Using nuclear magnetic resonance (NMR) spectroscopy in the study of metabolism has been immensely popular in medical- and health-related research but has yet to be widely applied to more fundamental biological problems. This review provides some NMR background relevant to metabolism, describes why 1H NMR spectra are complex as well as introducing relevant terminology and definitions. The applications and practical considerations of NMR metabolic profiling and 13C NMR-based flux analyses are discussed together with the elegant ‘enzyme trap’ approach for identifying novel metabolic pathway intermediates. The importance of sample preparation and data analysis are also described and explained with reference to data precision and multivariate analysis to introduce researchers unfamiliar with NMR and metabolism to consider this technique for their research interests. Finally, a brief glance into the future suggests NMR-based metabolism has room to expand in the 21st century through new isotope labels, and NMR technologies and methodologies.
Nuclear magnetic resonance spectroscopy (NMR) and mass spectrometry have become highly influential technologies over the last 20 years in the study of metabolomics in health and disease. There are many excellent examples and reviews on NMR-based metabolic studies in clinical settings and readers are encouraged to indulge their interests in this area through these associated references [1–5]. However, the perceived complexities associated with experimental procedures, sample preparation and data analysis have provided a barrier to many biochemical and biological laboratories who might otherwise be interested in using this incredibly useful and accessible resource more widely across non-clinical research.
The popularity of studying metabolism has expanded rapidly over recent years to become one of several keystone ‘-omic’ technologies. It should be considered complimentary to genomics, transcriptomics, proteomics and lipidomics where they underpin systems biology and provide a multifaceted interdisciplinary portfolio. As a result, it is imperative to embrace these technologies to maximise impact and remain at the forefront of bioscience research, but it is inefficient for a group leader to be a master of all methods. However, regardless of speciality, researchers need a comfortable level of understanding of these tools to amalgamate techniques efficiently. This article aims to introduce metabolic NMR and encourage bioscientists to consider discussing potential projects with a collaborative NMR spectroscopist.
Nuclear magnetic resonance spectroscopy utilises samples that contain nuclei with non-zero magnetic moments that absorb electromagnetic radiation (r.f) at a precise frequency when placed in an external magnetic field. This is the point where most non-NMR interested scientists switch off but this definition underlines the fundamental utility of the technique with respect to metabolomics. First, local electronic environments in a molecule induce small, but measurable, changes in the precise frequency of r.f absorption; this provides the familiar NMR spectrum of chemical shifts by which different chemical species can be distinguished. Second, the precise frequency means different nuclei can be ‘tuned’ for observation by an NMR spectrometer; this provides the commonly encountered 1H or proton spectrum as well as other nuclei such as 13C, 15N and 31P. Third, the non-zero magnetic moment makes NMR isotope specific; hence the terminology of 1H, 13C, 15N and 31P. NMR is selective regarding isotopes and is only possible when the magnetic moment (sometimes referred to as spin) of these isotopes is non-zero. Furthermore, only spin ½ nuclei provide high-resolution NMR spectra that allow the deciphering of chemical shifts and quantitative detail. As always, caveats exist with these definitions but it should come as no surprise that 1H, 13C, 15N and 31P isotopes all have spins of ½. Spin also explains the complexity of isotopic enrichment; 1H (protium) works well as this primary NMR observable isotope of hydrogen because it is 99.99% abundant, so any molecules containing hydrogen should provide plenty of 1H NMR signal. This is also the case for 31P NMR although phosphorous is more eclectic in its role across metabolism with only a small proportion of metabolites being phosphorylated. However, the 13C isotope of carbon is 1.07% abundant with the bulk of carbon being present as the 12C isotope that has zero spin and is NMR inactive. Nitrogen is even more complex because the common 14N isotope has a spin of 1, but is an excellent example whereby a spin > ½ is ‘bad for detailed NMR’, forcing us to resort to the less common spin ½ 15N isotope that is only 0.36% abundant. Consequently, many biological-based NMR experiments use isotopically enriched 15N; i.e. this isotope is boosted to excess of 90% in a sample. This comes at a cost and a logistical design headache regarding how the 15N isotope is introduced into the experiment or sample. However, the 13C isotope is more forgiving and can still be detected by NMR at natural abundance levels, although isotopic enrichment of 13C is a possible way of improving metabolism NMR data.
