Biol. Rev. (2010), pp. 000–000. doi: 10.1111/j.1469-185X.2010.00157.x

1

An introduction to biological nuclear magnetic resonance spectroscopy John H. F. Bothwell1,2,3∗ and Julian L. Griffin4 1 Medical

Biology Centre, Queen’s University Belfast, 97 Lisburn Road, Belfast, BT9 7BL, UK Marine Biological Association of the UK, The Laboratory, Citadel Hill, Plymouth, PL1 2PB, UK 3 UMR 7139, Station Biologique, Place Georges Teissier, 29682, Roscoff Cedex, France 4 The Hopkins Building, Department of Biochemistry, Tennis Court Road, Cambridge, CB2 1QW, UK 2

(Received 26 April 2009; revised 30 July 2010; accepted 10 August 2010)

ABSTRACT Nuclear magnetic resonance (NMR) spectroscopy is one of the most powerful analytical techniques available to biology. This review is an introduction to the potential of this method and is aimed at readers who have little or no experience in acquiring or analyzing NMR spectra. We focus on spectroscopic applications of the magnetic resonance effect, rather than imaging ones, and explain how various aspects of the NMR phenomenon make it a versatile tool with which to address a number of biological problems. Using detailed examples, we discuss the use of 1 H NMR spectroscopy in mixture analysis and metabolomics, the use of 13 C NMR spectroscopy in tracking isotopomers and determining the flux through metabolic pathways (‘fluxomics’) and the use of 31 P NMR spectroscopy in monitoring ATP generation and intracellular pH homeotasis in vivo. Further examples demonstrate how NMR spectroscopy can be used to probe the physical environment of a cell by measuring diffusion and the tumbling rates of individual metabolites and how it can determine macromolecular structures by measuring the bonds and distances which separate individual atoms. We finish by outlining some of the key challenges which remain in NMR spectroscopy and we highlight how recent advances—such as increased magnet field strengths, cryogenic cooling, microprobes and hyperpolarisation—are opening new avenues for today’s biological NMR spectroscopists. Key words: 13 C, 1 H, hyperpolarization, magic angle spinning, magnetization transfer, metabolomics, multidimensional NMR, NMR spectroscopy, 31 P, pulse sequences. CONTENTS I. II. III. IV.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Running an NMR experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) Measuring intensity and optimizing the signal-to-noise ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a) Increasing B0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (b) Signal averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c) Increasing the population difference in a sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (d) Temperature reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (e) Volume reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2) Measuring energy—chemical shift, or δ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a) Shimming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (b) Sample spinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c) Solution-state studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (d) Magic angle spinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 3 4 6 6 7 7 7 8 8 8 9 9 9 9

* Address for correspondence: (Tel: +44 (0)28 90972269; Fax: +44 (0)28 90975877; E-mail: [email protected] or [email protected]) Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

John H. F. Bothwell and Julian L. Griffin

2

V.

VI. VII. VIII.

(3) Measuring relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4) Measuring phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data analysis, or what can the frequency domain tell us? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) How much is there? Metabolic foot- and finger-printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2) What is it? Metabolic profiling and protein NMR spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a) Chemical shift and spin-spin coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (b) Multidimensional NMR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c) Hyphenated NMR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3) Where is it? Metabolite environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4) What is it doing? Metabolite kinetics/behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

I. INTRODUCTION Nuclear magnetic resonance, or NMR, spectroscopy uses radiofrequency waves to reveal information about magnetic nuclei. Since the word ‘spectroscopy’ describes any technique in which electromagnetic (EM) radiation is used to probe atoms (Figure 1), NMR spectroscopy is only one of a range of spectroscopic methods in everyday biological use. However, for reasons which we will consider herein, NMR differs from other forms of spectroscopy in three important ways. First, NMR looks at how the nuclei of a specific, user-selected, chemical element are distributed amongst the molecules of a sample, giving NMR a broader range of targets than most spectroscopic techniques. Second, NMR signals are sensitive to the local surroundings of the nuclei under observation, providing a tool that can probe the chemical and physical environment of an atom and which can reveal more information about a given sample than most other spectroscopic techniques. Third, NMR is more highly penetrating—but, usefully, less damaging—than other forms of spectroscopy. Why is NMR spectroscopy so powerful? To answer this question, imagine standing at the Earth’s North Magnetic Pole—currently an ice floe in the Canadian Arctic—with all of the World’s compasses pointing towards us. Their needles can, with a little effort, be forced to point in another direction, but are in their most stable states when aligned with the Earth’s magnetic field and will relax back to pointing north as soon as they are released. A number of common biological elements have nuclei which behave much like these compasses; they will align themselves in a magnetic field, may be forced to point in another direction, and will relax back to point ‘north’ once released, usually within a few hundred milliseconds. There are two things which we need to add to this magnetic compass analogy to understand enough NMR for most biological applications. First, a magnetic compass can be deflected to point in any direction, but quantum mechanical laws restrict magnetic nuclei to pointing in a much more limited set of directions. These ‘directions’ are actually nuclear energy levels called ‘spin states’ where ‘spin’ refers to the fact that the magnetism of a given nucleus may be

10 10 10 11 12 12 13 14 14 14 15 16 16

considered to arise from the spinning motion of its electrical charge. For instance, when the 1 H and the 13 C nuclei are in a magnetic field, they each have only two allowed spin states, which are called + 1/2 (‘spin up’) and − 1/2 (‘spin down’). 1 H and 13 C are therefore known as ‘spin 1/ nuclei’. It is important 2 to realize that, in any population of 1 H and 13 C nuclei, not all of the nuclei will be in the same spin state; some will be in the higher energy, excited, + 1/2 state, the rest will be in the lower energy, ‘ground’, − 1/2 state, with their proportions depending on the energy difference between the two spin states—the higher the energy difference, the more nuclei will be in the − 1/2 state. Other nuclei, known as ‘quadrupolar’ nuclei snd including the biologically informative 2 H, 18 O and 23 Na, have spins >1/2 and will also have detectable NMR signals. It is, however, important to remember that, although most elements have at least one magnetic isotope, not all nuclei are magnetic and only those with non-zero soin will give a detectable NMR signal. The second point which we need to add to our compass analogy is that individual magnetic nuclei may be moved, or ‘flipped’, from one spin state to another by radiofrequency (RF) waves whose exact frequencies are diagnostic for the chemical element involved. This RF-waveinduced flipping of magnetic nuclei was first observed over 70 years ago by Isidor Rabi’s group at Columbia University (Rabi, 1937; Rabi et al., 1938) and named nuclear induction. However, because some of the early theories to describe this nuclear flipping employed the idea of nuclei resonating at the frequencies at which they absorbed RF waves, Rabi’s ‘nuclear induction’ quickly became known as nuclear magnetic resonance, or NMR (Gorter & Broer, 1942). We will refer to NMR spectroscopy throughout this review, although it should be noted that, in a clinical context, it is common practice to drop the n-for-nuclear to avoid troubling patients unnecessarily with unfounded fears of nuclear radioactivity. Under clinical conditions, therefore, NMR spectroscopy is usually shortened to magnetic resonance spectroscopy, or simply MRS. Following Rabi’s initial physics-based discoveries, NMR’s biological applications began to be realized in 1954, when a group at Stanford used 1 H NMR spectroscopy to show that DNA strands have large hydration shells

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

3 audience which is interested in the biological applications and potential of NMR spectroscopy, but which lacks the mathematical background to tackle traditional theoretical approaches to the subject. We will start by looking at the routines involved in running a typical NMR experiment, we will then explain how these routines reflect the physical basis of NMR and we will end by demonstrating what information these routines can give us. This review has three main sections—the first two (Sections III and IV) are a brief introduction to the principles behind running an NMR experiment, the third (Section V) is a more detailed look at the information which can be obtained from NMR spectroscopy. For those who wish to read more, we recommend the following basic (Claridge, 1999; Derome, 1987; Hore, 1995) and advanced (Hore, Jones & Wimperis, 2000; Keeler, 2005; Levitt, 2001) textbooks, all of which are excellent.

A

B

Fig. 1. Some physical properties of the elctromagnetic spectrum. (A) Electromagnetic (EM) radiation is packaged into photons. Each photon has an associated frequency, ν, and wavelength, λ. Matter can only absorb EM radiation at discrete wavelengths, because the energy, E, of any beam of EM radiation is related to its wavelength, λ, by a constant, h, in the formula E = h/λ. (B) Different regions of the EM spectrum are used for different types of spectroscopy. UV, ultraviolet; Vis visible; IR, infrared; RF, radiofrequency; NMR, nuclear magnetic resonance.

(Jacobson, Anderson & Arnold, 1954). Only three years later, biomolecular NMR had advanced to the stage where an entire intact protein, ribonuclease, was being examined (Saunders, Wishnia & Kirkwood, 1957) and, today, NMR spectroscopy has become one of the most powerful and flexible analytical techniques available to biologists, being one of the few methods which give analytical information from deep within living tissue. The solid theoretical foundations of the effect have been exploited in innovative techniques such as multi-dimensional spectroscopy (Kumar, Welti & Ernst, 1975), Magic Angle spinning of solids (Li, 2006; Lowe, 1959) and quantum computing (Cory, Fahmy & Havel, 1997), while high-throughput NMR is increasingly driving post-genomic functional studies (Bundy et al., 2007). This makes it a regrettable, and preventable, shame that many biologists still view NMR as a prohibitively mathematical technique. Not only is NMR grounded in a phenomenon which most biologists use without demur in other contexts—spectroscopy using EM radiation (Fig. 1B)—but many of the applications of NMR can be understood in non-mathematical terms. Indeed, modern NMR spectrometers allow their users to acquire and analyze complex spectra without needing any deep physical understanding of the concepts by which the data are generated. For these reasons, and because of the ever-increasing workload which modern ‘-omics’ and structural biology place on high-throughput analytical techniques, we have written this review for the large professional and student

II. RUNNING AN NMR EXPERIMENT NMR experiments are, in practical terms, fairly straightforward. For simplicity’s sake, we will break them down into three parts—sample preparation, data acquisition and data analysis—each of which will be covered in separate sections of this review. During sample preparation (Section III), the tissue of interest is prepared, most commonly as a solution in a narrow-walled glass tube, and put into a high-field electromagnet whose large coil generates a strong B0 –pronounced ‘B nought’—field (Section IV). Once inside the B0 magnet, the sample sits in a second, smaller, electromagnetic coil which is found at one end of a poster-tube-sized cylindrical probe. This second coil is, understandably, known as the probe coil (Section IV). During data acquisition (Section IV), the sample—still in the magnet’s B0 field—is irradiated with RF waves generated by the probe coil. RF waves have electrical and magnetic components, so these irradiating RF waves are often known by their ‘magnetic’ name, which is the B1 field. These RF waves—sometimes grouped together in a cluster called a pulse sequence (Sections IV.1.c and V.2.b)—flip nuclei in the sample from one spin state to another. The more power applied in the RF wave, the more nuclei are flipped, a process known as saturation (Section IV.1.c). The irradiating RF waves/B1 field are then turned off and the nuclei allowed to relax (Section IV.3) back to their equilibrium population, which they do by emitting RF waves whose parameters provide information about the sample. These emitted RF waves are collected as Free Induction Decays, or FIDs (Section IV.3) in the Time domain (Section IV.3). The FIDs are mathematically simplified and transferred into the Frequency domain, using a technique called Fourier Transformation (Section IV.4), where they are usually presented in the form of a onedimensional (1D) or two-dimensional (2D) spectrum—or transient—containing a number of resonances (Section IV

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

John H. F. Bothwell and Julian L. Griffin

4 A

A

B

B

Fig. 2. Sample one-dimensional (1D) and two-dimensional (2D) nuclear magnetic resonance (NMR) spectra of yeast metabolite extracts. (A) A 1D 1 H NMR spectrum of yeast metabolites, showing intensity on the y axis and resonant frequency (also known as δ or chemical shift, see Section IV.2a) on the x axis. (B) A 2D spectrum of the same sample as in A, with both axes now representing a different resonant frequency— 13C for the y axis and 1 H for the x axis—and the intensity of the peak being reflected by the closeness of packing in the contour plot. It should be noted that the y and x axes are often called by their spectroscopic names of F1 and F2, respectively. This spectrum comes from a heteronuclear single quantum coherence (HSQC) experiment, which is used to follow 1 H nuclei which are attached to 13 C nuclei. We thank Reza Salek, Duncan MacInnis and Juan Castrillo at the University of Cambridge for allowing us to present these spectra.

and Fig. 2). The emitted RF radiation has four parameters (Fig. 3): (1) intensity (Sections IV.1 and V.1), (2) frequency, which is more commonly expressed as chemical shift (Section IV.2.a), (3) half-life (Sections IV.3 and V.3) and (4) phase (Sections IV.4 and V.2b) and in Section V we will look in turn at how an NMR experiment extracts information from each of these parameters.

