Introduction

Gamma-aminobutyric acid (GABA) is a key metabolite involved in inhibitory neurotransmission. Despite advancements in human GABA spectral editing, it is frequently assumed that the high B₀ of preclinical scanners is sufficient to detect GABA without spectral editing. Often, in preclinical GABA MRS studies, conventional sequences (PRESS¹) are used. Comparison between preclinical PRESS and spectral editing methods have not previously been reported. We compared MEGAPRESS² and HERMES³ against PRESS and assessed their GABA discrimination efficiency.

Methods

Data were acquired on a 9.4T Bruker BioSpec 94/20 MR scanner with ParaVision 6.0.1, using an 86-mm volume resonator and phased array mouse head receive-only coil. Fourteen mice (control and drug-treated mice, n=7 per group) and three in vitro samples containing GABA (Sigma Aldrich) concentrations of 2 mM, 10 mM, and 20 mM in phosphate-buffered saline were scanned. An increase in GABA concentration levels was induced using aminooxyacetic acid⁴ (AOAA). Animal work was in accordance with the Animals (Scientific Procedures) Act 1986, under standard laboratory conditions. Four hours prior to their spectroscopy acquisition, animals were dosed with saline (10ml/kg i.p.) or AOAA, (100mg/kg i.p). Initially, mice were anesthetized with isoflurane (5%) in oxygen/air (4/10) and then switched to subcutaneous medetomidine bolus (0.05mg/kg) followed by medetomidine infusion (0.1mg/kg/h)) with 0.4% isoflurane. Internal body temperature (37±0.5^o C) and breathing rate (~70-150 breaths/minute) were continuously monitored and maintained with a rectal temperature probe, heated pad, and thoracic respiratory sensor.
The MR acquisition protocol consisted of localizers, structural scans and three spectroscopy acquisitions (PRESS, MEGAPRESS, HERMES). The spectroscopy voxel was placed in the phantom centre (3x3x3mm³) or over bilateral striata (2x1.5x6mm³, Dorsal/Ventral x Rostral/Caudal x Right/Left) (Figure 1). B0-map and localized shimming and frequency adjustments were applied before data acquisition. PRESS, MEGAPRESS and HERMES acquisition parameters were: TR 2s; TE 8.61/68/80ms; 4.4kHz bandwidth; 4096 data points; 640 averages per acquisition; working frequency 2.7ppm; VAPOR water suppression. MEGAPRESS and HERMES editing pulses were created using a template pulse shape (8.5ms duration, 200Hz bandwidth, 180° flip angle) with additional cosine modulation for dual-band pulses. Editing frequencies were 1.9ppm (GABA), 4.56ppm (GSH) and 7.5ppm (the non-edited condition). Acquisition time per sequence was 21’20”. Mean (SD) in vivo water linewidths were 17.73 (2.38) Hz.
Data were imported into Matlab for pre-processing with a custom-written pipeline based on FID-A⁵ scripts and functions. PRESS and spectrally edited data underwent conversion to complex data, digital filter delay correction, fast-decaying exponential filtering, receiver-coil combination, removal of motion corrupted averages, frequency and phase correction, and spectral registration⁶. PRESS and Hadamard-reconstructed GABA difference spectra were saved to RAW LCModel⁷ compatible format. The unsuppressed water data was recovered from the method file. Spectra were analyzed with LCModel (6.3-1L). For each set of acquisition parameters (TR, TE, RF pulses waveforms, pulse sequence timings, etc.) and in vivo/in vitro scanning, basis sets were simulated for 81³ spatial positions and control parameters were created.
The relationship between true and LCModel-detected GABA concentrations was assessed by Pearson correlation. GraphPad Prism (version 9.2.0) two-way ANOVA analysis with Šídák's multiple comparisons tests were employed to determine significant differences between mouse groups for all methods. For between-method comparison, the intraclass correlation coefficient (ICC) and 95% Confidence Interval (CI) were estimated using SPSS (version 27.0.1.0), and Bland-Altman plots with GraphPad Prism.

Results

Correlation results for phantom MRS concentrations versus true GABA concentrations are illustrated in Figure 2. Figure 3 shows AOAA-treated mouse example spectra. Figure 4.a displays mean and SD plots of GABA concentrations (i.u.) and Cramér–Rao lower bound error estimates (CRLBs,%), while Figure 4.b shows significantly higher GABA concentrations in AOAA-treated mice than in controls for all methods (p<0.001). The ICC (95% CI) was 0.938 (0.029–.982). Bland-Altman plots are displayed in Figure 5.

Discussion

Phantom correlation results show excellent agreement between LCModel estimated and true concentrations for all three methods. Distribution/violin plots display a greater spread in AOAA concentrations, especially for PRESS, while the ANOVA results show all methods successfully discriminate between the two groups. Considering MEGAPRESS (without macromolecule suppression) as the gold standard, these results also indicate PRESS leads to an overestimation of GABA levels. The high ICC implies consistency between methods, but Bland-Altman plots indicate systematic differences between PRESS and spectral editing techniques (PRESS data points above the average difference). One potential explanation for these differences is the incorrect fitting of the GABA basis function by LCModel in the presence of confounding spectral peaks. Despite using sequence-matched simulated basis sets and disabling LCModel automated corrections, there might be LCModel hidden control parameters not fully optimized for field strength, data type, or pulse sequence parameters. Although group differences were detected by all methods, spectral editing comes with the advantage of a metabolite-tailored data acquisition, where macromolecules and lipids contribution is reduced through sub-spectra subtraction. Future work includes the comparison of MRS results with NMR measurements of extracted brain tissue.

Conclusions

PRESS and spectral editing methods (MEGAPRESS and HERMES) were compared at 9.4T for the first time. All methods successfully distinguished between the treatment groups, but PRESS appears less precise and more prone to systematic bias. PRESS GABA estimates may be sufficient when additional metabolites and group differences are more important than absolute quantification, otherwise, spectral editing should be used.

Acknowledgements

No acknowledgement found.

References

1. Bottomley PA, Ann N Y Acad Sci, 1987, 508, 333–348. 2. Mescher M, & Garwood M, NMR Biomed, 1998,11, 266–272. 3. Chan K. et al, Magn Reson Med, 2016, 76, 11–19. 4. Löscher W, Hörstermann D, Naunyn-Schmiedeberg's Arch Pharmacol, 1994, 349, 270–278. 5. Simpson R et al, Magn Reson Med, 2017, 77, 23–33. 6. Near J, Magn Reson med, 2014, 73, 44-50. 7. Provencher S, Magn Reson Med, 1993, 30, 672–679.