The logical starting point is 1H NMR because metabolites and pathway intermediates are hydrogen rich; for example, sucrose has the molecular formula of C12H22O11 and even hydrogen ‘starved’ pyruvate is C3H3O3. As outlined above, local structural differences in each metabolite alter the local electron environment to create 1H NMR peaks that reflect each distinct hydrogen environment. This results in a fingerprint of chemical shifts for each metabolite that can be unique in many cases, as shown in Figure 1A for the amino acid alanine [CH3CH(NH3)+COO−]. The alanine spectrum is acquired such that labile NH3+ protons are not observed and the two peaks are from the CH (left) and CH3 (right) environments. This sounds promising until one considers the large numbers of unique metabolites in a cell, media preparation or biological system; adding these collective ‘fingerprints’ explains the crowded nature of a 1H NMR profile spectrum, as shown in Figure 1B for a cell extract of Saccharomyces cerevisiae. This article focuses on the use of high-resolution solution state NMR to investigate metabolites, like those found in the cell extract in Figure 1, but solid samples such as cells and tissues can be interrogated using high-resolution magic angle spinning (HR-MAS) techniques for clinical and other specific biological applications [3,6–8]. NMR-based medical metabolomics involves the study of complex biological fluids such as urine, plasma and blood, and forms the basis of improving health and diagnosis through metabolic phenotyping with National and International Phenome Centres being established across the globe.
1H NMR spectra of alanine (
A) in water (using a pulse program to reduce the residual signal from H2O) and S. cerevisiae cell extract ( B) in D2O (heavy water) acquired at 25°C on a 14.1 tesla (600 MHz 1H) Bruker AV3 NMR spectrometer with a QCI-F cryoprobe.
As most metabolites have more than one ‘H’ environment, each metabolite spectrum is complex in its own right. Once multiplied by the number and variety of metabolites, it is now much easier to understand why a cell extract spectrum is extremely busy. The subject of crowded spectra leads to the choice of spectrometer. NMR spectrometers are generally referred to by their r.f. carrier frequency for proton (1H) at their particular magnetic field; 600, 700, 800 etc. corresponds to the carrier proton frequency in megahertz (MHz) at magnetic field strengths of 14.1, 16.5 and 18.8 tesla respectively. NMR spectrometer resolution goes hand-in-hand with frequency; therefore, using higher field spectrometers increases peak separation and allows more metabolite peaks to be unambiguously identified. As a guide, a 600 MHz system is a good workhorse but if you have access to 700, 800 or 950 MHz then use them. However, do not be deterred by the complexity of NMR spectroscopy or the data it provides because the spectrum in Figure 1B is a metabolic profile in its own right and deviations from this profile can be scrutinised in two ways; analytically, for specific changes in individual metabolites or holistically, to observe trends and changes as a result of a biological process or event.
As with all research, the choice of experimentation and analysis is project specific and should always begin with an understanding of all questions to be answered. For example, if your interests revolve around a cellular or biological process that may generate a metabolic response, start with a metabolic profiling study using either cell contents or media extract, depending on whether the response is intracellular or extracellular.
Profiling is extremely powerful because it doesn't require any prior knowledge of any potential metabolic changes during the initial study, although if suspicions exist, specific metabolites can be targeted for analysis. As it is important to assess whether significant metabolic changes occur between two or more conditions or processes, and that these changes may relate to controllable genetic, proteomic or lipidomic differences, this can provide a systems biology link to your study. In the first instance, consider two conditions of interest and observe the NMR metabolic profiles for any changes. Examples include observing profile differences between wild-type and mutant strains, growth conditions or even media types. This is illustrated in Figure 2, which shows 1H NMR data from two cell extracts of identical cells grown in different media. The 1H NMR spectrum highlights a similar profile for both extracts but cellular histidine levels are perturbed between the two samples. This demonstrates the power and detail available from NMR-based metabolomics despite the complexity of the spectra obtained.
Expanded regions of the 1H NMR spectra of cell extracts from a single yeast cell line grown in cell culture media recipes (
A and B).