III. SAMPLE PREPARATION One of the most striking things about NMR is its ability to acquire information non-invasively and non-destructively from an exceptionally broad range of samples, whether in vivo from within living organisms, such as the human body (Hoult et al., 1974), ex vivo from isolated organ preparations (Bittl & Ingwall, 1985; Brindle & Radda, 1985) and intact tissue samples, such as tumours (Griffin et al., 2003a), or in vitro from homogenous tissue extracts in solid, liquid or gaseous form. NMR spectroscopy displays this versatility because it involves nuclear absorption and emission of RF radiation. Since the

C

Fig. 3. Building free induction decays (FIDs) and onedimensional (1D) spectra from the four common nuclear magnetic resonance (NMR) parameters. (A) Nuclei (filled circles) which have been excited to a high energy spin state will relax back over time (x axis) to a lower energy, ground, state through the emission of photons of electromagnetic (EM) radiation (wavy lines). (B) When these EM photons are summed over a population of relaxing atoms, the decay in the intensity of emitted EM radiation over time will produce a FID. FIDs may be described using four parameters: (1) intensity, (2) frequency, (3) half-life, or t 1/2, and (4) phase. NMR is one of the few spectroscopic techniques which allows information to be extracted from all four of these parameters. RF radiofrequency. (C) Using a technique called Fourier transformation, the FID, or plot of intensity versus time in the time domain, is converted into a plot of intensity versus frequency in the frequency domain.

nucleus makes up only a very small part of any atom, RF radiation directed into tissue will have only a small chance of hitting a nucleus and so RF waves will tend to penetrate tens of centimetres into tissue before they are absorbed, allowing both imaging and spectroscopy inside the human body for example. Furthermore, because the flipping of a magnetic nucleus from one spin state to another involves only a very small amount of energy, any tissue irradiated by RF radiation will suffer very little damage from subsequent heating as nuclei relax back into their original ground state. A technique such as ultraviolet (UV) spectroscopy, by contrast, involves the absorption and emission of radiation by the electron

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

5

clouds which make up the bulk of an atom’s volume, so UV radiation is absorbed rapidly and cannot penetrate as deeply into tissue as RF radiation. Similarly, were microwaves used then they would be absorbed and released by the covalent bonds between atoms, leading to increased molecular motion and significant—and damaging—sample heating. Nonetheless, despite this enormous potential for in vivo work, the bulk of NMR spectroscopy is carried out on samples in solution, for reasons which we will consider in Section IV.2c. Some ex vivo samples—such as blood, urine or cerebrospinal fluid—are already in solution and may be observed almost directly, although care must be taken to ensure that these samples are not compromised in other ways. The haemoglobin in blood, for instance, is paramagnetic—it becomes magnetic in a magnetic field—and so can distort the applied B0 field enough to broaden resonances in the NMR spectrum to an extent where they overlap and much meaningful information is lost, for reasons discussed in Section IV.2. Blood, therefore, is usually analysed as serum or spun down to remove the haemoglobin-containing red blood cells and leave plasma. When this is done, the osmolarity of the serum or plasma must be maintained with salts during any dilution to preserve structures such as lipoproteins. Other fluids present their own idiosyncrasies; NMR spectroscopy of urine can be impaired by the presence of proteins and metal ions such as Mg2+ and Ca2+ which bind various metabolites and broaden their resonances. To overcome this, one can use chelating agents such as ethylenediaminetetraacetic acid (EDTA), which bind these cations (Nicholson, Buckingham & Sadler, 1983). In addition, the pH of urine samples must be fixed by balancing the ratio of sodium phosphate (NaH2 PO4 ) to orthophosphate (Na2 HPO4 ) to prevent the chemical shift of certain resonances varying from sample to sample (Beckwith-Hall et al., 1998). For in vitro tissue extracts, of course, solubilization is much less of a problem and sample preparation is able to focus on the efficient extraction of any molecules present. This is made easier because, in contrast to methods such as gas chromatography mass spectrometry (GC-MS), NMR samples do not need chemical derivatization to make them detectable. A range of extraction methods are available and have previously been summarized (Le Belle et al., 2002), with a typical extraction protocol reading something like this: Frozen [rat brain] slices were transferred to a mortar kept on dry ice. The slices were ground to a powder, and suspended in 5 mL of ice-cold 6% perchloric acid. This suspension was spun at 1500 g for 5 min and the supernatant removed [. . .], neutralized with 1 M KOH and freeze-dried. The freeze-dried residue was resuspended in 650 μl of deuterated water (D2 O) containing 2 mM 3-trimethylsilyl-deuterosodium propionate (d 4 -TSP) as a reference standard. (Bothwell et al., 2001, p1634). Perchloric and trichloric acids are a popular choice for aqueous extracts, especially—as in this example—for brain

tissue in which metabolism must be quenched rapidly. However, acid extraction has the unfortunate disadvantage of oxidizing many metabolites and proteins, so alternative extraction protocols are also used (Pears, 2007). The most common of these is chloroform/methanol extraction, with other widely used procedures including acetonitrile/water extraction and methanol/ethanol/water extraction. These alternatives have the advantage of partitioning metabolites among several distinct fractions—usually a polar, aqueous metabolite fraction, a non-polar, lipophilic metabolite fraction and an intact and unoxidized protein pellet—allowing more complete extraction of metabolites and better characterization of the sample (Fig. 4). As in the example protocol, above, extracted fractions are usually diluted into—or freeze-dried and reconstituted

B

A

C

D

Fig. 4. Different nuclear magnetic resonance spectra from different fractions from a biological tissue. (A) A chloroform/methanol extraction produces three fractions. (B) The lower chemical shift region of the 1 H NMR spectrum from the aqueous, methanol fraction of extracts from the Hwacheong rice cultivar (Oryza sativa L.). This region of the spectrum contains aliphatic polar metabolites, including (1) Trigonelline, (2) overlapping sugar resonances, mostly from sucrose, fructose and glucose, (3) glycerophosphocholine, (4) O-acetyl carnosine, (5) asparagine, (6) methionine, (7) glutamine, (8) glutamate, (9) acetate, (10) lysine, (11) alanine, (12) lactate and (13) overlapping resonances from leucine, isoleucine and valine. (C) Any protein in the sample is less dense than chloroform and forms a layer between the chloroform and aqueous phases. (D) The lower chemical shift region of the 1 H NMR spectrum from the corresponding organic, chloroform fraction extracted from the same sample as B. This fraction contains non-polar metabolites, including resonances from unsaturated fats (15, 16, 19), saturated fats (17, 18), common to all fatty acids (20) and unidentified metabolites too close to the baseline (14). We thank Oliver Jones at the University of Cambridge for allowing us to present these spectra.

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

6 into—solvents in which protons (1 H) have been replaced by deuterons (2 H). Deuterons produce NMR signals at a different frequency to those of protons, so these deuterated solvents serve two purposes. First, they lower the solvent’s contribution to the observed 1 H spectrum by reducing the amount of protonated solvent, which is a far from trivial effect when we consider that the concentration of water-as-solvent is usually around 50 mol l−1 , as opposed to millimolar metabolite concentrations. Solvent effects can also be removed by simple solvent suppression pulse sequences, as explained in Section IV.1b. Second, most modern NMR spectrometers are operated in a ‘deuterium locked mode’ where the known frequency of the deuterium signal is used to calibrate the other frequencies observed during the NMR experiment. Deuterium locking is usually supplemented by adding a chemical shift reference such as, for 1 H NMR spectra, 3-trimethylsilyl-deuterosodium propionate (TSP) or Tetramethylsilane (TMS) to define more accurately chemical shifts (Harris et al., 2002). It should be noted that deuterium locking is less common in older NMR studies, in which external reference standards are used exclusively; this is because older machines were unable to detect the deuterium signal at the same time as the observed nucleus or, for protein structure determination, the observed nuclei. In our example, above, tissue was reconstituted into deuterated water (D2 O), but many other solvents are also available: deuterated chloroform (CDCl3 ) is commonly used for lipid metabolites, deuterated di-methyl sulfoxide (DMSO) has been used to solubilise components in the humin fraction of soil (Simpson et al., 2007) and a range of deuterated solvents are used during combined liquid chromatography and NMR spectroscopy experiments (Section V.2c), according to the chromatographic separations that are being used (Dunn, Bailey & Johnson, 2005). We have emphasized sample preparation for one very good reason: so long as an NMR spectrometer is configured and operated correctly, multiple runs on the same sample show an exceptionally high degree of reproducibility, making high-resolution NMR spectroscopy a more robust analytical approach than those based on chromatographic and/or mass spectrometry based methods, such as high performance liquid chromatography (HPLC) or GC-MS. In practical terms, this means that, so long as a standard operating protocol is followed, researchers can be confident that variation between samples reflects biological, and not instrumental, variation. This is particularly important if spectra are to be analysed by automated pattern recognition tools. Indeed, NMR is such a powerful global analytical approach that it can often detect unexpected biological variation in a dataset and variations in diet, age, growth conditions, hormonal status and strain background have all unexpectedly muddied the interpretation of studies in plants, animals and micro-organisms (Bollard et al., 2005; Griffin & Nicholls, 2006; Gulston et al., 2008). The more standardized the sample preparation can be made, therefore, the better.

