COMMUNICATIONS

Magnetic Resonance in Medicine 51:435– 440 (2004)

Measurement of Brain Glutamate Using TE-Averaged PRESS at 3T Ralph Hurd,1* Napapon Sailasuta,1 Radhika Srinivasan,2 Daniel B. Vigneron,2 Daniel Pelletier,2 and Sarah J. Nelson2 A method is introduced that provides improved in vivo spectroscopic measurements of glutamate (Glu), glutamine (Gln), choline (Cho), creatine (Cre), N-acetyl compounds (NAtot , NAA ⴙ NAAG), and the inositols (mI and sI). It was found that at 3T, TE averaging, the f1 ⴝ 0 slice of a 2D J-resolved spectrum, yielded unobstructed signals for Glu, Glu ⴙ Gln (Glx), mI, NAtot, Cre, and Cho. The C4 protons of Glu at 2.35 ppm, and the C2 protons of Glx at 3.75 ppm were well resolved and yielded reliable measures of Glu/Gln stasis. Apparent T1/T2 values were obtained from the raw data, and metabolite tissue levels were determined relative to a readily available standard. A repeatibility error of <5%, and a coefficient of variation (CV) of <10% were observed for brain Glu levels in a study of six normal volunteers. Magn Reson Med 51:435– 440, 2004. © 2004 Wiley-Liss, Inc. Key words: TE averaging; glutamate; glutamine; J-resolved spectroscopy; T2 relaxation

Glutamate (Glu) is the principal excitatory neurotransmitter of the central nervous system (CNS). The neurotoxic properties of excess Glu have been linked to a wide range of degenerative diseases, such as multiple sclerosis (MS), amyotrophic lateral sclerosis (ALS), Huntington’s disease (HD), and Alzheimer’s disease (AD) (1– 4), and to secondary apoptosis associated with stroke, head trauma, and spinal injury (5). Even malignant gliomas have responded to treatment with Glu antagonists (6). In short-TE proton spectra, the overlapped signals of Glu and glutamine (Gln) are readily measured as total Glu ⫹ Gln (Glx). Some studies have shown increases in Glx for severe abnormalities, such as hypoxic encephalopathy (7), acute MS lesions (8), HD (9), ALS (10), and certain tumors (11). Decreases in the Glx signal have also been reported (12). Unfortunately, Glx doesn’t measure changes in Glu/Gln stasis, and is therefore unlikely to be an adequate marker for less profound changes in the intracellular conditons that may precede or accompany conditions of toxic Glu. The spectral overlap of Gln, Glu, N-acetyl (NA), and myo-Inositol (mI) makes it difficult to reliably sort out Glu and Gln by themselves in conventional spectra, even with prior knowledge fitting (13). Multiple quantum editing (14) has been used to resolve Glu; however, this technique has less clinical value because it results in the loss of Gln, mI, and NAtot, which

are also important to assess. More recently, 2D constant time point-resolved spectroscopy (CT-PRESS) (15) has been used to resolve Glu at 3T. While this method preserves the other metabolite markers, it requires over 120 points in the f1-dimension to resolve the Glu signal. 2D J-resolved spectroscopy (16,17) has also been used to resolve these metabolites. In this study, TE-averaged PRESS, the f1 ⫽ 0 component of a 2D J-resolved acquisition, is demonstrated to be a reliable spectroscopic tool. MATERIALS AND METHODS 3T Data Acquisition Data were acquired on a 3T Signa scanner (GE Medical Systems, Waukesha, WI) using the standard quadrature head coil, PROBE/SVQ™ (automated PRESS), and a modification of that sequence, PRESS-2D J, which enables the collection of TE averaged data. TR was set at 2 s for the human studies, except for a few cases in which fully relaxed TR ⫽ 8 data were collected to determine partial spin-echo saturation. Reference PRESS data were collected with 128 total scans at TE ⫽ 35. Very selective saturation pulses with PRESS overselection were used to define the volume of interest (18). PRESS-2D J data were collected by averaging 16 or 32 steps with an increment time of 10 ms, or 128 steps with an increment time of 2.5 ms—all with an initial TE of 35 ms. Sixteen steps with an increment of 10 ms were used for the validation study. Spectra were acquired from an 8-cc mid-parietal “mostly” gray matter (GM) region and a left parietal “mostly” white matter (WM) region, as illustrated in Fig. 1. A spectral width of 5000 Hz and 2K points provided an f2-sampling time of 410 ms. The number of excitations was adjusted to keep total data collection time to about 8 min. Basis set spectra were acquired under matched conditions. The volunteer group consisted of six normal, consenting adults (three males (22–52 years old) and three females (22–52 years old)). The volunteer studies were all conducted under IRB guidelines, and informed consent was obtained from all of the subjects. Basis Set Preparation and Spectra

1

GE Medical Systems, Menlo Park, California. University of California–San Francisco, San Francisco, California. *Correspondence to: Ralph Hurd, 333 Ravenswood Ave, Building 307, Menlo Park, CA 94025. E-mail: [email protected] Received 11 June 2003; revised 22 October 2003; accepted 24 October 2003. DOI 10.1002/mrm.20007 Published online in Wiley InterScience (www.interscience.wiley.com). 2

© 2004 Wiley-Liss, Inc.

