NAA is formed by the transamination of glutamate (Glu)with oxaloacetate which leads to aspartate (Asp) (Cooper et al. 1970), and in the presence of acetyl coenzyme A and NAA synthase, (l-aspartate N-acetyltransferase), aspartate (Asp) is converted to NAA. This process occurs primarily, but not exclusively, in neurons, where it exhibits a very high intracellular–extracellular gradient. NAA is hydrolyzed back into aspartate and acetate by aspartoacylase (N-acetyl-l-aspartate amidohydrolase) in mature oligodendrocytes only. In some neurons, a portion of NAA is turned into N-acetylaspartylglutamate (NAAG) in the presence of glutamate (Glu) by an NAAG synthase (N-acetylaspartate-l-glutamate ligase) (Baslow 2000). NAA and NAAG are major neuronal osmolytes, and the large amount of NAA and NAAG present in the brain can serve as cellular reservoirs for Asp and Glu. Since both are polar and ionizable hydrophilic molecules that undergo a regulated efflux into extracellular fluid, they can also play a role in water movement out of neurons (Baslow 2000). In addition, the fact that the NAA-metabolizing enzyme aspartoacylase is an integral component of the myelin sheath suggests that intraneuronal NAA may supply the acetyl groups for myelin lipid synthesis (Chakraborty et al. 2001). Finally, the cell type-specific metabolic enzymes in the NAA and NAAG cycle indicate a three-cell compartmentalization involving neurons, astrocytes, and oligodendrocytes. Those neurons that may synthesize NAAG from NAA and Glu target the release of NAAG to astrocytes, where it is cleaved into NAA and Glu. The Glu is taken up by astrocytes, where it may be returned to neurons via the glutamate–glutamine cycle. The residual NAA extracellular products are removed and hydrolyzed by oligodendrocytes (Baslow 2000). This unique tricellular metabolic cycle, as shown in Fig. 6.2, involving NAA and NAAG, provides a potential glial cell-specific signaling pathway, which can also be reflected in changes in different diseases (Baslow 2010).

As a “virtual biopsy,” localization is required when examining specific regions of the brain involved in disease. Localization is defined by a region of interest, or voxel, from which the spectrum is acquired and, hence, described as single voxel spectroscopy (SVS). This region is a cube within the brain that can be visualized and placed in the sagittal, coronal, and axial planes as shown in Fig. 6.1. The localization is achieved by utilizing a pulse sequence that acquires the MRS signal. Though several sequences, and permutations of sequences, exist which are capable of yielding an in vivo MRS spectra, many of them are research techniques and not widely available (although discussed later in this chapter). The two most commonly used sequences are PRESS and STEAM as described below:

One of the drawbacks of SVS is the natural limits of a single area of acquisition. Obtaining spectra in many different areas of the lesion would be time consuming. Multivoxel spectroscopy, also called chemical shift imaging (CSI) or spectroscopic imaging, overcomes this issue by adding an additional phase-encoding step that allows for spatial encoding of a larger volume such that it is divided into smaller voxels that can be summed or selected chemical shifts can be color-coded into chemical shift maps. In experimental duration, some two to three times longer than that in which a single voxel method acquires the spectrum of a single ROI, the CSI technique (Maudsley et al. 1983) can collect an array of spectra from a single plane.

As described above, single voxel and CSI methods suffer from issues such as lipid contamination, voxel registration, as well as assumptions of T1 and T2 relaxation times. One method of addressing the aforementioned issues is to take advantage of the fact that NAA is the most prominent peak within the MRS spectrum (thus having a high signal-to-noise ratio) that acquires signal from the entire head and is thus named whole-brain NAA (WBNAA) spectroscopy (Rigotti et al. 2007). Unlike the single voxel and CSI methods, WBNAA does not utilize localization and instead acquires data immediately after excitation (TE = 0 ms) with a very long repetition time (TR = 10 s) and a short inversion time (TI = 940 ms). The inversion pulse is alternated between each acquisition where the TI is designed to null the NAA signal. The even and odd acquisitions are then subtracted from one another which nulls those metabolites, including lipid, that have a short T1 relaxation time. As NAA has a long T1 relaxation time (1.4 s), it remains visible. As a result of the long TR, there are no T1- or T2-weighting effects. The WBNAA signal is then normalized to the total brain volume as can be obtained from brain segmentation. This method operates under the assumption that all NAA arises from within the brain, and given the lack of localization, results of such studies would imply global effects of the disease.

