Localized in vivo¹³C NMR spectroscopy of the brain

Abstract

Localized ¹³C NMR spectroscopy provides a new investigative tool for studying cerebral metabolism. The application of ¹³C NMR spectroscopy to living intact humans and animals presents the investigator with a number of unique challenges. This review provides in the first part a tutorial insight into the ingredients required for achieving a successful implementation of localized ¹³C NMR spectroscopy. The difficulties in establishing ¹³C NMR are the need for decoupling of the one-bond ¹³C–¹H heteronuclear J coupling, the large chemical shift range, the low sensitivity and the need for localization of the signals. The methodological consequences of these technical problems are discussed, particularly with respect to (a) RF front-end considerations, (b) localization methods, (c) the low sensitivity, and (d) quantification methods. Lastly, some achievements of in vivo localized ¹³C NMR spectroscopy of the brain are reviewed, such as: (a) the measurement of brain glutamine synthesis and the feasibility of quantifying glutamatergic action in the brain; (b) the demonstration of significant anaplerotic fluxes in the brain; (c) the demonstration of a highly regulated malate-aspartate shuttle in brain energy metabolism and isotope flux; (d) quantification of neuronal and glial energy metabolism; and (e) brain glycogen metabolism in hypoglycemia in rats and humans. We conclude that the unique and novel insights provided by ¹³C NMR spectroscopy have opened many new research areas that are likely to improve the understanding of brain carbohydrate metabolism in health and disease.

INTRODUCTION

Direct detection of ¹³C label provides a wealth of highly specific information on metabolites and metabolic rates,^1–7 such as the measurement of resolved carbon resonances of Glu and Gln in the brain,^8,9 and the reliable measurement of tissue glycogen^10–18 and tissue glucose content.^19–24 Most of these studies have involved the administration of a ¹³C-enriched precursor. When the enriched ¹³C label is transferred to molecules in the metabolic pathway, sensitivity can not only be increased, but important information on metabolic pathways can also be obtained. For example, recent studies showed that the information content of ¹³C NMR spectroscopy can be amplified considerably, such as the resolved observation of GABA labeling in the human brain, as well as the first detection of lactate labeling in normal human brain.⁸ In addition to its importance in assessing metabolism in intact brain, ¹³C NMR spectroscopy has become an important and useful tool in assessing compartmentation of metabolism in brain cells using extracts.^25,26

In addition to its low sensitivity, ¹³C NMR spectroscopy is methodologically more challenging than ¹H or even ³¹P NMR spectroscopy. However, ¹³C NMR provides its own unique insight and advantages over ¹H NMR spectroscopy. The uniqueness of ¹³C NMR stems mainly from its increased chemical shift dispersion, which can, for example, be used in two-dimensional NMR to increase spectral resolution following specific labeling of the protein.²⁷ As shall be elaborated further below, ¹³C NMR spectroscopy in living tissue provides a unique window on in vivo metabolism as it occurs. In the context of this review, we will consider mainly in vivo ¹³C NMR spectroscopy of the intact brain, which places specific requirements on the methodology. In contrast, the important research done using ¹³C NMR spectroscopy of cell cultures and suspensions ^28–30 requires methodology that is equivalent to that of in vitro work of body fluids and tissue extracts ^31–34 and to some extent isolated organs,³⁵ since these studies are often performed in high-resolution spectrometers with a test-tube-type setup, and usually localization is confined to everything that is detected by the RF coil. Most in vivo ¹³C NMR spectroscopy studies of intact organs have been performed using the surface coil as the only means to ‘localize’ the signals.^6,22,36–40 In contrast to the aforementioned cell culture and suspension studies and related important work, in vivo ¹³C NMR spectroscopy of intact organs faces a number of problems that make its application more challenging. Complete three-dimensional localization seems necessary in the brain to eliminate the intense triacylglycerol resonances from the scalp and other signals outside the brain, which adds to the challenges of in vivo ¹³C NMR. Localized ¹³C NMR spectroscopy thus is a novel investigative modality, the methodological requirements for which are the focus of the first part of this review.

LOCALIZED IN VIVO ¹³C NMR SPECTROSCOPY OF THE BRAIN. A BRIEF HISTORY IN TIME

Localized in vivo ¹³C NMR spectroscopy has undergone an impressive development in the past decade. A number of historical developments leading to the metamorphosis of this remarkable metabolic tool can be discerned. The goal of this section is to provide a historical perspective of the achievements of localized ¹³C NMR spectroscopy. The first application of ¹³C NMR spectroscopy to a living system was the metabolism of E. coli.²⁸ The first in vivo application of ¹³C NMR to the head was reported in 1986³⁷ and this early paper was admittedly struggling with many technical problems. Most notably were the limitations in localized shimming, which prohibited the separation of the glutamine from the glutamate resonances in vivo. In 1991–1992, it was shown for the first time that sufficient ¹³C-labeled glucose can be administered to humans to detect resonances from glutamate and glucose.^20,41,42 In retrospect, it is clear from the detection of intense lipid signals in these and other early in vivo spectra obtained from the head, that signals from outside the brain were dominant.^37,38,42,43 With the introduction of full three-dimensional localization to ¹³C NMR spectroscopy of the brain,^20,44 it rapidly became clear in 1991–1992 that the concentration of mobile lipids is generally too low to be detected in vivo in the normal brain (although this can be done in extracts⁴⁵). Therefore, the lipid resonances seen in ¹³C NMR spectra of the head must be attributed to extracerebral fat tissue, such as subcutaneous fat. The use of automated, localized shimming (i.e. in vivo optimization of the main static magnetic field, B₀, such that it becomes largely independent of the spatial coordinates⁴⁶) of all first- and second-order terms using FASTMAP dramatically improved sensitivity by narrowing linewidths in ¹³C NMR spectra.⁴⁷ These two methodological advances contributed to the then rather surprising observation that natural abundance signals from brain metabolites such as those from myo-inositol can be detected in vivo⁴⁴ and the discovery that labeling of glutamine can be detected in the brain in vivo.^47,48 The measurement of glutamine and glutamate turnover has been recognized as a window to study cerebral metabolic compartmentation.^49–53 A few years later it was demonstrated that the high demands of ¹H decoupling are not detrimental for the application of ¹³C NMR spectroscopy at higher fields⁵⁴ when using a novel RF coil design (see below for further discussion). At about the same time, it was also shown that three-dimensional localization based on the ¹H magnetization can be achieved in the brain in vivo.⁵⁵ Another, more recent development of in vivo ¹³C NMR spectroscopy was the demonstration that localized ¹³C NMR detection of glycogen can be achieved,⁵⁶ despite the relatively short relaxation times of the glycogen ¹³C resonances. This advance led to the first non-invasive detection of brain glycogen metabolism in the rat¹⁷ and signals from glycogen were recently quantified in the human brain as well,¹⁸ thereby opening a whole new field for investigating brain metabolism. These advances in NMR methodology and technology have led to further insights into brain metabolism and placed the ¹³C NMR method into the neuroscience theater to provide a unique in vivo window on the brain. ¹³C NMR spectroscopy holds promise to study a number of neuro-chemical events that are otherwise inaccessible by non-invasive means.

RELATIONSHIP TO OTHER NON-INVASIVE MODALITIES. WHAT MAKES IN VIVO ¹³C NMR UNIQUE

The administration of a tracer, whether stable or radioactive, and the ability to follow its metabolism in the brain provides neuroscientists with tools to study in vivo metabolism non-invasively. When using radiotracers, label in different metabolic pools cannot be distinguished, which has led to the use of non-metabolizable analogs, such as deoxy-glucose.

On the other hand, tracer studies can be performed using stable isotopes. The metabolism of stable isotopes can be followed non-invasively using NMR, although administration of the ‘tracer’ requires a higher isotopic enrichment, for ¹³C, typically above 50% in the precursor pool. Because NMR spectroscopy can be used to detect label in different molecules and different chemical positions, it offers the attractive possibility of following metabolism of the precursor, labeled at one or more specific positions. For example, glucose can be labeled with ¹³C at the C1 position and flow of the label into metabolic pools further downstream can be followed non-invasively. Applications are predominantly focused on nuclei where a stable isotope is present at low natural abundance, examples include ²H, ¹⁵N, ¹⁹F and ¹³C. The direct detection of the latter shall be dealt with in this review.

The dominant method to measure cerebral glucose consumption is by measuring the activity accumulated in the phosphorylation product of a glucose analog, e.g. the widely used autoradiography of deoxyglucose,⁵⁷ or the non-invasive fluoro-deoxy-glucose positron emission tomography.⁵⁸ In deriving the glucose metabolic rate from the uptake of a glucose analog, the kinetics of glucose transport is important and have been assessed in several studies.^59–61 In vivo NMR spectroscopy has the capability of providing a direct, localized measurement of brain glucose content non-invasively.^20,21,24

Of course, sensitivity for NMR is low compared with some radioactivity-based methods, and the relative sensitivity of ¹³C NMR is even lower. Nonetheless, despite the sensitivity disadvantage, as shall be discussed in the second half of this review, ¹³C NMR spectroscopy can provide unique insights into brain metabolism, albeit at a low spatial resolution. Many consider a low sensitivity and thus low spatial resolution a significant disadvantage. However, it is our belief that there are a significant number of biomedical problems/questions that can be uniquely addressed using in vivo ¹³C NMR spectroscopy.

Ideally, an investigative method applied to biomedical problems is fully developed and ‘mature’. Unfortunately, the development of ¹³C NMR spectroscopy in vivo has been limited to a handful of sites worldwide^8,43,51,62 and largely requires further development, as biomedical problems in general and neurochemical questions in particular develop. When considering these constraints on the application of this method, it is clear that methodological advances are probably required and expected when addressing important biomedical questions with ¹³C NMR spectroscopy.

