The small pilot investigation, Experiment 1, indicated elevated Glx in pACC in ASD. A larger follow-up study, Experiment 2, in which the sample with ASD and the control sample were now better matched for gender and IQ again found elevated Glx in ASD, albeit restricted to right pACC. Thus, two separate studies using independent samples of children with ASD and controls and scanning on two different systems (GE, Siemens) both support the idea of pACC hyperglutamatergia in ASD.
As discussed in our review (Levitt et al., in press) 
, prior MRS studies reporting Glu or Glx in ASD have yielded mixed results, presumably due to differences in subject samples and MRS methods. Using rigorous MRSI methods, Friedman et al. (2003, 2006) 
found no effects of ASD on Glx in the cingulate. Unfortunately, these authors do not specify which subregion(s) of the cingulate they sampled. Also, all the subjects with ASD in their study underwent propofol sedation at time of scan, whereas that was the case for only one subject in Experiment 1 and for none in Experiment 2. Shulman et al. (2003, 2009) point out that brain energetic metabolism, and thence the rate of Glu-Gln cycling and thereby possibly Glx levels, is intimately related to anesthesia-induced level of consciousness 
. Finally, the subjects in Friedman et al. (2003, 2006) were considerably younger (<5 years old) than ours. Anatomic neuroimaging suggests differences between ASD in early and later childhood; for example, the well-known observation of brain overgrowth in ASD between 2–4 years 
that arrests or normalizes in later childhood and adolescence 
. Changes in regional neurometabolite levels in the developmental course of ASD may similarly underlie differences between our Glx findings and those of the Seattle Group. DeVito et al. (2007) 
obtained below-normal Glx and tNAA diffusely in cortical gray matter in subjects with ASD. In particular, endpoint values were extrapolated from linear plots of metabolite levels sampled from voxels across entire cerebral lobes. Hence, it is not clear what levels of this metabolite were locally in pACC. Also, their TE of 135 ms leaves a rather small Glx signal that is perhaps less reliable than the signals of other metabolites. Sampling the amygdala-anterior hippocampus, Page et al. (2006) 
measured elevated Glx in their sample with ASD. Experiments 1 and 2 are not strictly comparable with these prior studies as the latter did not specifically target the pACC using short-TE (25–30 ms) spectroscopy; to that extent, the present findings are novel. Using methods similar to our, Bernardi et al. (2011) 
observed depressed Glx in right pACC, the opposite of the present findings, but in adult subjects with ASD. On the one hand, these opposite findings may reflect differences between adults and children. On the other hand, metabolite levels in our study are CSF-corrected and take account of MRSI voxel tissue composition while Bernardi et al. (2011) 
do not provide this information. As metabolite levels vary between tissue types, it is possible that metabolite effects in Bernardi et al. (2011) 
are confounded with intersubject variation in voxel tissue composition. Thus, while our studies of ASD offer anatomically highly targeted assays of pACC metabolite levels at short-TE accounting for voxel tissue composition, no previous study we are aware of offers comparable methodological rigor in the same age range (middle-childhood to adolescence).
Hence, we think that our two experiments add support to the notion that hyperglutamatergia exists at least in pACC and at least in middle to late childhood and adolescence in ASD. To determine whether this applies also in early childhood, a subsampling of the Seattle data within the pACC portion of their MRSI slab might be helpful; a more stringent and more challenging test would be to acquire from the region in non-anesthetized children, perhaps while sleeping or using very rapid pulse sequences. To determine whether hypo- or hyperglutamatergia applies in pACC in adults with ASD, results of an analysis of the Bernardi et al. (2011) data taking account of voxel tissue composition might be helpful.
The findings of hyperglutamatergia in our studies are generally consistent with recent broader theories of autistic spectrum disorders that emphasize an imbalance in cortical excitation vs. inhibition 
. Elevated glutamatergic metabolism, as seen in the present results, may be a consequence of deficient central GABAergic inhibition, as proposed in these theories. Based on human post-mortem and genetic and rodent ASD model investigations, inadequate GABAergic inhibition may have multiple developmental causes including underexpression of the GAD65 and GAD67 enzymes (that synthesize GABA from Glu), atypical GABAA
receptor subunit composition, and failure of neuroligin and neurexin synaptic binding proteins to appose post-synaptic GABA receptors precisely to presynaptic release sites. In the past few years, translational neuroscience has been making a mounting case for hypofunctioning of brain GABAergic systems ASD that could lead to excess brain glutamatergic metabolism.