Metabolically, glutamate (Glu) is stored as glutamine (Gln) in the glia, and the balanced cycling between these two neurochemicals is essential for normal functioning of brain cells. Glu and Gln are compartmentalized in neurons and glia, respectively, and this chemical interconversion reflects an important aspect of metabolic interaction between these two types of cells. In vivo studies have revealed that the neuronal/glial Glu/Gln cycle is highly dynamic in the human brain and is the major pathway of both neuronal Glu repletion and astroglial Gln synthesis (Gruetter et al. 1994; Mason et al. 1995). After its release into the synaptic cleft, Glu is taken up by adjoining cells through excitatory amino acid transporters. Astrocytes are responsible for the uptake of most extracellular Glu via Glu transporters and additionally have a vital role in preserving the low extracellular concentration of Glu needed for proper receptor-mediated functions, as well as to prevent excitotoxicity (Schousboe 2003; Schousboe and Waagepetersen 2005). Once taken up into the astrocyte, Glu is rapidly converted to Gln by the enzyme Gln synthetase. Small quantities of Gln are also produced de novo or from GABA (Hertz and Zielke 2004; Bak et al. 2006). Gln is released from astrocytes, accrued by neurons, and converted to Glu by the neuron-specific enzyme phosphate-activated glutaminase (Bak et al. 2006). Gln is the main precursor for neuronal Glu and GABA (Hertz and Zielke 2004), but Glu can also be synthesized de novo from tricarboxcylic acid cycle intermediates (De Graaf et al. 2011). The rate of Glu release into the synapse and subsequent processes are dynamically modulated by neuronal and metabolic activity via stimulation of extrasynaptic Glu receptors, and it has been estimated that the cycling between Gln and Glu accounts for more than 80 % of cerebral glucose consumption (Sibson et al. 1998). The tight coupling between the Glu/Gln cycle and brain energetics is largely tied to the nearly 1:1 stoichiometry between glucose oxidation and the rate of astrocytic Glu uptake. This relationship was first determined by Magistretti et al. in cultured astroglial cells where the addition of Glu resulted in increased glucose consumption (Pellerin and Magistretti 1994). These results provided the hypothesis that glycolysis in the astrocytes results in a production of two molecules of ATP, which are then consumed by the formation of Glu from Gln, which suggests a tight coupling between the two mechanisms.

The molecular structures of Glu and Gln are very similar and, as a result, give rise to similar magnetic resonance spectra (Govindaraju et al. 2000). The four protons from the two methylene groups of glutamate Glu and Gln are located at 2.04–2.35 ppm and 2.12–2.46 ppm, respectively. Similarly, the methine group resonates at 3.74 ppm and 3.75 ppm for glutamate Glu and Gln, respectively. Thus, even though Glu has a relatively high concentration in the brain, its major resonances are usually contaminated by contributions from Gln, GABA, glutathione (GSH), and NAA. To avoid confusion in spectral assignment of Glu and Gln, the term “Glx” has traditionally been used to reflect the combined Glu and Gln concentrations. However, this approach does not allow for the evaluation of conditions where the concentrations of Gln and Glu are in opposing directions, nor does this approach allow for the evaluation of Gln and Glu separately. As our understanding of the importance of Glu/Gln system in the human brain has increased, much endeavor has been invested in being able to quantify Glu or Gln separately. The sections below detail methods for measuring Glu utilizing various pulse sequences to achieve the separation of Glu from co-resonating metabolites.

LCModel (Provencher 1993) is a software package often used for post-processing of MRS data. It utilizes prior knowledge in the form of a basis set that is made up of a linear combination of in vitro, or simulated, spectra from individual metabolite solutions. The advantage of this method is that it is almost completely automated and utilizes algorithms for baseline correction and peak fitting without imposing restrictive parameterization and without subjective input. It is often used in the literature to calculate concentrations of individual metabolites including Glu and Gln. LCModel provides a quantitative measure of the accuracy of the measure by using the estimated standard deviations or Cramer–Rao lower bound (CRLB) that provides a measure of how reliable the measurement is. A standard deviation of <20 % has been used throughout the literature as the criteria for adequate reliability. While Glu standard deviation measures tend to fall below 20 %, Gln standard deviation measures do not. Given the overlap between the two molecules, it is unclear whether LCModel can accurately discern Glu from Gln.