Solutions of 50 mM NAA, 100 mM Glu, 50 mM Gln, 50 mM Cre, 50 mM Cho, and 150 mM mI were prepared in 50 mM phospate buffer, pH 7.2, containing 0.1% azide, 0.5 mM DSS, and approximately 0.5 mM chelated Gadolinium (0.1% v/v Magnavest). The solutions were placed into 600-ml polyethylene spheres, and data were collected

435

436

Hurd et al.

FIG. 1. Location of mostly GM (white) and mostly WM (black) volumes selected for study.

to match in vivo conditions. The standard GE MRS HD sphere (12.5 mM NAA, 10 mM Cre, 3 mM Cho, 5 mM lactate, 12.5 mM Glu, and 7.5 mM mI, pH 7.2, 0.1% Magnavist) was used as the quantification standard.

FIG. 2. 3T PRESS spectrum of normal mid-parietal WM, and scaled spectra of NAA, Glu, Gln, and mI—all collected under identical conditions at TE ⫽ 35 and TR ⫽ 2000. Note the overlap of signals from NAtot, Glu, Gln, and mI.

Simulations GAMMA software (19) was used to simulate both TE ⫽ 35 PRESS and TE- averaged PRESS spectra using previously published values for Glu and Gln chemical shift and coupling constants (21). A Gaussian apodization of 0.04 ppm was added to each spectrum to approximate a typical in vivo performance. Data Processing In vivo spectra were all processed by averaging in t1, application of 4-Hz Gaussian apodization, fast Fourier transform (FFT), and automatic zero-order phase correction. The linewidth of Cre methyl at 3.02 ppm was mea-

sured and used to determine the matching apodization for processing the reference spectra. T2 and K, the peak intensity at TE ⫽ 0, were then determined for both reference and in vivo spectra using SAGE™ T2 fitting, limited to a single exponential. Direct measurements were made for NAtot, Cho, Cre, and residual water. In the case of Glu (2.35 ppm), peak height was used. Gln levels were estimated as Glx (3.75 ppm) – Glu (2.35 ppm). The mI signal, which resolves as an apparent doublet at 3.6 ppm in TEaveraged PRESS, was estimated using an average of the double peak. The basis sets and standard GE MRS HD sphere all had T1 relaxations of ⬍400 ms and were considered fully relaxed under the TR conditions used. GM

FIG. 3. GAMMA simulation of TE ⫽ 35 PRESS spectra of Glu (dark) and Gln (light), using chemical shift and J-values from Ref. 19. Spectra at 0.5–7T with 0.04 ppm of Gaussian apodization. Left: Constant Hz scale. Right: Constant ppm scale.

Measurement of Brain Glutamate

437

FIG. 4. GAMMA simulation of TE-averaged PRESS spectra of Glu (dark) and Gln (light), using chemical shift and J-values from Ref. 19. Spectra at 0.5T to 7T with 0.04 ppm of Gaussian apodization. Left: Constant Hz scale. Right: Constant ppm scale.

