The amygdala plays a key role in emotional learning and emotion processing. Since disturbed amygdala function has been linked to psychiatric conditions such as anxiety, autism and major depression, insight into its biochemistry, particularly the neurotransmitter levels, is highly desirable to improve the understanding of the pathophysiology of these disorders¹. The small size of the amygdala and its difficult position regarding field homogenization often lead to low signal-to-noise ratio (SNR) and poor spectral resolution, hampering MRS-based neurochemical profiling in vivo. Thus, previous MRS studies of the amygdala did not achieve separation of glutamate (Glu) and glutamine (Gln) nor an estimation of glutathione (GSH), all metabolites of utter relevance in the above context. We combined the spin echo full intensity acquired localized (SPECIAL)² technique with an optimized shimming protocol, carefully selected postprocessing steps and data analysis including a basis set with a measured macromolecule (MM) spectrum for quantification of metabolites in the human amygdala, including glutamatergic metabolites and GSH.Twenty healthy volunteers (34.1 ± 8.7 years, 14 female) participated in the study. MR measurements were performed on a 3T Verio scanner (Siemens Healthcare, Erlangen, Germany) using a quadrature transmit-receive head coil (Rapid Biomedical, Rimpar, Germany). Subsequent to MPRAGE imaging a voxel of 15 x 15 x 15 mm³ (figure) was carefully placed on the left amygdala using the hippocampal head as a landmark and avoiding adjacent cerebrospinal fluid (CSF) and white matter. All first and second order shims were adjusted using optimized FAST(EST)MAP³; when the water linewidth (full width at half maximum) after shimming was larger than 11 Hz, shimming was repeated. MR spectra were then acquired using SPECIAL with TR = 3 s, TE = 6 ms, 256 averages and acquisition time 1024 ms. Both metabolite and unsuppressed water raw data were processed using in-house software written in MATLAB, applying the SPECIAL add/subtract scheme, frequency drift correction and averaging. Resulting spectra were analyzed using LCModel with a simulated basis set containing 18 metabolites and, to improve the accuracy of the analysis⁴, a measured MM spectrum (averaged from five healthy volunteers). Metabolite amplitudes were referenced to an unsuppressed water scan. Concentrations were corrected for relaxation using measured and literature values for T1 and T2, and for the amount of CSF in the voxels deduced from the segmented T1-weighted images using SPM8⁵.In good agreement with previous studies^6,7, the mean fractional volumes (± SD) of the amygdala voxel were 0.861 ± 0.027 gray matter, 0.071 ± 0.024 white matter and 0.068 ± 0.024 CSF. The mean linewidth (± SD) of the water signal after shimming was (8.1 ± 1.5) Hz; no water spectrum had a linewidth above 10.7 Hz. Frequency drift correction improved the SNR by up to 23 % (mean 9.6 %). The quality of the resulting metabolite spectra is exemplified in the figure, which includes the overall fit and the fit results for the metabolites of interest. The table shows the metabolite concentrations and corresponding Cramér-Rao lower bounds (CRLB). None of the individual CRLB for total N-acetylaspartate, creatine and choline (tNAA, tCr, tCho), myo-inositol (Ins) and, most importantly, Glu exceeded 11 %. Some individual Gln and GSH values had CRLB above 30 % and were excluded. Repeating, for comparison, the analysis without measured MM in the basis set, i.e. with the MM simulated by LCModel, resulted in higher metabolite concentrations, perhaps due to systematic overestimation, thus demonstrating the importance of this modification.A combination of a carefully executed shimming approach with very short echo time spectroscopy at 3 T on the acquisition side with data processing using frequency drift correction and LCModel fitting with measured MM spectra as prior knowledge proved a successful approach to assess integral parts of the neurochemical profile in a brain region as difficult to investigate as the amygdala. Beyond quantification of the three main metabolites plus myo-inositol with excellent precision, for the first time glutamate was determined reliably and separately from glutamine. An attempt to use a basis set without macromolecules introduced a systematic bias into the analysis (possibly overestimated concentrations). Due to low SNR, quantification of glutamine and glutathione was not possible in all spectral datasets. As might be expected, the choice of the MM approach strongly influenced the analytical precision for these low-concentrated metabolites. Further improvement should be achievable by using higher order shimming to enhance spectral dispersion. The devised method may help to elucidate metabolic alterations in the amygdala in psychiatric disorders.