The separation of Gln and Glu in STEAM: a comparison study using short and long TEs/TMs at 3 and 7 T

Transcription

1 DOI /s RESEARCH ARTICLE The separation of Gln and Glu in STEAM: a comparison study using short and long TEs/TMs at 3 and 7 T Weiqiang Dou Jörn Kaufmann Meng Li Kai Zhong Martin Walter Oliver Speck Received: 2 September 2014 / Revised: 12 December 2014 / Accepted: 15 December 2014 ESMRMB 2014 Abstract Objectives This study aimed to determine the optimal echo time (TE) and mixing time (TM) for in vivo glutamine (Gln) and glutamate (Glu) separation in stimulated-echo acquisition mode at 3 and 7 T. We applied a short TE/TM (20/10 ms) for a high signal-to-noise-ratio and a field-specific long TE/TM (3 T: 72/6 ms; 7 T: 74/68 ms) for optimal Gln and Glu separation of the Carbon-4 proton resonances. Materials and methods Corresponding Gln and Glu spectra were simulated using VeSPA software, and measured in a phantom and human brains at 3 and 7 T. W. Dou (*) O. Speck Faculty for Natural Sciences, Biomedical Magnetic Resonance, Otto-von-Guericke University, ZENIT Building, Leipziger Str. 44, Magdeburg, Germany douweiqiang@gmail.com J. Kaufmann Department of Neurology, Otto-von-Guericke University, Magdeburg, Germany M. Li M. Walter O. Speck Leibniz Institute for Neurobiology, Magdeburg, Germany K. Zhong High Magnetic Field Laboratory, Chinese Academy of Sciences, Hefei , China M. Walter Department of Psychiatry, Otto-von-Guericke University, Magdeburg, Germany M. Walter O. Speck Center for Behavioral Brain Sciences, Magdeburg, Germany O. Speck German Center for Neurodegenerative Disease (DZNE), Site Magdeburg, Magdeburg, Germany Results Higher spectral separation for Gln and Glu was achieved at 7 than 3 T. At 7 T, short TE/TM provided comparable spectral separation and in vitro Gln and Glu quantification compared to long TE/TM. Moreover, it showed greater reliability in in vivo Gln and Glu detection and separation than long TE/TM, with significantly lower Cramer Rao lower bounds (Gln: 14.9 vs. 75.8; Glu: 3.8 vs. 6.5) and correlation between Gln and Glu (p = 0.004). Conclusion Based on the optimal separation for Gln and Glu, a short TE/TM at 7 T is proposed for future in vivo Gln and Glu acquisition. Keywords Gln and Glu STEAM Short and long TEs/ TMs Separation 3 and 7 T Introduction Glutamate (Glu) is the major excitatory neurotransmitter in the central nervous system. Glutamine (Gln), the precursor of Glu, is synthesized from Glu in astrocytes and converted into Glu after its release into neurons [1, 2]. Non-invasive quantitative detection of Glu and Gln in human brains is of particular interest in clinical studies, since abnormal levels of synaptic Glu and Glu/Gln cycling are directly related to many neurological and psychiatric diseases, including Huntington s disease [3], major depressive disorder [4], bipolar disorder [5], or Alzheimer s dementia [6]. However, reliable detection of both metabolites is currently limited primarily by spectral overlap between the two due to similar chemical shifts and J-coupling effects [7]. Proton magnetic resonance spectroscopy ( 1 H-MRS) is a non-invasive technique for measuring Gln and Glu [8]. Based on 1 H-MRS, a number of methods, mainly between 1.5 and 4 T, have been reported to be capable of effectively

2 detecting Gln and Glu. These include J-refocused spectral editing [9], multiple quantum coherence filtering [10], spectrally selective refocusing [11], chemical shift selective filters [12], 2D J-resolved techniques [13, 14], and constant time point-resolved spectroscopy (PRESS) [15]. All these proposed spectral editing techniques, however, only focus on one metabolite while suppressing the other, and the 2D spectroscopy techniques usually drastically increase scan time for each voxel. Due to these drawbacks, the abovementioned techniques are not very suitable for application, especially in clinical MRS studies that require simultaneous multi-metabolite acquisition within a limited scan time. Stimulated echo acquisition mode (STEAM), as one widely used 1 H-MRS technique, is able to efficiently measure metabolites in parallel [16]. Yang et al. [7] proposed a spectral simplification method for Gln and Glu using STEAM with field-specific inter-pulse timings, i.e., echo time (TE) and mixing time (TM). By simulating the spectral responses of Gln and Glu in STEAM acquisitions with a range of TEs and TMs for T, they found field-specific long TE/TM settings to optimally resolve Gln and Glu signals of the Carbon-4 (C4) proton resonances, which are the main peaks of Gln and Glu spectra. On the other hand, a short TE STEAM sequence was also proposed for in vivo Gln and Glu detection at high or ultra-high field strengths, since short TE can provide spectra with a high signal-tonoise-ratio (SNR), and increased field strength