John H. F. Bothwell and Julian L. Griffin IV. DATA ACQUISITION After preparation, samples are put into the most distinctive feature of an NMR experiment—the barrel-shaped aluminium magnet—and data acquisition begins. The following example protocols give an idea of the various magnet and spectrometer settings which must now be considered: ‘‘All spectra were obtained at 30 ◦ C on a [. . .] spectrometer operating at a proton frequency of 400.15 MHz. Fully relaxed spectra were acquired with 90◦ pulses applied every 13s for 512 transients. The decoupler was gated on the water frequency, in the delay between pulses, to suppress the residual water peak.’’ (Bothwell et al., 2001, p1634). ‘‘1H spectra were acquired at 400 MHz into 40,000 data points using a 90◦ pulse, an acquisition time of 4 s, and a sweep width of 5 kHz. The overall pulse repetition time was 5 s. The samples were spun at 16 Hz and maintained at 30 ◦ C during data acquisition. The spectra were the sum of 128 transients.’’ (Raamsdonk et al., 2001, p 50). As we stated at the end of Section II, the purpose of these protocols is to measure four parameters (intensity, frequency, half-life and phase; Fig. 3) of the EM radiation which is released when a population of nuclei relax from one spin state to another. We will look in turn at how an NMR experiment measures each of these parameters. (1) Measuring intensity and optimizing the signal-to-noise ratio Any spectroscopic technique benefits from a high signal-tonoise ratio and an important limitation with NMR, relative to other spectroscopies, is its lack of sensitivity. The inherent ability of any nucleus to absorb EM radiation is given by its gyromagnetic ratio: nuclei with higher gyromagnetic ratios will absorb EM radiation more readily than nuclei with lower gyromagnetic ratios. Theorists in the late 1920s and early 1930s realized that the energy difference between nuclear spin states would depend on the strength of the magnetic field in which the nuclei were sitting. Unfortunately, in the Earth’s magnetic field, these energy differences are vanishingly small, so that, even for nuclei with higher gyromagnetic ratios, the small energy differences between nuclear spin transitions are of the order of thermal energy at room temperature. This means that ambient heat suffices to move many nuclei into higher spin states, making the population difference between high and low energy spin states very small—typically less than a few parts per million. This, in turn, reduces the number of low energy nuclei which may be excited in a sample and thus reduces the maximum achievable signal-to-noise. To overcome this low-sensitivity problem, spectroscopists use five common methods to increase the signal-to-noise

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

7

ratio for a given mass of sample: (a) they increase B0 , (b) they average together many signals, (c) they increase the saturation of the sample, (d) they reduce the temperature of the sample and (e) they reduce the volume in which the NMR signal is observed. (a) Increasing B0 Since theory predicted that energy differences between nuclear spin states varied with the external magnetic field strength, and since very low energy EM radiation was almost impossible to detect using the electronics available in the 1930s, Isidor Rabi’s pioneering group at Columbia placed nuclei in an extremely strong homogenous B0 field. This increased the energy differences between spin states so that the nuclei could absorb EM radiation—applied as a B1 field—in the more easily detectable RF range. These early experiments used molecular beams of nuclei, because these early workers were interested in nuclear physics, not in probing biological samples. However, Edward Purcell (Purcell, Torrey & Pound, 1946) and Felix Bloch (Bloch, Hansen & Packard, 1946), working independently at Stanford and MIT, soon realized that NMR signals could also be obtained from more everyday preparations. Paraffin wax and tubs of water were early favourites; we have already explained (Section III) that the strong 1 H signal generated by water remains one of the most common complications in biological NMR spectroscopy and is the reason why samples are usually reconstituted in deuterated solvents. Magnetic field strength is usually measured in Tesla, given the symbol T. Today’s B0 fields are generated by powerful electromagnets which generate fields between 2 and 21 T, several hundred times as strong as the Earth’s own 30–60 mT (Le Mouel, Kossobokov & Courtillot, 2005) field. The actual electromagnetic coil takes up only a small amount of the distinctive aluminium barrels, but in order to generate such high B0 fields, superconducting material must be used and this, in turn, must be cooled to liquid helium temperatures of around 4 K (−269 ◦ C). The bulk of the barrel consists, therefore, of two cooling jackets—a liquid helium one to cool the electromagnet and a liquid nitrogen one to cool the liquid helium. Furthermore, in modern magnets the superconducting B0 coil is surrounded by another coil which prevents, at least in part, the B0 field from extending outside the cylinder, so that spectroscopists can hug a 9.4 T magnet and still be certain that their credit cards will survive. The first way of improving signal-to-noise would simply be to increase the strength of the B0 field, to a current commercial maximum of 21 T, or 950 MHz, thereby increasing the population difference between low and high spin states. Unfortunately, this is usually a prohibitively expensive way to increase signal-to-noise ratios, as magnet costs increase rapidly with field strength. It should be noted that there is an alternative measure of magnet strength commonly used by spectroscopists. We mentioned in Section I that nuclei which absorb RF radiation of a particular frequency are said to resonate at that frequency, hence the R-for-Resonance in NMR

and hence why absorption peaks are commonly known as resonances. As we have explained, the EM frequency at which nuclei absorb, or resonate, depends upon the strength of the B0 field; in a 9.4 T magnet, for example, 1 H nuclei resonate at 400 MHz, in the same range as wireless local area network (LAN) and mobile phones, and 13 C nuclei resonate at 100.2 MHz. Magnets are, therefore, often described by the frequency at which 1 H nuclei resonate, so that the same magnet may equally correctly be referred to as a 9.4 T magnet or a 400 MHz magnet, according to taste. (b) Signal averaging Second, and very commonly, signal-to-noise ratio can be improved by averaging together individual spectra, or transients. However, while signal increases as a function of √ the number of transients, n, noise increases as a function of√ n, which √ means that the signal-to-noise ratio increases as n/ n, or n. A two-fold increase in signal-to-noise therefore comes at a four-fold cost in time, which can soon become prohibitive. (c) Increasing the population difference in a sample In order to maximize signal-to-noise in, for example, NMR spectroscopy with spin 1/2 nuclei such as 1 H and 13 C, the excess nuclei in the ground, − 1/2, state should be flipped into the + 1/2 state to give equal populations of nuclei in each spin state. This process of population equalization is called full saturation of the sample and is achieved by applying a 90◦ pulse from the B1 coil; ‘90◦ ’ reflects the fact that when we redistribute magnetic nuclei in this way, we also move the net magnetization of the sample from the z axis to the xy plane, i.e. through 90◦ . After saturation, we need to collect all of the emitted RF signal by waiting for the flipped + 1/2 nuclei to relax fully back to the ground − 1/2 state, restoring the original population difference between spin states. Unfortunately, it takes a long time to collect spectra using both full saturation, using a 90◦ pulse, and full relaxation, using a betweenpulse delay—the relaxation delay—long enough to allow full restoration of the original populations. This is is a particular problem if many transients are being averaged together. For these reasons, spectroscopists usually aim to strike an optimal balance between minimizing run time and maximizing signal-to-noise, and they do this by applying less-than-90◦ RF pulses (often around 60◦ ) and allowing only partial relaxation. Most NMR spectra are, therefore, acquired under partial saturation and partial relaxation. For most purposes, partial saturation is beneficial, although it can complicate the quantification of some compounds, especially those with long relaxation times, such as the carbonyl 13 C nuclei (R2 -13 C=O) found in carboxylic acids. As an aside, saturation is also used to increase signal-tonoise ratio by reducing noise; specifically, noise from solvent protons (Section III). Because the water resonance is so large, it can swamp other resonances and so is removed from the spectrum by irradiating the water resonance continually, so

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

8 that the population difference between high and low energy states is zero. This means during any subsequent irradiation by EM radiowaves there are no protons in water molecules to contribute a net magnetisation for the NMR effect. The water resonance has effectively been decoupled from the sample resonances, so this process is known as ‘decoupling’. Finally, we can see that decoupling is the simplest example of a ‘pulse sequence’, since it involves using one RF pulse to suppress water and then a second RF pulse to excite the sample nuclei-of-interest. We will meet more complex pulse sequences later (Section V.2b).

John H. F. Bothwell and Julian L. Griffin Even without the improvements associated with cryoprobes and microprobes, typical sample acquisition times for a 1 H NMR experiment where the sample is not limited are between 6 and 15 min. Furthermore, the approach is cheap on a per sample basis and thus large datasets can be rapidly acquired, especially if automation is used, with more than 150 samples in 24 hours being easily achievable. Hence, the tool has been widely used in the drug safety assessment area of toxicology during drug screening (Beckwith-Hall et al., 1998). (2) Measuring energy—chemical shift, or δ

(d) Temperature reduction The signal-to-noise ratio can be increased by temperature reduction in two different ways. Either the temperature of the sample may be reduced by cooling—although this is limited by the temperature at which the solvent freezes—or the temperature of the equipment may be reduced using cryogenic probes (Styles et al., 1984). Sample cooling reduces thermal motion so that more nuclei will sit in lower energy spin states. This increases the population difference between ground and excited spin states and thus increases signal-tonoise ratio by increasing the maximum signal. Probe cooling, on the other hand, reduces the contribution of electronic and thermal noise and provides a theoretical up-to-four-fold increase in signal-to-noise ratio. The increased sensitivity of cryogenic probes has the added benefit of allowing the use of nuclei with lower sensitivity (i.e. lower gyromagnetic ratio) than 1 H, which would normally be prohibited due to the time required to acquire a sufficient signal (Keun et al., 2002). (e) Volume reduction Finally, the more homogenous the B0 field, the better the signal resolution. Since a homogenous field is easier to generate over a small volume than a large, most magnets have narrow bores of only a few centimetres, although those used for in vivo work must, of course, have bores large enough for living organisms to fit inside. These larger bores range from tens of centimetres for small animals to the wide bore medical magnets used for patient studies. Provided the individual is stationary, the effect of magnetic fields on patients is minimal, although recently the EU has tried to place exposure limits for magnetic fields. So little is known about magnetobiology, and even whether magnetic fields can induce long-term effects, that these limits are, to a large extent, arbitrary. Additionally, and for esoteric reasons connected to the geometry of probe coils, small probe coils are inherently more sensitive than larger coils for a given mass of sample, so that microprobes which hold only a few microlitres of sample can give signal-to-noise ratios which rival cryoprobes (Sakellariou, Le Goff & Jacquinot, 2007). Microprobes are also useful, of course, when sample volume is limited for biological reasons, for example in the analysis of the few microlitres of cerebrospinal fluid in the brains of individual mice (Griffin et al., 2002).

We noted that nuclei of the same element absorb EM radiation of a characteristic frequency in a B0 field (Section I). This is not strictly correct and would make for a poor analytical technique, as we would only be able to distinguish different elements within a mixture. We should, more accurately, have said that nuclei of the same element absorb EM radiation over a characteristic range of frequencies. This is demonstrated by the experiment which turned NMR into a versatile analytical tool: the observation that the 1 H spectrum of ethanol —CH3 CH2 OH— consisted of three resonances (Proctor & Yu, 1950). This was unexpected, because it had been assumed that all the 1 H nuclei in a sample would absorb RF radiation of the same energy when moving from the − 1/2 to the + 1/2 spin state. Instead, it swiftly became apparent that the resonance of a nucleus shifted according to its chemical environment, with ethanol’s three resonances reflecting its three 1 H-containing chemical groups; -CH3 , -CH2 - and -OH (Arnold, Dharmatti & Packard, 1951). The discovery of this so-called chemical shift, often abbreviated to the Greek letter δ, brought NMR into many chemistry departments as a highly discriminatory analytical probe and this, together with its penetrance and non-invasiveness, meant that it was soon adopted in the analysis of biological samples (Odeblad, Bhar & Lindstr¨om, 1956). 31 P NMR spectroscopy provides a striking illustration of how chemical shift may be used in experiments on muscle function in vivo (Fig. 5A), with 31 P chemical shift differences readily allowing the spectroscopist to distinguish resonances from phosphocreatine, inorganic phosphate and ATP (Fig. 5B). Notice that, in in vivo spectra from muscle (Fig. 5B), there is no contribution from ADP, which would produce two resonances in 31 P NMR spectra. This was an important early finding of biological NMR, demonstrating that the concentration of ADP measured in assays in vitro does not reflect the low intracellular free concentrations of ADP, because ADP binds readily to proteins within the cell and its effective concentration is reduced accordingly (Meyer, Brown & Kushmerick, 1985). Moreover, because the inorganic phosphate resonance is actually an average of muscle phosphate (H2 PO4 − ) and orthophosphate (HPO4 2− ), it is sensitive to changes in pH, producing a change in chemical shift which can be used to measure pH changes during metabolic insults such as ischaemia (lack of blood flow), hypoxia (lack of oxygen) or cold (Fig. 5B) (Sartoris, Bock & Portner, 2003).

