The authors studied 54 carefully screened healthy Caucasian American comparison subjects of European ancestry. Demographic information is presented in a table available online (Data Supplement 1 at http://ajp.psychiatryonline.org). All of the participants underwent a structured diagnostic interview and a neurological evaluation, and subjects with any current or past psychiatric or neurological diagnosis, current medical illness, or family history of psychosis were excluded. Written informed consent was obtained from all subjects. The subjects were genotyped at rs6465084 in intron 2 of the GRM3 gene, as described by Egan et al.
(4) . The subjects completed a battery of tests, including verbal fluency (letters and categories), and an IQ test (WAIS–Revised; see reference 4 for more details). Proton magnetic resonance spectroscopic imaging (MRSI) was performed with a 3 T GE magnet (General Electric Medical Systems, Milwaukee), with a multislice imaging technique similar to that in previous publications by our group (four slices [7.5 mm cubic voxel dimensions]; spin echo slice selection; TR=2300 msec, TE=280 msec, no water suppression, and lipid signal from the scalp suppressed with outer volume saturation pulses)
(5) . The MRSI acquisition was conducted on a plane parallel to the main axis of the hippocampi. Regions of interest were drawn on the left and right dorsolateral prefrontal cortex, the cingulate cortex, the white matter of the centrum semiovale, the left and right hippocampal formation, and the occipital cortex on structural magnetic resonance imaging (MRI) scans registered to the MRSI slices. Metabolite signals were calculated as the integral of the magnitude spectrum in a range of 0.25 ppm surrounding peaks for
N -acetylaspartate, creatine, and choline and were reported as metabolite ratios averaged over the voxels in the regions of interest. Extensive quality control procedures were undertaken to eliminate voxels with obvious artifacts. A low number of voxels in the regions of interest (less than three voxels in slice one and less than seven voxels in slice four) resulted in rejection of that particular region of interest for further statistical analysis. All quality control procedures were performed blind to genotype status. We collapsed the A/G (N=21) and G/G (N=4) genotype subjects into a G carrier group in the analysis. The differences in metabolite ratios and letter and category fluency task scores between the two genotype groups were analyzed by unpaired t tests. One-tailed statistics with a significance level set at p<0.05 were used for the dorsolateral prefrontal cortex because of the directional hypothesis based on prior findings by our group
(4) (no correction for multiple comparisons). Potential confounding factors, such as age, sex, handedness, education, and IQ, were also examined.