is able to enhance spectral sensitivity and chemical shift dispersion to partially resolve the spectral overlap problem [16, 17]. Moreover, Stephenson et al. [18] reported that STEAM with a short TE/TM (16/17 ms) can provide significantly lower Cramer Rao lower bound (CRLB) values than the 7 T-specific long TE/TM (74/68 ms) proposed by Yang et al. [7], when measuring in vivo Gln and Glu signals at 7 T. However, based on their respective advantages for Gln and Glu detection, it remains unknown whether the optimal spectral separation of Gln and Glu and, thus, the most accurate metabolite quantification, can be achieved by the field-specific long TE/TM or with the higher spectral SNR at a short TE/TM in high/ultra-high field. Therefore, to determine the best TE/TM for Gln and Glu separation, we systematically simulated the signals of Gln and Glu in STEAM with a short TE/TM and the field-specific long TE/TM settings proposed by Yang et al. [7], and measured the corresponding in vitro and in vivo Gln and Glu signals at 3 and 7 T. Besides the investigation of spectral overlap and CRLB estimations for Gln and Glu, we further compared and assessed the concentration relationships and the degrees of correlation between Gln and Glu at short and long TEs/TMs. In addition, we also acquired in vivo spectra using PRESS with a short TE of 35 ms at 3 T to further explore the relationship between spectral SNR and quantification of Gln and Glu. Materials and methods Spectral simulation The software Versatile Simulation, Pulses and Analysis for Magnetic Resonance Spectroscopy (VeSPA) version [19, 20] was applied to simulate the spectral responses of Gln and Glu in a standard STEAM sequence with short and long TE/TM settings at 3 and 7 T. The chemical shift values and J-coupling constants of Gln and Glu were obtained from Govindaraju et al. [21]. The resonance frequencies and short and long TE/TM values were separately set as MHz, 20/10 ms, and 72/6 ms for 3 T, respectively, and MHz, 20/10 ms, and 74/68 ms, respectively, for 7 T, in which the field-specific long TE/TM settings were proposed by Yang et al. [7]. The embedded standard STEAM sequence with ideal hard radio frequency (RF) pulses was chosen to simulate spectra for simplicity, since the choice of realistic or ideal hard selective 90 pulses was proven not to be a critical factor governing spectral responses in STEAM sequences [22]. In vitro experiments A spherical phantom with a 40-mm diameter was used in in vitro experiments, consisting of 100-mM Glu and 17-mM Gln in a buffered solution (ph 7.2). The concentration ratio of Glu to Gln (Glu/Gln) is thus 5.88, agreeing with the recently proposed Glu/Gln ratio in the posterior cingulate cortex (PCC) of healthy human brains [23]. All in vitro experiments were performed using a 3 T Siemens MAGNETOM Trio scanner with an 8-channel phased-arrayed head coil and a 7 T MR scanner (Siemens MAGNETOM) with a 32-channel head array coil. All MR spectroscopy (MRS) voxels were placed in the phantom center and manually shimmed to improve magnetic field homogeneity according to the vendor-provided automatic shim procedure. After these voxel-wise adjustments, standard STEAM spectra [128 averages, repetition time (TR) = 2,000 ms, short TE/TM = 20/10 ms, long TE/TM = 72/6 ms, voxel size = mm 3, data size = 2,048 points, bandwidth (BW) = 2,000 Hz] were acquired at 3 T. Spectrum acquisition for each voxel took 4 min and 24 s. STEAM sequences with optimized variable rate selective excitation (VERSE) RF pulses [24], which allows reduction of the peak power requirements of RF pulses at ultra-high field strengths [25], were employed to acquire spectra at 7 T. The scan parameters included 128 averages, TR = 3,000 ms, short TE/TM = 20/10 ms, long TE/TM = 74/68 ms, voxel size = mm 3, data size = 2,048 points, and BW = 2,800 Hz. The acquisition time for each voxel was 6 min and 36 s. Corresponding water reference spectra for

3 Fig. 1 Voxel placement in the rpcc region (a) and cpcc region (b) in representative 3D MPRAGE anatomical images acquired at 3 T eddy current correction and absolute metabolite concentration quantification were acquired using four averages at 3 T and one average at 7 T. In vivo experiments After giving informed consent, six healthy male subjects (28 ± 4 years old) were recruited for human brain scans at 3 T, and eight other healthy male subjects (26 ± 3 years old) were enrolled at 7 T. To rule out physical illnesses and psychiatric abnormalities, all subjects were assessed before participation using self-report questionnaires. The study was approved by the local Institutional Review Board. Similar to the in vitro measurements, the human experiments were performed in the same 3 T system with an 8-channel head coil and in the same 7 T system, albeit with a 24-channel head array coil. A 