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy A

9 older magnets will require the sample to be spun, usually at around 20 Hz. This spinning averages out any remaining small B0 field inhomogeneties in the horizontal (often referred to as xy) plane, improving the final B0 field homogeneity.

B

(c) Solution-state studies

Fig. 5. The uses of chemical shift: 31 P measurement of muscle pH in vivo. (A) pH changes play an important role in allowing animals to adapt to changes in temperature. This has been investigated in two species of eelpout, a small bony fish, using 31 P NMR spectroscopy. Sartoris et al. (2003) compared two species : the temperate Zoarces viviparus (shown in A) and the cold-adapted Pachycara brachycephalum (Pappenheim, 1912). (B) Muscle energetics were assessed in both species to measure intramuscular pH. This is shown for Zoarces viviparus in normal water (top spectrum) and warm water (bottom spectrum). Intramuscular pH decreases when the fish is heated. Adapted, with permission, from (Sartoris et al., 2003).

As 31 P spectroscopy demonstrates, chemical shift is most useful when signals from nuclei in different chemicals can be distinguished. This separation may be optimized in four ways: (a) shimming, (b) sample spinning, (c) using solutionstate studies and (d) magic angle spinning. (a) Shimming The EM radiation absorbed and emitted by each nucleus is a function of both the B0 field and the chemical environment, so good signal separation requires both of these to be as homogenous as possible. Because no electromagnet, regardless of cost, can ever generate a truly homogeneous B0 field, and because the actual sample itself disturbs the B0 field, the B0 field must be re-homogenized, or shimmed, after sample introduction; a practice which takes its name from the engineering trick of using small wedges of material to adjust the position of a larger object. Shimming used to be a time-consuming manual process, but is invariably automated on modern machines, allowing the B0 field homogeneity to be fine-tuned once the sample is sitting in the magnet. (b) Sample spinning When combined with the auto-shimming mechanisms on modern NMR spectrometers, modern electromagnets can deliver acceptably homogenous B0 fields. However, many

The need for field homogeneity also explains the usual use of samples in solution. In a heterogenous sample, different sample regions will experience different magnetic environments, most commonly through two phenomena called magnetic susceptibility and chemical shift anisotropy. These give rise to local distortions of the B0 field which will, in turn, produce a smear of energy differences between two spin states, rather than a sharp line. In solution, the free tumbling of solutes means that the chemical environments of those solutes are as similar as possible, which reduces this line-broadening to give distinct, nonoverlapping resonances. If shimming, spinning and solutions are effective, resonances will be narrow and well defined, with a good 1 H linewidth—the width of the resonance at half its height—being less than 1 Hz. We should point out here that the cell is not a true solid, but rather a semi-viscous liquid, and its metabolite tumbling rates are fast enough to ensure reasonably narrow resonances for in vivo tissue studies. Furthermore, many tissues are homogeneous enough for the effects of chemical shift anisotropy and magnetic susceptibility to be ignored to a first approximation. This has allowed clinicians to probe every organ of the human body by magnetic resonance imaging and many of these have also been studied by NMR spectroscopy. (d) Magic angle spinning Line-broadening is a particular problem with solid tissue samples, in which compounds cannot ‘tumble’ as freely as they can in solution. A combination of slow relaxation rates and magnetic susceptibility and chemical shift anisotropy effects tends to produce a broad amorphorous resonance in any solid examined by NMR and, at least to a first approximation, would suggest that in vivo spectroscopy would be uninformative for following much solid-state biology. However, resolution of resonances in solid samples can be improved by the process of magic angle spinning (MAS) NMR spectroscopy (Lowe, 1959). Andrew, Bradbury & Eades (1958) observed that many of the line-broadening effects in solids depended on the angle at which the sample was placed with respect to the B0 field. They went on to demonstrate that rapidly spinning a solid sample at the so-called ‘magic angle’, 54.7◦ , resulted in spectra with significantly reduced linewidths. This is caused by a combination of physical effects whose explanation is so complicated that ‘magic’ really is the most descriptive adjective! Briefly, various physical effects scale according to the angle at which the sample is placed with respect to the B0 field; at the magic angle the angular term becomes zero and a narrow resonance is produced. Applying this approach

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

10 to a range of biological tissues, in which most metabolites are in the semi-viscous cell cytosol, linewidths comparable to solution-state NMR spectroscopy have been produced in many different tumour types and neurological tissues taken either during surgery or post mortem in humans (Cheng et al., 1997; Griffin et al., 2003b; Millis et al., 1997). This has been extended further to the spinning of whole organisms, with mice being technically the most convenient (Wind, Hu & Rommereim, 2003). (3) Measuring relaxation Nuclei in excited spin states usually relax with distinctive half-lives of around 100–1000 ms and, like chemical shift, changes in the exact value of the half-life can give information about the environment of the molecules involved. The relaxation signal is measured as a decay in signal intensity over a few seconds—the free induction decay (Fig. 3)—and we note in passing that an exactly analogous, although less informative, phenomenon is seen for light spectroscopy, in which fluorescence lifetime imaging is used to increase the information obtainable from visible light excitation of fluorescent molecules (Suhling, French & Phillips, 2005). The NMR signal decays through two mechanisms: T1 and T2 . The T1 relaxation time (also called the spinlattice relaxation time) refers to the restoration of the equilibrium population of the spins along the B0 field axis (i.e. the superconducting magnet for most NMR applications) following the RF pulse. This process takes place largely through nuclear interactions with the surrounding environment and, because it determines how many nuclei are available for excitation, affects the rate at which exciting, B1 , pulses can be applied to a sample during an NMR experiment. By definition, therefore, the T1 relaxation rate determines the extent to which a sample may be saturated during data acquisition: if the T1 relaxation is long, a long relaxation delay is needed between each pulse (Section IV.1c). The T2 relaxation time (also referred to as spin-spin relaxation time), on the other hand, refers to the loss of magnetisation coherence that creates the EM radiation which is emitted during the NMR experiment. This arises from quantum mechanical effects and the homogeneity of the magnetic field, and determines how broad the NMR resonances are. For various reasons, this T2 relaxation takes place in the xy plane (i.e. perpendicular to the magnet) and is always faster than T1 relaxation. (4) Measuring phase By the late 1950s, NMR spectroscopy had become established as a useful analytical technique, with its main disadvantages, then, as now, lying in its low sensitivity and spatial resolution. This low sensitivity was a particular problem because spectra were acquired using what was called ‘continuous wave’ NMR, in which samples were excited using a series of single-wavelength RF waves, each of slightly longer wavelength than the last, in a manner very

John H. F. Bothwell and Julian L. Griffin similar to modern visible light spectroscopy. Nevertheless, NMR spectroscopy’s ability to penetrate deeply into samples and to identify their composition without damaging them made it particularly well suited to certain in vivo biological applications. If this were all, NMR would be simply another type of spectroscopy; a useful, but not essential, addition to modern biology’s toolkit. There is, however, a particularly useful feature of NMR: because RF waves are low energy, and therefore long wavelength, technical considerations mean that both their intensities and their phases may be determined, in contrast to the majority of EM radiation used by biologists, such as the visible and near-visible light wavelengths, for which only information about intensities may be determined. This means that pulse sequences can be used to simplify spectra and to add information—visible light is only slowly being manipulated in this way with the new generation of ultrafast pulsed lasers (Jonas, 2003; Kroll et al., 2007). There are two ways in which NMR’s phase information has been particularly useful. One of the biggest advances in NMR spectroscopy during the early years of its development was the fast Fourier transform (Ernst & Anderson, 1966) which takes a free induction decay in the time domain and converts it into the frequency domain to produce a spectrum (Fig. 3). This development significantly speeded up spectroscopy as it allowed the acquisition of a range of frequencies in a single pulse, rather than necessitating a laborious scan across a range of frequencies (as occurs in many UV/visible spectrophotometers). This leaves more time for the biggest bugbear of most NMR spectroscopists—the inherent lack of sensitivity of the approach. The second application of phase has been the use of pulse sequence and multidimensional spectroscopy. This will be dealt with further in Section V.2b.

V. DATA ANALYSIS, OR WHAT CAN THE FREQUENCY DOMAIN TELL US? The four parameters of the emitted EM radiation collected from an NMR experiment—intensity, frequency, relaxation and phase (Fig. 3)—contain a great deal of information and NMR spectroscopy therefore provides a highly quantitative and discriminatory analytical tool, capable of distinguishing metabolites in a mixture, or individual amino acids in a protein. Furthermore, multidimensional outputs can be generated to show the dependence of one parameter on one, two or all three of the others, and these inter-relationships can shed light on a wealth of other information, including the environment the molecules are found in and whether they are bound to proteins or to other macromolecules. Broadly speaking, the information from an NMR experiment can tell us four things about a sample: (1) how much of it is present, (2) what it is, (3) where it is and (4) what it is doing. We will now examine some examples.

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

11

(1) How much is there? Metabolic foot- and finger-printing The simplest, and most common, type of NMR spectroscopy produces a spectrum of intensity on the y axis against frequency on the x axis—a so-called 1D plot, or frequency domain (Fig. 3). This is analogous to the intensity versus wavelength plots generated by other forms of spectroscopy. Resonances in 1D plots are mathematically multiplied in a process called line-broadening, which simplifies subsequent quantification by smoothing the raw data to reduce the contribution noise makes to the resultant NMR spectrum. Because the absolute frequency at which nuclei resonate is primarily a function of the applied B0 field, resonances are normalized by plotting the x axis as a shift from a reference standard, so that published chemical shifts are independent of the magnet strength and so that spectra acquired in different B0 fields may be compared. Indeed, the use of such a relative frequency scale is even needed to compare results between magnets of the same nominal B0 field strength. In order to ensure that magnets of a given strength do not interfere with one another if placed in the same facility they will have a small offset in terms of the magnet field strength. These B0 differences between apparently equivalent magnets are often greater than proton chemical shifts. As mentioned in Section IV, chemical shift differences are usually small, so that if the 1 H nuclei in the external shift reference, TSP (Section III), resonate at 400 MHz in a 9.4 T magnet, the C1 1 H nuclei in glucose will resonate at 400.0021 MHz. Chemical shift is, therefore, given in parts per million (ppm) so that, in this particular example, glucose would have a set of C1 1 H resonances centred at 1,000,000 ∗ (400.0021 − 400)/400 = 5.25 ppm. Any given experiment will cover a certain ppm range, called its ‘sweep width’. Because NMR is a reproducible quantitative approach, 1D spectra can be used even without identification of the molecules which have contributed to the spectrum. As we noted in Section IV.1c, one of the ways in which signal-to-noise ratios are increased is by performing NMR spectroscopy in a mode that is termed partially saturated, where nuclei are excited at a quicker rate than would allow the NMR signal to decay completely to zero. Fortunately, saturation factors can be calculated readily, so a simple linear relationship exists between the area under a partially saturated peak and the concentration of the nuclei that produce the signal. This gives us a characteristic metabolic phenotype from every sample—akin to a barcode, or a snapshot—which can be statistically separated from another set of samples using multivariate analysis such as principal components analysis, even if the individual resonances cannot be identified (Bundy et al., 2007; Raamsdonk et al., 2001). This snapshot approach has proved itself to be particularly useful in functional genomic studies, in which genes are altered and the phenotype of the resulting mutants is examined to provide clues about gene function. Phenotypes have traditionally been scored using a very limited range