Another powerful application of 1H NMR metabolic profiling is in bioprocessing. An approach called Fermentanomics , which studies mammalian cell culture media using NMR, has been shown to optimise amino acid supplementation in order to achieve optimal antibody production . Another profiling study was used to monitor both cell extracts and media during cold-shock and recovery of adherent (CHOK1) and suspension (CHOS) Chinese hamster ovary cell lines to gather information on metabolic load and production .
In all metabolic profile studies, it is crucial to identify ‘significant’ metabolic changes when observing profiles; careful choice of repeat experiments and controls are necessary to classify differences as significant. To maximise research impact, many studies scrutinise the spectral profiles and describe specific metabolite changes in a process that requires assignment of specific NMR peaks to individual metabolites. As a result, many studies use explicit identification of metabolites through additional NMR methods; typically, 2D analytical approaches that separate peak overlap and aid identification. Alternatively, metabolites can be identified using online databases or software and an overview of such tools was published by Ellinger and co-workers in 2013 .
1H NMR profile spectra can be cross-referenced with NMR spectra obtained from separated metabolites or even by spiking metabolites into the extract. For separated metabolites, it is best to follow an established protocol such as that published for the Birmingham Metabolite Library . Metabolic profiling also has the advantage of being able to measure the concentration of each metabolite by adding a chemical standard that is present in a known concentration. This is normally achieved by adding the sodium salt of either 3-(trimethylsilyl)propionic acid (TSP) or 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS) at a concentration between 0.01–0.05 mM. Both TSP and DSS provide a strong, isolated peak at 0 ppm corresponding to nine protons; TSP is labelled in Figure 1. TSP and DSS enable quantitation because metabolite peaks in standard 1D 1H NMR spectra have peak areas proportional to their concentrations in the majority of cases. However, quantitative NMR requires the spectrometer to be set-up correctly and particular attention must be paid to water suppression, data points/acquisition time and relaxation delay. Water suppression is a particular issue because pure water contains molecules at a concentration of 55 M at 25°C and even in a buffer, millimolar to micromolar concentrations of metabolites are dwarfed by the 1H water signal. NMR water suppression is a complex methodology beyond the scope of this review but several very efficient approaches exist that are reviewed elsewhere . In our experience, water suppression is best achieved using presaturation when metabolite signals exist in close spectral proximity to the water 1H NMR resonance, but labile protons are also ‘at risk’ of attenuation using this technique. Other suppression methods also exist, such as WaterGATE and excitation sculpting, but their binomial approach to water suppression can attenuate metabolite signals close to the water resonance, thus obscuring potentially important species from the subsequent analysis.
As with all analytical techniques, sample preparation is important, and solutions of metabolites have to be extracted or purified with care to prevent the unwanted introduction of false-positive information. Samples often need to be treated in such a way as to prevent further metabolic processes that would otherwise interfere with the subsequent analyses, be it through the addition of antimicrobial agents and/or snap freezing and storage at -80°C until immediately prior to data acquisition . It is important that metabolic differences observed by NMR are due to biological processes and not, for example, due to differences in cell number, cell volume or quantity of harvested growth media. In adiition, be aware there are other variables including the method of metabolite extraction and the need to standardise pH. Many approaches use organic extractions that have been shown to be robust and convenient, whereby the organic extraction solvent is removed under vacuum and the sample resuspended in H2O or D2O prior to NMR analysis, although additional freeze-drying processes may be used. Lin and collaborators published a study evaluating metabolite extractions and concluded that methanol/chloroform/water extraction was the preferred method . Should the sample be contaminated with large molecules (e.g. proteins or lipids) NMR methods can be employed to reduce or remove peaks from such molecules by spectral editing of their fast transverse relaxation rates. Such methods utilise a ‘spin-lock’ or Carr–Purcell–Meiboom–Gill (CPMG) pulse train within the experiment and can be set-up by your collaborating spectroscopist.
