Regional Neurochemical Profiles in the Human Brain measured by ¹H MRS at 7 Tesla using Local B₁ Shimming

Abstract

Increased sensitivity and chemical shift dispersion at ultra-high magnetic fields (UHF) enable precise quantification of an extended range of brain metabolites from ¹H MR spectra. However, all previous neurochemical profiling studies using single-voxel MRS at 7 T were limited to data acquired from the occipital lobe with half-volume coils. Challenges of ¹H MRS of the human brain at 7 T include short T₂ and complex B₁ distribution that imposes limitations in maximum achievable B₁ strength. In this study, the feasibility of acquiring and quantifying short-echo (TE = 8 ms), single voxel ¹H MR spectra from multiple brain regions was demonstrated by utilizing a 16-channel transceiver array coil with 16 independent transmit channels allowing local transmit B₁ (B₁⁺) shimming. Spectra were acquired from volumes-of-interest of 1 – 8 mL in brain regions that are of interest for various neurological disorders: frontal white matter, posterior cingulate, putamen, substantia nigra, pons and cerebellar vermis. Local B₁⁺ shimming substantially increased transmit efficiency, especially in the peripheral and ventral brain regions. By optimizing a STEAM sequence for utilization with a 16-channel coil, artifact-free spectra were acquired with a small chemical shift displacement error (< 5% /ppm/direction) from all regions. The high SNR enabled the quantification of neurochemical profiles consisting of at least 9 metabolites including GABA, glutamate and glutathione in all brain regions. Significant differences in neurochemical profiles were observed between brain regions. For example, GABA levels were highest in the substantia nigra, total creatine highest in the cerebellar vermis and total choline highest in the pons, consistent with known biochemistry of these regions. These findings demonstrate that single voxel ¹H MRS at UHF can reliably detect region-specific neurochemical patterns in the human brain and has the potential to objectively detect alterations in neurochemical profiles associated with neurological diseases.

Introduction

The advantages of ultra-high field (UHF) MR have been demonstrated in a variety of MR acquisition modalities (1). MRS can particularly benefit from substantial gains in signal-to-noise ratio (SNR) and spectral resolution at UHF, enabling the quantification of numerous metabolites from small volumes-of-interest (VOIs). For example, concentrations of up to 18 neurochemicals can be quantified from selected regions in the human brain at 7 T (2–4). SNR is approximately two-fold and resolution 14% higher at 7 T relative to 4 T in the occipital cortex (2). As a result, metabolites are quantified with higher precision (lower Cramér-Rao Lower Bounds, CRLB) at 7 T than at 4 T, which is particularly critical for weakly represented neurochemicals (2). Despite this potential, short-echo ¹H MRS studies that reported neurochemical profiles of the human brain at 7 T were primarily focused on the occipital lobe (2–8) due to multiple challenges faced when using a standard volume head coil at UHF. First, the power to generate B₁ increases approximately linearly with magnetic field (9). As a result, the maximum magnitude of B₁⁺ achievable with conventional volume head coils and available RF power are insufficient to generate conventional 90° or 180° RF pulses with bandwidths large enough to minimize chemical shift displacement errors in spectroscopy. This problem is exacerbated in large regions of the brain due to the destructive interferences that result in submaximal B₁⁺ when volume coils circumscribing the whole head are used (10). Second, T₂ relaxation times of metabolites in the human brain are relatively short at this field (11,12) and the SNR advantages quickly disappear with increasing echo times. Therefore, short echo times are critical for quantifying neurochemical profiles and are particularly important for reliable quantification of metabolites with coupled spin systems, such as glutamine (Gln), glutamate (Glu) and glutathione (GSH).