In our lab we conducted a series of experiments where different solutions, or phantoms, were created that had different concentrations of Glu and Gln. In three of the phantoms, Glu was increased from 6 to 18 mM concentrations and Gln was maintained at 6 mM, which is equivalent to the physiological concentrations in the brain (3–5.8 mM (Govindaraju et al. 2000)). In the other three phantoms, Glu was maintained at 12 mM (equivalent to in vivo concentrations of 6–12.5 mM (Govindaraju et al. 2000)), but Gln was increased from 3 to 12 mM. Spectra were then acquired from these solutions using a clinical MRI scanner (Siemens 3 T TIM Trio) using conventional PRESS MRS (echo time of 30 ms, repetition time of 2 s). LCModel was then used to calculate the concentrations of each phantom as shown in Fig. 6.5. Our results showed that the variable concentration of Gln has a direct effect on the reported Glu concentrations and therefore demonstrates that this method is not sufficient for discerning Glu from Gln.

In order to better separate Glu and Gln, the echo time can be altered to minimize the contribution of the Gln signal to the Glx complex. Schubert et al. (Schubert et al. 2004) were able to selectively detect C4 proton resonances of Glu using a PRESS sequence at 3 T with a TE of 80 ms and analyzing the data using both the time and the frequency domains with prior knowledge obtained from phantom spectra. Using this approach, in vivo spectral features were similar to in vitro Glu spectral features collected from a phantom, and Glu signal was well resolved and separated from major interferences, such as Gln and NAA. Similarly, Jang et al. acquired PRESS spectra at 1.5 T in vitro and in vivo at four TE values: 30, 35, 40, and 144 ms (Jang et al. 2005), with the resulting spectra analyzed with LCModel (Provencher 2001). In vitro and in vivo spectra yielded the lowest Cramer–Rao lower bounds (CRLB) for Glu quantitation when TE was set to 40 ms. TE optimization for the purpose of Glu and Gln detection has also been demonstrated using stimulated echo acquisition mode (STEAM) spectroscopy by Yang et al. (2008), who optimized TE and mixing time (TM) to resolve the C4 proton resonances of Glu and Gln. Optimal TE methods are appealing acquisition strategies due to their ease of implementation for both acquisition and processing and the ease with which appropriate parameter timings can be selected at the scanner interface. Additionally, the resulting spectra can be analyzed using scanner software or commercially available software (e.g., LCModel, jMRUI) after simulating an appropriate basis set.

Due to the fact that Glu is a strongly coupled system, short echo time acquisitions are usually preferred over long echo time acquisitions for the purpose of reducing peak phase modulation. The ultrashort TE approach in such coupled systems gives rise to spectra where the majority of peaks are inphase and thus reduces signal cancellations due to J-modulations at longer echo times which in turn lead to more robust signal quantification. Wijtenburg and Knight-Scott (2011) quantified Glu/tCr at 3 T by combining short TE STEAM (6.5 ms) with phase rotation (Hennig 1992; Ramadan 2007) to detect Glu in human brain. This approach has been compared to 40 ms TE PRESS, 72 ms TE STEAM, and TE averaging and demonstrated that the short TE phase-rotation-STEAM approach yielded the greatest precision for Glu/tCr ratio quantification followed by TE averaging, PRESS 40 ms, and STEAM 72 ms (Wijtenburg and Knight-Scott 2011).

The implementation of very short echo time techniques requires substantial modification of standard vendor-supplied localization sequences. These spectra are usually complicated by a broad baseline extending over the whole spectral width, and care should be taken during analysis to eliminate the contribution of the baseline to the desired metabolite signal—which is usually done by utilizing T1 differences (Starcuk et al. 2001). The use of third-party spectral analysis software and, in some cases, the collection or simulation of metabolite basis set are essential to yield results of high accuracy (Wijtenburg and Knight-Scott 2011).

In this method, a number of 1D PRESS spectra are acquired at variable TE values and co-added (averaged) in real time to produce a single 1D, TE-averaged spectrum. Alternatively, the same TE-averaged spectrum can be obtained if these time-domain FIDs are ordered in a 2D matrix and a two-dimensional Fourier transform (2DFT) applied. The resultant 1D spectrum corresponding to the middle of the F1 axis (F1 = 0 Hz) is equivalent to the TE-averaged spectrum obtained above (Dreher and Leibfritz 1995; Bolan et al. 2002). This 1D spectrum offers relatively uncontaminated spectral peaks for C4 protons at 2.35 ppm for Glu and for C2 at 3.75 ppm for Glx (Hurd et al. 2004). Representative spectra are shown in Fig. 6.6a.