Pathway monitoring and metabolic fate
Where a metabolic pathway or specific metabolite is the focus of a study, NMR-based metabolic profiling can still be utilised but the resultant spectrum is interrogated for the identification and quantitation of specific resonances belonging to the metabolite(s) of interest. Spectra automatically include ‘control’ metabolites across the complex profile to support reproducibility, but these peaks also create overlap and reduce the ability to identify and quantify metabolite peaks in highly populated regions of the spectrum. This ‘dynamic range’ issue with 1H NMR metabolic profiling can be circumvented by using 2D NMR methods such as J-resolved (J-res), total-correlation spectroscopy (TOCSY) or 13C,1H-heteronuclear single quantum correlation (HSQC) experiments to reduce signal overlap. However, quantifying metabolites from 2D NMR spectra provides new challenges, as peak intensities in many 2D NMR experiments are influenced by spin-spin (J) coupling as well as concentration. The process of monitoring specific metabolites is considered a ‘targeted’ approach and has been successfully applied across medical metabolomics applications such as the deliberate spiking of blood plasma with 32 specific metabolites to provide phenotyping in breast cancer . However, this should not be confused with a methodological ‘targeted approach’, presented by Weljie and co-workers , where individual NMR resonances are mathematically modelled from pure compound spectra. This method forms the basis of operation within the software package Chenomx and is commonly applied to metabolic analysis such as the recent study of the mechanisms of action of antibiotics through NMR metabolomics . This targeted approach is invaluable in the study of complex biofluids, such as urine, which was shown in a quantitative metabolic study of human urine using both 1H and 13C NMR .
Alternatively, specific metabolites may be followed using tracers, whereby a metabolic precursor is introduced and its fate observed at time points following induction. This is widely described as metabolic flux analysis (MFA) and typical precursors used include glucose and pyruvate. These molecules are ‘tracers’ because they are 13C enriched and their fate is therefore monitored using 13C NMR. This approach is possible because the natural abundance of the 13C isotope is only 1.07% in non-enriched metabolites but is >90% in enriched glucose. As a result, metabolites originating from 13C glucose have significantly larger signals in 13C NMR, but this approach has its limits because as time progresses the 13C labelled carbon atoms from the tracer molecule become diluted and move down a multitude of metabolic pathways. However, careful experiment design can yield amazing results and the strength of this approach is demonstrated when specific carbon atoms in a tracer molecule are labelled with 13C, which can be used to identify where metabolites separate down different pathway branches. Examples include the study of astrocytes in brain metabolism  as well as microbiological studies of CO2 effects on anaerobic succinate production . Although details of experimental 13C flux analysis labelling have been published [22,23] they have been typically geared toward gas chromatography-mass spectrometry (GC-MS) studies to date. Sample preparation for flux analysis typically mirrors that for metabolic profiling.
Specific metabolite and pathway intermediate identification
Nuclear magnetic resonance is pivotal in deducing the molecular make-up of a new metabolic intermediate as well as establishing new theories regarding the enzymatic processes by which a pathway proceeds. Here, it is critical to isolate a pure specific metabolic component in sufficient quantities for analysis. New pathway intermediate information is useful for future metabolic studies, designing competitive inhibitors and delivering enzymatic knowledge for targeted drug discovery. Novel structural information can either be deduced through individual studies of each pathway intermediate, or via the use of specific isotopic labelling to follow the metabolic pathway and register the chemistries involved. Whatever approach is used, the process is both time- and cost-intensive due to the need to produce milligram quantities of relatively pure and potentially chemically fragile material that may require isotopic enrichment.
