487
  17
  Genome-wide association study with targeted and non-targeted nmr metabolomics identifies 15 novel loci of urinary human metabolic individuality
  Raffler, Johannes; Friedrich, Nele; Arnold, Matthias; Kacprowski, Tim; Rueedi, Rico; Altmaier, Elisabeth; Bergmann, Sven; Budde, Kathrin; Gieger, Christian; Homuth, Georg; Pietzner, Maik; Roemisch-Margl, Werner; Strauch, Konstantin; Voelzke, Henry; Waldenberger, Melanie; Wallaschofski, Henri; Nauck, Matthias; Voelker, Uwe; Kastenmueller, Gabi; Suhre, Karsten
  PLoS Genetics (2015), 11 (9), e1005487/1-e1005487/28CODEN: PGLEB5; ISSN:1553-7404. (Public Library of Science)
  Genome-wide assocn. studies with metabolic traits (mGWAS) uncovered many genetic variants that influence human metab. These genetically influenced metabotypes (GIMs) contribute to our metabolic individuality, our capacity to respond to environmental challenges, and our susceptibility to specific diseases. While metabolic homeostasis in blood is a well investigated topic in large mGWAS with over 150 known loci, metabolic detoxification through urinary excretion has only been addressed by few small mGWAS with only 11 assocd. loci so far. Here we report the largest mGWAS to date, combining targeted and non-targeted 1H NMR anal. of urine samples from 3,861 participants of the SHIP-0 cohort and 1,691 subjects of the KORA F4 cohort. We identified and replicated 22 loci with significant assocns. with urinary traits, 15 of which are new (HIBCH, CPS1, AGXT, XYLB, TKT, ETNPPL, SLC6A19, DMGDH, SLC36A2, GLDC, SLC6A13, ACSM3, SLC5A11, PNMT, SLC13A3). Two-thirds of the urinary loci also have a metabolite assocn. in blood. For all but one of the 6 loci where significant assocns. target the same metabolite in blood and urine, the genetic effects have the same direction in both fluids. In contrast, for the SLC5A11 locus, we found increased levels of myo-inositol in urine whereas mGWAS in blood reported decreased levels for the same genetic variant. This might indicate less effective re-absorption of myo-inositol in the kidneys of carriers. In summary, our study more than doubles the no. of known loci that influence urinary phenotypes. It thus allows novel insights into the relationship between blood homeostasis and its regulation through excretion. The newly discovered loci also include variants previously linked to chronic kidney disease (CPS1, SLC6A13), pulmonary hypertension (CPS1), and ischemic stroke (XYLB). By establishing connections from gene to disease via metabolic traits our results provide novel hypotheses about mol. mechanisms involved in the etiol. of diseases.
18. 18
  Ihmels, J.; Bergmann, S.; Barkai, N. Defining Transcription Modules Using Large-Scale Gene Expression Data. Bioinformatics 2004, 20 (13), 1993– 2003, DOI: 10.1093/bioinformatics/bth166
  18
  Defining transcription modules using large-scale gene expression data
  Ihmels, Jan; Bergmann, Sven; Barkai, Naama
  Bioinformatics (2004), 20 (13), 1993-2003CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Large-scale gene expression data comprising a variety of cellular conditions hold the promise of a global view on the transcription program. While conventional clustering algorithms have been successfully applied to smaller datasets, the utility of many algorithms for the anal. of large-scale data is limited by their inability to capture combinatorial and condition-specific co-regulation. In addn., there is an increasing need to integrate the rapidly accumulating body of other high-throughput biol. data with the expression anal. In a previous work, the signature algorithm was introduced, which overcomes the problems of conventional clustering and allows for intuitive integration of addnl. biol. data. However, this approach is constrained by the comprehensiveness of relevant external data and its lacking ability to capture hierarchical modularity. A novel method has been presented for the anal. of large-scale expression data, which assigns genes into context-dependent and potentially overlapping regulatory units. Authors introduce the notion of a transcription module as a self-consistent regulatory unit consisting of a set of co-regulated genes as well as the exptl. conditions that induce their co-regulation. Self-consistency is defined by a rigorous math. criterion. Authors propose an efficient algorithm to identify such modules, which is based on the iterative application of the signature algorithm. A threshold parameter that dets. the resoln. of the modular decompn. is introduced. The method is applied systematically to over 1000 expression profiles of the yeast Saccharomyces cerevisiae, and the results are presented using two complementary visualization schemes developed. The av. biol. coherence, as measured by the conservation of putative cis-regulatory motifs between four related yeast species, is higher for transcription modules than for clusters identified by other methods applied to the same dataset. This method is related to singular value decompn. (SVD) and to the pairwise av. linkage clustering algorithm. It extends SVD by filtering out noise in the expression data and offering variable resoln. to reveal hierarchical organization. It furthermore has the advantage over both methods of capturing overlapping modules in the presence of combinatorial regulation.