Whereas Experiment 1 found diminished pACC Cr in the sample with ASD, Experiment 2 found elevated Cr in right pACC. Thus, abnormal Cr in pACC in ASD, if it does exist, is more variable and harder to reproduce than abnormal Glx. This variability might reflect unstable local cell energetics that induce an expanded creatine-phosphocreatine ATP buffer in some subjects and a contracted buffer in others. Page et al. (2006) 
also observed elevated Cr in amygdala-hippocampus in ASD, and Levitt et al. (2003) 
observed effects of ASD diagnosis on Cr in occipital cortex and caudate, so there is precedence for abnormal Cr in ASD, albeit in other brain regions.
Elevated tNAA was found in pACC in ASD in Experiment 2 only. Again, abnormal tNAA in ASD may be harder to reproduce than elevated Glx. In prior work, Oner et al. (2007) 
registered higher tNAA/Cr and tNAA/Cho in right anterior cingulate cortex in subjects with Asperger's syndrome than in controls and Fujii et al. (2010) 
found lower tNAA/Cr in anterior cingulate in subjects with autism than in controls. Interpretation of these results is partially obscured by normalization to Cr, which itself may vary, but they do suggest heterogeneous effects of ASD on tNAA. In other brain regions (reviewed in Levitt et al, in press) 
, investigators have often found below-normal tNAA or its ratios in ASD, although findings of above-normal and no difference also exist. How plausible is a local elevation of tNAA in the pACC? In addition to the above-cited MRS results, data from recent fMRI and hybrid fMRI-MRS experiments do, in fact, strongly suggest a special role for the pACC in ASD and autistic symptomatology. The pACC, for example, was one of the few brain regions demonstrating significant effects of ASD diagnosis in a recent metaanalysis of fMRI studies 
. Working in healthy subjects, the same researchers related fMRI functional connectivity with the pACC with elevated levels of autistic traits 
. Also in healthy controls, Duncan et al. (2011) 
found correlations localized to pACC between MRS Glx and an fMRI effect related to subject empathy, low empathy being a common symptom of ASD. Finally, elevated intensity was observed in at-risk carriers of an autism-associated CNTNAP2 allele in pACC 
. These and other neuroimaging results give ample evidence for focal effects of ASD diagnosis and autistic traits and autistic symptoms in the pACC. It is therefore not surprising to find MRS metabolic effects particular to that brain region.
Experiment 2 alleviated several but not all limitations of Experiment 1. Both studies were still conducted at low-field (1.5 T) and expressed their results as Glx rather than as Glu and Gln separately. Based on low field strength and, in the case of MRSI, small voxel size, our quality control procedures used the standard 20% SD criterion of the LCModel fitting package and a SNR cut-off of 3 for MRSI and 5 for single-voxel MRS. Although some spectroscopists might prefer stricter cut-offs, working with these values we found that individual metabolite peaks were typically readily identified by eye and easily fit by automated routines. Also single-subject data quality was frequently higher than the cut-off values. In neither study was it possible to match between-group voxel tissue-composition thoroughly (although differences were dealt with through statistical covariates and co-sampling proximal white matter). Efforts to match tissue composition may have been aggravated by putative effects of ASD on anterior cingulate cortical volume or thickness 
. Future MRS and MRSI studies at 3 T will allow smaller, hopefully more tissue-pure voxels and also better spectral segregation of Glu and Gln. Regarding the latter, better segregation might also be achieved by acquiring spectra at TE
80 ms, thought to be optimal for quantifying Glu 
. Future investigations should also include MR relaxation studies, as autism may affect metabolite and water relaxation times 
. Finally, in both Experiments, several subjects with ASD were undergoing treatment with psychotropic medication at time of scan. Ideally, one would test only drug-naïve subjects, although, given clinical realities, this can be difficult to achieve on a practical time scale. These limitations notwithstanding, the present findings suggest that Glx is elevated locally in pregenual anterior cingulate cortex in subjects with ASD. This elevation, combined with less certain effects of autism on Cr and tNAA may reflect disturbances of cell-energy metabolism.