1.0-mm isotropic resolution three dimensional (3D) magnetizationprepared rapid gradient echo (MPRAGE) sequence was applied for T 1 -weighted anatomical image acquisition at 3 T with TE = 4.77 ms, TR = 2,500 ms, inversion time (TI) = 1,100 ms, flip angle (FA) = 7, BW = 140 Hz/ pixel, and acquisition matrix = A 0.8-mm isotropic resolution 3D MPRAGE sequence was employed for acquiring T 1 -weighted brain anatomical images at 7 T with TE = 2.73 ms, TR = 2,300 ms, TI = 1,050 ms, FA = 5, BW = 150 Hz/pixel, and acquisition matrix = All MPRAGE images were reconstructed into the anterior commissure posterior commissure plane. Based on three views (i.e., sagittal, coronal, and transverse directions) of the MPRAGE images, MRS voxels with voxel size mm 3 = 3.75 ml were then carefully placed in the rostral PCC (rpcc) region (Fig. 1a) for all 3 and 7 T subjects, as well as an additional caudal PCC (cpcc) region (Fig. 1b) for 3 T subjects. Manual shimming was used to improve the magnetic field homogeneity for voxel-specific regions at 3 T, and a vendor-provided double-gradient echo shim technique was implemented to shim all voxels at 7 T. STEAM and STEAM VERSE sequences with identical TR/TE/TM

4 settings as in the in vitro experiments were applied for in vivo spectrum acquisitions at 3 and 7 T. In addition, PRESS with short TE = 35 ms was employed in the 3 T measurements for further exploring the relationship between high SNR and accurate metabolite quantification, although it requires much higher RF peak power than STEAM with an identical TE and, thus, is not suitable for the application in ultra-high field 7 T [26]. The scan parameters were as follows: 128 averages, TR = 2,000 ms, TE = 35 ms, voxel size = mm 3 = 3.75 ml, data size = 2,048 points, BW = 2,000 Hz. The acquisition took 4 min and 24 s per voxel. Data analysis Spectral overlap between simulated Gln and Glu spectra was calculated to evaluate the spectral separation of the two by computing the corresponding geometrical overlap ratios as the Gln and Glu intersection relative to the set union of Gln and Glu (Gln Glu/Gln + Glu) in the range of ppm. The correspondingly calculated set unions (Gln + Glu) for short and long TEs/TMs were also compared. The SNRs of in vitro and in vivo spectra were calculated according to the following equations: SNR in vitro = peak height of Glu/standard deviation of spectral noise and SNR in vivo = peak height of NAA/standard deviation of spectral noise. LCModel version [27] was applied to analyze all in vitro and in vivo spectra in this study. Four TE/TM- and field-specific model spectral basis sets, for analyzing the corresponding spectra acquired using STEAM, consisted of measured Gln and Glu spectra and fifteen simulated spectra of alanine, aspartate, N-acetylaspartate (NAA), N-Acetylaspartylglutamic acid, choline, creatine (Cr), GABA, glucose, glycerophosphocholine, gluthatione, myo-inositol, phosphocreatine (PCr), lactate, phosphocholine, and scyllo-inositol. To measure the corresponding Gln and Glu spectra, two phantoms identical to the one for in vitro measurements were separately injected with 100-mM Gln or Glu resolved in a buffered solution. The solution ph in each phantom was 7.2. Both phantoms were measured at room temperature using the above-mentioned STEAM and STEAM-VERSE sequences with identical TR/TE/TM settings at 3 and 7 T. While most basis spectra were simulated to avoid time consuming measurements, the experimental basis spectra of these two metabolites were included as these improved the fitting of the main metabolites of interest with lower CRLB values. In addition, the PRESS-specific in vitro basis set was measured using an identical TR/TE set at 3 T. It included sixteen major metabolite spectra (alanine, aspartate, NAA, N-Acetylaspartylglutamic acid, Cr, GABA, Glu, Gln, glucose, glycerophosphocholine, gluthatione, myo-inositol, PCr, lactate, phosphocholine, and taurine). Instructions in the LCModel manual [28] for analyzing phantoms without dominating landmark resonances from NAA, Cr, or Choline were followed to analyze the in vitro spectra. The corresponding concentrations of Gln and Glu, as well as their respective CRLB values, were obtained and the concentration ratios of Glu to Gln (Glu/Gln) were calculated. For the in vivo spectra, the absolute concentrations of target metabolites (e.g., Gln and Glu) with respective CRLBs and full width at half maximum (FWHM) values for spectral line-width estimation were obtained. In addition, correlation coefficients between fitted metabolite concentrations, derived from a standard least-squares variance covariance matrix of LCModel analysis, can be found in the detailed LCModel output [28, 29]. The values between in vivo Gln and Glu were used to evaluate their correlations. Under the assumption that these metabolites are not