and WM metabolite T1’s were calculated from partial saturation estimated as S/So ⫽ (1 – 2e–(TR–TEeff/2)/T1) ⫹ e–TR/T1), where So is the fully relaxed signal intensity at TR 8, and S is the signal intensity at TR ⫽ 2. Underestimation of T1 due to transition band saturation differences was minimized by the use of very selective saturation (VSS) and overPRESS (18). For the population study, the percent coefficient of variation (CV) is presented as the 100 * standard deviation (SD)/average. RESULTS A conventional 3T PRESS spectrum of human brain at TE ⫽ 35, along with basis set spectra of NAA, Glu, Gln, and mI, are shown in Fig. 2. In this typical GM spectrum, the N-acetyl group of NAtot (2.02 ppm) is overlapped with Glu C3 protons. The Glu C4 protons (2.35 ppm) are contaminated with NAA C3 protons, as well as Gln C3 and C4 protons. The co-resonant signals from the Glu and Gln C2 protons (3.75 ppm) are partly obscured by mI. At this TE, some macromolecule signal is also co-resonant, adding to the uncertainty of any deconvolution attempt. The issue of Glu/Gln overlap is illustrated in Fig. 3, which shows TE ⫽ 35 PRESS simulations from 0.5T to 7T. With the exception of 0.5T, where SNR and magnet homogeneity may limit the resolution of Glu and Gln in vivo, overlap is significant. In comparison, TE-averaged PRESS (Glu/Gln simulations shown in Fig. 4) fully resolves the Glu C4 protons (2.35 ppm) from overlap at 3T and higher. At 1.5T and below, not enough of the C4 proton multiplet is fully coincident with the C4 proton chemical shift to remain unmodulated. In vivo 3T results are shown in Fig. 5. The Glu C4 protons at 2.35 ppm resolve from Gln and NAA. The signal from Glx C2 protons (3.75 ppm) is resolved from overlap with mI, and the NAtot signal at 2.02 ppm can be quantified without the added problem of overlapped Glu C3 protons. The TE-averaged PRESS data shown in Fig. 5 were collected from TE ⫽ 35–352, with average TE ⫽ 194 ms, in effect trading off T2 loss for resolution. How-

ever, as shown in Fig. 6, 16 points sampled from 35 to 185 ms still provides sufficient resolution, and reduces effective TE (TE at mean signal intensity during T2 decay) to about 100 ms. In this case the loss of signal relative to TE ⫽ 35 PRESS is modest, as illustrated in Fig. 7. The choice of step size and number is relatively unimportant for the protocol used in this study. A step size of 20 ms and seven increments provides sufficient bandwidth in t1 to cover the J-dimension, and provides results equivalent to those shown with 16 increments of 10 ms, or 64 increments of 2.5 ms. Fewer steps can be useful for spectroscopic imaging applications, while oversampling in t1 can

FIG. 5. TE-averaged spectrum of mid-parietal GM. The peak at 2.35 ppm is exclusively from Glu. The resolved signal at 3.75 ppm is Glx, the sum of Glu and Gln. In conventional PRESS, Glu, Gln, and NAA are overlapped. The spectra were all collected by averaging TEs from 35–335 ms, in 128 steps of 2.5 ms, and NEX ⫽ 2. TR was set to 2 s.

438

Hurd et al.

found to give less variance. A summary of the metabolite tissue levels and T2 values is given in Table 1.

DISCUSSION

FIG. 6. TE-averaged spectrum of mid-parietal GM along with spectra of NAA, Glu, Gln, and mI collected by TEs averaging 35–195 ms. The peak at 2.35 ppm is exclusively from Glu. The resolved signal at 3.75 ppm is Glx, the sum of Glu and Gln. In conventional PRESS, Glu, Gln, and NAA are overlapped. Data were collected in 16 steps of 10 ms, and NEX ⫽ 8. TR was set to 2 s.

reduce artifacts in data collected without water suppression (16) or in regions with high lipid content (21). The concomitant measurement of T2 for NAtot, Cre, and Cho, which is inherent in the TE-averaged PRESS method, is repeatable to an SD of ⬍6%. Therefore, biological deviations exceeding that can be included in patient-specific absolute tissue-level calculations. Typical T2 fits of in vivo and standard reference spectra are illustrated in Fig. 8. This approach provides a T2-compensated quantification of the resolved singlets directly. For the isolated Glu signal at 2.35 ppm, the modulation pattern from the simulation can be used in the estimation of standard and in vivo T2 and K; however, the use of peak intensity in this case was

In this work we have shown that TE-averaged PRESS at 3T can provide spectra optimized for quantitative measurements of Glu, Gln, mI, Cho, Cre, and NAtot. Good baseline and sufficient resolution are achieved at relatively short effective-TE values, and with relatively few points in t1. With respect to SNR, effective TE depends on metabolite T2. For example, the effective TE for GM creatine (Cre) (T2 ⫽ 143 ms) will be 150 ms for a TE ⫽ 35–345 ms t1 sample, and 100 ms for a TE ⫽ 35–185 ms sample. The elimination of Glu-NAA signal overlap simplifies the quantitative measurement of the NAtot N-acetyl spectral region. It should be noted, however, that this approach does not improve separation of the NAAG and NAA signals, and it may still be preferable to obtain a composite measurement of the N-acetyl region at 2 ppm. This improvement in the measurement of Glu is clearly not possible at 1.5T, but based on the GAMMA simulations, 4T and 7T may benefit. NAtot, Cre, and Cho T2 measurements are directly obtainable from the PRESS-2D J data, with repeatability at the SNR limit. Single exponential linear least-square fits to these signals at 3T were repeatable to an SD of ⬍6%. The methyl signal from Cre showed the smallest CV across our small test population, with %CV at the repeatability limit, and GM had the lowest T2 at 143 ⫾ 7 ms. Cre in the WM region of interest (ROI) also demonstrated a relatively narrow biological variation, with an average T2 of 176 ⫾ 13. WM NAtot T2 at 328 ⫾ 51, and GM NAtot T2 at 213 ⫾ 21 showed a biological variation in excess of the repeatability limit, but not nearly as much as Cho. GM Cho T2 at 230 ⫾ 74 ms, and WM Cho T2 at 236 ⫾ 29 ms showed enough biological variation to justify the expansion of this limited preliminary population data. If the magnitude of this vari-

