FIGURE S1 Parameterization of the downfield spectrum acquired with “-selref” (A) and (B) sequences. Parameterization was performed using the AMARES (advanced method for accurate, robust and efficient spectral fitting) algorithm in jMRUI 6.0 using nine (A) and seven (B) Gaussian peaks, respectively. Black, spectrum; red, modeled estimate; pink, residual; top, individual parameterized downfield peaks. The different spectral pattern observed in (A) and (B) is caused by different magnetization profiles of the RF pulses

FIGURE S2 Spectral grid of downfield and upfield -selref acquisition for subject 2 is shown with the corresponding metabolic maps. The quantification artifact is visible in some of the maps (white arrow). Only spectra that passed quality control are displayed in the spectral grid. The downfield and upfield data set were acquired in two separate 2D-MRSI scans within one scanning session

FIGURE S3 Spectral grid of downfield and upfield -selref acquisition for subject 3 is shown with the corresponding metabolic maps. The quantification artifact is visible in some of the maps (white arrow). Only spectra that passed quality control are displayed in the spectral grid. The downfield and upfield data set were acquired in two separate 2D-MRSI scans within one scanning session

FIGURE S4 Spectral grid of downfield and upfield -selref acquisition for subject 4 is shown with the corresponding metabolic maps. Only spectra that passed quality control are displayed in the spectral grid. The downfield and upfield data set were acquired in two separate 2D-MRSI scans within one scanning session

FIGURE S5 Spectral grid of downfield and upfield -selref acquisition for subject 5 is shown with the corresponding metabolic maps. Only spectra that passed quality control are displayed in the spectral grid. The downfield and upfield data set were acquired in two separate 2D-MRSI scans within one scanning session

FIGURE S6 Representative downfield -selref, upfield -selref, and spectra shown at full spectral width without any processing steps applied. The spectra were normalized with respect to the water residual

FIGURE S7 Mean correlation matrix for downfield -selref acquisition. The correlation coefficient was calculated as a mean of correlation coefficients across all subjects and voxels (N = 5) as reported by LCModel. A weak correlation (r ~0.15–0.3) was observed for peaks around 7 ppm (i.e., 6.83 ppm, 7.09 ppm, 7.30 ppm, 7.48 ppm, and 7.90 ppm), and a moderate correlation (r ~0.5) was observed for 8.18 ppm, 8.24 ppm, and 8.37 ppm peaks. These peaks were also reported as DF8.X in the main manuscript

FIGURE S8 Metabolite maps of peaks with a weak correlation as reported from LCModel. DF 7.X1 (combined signal of DF6.83, DF7.09, DF7.30, DF7.48. DF 7.X2) combined the signal of DF6.83, DF7.09, DF7.30, DF7.48, and DF7.90