directly coupled, a lower correlation represents better separation. Because T 1 and T 2 relaxation effects of metabolites can drastically influence the quantification of metabolite concentrations, especially for a long TE/TM [28], the correction for relaxation effects was taken into account when using the LCModel for metabolite quantification. For simplicity, the following field-specific T 1 and T 2 values of Cr (the reference metabolite in LCModel analysis) were selected: 1,380 and 151 ms for 3 T, and 1,760 and 121 ms for 7 T [30]. The metabolite concentrations were expressed using institutional units (i.u.). In vivo spectra with poor quality were excluded, following standard criteria of SNR < 15, FWHM > 12 Hz (for 3 T) and > 25 Hz (for 7 T), CRLBs for Gln + Glu (Glx), and Gln and Glu > 100 % [18, 29]. In total, two Gln values acquired with the 7 T-specific long TE/TM were discarded from the whole data sample because of their CRLB values larger than 100 %. Additionally, the paired t test function in SPSS 18 (SPSS for Windows, Chicago, IL, USA) was applied for the full sample to test statistical significance, with the threshold offset to p = The 3 T rpcc and cpcc spectra were combined as PCC spectra for paired t test analysis, since the concentration levels of Gln and Glu at both sub-regions in PCC were reported to have no significant difference [23]. Results Simulation The geometrical overlap ratios (Gln Glu/Gln + Glu) of Gln and Glu spectra simulated for 3 T were substantially larger for short TE/TM than for long TE/TM (28 vs. 22 %; Fig. 2a, b). At the increased field strength of 7 T, spectral separation of the two was higher at either short or long TE/ TM (Fig. 2c, d) and both overlap ratios became comparable (11 vs. 10 %). In addition, the set unions (Gln + Glu) at short TE/TM were significantly larger (3 T: 1.4 times larger; 7 T: 1.8 times larger) than those at long TE/TM.

5 Fig. 2 Simulated Glu (red) and Gln (black) spectra of the C4 proton resonances (around 2.35 ppm for Glu and 2.45 ppm for Gln) in STEAM acquisitions with a short TE/TM (20/10 ms) at 3 T (a), a long TE/TM (72/6 ms) at 3 T (b), a short TE/ TM (20/10 ms) at 7 T (c), and a long TE/TM (74/68 ms) at 7 T (d). Signal intensities are expressed using arbitrary units (a.u.) Fig. 3 LCModel-analyzed phantom spectra: a, b show the spectra acquired using STEAM with a short TE/TM (20/10 ms) and a long TE/TM (72/6 ms) at 3 T; c, d show the spectra acquired using STEAM with VERSE at a short TE/TM (20/10 ms) and a long TE/TM (74/68 ms) at 7 T In vitro Phantom results are shown in Fig. 3 and Table 1. Unlike the values for 3 T, both Glu/Gln concentration ratios obtained with short and long TEs/TMs for 7 T were similar and much closer to the true value 5.88 (3 T: 8.64 vs. 6.86; 7 T: 5.99 vs. 5.87), while short TE/TM provided higher SNRs and lower CRLBs than long TE/TM. In vivo 3 T Using paired t tests for all 3 T spectra (Fig. 4a c), we detected no significant differences in spectral FWHMs between PRESS, short, and long TE/TM spectra (Fig. 5a). The SNRs of PRESS spectra were the highest compared to

6 Magn Reson Mater Phy Table 1 Quantitative phantom results from STEAM acquisitions with short and long TEs/TMs at 3 T and STEAM VERSE acquisitions with short and long TEs/TMs at 7 T Short TE/TM Long TE/TM CRLB (%) CRLB (%) SNR Glu/Gln Gln Glu 3T T SNR Glu/Gln Gln Glu Fig. 4 Example LCModel-analyzed 3 T in vivo spectra from the cpcc region using STEAM with a short TE/TM (20/10 ms) (a), a long TE/TM (72/6 ms) (b), using PRESS with short TE = 35 ms (c), 7 T in vivo spectra from the rpcc region using STEAM with VERSE with a short TE/TM (20/10 ms) (d) and a long TE/TM (74/68 ms) (e) STEAM spectra at short and long TEs/TMs, while STEAM spectra had significantly higher SNRs with short TE/TM than with long TE/TM (Fig. 5b). 13 As shown in Fig. 5c, no significant differences in Glx concentrations are revealed between short and long TEs/TMs and between long TE/TM and PRESS,

7 Fig. 5 3 T in vivo results acquired using STEAM with short and long TEs/TMs and PRESS with short TE are shown: FWHMs (a), SNRs (b), concentrations (c), and CRLBs (d) of metabolites ( S : short TE/TM; L : long TE/TM; P : PRESS), and correlation coefficients between Gln and Glu (e). Data are shown as mean ± SD (***p < 0.001; *p < 0.05). Metabolite concentrations are expressed using institutional units (i.u.) although a marginal difference is found between short TE/TM and PRESS. However, the assignments of Glx signals to either Gln or Glu are found to be significantly different between short and long TEs/TMs and PRESS. The correlation coefficients between Gln and Glu (Fig. 5e) were comparable between short and long TEs/ TMs, and between short TE/TM and PRESS, while significantly different correlation coefficients were found between long TE/TM and PRESS.