19. 19
  Bergmann, S.; Ihmels, J.; Barkai, N. Iterative Signature Algorithm for the Analysis of Large-Scale Gene Expression Data. Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top. 2003, 67 (3), 031902 DOI: 10.1103/PhysRevE.67.031902
  There is no corresponding record for this reference.
20. 20
  Xiong, X.; Liu, D.; Wang, Y.; Zeng, T.; Peng, Y. Urinary 3-(3-Hydroxyphenyl)-3-Hydroxypropionic Acid, 3-Hydroxyphenylacetic Acid, and 3-Hydroxyhippuric Acid Are Elevated in Children with Autism Spectrum Disorders. BioMed Res. Int. 2016, 2016, 9485412, DOI: 10.1155/2016/9485412
  20
  Urinary 3-(3-Hydroxyphenyl)-3-hydroxypropionic Acid, 3-Hydroxyphenylacetic Acid, and 3-Hydroxyhippuric Acid Are Elevated in Children with Autism Spectrum Disorders
  Xiong Xiyue; Liu Dan; Wang Yichao; Zeng Ting; Peng Ying
  BioMed research international (2016), 2016 (), 9485412 ISSN:.
  Autism spectrum disorders (ASDs) are a group of mental illnesses highly correlated with gut microbiota. Recent studies have shown that some abnormal aromatic metabolites in autism patients are presumably derived from overgrown Clostridium species in gut, which may be used for diagnostic purposes. In this paper, a GC/MS based metabolomic approach was utilized to seek similar biomarkers by analyzing the urinary information in 62 ASDs patients compared with 62 non-ASDs controls in China, aged 1.5-7. Three compounds identified as 3-(3-hydroxyphenyl)-3-hydroxypropionic acid (HPHPA), 3-hydroxyphenylacetic acid (3HPA), and 3-hydroxyhippuric acid (3HHA) were found in higher concentrations in autistic children than in the controls (p < 0.001). After oral vancomycin treatment, urinary excretion of HPHPA (p < 0.001), 3HPA (p < 0.005), and 3HHA (p < 0.001) decreased markedly, which indicated that these compounds may also be from gut Clostridium species. The sensitivity and specificity of HPHPA, 3HPA, and 3HHA were evaluated by receiver-operating characteristic (ROC) analysis. The specificity of each compound for ASDs was very high (>96%). After two-regression analysis, the optimal area under the curve (AUC, 0.962), sensitivity (90.3%), and specificity (98.4%) were obtained by ROC curve of Prediction probability based on the three metabolites. These findings demonstrate that the measurements of the three compounds are strong predictors of ASDs and support the potential clinical utility for identifying a subgroup of ASDs subjects.
21. 21
  Nielsen, H. R.; Killmann, S. A. Urinary Excretion of Beta-Aminoisobutyrate and Pseudouridine in Acute and Chronic Myeloid Leukemia. J. Natl. Cancer Inst. 1983, 71 (5), 887– 891
  21
  Urinary excretion of β-aminoisobutyrate and pseudouridine in acute and chronic myeloid leukemia
  Nielsen, Henrik R.; Killmann, Sven Aage
  JNCI, Journal of the National Cancer Institute (1983), 71 (5), 887-91CODEN: JJIND8; ISSN:0198-0157.
  β-Aminoisobutyrate (β-AIB) and pseudouridine (ψ-Urd) urinary excretion was investigated in abnormal hematopoietic conditions including 26 patients with acute myeloid leukemia (AML) and chronic myeloid leukemia (CML) and was compared to that of 25 healthy controls. β-AIB excretion in CML was directly correlated to the leukocyte count, the indicator of tumor cell mass. β-AIB excretion was elevated in 27 and 75% of untreated AML and CML cases, resp. Marrow blast cell content tended to correlate pos. with ψ-Urd excretion in AML. ψ-Urd excretion was elevated in 82 and 87% of untreated AML and CML, resp. Turnover of hematopoietic cells seemed to be a determinant for β-AIB excretion, including higher cell turnover in CML patients compared to that in AML patients and in controls. With cytostatic treatment, excretion levels of β-AIB and/or ψ-Urd decreased after a transient rise.
22. 22
  Ziegler, E. E., Ed.; Present Knowledge in Nutrition; Filer, L. J. J., Engl. Ed.; International Life Sciences Inst.: Washington, D.C., 1996.
  There is no corresponding record for this reference.
23. 23
  Nicholas, P. C.; Kim, D.; Crews, F. T.; Macdonald, J. M. Proton Nuclear Magnetic Resonance Spectroscopic Determination of Ethanol-Induced Formation of Ethyl Glucuronide in Liver. Anal. Biochem. 2006, 358 (2), 185– 191, DOI: 10.1016/j.ab.2006.08.033
  23
  Proton nuclear magnetic resonance spectroscopic determination of ethanol-induced formation of ethyl glucuronide in liver
  Nicholas, Peter C.; Kim, Daniel; Crews, Fulton T.; Macdonald, Jeffrey M.
  Analytical Biochemistry (2006), 358 (2), 185-191CODEN: ANBCA2; ISSN:0003-2697. (Elsevier)
