1H-MRS (magnetic resonance spectroscopy) is an MRI technique that particularly benefits from moving to higher magnetic field strength, because of SNR improvements that allow the use of smaller voxels, mitigating line-broadening due to magnetic field inhomogeneities. There is strong interest in quantifying GABA and Glutamate (Glu), because they are altered in many neurological pathologies. However, these metabolites are among the most difficult to detect, due to their low concentration, and overlapping peaks. Detection of GABA and Glu improves with a short echo-time (TE), because of short T2 times and J-coupling¹. The purpose of this work was to develop an optimized 1H-MRS pulse sequence for quantifying GABA and Glu in selected anatomical sub-structures of the mouse brain as small as 3.6mm³ in less than 20 minutes.

Setup and sequence optimization: All MRS measurements were performed on a 9.4 T Bruker Biospec scanner with a horizontal bore of 20 cm diameter (Biospec 94/20 USR, Bruker Biospin, Germany). We decided to employ a dual-channel cryogenic surface transceiver RF coil (CP) for improved signal-to-noise ratio. However, a limitation of this coil is that its power-limitations are relatively low (max. 20W) rendering the 180^o pulses of the widely used PRESS sequence relatively long. In particular, due to the inhomogeneous flip angle profile of the CP, more pulse power is required to create a desired flip angle physically farther away from the coil (ventral) than closer (dorsal). The STEAM sequence employs only 90^o pulses, allowing a much shorter TE with the same pulse power, but suffers from a reduction of the SNR by a factor of 2 compared to PRESS (at the same TE). To compare the two sequences we optimized them with respect to minimum TE using a 3.6mm³ voxel placed in the basal ganglia (BG; ventral location) resulting in TE=2.6ms for STEAM and TE=11.5ms for PRESS. Other parameters were identical: VAPOR 250Hz; 4096 points; 6010Hz. The sequence TR was optimized to yield the highest GABA/Glu signal per time interval resulting in TR=2s for T1=1.5s.² With 512 averages this resulted in a measurement time of 17:12 min:sec. The PRESS sequence had a maximal (fat-water) spatial displacement of 1.19mm, the STEAM sequence 0.38mm.

Data acquisition: Experiments were performed in a SJL/J mouse and were approved by our Institutional Animal Care & Use Committee. Animals were anesthetized using 1-3% isoflurane under monitoring of respiration rate and body temperature. We applied both optimized sequences in two different brain regions with exactly equal voxel prescriptions: frontal cortex and BG with voxel volumes of 3.8 and 3.6mm³, respectively. Localized iterative shimming was performed before each MRS acquisition. A water spectrum was acquired before each sequence for eddy-current correction. To understand the benefit of using CP, we repeated the experiment with a standard room temperature (RT) cross-coil configuration employing a quadrature volume coil for excitation and a four-channel surface array coil for signal detection.

Data analysis: All spectra were analyzed using LCModel(v6.3) using appropriate basis sets for the two sequences. SNR of spectra was evaluated using the following relation [cf. Eq. section 2.4 in Ref. 5]: $$\frac{\textrm{maximum peak height}-\textrm{baseline-fit}}{\textrm{analysis window} _{ppm}*2*\textrm{RMS}(\textrm{fit residual})}$$

Excellent spectral quality was obtained with both sequences using the CP configuration (Fig.1). The PRESS spectrum had a lower noise level but the overlapping peaks of GABA and Glu were better discernible in the STEAM spectrum due to the lower TE³. SNR levels of STEAM and PRESS were 9 and 10, respectively. The LCModel quantification yielded similar Cramér–Rao bounds (CRB) for GABA and Glu with both sequences, between 2% and 8%.

Fig.2 shows a comparison of the CP setup with the RT setup. The SNR was significantly reduced compared to the CP, with 3 for STEAM and 4 for PRESS. Due to the low SNR quantification of GABA and Glu was not possible (CRB>20%).

Fig.3 and 4 show voxel prescriptions and exemplary LCModel fits of the CP-based STEAM data in the BG and cortex.

We demonstrated the feasibility of GABA and Glu quantification in anatomical sub-structures of the mouse brain with very low CRBs in 17 minutes acquisition time. STEAM and PRESS sequences yielded similar CRBs, but the PRESS had rather poor voxel localization compared to the STEAM sequence⁴, with a maximum displacement as large as the voxel edge length. The quantification of GABA and Glu was ultimately enabled by using the CP, which yielded a three fold improved SNR. To achieve the same SNR with RT, the number of averages would have to be an order of magnitude higher, leading to an impractical acquisition time of 3 hours.