of parameters, such as growth rate in yeast. However, changes in gene activity will often result in apparently silent mutations, as other regulatory processes compensate to produce no discernible change in gross phenotype (Teusink et al., 1998). In such cases, a metabolic snapshot, quantifying resonances without needing to identify them, may be used as a multiparametric and extremely sensitive phenotype. If the metabolic snapshot comes from cell, tissue or organism preparations, this approach is known as metabolic fingerprinting; if the snapshot looks at cell media or excretory products, it is known as metabolic footprinting (Allen et al., 2003). Both finger- and footprinting work best with sensitive and abundant nuclei, such as 1 H; one of the most famous examples of this approach is the highly cited functional analysis by co-responses in yeast (FANCY) manuscript by Raamsdonk et al., 2001. This work showed that a number of Saccharomyces cerevisae (yeast) strains with similar growth rates had markedly different 1 H NMR spectra which could be used to distinguish glycolytic mutants from oxidative phosphorylation mutants. FANCY has recently been extended to identify metabolic modules in S. cerevisiae (Bundy et al., 2007). Briefly, extracellular metabolite profiles, metabolic footprints, were quantified by 1 H NMR spectroscopy and analyzed using principal components analysis (PCA) (Fig. 6). This is an unsupervised statistical analysis technique that fits a large multivariate dataset onto a multidimensional set of orthogonal axes. These axes, which are called principal components, are chosen so that each axis explains as much of the variance remaining in the dataset as possible. The overall aim of PCA is to reduce the number of parameters needed to describe the differences among datasets and, properly applied, PCA is a powerful way to classify mutants and their associated genes (Fig. 6). As well as the convenience of not having to identify metabolites, snapshot approaches can also be used in high throughput screens where metabolite identification can be postponed until an effect has been observed. Using such an approach, the consortium for metabonomic toxicology (COMET) investigated around 150 model liver and kidney toxins during a three-year study using NMR-based analysis of urinary metabolites (Lindon et al., 2005; 2003). It is hoped that such an approach will allow the generation of expert systems where liver and kidney toxicity can be predicted for model drug compounds, with the databases being easily transferable among laboratories. In addition the approach has also been used to screen human populations to understand disease processes (Brindle, 2002; Kirschenlohr et al., 2006) and monitor the effects of diet (Lenz et al., 2004). This metabolomic approach to analyzing gene function has three advantages over gene- or microarray-based approaches. First, it measures functional entities, which reduces biological noise. Second, there are a smaller number of metabolites than there are genes. Which simplifies analysis: at the last count, around 1500 metabolites were found in the yeast S. cerevisiae as opposed to 6000 genes (Goffeau et al., 1996; Herrgard et al., 2008). Indeed, the components of

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

John H. F. Bothwell and Julian L. Griffin

12

A

0.6

B

PUT3

Glucose

PUT2 HO

0.4

0.5

PUT1 URA3

PC1 loadings

0.2

PC2

1.0

URA8

URA5

0.0

URA7

-0.2

0.0

-0.5

PPR1

-0.4 URA2 URA10

-0.6

-1.0 -1.0 -0.5

0.0

0.5

1.0

1.5

2.0

2.5

PC1

Variable order

Fig. 6. Metabolic footprinting of yeast mutants. (A) principal component 1 (PC1) (x axis) versus principal component 2 (PC2) (y axis) scores plotted for 1 H nuclear magnetic resonance (NMR) footprints from eleven Saccharomyces cerevisiae yeast strains: three proline utilization mutants (PUT1, PUT2, and PUT3; red), seven pyrimidine biosynthesis mutants (URA2, URA3, URA5, URA7, URA8, URA10, PPR1; blue), and one control (HO, black). The centre of each ellipse is the mean of the principal component on that axis, with the margins indicating one standard deviation. The principal components of most strains cluster together, but those of URA7 and URA8 sit well away from the others, on the right of the plot. (B) The difference between URA7 and control spectra is plotted as a loadings plot, with the variable order (the pattern recognition software’s term for chemical shift) on the x axis. Loadings plots highlight which metabolites are most discriminatory in the PCA plot. In this example, the regions containing glucose are all shown to have increased in the loading for PC1, demonstrating that URA7 has increased glucose concentrations relative to the control strain. Taken together, these results suggest that the URA7 and URA8 gene products are subject to different regulatory controls than the other genes. Adapted, with permission, from Bundy et al., (2007).

central metabolism which dominate the metabolome only number a hundred or so metabolites. Third, it is contextdependent because it can measure how the plasticity of a mutant phenotype in response to environmental and external stimuli. When considering large-scale analyses of biological mixtures, the main alternative to NMR spectroscopy has traditionally been mass spectrometry. Mass spectrometry is undoubtedly the more sensitive technique and in tissue extracts in which 30–60 metabolites might be detected by NMR, a relatively cheap GC-MS instrument will often find more than 100 (Atherton et al., 2006), with the upper limit to metabolite numbers thought to depend on the constraints of cellular osmolarity (Krishnan, Kruger & Ratcliffe, 2005). However, despite recent improvements in mass spectrometry sensitivity and robustness GC-MS still identifies less than 20% of the metabolites present, and methods such as liquid-chromatography-based MS and Fourier transform MS, as well as niche techniques such as Coulombic arrays, fare little better. The fact that these techniques can still measure only a small proportion of the total metabolome (see Ellis et al., 2007) for a broad review of metabolomic techniques suggests that there are problems associated with metabolite extraction and resolution, rather than technique sensitivity. In any case, because NMR analyses are reproducible, can detect certain critical metabolites which define the regulatory metabolic networks (Delgado et al., 2004) and, crucially, may be used on a sample before derivitization for GC-MS, there is increasing consensus that NMR spectroscopy and

mass spectrometry are best used in conjunction, rather than in competition, to maximize the coverage of the metabolome. (2) What is it? Metabolic profiling and protein NMR spectroscopy Metabolic footprinting assigns a quantity to each peak, but 1D spectra of the frequency domain also carry qualitative information about a sample. We noted in Section IV.2 that the spectrum of ethanol gave three distinct resonances and, in general, each chemical will have its own characteristic pattern of peaks. There are a number of ways in which we can identify these resonances, drawn from analytical chemistry. These are: (a) chemical shift, (b) multidimensional NMR, and (c) hyphenated NMR. (a) Chemical shift and spin-spin coupling As we noted in Section IV.2, the energy difference between spin states is dependent upon nuclear environment, so excited nuclei from one element, but in different nuclear environments, will emit slightly different RF wavelengths of EM radiation. In NMR, the energy of the emitted EM radiation is expressed as the chemical shift, δ, and the exact value of δ gives a reasonable indication of the molecule’s identity. This identification may be narrowed down still further if we consider each resonance’s splitting pattern. Many resonances at a given chemical shift will be split into two,

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

13

three, or more sub-peaks, called, respectively, doublets, triplets and multiplets. These splittings arise through a phenomenon called spin-spin coupling; each nucleus behaves as a tiny magnet within the molecule, so that nucleus A will affect the energy of nucleus B, and vice versa, so long as they are positioned close to one another. Because each molecule has a characteristic number and pattern of these sub-peaks, a simple comparison of unidentified peaks with standard reference spectra will often allow identification of the metabolites in a sample. Accordingly, a number of NMR spectral databases have been set up to aid this metabolite assignation; these databases include the Human Metabolome Database (Wishart et al., 2007) and the Madison Metabolomics Consortium Database (Cui et al., 2008), as well as a number of proprietary software tools.

can use RF pulse sequences to add information to a spectrum which might help with assignation. For example, one of the most efficient methods for determining which compounds are in a mixture is to use heteronuclear pulse sequences which detect which 1 H nuclei are connected to which 13 C nuclei. A variety of such pulse sequences exist, which have unfortunately opaque names like heteronuclear single quantum coherence (HSQC), heteronuclear multiple quantum coherence (HMQC) and heteronuclear multiple bond coherence (HMBC) (Braun, Kalinowski & Berger, 1998). These 1 H-13 C heteronuclear pulse sequences are especially useful because they combine the relatively high sensitivity of 1 H NMR spectroscopy with the increased dispersion of 13 C NMR spectroscopy, the latter having a chemical shift range of approximately 250 ppm compared with approximately 15 ppm in 1 H NMR spectroscopy. Thus by detecting a combination of 1 H and 13 C resonances in a metabolite, we can rapidly narrow down the identity of the compounds that could be present in that mixture. For instance, in order to understand tissue energetics, it is important to know the relative levels of highenergy phosphates. The sperm of many animals contain high levels of ATP, but mammalian sperm do not, containing instead phosphodiesters and phosphomonoesters. Because phosphomonoester resonances are poorly disperse, multidimensional heteronuclear NMR spectroscopy has been used to identify the exact species of phosphomonoester present and to further our understanding of mammalian fertilization and reproduction (Fig. 7).

(b) Multidimensional NMR The previous section showed that, in theory, unknown compounds in a spectrum can be identified by their chemical shift and by the number and pattern of their sub-peaks in the frequency domain. Unfortunately, because the differences in chemical shift are relatively small for 1 Hcontaining molecules, biological 1 H NMR spectra are often cluttered with many overlapping resonances that make peak identification difficult: we say that 1 H peaks are ‘poorly dispersed’. This dispersion problem occurs in most forms of spectroscopy, but highlights one of the strengths of NMR; because it is possible to shape RF waves and measure their phase, and because it is also possible to detect the nuclei of different elements simultaneously in a single experiment, we A

B

Fig. 7. Pulse sequences in multidimensional nuclear magnetic resonance (NMR) spectroscopy. (A) Two-dimensional (2D) 1 H, 31 P heteronuclear multiple bond correlation (HMBC) NMR spectroscopy of lyophilized boar sperm extracts. HMBC is a pulse sequence which detects 1 H nuclei separated by a given number of bonds from 31 P nuclei and, in this case, shows a chemical shift correlation between the 31 P phosphomonoester peak (PME) and a given 1 H peak at around 4 ppm. The inset shows an expansion for the PME cross-peak ; glycerol 3-phosphorylcholine (GPC) and inorganic phosphate (Pi ) peaks are also labelled on the 31 P spectrum. (B) 2D 1 H, 13 C gradient-selected heteronuclear single quantum coherence (HSQC) of boar sperm extracts. HSQC is a pulse sequence which observes 1 H nuclei attached to 13 C nuclei and this spectrum shows that the 1 H peak identified in A is linked to an adenine base, which identifies the major phosphomonoester component as AMP. Adapted, with permission, from Kalic et al., (1997). Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