TECHNICAL ISSUES. THE CRUX OF IN VIVO ¹³C NMR SPECTROSCOPY

The application of ¹³C NMR spectroscopy to the brain in vivo faces some challenges on the technical level that are unique to the ¹³C nucleus. The reasons for ¹³C NMR spectroscopy to be such a challenging modality stem mainly from three roots:

• low inherent sensitivity and low natural abundance;

• exogenous administration of expensive ¹³C labeled precursors;

• technically challenging methodology.

The first reason why ¹³C NMR spectroscopy is so challenging lies in its relatively low sensitivity compared with two other major nuclei used for in vivo NMR spectroscopy, namely ¹H and ³¹P. Although the inherent sensitivity is comparable to that of, for example, ²³Na [gyromagnetic ratio γ(²³Na) = 1.05γ (¹³C)], the problem of relative sensitivity is compounded by the fact that the ¹³C isotope has a natural abundance on the order of 1%. Hence in many cases, administration of exogenous ¹³C-enriched precursors is not only mandatory but also desired to obtain metabolic information. The administration of labeled precursors adds to the barrier, making a broad application of this method more difficult compared with ¹H or ³¹P NMR spectroscopy, as it increases the cost and experimental complexity. Even if the problem of low sensitivity can be alleviated somewhat by the administration of ¹³C-enriched substrate, considerable technical difficulties remain, rooted in the desire to improve the sensitivity. For instance, ¹H decoupling and accurate localization of the ¹³C NMR signals are necessary in vivo. This section deals with the last of the aforementioned three challenges, first by examining the reverberations of the requirement to decouple the spectrum during acquisition, second by discussing the localization requirements, and lastly by reviewing issues related to sensitivity and quantification.

RADIOFREQUENCY. THE LEGACY OF DECOUPLING

Optimization of sensitivity is critical for successful in vivo ¹³C NMR spectroscopy. To maximize the signal-to-noise ratio and spectral resolution in ¹³C NMR spectra, ¹H decoupling is generally applied during data acquisition, resulting in a simplified spectral pattern. The application of rather intense RF power during acquisition requires that the RF coil be capable of receiving the ¹³C NMR signal while transmitting ¹H RF power. This places several requirements on the RF console, RF coils and RF filters, all of which shall be discussed below.

RF coil design

The need for decoupling results in two requirements for RF coils: first, the ¹³C coil (operating at a frequency with a wavelength approximately four times that of the ¹H frequency) should not interfere with the RF profile of the ¹H coils. Second, the two RF circuits should be sufficiently isolated electrically. While it is in principle possible that surface coils with a resonance mode at the ¹H as well as at the ¹³C frequency can be designed, these designs typically result in a reduced RF efficiency at least at one of the two frequencies, if not both. This may have not been given much attention, since the performance of the ¹H circuit does not affect the sensitivity or the direct-detected ¹³C NMR experiment per se. However, a less than optimal ¹H coil design can result in vastly increased RF power deposition and increased local specific absorption rates (SAR).

Sensitivity requirements for ¹³C NMR have almost invariably led to the use of surface coils for detection.^36,63–65 The sensitivity of the experiment can be further optimized with B₁-insensitive pulses^44,65–67 to alleviate the drawbacks of an inhomogenous RF field for excitation. This discussion, therefore, will be focused on the use of surface coils for ¹³C NMR spectroscopy.

Typically, surface coils consist of an inductor combined with lumped capacitors. It is important to recognize the fact that, at higher magnetic fields, the impedance of the capacitor (1/iωC) is decreased at the ¹H frequency and approaches zero, which results in the RF of the ¹H coil being effectively blocked due to induced currents. The increased flux blockage at higher field was especially a problem for concentric ¹³C and ¹H coils, previously used at lower fields.^20,44 To overcome this problem, the ¹³C coil can be geometrically decoupled from the ¹H coil. Until 1996, several RF coil designs used a linearly polarized, figure-8-type geometry (‘butterfly design’) for the ¹H decoupling coil, where the proton decoupling RF field, B₂, is parallel to the ¹³C coil plane. An inherent shortcoming in all figure-8-type coil designs was the rapidly decreasing B₂ along the ¹³C coil axis, which required excessive ¹H RF power to decouple the entire volume of the ¹³C coil and thus led to excessive local SAR values close to the crossing point of the coil loops. At the time, it was considered impossible to perform broadband decoupled ¹³C NMR spectroscopy, especially at high fields such as 4 T.

Recognizing that it was highly desirable to use a quadrature decoupling coil that can reduce the RF power requirements up to two-fold, it was demonstrated that two surface coils can be used to generate a circularly polarized RF field for ¹H decoupling and imaging.⁵⁴ The resulting coil design (Fig. 1) was a combination of three basic transmit/receive coils; one was a linearly polarized ¹³C coil that minimally affects the B₂ field distribution of the quadrature ¹H RF field generated by the other two RF coils. The end result of this design was that it allowed efficient low-power decoupling at 4 and 9.4 T with negligible performance loss on either channel.⁵⁴

Figure 1.

Cross-sectional view of a half-volume ¹³C–¹H coil. From Adriany and Gruetter.⁵⁴ The ¹H coil consists of two surface coil loops with distributed capacitance. The geometric arrangement of the two coils in conjunction with a quadrature hybrid generates a circularly polarized RF field over the field of view of the smaller ¹³C surface coil, which is placed above the intersection of the two ¹H coils. The ¹³C coil overlaps partially with each of the ¹H coils, thereby minimizing the voltage induced by the ¹H coil in the ¹³C coil. The T₁-weighted MDEFT image¹⁸⁰ of a human head is shown to illustrate the excellent quality and relative homogeneity of the resulting ¹H RF field

Such a half-volume quadrature ¹H decouple coil design was found to be more efficient in terms of local SAR than previous coil designs for three reasons: first, the quadrature polarized coil has two-fold reduced power requirements for a given γB₁ than a linearly polarized coil. Second, to achieve an optimal quadrature field in the ¹³C coil sensitive volume, the two ¹H coil loops were placed as perpendicular as possible (Fig. 1). The increased distance between the wires and the tissue reduced the potential for so-called ‘hot spots’ of power deposition. Third, the resulting ¹H RF field had a very modest decrease along the y-axis. This novel coil design enabled broadband decoupling with 30 W peak power using WALTZ-16 at 4 T.⁵⁴ With typical duty cycles of 10% in vivo, this led to many applications being possible within FDA guidelines at 4 T. Not surprisingly, this coil design has been successfully adapted to ¹³C NMR spectroscopy of the human brain with corresponding decreases in power deposition over previous designs at 1.5⁶⁸ and 2.1 T⁶⁹ and has been used in many other studies.^70,71 The same coil design principle has been used for indirect detection of ¹³C label in humans,^62,70,72 where the larger ¹³C coils used for decoupling were driven in quadrature. The same coil design was used in in vivo studies of the rodent brain at 9.4 T,^17,56,73 for inverse detection in the rat brain at 9.4 T,⁷⁴ and with a larger ¹³C quadrature coil more recently also at 7 T in the rat.⁷⁵

Double-tuned volume coils have the advantage of providing coverage for the whole head and relatively uniform RF fields. Demands on RF power, filter performance and electrical isolation may critically increase when using volume coils due to the higher RF power required to generate a given γB₁ RF field. To avoid substantial coil flux coupling, the coils or the coil fields can be arranged in an orthogonal fashion.⁷⁶ Volume coils typically have a reduced sensitivity. These requirements can be substantially alleviated by using, for example, the TEM (transverse electromagnetic) resonator design, which allows the generation of quadrature polarized RF fields at both frequencies with minimal performance losses on either circuit.⁷⁷ Such coils have been used with great success in ¹H-detected ¹³C labeling studies of the human brain.^78,79

Console and RF filters

In addition to increased demands on an efficient RF coil design, the inherent requirement to apply RF at the ¹H and at the ¹³C frequencies demands that the spectrometer should be equipped with at least one additional broadband RF channel. This is a requirement that may not be easily compatible with the design of MRI systems for diagnostic purposes and hence may pose a direct tangible barrier for clinical research applications using ¹³C NMR spectroscopy.

Having the spectrometer capable of performing two-channel broadband NMR spectroscopy does not guarantee a successful application of ¹³C NMR spectroscopy. Even when using a good RF coil design, RF coils rarely provide sufficient isolation between the observe and the decouple RF channel to allow ¹H decoupling during signal detection. Therefore, additional isolation is required, which can be achieved by the use of RF filters. Two characteristics of RF filters must be considered, namely the need to effectively filter unwanted frequency bands, as well as the applicable RF power. In this context it is important to recognize that some filters contain iron cores, which are ferromagnetic and thus not useful close to the magnet. In terms of RF power requirements, it is important to state the required peak RF power under which the filter must perform and the allowable continuous wave power the filter must withstand. The recommendation is to overdesign the peak RF power a filter can handle to avoid the potential for filter breakdown.

Placement of the RF filter is equally important and in many cases site-specific (Fig. 2). For instance, it may be advantageous to place a bandpass filter at the input of the RF amplifier to eliminate potential broadband noise emanating from the modulator. In addition, it is typically necessary to place bandpass filters at the output of the RF amplifier, preferably after the RF has been passed into the Faraday shield. Bandpass filters typically have a higher insertion loss (1–2 dB) than low- or highpass filters and thus can in principle be used for the decouple channel, provided the RF power available is not limiting. On the detection side, however, it is desirable to minimize insertion losses, which can be achieved using low-pass filters in the ¹³C channel. When using volume coils for either channel demands on RF power, the requirements on filter performance and electrical isolation may critically increase.