New excitation and saturation pulses can be designed to partially overcome problems resulting from limitations in available B₁⁺ magnitude. For example, MR spectroscopic imaging (MRSI) data were recently obtained using the FIDLOVS technique (13), based on 2D pre-localization by outer volume suppression (OVS) and slice-selective excitation with a newly designed broadband frequency-modulated RF pulse. Alternatively, the RF power required to achieve a given magnitude of B₁⁺ in a local area can be reduced by utilizing an array of transmit coil elements and optimizing the phase and/or magnitude of RF delivered to each transmit element (B₁⁺ shimming) (10,12,14,15). Recently, a similar approach has been employed to acquire MRSI data from brain slices at 7 T, demonstrating the feasibility of detecting myo-inositol (myo-Ins), Glu and Gln, in addition to N-acetylaspartate (NAA), creatine and choline (16,17). In these MRSI studies (13, 16, 17) relative metabolite concentrations were reported because acquisition of water reference data needed for absolute quantification would double the scan time. Another challenge with obtaining MRSI data for metabolite quantification is that efforts for improving the homogeneity of B₁⁺ over a large volume generally result in decreased RF efficiency (18). On the other hand, in the single voxel spectroscopy approach, high RF efficiency to maximize B₁⁺ over small VOIs can be readily achieved using local B₁⁺ shimming, as was demonstrated in the human head at 9.4 T (12).

Another advantage of the single voxel approach is the ease with which the static magnetic field homogeneity can be adjusted. B₀ shimming is a key factor for successful MRS at 7 T, because it determines the achieved spectral resolution, which consequently affects the reliability of metabolite quantification. It has been demonstrated that a strong second-order shim system is sufficient to compensate B₀ inhomogeneities in reasonably small volumes (2,7). On the other hand, to achieve acceptable linewidths over the slice, the MRSI studies mentioned above (13,16,17) utilized third order B₀ shimming, a feature which is absent from most 7 T scanners, including our equipment. Therefore, in the absence of third or even higher order shims to attain sufficient B₀ homogeneity over large volumes in the brain, single voxel spectroscopy provides an alternative to MRSI for acquisition of very high quality spectra.

Finally, single voxel MRS might be preferable over the MRSI approach for clinical applications where the affected brain region is known and where motion artifacts are common (19–21). The phase and frequency instabilities induced by patient motion can be eliminated from single voxel MRS data by post-processing, provided that single scan spectra are clean, and uncorrectable scans can be removed from the final FID summation (22). In addition, the single voxel approach allows one to abort the acquisition at any time and use the limited data set for metabolite quantification.

In this study, we tested the feasibility of acquiring and quantifying short-echo (TE = 8 ms), single voxel ¹H MR spectra from multiple brain regions utilizing the STEAM sequence (7,22), a 16-channel transceiver array coil (23) and local B₁⁺ shimming at 7 T. Our goals were 1) to determine if artifact free spectra of similar quality to those reported in the occipital lobe (2,7) could be obtained from brain regions that are of interest for various neurological disorders, namely the frontal white matter (demyelinating white matter diseases), posterior cingulate (Alzheimer’s disease), putamen (Huntington’s and Parkinson’s diseases), substantia nigra (Parkinson’s disease), pons (spinocerebellar ataxias, motor neuron diseases) and cerebellar vermis (spinocerebellar ataxias), and 2) to quantify neurochemical profiles from these VOIs and to compare them to prior biochemical literature regarding neurochemical patterns in these regions.

Methods

Subjects

Fourteen healthy volunteers (8 females, 6 males, age (mean ± SD) = 32 ± 10) participated in this study after giving informed consent using procedures approved by the Institutional Review Board: Human Subjects Committee of the University of Minnesota.