Examples of pathway intermediate analysis using NMR can be found within recent studies on vitamin B12 (cobalamin) biosynthesis. Cobalamin is a complex biomolecule with a core tetrapyrrole structural unit which is only produced by bacteria and archaea despite being essential for higher order animals. It is a water-soluble vitamin with a multitude of essential roles in nature including the correct functioning of brain and nervous systems as well as the formation of red blood cells, and is of nutritional, biomedical and pharmaceutical importance. Both aerobic and anaerobic metabolic pathways exist in vitamin B12 biosynthesis but the anaerobic pathway has always been perplexing because the isolation and identification of intermediates is challenging. However, selective overproduction of specific enzymes enabled the isolation and NMR analytical identification of all metabolic pathway intermediates between uroporphyrinogen III and cobyrinic acid . This elegant approach utilises an ‘enzyme-trap’ to isolate intermediates  and has also been used to study processes of chemical interest including tetrapyrrole ring contraction  and the biosynthesis of heme and heme-d1 . These projects utilised an explicit suite of NMR experiments to enable identification of each macrocyclic pathway intermediate and a selection of such experiments is shown in Figure 3 for vitamin B12. Of particular interest are 2D experiments such as 1H,1H-TOCSY (Figure 3B), 1H,1H-NOESY (Figure 3C), 13C,1H-HSQC (Figure 3D) and 13C,1H-HMBC (Figure 3E) which provide molecular connectivity and through space proximity information. Although these experiments appear complex, the role of 2D NMR is to simplify data and help identify the molecule in question. 1H,1H-TOCSY provides information that links hydrogen atoms separated by 3-bonds or less in the molecule, 13C,1H-HSQC and HMBC enable carbon atoms to be linked to hydrogen atoms and 1H,1H-NOESY links hydrogen atoms close in space. The combination of all of these datasets is immensely powerful in studies of metabolites as these spectra separate a crowded 1D 1H NMR spectrum into a second dimension, thus providing a unique fingerprint for each pathway intermediate and greatly aiding the assignment process. Further understanding of these experiments can be gained from textbooks and we particularly recommend Claridge's text for both chemists and biochemists .
A selection of vitamin B12 NMR spectra used in pathway intermediate identification:
No isotopic enrichment was used in the experiments shown in Figure 3 and all of the 13C NMR spectra were obtained using the natural abundance of this isotope. This was possible by using a 14.1 tesla (600 MHz 1H NMR) QCI-F cryoprobe to provide a much-needed boost in sensitivity where only 0.5–2.5 mg of the pathway intermediate sample was available, which was dissolved in 0.6 mL of solvent. Also, certain intermediates were air sensitive and Septum Screw-Cap or Schlenk Line NMR tubes were used with the sample sealed under argon or nitrogen gas.
A cryogenically cooled probehead (cryoprobe) works well to improve sensitivity for pathway intermediate applications and could be considered an equally useful resource for signal boosting in metabolic profiling. However, room temperature NMR probeheads are often preferred when reproducible precision is paramount because cryoprobes can introduce minute variations in chemical shift from sample to sample. Such reproducibility in precision is not as critical for the identification of intermediates when NMR is employed in tandem with biochemistry and molecular biology to provide a systems approach to solving the fundamental scientific questions. Such research requires careful planning, resources and a high-quality interdisciplinary team to maximise impact and success.
Nuclear magnetic resonance-based metabolic data comes in many forms. Specific metabolite and 13C flux investigations generally utilise analytical organic chemistry NMR procedures that are described in detail within research texts [27,28]. Data interpretation of metabolic profiles was previously outlined with respect to using databases and reference spectra to identify metabolites but this complex data can also be statistically analysed for confidence and significance. The most common approach is to use multivariate statistics, which embraces the complexity of data where more than a single outcome variable exists. Many researchers opt for the branch of multivariate statistics called principal component analysis (PCA) that analyses the data and then creates a different group of orthogonal variables that contain identical overall information as the original set. Unfortunately, explaining PCA operation is beyond the scope of this review but there are many resources available to describe it, e.g. see reference . Suffice to say, PCA has the ability to expose the internal data structure in a manner that best explains variance in the data. PCA manipulated data are usually shown graphically using two or three orthogonal principal component axes with the data mapped upon it. Data with common principle components tend to form clusters such that if any multivariable differences exist, these will form a separate cluster. As a result, PCA is a way of ‘seeing the wood for the trees’ when the raw metabolic spectrum data are complex. However, a word of warning: clustering in PCA needs to be focused for it to be meaningful and thus concomitant datasets must have low variance. This is the reason that sample preparation must be tightly monitored and the spectrometer set-up identically for all data acquisitions, thus ensuring that meaningful results are obtained by minimising variance from data acquisition which drastically reduces the effectiveness of PCA and its degree of significance. Loss of variance within a known cluster is most probably due to either chemical shift or concentration variance. The latter is most probably due to poor repetitive sample preparation whereas the former is influenced by sample pH differences or inconsistent spectrometer operation. Chemical shift variance between equivalent datasets is often higher, and water suppression more difficult, when using cryoprobes which is why there is a call to return to room temperature probeheads for the next generation metabolism NMR spectrometers. In reality, cryoprobes can offer an excellent sensitivity boost and should not be completely discouraged from metabolic profile studies (see Figures 1, 2 and 3).