  Et glucuronide (ethyl-β-D-6-glucosiduronic acid, EtG), a unique metabolite of EtOH, has received much recent attention as a sensitive and specific biol. marker of EtOH consumption. Formed in the liver via conjugation of EtOH with activated glucuronate, EtG remains detectable in serum, plasma, and hair for days after EtOH abuse. Thus far, gas chromatog.-mass spectrometry and enzyme-linked immunosorbent assays were developed to detect trace quantities of EtG for forensic purposes, but reports of the NMR properties of EtG were scarce. Herein the authors present the first report of EtG detn. using proton NMR spectroscopy. The authors collected 700-MHz proton spectra of liver exts. from rats treated with a 4-day binge EtOH protocol (av. EtOH dose: 8.6 g/kg/day). An unexpected signal (triplet, 1.24 ppm) appeared in EtOH-treated liver exts. but not in control samples; based on chem. shift and multiplicity, the authors suspected EtG. The authors obsd. quant. hydrolysis of the unknown species to EtOH while incubating the samples with β-glucuronidase, confirming that the Me protons of EtG were responsible for the triplet at 1.24 ppm. This study demonstrates that proton NMR spectroscopy is capable of detecting EtG and that future NMR-based metabolomic studies may encounter this metabolite of EtOH.
24. 24
  Kim, S.; Lee, M.; Yoon, D.; Lee, D.-K.; Choi, H.-J.; Kim, S. 1D Proton NMR Spectroscopic Determination of Ethanol and Ethyl Glucuronide in Human Urine. Bull. Korean Chem. Soc. 2013, 34 (8), 2413– 2418, DOI: 10.5012/bkcs.2013.34.8.2413
  24
  1D proton NMR spectroscopic determination in ethanol and ethyl glucuronide in human urine
  Kim, Siwon; Lee, Minji; Yoon, Dahye; Lee, Dong-kye; Choi, Hye-Jin; Kim, Suhkmann
  Bulletin of the Korean Chemical Society (2013), 34 (8), 2413-2418CODEN: BKCSDE; ISSN:0253-2964. (Korean Chemical Society)
  Forensic and legal medicine require reliable data to indicate excessive alc. consumption. Ethanol is oxidatively metabolized to acetate by alc. dehydrogenase and non-oxidatively metabolized to Et glucuronide (EtG), Et sulfate (EtS), phosphatidylethanol, or fatty acid Et esters (FAEE). Oxidative metab. is too rapid to provide biomarkers for the detection of ethanol ingestion. However, the nonoxidative metabolite EtG is a useful biomarker because it is stable, non-volatile, water sol., highly sensitive, and is detected in body fluid, hair, and tissues. EtG anal. methods such as mass spectroscopy, chromatog., or ELISA techniques are currently in use. We suggest that NMR (NMR) spectroscopy could be used to monitor ethanol intake. As with current conventional methods, NMR spectroscopy doesn't require complicated pretreatments or sample sepn. This method has the advantages of short acquisition time, simple sample prepn., reproducibility, and accuracy. In addn., all proton-contg. compds. can be detected. In this study, we performed 1H NMR analyses of urine to monitor the ethanol and EtG. Urinary samples were collected over time from 5 male volunteers. We confirmed that ethanol and EtG signals could be detected with NMR spectroscopy. Ethanol signals increased immediately upon alc. intake, but decreased sharply over time. In contrast, EtG signal increased and reached a max. about 9 h later, after which the EtG signal decreased gradually and remained detectable after 20-25 h. Based on these results, we suggest that 1H NMR spectroscopy may be used to identify ethanol non-oxidative metabolites without the need for sample pretreatment.
25. 25
  Solomons, H. D. Carbohydrate Deficient Transferrin and Alcoholism. GERMS 2012, 2 (2), 75– 78, DOI: 10.11599/germs.2012.1015
  25
  Carbohydrate deficient transferrin and alcoholism
  Solomons, Hilary Denis
  GERMS (2012), 2 (2), 75-78CODEN: GERMCF; ISSN:2248-2997. (European Academy of HIV/AIDS and Infectious Diseases)
  A review. Alc. abuse is an important public health problem, with major implications in patients with a preexisting liver pathol. of viral origin. Hepatitis C, for example, is one of the diseases in which alc. consumption can lead to the transition from a fairly benign outline to a potentially life-threatening liver disease. Alc. abuse is usually identified on the basis of clin. judgment, alcoholism related questionnaires, lab. tests and, more recently, biomarkers. Also on this list of tests, carbohydrate deficient transferrin (CDT) is widely available and useful for detg. recent alc. consumption, particularly when corroborated with elevation of other liver-assocd. enrymes. Clinicians should be aware of the indications and limitations of this test in order to better evaluate alc. consumption in their patients.
Supporting Information
Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jproteome.9b00295.
- Supporting figures and table (PDF)
- pr9b00295_liveslides.mp4 (5.62 MB)
- pr9b00295_si_001.pdf (5.25 MB)
Terms & Conditions

Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.

Automated Analysis of Large-Scale NMR Data Generates Metabolomic Signatures and Links Them to Candidate MetabolitesClick to copy article linkArticle link copied!

Journal of Proteome Research

Abstract

Introduction

Methods

Preprocessing

Confounding

Metabomatching

ACP: Averaged Correlation Profile

ISA: Iterative Signature Algorithm

PCA: Principal Component Analysis

Identification of Metabolites

Pseudoquantification of Metabolite Concentrations

Analysis and Results

Figure 1

Correlation-Based Pseudospectra

Iterative Signature Algorithm

Principal Component Analysis

Many Pseudospectra Defined by the ACP Method and ISA Match to Urine Metabolites

Figure 2

Metabolite Concentration Pseudoquantification with NMR Features of Matched Pseudospectra

Figure 3

Conclusions and Discussion

Supporting Information

Terms & Conditions

Author Information

Acknowledgments

References

Cited By

Journal of Proteome Research

Article Views

Altmetric

Citations

Recommended Articles

Abstract

Figure 1

Figure 2

Figure 3

References

Supporting Information

Supporting Information

Terms & Conditions

Automated Analysis of Large-Scale NMR Data Generates Metabolomic Signatures and Links Them to Candidate Metabolites
Click to copy article linkArticle link copied!