8 In addition, as major metabolites, Cr + PCr (tcr) showed comparable concentrations at both short and long TEs/TMs, but NAA concentrations at short TE/TM were significantly lower than those measured at long TE/TM (Fig. 5c). 7 T 7 T in vivo results (Fig. 4d, e) showed significantly higher spectral SNRs with short TE/TM than with long TE/TM, while the FWHM values were not significantly different between short and long TEs/TMs (Fig. 6a, b). Short TE/TM showed no significantly different Glx concentrations compared to long TE/TM, but did show significantly different Gln and Glu concentrations compared to long TE/TM (Fig. 6c). In addition, the correlations between Gln and Glu with short TE/TM were significantly lower than those with long TE/TM (Fig. 6e). As shown in Fig. 6c, NAA and tcr have comparable concentrations when measured with short and long TEs/ TMs. Discussion In this study, we simulated 3 and 7 T STEAM spectra of Gln and Glu at short and long TEs/TMs and measured the corresponding in vitro and in vivo Gln and Glu signals. The field-specific Gln and Glu signals at short TE/TM for high SNRs and long TE/TM for best separation of both main peaks were compared in order to determine the optimal TE/ TM setting for Gln and Glu separation, and, thus, for in vivo Gln and Glu acquisitions. While long TE/TM provided less spectral overlap for the simulated Gln and Glu spectra and allowed more reliable quantification for the in vitro spectra than short TE/TM at 3 T, this advantage became insignificant in the simulated and in vitro spectra at 7 T, which at both short and long TEs/TMs achieved much lower spectral overlap and more accurate metabolite quantification. All in vitro spectra were acquired with sufficient SNRs (Table 1). More accurate quantification for long TE/TM at 7 than 3 T is mainly because of the increased chemical shift dispersion, allowing higher spectral separation between Gln and Glu. At 3 T, short TE/TM caused more overlapped spectra as opposed to long TE/TM. Therefore, more deviated quantification was introduced when spectra were acquired with short TE/TM. Due to much lower concentrations and a shorter T 2 relaxation time, in vivo spectra usually cannot be measured with comparable SNRs as in vitro spectra, especially at a long TE. Compared to short TE/TM, long TE/TM with significantly lower SNRs provided similar Glx quantification but significantly different assignments to either Gln or Glu in 3 and 7 T in vivo spectra. Based on much smaller CRLBs and less correlated Gln and Glu values (Figs. 5d, e, 6d, e), the in vivo Gln and Glu acquired with short TE/TM were considered more reliably separated. Additionally, the in vivo spectra measured with a short TE in PRESS, which had highest SNRs across all 3 T spectra, show Glu/Gln concentration ratios consistent with the literature (3.62 vs. 3.65) [31], while the ratio is 2.4 for short TE/TM and 1.0 for long TE/TM. Also, with twice the SNR, 7 T rpcc spectra at short TE/TM have Glu/Gln concentration ratios more comparable to values reported for the anterior cingulate cortex (ACC) (6.0 vs. 4.8, 4.2) [18, 32] than those at long TE/TM (16.7). Small deviations of the ratios at short TE/ TM from the literature are likely because of the reported regional variation of Glu/Gln ratios between ACC and PCC regions [23]. Therefore, we can infer that a high spectral SNR is the dominant contribution for accurate Gln and Glu separation and, thus, quantification in in vivo spectra. Most of the CRLB values presented for Glu and Gln at short and long TEs/TMs are slightly larger as compared to the literature [7, 18]: 4 versus 2 and 7 versus 8 (Glu), 15 versus 6 and 76 versus 28 (Gln) at 7 T; 13 versus 8 and 28 versus 10 (Glu), 31 versus 24 and 27 versus 5 (Gln) at 3 T. The most likely reason is that for high spectral SNRs, both previous studies acquired spectra with a voxel size more than two times larger and twice as many measurement repetitions than used here (8, 9 vs ml; 256, 288 vs. 128). However, in most clinical MRS studies, physicians require a small voxel size to limit partial volume effects; they also