14 The field where multi-dimensional NMR spectroscopy has most impressively been used is its use in the determination of protein structures as first suggested by among others Kurt Wuthrich (Wuthrich, 1990). Proteins may have thousands of NMR resonances associated with them, and thus simple one-dimensional 1 H NMR spectra are highly congested. To circumvent this dispersion problem, spectra from proteins are usually collected across multiple dimensions often associated with different nuclei. In particular, because the backbone of peptides consists of amide linkages, the use of 1 H, 13 C and 15 N multi-dimensional NMR spectra is common in protein NMR spectroscopy. These experiments can consist of two-dimensional NMR spectra such as 1H-13 C HSQC and 1 H-15 N HSQC experiments which take hours to 2–3 days to acquire depending on the protein, whether it has been artificially labelled with amino acids containing 13 C and 15 N and the field strength of the magnet used. Alternatively, more time-consuming three-dimensional experiments can be performed that simultaneously measure the interactions between 13 C, 15 N and 1 H or experiments. Furthermore, there are certain NMR approaches that do not rely on magnetisation being carried through bonds as occurs in the HSQC experiment but move through space. The nuclear Overhauser effect spectroscopy (NOESY) pulse sequence is used to measure how close atoms are in proteins and thus useful in placing distance constraints on the size and shape of a protein. The field of protein spectroscopy is a large and complex one and we direct the interested reader to specialist reviews in this area, notably Wuthrich (1990), Reid et al., (1997) and Shin, Lee & Lee (2008). (c) Hyphenated NMR Multidimensional NMR adds another dimension (or dimensions) to a spectrum and effectively adds more information to aid peak assignation. This extra dimension does not have to be an NMR one and it is often desirable to combine NMR with other techniques. For example, we can include online chromatographic separation, which fractionates the sample before running a sequence of NMR spectra. This is analogous to GC-MS and liquidchromatography mass spectrometry (LC-MS). In addition to separating metabolites according to their chromatographic properties, liquid chromatography also concentrates metabolites into chromatographic peaks. By increasing the local concentrations of metabolites, the sensitivity of the NMR experiment can be increased (Bailey et al., 2002). (3) Where is it? Metabolite environment Differences in chemical shift reflect differences in the environment of a nucleus. So far, we have assumed that different environments mean similar nuclei in different molecules. However, they may also reflect the same molecules in different environments. A number of properties of the spectral peak may be affected by environment. Most simply, the chemical shift may change, as is the case with the inorganic phosphate peak in 31 P spectroscopy, whose

John H. F. Bothwell and Julian L. Griffin chemical shift is pH dependent (Section IV.2 and Fig. 5). However, the relaxation times T1 and T2 (Section IV.3) and the diffusion properties (which can be measured by NMR) can also be used to provide information about molecular environments. For relaxation times, both T1 and T2 are influenced by how fast individual metabolites tumble in the B0 field. Metabolites in a restricted environment, and which cannot tumble freely, produce broad resonances which are most easily distinguished by direct measurement of T1 and T2 relaxation times. This approach has been used to demonstrate the restrictive environment of the mitochondrion for both proteins (Haggie & Brindle, 1999) and metabolites (Bollard et al., 2003). In addition, so-called relaxation editing has been used to distinguish different metabolites in complex biofluids such as plasma using pulse sequences (Liu, Nicholson & Lindon, 1996). An alternative to relaxation-based approaches is to distinguish metabolites by their diffusion properties. This approach uses a magnetic gradient in which the B0 field varies, usually linearly, across the sample. A second B0 gradient is then applied that is equal in magnitude but opposite in sign. For metabolites that do not move rapidly these two magnetic gradients will cancel out exactly. However, the nuclei of rapidly moving metabolites will receive a net magnetisation which will vary according to the position of the nuclei. Averaging over the whole sample, this results in a broadening of the resonance. This approach has also been used to distinguish low molecular weight metabolites from lipoproteins and bound metabolites in blood plasma and biological tissues such as intact tumours using so-called diffusion weight spectroscopy (Liu et al., 1996; Griffin et al., 2003a). (4) What is it doing? Metabolite kinetics/behaviour As with any form of spectroscopy, NMR can follow changes in peak intensity over time to give an idea of the timecourse of metabolite reactions in a way which is not too different to the use of radiolabels or similar metabolite tags. However, NMR is once again set above other forms of spectroscopy by our ability to measure the phase of EM radiation and thus to use pulse sequences. In particular, using an experiment referred to as magnetization transfer, NMR spectroscopy may, effectively, introduce a metabolite label at any point during the course of an experiment (Bittl & Ingwall, 1985; Degani et al., 1985). For example, magnetization transfer has been used to examine the transfer of phosphate between phosphocreatine and ADP (Brindle & Radda, 1985). Here, the phosphocreatine phosphorus nucleus is irradiated by RF waves in such a manner that its population of spins is cancelled out (and hence the resonance would be invisible in the NMR effect). When this phosphate group is transferred from phosphocreatine to ADP to produce ATP, it carries its magnetisation with it. This shows up as a reduction in the signal of the ATP γ phosphate. The rate of reduction in the intensity of the ATP γ phosphate resonance can then be used to estimate the activity of the creatine kinase which

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

15

catalyses the conversion of phosphocreatine to ATP (Brindle & Radda, 1985). With the development of new protein labeling schemes, NMR spectroscopy can now reveal structural and binding details of large protein complexes. Binding of substrates will induce changes in the structure of the enzyme and this in turn will alter the chemical shift of the resonances observed in that protein, especially of those involved in the binding site. For instance, Velyvis et al., (2007) studied ligand binding to the 306 kDa aspartate transcarbamoylase (ATCase) enzyme using highly deuterated, 1 H,13 C-methyl labelled ATCase together with methyl-transverse relaxation optimized NMR spectroscopy. Previous research had demonstrated that the cooperative binding of active site ligands follows the Monod-Wyman-Changeux (MWC) model of allostery, which proposes the existence of two conformations of an enzyme in equilibrium, T and R. The T form has a lower affinity for substrates, and as more sites in the enzyme bind substrate, the equilibrium shifts toward the R state. The authors established that for ATCase, which catalyzes the first step in pyrimidine biosynthesis, only the T state of the enzyme can be observed in NMR spectra of unliganded ATCase. However, the binding of nucleotides such as ATP or CTP shifts the T-R equilibrium so that correlations from the R state become visible, consistent with the MWC model, with these resonances shown in black in Fig. 8. This study emphasizes the utility of modern solution NMR spectroscopy in understanding protein function, even for systems with high molecular masses (Velyvis et al., 2007). Finally, in just the same way that 14 C-labelled compounds have been used to determine substrate metabolism in perfused tissues and the whole organism, 13 C-labelled compounds can be used and detected by 13 C NMR spectroscopy. Although 13 C NMR spectroscopy is of inherently low sensitivity when compared to 1 H NMR spectroscopy— 13 C’s low gyromagnetic ratio making it an order of magnitude less detectable than 1 H—it can be performed in vivo or in situ in real time allowing the measurement of the rates of processes such as the tricarboxylic acid (TCA) cycle (Malloy, Sherry & Jeffrey, 1990), or even mitochondrial transport (Yu et al., 1996). Furthermore, unlike radioactivity-based approaches, one can also determine the exact position of the label, which has allowed the study of metabolic compartmentation within the brain (Badar-Goffer et al., 1990; Griffin et al., 1998) and substrate selectivity within the heart (Malloy, Sherry & Jeffrey, 1988). Moreover, the problem of poor sensitivity is now being addressed for some applications involving 13 C NMR spectroscopy, thanks to a resurgence of interest in a technique called hyperpolarization (Golman et al., 2003; Schroeder et al., 2008), in which huge population differences can be built up between ground and excited spin states in certain molecules. In hyperpolarization, the sample is mixed with a polarization transfer agent before being frozen solid. The polarization transfer agent is typically something with an unpaired electron which can be excited by microwave radiation. This, incidentally, is electron spin resonance,

Fig. 8. Protein spectroscopy. Two-dimensional (2D) 1 H-13 C methyl-transverse relaxation optimized spectroscopy (TROSY) heteronuclear multiple quantum correlation (HMQC) spectrum of U-[2 H] Ile-[δ113 CH3 ]-labelled aspartate transcarbamoylase (ATCase) recorded at 800 MHz, showing 27 resonances which correspond to the 27 Ile residues in ATCase. TROSY is a pulse sequence that specializes in the analysis of large proteins. For normal one-dimensional (1D) and 2D spectroscopy, resonance linewidths increase with molecular mass, producing short T2 relaxation times, broad resonances and correspondingly poor resolution. TROSY avoids this line-broadening and produces narrow peaks for large proteins, allowing the discrimination of resonances not possible using other pulse sequences. By way of demonstration, the arrows point to the cross-peaks through which the four 1D traces have been drawn. These four traces show the high signal-to-noise ratio and illustrate the sensitivity of the spectrum, which comes from solution studies and the very high B0 field strength (800 MHz). Peaks from the r and c chains of ATCase are colored red and black, respectively, with the change in structure induced by ligand binding. Adapted, with permission from (Velyvis et al., 2007).

which is analogous to NMR except that it occurs at a higher EM frequency, since electrons have larger energy differences between their two spin states. In the solid, frozen, state magnetization is transferred from the polarisation transfer agent onto the sample. The sample is then rapidly thawed before being used for conventional NMR spectroscopy. Hyperpolarization can build up huge population differences, but with the disadvantage that differences begin to disappear through T1 relaxation as soon as the sample is thawed (Golman et al., 2003). As a result most applications have so far involved carbonyl 13 C resonances, as these tend to have relatively long T1 times (tens of seconds) and thus allow the magnetization to be maintained during the experiment. Despite this limitation, this approach has been used to follow metabolism in tumours and heart tissue with exquisite temporal resolution (Fig. 9).

VI. CONCLUSIONS (1) NMR spectroscopy is a powerful and flexible technique which allows biologists to study the molecular and

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

16

John H. F. Bothwell and Julian L. Griffin

A

(7) The frequencies, or chemical shifts, of peaks in an NMR spectrum can be used to identify which compounds are present in a mixture. (8) The half-lives, or relaxation times, of peaks in an NMR spectrum can give information about molecular environments. (9) NMR spectroscopy has more analytical power—and is more complicated—than other forms of spectroscopy because we are able to measure the phase of RF radiation. This allows us to study the connectivities between magnetic nuclei, giving a tool with which to study protein structure and dynamics. (10) We have focussed mainly on small-molecule spectroscopic applications, despite huge literatures on magnetic resonance imaging and protein NMR spectroscopy. However, we hope that interested readers will now be armed with the key concepts of NMR spectroscopy and will be in a position to benefit from more specialised reviews and primary publications.

B

Fig. 9. Hyperpolarization. (A) An example time course of spectra acquired from a rat heart in vivo over 60 s of data acquisition. The arrival of [1-13 C]pyruvate can be seen, followed by its subsequent decay back to thermal equilibrium levels. The appearance of the metabolic products, [1-13 C]lactate, [1-13 C]alanine and bicarbonate (H13 CO3 − ) follows shortly afterwards through [1-13 C]pyruvate metabolism. (B) An example of the temporal variation in the fitted peak areas of each metabolite over the 60 s time course. The pyruvate area has been reduced by a factor of 10 to improve visualization. This figure was supplied to the authors by Damian Tyler at the University of Oxford.