While TE averaging is a simple and reliable method for measuring Glu and Gln, it sacrifices spectral information arising from the J-evolution of other metabolites. The method also suffers from differential T2 weighting per individual TE value, which can be corrected for if a basis set is similarly simulated or experimentally acquired. TE-averaging sequences are not usually available from major MRI vendors but can be easily modified from standard PRESS sequence source code.

A large sensitivity enhancement for the detection of Glu can be achieved by careful design of MR exchange experiments. A recent example of this technique was recently implemented by Cai et al. where the chemical exchange saturation transfer (CEST) between bulk water and Glu was utilized for the detection of Glu (Cai et al. 2012). In such CEST experiments, advantage is taken from the protons on the Glu amine group, which are labile and in constant exchange with the protons of bulk water. When the Glu NH₂ protons are saturated by RF irradiation at 7.7 ppm (i.e., 3 ppm higher than water), the resultant saturation effect is transferred to water due to the ongoing chemical exchange between the protons of NH₂ and H₂O. This in turn leads to a reduction in the amplitude of water resonance due to the saturation of its protons and the saturation RF field applied to the Glu amine group. An excellent review of chemical exchange methodology can be found here (Zhou and van Zijl 2006).

An alternative to separating Glu from Gln is to use a second chemical shift dimension: 2D Fourier NMR spectroscopy utilizes a simple two-pulse sequence, 90x-t1-90x-Acq(t2), creating a dataset that is a series of one-dimensional (1D) spectra with traditional time readout (t2), but each with increments in delay (t1), inserted before the terminal readout 90° RF pulse (Jeener et al. 1979). Two-dimensional Fourier transformation (FT) of t1 and t2 produces a 2D spectrum where the x- and y-axes are the frequencies F2 and F1, respectively, known as correlated spectroscopy (COSY). In a 2D COSY spectrum, scalar coupling between protons in molecules results in cross peaks that allow for unambiguous identification of different metabolites (Thomas et al. 2001; Cocuzzo et al. 2011; Ramadan et al. 2011). This technology has been translated to clinical MR scanners and has been applied in vivo (Schulte et al. 2006), demonstrating detection of GABA, Glu, Gln, glutathione, as well as other metabolites (Fig. 6.7).

While the aforementioned methods have been utilized throughout the literature to study Glu, these measures provide static measures of Glu, measuring the total concentration of Glu in the brain without a sense of the dynamic changes that result from neurotransmission and brain energetics. Carbon 13 (¹³C) MRS is presently the only method that provides noninvasive measurements of neuroenergetics and neurotransmitter cycling in the human brain. ¹³C MRS, combined with the administration of ¹³C-labeled substrates, allows the detection of ¹³C incorporation from ¹³C-labeled precursors into various carbon positions of metabolites, including the Glu cycles between neuronal and glial compartments (De Graaf et al. 2011). For example, if glucose, the primary fuel for the brain, is labeled with a nonradioactive isotope of ¹³C on the first carbon, that label is retained as it is metabolized through the TCA cycle. At the point of alpha-ketoglutarate, the label is transaminated to Glu and Gln via the Glu/Gln cycle as shown in Fig. 6.8. As the label is incorporated separately into the fourth carbon group of Glu and Gln, real-time acquisition of the brain will result in the observation of the Glu C4and Gln C4 resonances at 34.4 ppm and 31.8 ppm, respectively. In the second turn of the TCA cycle, Glu C2 and C3 are labeled. Thus, ¹³C MRS allows continuous, noninvasive monitoring of metabolic fluxes under different physiological or pathophysiological conditions.

The ¹³C MRS acquisition consists of several different steps providing multiple means by which Glu can be measured. These include choice of substrate (glucose, acetate, label position), method of delivery (oral, IV, glucose clamped), data acquisition (direct detection, indirect detection), and method of analysis (dynamic difference spectroscopy, isotopomer analysis) (Ross et al. 2003). It is important to note that the natural abundance of ¹³C is approximately 1.1 % which implies that there is very little background signal of indigenous ¹³C Glu, therefore providing excellent contrast to noise ratio of signals that arise from the incorporation of the ¹³C label via the TCA cycle and subsequent transamination from alpha-ketoglutarate to the Glu/Gln cycle.