restrict the scan time for patients. On the other hand, it was reported that the mobile lipid or the macromolecule signals might interfere with the Glu and Gln signals of the C4 proton resonances at 2.35 and 2.45 ppm when using a short TE/TM [7, 33]. To accurately quantify Gln and Glu using the LCModel, Schaller et al. [34, 35] proposed including an experimentally measured macromolecule spectrum into the basis set. Together with the mathematical approximation (spline baseline) in the LCModel, this extended basis set can improve quantification, especially for 7 T in vivo Gln and Glu. In addition, NAA was reported to have increased spectral interference with Glx at a short TE/TM [7, 36]. Our results indeed find that at 3 T, the estimated concentrations of NAA measured with short TE/TM are significantly lower than those measured at long TE/TM, which might indicate significant signal interference between NAA and Glx using a short TE/ TM at 3 T. However, while 7 T results provided more reliable Gln and Glu separation, such interference does not occur at this field strength with similar NAA concentrations at short and long TEs/TMs (p > 0.05). Identical metabolites were contained in a total of four STEAM-specific basis sets. Taurine was not included in

9 Fig. 6 7 T in vivo results acquired using STEAM with short and long TEs/TMs are shown: FWHMs (a), SNRs (b), concentrations (c), and CRLBs (d) of metabolites ( S : short TE/TM, L : long TE/TM), and correlation coefficients between Gln and Glu (e). Data are shown as mean ± SD (**p < 0.01; *p < 0.05) any of them. The absence of taurine is not considered crucial for in vivo Gln and Glu estimation, since its signals appearing at 3.2 and 3.4 ppm [21] do not fall into the peak ranges of Gln and Glu. Moreover, the taurine concentration is usually low in healthy brains (<2 mmol/kg) [35]. In addition, the applied PRESS-specific basis set did not include choline and scyllo-inositol. The absent choline signal was replaced by the combination of phosphocholine and glycerophosphocholine signals [35]. The missing scylloinositol is considered to be not important for in vivo Gln and Glu analysis, as its signal appearing at 3.3 ppm [21] does not overlap with either Gln or Glu signals and its concentration in healthy human brain is minor (<0.4 mmol/kg) [35]. While increased chemical shift dispersion improves spectral separation, an increased chemical shift effect

10 (spatial displacement between metabolites) is also introduced at 7 T. However, the impact is considered to be small in our study, since high BW (2,800 Hz) frequency modulated slice selective RF pulses were used in our STEAM sequence, which caused only a 1 % chemical shift displacement error (<1 mm) for the 30-Hz chemical shift difference between the main peaks of Gln and Glu. Besides the STEAM technique, other 1 H-MRS techniques have also been previously applied at 3 or 4 T for in vivo Gln and Glu detection, such as MEGA-PRESS spectral editing [37], semi-adiabatic LASER [38], and 2D J-resolved PRESS [39]. Kakeda et al. [37] used MEGA- PRESS with TE = 68 ms to acquire Gln and Glu signals in human frontal lobe and parieto-occipital lobe regions at 3 T. The quantified Glu/Gln ratio was 1.1 with CRLBs of 70 and 11 for Glu and Gln estimation, respectivley. Öz et al. [38] applied the semi-adiabatic LASER sequence with a short TE = 24 ms for full signal intensity to measure human cerebellum spectra at 4 T. A mean Glu/Gln ratio of 2.6 was obtained. In addition, as suggested by Henry et al. [40], 2D J-resolved PRESS is an effective and stable technique for in vivo Gln and Glu acquisitions. Walter et al. [39] utilized this technique at 3 T to detect the concentrations of Gln and Glu in the human pregenual ACC region with a 16-min acquisition time. The resulting Glu/Gln ratio was 4.0. Compared to these above-mentioned techniques, application of STEAM with a short TE/TM at 7 T shows the best separation for Gln and Glu with a high and reasonable Glu/Gln ratio. Moreover, its much faster spectrum acquisition, as compared to 2D J-resolved PRESS, is particularly beneficial for