MR Protocol

All measurements were performed on a 90-cm bore 7 T magnet (Magnex Scientific, Abingdon, UK) equipped with a head or body gradient coil (a maximum gradient amplitude of 40 mT/m and a slew rate of 170 mT/m/ms) and interfaced to a Siemens Syngo console (Siemens, Erlangen, Germany) with sixteen independent RF receivers. The body gradient coil included second-order shim coils with maximum shim strengths of Z² = 6.2 mT/m² and ZX, ZY, XY and X²-Y² = 2.5 mT/m² at a current of 20 A, while all second order shim coils in the head gradient system had a strength of 6.2 mT/m² at a current of 20 A. A 16-channel transceiver array coil (23) was driven by sixteen 1 kW RF amplifiers (CPC, Brentwood, NY) which allowed independent control of phases and/or magnitudes of each transmit element of the array for B₁⁺ shimming (24). 1 kW RF amplifiers were calibrated to deliver 900 W peak power, however only half of the transmitted power was actually available at the coil ports due to losses in cables and connectors. An in house built 16-channel monitoring system was used to continuously monitor the forward and reflected RF power on each channel which was averaged over two moving windows of 10 seconds and 10 minutes (12). If the power input into any coil element exceeded the predefined FDA threshold (3 / 16 Watts/kg/channel, conservatively assuming power deposition in only the head weighing ~ 4 kg), the scan was aborted. Besides the continuous RF power monitoring used during in vivo experiments, numeric electromagnetic simulations were also used to verify that global and local specific absorption rates (SAR) corresponding to our B₁⁺ shimming approach were in compliance with U.S. Food and Drug Administration (FDA) guidelines. The 16-channel transceiver array coil loaded with a human head was modeled in the finite difference time domain (XFDTD, Remcom Inc, PA, USA). B₁⁺ shimming solutions were performed within this model for VOIs defined in each of the anatomical target used in experiments, using the same B₁⁺ phase shimming algorithm. For each corresponding B₁⁺ shim solution the input power was scaled to match the experimentally measured B₁⁺ value. Using the exact shape, amplitude and duration of each RF pulse used in the sequence for in vivo experiments, with same TR, SAR was then calculated over the head model. For each B₁⁺ solution, global (over the whole head) and local (averaged over 1g of tissue) SAR values were always found smaller than 3 Watts/kg and 8 Watts/kg, respectively, within FDA guidelines (data not shown) (25).

Before the MRS acquisition from each VOI, local B₁⁺ shimming was performed over a region of interest (ROI) that encompassed the VOI. Low flip angle gradient echo (GRE) images were acquired for B₁⁺ shimming when pulsing the RF power through one coil at a time, while the receive signal was sampled on all 16 channels (field of view (FOV) = 168 × 256 mm², repetition time (TR) = 60 ms, echo time (TE) = 3.6 ms, nominal flip angle (FA) = 15°, slice thickness 4 mm). The relative transmit B₁⁺ phases were calculated using the B₁⁺ phase formalism introduced by Van de Moortele et al (10). Local B₁⁺ phase shimming was performed by averaging the phase of each channel over a ROI and then subtracting the normalized phase from the respective channel to minimize the destructive phase cancellation and maximize the fraction of available B₁⁺. Note that the fraction of available B₁⁺, which was defined as the ratio between the magnitude of the sum of the complex B₁⁺ of each transmit channel and the sum of their magnitudes (14), can vary between 0 (when interferences are completely destructive) and 1 (when interferences are completely constructive). The predicted relative transmit efficiency (RTE) achievable with the local B₁⁺ phase shimming technique relative to an initial transmit phase (set in 22.5 degree steps between adjacent elements) was calculated, based on the relative B₁⁺ maps, from the postshim to preshim ratio of fraction of available B₁⁺ averaged over the ROI (14). Note that only the phase, but not the magnitude, of each transmit channel was modulated relative to the others with our B₁⁺ shimming algorithm.

Sagittal and transverse multislice images were acquired with a turbo spin echo sequence (FOV = 180 × 180 mm², TR = 3 s, TE = 93 ms, FA = 150 °, slice thickness 3 mm, 48 slices, one average) for selection of the VOI. First- and second-order shims were adjusted using FASTMAP (fast, automatic shimming technique by mapping along projections) with echo-planar imaging (EPI) readout (26). Spectra were measured by a short-echo STEAM sequence (TR = 5 s, TE = 8 ms, mixing time (TM) = 32 ms). The STEAM sequence combined with 3-D OVS and VAPOR (variable-power RF pulses with optimized relaxation delays) water suppression previously developed using a half-volume coil (22) was utilized with the following modifications: The asymmetric RF pulses used for STEAM slice selection were replaced by symmetric sinc RF pulses with broader bandwidth (1.5 ms, 5.9 kHz bandwidth) requiring a peak power B₁ of 34 µT (γB₁/2π = 1.46 kHz). In the previous work, a higher RF power was used for the OVS slab in y-dimension further away from the coil plane than the slab near the coil to account for the decrease in B₁⁺ field with distance from the half-volume coil, whereas the same power magnitudes were used for the OVS pulses (hyperbolic secant, 5.0 ms, 8.4 kHz bandwidth) in each dimension here. The VAPOR water suppression parameters were unchanged from before (22) while crusher gradients were further optimized to provide artifact free spectra from all VOI. After the local B₁⁺ phase shimming procedure for each VOI, the RF power magnitude for the 90° STEAM pulses was calibrated by monitoring the signal intensity while increasing the RF power and choosing the RF power setting that produced the maximum signal. The power magnitudes of OVS and VAPOR RF pulses were automatically set relative to the 90° STEAM pulses. Excellent water suppression in all brain regions, using the RF power setting derived from the STEAM RF power calibration, confirmed that a 90° flip angle of STEAM pulses, requiring a B₁⁺ field of 34 µT was always reached. Spectra were acquired from 7 VOIs: occipital cortex (20 × 20 × 20 mm³, number of transients (NT) = 128), posterior cingulate (20 × 20 × 20 mm³, NT = 128), frontal white matter (20 × 20 × 20 mm³, NT = 128), putamen (13 × 13 × 15 mm³, NT = 256), substantia nigra (6 × 13 × 13 mm³, NT = 512), pons (15 × 15 × 15 mm³, NT = 256) and cerebellar vermis (10 × 25 × 25 mm³, NT = 128), each from 5 healthy volunteers. The unsuppressed water signal acquired from the same VOI was used to remove residual eddy current effects and to reconstruct the phased array spectra (27). This included weighting the spectra based on the sensitivity of each receive element at the VOI and correcting for the different constant phase shift terms of the complex spectra prior to the summation. Single scan spectra summed from 16 channels were corrected for frequency and phase variations induced by subject motion and then summed. When insufficient SNR did not permit correction of single acquisitions (in the pons, putamen and substantia nigra) data were averaged over 4 (pons and putamen) or 32 (substantia nigra) transients, corrected for frequency shifts and summed.

Metabolite quantification

Metabolites were quantified using LCModel (28,29). The model spectra of alanine (Ala), aspartate (Asp), ascorbate/vitamin C (Asc), glycerophosphocholine (GPC), phosphocholine (PC), creatine (Cr), phosphocreatine (PCr), γ-amino-butyric acid (GABA), glucose (Glc), Gln, Glu, GSH, myo-Ins, lactate (Lac), NAA, N-acetylaspartylglutamate (NAAG), phosphoethanolamine (PE), scyllo-inositol (scyllo-Ins) and taurine (Tau) were generated based on previously reported chemical shifts and coupling constants (30,31). Macromolecule spectra were acquired from the occipital cortex of 5 volunteers using an inversion recovery sequence (TR = 2 s, inversion time = 0.680 s) (32). Metabolite concentrations were obtained relative to an unsuppressed water spectrum acquired from the same VOI assuming a water content of 72% for frontal white matter and pons, 78% for putamen, 76% for substantia nigra and 82% for occipital cortex, posterior cingulate and cerebellar vermis, which primarily contain gray matter (33–35). Concentrations were not corrected for T₁ and T₂ effects and cerebrospinal fluid contribution to the VOI. Metabolites quantified with CRLB (estimated error of the metabolite quantification) > 50% were classified as not detected. Only metabolites quantified with CRLB ≤ 50% in at least three of the five spectra from a brain region were included in the final neurochemical profile. If the correlation between two metabolites was consistently high (correlation coefficient < −0.5) in a given region, their sum was reported, e.g. total creatine (Cr+PCr), total choline (GPC+PC) and Glc+Tau.

Statistics

One-way analysis of variance (ANOVA) with Tukey post hoc test was used to compare the concentrations at each of the seven locations for each metabolite.

Results

B₁⁺ optimization

ROIs larger than and concentric with the selected VOIs were chosen to optimize the phases of the B₁⁺ fields transmitted from each of the 16 channels. Two typical ROIs defined for local B₁⁺ shimming are shown on transverse anatomical images in Figure 1A (occipital cortex) and D (substantia nigra). The phases of channels 1 through 16 were set to 0 to 360° in increments of 22.5° before local B₁⁺ shimming. Low flip angle GRE images were utilized to generate relative B₁⁺ maps both before and after implementation of local B₁⁺ shim phases to evaluate the predicted RTE within the ROI. For instance, in the occipital cortex, a region in the periphery of the brain, 34% of the available B₁⁺ was realized before B₁⁺ shimming (Fig. 1B). After B₁⁺ shimming, 89% of the available B₁⁺ was realized within the same ROI (Fig. 1C), corresponding to a predicted RTE of 2.64. On the other hand, in the substantia nigra, a more central region in the brain, the fractions of available B₁⁺ before and after B₁⁺ shimming were 73% and 84%, respectively, and the predicted RTE was 1.16. The averages of predicted RTE values are shown in Table 1 for all brain regions. These values were even higher when calculating the average B₁⁺ after local B₁⁺ shimming only within the VOI used for spectroscopy, since the maximum B₁⁺ efficiency was obtained in the center of the ROI (Fig. 1). That a B₁⁺ of at least 34 µT was available for all brain regions was confirmed directly by calibration of the RF pulses in STEAM.

Figure 1.

Local B₁⁺ shimming at the periphery (a) and center of one subject’s brain (d). The ROIs (red circles) were used to shim B₁⁺ for the VOIs (white squares); (a) occipital cortex and (d) substantia nigra. Maps of the fraction of available B₁⁺ before (b, e) and after (c, f) B₁⁺ shimming are shown for the occipital cortex (b,c) and substantia nigra (e, f).

Table 1.

Relative transmit efficiencies (RTE), i.e. ratios (mean ± SD) of the B₁⁺ maps after to before B₁⁺ shimming, averaged over the different regions of interest.

Region of Interest	RTE (After / Before B1+ shimming)
Occipital cortex	2.5 ± 0.2
Posterior cingulate	1.4 ± 0.4
Frontal white matter	1.9 ± 0.2
Putamen	1.6 ± 0.2
Substantia nigra	1.2 ± 0.1
Pons	1.3 ± 0.1
Cerebellar vermis	3.3 ± 0.3

Pulse sequence optimization

The modifications in the STEAM sequence relative to the one previously developed using a half-volume coil (22) provided optimum performance with the 16-channel transceiver array coil. First, the broader bandwidth symmetric sinc RF pulses used for slice selection resulted in slightly reduced chemical shift displacement error (< 5% /ppm/direction). Second, equal OVS power magnitudes in each dimension provided optimum suppression of signal from outside of the VOI because the B₁⁺ field of the 16-channel coil after local B₁⁺ shimming was almost symmetric around the VOI in every dimension. Third, the VAPOR water suppression parameters previously optimized for minimum sensitivity to T₁ and B1⁺ variation (22) provided optimum water suppression with the 16-channel coil as well. Finally, crusher gradients used for STEAM, OVS and VAPOR adjusted to suppress unwanted stimulated echoes resulted in artifact-free spectra from all brain regions studied (Fig. 2).

Figure 2.

¹H MR spectra obtained with STEAM (TR = 5 s, TE = 8 ms, TM = 32 ms) from seven VOIs. The spectra were scaled based on neurochemical concentrations obtained by LCModel. Processing: Reconstruction of single scan FIDs from phased array data, frequency and phase correction of FID arrays, FID summation, correction for residual eddy current effects, Gaussian multiplication (σ = 0.1 s), Fourier Transform (FT), zero-order phase correction. Positions of the VOIs are shown on T₂ – weighted images. (NT = number of transients)

Spectral quality

Figure 2 shows representative spectra obtained from different brain regions at 7 T. In all brain regions, the residual water signal was smaller than the major metabolite peaks. In addition, the double localization accomplished by STEAM and OVS eliminated signals from outside the VOI, such as lipid signals from the subcutaneous tissue, which resulted in artifact-free spectra with a flat baseline. This excellent quality was consistently achieved in all brain regions (Fig. 2). Localized B₀ shimming using FASTMAP resulted in water linewidths (full width at half maximum (FWHM)) ranging from 11 Hz (in posterior cingulate) to 18 Hz (in substantia nigra) (Table 2).

Table 2.

Water signal linewidths (mean ± SD) achieved in the different volumes of interest.

Volume of Interest	FWHM (Hz)
Occipital cortex	12.7 ± 0.3
Posterior cingulate	11.3 ± 0.7
Frontal white matter	15.7 ± 0.8
Putamen	15.9 ± 1.2
Substantia nigra	18.0 ± 1.0
Pons	14.3 ± 1.9
Cerebellar vermis	14.5 ± 0.6

Neurochemical profiles

The mean concentrations of metabolites that met our criteria for reliable quantification in at least six of the seven brain regions are shown in Figure 3A. Both the sum and individual values are reported for NAA and NAAG because these metabolites could be reliably distinguished (correlation coefficient between concentrations less negative than −0.5, Fig. 4) in all regions with CRLB < 10% except substantia nigra and putamen (correlation coefficient ~ −0.65, CRLB < 30%). Eight other metabolites (GABA, Glu, GSH, myo-Ins, scyllo-Ins, Cr+PCr, GPC+PC and Glc+Tau) were quantified with average CRLB ranging between 1–35% in all brain regions (Fig. 3B). Asc, Gln and Lac were quantified reliably in all brain regions (CRLB < 30%) except substantia nigra.

Figure 3.

Neurochemical profiles and corresponding CRLBs from different brain regions determined by LCModel fitting of short-echo STEAM spectra acquired at 7 T. (a) Only metabolites that were quantified reliably (with CRLB ≤ 50% in at least three of the five spectra from a brain region) in at least six of the seven brain regions are shown. Metabolites that were significantly different between regions are marked with: * p < 0.005 (ANOVA). (b) CRLBs expressed in relative units (%). Error bars: inter-subject SD (N=5 for all brain regions).

Figure 4.

Mean correlation coefficients matrices between the metabolite concentrations in different brain regions determined by LCModel fitting of short-echo STEAM spectra acquired at 7 T. Non-zero off-diagonal elements of the correlation coefficients matrices indicate the level of covariance between the fitting results for the corresponding two metabolites.

The comparison of metabolite concentrations revealed neurochemical profiles characteristic of different brain regions (Fig. 3A). For instance, the GABA level in the substantia nigra (2.5 ± 0.15 µmol/g) was 2–3 fold higher (p < 0.001) than the other brain regions (0.5 – 1.8 µmol/g). The cerebellar vermis had the highest total creatine (12.8 ± 0.8 µmol/g, p < 0.001) and Glc+Tau level (5.1 ± 0.1, µmol/g, p < 0.001). A high (GPC+PC)/(Cr+PCr) ratio was apparent from the pons spectrum (Fig. 2), which resulted primarily from higher total choline (2.9 ± 0.3 µmol/g) than other regions. Total NAA (NAA + NAAG) was fairly uniformly distributed between regions (11.6 – 13.6 µmol/g). The lowest Glu level was in the substantia nigra (5.9 ± 0.5 µmol/g) and the highest Glu levels were observed in the posterior cingulate and putamen (10.8 ± 0.5 and 10.7 ± 0.3 µmol/g, respectively). The highest myo-Ins levels were quantified in the pons and the cerebellar vermis (8.4 ± 0.7 and 9.1 ± 1.1 µmol/g, respectively). The GSH level varied from 0.50 ± 0.1 µmol/g (occipital cortex) to 1.2 ± 0.2 µmol/g (cerebellar vermis) and the scyllo-Ins varied from 0.2 ± 0.1 µmol/g (frontal white matter) to 0.5 ± 0.1 µmol/g (substantia nigra).

Discussion

This study demonstrated that high quality single-voxel short-echo STEAM spectra can be obtained from clinically relevant brain regions at 7 T using a transceiver array coil and local B₁⁺ shimming. Importantly, our 7 T scanner is operated with a clinical console, demonstrating the feasibility of implementing the technology on clinical platforms. The spectral quality previously reported in the occipital cortex at 7 T using a half-volume RF coil (2,7) was reproduced with a 16-channel transceiver array coil (Fig. 2). In addition, spectra of high quality were acquired from other brain regions affected in various neurological disorders. Neurochemical profiles quantified from these spectra were consistent with biochemical literature (see below). To our knowledge, these extended profiles of region specific neurochemical concentrations are reported at 7 T for the first time.

Previously, single voxel, short-echo ¹H MRS studies at 7 T utilized half-volume coils and therefore reported neurochemical profiles of only the occipital cortex in humans (2–7,22). Flexibility in selecting VOIs, especially deeper in the brain requires the use of volume head coils for both transmission and detection. This, however, creates challenges because the B₁ distribution in the human head for a volume coil is extremely non-uniform (9,10,36,37). In the center of the brain, transmit and receive B₁ associated with individual current carrying elements of the volume coil add approximately in phase resulting in highest possible B₁ (10,36); however, B₁ fields generated by the individual current carrying elements tend to cancel each other in the periphery and ventral regions of the brain due to destructive interferences. In this study, we utilized a transceiver array coil driven by a multichannel transmit system, which allowed local B₁⁺ shimming to minimize the B₁⁺ losses. Optimizing the transmit phases in the individual coil channels enabled to maximize the available B₁⁺ magnitude in the selected VOIs (Fig. 1). The predicted RTE values ranged from 1.2 in the substantia nigra to 3.3 in the cerebellar vermis (Table 2). These results are in agreement with the fact that at UHF, constructive interferences yield increased magnitude of B₁⁺ field in the center of the brain, whereas destructive interferences cause a reduction in magnitude of B₁⁺ field in the periphery of the brain (10). Thus, in this study, VOIs selected in the periphery of the brain and especially in the cerebellum benefited more substantially from B₁⁺ shimming than VOIs in the center of the brain.

The hardware available in this study (a 16-channel transceiver array coil driven by sixteen 1 kW RF amplifiers) and improved RTE by local B₁⁺ shimming provided a transmit field of at least 34 µT in all studied brain regions, which was confirmed directly by calibration of the RF pulses in STEAM. The available peak B₁⁺ field combined with high bandwidth RF pulses resulted in a reasonably low chemical shift displacement error (< 5% /ppm/direction). In general, if sufficiently high B₁⁺ is not achievable, extending the duration of RF pulses in the MRS localization sequence is necessary. However, chemical shift displacement errors at UHF quickly become unacceptably large with increasing RF pulse length and decreasing bandwidths when conventional, amplitude modulated RF pulses are utilized. Another possibility for reducing the chemical shift displacement error when B₁⁺ is low is to use broadband adiabatic RF pulses for slice selection (38,39). However, slice selective adiabatic excitation pulses are currently not available; therefore the localization must be achieved with multiple inversion pulses at the expense of much longer TE as in the LASER technique (38) or by forgoing single shot acquisition as in the ISIS method (40,41).

For this study, the STEAM sequence was chosen despite the loss of half of the available signal, primarily to minimize relaxation effects at an ultra-short echo time, but also to have a low chemical shift displacement artifact and to compare spectral quality and metabolite quantification to previously published work in the occipital cortex at 7 T (2–4,7,22). Transverse relaxation at long TE results in substantial signal loss at higher fields since the Hahn T₂ of metabolites decreases with increasing field strength due to an increased dynamic dephasing contribution (11,12). For instance, the T₂ of NAA decreases from 240 ms at 4 T to 160 ms at 7 T (11). Thus, even a modest increase of TE from 8 to 30 ms would lead to a 12% lower NAA signal at 7 T. In addition, J-evolution was minimized at the ultra-short TE, which facilitated the quantification of metabolites with coupled spin systems.

To achieve high spectral resolution at 7 T, efficient B₀ shimming is very important. In this study, the strength of the second-order shim system combined with efficient B₀ shimming (FASTMAP) resulted in highly reproducible linewidths in the selected VOIs. This was shown by less than 15% standard deviation of the water signal linewidth in all brain regions (Table 2). Considering an increase in linewidth at 7 T by a factor of 1.5 relative to 4 T (2), the linewidths obtained in the substantia nigra were in agreement with published results for substantia nigra at 4 T (21). Similarly, the previously reported linewidths for the occipital cortex at 7 T (2) were reproduced in the current study. Linewidths obtained in the substantia nigra, putamen and frontal white matter were substantially higher than those in the other VOIs due to the effects of high iron content in the substantia nigra and putamen (42) and global magnetic field inhomogeneity in frontal white matter due to the close proximity of the nasal air cavities.

The achieved spectral quality (high spectral resolution, SNR, excellent localization performance, efficient water suppression, and a distortionless baseline) allowed reliable quantification of 17 metabolites in occipital cortex (VOI = 8 mL) using LCModel analysis. Concentrations of metabolites quantified in this study were in good agreement with recently reported values acquired from the same brain location using a very similar approach except for the RF coil (2). The average difference between concentrations obtained in the current study and the prior report was only 0.25 µmol/g.

Nine metabolites were reliably estimated using LCModel analysis in all studied brain regions. In addition, nine metabolites were quantified with average CRLB less than 20% in 5 of the 7 brain regions studied (Fig. 3B), indicating that these were the most robustly quantified neurochemicals. In general, changes in neurochemical levels greater than twice the CRLB can be detected in individual spectra with 95% confidence (43). Hence, disease related alterations in those metabolites quantified with CRLB above 20% can only be detected if the effect sizes are quite large (more than twice the CRLB).

Comparison of metabolite concentrations between different brain regions revealed significant regional variations (Fig. 3). The findings were in excellent agreement with prior neurochemical and MRS studies of anatomical locations most similar to those presented here. For example, the high cerebellar total creatine level was in agreement with previous reports (20,44–46). In addition, the highest Glc+Tau concentration obtained in cerebellar vermis was consistent with high cerebellar Tau (47) and Glc (48) levels reported relative to other brain regions. Confirming previous MRS studies (44,49), the highest total choline levels were observed in the pons and the lowest total choline levels were measured in the occipital cortex. The concentration of total NAA had a fairly homogenous distribution, consistent with previous MRS studies (44,49). The GABA level was significantly higher in the substantia nigra than other regions, also in agreement with both MRS (21) and neurochemistry literature (47). High myo-Ins concentrations in the pons and vermis were also consistent with prior MRS literature (46).

In conclusion, the combination of a multichannel transceiver array coil and local B₁⁺ shimming enabled us to acquire excellent single voxel, short-echo STEAM spectra from multiple brain regions at 7 T. The high spectral quality facilitated the quantification of neurochemical profiles consisting of at least nine metabolites including GABA, Glu and GSH. These findings demonstrate that single voxel ¹H MRS at 7 T can reliably detect region-specific neurochemical patterns in the human brain and has the potential of becoming an objective tool to monitor neurochemical profile alterations associated with several neurological diseases.

Abbreviations Used

Ala

alanine

Asc

ascorbate

Asp

aspartate

Cr

creatine

CRLB

Cramér-Rao lower bounds

FA

flip angle

FOV

field of view

FWHM

full width at half maximum

GABA

γ-aminobutyric acid

Glc

glucose

Gln

glutamine

GPC

glycerophosphocholine

GRE

gradient echo

GSH

glutathione

Glu

glutamate

Lac

lactate

MRSI

magnetic resonance spectroscopic imaging

myo-Ins

myo-inositol

NAA

N-acetylaspartate

NAAG

N-acetylaspartylglutamate

NT

number of transients

OVS

outer volume suppression

PC

phosphocholine

PCr

phosphocreatine

PE

phosphoethanolamine

ROI

region of interest

RTE

relative transmit efficiency

SAR

specific absorption rate

scyllo-Ins

scyllo-inositol

SNR

signal-to-noise ratio

Tau

taurine; TE, echo time

TE

echo time

TM

mixing time

TR

repetition time

UHF

ultra high field

VAPOR

variable-power RF pulses with optimized relaxation delays

VOI

volume of interest