(2)

(3)

(4)

(5)

(6)

cellular environments of magnetic nuclei. Commonly studied magnetic nuclei include 1 H, 13 C and 31 P. In NMR spectroscopy experiments, magnetic nuclei are placed in a strong B0 electromagnetic field and excited from lower energy spin states to higher energy spin states by the absorption of EM radiation generated by a second, B1 , electromagnetic field. These excited nuclei then relax by emitting EM radiation of slightly less energy; both exciting and emitted radiation are in the RF range. Because NMR spectroscopy uses low-energy RF radiation which is absorbed by nuclei, it can penetrate deeply into tissue and is non-invasive. Unfortunately, the low energy of RF radiation also makes NMR spectroscopy a relatively insensitive technique. There are a number of practical ways around this low-sensitivity problem, including hyperpolarization. The emitted RF radiation has four parameters which may be analyzed for information about the sample: intensity, frequency, half-life and phase. The intensities of peaks in an NMR spectrum reflect molecular concentrations, so NMR spectroscopy is often used as a standard, automated, highthroughput analytical technique, especially suitable for metabolomics.

VII. ACKNOWLEDGEMENTS J.H.B. is funded by a Leverhulme Early Career Fellowship. Additionally, we are grateful to the BBSRC, MRC, NERC and Royal Society for financial support. We would like to thank Reza Salek, Duncan MacInnis and Juan Castrillo at the University of Cambridge for the yeast HSQC spectrum in Fig. 2, Oliver Jones at the University of Cambridge for the rice spectra in Fig. 4, Franz-Josef Sartoris at the AlfredWegener-Institut for the eelpout spectra in Fig. 5, Kevin Brindle at the University of Cambridge for the yeast spectra in Fig. 6, G¨unter Kamp at Johannes Gutenberg University for the sperm spectra in Fig. 7, Lewis Kay and Algirdas Velyvis at the University of Toronto for the ATCase spectrum in Fig. 8 and Damian Tyler at the University of Oxford for the spectra in Fig. 9.

VIII. REFERENCES Allen, J., Davey, H. M., Broadhurst, D., Heald, J. K., Rowland, J. J., Oliver, S. G. & Kell, D. B. (2003). High-throughput classification of yeast mutants for functional genomics using metabolic footprinting. Nature Biotechnology 21, 692–6. Andrew, E., Bradbury, A. & Eades, R.G. (1958). Nuclear magnetic resonance spectra from a crystal rotated at high speed. Nature 182, 1659. Arnold, J., Dharmatti, S. & Packard, M. (1951). Chemical effects on nuclear induction signals from organic compounds. Journal of Chemical Physics 19, 507. Atherton, H. J., Bailey, N. J., Zhang, W., Taylor, J., Major, H., Shockcor, J., Clarke, K. & Griffin, J. L. (2006). A combined H-1-NMR spectroscopyand mass spectrometry-based metabolomic study of the PPAR-alpha null mutant mouse defines profound systemic changes in metabolism linked to the metabolic syndrome. Physiological Genomics 27, 178–186. Badar-Goffer, R. S., Bachelard, H. S. & Morris, P. G. (1990) Cerebral metabolism of acetate and glucose studied by 13C-NMR spectroscopy. A technique for investigating metabolic compartmentation in the brain. Biochemistry Journal 266, 133–9. Bailey, N. J. C., Sampson, J., Hylands, P. J., Nicholson, J. K. & Holmes, E. (2002). Multi-component metabolic classification of commercial feverfew preparations via high-field H-1-NMR spectroscopy and chemometrics. Planta Medica 68, 734–738.

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

Biological NMR spectroscopy

17

Beckwith-Hall, B. M., Nicholson, J. K., Nicholls, A. W., Foxall, P. J. D., Lindon, J. C., Connor, S. C., Abdi, M., Connelly, J. & Holmes, E. (1998). Nuclear magnetic resonance spectroscopic and principal components analysis investigations into biochemical effects of three model hepatotoxins. Chemical Research in Toxicology 11, 260–272. Bittl, J. A. & Ingwall, J. S. (1985) Reaction rates of creatine kinase and ATP synthesis in the isolated rat heart. A 31P NMR magnetization transfer study. Journal of Biological Chemistry 260, 3512–7. Bloch, F., Hansen, W. W. & Packard, M. (1946). Nuclear Induction. Physical Review 69, 127–127. Bollard, M. E., Murray, A. J., Clarke, K., Nicholson, J. K. & Griffin, J. L. (2003). A study of metabolic compartmentation in the rat heart and cardiac mitochondria using high-resolution magic angle spinning H-1 NMR spectroscopy. Febs Letters 553, 73–78. Bollard, M. E., Stanley, E. G., Lindon, J. C., Nicholson, J. K. & Holmes, E. (2005). NMR-based metabonomic approaches for evaluating physiological influences on biofluid composition. NMR in Biomedicine 18, 143–62. Bothwell, J. H., Rae, C., Dixon, R. M., Styles, P. & Bhakoo, K. K. (2001). Hypo-osmotic swelling-activated release of organic osmolytes in brain slices: implications for brain oedema in vivo. Journal of Neurochemistry 77, 1632–1640. Braun, S., Kalinowski, H. O. & Berger, S. (1998). 150 and more basic NMR experiments : a practical course. Wiley-VCH, Weinheim : Chichester. Brindle, K. M. (2002). Special issue: Assessing tumor vascular function using magnetic resonance imaging. Nmr in Biomedicine 15, 87–88. Brindle, K. M. & Radda, G. K. (1985). Measurements of exchange in the reaction catalysed by creatine kinase using 14C and 15N isotope labels and the NMR technique of saturation transfer. Biochimica Biophysica Acta 829, 188–201. Bundy, J. G., Papp, B., Harmston, R., Browne, R. A., Clayson, E. M., Burton, N., Reece, R. J., Oliver, S. G. & Brindle, K. M. (2007). Evaluation of predicted network modules in yeast metabolism using NMR-based metabolite profiling. Genome Research 17, 510–9. Cheng, L. L., Ma, M. J., Becerra, L., Ptak, T., Tracey, I., Lackner, A. & Gonzalez, R. G. (1997). Quantitative neuropathology by high resolution magic angle spinning proton magnetic resonance spectroscopy. Proceedings of the National Academy of Sciences, U S A 94, 6408–13. Claridge, T. (1999). High-resolution NMR techniques in Organic Chemistry. Pergamon. Cory, D. G., Fahmy, A. F. & Havel, T. F. (1997). Ensemble quantum computing by NMR spectroscopy. Proceedings of the National Academy of Sciences of the United States of America 94, 1634–1639. Cui, Q., Lewis, I. A., Hegeman, A. D., Anderson, M. E., Li, J., Schulte, C. F., Westler, W. M., Eghbalnia, H. R., Sussman, M. R. & Markley, J. L. (2008). Metabolite identification via the Madison Metabolomics Consortium Database. Nature Biotechnology 26, 162. Degani, H., Laughlin, M., Campbell, S. & Shulman, R. G. (1985). Kinetics of creatine kinase in heart: a 31P NMR saturation- and inversion-transfer study. Biochemistry 24, 5510–6. Delgado, T. C., Castro, M. M., Geraldes, C. F. & Jones, J. G. (2004). Quantitation of erythrocyte pentose pathway flux with [2-(13)]Glucose and H-1 NMR analysis of the lactate methyl signal. Magnetic Resonance in Medicine 51, 1283–1286. Derome, A. (1987). Modern Nuclear Magnetic Resonance techniques for chemists. Pergamon. Dunn, W. B., Bailey, N. J. & Johnson, H. E. (2005). Measuring the metabolome: current analytical technologies. Analyst 130, 606–25. Ellis, D. I., Dunn, W. B., Griffin, J. L., Allwood, J. W. & Goodacre, R. (2007). Metabolic fingerprinting as a diagnostic tool. Pharmacogenomics 8(9), 1243–66. Ernst, R. R. & Anderson, W. A. (1966). Application of Fourier transform spectroscopy to magnetic resonance. Review of Scientific Instruments 37, 93–102. Goffeau, A., Barrell, B. G., Bussey, H., Davis, R. W., Dujon, B., Feldmann, H., Galibert, F., Hoheisel, J. D., Jacq, C., Johnston, M., Louis, E. J., Mewes, H. W., Murakami, Y., Philippsen, P., Tettelin, H. & Oliver, S. G. (1996). Life with 6000 genes. Science 274, 546–7. Golman, K., Ardenaer-Larsen, J. H., Petersson, J. S., Mansson, S. & Leunbach, I. (2003). Molecular imaging with endogenous substances. Proceedings of the National Academy of Sciences of the United States of America 100, 10435–10439. Gorter, C. J. & Broer, L. J. F. (1942). Negative result of an attempt to observe nuclear magnetic resonance in solids. Physica (The Hague) 9, 591. Griffin, J. L., Rae, C., Dixon, R. M., Radda, G. K. & Matthews, P. M. (1998) Excitatory amino acid synthesis in hypoxic brain slices: does alanine act as a substrate for glutamate production in hypoxia? Journal of Neurochemistry 71, 2477–86. Griffin, J. L., Lehtimaki, K. K., Valonen, P. K., Grohn, O. H., Kettunen, M. I., Yla-Herttuala, S., Pitkanen, A., Nicholson, J. K. & Kauppinen, R. A. (2003a). Assignment of 1H nuclear magnetic resonance visible polyunsaturated fatty acids in BT4C gliomas undergoing ganciclovir-thymidine kinase gene therapyinduced programmed cell death. Cancer Research 63, 3195–201. Griffin, J. L. & Nicholls, A. W. (2006). Metabolomics as a functional genomic tool for understanding lipid dysfunction in diabetes, obesity and related disorders. Pharmacogenomics 7, 1095–1107.

Griffin, J. L., Nicholls, A. W., Keun, H. C., Mortishire-Smith, R. J., Nicholson, J. K. & Kuehn, T. (2002). Metabolic profiling of rodent biological fluids via H-1 NMR spectroscopy using a 1 mm microlitre probe. Analyst 127, 582–584. Griffin, J. L., Pole, J. C., Nicholson, J. K. & Carmichael, P. L. (2003b). Cellular environment of metabolites and a metabonomic study of tamoxifen in endometrial cells using gradient high resolution magic angle spinning 1H NMR spectroscopy. Biochimica Biophysica Acta 1619, 151–8. Gulston, M. K., Rubtsov, D. V., Atherton, H. J., Clarke, K., Davies, K. E., Lilley, K. S. & Griffin, J. L. (2008). A combined metabolomic and proteomic investigation of the effects of a failure to express dystrophin in the mouse heart. Journal of Proteome Research 7, 2069–77. Haggie, P. M. & Brindle, K. M. (1999). Mitochondrial citrate synthase is immobilized in vivo. Journal of Biological Chemistry 274, 3941–3945. Harris, R. K., Becker, E. D., Cabral de Menezes, S. M., Goodfellow, R. & Granger, P. (2002). NMR Nomenclature: Nuclear Spin Properties and Conventions for Chemical Shifts. IUPAC Recommendations 2001. Solid State Nuclear Magnetic Resonance 22, 458–483. Herrgard, M. J., Swainston, N., Dobson, P., Dunn, W. B., Arga, K. Y., Arvas, M., Buthgen, N., Borger, S., Costenoble, R., Heinemann, M., Hucka, M., Le Novere, N., Li, P., Liebermeister, W., Mo, M. L., Oliverra, A. P., Petranovic, D., Pettifer, S., Simeonidis, E., Smallbone, K., Spasic, I., Weichart, D., Brent, R., Broomhead, D. S., Westerhoff, H. V., Kurdar, B., Penttila, M., Klipp, E., Palsson, B. O., Sauer, U., Oliver, S. G., Mendes, P., Nielsen, J. & Kell, D. B. (2008). A consensus yeast metabolic network reconstruction obtained from a community approach to systems biology. Nature Biotechnology 26, 1155–1160. Hore, P. (1995). Nuclear Magnetic Resonance. Oxford University Press, Oxford. Hore, P., Jones, J. & Wimperis, S. (2000). NMR: the toolkit. Oxford University Press, Oxford. Hoult, D. I., Busby, S. J., Gadian, D. G., Radda, G. K., Richards, R. E. & Seeley, P. J. (1974). Observation of tissue metabolites using 31P nuclear magnetic resonance. Nature Biotechnology 252, 285–7. Jacobson, B., Anderson, W. & Arnold, J. (1954). A proton magnetic resonance study of the hydration of deoxyribonucleic acid. Nature 173, 772–3. Jonas, D. M. (2003). Chemistry. Optical analogs of 2D NMR. Science 300, 1515–7. Kalic, M., Busselmann, G., Lauterwein, J. & Kamp, G. (1997). Phosphorous metabolites in boar spermatozoa. Identification of AMP by multinuclear magnetic resonance. Biochimica Biophysica Acta 1320, 208–16. Keeler, J. (2005). Understanding NMR spectroscopy. John Wiley and Sons Ltd. Keun, H. C., Beckonert, O., Griffin, J. L., Richter, C., Moskau, D., Lindon, J. C. & Nicholson, J. K. (2002). Cryogenic probe C-13 NMR spectroscopy of urine for metabonomic studies. Analytical Chemistry 74, 4588–4593. Kirschenlohr, H. L., Griffin, J. L., Clarke, S. C., Rhydwen, R., Grace, A. A., Schofield, P. M., Brindle, K. M. & Metcalfe, J. C. (2006). Proton NMR analysis of plasma is a weak predictor of coronary artery disease (vol 12, pg 705, 2006). Nature Medicine 12, 862–862. Krishnan, P., Kruger, N. J. & Ratcliffe, R. G. (2005). Metabolite fingerprinting and profiling in plants using NMR. Journal of Experimental Botany 56, 255–65. Kroll, J., Darmo, J., Dhillon, S. S., Marcadet, X., Calligaro, M., Sirtori, C. & Unterrainer, K. (2007). Phase-resolved measurements of stimulated emission in a laser. Nature 449, 698–701. Kumar, A., Welti, D. & Ernst, R. R. (1975). Nmr Fourier Zeugmatography. Journal of Magnetic Resonance 18, 69–83. Le Belle, J. E., Harris, N. G., Williams, S. R. & Bhakoo, K. K. (2002). A comparison of cell and tissue extraction techniques using high-resolution H-1-NMR spectroscopy. NMR in Biomedicine 15, 37–44. Le Mouel, J. L., Kossobokov, V. & Courtillot, V. (2005). On long-term variations of simple geomagnetic indices and slow changes in magnetospheric currents: The emergence of anthropogenic global warming after 1990? Earth and Planetary Science Letters 232, 273–286. Lenz, E. M., Bright, J., Wilson, I. D., Hughes, A., Morrisson, J., Lindberg, H. & Lockton, A. (2004). Metabonomics, dietary influences and cultural differences: a H-1 NMR-based study of urine samples obtained from healthy British and Swedish subjects. Journal of Pharmaceutical and Biomedical Analysis 36, 841–849. Levitt, M. (2001). Spin dynamics:basics of nuclear magnetic resonance. John Wiley and Sons Ltd. Li, W. (2006). Multidimensional HRMAS NMR: a platform for in vivo studies using intact bacterial cells. Analyst 131, 777–781. Lindon, J. C., Keun, H. C., Ebbels, T. M. D., Pearce, J. M. T., Holmes, E. & Nicholson, J. K. (2005). The Consortium for Metabonomic Toxicology (COMET): aims, activities and achievements. Pharmacogenomics 6, 691–699. Lindon, J. C., Nicholson, J. K., Holmes, E., Antti, H., Bollard, M. E., Keun, H., Beckonert, O., Ebbels, T. M., Reilly, M. D., Robertson, D., Stevens, G. J., Luke, P., Breau, A. P., Cantor, G. H., Bible, R. H., Niederhauser, U., Senn, H., Schlotterbeck, G., Sidelmann, U. G., Laursen, S. M., Tymiak, A., Car, B. D., Lehman-McKeeman, L., Colet, J. M., Loukaci, A. &

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society

18 Thomas, C. (2003). Contemporary issues in toxicology—The role of metabonomics in toxicology and its evaluation by the COMET project. Toxicology and Applied Pharmacology 187, 137–146. Liu, M., Nicholson, J. K. & Lindon, J. C. (1996). High-resolution diffusion and relaxation edited one- and two-dimensional 1H NMR spectroscopy of biological fluids. Analytical Chemistry 68, 3370–6. Lowe, I. (1959). Free induction decays of rotating solids. Physical Review Letters 2, 285–287. Malloy, C. R., Sherry, A. D. & Jeffrey, F. M. (1988). Evaluation of carbon flux and substrate selection through alternate pathways involving the citric acid cycle of the heart by 13C NMR spectroscopy. Journal of Biological Chemistry 263, 6964–71. Malloy, C. R., Sherry, A. D. & Jeffrey, F. M. (1990). Analysis of tricarboxylic acid cycle of the heart using 13C isotope isomers. American Journal of Physiology, 259, H987–95. Meyer, R. A., Brown, T. R. & Kushmerick, M. J. (1985). Phosphorous nuclear magnetic resonance of fast- and slow-twitch muscle. American Journal of Physiology 248, C278–87. Millis, K. K., Maas, W. E., Cory, D. G. & Singer, S. (1997). Gradient, highresolution, magic-angle spinning nuclear magnetic resonance spectroscopy of human adipocyte tissue. Magnetic Resonance in Medicine 38, 399–403. Nicholson, J. K., Buckingham, M. J. & Sadler, P. J. (1983). High resolution 1H NMR studies of vertebrate blood and plasma. Biochemistry Journal 211, 605–15. Odeblad, E., Bhar, B. N. & Lindstrom, ¨ G. (1956). Proton magnetic resonance of human red blood cells in heavy water exchange experiments. Archives of Biochemistry and Biophysics 63, 221–225. Pears, M. (2007). A metabolomic study of Batten disease. Cambridge University Thesis. Proctor, W. & Yu, F. (1950). The dependence of nuclear magnetic resonance frequency upon chemical compound. Physical Review 77, 717. Purcell, E. M., Torrey, H. C. & Pound, R. V. (1946). Resonance Absorption by Nuclear Magnetic Moments in a Solid. Physical Review 69, 37–38. Raamsdonk, L. M., Teusink, B., Broadhurst, D., Zhang, N. S., Hayes, A., Walsh, M. C., Berden, J. A., Brindle, K. M., Kell, D. B., Rowland, J. J., Westerhoff, H. V., van Dam, K. & Oliver, S. G. (2001). A functional genomics strategy that uses metabolome data to reveal the phenotype of silent mutations. Nature Biotechnology 19, 45–50. Rabi, I. I. (1937). Space Quantization in a Gyrating Magnetic Field. Physical Review 51, 652–654. Rabi, I. I., Zacharias, J. R., Millman, S. & Kusch, P. (1938). A New Method of Measuring Nuclear Magnetic Moment. Physical Review 53, 318–318. Reid, D. G., Maclachlan, L. K., Edwards, A. J., Hubbard, A. J. & Sweeney, P. J. (1997). Introduction to the NMR of proteins. Methods in Molecular Biology 60, 1–28. Sakellariou, D., Le Goff, G. & Jacquinot, J. F. (2007). High-resolution, highsensitivity NMR of nanolitre anisotropic samples by coil spinning. Nature 447, 694–7.

John H. F. Bothwell and Julian L. Griffin Sartoris, F. J., Bock, C. & Portner, H. O. (2003). Temperature-dependent pH regulation in eurythermal and stenothermal marine fish: an interspecies comparison using P-31-NMR. Journal of Thermal Biology 28, 363–371. Saunders, M., Wishnia, A. & Kirkwood, J. (1957). The nuclear magnetic resonance spectrum of ribonuclease. Journal of the American Chemical Society 79, 3289–90. Schroeder, M. A., Cochlin, L. E., Heather, L. C., Clarke, K., Radda, G. K. & Tyler, D. J. (2008). In vivo assessment of pyruvate dehydrogenase flux in the heart using hyperpolarized carbon-13 magnetic resonance. Proceedings of the National Academy of Sciences U S A 105, 12051–6. Shin, J., Lee, W. & Lee, W. (2008). Structural proteomics by NMR spectroscopy. Expert Reviews Proteomics 5(4), 589–601. Simpson, A. J., Song, G., Smith, E., Lam, B., Novotny, E. H. & Hayes, M. H. (2007). Unraveling the structural components of soil humin by use of solutionstate nuclear magnetic resonance spectroscopy. Environmental Science Technology 41, 876–83. Styles, P., Soffe, N. F., Scott, C. A., Cragg, D. A., Row, F., White, D. J. & White, P. C. J. (1984). A High-Resolution Nmr Probe in Which the Coil and Preamplifier Are Cooled with Liquid-Helium. Journal of Magnetic Resonance 60, 397–404. Suhling, K., French, P. M. & Phillips, D. (2005). Time-resolved fluorescence microscopy. Photochemistry Photobiology Science 4, 13–22. Teusink, B., Baganz, F., Westerhoff, H. V. & Oliver, S. G. (1998). Metabolic control analysis as a tool in the elucidation of the function of novel genes. Yeast Gene Analysis 26, 297–336. Velyvis, A., Yang, Y. R., Schachman, H. K. & Kay, L. E. (2007). A solution NMR study showing that active site ligands and nucleotides directly perturb the allosteric equilibrium in aspartate transcarbamoylase. Proceedings of the National Academy of Sciences of the United States of America 104, 8815–8820. Wind, R. A., Hu, J. Z. & Rommereim, D. N. (2003). High-resolution 1H NMR spectroscopy in a live mouse subjected to 1.5 Hz magic angle spinning. Magnetic Resonance in Medicine 50, 1113–9. Wishart, D. S., Tzur, D., Knox, C., Eisner, R., Guo, A. C., Young, N., Cheng, D., Jewell, K., Arndt, D., Sawhney, S., Fung, C., Nikolai, L., Lewis, M., Coutouly, M. A., Forsythe, I., Tang, P., Shrivastava, S., Jeroncic, K., Stothard, P., Amegbey, G., Block, D., Hau, D. D., Wagner, J., Miniaci, J., Clements, M., Gebremedhin, M., Guo, N., Zhang, Y., Duggan, G. E., Macinnis, G. D., Weljie, A. M., Dowlatabadi, R., Bamforth, F., Clive, D., Greiner, R., Li, L., Marrie, T., Sykes, B. D., Vogel, H. J. & Querengesser, L. (2007). HMDB: the human metabolome database. Nucleic Acids Research 35, D521–6. Wuthrick, K. (1990). ‘‘Protein structure determination in solution by NMR spectroscopy’’. Journal of Biological Chemistry 265(36), 22059–62. Yu, X., White, L. T., Alpert, N. M. & Lewandowski, E. D. (1996). Subcellular metabolite transport and carbon isotope kinetics in the intramyocardial glutamate pool. Biochemistry 35, 6963–8.

Biological Reviews (2010) 000–000 © 2010 The Authors. Biological Reviews © 2010 Cambridge Philosophical Society