human studies. Conclusion In summary, together with the increased spectral dispersion in ultra-high field 7 T, the superiority of Gln and Glu separation using the previously proposed long TE/TM is outweighed by the high SNR of a short TE/TM. The short TE/TM can provide more reliable Gln and Glu separation, especially for in vivo spectra. Therefore, the application of a short TE/TM in STEAM at 7 T is proposed for future Gln- and Glu-oriented in vivo MRS studies. Acknowledgments This work was supported by the Deutsche Forschungsgemeinschaft (DFG: Wa2673/3-1), the ERA-NET Neuron Project SuppHab (MW), and Sonderforschungsbereich (SFB)-779 (MW, OS). Conflict of interest The authors have no conflict of interest. Ethical standards Measurements on human subjects in this study were approved by the local ethics committee and were, therefore, performed in accordance with the ethical standards laid down in the Declaration of Helsinki. All involved subjects gave their informed consent before recruitment in the study. References 1. Albrecht J, Sidoryk-Węgrzynowicz M, Zielińska M, Aschner M (2010) Roles of glutamine in neurotransmission. Neuron Glia Biol 6: Daikhin Y, Yudkoff M (2000) Compartmentation of brain glutamate metabolism in neurons and glia. J Nutr 130:1026S 1031S 3. Taylor-Robinson SD, Weeks RA, Bryant DJ, Sargentoni J, Marcus CD, Harding AE, Brooks DJ (1996) Proton magnetic resonance spectroscopy in Huntington s disease: evidence in favour of the glutamate excitotoxic theory. Mov Disord 11: Horn DI, Yu C, Steiner J, Buchmann J, Kaufmann J, Osoba A, Eckert U, Zierhut KC, Schiltz K, He H, Biswal B, Bogerts B, Walter M (2010) Glutamatergic and resting-state functional connectivity correlates of severity in major depression the role of pregenual anterior cingulate cortex and anterior insula. Front Syst Neurosci 4:33 5. Brennan BP, Hudson JI, Jensen JE, McCarthy J, Roberts JL, Prescot AP, Cohen BM, Pope HG, Renshaw PF, Ongür D (2010) Rapid enhancement of glutamatergic neurotransmission in bipolar depression following treatment with riluzole. Neuropsychopharmacology 35: Moats RA, Ernst T, Shonk TK, Ross BD (1994) Abnormal cerebral metabolite concentrations in patients with probable Alzheimer disease. Magn Reson Med 32: Yang S, Hu J, Kou Z, Yang Y (2008) Spectral simplification for resolved glutamate and glutamine measurement using a standard STEAM sequence with optimized timing parameters at 3, 4, 4.7, 7, and 9.4 T. Magn Reson Med 59: Kaiser LG, Schuff N, Cashdollar N, Weiner MW (2005) Agerelated glutamate and glutamine concentration changes in normal human brain: 1 H MR spectroscopy study at 4 T. Neurobiol Aging 26: Lee HK, Yaman A, Nalcioglu O (1995) Homonuclear J-refocused spectral editing technique for quantification of glutamine and glutamate by 1 H NMR spectroscopy. Magn Reson Med 34: Thompson RB, Allen PS (1998) A new multiple quantum filter design procedure for use on strongly coupled spin systems found in vivo: its application to glutamate. Magn Reson Med 39: Choi C, Coupland NJ, Bhardwaj PP, Malykhin N, Gheorghiu D, Allen PS (2006) Measurement of brain glutamate and glutamine by spectrally-selective refocusing at 3 tesla. Magn Reson Med 55: Schulte RF, Trabesinger AH, Boesiger P (2005) Chemical-shiftselective filter for the in vivo detection of J-coupled metabolites at 3 T. Magn Reson Med 53: Ryner LN, Sorenson JA, Thomas MA (1995) Localized 2D J-resolved 1 H MR spectroscopy: strong coupling effects in vitro and in vivo. Magn Reson Imaging 13: Thomas MA, Ryner LN, Mehta MP, Turski PA, Sorenson JA (1996) Localized 2D J-resolved 1 H MR spectroscopy of human brain tumors in vivo. J Magn Reson Imaging 6: Mayer D, Spielman DM (2005) Detection of glutamate in the human brain at 3 T using optimized constant time point resolved spectroscopy. Magn Reson Med 54: Bartha R, Drost D, Menon R, Williamson P (2000) Comparison of the quantification precision of human short echo time 1 H spectroscopy at 1.5 and 4.0 tesla. Magn Reson Med 44: Tkác I, Andersen P, Adriany G, Merkle H, Ugurbil K, Gruetter R (2001) In vivo 1 H NMR spectroscopy of the human brain at 7 T. Magn Reson Med 46:

11 18. Stephenson MC, Gunner F, Napolitano A, Greenhaff PL, Macdonald IA, Saeed N, Vennart W, Francis ST, Morris PG (2011) Applications of multi-nuclear magnetic resonance spectroscopy at 7 T. World J Radiol 3: Smith SA, Levante TO, Meier BH, Ernst RR (1994) Computer simulations in magnetic resonance. An object-oriented programming approach. J Magn Reson A 106: Soher B, Semanchuk P, Young K, Todd D (2014) VeSPA simulation user manual and reference. vespa/simulation. Accessed 17 July Govindaraju V, Young K, Maudsley AA (2000) Proton NMR chemical shifts and coupling constants for brain metabolites. NMR Biomed 13: Thompson RB, Allen PS (2001) Response of metabolites with coupled spins to the STEAM sequence. Magn Reson Med 45: Dou W, Palomero-Gallagher N, van Tol MJ, Kaufmann J, Zhong K, Bernstein HG, Heinze HJ, Speck O, Walter M (2013) Systematic regional variations of GABA, glutamine and glutamate concentrations follow receptor fingerprints of human cingulate cortex. J Neurosci 33: Elywa M, Mulla-Osman S, Godenschweger F, Speck O (2012) Proton magnetic resonance spectroscopy in deep human brain structures at 7 T. J Appl Spectrosc 79: Hargreaves BA, Cunningham CH, Nishimura DG, Conolly SM (2004) Variable-rate selective excitation for rapid MRI sequences. Magn Reson Med 52: Mandal PK (2007) Magnetic resonance spectroscopy (MRS) and its application in Alzheimer s disease. Concepts Magn Reson 30: Provencher SW (1993) Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn Reson Med 30: Provencher SW (2014) LCModel and LCMgui user s manual Tkác I, Öz G, Adriany G, Ugurbil K, Gruetter R (2009) In vivo 1 H NMR spectroscopy of the human brain at high magnetic fields: metabolite quantification at 4 T versus 7 T. Magn Reson Med 62(4): Li Y, Xu D, Chen AP, Vigneron DB, Nelson SJ (2008) Proton spectroscopy of human brain at 3 T and 7 T: signal-to-noise ratio, spectral linewidth and relaxation times. In: Proceedings of the 16th scientific meeting, International Society for Magnetic Resonance in medicine, Toronto, p Hurd R, Sailasuta N, Srinivasan R, Vigneron DB, Pelletier D, Nelson SJ (2004) Measurement of brain glutamate using TEaveraged PRESS at 3 T. Magn Reson Med 51: Wijtenburg S, Rowland L, Edden R, Barker P, Phil D (2013) Reproducibility of brain spectroscopy at 7 T using conventional localization and spectral editing techniques. J Magn Reson Imaging 38(2): Seeger U, Klose U, Mader I, Grodd W, Nagele T (2003) Parameterized evaluation of macromolecules and lipids in proton MR spectroscopy of brain diseases. Magn Reson Med 49: Schaller B, Xin L, Cudalbu C, Gruetter R (2013) Quantification of the neurochemical profile using simulated macromolecule resonances at 3 T. NMR Biomed 26(5): Schaller B, Xin L, Gruetter R (2014) Is the macromolecule signal tissue-specific in healthy human brain? A 1 H MRS study at 7 tesla in the occipital lobe. Magn Reson Med 72: Clementi V, Tonon C, Lodi R, Malucelli E, Barbiroli B, Iotti S (2005) Assessment of glutamate and glutamine contribution to in vivo N-acetylaspartate quantification in human brain by (1)H-magnetic resonance spectroscopy. Magn Reson Med 54: Kakeda S, Korogi Y, Moriya J, Ohnari N, Sato T, Ueno S, Yanagihara N, Harada M, Matsuda T (2011) Influence of work shift on glutamic acid and gamma-aminobutyric acid (GABA): evaluation with proton magnetic resonance spectroscopy at 3 T. Psychiatry Res 192: Öz G, Tkáč I (2011) Short-echo, single-shot, full-intensity proton magnetic resonance spectroscopy for neurochemical profiling at 4 T: validation in the cerebellum and brainstem. Magn Reson Med 65: Walter M, Henning A, Grimm S, Schulte RF, Beck J, Dydak U, Schnepf B, Boeker H, Boesiger P, Northoff G (2009) The relationship between aberrant neuronal activation in the pregenual anterior cingulate, altered glutamatergic metabolism, and anhedonia in major depression. Arch Gen Psychiatry 66: Henry ME, Lauriat TL, Shanahan M, Renshaw PF, Jensen JE (2011) Accuracy and stability of measuring GABA, glutamate, and glutamine by proton magnetic resonance spectroscopy: a phantom study at 4 tesla. J Magn Reson 208: