In each subject, single-voxel proton MRS data were acquired at rest from a volume of interest of 20 mm × 25 mm × 35 mm = 17.5 mL, which was placed in the right pgACC (Figure 1). Magnetic resonance spectroscopic acquisitions were performed using a Philips Intera 3T whole-body magnetic resonance unit (Philips Healthcare, Best, the Netherlands) equipped with a transmit-receive volume coil. To enable unambiguous, separate, and simultaneous quantification of GABA, Gln, Glu, Glc, and NAA, data were acquired using a maximum echo–sampled 2-dimensional J-resolved point-resolved spectroscopy (JPRESS) sequence,26 which encodes the J coupling along the indirect spectral dimension by acquiring data with multiple echo times (eFigure 1). This approach allows for a significant reduction of spectral overlap by spreading multiplets along 2 frequency axes. The echo times for the JPRESS experiment ranged from 31 to 229 milliseconds, with a step size of 2 milliseconds, a phase cycling of 4 for each echo time, a bandwidth in the direct dimension of 2 kHz, and 2048 sample points. Using 100 encoding steps and 4 averages per encoding step at a repetition time of 2500 milliseconds, the acquisition time for 1 voxel added up to 16 minutes. The sequence was preceded by water suppression using frequency-selective excitation and gradient spoiling followed by adiabatic frequency-selective rephasing and gradient spoiling. The 2-dimensional JPRESS data were quantified using ProFit,25 a 2-dimensional fitting procedure, which applies the full amount of prior knowledge by fitting a linear combination of simulated 2-dimensional basis metabolite spectra for the following 19 brain metabolites: alanine, ascorbic acid, aspartate, creatine, GABA, Glc, Gln, Glu, glycine, glycerolphosphorylcholine, glutathione, lactate, myo-inositol, NAA, N-acetylaspartylglutamate, phosphorylcholine, phosphorylethalonamine, scyllo-inositol, and taurine. Simulation of the basis metabolite spectra was performed with GAMMA.34 Cramer-Rao lower bounds,35 an estimate of the fitting error, were used as a quality criterion to exclude data sets with unreliable quantification results. Hence, analyses of group effects and correlational interdependence were restricted to subjects who met strict quality criteria to indicate reliable spectral quantification (Cramer-Rao lower bounds < 20%)25 for each metabolite. This resulted in reduced and different sample sizes for distinct substances, since some metabolites give rise to more prominent resonance lines than others owing to differences in absolute concentrations and coupling behavior. In addition, covariance coefficients were determined in order to exclude systematic correlations due to spectral overlap for reported physiological correlations between metabolites (eFigure 1). Because determination of absolute metabolite concentrations in millimolars requires a reliable T1 and T2 relaxation correction, while relaxation times of coupled metabolites are hardly known for spectroscopy at 3 T, all metabolite concentrations are given relative to creatine levels. Creatine was proven to be an appropriate internal reference for the ProFit analysis in MDD as previously suggested21; absolute creatine concentrations in patients with MDD and healthy volunteers (mean [SD], MDD, n = 19: 6.32 [1.20] mM; healthy volunteers, n = 24: 6.30 [0.96] mM; P < .95) were determined by using the internal water reference method. Additionally, recorded 1-dimensional point-resolved spectroscopy spectra (echo time = 30 milliseconds, 128 averages, 5-minute acquisition time) (eFigure 1) from the same volume of interest were analyzed by LCModel (S.W. Provencher, PhD).