Figure 2.

Placement of RF filters for direct-detected ¹³C NMR spectroscopy, adapted from Adriany and Gruetter.⁵⁴ The filters are designed to minimize ¹H RF breakthrough at the ¹³C (observe) channel. The ¹³C bandpass filter after the ¹³C pre-amplifier effectively renders the pre-amplifier narrow banded

Decoupling

Many ¹³C nuclei are directly bonded to protons. The resulting magnetic coupling between the ¹³C nucleus and ¹H nucleus results in a splitting of the ¹³C resonance into multiplets separated by J_CH Hz. The process of decoupling collapses the multiplets due to heteronuclear coupling into singlets, thereby simplifying the spectra and effectively increasing the sensitivity. Most ¹³C NMR applications have relied on WALTZ decoupling of the protons.⁸⁰ Recently, it has been demonstrated that using frequency-swept RF pulses as the basic building block, such as the hyperbolic secant pulse,⁸¹ dramatically increase the decoupling bandwidth. These decoupling methods feature only modest increases in peak RF power demand,^82–84 while maintaining a favorable ratio of cycling sideband intensity to center peak intensity.⁸⁴ We recently demonstrated that broadband adiabatic ¹³C decoupling was achieved at 9.4 T using such cycles at a moderate peak γB₁/2πof ~ 1 kHz.⁷⁴

Specific absorption rates (SAR). RF power is important

One potential risk to the MR examination is the fact that the RF power absorbed by the tissue is converted to heat, which can potentially lead to excessive heating of the body and local tissue, resulting in damage. It is therefore important to limit the RF power administered to the brain, especially for human studies.

To minimize the SAR requires that the pulse sequences and decoupling power are adjusted to operate at the minimum power threshold at which sequence performance is still acceptable. When using surface coils, optimization of RF power is most reliably achieved with an external reference sphere filled with a suitable ¹³C-labeled compound, such as formic acid, placed at the coil center. Such an external standard for RF power adjustment can also be an integral part of the external quantification procedure (see below). Highly efficient and accurate methods for adjusting the decoupling power have been described.^85,86

When measuring RF power it is of course important to recognize that RF power lost in the cables, T/R switches and filters will not reach the subject. Hence these factors should be considered when calibrating the SAR monitor. A practical approach is to measure the RF power at the coil port with the spectrometer configuration set to what will be used in the experiments. Additional factors that will reduce the amount of RF power delivered to the subject are: (i) the coil efficiency (which can be measured by comparing the loaded vs the unloaded coil Q factor); (ii) the power reflected by the RF coil (caused by, for example, amplified RF noise transmitted outside the bandwidth of the RF coil, acting as a bandpass filter); and (iii) radiation losses, which are increasingly important with increased RF frequency, i.e. B₀. In general the majority of the SAR is generated by the decoupling.

The Center for Devices and Radiological Health of the United States Food and Drug Administration has stated that studies exceeding SAR of 4 W/kg averaged over 15 min over the entire body or 3 W/kg averaged over the head averaged over 10 min or 8 W/kg in any gram of the head must be considered studies of significant risk (www.fda.gov/cdrh/ode/guidance/793.html).

The first criterion for significant risk is rarely exceeded in human studies using head volume coils or surface coils and thus is less important. The second criterion is also rarely exceeded in studies using surface coils, leaving the last criterion as the most difficult to comply with.

LOCALIZATION. THE CHEMICAL SHIFT DISPLACEMENT ARTIFACT

The full chemical shift range of most biologically interesting ¹³C resonances is approximately 160–200 ppm wide. Even when considering the four-fold lower gyromagnetic ratio than ¹H, this is a range that (in Hz) exceeds the range for ¹H resonances of main biological interest (7 ppm) by at least a factor of six. Thus, chemical shift displacement error problems are important factors to consider when wishing to localize ¹³C NMR signals.

Is it necessary to localize NMR signals? For most applications in the brain, the answer to this question must be emphatically yes. In ¹H NMR spectroscopy, water suppression and the intense lipid signal from subcutaneous fat cause dangerous artifacts that are likely to obscure and degrade the signal from the brain. In ³¹P NMR spectroscopy, for example, it is well known that ATP and PCr concentrations in extracerebral muscle (such as the signals from the temporalis muscle) are vastly different than those in the brain and thus can lead to inaccurate interpretation. In ¹³C NMR spectroscopy, water suppression is obviously not an issue and for some resonances the lipid signals may not be a significant problem. Non-localized ¹³C NMR has found important applications, especially when considering the need for signal (which undoubtedly is reduced when using localization methods) and experimental simplicity.^39,87–89 However, for some resonances, such as glucose, lactate, glutamate and especially glycogen, the concentration in extracerebral tissue can be significant.

Therefore, localization is important for ¹³C NMR spectroscopy of the brain, because the frequently employed technique of subtracting, for example, a pre-infusion natural abundance ¹³C spectrum has several drawbacks. Among the potential problems of subtracting a pre-infusion spectrum are the reduced sensitivity, increased measuring time, the susceptibility to motion artifacts and small changes in linewidth over the rather long measuring times.

Spatial localization does not depend on chemical shift when using the one-dimensional surface-spoiling gradient^63,90 or spectroscopic imaging,^91,92 which is an elegant technique to solve the chemical shift displacement problem when many transients can be acquired.⁹³ The low spatial resolution may require dedicated solutions to minimize Gibbs ringing from superficial fat signals, and several approaches have been described.^94,95

Nevertheless, one of the challenges of localized ¹³C NMR is the localization error induced with slice selection techniques due to the demanding ¹³C chemical shift range. In the following, we will concentrate on the various methods for full three-dimensional localization of ¹³C NMR signals using gradient-based methods.

Direct localization using slice-selection

The use of direct localization of ¹³C has been discounted on the basis of the large chemical shift requirement leading in general to increased demand on the RF power through the need for an increased RF bandwidth. In some cases, even when the demands on the RF bandwidth can be met, the required gradient strength may very well be limiting, because of the lower gyromagnetic ratio γ. The following example illustrates this case: true localization of the entire ¹³C chemical shift range including the carboxyl resonances at around 180 ppm and the lactate methyl resonance at 20 ppm requires a range of 160 ppm to be covered, equivalent to a frequency spread of ~1700 Hz/T B₀ field. To maintain the chemical shift displacement error below 10% of the voxel dimension, the bandwidth of the RF pulse thus has to be 17000 Hz/T. When localizing a slice of, for example, 4 cm, the gradient strength must be ~4250 Hz/cm, which corresponds to a gradient strength of 40 mT/m per Tesla B₀, because the γ of ¹³C is four-fold smaller than that of ¹H. For low magnetic fields such as 1.5 T, a gradient strength of 60 mT/m is required and this is pushing the envelope of what is currently possible, even more so at higher B₀.

However, when contemplating the large chemical shift range of ¹³C, it is worth considering that the chemical shift range that needs to be covered for specific applications may be much smaller. Such reduced chemical shift dispersion is summarized for several metabolites in Table 1. For example, all the inositol resonances are observed within 3 ppm⁹⁶ and it is obvious that the chemical shift displacement error for myo-inositol is small, even when using direct localization. The first three-dimensional localization of ¹³C NMR signals was achieved using ISIS localization on the ¹³C z-magnetization,⁴⁸ used for the first direct measurement of brain glucose C1 in the human brain.²⁰ In addition, directly localized ¹³C NMR spectroscopy was used for the first detection of the natural abundance signals of small molecules, as illustrated with myo-inositol.⁴⁴

Table 1.

Chemical shift ranges for ¹³C MRS per Tesla (expressed in Hz/T) including the range of the corresponding coupled ¹H resonances

Compound class	Chemical shift range (ppm)	13C Frequency range (Hz/T)	1H frequency range (Hz/T)
Glycogen/glucose (all)	101–61	430	94
Myo-inositol (all)	76–72	43	34
Amino acids (CHn)	56–22	363	85
Amino acids (CH)	56–53	32	20
Lipids (CHn)	131–14	1248	196

Polarization transfer

Localization on the ¹³C magnetization using ISIS must be considered a valid localization method for a restricted spectral region. However, when using the ¹³C longitudinal magnetization for localization, signal enhancements are commonly achieved using the nuclear Overhauser effect (NOE), which has a theoretical upper limit of a three-fold enhancement of the ¹³C signals. However, for resonances with sufficiently long T₂, such as those of most brain metabolites as judged from their narrow line widths of a few Hz,⁴⁷ it is feasible to use polarization transfer to recover the maximal sensitivity gain of a fourfold enhancement and to localize on the ¹H magnetization, thereby greatly reducing the chemical shift displacement error due to the much smaller chemical shift range (Table 1). Heteronuclear polarization transfer combined with localization on the proton magnetization, as proposed earlier,⁹⁷ was shown to minimize the chemical shift displacement error in ¹³C MRS of the human brain to a level beyond concern even at 4 T.⁵⁵ To minimize the number of pulses needed for generating the in-phase ¹³C signal enhancement and to minimize phase distortions in the spectrum, distortionless enhanced polarization transfer, DEPT,⁹⁸ was used. Another alternative was INEPT,^69,99 which required more RF pulses compared with DEPT, having more potential for signal loss in inhomogeneous RF fields. Because of the number of RF pulses used, polarization transfer sequences tend to be more sensitive to B₂ inhomogeneities such as those present when using surface coils. However, surface coils were used to optimize sensitivity as in almost all ¹³C studies, and the polarization transfer sequence had to be carefully optimized for the volume of interest, which was shown to be possible when using localized pre-calibrations of decoupler and transmitter RF power.⁵⁵ As a result, a high sensitivity can nevertheless be achieved, especially when the VOI dimensions are small enough that RF variation across the VOI can be neglected.^54,55 While excellent sensitivity was demonstrated with the localized detection of natural abundance myo-inositol, scyllo-inositol, glutamate, taurine, glucose, glutamine and NAA signals,^21,54,55,96 further gains were expected when incorporating adiabatic pulses into the pulsed polarization transfer sequence, which was shown to improve reliability and sensitivity.^8,100 When performing direct-detected ¹³C NMR spectroscopy, the ¹³C coil is typically smaller than the ¹H coils and thus the ¹³C part of the polarization transfer is more prone to signal loss. Replacing the standard 90–τ–180–τ section of the ¹³C channel in DEPT with a segmented BIR-4¹⁰¹ pulse resulted in greatly minimized effects of RF inhomogeneity on the acquired signal.^8,102 Such a sequence is shown in Fig. 3.

Figure 3.

Localization using a semi-adiabatic DEPT sequence including coherence elimination by gradient dephasing. From Henry et al.¹⁰² Localization is performed on the ¹H z-magnetization using ISIS, complemented with outer volume suppression (OVS). The ¹H part of the coherence generation was achieved using standard hard pulses, with the last pulse flip angle set to 45° to allow for the simultaneous detection of all CH_n carbons. The two 90°–τ–180°–τ sequence was replaced by a segmented 0° BIR-4 pulse, rendering the performance of the sequence much less susceptible to the spatial variation of the ¹³C RF field, especially when using surface coils. The spoiling gradient (spoil) dephases unwanted coherences excited by the ¹H pulses when they deviate from their nominal flip angles indicated. The delay τ is determined by the heteronuclear J coupling, J_CH, which ranges in vivo from 127 to 167 Hz⁷⁴

The broadband localization achievable with polarization transfer methods has resulted in localization of all amino acid resonances in the human brain [Fig. 4(A)] as well as in efficient broadband localization in the rat brain for the first time [Fig. 4(B)], indicating complete elimination of extracerebral lipid signals.

Figure 4.

Examples of direct-detected ¹³C NMR spectroscopy from the brain. (A) ¹³C NMR detection of label incorporation into mostly cytosolic amino acids at 4 T, from Gruetter et al.⁹ Shown is a representative spectrum obtained from a 45 ml volume in the human visual cortex during an infusion of 67%-enriched [1-¹³C]glucose. In addition, resonances resulting from homonuclear ¹³C–¹³C coupling were readily detected at the positions of all glutamate resonances (indicated by the brackets). Processing consisted of a mild Lorentz–Gauss apodization (3 Hz) and the spectrum is shown without baseline correction. (B) In vivo ¹³C NMR spectra from a 400 μl volume in the rat brain, acquired using the modified DEPT sequence depicted in Fig. 3 during an infusion of 70%-enriched [1,6-¹³C₂]glucose, from Henry et al.¹⁰² Processing consisted of zero-filling, 2 Hz Lorentzian-to-Gaussian resolution enhancement and fast Fourier transform. No baseline correction was applied. Note the complete absence of the lipid signals over the entire spectral range. Resonance assignments are as follows: Glu C2 at 55.6 ppm; Gln C2 at 55.0 ppm; NAA C2 at 54.0 ppm; Asp C2 at 53.7 ppm; NAA C3 at 40.5 ppm; GABA C4 at 40.45 ppm; Asp C3 at 37.6 ppm; GABA C2 at 35.3 ppm; Glu C4 at 34.2 ppm; Gln C4 at 31.7 ppm; Glu C3 at 28.0 ppm; Gln C3 at 27.7 ppm

Three-dimensional localization of the ¹³C NMR signals of glycogen

Clearly, the chemical shift displacement error can be minimized by using the ¹H z-magnetization for localizing the signals for ¹³C NMR detection. However, methods such as those discussed in the previous section are not applicable in situations where T₂ is very short. Unfortunately, for some compounds such as glycogen, the T₁ and T₂ are relatively short. This is especially true for T₂ that is short relative to the delays required to generate polarization transfer, resulting in substantial signal loss. Therefore, it is impossible to realize the full sensitivity gains of the polarization transfer technique. Nevertheless, it has been pointed out that in muscle the signal gains with a frequency-tailored INEPT sequence are comparable to what can be achieved by proton decoupling and NOE generation (SINEPT).^15,103 While the SINEPT technique relies on performing the polarization transfer on-resonance and without refocusing the J evolution, thereby minimizing the T₂-losses, the method remains in principle susceptible to changes in T₂, which have been reported to occur depending on the glycogen molecule size and temperature.¹⁰⁴ Unfortunately, the potential four-fold signal gain is reduced by the short T₂ of ¹H and ¹³C signals on the order of 5 ms.^15,104,105 Given the uncertainty of potential changes in transverse relaxation times,¹⁰⁴ a robust method that is capable of achieving three-dimensional localization of the glycogen signal without transverse coherence generation was considered advantageous.

The method developed for the localized measurement of glycogen relies on outer-volume suppression, achieved with slice-selective inversion recovery nulling of the magnetization in two slices parallel to B₀ and perpendicular to the RF coil (Fig. 5) in conjunction with nominal 90° pulses applied along the six slices parallel to the volume-of-interest. In the brain (see below) such localization was crucial to minimize potential contamination of the detected signal with signals from extraneous muscle, especially in the rat. The performance of the sequence was validated from the post-mortem rapid elimination of glucose and glycogen in the brain, which is rapid compared with muscle [Fig. 6(A)], as well as from the elimination of subcutaneous lipid signals, as shown for the human head in Fig. 6(B).

Figure 5.

Localization of ¹³C NMR spectroscopy using OVS only, from Choi et al.⁵⁶ The localization of the magnetization starts with two adiabatic pulses that invert the z-magnetization in slabs along x adjacent to the voxel. This inverted z-magnetization approaches zero during the delay TI and when it is approximately minimized, a standard OVS sequence along all three dimensions with nominal 90° flip angles applied. Just before the adiabatic excitation pulse, an optional inversion pulse is applied on alternate scans, which together with the concomitant y-gradient and appropriate phase cycling selects a slice along y (parallel to the ¹³C coil plane) as in one-dimensional ISIS. Reproduced with permission from Mag. Reson. Med. Copyright © 2000 John Wiley & Sons, Ltd.

Figure 6.

Validation of the localization of ¹³C NMR signals using OVS (Fig. 5). (A) Twenty minutes after termination by KCl injection, glucose and glycogen resonances from extracerebral tissue are still observed in the post-mortem rat (top), whereas upon application of the three-dimensional localization method these signals were reduced to the noise level (bottom trace). Reproduced with permission from Choi et al.¹⁷ (B) In the human brain, where such post-mortem studies are not applicable, the efficiency of the adapted sequence was verified from the more than 100-fold suppression of the superficial lipid signals. From Oz et al.¹⁸ Copyright © 2003, with permission from Elsevier

SENSITIVITY. UNEARTHING TINY SIGNALS

Because the detection of ¹³C label is inherently insensitive, generally every effort is made to improve the sensitivity. In this section, we focus on the sensitivity gains achieved with increased magnetic field B₀, direct vs. indirect detection, choice of RF coil and shimming.

Effect of higher magnetic fields, B₀

It is well-known that increasing the static field B₀ increases the sensitivity of NMR detection.¹⁰⁶ The increase in sensitivity for non-loading samples has been reported as being an exponential function of B₀, which can be written as

$$S/N\hspace{0.17em}\alpha \hspace{0.17em}{B}_0^{\beta }$$ (1)

From theoretical considerations, it is clear that the signal S increases with B₀², hence β must be 2 or lower, amounting to an upper limit for the gains in sensitivity achievable due to increases in B₀ alone. Even when the noise present in the RF coil is the dominant source of noise, its root-mean-square amplitude may increase with frequency (B₀) and it is generally accepted that in this case β = 1.75. For in vivo spectroscopy, however, the sample is the most dominant source of noise and thus the noise detection efficiency increases with frequency and thus the sensitivity increase with B₀ is closer to linearity with β~1. Because the sample may not be the dominant source of noise, it is plausible that in vivo β may lie somewhere between 1 and 1.75. In practice β is probably closer to the former. In this discussion it was assumed that resonance linewidths and relaxation times are very similar between the different field strengths. An increase in resonance linewidth can offset some of the sensitivity gains, as can an increase in longitudinal relaxation time, although the latter effect influences sensitivity only with the square-root of the T₁. Lastly, increased spectral resolution can result in increased sensitivity due to the decreased signal overlap. Spectral overlap may in general not be a problem for direct-detected ¹³C NMR spectroscopy, but resolution of the spectral region of the C2 of amino acids and the overlap between GABA C2 and glutamate C4 stand to gain from increased spectral resolution. In practice, it is very difficult to determine the precise value of β, but it is possible to state the sensitivity gains achieved over previously published results.

Some studies have suggested that the sensitivity gains at lower frequencies may be supralinear,¹⁰⁷ indicating that the RF coil contributes significantly to the noise. In this case, the use of superconducting RF coils may potentially yield additional sensitivity gains,¹⁰⁸ as is increasingly being realized for high-resolution NMR of solutions. It is of interest to note that polarization transfer from hyperpolarized compounds has been recently used with specific compounds to image blood flow.¹⁰⁹ This points to the potential utility of hyperpolarization for ¹³C to enhance sensitivity, although how this will be achievable is at present not clear.

Direct vs indirect detection

A study of the sensitivity of ¹³C NMR spectroscopy in vivo would be incomplete without pointing out the two principal detection methods available, namely direct and indirect detection of ¹³C label. Direct detection of the ¹³C signals takes advantage of the large chemical shift dispersion of the ¹³C nucleus. In addition, the spin systems of the ¹³C nucleus are simpler, provided ¹H decoupling is applied.⁵⁵ However, the increased specificity of direct detection comes at the price of a substantial loss of sensitivity.⁷² For example, the signal-to-noise ratio from an equimolar concentration of ¹³C at a given static field B₀ is much lower for ¹³C than the corresponding ¹H signal in vivo. In theory the signal is γ³ stronger for ¹H (64-fold, which in the literature is often confused with the gain in sensitivity). However, under conditions of full sample loading (noise ~γ) this is reduced to a γ² (16-fold) sensitivity gain. Polarization transfer reduces the relative sensitivity gain to γ. In many cases, such as glutamate, glutamine and aspartate, the resonances in the ¹H spectrum are coupled to other ¹H, resulting in often complex resonances of appreciable width on the order of 20 Hz, as is the case for the γ-proton of glutamate. In contrast, the ¹³C resonances have linewidths on the order of a few Hz,⁴⁷ and this can translate into a further reduction of the estimated sensitivity gain of approximately two-fold. In addition, the multiplicity of the CH_n groups enhances the sensitivity of ¹H detection. Thus the sensitivity gain is estimated at ~2n:1. These estimated sensitivity gains may in some cases be offset by the limited spectral resolution for protons. Given these real advantages in sensitivity, it is attractive to exploit the fact that ¹H nuclei in close chemical proximity to a ¹³C nucleus are coupled through the heteronuclear J coupling. With appropriate ‘spin gymnastics’ this coupling can be exploited to extract the signal from these protons by eliminating the signal from protons that are not coupled to ¹³C.^{74,78,110–113} The ensuing ‘¹³C-edited’ ¹H spectrum stands to benefit from the improved sensitivity of the proton. It also stands to suffer from a major shortcoming of ¹H NMR spectroscopy, namely the limited spectral dispersion, as the corresponding chemical shifts in biomedical applications reside in a 10-fold reduced bandwidth compared with carbon, and additional efforts such as two-dimensional spectroscopy may be required to achieve adequate spectral resolution.⁹⁹

The spectral resolution in ¹H NMR spectroscopy is especially limiting at low magnetic fields, such as the widely deployed magnetic field of 1.5 T, which does not allow the accurate separation of the resonances of glutamate and glutamine. When increasing the magnetic field some of these limitations can be overcome due to improved spectral dispersion,¹¹⁴ allowing the detection of resolved signals from ¹³C labeled glutamate and glutamine by ¹H NMR spectroscopy.⁷⁴ Figure 7 shows that in ¹H NMR spectra the resolution of a number of resonances is limited even at 9.4 T, when comparing with the resolution in the ¹³C NMR spectrum [Fig. 4(B)]. A recent study at 7 T recently confirmed that the spectral resolution decreased with lower field in the ¹H spectrum compared with direct-detected ¹³C NMR.⁷⁵ Nonetheless, the C2 resonances of glutamate and glutamine are not easily separated in the ¹H NMR spectra (Fig. 7), whereas at 4 T (Fig. 4),^8,9 and to some extent at 1.5 and 2.1 T, the C2 of glutamate and glutamine are readily separated in direct-detected ¹³C NMR spectra.^43,47,69 In addition, direct detection offers the potential to detect ¹³C–¹³C couplings (Fig. 4),^56,102,115 containing important metabolic information.^116,117

Figure 7.

Illustration of the information content achievable in vivo by ¹H-detected ¹³C NMR at 9.4 T. The ¹H NMR spectrum was obtained from a 130 μl volume in the rat brain in the first 1 h of glucose infusion showing resonances coupled to ¹³C only, from Pfeuffer et al.⁷⁴ The improved sensitivity allowed the detection of label incorporation into alanine C3 (Ala). Natural abundance signal is detected for creatine (Cr_tot) and NAA. Reprinted from Magn. Reson. Med. Copyright © 1999 John Wiley & Sons, Ltd.

RF coil sensitivity considerations

When using surface coils it is advantageous to use adiabatic RF pulses to overcome the problems arising from inhomogenous RF fields, which have been used in a number of studies for ¹³C NMR spectroscopy.^{8,17,18,20,44,65,66,69,102,118} In principle it may be desirable to have a homogenous RF field covering the entire brain, as is being widely used for head imaging on conventional MR scanners. When compared with surface coils, the volume coils require increased RF power since the RF power is distributed into a larger volume.¹⁰⁶ Increased RF power implies decreased sensitivity from the reciprocity principle. It is not surprising that volume coils typically are two to three times less sensitive than surface coils with optimized dimensions, even when considering the most efficient coil design. This sensitivity loss can be partially compensated by imaging the ¹³C label indirectly,^79,99,119 which may require simplified models of brain glutamate metabolism (see below).

Shimming

Even though the effect of B₀ inhomogeneity on the linewidth of a ¹³C resonance is reduced four-fold compared with that of ¹H, shimming remains an important issue: the low sensitivity of ¹³C NMR typically precludes the measurement of signals from small volumes, leading to comparatively large volumes. The measurement of substantially larger volumes probably results in the need for adjusting the currents in the second-order shim coils, which can be achieved using quantitative shim methods.^47,120,121 The need to adjust second-order shim coils can be appreciated from the fact that the spatial distribution of the B₀ field of second-order shim coils results in significant signal intensity being distributed in the wings of the resonance (Fig. 8). Such signal distribution can be easily missed in peak integration or even peak fitting at low signal-to-noise ratios, which are typical for in vivo ¹³C NMR spectroscopy, or when using even modest resolution enhancement. Shimming with second-order shim coils typically has only a modest effect on the full-width at half-maximum of the resonance. Nonetheless, a significant fraction of the total signal in the wings is shifted under the main resonance, thereby reducing the potential for quantification errors and increasing the sensitivity of the experiment further. These effects are expected to be of increased importance with increased B₀. In summary, it will be important to adjust second-order shim coils to harness the full sensitivity gains at higher field.

Figure 8.

Effect of shimming on lineshape and width. Shown is the effect of a second-order shim coil (yz) on the field distribution in a cubic volume. Upon elimination of this term (by shimming), the intensity in the wings is moved underneath the central peak indicated by the arrows, thereby increasing sensitivity and reducing potential quantification errors

QUANTIFICATION METHODS FOR IN VIVO ¹³C NMR SPECTROSCOPY

This section focuses on the methods available to calibrate the in vivo signal intensity. How to measure the signal intensity using peak fitting, integration or other computationally even more involved methods is not part of this review. The reader may be interested in a companion article in this issue which considers issues of peak intensity measurement.¹²²

Internal reference methods

Internal reference methods have become the method of choice for quantification of ¹H NMR spectroscopy. These methods use either referencing to the signal of a compound measured in the spectrum, such as creatine, or referencing to tissue water. These methods are not easily applicable for ¹³C NMR spectroscopy. The main reason is that ¹³C NMR spectra of the brain typically lack a signal from a natural abundance compound that is present in high concentration, although it appears that, especially at high fields and longer acquisition times, the signal of some brain metabolites might be sufficient, this has been used by some investigators. However, these signals are on the order of 0.1–0.2 μmol/g in ¹³C concentration and the signal-to-noise ratio of such reference measurements is typically low, especially at low fields. Errors in the reference measurement are expected to propagate into the quantification of the ¹³C label in the ¹³C NMR spectrum.

External reference methods

Because of the complexity of the ¹³C NMR measurements and the frequent use of the surface coil as transceiver, the external reference method has become an important tool to quantify ¹³C label in the brain. The method is based on the idea of repeating the identical NMR experiment (under identical experimental conditions as in vivo) in a phantom containing an aqueous solution of a reference compound (typically identical to that being measured, such as glucose, glycogen, etc.) of known concentration. The principle of the external reference method is illustrated in the scheme in Fig. 9. The RF power is adjusted based on the reference signal provided at the ¹³C coil (see above) and corrections for the effect of coil loading on the signal and of differential relaxation effects are measured and taken into account in the quantification. A major advantage of this method is that, by repeating the reference experiment under identical conditions (i.e. applying RF at the identical chemical shift with an identical γB₁ at the coil center), some imperfections are inherently corrected, such as off-resonance effects in the sequence and effects of inhomogenous B₁.

Figure 9.

Scheme of the external reference method. The in vivo experiment is scaled by the reference intensity from, for example, 99% ¹³C-formic acid (FA) placed at the ¹³C coil center to correct for differences of the effect of sample loading between the in vivo and the reference experiment. The signals are further corrected by the correction factor CF that takes into account the relaxation effects on the signal (T₁, T₂ and NOE) in vivo, CF^{in vivo}, and in the phantom, CF^ref, to yield the corrected signal intensity I_corr. From the known concentration in the phantom the in vivo concentration can be determined, as follows: $\left[C^{\text{in\hspace{0.17em}vivo}}\right]=\left[C^{\text{ref}}\right]\frac{I^{\text{in\hspace{0.17em}vivo}}{\text{CF}}^{\text{in\hspace{0.17em}vivo}}}{{\text{FA}}^{\text{in\hspace{0.17em}vivo}}}\frac{{\text{FA}}^{\text{ref}}}{I^{\text{ref }}{\text{CF}}^{{}^{\text{ref}}}}$

Quantification of brain metabolites using natural abundance ¹³C NMR of the brain

Studies using natural abundance ¹³C NMR certainly are not expected to compete with ¹H NMR measurements of the respective compound; however, they prove invaluable in estimating concentrations, which can aid the quantification of the ¹H NMR spectrum.^88,123 For example, the difficulty in detecting the natural abundance signal of glutathione (GSH) indicates a concentration below 3 μmol/g in agreement with ¹H NMR quantification¹²⁴ and, similarly, the concentrations of taurine, aspartate and GABA are expected to be below 2 μmol/g.

From the sensitivity and the quantification of the localized ¹³C NMR signals of glucose in the human brain,²⁰ it was suggested that signals from natural abundance myo-inositol should be detectable at 2.1 T. At the time, the only compounds measured by natural abundance ¹³C NMR were glycogen in muscle ¹²⁵ and liver,¹¹ creatine in skeletal muscle¹⁴ and subcutaneous lipid signals, all of which are present in quantities of tens of μmol/g or above. Indeed, it was shown early on that quantification of natural abundance myo-inositol was possible.⁴⁴ Improvements in sensitivity and magnetic field strength have permitted the localized quantification of glucose,²¹ as well as glutamine, glutamate, N-acetyl-aspartate (Fig. 10 ) and elevated scyllo-inositol.⁹⁶

Figure 10.

Localized ¹³C NMR detection of natural abundance resonances in the human brain. The glucose resonances detected during hyperglycemia are indicated by the vertical dashed lines and identified by the comparison with the glucose phantom (bottom trace). In addition to glucose and myo-inositol, resonances from glutamate, glutamine and N-acetyl-aspartate were also discernible. From Gruetter et al.²¹

APPLICATIONS: FINDINGS FROM LOCALIZED ¹³C GLUCOSE LABELING STUDIES OF THE BRAIN

The detection of signal from natural abundance compounds in the brain certainly represents a useful complement to in vivo quantification by ¹H NMR spectroscopy, albeit of limited practical value. The real power of ¹³C NMR spectroscopy is more obvious when considering the administration of ¹³C labeled precursors such as glucose to follow the redistribution of the label in metabolic products of glucose consumption. Localized ¹³C NMR spectroscopy to date has provided insights into brain metabolism in vivo. In addition to measuring the turnover of glutamate and glutamine, other observations provided by localized ¹³C NMR spectroscopy include: (i) the demonstration of reversible Michaelis–Menten kinetics of glucose transport in human brain, and that a brain glucose close to zero is the point where cerebral blood flow increases during hypoglycemia; (ii) the measurement of brain glycogen metabolism during hypoglycemia in animals, demonstrating that brain glycogen is a significant source of fuel during hypoglycemia; (iii) the detection of very slow brain glycogen metabolism in the human brain; (iv) the revelation that pyruvate carboxylase flux (anaplerosis) is substantial in the in vivo human brain; (v) the demonstration that glial metabolism is significant at rest and mostly oxidative in vivo; (vi) the measurement of the malate-aspartate shuttle flux as a major controlling step in brain oxidative metabolism and isotope flux; and (vii) the observation that glutamate metabolism is affected by physiological focal stimulation in human brain. In addition to these studies, localized ¹³C NMR spectroscopy was used to detect in vivo GABA labeling as well as significant lactate metabolism in the resting, healthy brain.^8,74

These observations make it clear that localized ¹³C NMR spectroscopy provides a unique window on in vivo brain metabolism, with a chemical specificity and diversity of potential measurements not possible by other methods. Some of these achievements shall be highlighted below.

GLUCOSE TRANSPORT

In principle, glucose concentrations can be measured using ¹H NMR spectroscopy.^21,126 However, for a CH group with homonuclear J-coupling, the sensitivity advantage compared to ¹³C is expected to be only two- to three-fold as discussed above. The proximity of the ¹H resonance of glucose H1 to water adds to the difficulty in measuring brain glucose using ¹H NMR spectroscopy.¹²⁷ In cases where metabolism is followed by ¹³C NMR, it may be advantageous to measure brain glucose content as well. Unless labeled glucose is injected directly into the brain, the administered ¹³C label must cross the blood–brain barrier before it can be metabolized by the brain cells. Aside from lactate and possibly glycogen,¹⁷ brain glucose is the only sizable kinetic pool that is capable of influencing the labeling kinetics of pyruvate and thus ultimately of acetyl-CoA. Therefore, precise knowledge of the size of the brain glucose pool and its physical distribution space is important for the derivation of absolute metabolic fluxes from, e.g. glutamate labeling curves.^6,22 It has been shown that steady-state glucose transport kinetics can be derived from the relationship between brain and plasma glucose, which can provide insights into the kinetics of the brain glucose pool.^20–22,24

Glucose is the single most important substrate for normal function, and the brain relies on a continuous import of glucose from the blood, which must occur across the blood–brain barrier. Glucose transport rates into the brain are thus indicative of the maximal sustainable rate of glucose consumption, CMR_glc.

Traditionally, glucose transport kinetics has been analyzed with a model of brain glucose transport that was based on standard Michaelis–Menten kinetics. However, Michaelis–Menten kinetics is based on the assumption that initial rate of unidirectional product formation is measured, e.g. immediately after substrate and enzyme have been combined. This experimental condition would require the elimination of the brain glucose, which is difficult to achieve without interfering with normal brain function. Hence it is reasonable to expect that reversible Michaelis–Menten kinetics is more appropriate in describing brain glucose transport. Such a model has been proposed,^21,128 and it was shown that one implication of the reversible model of brain glucose transport is that the relationship between brain and plasma glucose is linear.²¹ Many measurements of brain glucose content as a function of plasma glucose have in the meantime corroborated the observation that brain glucose concentrations are a linear function of plasma glucose,^{24,115,129,130} and such a case is illustrated for two different anesthetic regimes, α-chloralose and pentobarbital in Fig. 11(A). These studies indicated that decreased electrical activity and thus decreased energy metabolism resulted in increased brain glucose concentrations. The increment in brain glucose was consistent with an approximately twofold reduction in brain glucose utilization. The data furthermore indicated that during deep pentobarbital anesthesia (a condition known to cause isoelectricity), the brain glucose concentration was still considerably below that expected when glucose consumption was close to zero [indicated by the dashed line in Fig. 11(A)]. The presence of a sizable concentration gradient between brain and plasma glucose implies that net glucose uptake (i.e. glucose consumption at steady state) was appreciable even under conditions close to isoelectricity.

Figure 11.

Brain glucose transport kinetics from the measurement of the brain glucose content as a function of plasma glucose concentration. (A) Demonstration of a linear relationship between brain and plasma glucose concentrations, as well as the effect of increased anesthesia (decreased electrical activity) on brain glucose content in vivo. From Choi et al.¹¹⁵ (B) Comparison of ¹³C NMR quantification with ¹H NMR quantification of brain glucose concentrations during hypoglycemia. From Choi et al.²⁴

The importance of measuring the brain glucose concentration can be appreciated from its role in regulating brain glucose metabolism: glucose becomes rate-limiting for metabolism when its concentration approaches that of the K_m of the first step in its metabolism, which is phosphorylation by hexokinase. Since the K_m of brain hexokinase is very low (~50 μm) and NMR sensitivity in vivo generally is too low to detect such small concentrations of glucose, brain glucose concentrations measured by NMR that are close to zero indicate that metabolism is limited by the glucose available to the brain cell. The general consensus is that brain glucose transport is not rate-limiting for metabolism under normal circumstances. We have recently shown in the conscious human and the α-chloralose-anesthetized rat that the maximal sustainable rate of glucose consumption is approximately 60–90% above the basal rate of glucose metabolism.^21,24 This may, however, not be the case under conditions of extreme metabolic activation or during hypoglycemia.

Previously, models of brain glucose transport have been evaluated at normal or hyperglycemic conditions only.^{21,22,115,129,130} A recent study extended the brain glucose concentrations measurements to hypoglycemia using ¹³C NMR spectroscopy.²⁴ The concentrations measured by ¹³C NMR were found to be in excellent agreement with those predicted by the reversible Michaelis–Menten model as well as those measured by ¹H NMR spectroscopy [Fig. 11(B)]. Interestingly, when the brain glucose concentration approached zero, CBF was acutely increased ²⁴ and glycogen degradation started,¹³¹ all of which points to brain glucose being important in activating cerebral defenses against a deficiency in fuel supply.

Studies have reported that over a 45 min period of forepaw stimulation, oxidative glucose metabolism was increased by more than three-fold.^112,132 Such an increase in cerebral glucose metabolism is clearly beyond what transport across the blood–brain barrier can sustain alone and the implication is that other sources of fuel must have been increasingly utilized.

BRAIN GLYCOGEN, THE FORGOTTEN ENERGY STORE

One fuel source that is endogenous to the brain is glycogen, which is present in the brain in measurable quantities and appears to be essential for brain function. However, the brain glycogen concentration is small compared to the basal metabolic rate of the brain, even though brain glycogen is typically present in quantities that exceed those of tissue glucose in the brain. Textbook reasoning has been that brain glycogen is unlikely to play a role as a significant glucose reservoir, as it may be consumed within minutes during, e.g. hypoglycemia. However, during hypoglycemia, glycogen needs to account for only part of the total glucose metabolic rate and hence can survive longer periods of sustained hypoglycemia. Similar to glucose, brain glycogen is rapidly eliminated in post-mortem tissue,^17,133,134 therefore, its direct biochemical measurement is difficult. Recent studies suggest that, traditionally, brain glycogen content may have been underestimated.^135,136 Localized ¹³C NMR spectroscopy has the unique capability of following brain glycogen metabolism longitudinally, employing a much smaller number of animals than would be used with biochemical extraction. Our results showed that brain glycogen indeed was only slowly degraded during hypoglycemia. This degradation started when brain glucose approached zero [Fig. 12(A)] and became rate-limiting for metabolism,^131,137 as discussed above. Interestingly, at this point cerebral blood flow was also increased abruptly, indicating an attempt by the brain to increase fuel supply for glycolysis, apparently by decreasing the arterio-venous gradient for glucose.²⁴ The rate of brain glycogen degradation during hypoglycemia implied that brain glycogen accounted for the majority of the glucose supply deficit during the hypoglycemic period.¹³¹ Together with the apparent stability of glycogen in the non-stimulated brain at eu- or hyperglycemia,^17,138 these data suggest that brain glucose plays an important regulatory role in cerebral glycogenolysis. These studies also showed that brain glycogen increased above the basal level and beyond following a single episode of hypoglycemia [Fig. 12(A)]. This rebound or super-compensation of brain glycogen may result in increased neuroprotection. In this context it is interesting to note that glycogen metabolism also seems to be insulin-sensitive in the brain, as in most tissues. Therefore, brain glycogen metabolism is likely to be influenced by factors such as insulin and glucagon that are known to be deranged in diabetes. It has been proposed that brain glycogen metabolism may be a factor involved in the mechanism of the hypoglycemia unawareness syndrome observed clinically in patients with type I diabetes,^131,138 perhaps through the enhanced neuroprotective effect of increased brain glycogen.

Figure 12.

Brain glycogen metabolism in the rat. (A) Time-course of glycogen C1 and glucose C1 before, during and after hypoglycemia, which was induced by administering insulin, starting at the point indicated by the arrow. During hypoglycemia, plasma glucose concentration was below 2 mm for 2 h. The vertical dotted line indicates the start of glycogenolysis during hypoglycemia, which coincided with the time point where brain glucose approached zero. The dashed line highlights the slow rate of glycogenolysis during hypoglycemia, expressed as percentages of the pre-hypoglycemic glycogen C1. From Choi et al.¹³¹ Reprinted from J. Neurosci. Res. Copyright © John Wiley & Sons, Ltd. (B) Label incorporation into glycogen and glucose C1, as well as several metabolites was observed in a ¹³C NMR spectrum acquired from a rat brain after 99% enriched [1-¹³C] glucose had been administered for over 48 h ad libitum. Processing consisted of 10 Hz line-broadening and zero-filling prior to Fourier transformation. The spectrum is shown without baseline correction. (C) Comparison between label incorporation into brain glycogen C1 and NAA C6 (solid circles) indicating slow turnover of brain glycogen in the awake rat. The solid line indicates the result of linear regression (r = 0.93, p < 0.01, n = 6). From this relationship, total brain glycogen content was estimated at 3.3 μmol glucosyl units/g wet weight. (B) and (C) are from Choi et al.¹³⁸ Reprinted from Neurochem. Int. Copyright © 2003, with permission from Elsevier

Thus glycogen probably is a viable and important store of glucose equivalents in the brain, whose metabolism is affected by hormones, neurotransmitters and second messengers.¹³⁹

Studies have suggested that brain glycogen metabolism may be altered during focal activation.^140,141 Our studies are consistent with a role for brain glycogen during extreme activation for the following reason: under conditions of extreme local glucose metabolic demand, it is conceivable that the brain glucose concentration could briefly fall into the range of the K_m of hexokinase leading to glycogenolysis and subsequent resynthesis at some later time point. In this scenario, glucose must become rate-limiting for metabolism, which can be inferred from a study that reported during electrical stimulation of the forepaw^112,132 increases in oxidative glucose consumption that were likely to exceed the sustainable supply in brain glucose transport across the blood–brain barrier.

Anesthesia or depressed electrical activity have been reported to have an effect on brain glycogen concentrations.^115,140,142 It has therefore been argued that small changes in brain lactate during stimulation are due to a futile cycling of glucose in and out of glycogen (brain glycogen shunt),¹⁴³ which would link brain glycogen metabolism to brain activity, even when brain glucose is not rate-limiting for metabolism. However, in the awake rat brain, extremely slow rates of bulk brain glycogen turnover were observed,¹³⁸ as illustrated in Fig. 12(B) and (C). These observations together with the apparent influence of low brain glucose on glycogenolysis (see above) make this a rather unlikely scenario. Nevertheless, during severe focal activation it is possible that brain glycogen could in part be activated. Changes in labeled glucose incorporation reported previously¹⁴⁴ may not reflect the entire glycogen molecule.

Because all previous studies measured brain glycogen metabolism in animals, the question remained as to whether brain glycogen metabolism may be faster in the conscious human brain. Brain glycogen metabolism has never been measured in the human brain and ¹³C NMR is the only technique that can provide this insight. We have recently adapted the localization method (see above on localization methods and Fig. 5) for measuring brain glycogen in humans and demonstrated that a reproducible measurement of the brain glycogen signal was indeed possible in the human brain¹⁸ [Fig. 13(A)]. These initial results furthermore demonstrated that brain glycogen metabolism was extremely slow in subjects measured in the awake, resting condition [Fig. 13(B)]. This observation was in excellent agreement with previous studies showing that, under the conditions of this study (plasma glucose at euglycemia or higher with concomitant hyperinsulinemia), the brain glucose concentration is well above the K_m of hexokinase,^20,21,129 thereby eliminating the need for appreciable glycogen activation. In fact, the flux through glycogen synthase at hyperinsulinemia was estimated at 0.1–0.2 μmol/g/h. As a consequence, a brain glycogen pool of a few mm is expected to have a turnover time on the order of several days to a week. These findings suggest that glycogen metabolism is a negligible factor in the energy metabolism of the conscious unstimulated human brain at euglycemia and above. Since concentration changes are induced by a mismatch in catabolic and anabolic flux of glycogen, it is, however, reasonable to assume that elevated brain glycogen concentrations, such as the ones seen after hypoglycemia in the rat, may take days to weeks to be normalized, which incidentally is consistent with the time it takes to restore the hypoglycemia unawareness syndrome in diabetes.¹⁴⁵

Figure 13.

Measurement of glycogen in the human brain. (A) Demonstrates the detection of the brain glycogen signal in four different subjects (arrows) along with the glucose C1 resonances. Shown is the spectral region containing the glycogen C1 and glucose C1 resonances. (B) The increase in the quantified glycogen C1 signal represents the accumulation of [1-¹³C] glycogen, which occurred at an extremely slow rate on the order of 0.15 μmol/g/h in the human brain, as illustrated in the graph containing measurements from three different studies. From Oz et al.¹⁸ Reprinted from Neurochem. Int. Copyright © 2003, with permission from Elsevier

GLUTAMATE C4 TURNOVER: OXYGEN METABOLISM

Because of the ever increasing importance of functional MRI, which depends on an activation-dependent change in the venous concentration of deoxyhemoglobin, the question whether there is tight coupling between glucose and oxygen consumption in the brain has become of paramount importance. The landmark study by Fox and Raichle in the late 1980s suggested that there is indeed a large increase in glucose metabolism that exceeds the changes in oxygen metabolism.¹⁴⁶ The concept of uncoupled oxygen metabolism has been supported by studies reporting small increases in brain lactate during focal activation^147,148 that initially were very controversial¹⁴⁹ and that are very difficult to perform. The relatively small magnitude of change in brain lactate is difficult to reconcile with the reported large uncoupling between oxygen and glucose consumption¹⁵⁰ and explanations linking the lactate increase to brain glycogen at present appear unlikely (see above). To address this question, it is useful to measure the TCA cycle activity in the brain. Because glucose is the dominant substrate for metabolism, the flow of ¹³C label from glucose into the TCA cycle is likely to reflect the cerebral oxygen consumption. In intact tissue the transfer of ¹³C label into the glutamate pool has been linked to TCA cycle flux. The effect of focal activation on cerebral glutamate turnover thus is expected to reflect cerebral oxygen metabolism, at least in part. Using hemi-field activation, glutamate turnover was measured in the activated hemisphere and simultaneously in a control area placed symmetrically to the midline separating the two brain hemispheres. Comparison of the rate of label incorporation indicated a significant difference between the activated and the resting voxels. Modeling of the data indicated that oxygen consumption increased at most by 30%, which is approximately half of the cerebral blood flow increase measured using this stimulation paradigm.⁶² This study supports the idea that oxygen consumption increases are less than the associated cerebral blood flow increases, leading to a net decrease in deoxyhemoglobin content during focal activation, which forms the basis of blood-oxygen-level-dependent functional MRI.¹⁵¹ Other studies have also reported changes in the glutamate C4 turnover rate in response to altered electrical activity.^39,132,152 However, it has been noted in the heart that the labeling of glutamate can change without a change in oxygen consumption,¹⁵³ which was explained by changes in a subcellular isotope flux that contributes to the labeling rate of glutamate, as discussed below.

THE IMPORTANCE OF THE MALATE–ASPARTATE SHUTTLE IN REGULATING CEREBRAL GLUCOSE METABOLISM AND ISOTOPE FLUX

Measurements of Krebs cycle flux from the flow of label from glucose C1 to glutamate C4 are inherently affected by the assumptions made in the modeling. Of critical importance in that regard is the fact that most amino acids are located in the cytosol, whereas the reactions leading to the scrambling of the label take place in the mitochondrion. Therefore, transport has to occur across the highly charged inner mitochondrial membrane. It is likely that the transport of glutamate (a charged amino acid, whose transport is generally associated with concomitant ion transport) or 2-oxoglutarate is controlled, since rapid transport of these acids is likely to interfere with chemiosmosis. Indeed, this exchange rate V_x was found to be comparable to the flux through pyruvate dehydrogenase, V_PDH, in most recent studies of the intact brain,^9,115,154 as illustrated by Fig. 14(A), consistent with what has been reported in most other tissues.^{40,116,153,155} In the brain and other tissues, isotope exchange can occur via the malate-aspartate shuttle [illustrated by the scheme in Fig. 14(B)]. The observation that V_x was comparable to the flux through pyruvate dehydrogenase (Table 2) implied that the malate–aspartate shuttle may be a major mechanism mediating the exchange of label across the mitochondrial membrane,¹⁵⁶ also pointed out elsewhere.^9,53,153,157 Even transamination of glutamate may not be a very fast reaction in vivo: some studies suggest that the transamination rates in the brain are within one order of magnitude of the metabolic rates.¹⁵⁸

Figure 14.

Modeling of cerebral neurotransmitter and amino acid metabolism labeling from glucose. (A) Direct experimental evidence of a slow exchange between 2-oxoglutarate and glutamate in vivo by comparing the rate of labeling for the C3 and the C4 of glutamate. The solid line indicates calculation using the parameters from Choi et al.¹¹⁵ The dashed line was calculated after fitting to the ¹³Glu₄(t) and ¹³Gln₄(t) data only, which resulted in V_PDH = 0.15 μmol/g/min when assuming V_x = 57 μmol/g/min, as in Sibson et al.³⁹, but this was clearly inconsistent with the relative rate of labeling of the C3 resonance in glutamate. (B) The link between oxidative glucose consumption and glutamate labeling is established by active exchange between 2-oxoglutarate (OG) and glutamate (Glu), this flux is indicated by V_x. This exchange, as well as that between oxaloacetate and aspartate, is mediated by the malate-aspartate shuttle.⁵³ In short, the NADH produced by oxidative metabolism of pyruvate at the rate V_PDH must be recycled. This is accomplished by the malate-aspartate shuttle, which transports reducing equivalents produced from the cytosol to the mitochondrion, indicated by the fluxes. The solid lines indicate the isotope flow between the TCA cycle and the cytosolic amino acids, the dashed lines the oxidative metabolism of pyruvate, the dotted lines the NADH/NAD⁺ redox reaction, and the subscripts the flow of label from glucose C1 (or C6) to glutamate. (C) A mathematical model of compartmentalized brain metabolism, from Gruetter et al.⁵³ Reprinted from Neurochem. Int. Copyright ©2003, with permission from Elsevier. The glial compartment is on the left and the neuronal on the right. Abbreviations for metabolic fluxes: V_g + V_PC, glial pyruvate dehydrogenase flux; V_efflux, loss of Gln from the glial compartment; V_out, V_Lout, label dilution and exchange of lactate across the blood–brain barrier. The metabolites in bold are the signals measurable by NMR in vivo. The subscripted numerals indicate the positions labeled due to metabolism of glucose labeled at the C1 position, i.e. Glc₁

Table 2.

Relative fluxes in human visual cortex (mean ± SE) as measured by ¹³C NMR following glucose infusion (from Gruetter et al.⁹)

	Relative flux rate
VNT/CMRglca	41 ± 14%
Oxidative ATP/total ATP production in glia	80 ± 5%
Glial/neuronal ATP synthesis	16 ± 9%
Vx/VPDH	100 ± 35%

^aFor a definition of fluxes see Fig. 14(C) and Gruetter et al.⁹

GLUTAMINE TURNOVER: ANOTHER HALLMARK OF CEREBRAL COMPARTMENTATION

Brain metabolism is exquisitely compartmentalized, with at least two major compartments (attributed to the ‘neuronal’ and ‘glial’ compartments) that are differentiated by the size of the respective glutamate pools associated with the Krebs cycle and that are metabolically linked by the so-called glutamate-glutamine cycle.^159,160 It has now become accepted that the large glutamate pool is in the neuronal compartment and the small glutamate pool in the glial compartment.^161,162 The compartmentation extends further to a number of enzymes, such as glutamine synthetase¹⁶³ and pyruvate carboxylase,¹⁶⁴ being almost exclusively in the glial compartment. Furthermore, glycogen appears to be mainly in the glial compartment in the normal adult brain.¹⁶⁵ The compartmentation of brain metabolism and associated enzymes is summarized in more detail elsewhere.^9,49,50 The scheme in Fig. 14(C) summarizes some of the salient features that can be exploited in modeling brain metabolism. It is of interest to note that the exclusive glial localization of glutamine synthetase implies that the observation of glutamine synthesis in vivo, first achieved in the human brain,^47,48 is a direct manifestation of glial metabolism, whereas the observation of label incorporation into glutamate implies a mainly neuronal event. Clearly, glutamine synthesis can be measured non-invasively by ¹³C NMR spectroscopy, as shown in Fig. 4, which demonstrates the ability of ¹³C NMR spectroscopy to study cerebral compartmentation non-invasively.

ASTROCYTIC ENERGY METABOLISM AND ANAPLEROSIS

A key feature of cerebral compartmentation is the inactivation of neurotransmitter glutamate by uptake into perisynaptic astrocytes and conversion into electrophysiologically inactive glutamine.^{158,166–168} Clearly, this mechanism implies a much more active role for astrocytes than is conventionally assumed, since the conversion of neurotransmitter glutamate to glutamine involves glial energy metabolism.^169–171 The neuron–astrocyte pair thus has to be considered the functional unit (the ‘tripartite synapse’ containing the pre- and postsynaptic neuron and the astrocyte) intimately involved in achieving chemical transmission, as has been proposed in the last decade by Magistretti and others.^52,172,173

The link between astrocytes and neurons is generally accepted from a metabolic and from a neurophysiological standpoint,¹⁷⁴ yet differences exist as to the precise relationship between the metabolic rates and the specific energetics involved.^9,39,115,175 For example, a recent proposal suggested that the rate of glutamate/glutamine inter-conversion (‘glutamate-glutamine cycle’), identified in the scheme in Fig. 14(C) by V_NT^app, is equal to the glucose consumption rate.³⁹ Our recent measurements (summarized in Table 2) imply that, even when assuming that changes in glutamate neurotransmission result in a 1:1 stoichiometric change of glucose metabolism with V_NT^app (which is different from what has been proposed³⁹), glucose metabolism at V_NT = 0 (‘isoelectricity’) is approximately half of that at normal resting conditions, which is in good agreement with reported rates of oxygen consumption during isoelectric pentobarbital anesthesia in humans.¹⁷⁶ The proposal put forth by Shulman and coworkers^39,51 is intriguing as it emphasizes further the coupling between neurons and glia at the level of energy metabolism as put forward earlier by others.¹⁷³

Astrocytes have significant oxidative capacity to generate ATP, which shall be illustrated with two experimental observations. First, anaplerosis, a mitochondrial reaction confined to the glial compartment, mediated by pyruvate carboxylase, is now generally accepted to have significant activity in vivo.^8,9,31,69 Anaplerosis occurs in the glial compartment and labels the TCA cycle intermediates in a different way than does pyruvate dehydrogenase. If pyruvate carboxylase was a substantial flux, V_PC, in the brain in vivo, this should result in a differential labeling of glutamate and glutamine due to their differential distribution between astrocytes and neurons. Indeed, it was concluded from the differential label distribution in glutamate and glutamine (Table 3) that pyruvate carboxylase activity was substantial^8,9,31,33 and contributed approximately 25% to the flux through glutamine synthetase V_syn,⁸ which was comparable to the numbers subsequently reported in the rat brain using [2-¹³C] glucose infusions.⁸⁹ Net glutamine synthesis (via pyruvate carboxylase) was shown to generate excess amounts of ATP oxidatively, thereby generating approximately one-third of the ATP per glucose molecule compared to that from complete glucose oxidation.⁵³ The second experimental evidence supporting the presence of significant oxidative metabolism in astrocytes stems from the fact that acetate is known to be exclusively metabolized in astrocytes by the TCA cycle,^177,178 which can only occur by oxidative metabolism.¹⁷⁹ Interestingly, the metabolic rates of the glial compartment relative to the neuronal compartment reported from the acetate studies in the human brain^71,87 confirmed the earlier measurement of relative glial ATP synthesis in the human brain,⁹ tabulated in Table 2.

Table 3.

Label incorporation into glutamine (Gln) relative to glutamate (Glu) in human visual cortex as measured by ¹³C NMR following glucose infusion (mean ± SEM)

	C2	C3	C4
Gln/Glu (45 ml volume9)	0.41 ± 0.02	0.28 ± 0.04*	0.30 ± 0.01
Gln/Glu (72 ml volume8)	0.39 ± 0.03	0.18 ± 0.01*	0.26 ± 0.02

^*Significantly different between C2 and C3 (p < 0.05).

The mainly oxidative glial metabolism is compatible with the proposal that lactate produced in astrocytes is a fuel for oxidative metabolism in neurons. For example, if one-sixth of the ATP production from glucose is in the glial compartment, five-sixths of the lactate remain for export to neurons, when assuming the extreme case that phosphorylation of glucose is an exclusively glial process.

OUTLOOK

As illustrated above and summarized in the abstract, localized ¹³C NMR spectroscopy has already made many substantial contributions to our understanding of brain metabolism, e.g. understanding glucose transport, measuring brain glutamate and glutamine turnover, quantifying astrocytic and brain glycogen metabolism. Pivotal achievements include the demonstration that pyruvate carboxylase activity is substantial in vivo, and that glutamatergic neurotransmission can potentially be quantified in vivo. Future developments on the technical side probably will include the extension of these studies to GABA metabolism in vivo, and the characterization of the various fluxes that can be measured in different brain areas. Some investigators have already begun applying ¹³C NMR spectroscopy to disease states in humans^88,123 and animal models of disease,¹⁵⁴ and future expansion is to be expected in this area as well. The low sensitivity and technical complexity, including the administration of ¹³C label, will probably remain obstacles that need to be overcome for a successful ¹³C NMR line of research. The most important issue for this field to flourish, however, is the emergence of a critical mass of several independent research groups whose competition will foster the development of this new investigative tool of localized in vivo ¹³C NMR spectroscopy.