MRS data analysis

Raw MRS data were post-processed using the MATLAB (version R2018a) toolbox MRspa version 1.5f. (https://www.cmrr.umn.edu/downloads/mrspa/). Motion-corrupted data were removed to improve the spectral quality. To quantify the proportion of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) fractions in the volume of interest, segmentation in SPM (version 12) was applied to the T1w MPRAGE images. All neurotransmitter concentrations were corrected for GM and WM concentrations, as previously described⁴². Eddy current and frequency corrections were performed using a water reference scan, and the zero- and first-order phases of the array coil were aligned using the cross-correlation method of MRspa. Subsequently, LCModel (version 6.3-1N; Stephen Provencher, Inc., Oakville, ON, Canada) was used to quantify the concentration of neurochemicals within the chemical shift range of 0.5–4.1 ppm^43,44. Other parameters in the LCModel were as reported previously⁴⁵. The metabolites were estimated with a simulated basis set of 7T STEAM sequence provided by LCModel. The basis set includes the spectra for the following metabolites: alanine, aspartate (Asp), Cr, phosphocreatine (PCr), GABA, glucose, glutamine (Gln), Glu, glycerophosphocholine, phosphocholine, choline (Cho), myo-inositol (Ins), N-acetyl aspartate (NAA), N-acetyl aspartylglutamate (NAAG), scyllo-inositol, taurine (Tau), and macromolecules. The concentrations of GABA and Glu were normalized to that of tCr, which is known to be a high and stable concentration in the human brain²⁷. We found significant positive correlation between learning related change in GABA and Glu (Fig. 2B). As GABA is generated from Glu²⁷, we calculated the change in the GABA/Glu ratio after the motor sequence learning task using the following equation:

where GABA/Glu_pre and GABA/Glu_post indicate the GABA/Glu ratio at pre- and post-task, respectively. Two-way repeated-measures ANOVA was conducted using SPSS, with the concentrations of GABA and Glu at different time points (pre-, during-, and post-task) as a factor. The CRLB, water linewidth at FWHM, and signal-to-noise ratio (SNR) were used for the quality control of spectra⁴⁴. The CRLB and SNR were calculated using LCModel, and water linewidth was obtained by fitting to the additional water spectrum using MATLAB. Data were excluded when CRLB > 15% (n = 2), linewidth > 19 Hz (n = 1), or SNR < 30. Two-way repeated-measures ANOVA was performed on the CRLB, water linewidth, and SNR with the experimental group (learning vs. movement) as a between-subjects factor and time points (pre-, during-, and post-task) as a within-subjects factor, using SPSS.