FIG. 7. Comparison of TE ⫽ 35 PRESS (left) with a scaled TE-averaged PRESS spectrum (right). TE-averaged PRESS was sampled with 16 increments of 10 ms starting at TE ⫽ 35, and NEX ⫽ 4; 128 total averages were collected for both fixed TE ⫽ 35 and TE averaged runs. With TR ⫽ 2000, each spectrum took 4.27 min to collect.

Measurement of Brain Glutamate

FIG. 8. Quantification of metabolites using a standard MRS HD sphere as reference. Reference spectra were apodized to match in vivo, and volume-selected to match transmit gain within ⫾0.5 dB. a: Cho. Left: normal GM. Right: Reference. T1 was corrected for a value of 1.1 s, as measured using TR ⫽ 8, NEX ⫽ 2 as a fully relaxed reference. b: Cre corrected for a T1 of 1.55 s. c: NAtot corrected for a T1 of 1.65 s.

439

440

Hurd et al.

Table 1 Metabolite Levels and T2’s from Test Group ROI

NAtot

Cre

Cho

Glu

Gln

WM (N ⫽ 6) GM (N ⫽ 6)

7.4 (0.4) 9.5 (0.7)

5.6 (0.5) 7.2 (0.8)

1.7 (0.1) 1.2 (0.1)

7.1 (0.5) 11.7 (1.1)

1.7 (1.6) 3.2 (1.6)

ROI

NAtot T2

Cre T2

Cho T2

WM (N ⫽ 6) GM (N ⫽ 6)

328 (51) 213 (21)

176 (13) 143 (7)

236 (29) 230 (74)

Average and SDs given in mmoles/kg wet weight for metabolite levels and in milliseconds for T2 values.

ation is confirmed in a larger study, compensation for individual Cho T2’s, even in studies with relatively short TEs (e.g., TE ⫽ 35), will be required for accurate quantitative measurements. Measuring the neuronal marker (NAA) and a putative glial marker (mI) concomitantly may enable investigators to distinguish between neuronal injury and other sources of Glu/Gln shift. Further, an inherent measure of metabolite T2 also provides a measure of the intracellular environment, which may change due to swelling (22). The metabolite levels observed in this study are within the distribution range of previously reported MRS and ex vivo values (23–25). However, GM Glu at 11.7 mmoles/kg is at the high end of this range, and the WM NAA level at 7.4 mmoles/kg is in the lower range of reported values. A possible source of Glu overestimation is the macromolecule content that remains from the short-echo portion of the TE average. Although spectral simplification and flat baselines provided by TE-averaging make direct single peak fitting reliable, the use of prior knowledge fitting (13) may further improve quantitative accuracy. Although the TE-averaged method avoids overestimation of NAtot due to Glu contamination, underestimation (especially in WM) may arise due to the assumption of a single T2 relaxation for NAA ⫹ NAAG. Unlike conventional MRS, the method described herein can reliably measure a change in Gln/Glu ratio. An additional advantage of this approach is that the source of the Gln/Glu shift in any given disease state can potentially be determined. The co-measurement of the neuronal viability marker (NAA), glial marker (mI), and cellular environment changes via metabolite T2 could help refine the source of change. In our study, Glu/NAtot had a relatively narrow variation in normal brain (8%). NAtot, mI, Cho, and Cre were co-measured, with small %CV, along with T2-relaxivity measures of the water and the metabolite methyl groups, which provided added specificity. REFERENCES 1. Pitt D, Werner P, Raine CS. Glutamate excitotoxicity in a model of multiple sclerosis. Nat Med 2000;6:67–70. 2. Benveniste H, Drejer J, Schousboe A, Diemer N. Elevation of the extracellular concentrations of glutamate, and asparate in rat hippocampus during transient cerebral ischemia monitored by intracerebral microdialysis. J Neurochem 1984;43:1369 –1374. 3. Auger C, Attwell D. Fast removal of synaptic glutamate by post synaptic transporters. Neuron 2000; 28:547–558. 4. Doble A. The role of excitotoxicity in nerodegenerative disease: implications for therapy. Pharmacol Ther 1999;81:163–221. 5. Zipfel G, Babcock DJ, Lee JM, Choi DW. Neuronal apoptosis after CNS injury: the roles of glutamate and calcium. J Neurotrauma 2000;17:857– 869.