As an alternate approach to PCA, some researchers opt for partial least squares discriminant analysis (PLS-DA) that produces a linear regression model by projecting predicted variables and observed variables into a ‘new space’. An excellent review on PLS-DA was recently published by Gromski and colleagues , describing a method that lends itself to NMR metabolomics by targeted analysis. This paper also discusses the relative ease of PLS-DA due to the wealth of available software, but equally warns against the pitfalls of using such approaches in an unguided manner. On that note, it is also worth reading Broadhurst and Kell's article which highlights that many metabolomics statistical approaches are never described in sufficient detail or suitably validated .
Challenges and the future of metabolism studies using NMR spectroscopy
The future for NMR-based metabolomics beyond clinical applications looks very bright whereby the use of such methodologies will enrich multi-omics research projects. Metabolic responses in biological systems are more focused and will suffer less from the high workflows that hamper clinical metabolomics. As the application of NMR-based metabolic studies expands through the biosciences, sample preparation, study design, interpretation and data analysis will improve. This is before new NMR methodologies will assist in answering specific questions; here we predict that NMR will move beyond 1H, 13C (and 31P) as primary target nuclei and begin to use more exotic species including non-biochemical elements such as 19F that have sensitivity approaching 1H whilst providing a significant leap in data simplification. There is also additional scope for technological advancements in NMR for metabolomics through reducing chemical shift variance and improved water-suppression from cryoprobes, including the use of liquid nitrogen cooled probeheads.
Software and online resources
Databases that include NMR data are available from the Metabolomics Centre in Alberta, Canada (http://www.metabolomicscentre.ca/software). These currently cover the metabolomes of human, bovine, E. coli, S. cerevisiae as well as housing specialist repositories for drug bank, toxin and target, small molecule pathway, cerebral spinal fluid and cybercell. This website also has links to software including Metabominer that can be used to assign complex 2D NMR spectra .
The BioMagResBank (http://www.bmrb.wisc.edu) , The Human Metabolome Database (www.hmdb.ca)  and the Birmingham Metabolite Library (http://www.bml-nmr.org/)  have repositories of data for metabolites and other biological molecules.
A list of certain metabolomics software for NMR and MS can be found at http://metabolomicssociety.org/resources/metabolomics-software.
Chenomx commercial software (www.chenomx.com) is widely utilised and enables uploading NMR spectra for referencing and quantitative analysis. It also includes an expandable database of metabolites.
Multivariate statistical packages come in a variety of forms but many groups use SIMCA (http://umetrics.com/products/simca) or AMIX (https://www.bruker.com/products/mr/nmr/nmr-software/software/amix/overview.html).
For further details of software and online resources, see reference .
NMR spectroscopy provides a useful resourceful tool in the study of cellular metabolic processes through global profiling, specific monitoring and pathway intermediate identification.
Cells can be monitored for cellular responses to their environment, nutrition, infection, or physical, chemical and biochemical stresses.
Metabolic variations can be monitored between wild-type and mutant strains to link the metabolome to genome and proteome.
Initial investigations are relatively simple to test, providing access is available to a NMR spectrometer and collaborative technical spectroscopist.
Sample preparation and data analyses do need careful consideration to deliver robust results for a rigorous study. This should not be seen as a barrier and there are many tools available to assist in the data analysis and metabolome mining process.
vitamin B12, cobalamin
Chinese hamster ovary
heteronuclear multiple bond correlation
heteronuclear single quantum correlation
metabolic flux analysis
megahertz=1 × 106 Hz
nuclear magnetic resonance
nuclear Overhauser effect spectroscopy
principal component analysis
partial least squares discriminant analysis
Quadruple resonance Cryoprobe with Inverse detection and Fluorine
total correlation spectroscopy
Water suppression by GrAdient Tailored Excitation
The authors confirm they have no competing interests in relation to this manuscript