6. Rzeski W, Turks L, Ikonomidou C. Glutamate antagonists limit tumor growth. Proc Natl Acad Sci USA 2001;98:6372– 6377. 7. Kreis R, Arcinue E, Ernst T, Shonk TK, Flores R, Ross BD. Hypoxic encephalopathy after near-drowning studied by quantitative 1H-MRS. J Clin Invest 1996;97:1142–1154. 8. Helms G. Volume correction for edema in single-volume proton MR spectroscopy of contrast-enhancing multiple sclerosis lesions. Magn Reson Med 2001;46:256 –263. 9. Taylor-Robinson SD, Weeks RA, Bryant DJ, Sargentoni J, Marcus CD, Harding AE, Brooks DJ. Proton MRS in HD: evidence in favour of the glutamate excitotoxic theory. Mov Disord 1996;11:167–173. 10. Pioro EP, Majors AW, Mitsumoto H, Nelson DR, Ng TC. 1H-MRS evidence of neurodegeneration and excess glutamate plus glutamine in ALS medulla. Neurology 1999;53:71–79. 11. Rijpkema M, Schuuring J, Van Der Meulen Y, Van Der Graaf M, Bernsen H, Boerman R, Van Der Kogel A, Heerschap A. Characterization of oligodendogliomas using short echo time 1H MR spectroscopic imaging. NMR Biomed 2003;16:12–18. 12. Hanstock CC, Thompson RB, Wieler M, Allen PS, Martin WRW. Depletion of glutamate in the motor cortex in Huntington’s disease measured using short echo STEAM at 3 Tesla. In: Proceedings of the 7th Annual Meeting of ISMRM, Philadelphia, 1999. p 1445. 13. Provencher SW. Estimation of metabolite concentration from localized in vivo proton NMR spectra. Magn Reson Med 1993;30:38 – 44. 14. Thompson RB, Allen PS. A new multiple quantum filter design procedure for use on strongly coupled spin systems found in vivo: its application to glutamate. Magn Reson Med 1998;39:762–771. 15. Mayer D, Spielman DM. Detection of glutamate in the human brain at 3 Tesla using optimized CT-PRESS. In: Proceedings of the 10th Annual Meeting of ISMRM, Honolulu, 2002. p 785. 16. Hurd RE, Gurr D, Sailasuta N. Proton spectroscopy without water suppression: the oversampled J-resolved experiment. Magn Reson Med 1998;40:343–347. 17. Thomas MA, Ryner LN, Mehta MP, Turski PA, Sorenson JA. Localized 2D J-resolved 1H MR spectroscopy of human brain tumors in vivo. J Magn Reson Imaging 1996;6:453– 459. 18. Khan T-K, Vigneron D, Sailasuta N, Tropp J, Le Roux P, Nelson S, Hurd RE. Very selective suppression pulses for clinical MRSI studies of brain and prostate cancer. Magn Reson Med 2000;43:23–33. 19. Smith SA, Levante TO, Meier BH, Ernst RR. Computer simulations in magnetic resonance. An object oriented programming approach. J Magn Reson 1994;106a:75–105. 20. Govindaraju V, Basus VJ, Matson GB, Maudsley AA. Measurement of chemical shifts and coupling constants for glutamate and glutamine. Magn Reson Med 1998;39:1011–1013. 21. Bolan PJ, DelaBarre L, Baker EH, Merkle H, Everson LI, Yee D, Garwood M. Eliminating spurious lipid sidebands in 1H MRS of breast lesions. Magn Reson Med 2002:48:215–222. 22. Hansson E, Johansson B, Westergren I, Ronnback L. Glutamate-induced swelling of single astroglial cells in primary culture. Neuroscience 1994;63:1057–1066. 23. Kreis R. Quantitative localized 1H MR spectroscopy for clinical use. Prog NMR Spectrosc 1997;31:155–195. 24. Pouwels PJW, Frahm J. Regional metabolite concentrations in human brain as determined by quantitative localized proton MRS. Magn Reson Med 1998;39:53– 60. 25. Perry TL, Hansen S, Berry K, Mok C, Lesk D. Free amino acids and related compounds in biopsies of human brain. J Neurochem 1971;18: 521–528.