Systematic Regional Variations of GABA, Glutamine, and Glutamate Concentrations Follow Receptor Fingerprints of Human Cingulate Cortex

Discussion

In this study, we showed that single voxel ¹H-MRS using STEAM VERSE with short TE/TM at 7T led to solid results based on data with high retest reliability, when standard spectral quality criteria were applied. We for the first time obtained reliable spectra across different anatomical CC partitions. This enabled us to find robust within-subject regional variations in absolute and relative concentrations of GABA, Gln, and Glu, even under consideration of potential effects of GM content variations across the CC.

So far, metabolic imaging studies using MRS were highly limited in their interpretation of observables in regard to specific cellular compartments. Although contributions of GM and WM could, in principle, be discerned by considering the individual single-voxel compositions, we could not infer whether transmitter concentrations reflect intracellular or synaptic pools, or are all attributable to neurons. This is of specific concern, given that both GABA and Glu are converted from Gln, synthesized almost exclusively in neighboring astrocytes. In the case of tripartite synapses, synaptic Glu is taken up into the astrocyte and then converted into the non-neurotoxic intermediate Gln. Although two recent pharmacological pilot studies in bipolar depression (Brennan et al., 2010) and major depressive disorder (Salvadore et al., 2012) were important to substantiate the role of differential assessment of Glu and Gln, they were very long, with acquisition time over 1 h, and particularly in the second study, regionally quite imprecise. Overall, no evidence currently exists, in which extent the expected glial toward neuronal contribution could be set into the context of glutamatergic receptor densities.

Here, we showed that metabolite concentrations and ratios indeed covary with known receptor distributions, when anatomically defined subregions are investigated. For the distribution of AMPA receptors, Palomero-Gallagher et al. (2009) reported a gradual decline from above-average densities in pgACC toward average levels in aMCC, and following in caudal direction, even subaverage densities in the regions around the central sulcus. Although ketamine is an NMDA receptor antagonist, its relevant antidepressant action is thought to be mainly mediated via activation/translocation of AMPA receptors, similar to other glutamate-modulating agents, such as riluzole or lamotrigine (Du et al., 2007). The suggestion speaks in favor for the observed treatment effects in pgACC, mirrored by increased Gln cycling and predicted by local estimates of Gln (Salvadore et al., 2012). Although this is still a very indirect connection, it has to be considered as the first link between local metabolite concentration and a receptor subtype density related to pharmacologically increased metabolite levels and Gln turnover. Our method here is evidenced by its robust estimation for comparably small and thus anatomically specific ROIs at very short acquisition times, allowing for assessment of several regions in one scanning session. Another major advantage is the parallel acquisition of all three metabolites with 95% success rates, thus enabling estimation of Gln/Glu ratios and inhibitory GABAergic mechanisms simultaneously. As described above for glutamatergic neurotransmission, a link between GABA concentrations and their relevance for actual synaptic GABAergic activity was currently not established in humans. Increases in pgACC GABA as found here, reflect the higher GABA_B receptor densities toward the other three cingulate compartments (Palomero-Gallagher et al., 2009). Conversely, GABA concentrations are inversely correlated with GABA_A receptor densities, which present considerably lower concentrations in pgACC than in the remaining cingulate regions (Palomero-Gallagher et al., 2009). Thus, GABA release in pgACC would mainly result in slow synaptic inhibition via GABA_B receptors, whereas mid-CC and PCC would be subject to the fast component of GABAergic inhibition, which is mediated to a greater extent by GABA_A receptors. Interestingly, activation of presynaptic GABA_B receptors suppresses multivesicular release and thus decreases synaptic Glu concentration (Chalifoux and Carter, 2010), which would explain why we found the highest GABA/Glu ratios in pgACC. This characteristic ratio, which next to its receptor fingerprints, would indicate that pgACC is subject to a higher degree of inhibition than mid-CC or PCC are, according to its functional characterization as a default mode network region with high baseline activity. Indeed, Northoff et al. (2007) could show that across subjects, the amount of negative blood oxygen-level-dependent responses correlates with regional GABA concentrations in pgACC. Although a direct translation of between-subject transmitter variances to variations across the brain needs be treated as carefully as the above mentioned line of evidence for AMPA receptor-mediated increases in Gln cycling, it needs however be noted, that we here found a similar effect of covariance of markers of an excitation/inhibition balance and the known regional baseline mechanisms of task-induced activations and deactivations.

Although direct pharmacological follow up studies are needed to elaborate the above depicted correlations, a second main achievement of this study is the definition of robustness of the multivoxel, multimetabolite protocol. As 7T scanners recently became available at many institutes world-wide, our findings are encouraging that similar protocols can be applicable in such studies.

2D J-resolved spectroscopy at 3T was also applied in the pgACC region to measure GABA, Gln, and Glu (Walter et al., 2009), although long acquisition times were not suitable for our multivoxel acquisitions. In comparison, our metabolite concentrations relative to creatine (Cr) in pgACC are similar to this literature's result: Gln/Cr (0.35 vs 0.34), Glu/Cr (1.47 vs 1.35), and GABA/Cr (0.29 vs 0.21). In addition to, we found highly reliable metabolite estimates in our study. However, as a cautionary remark, this applies only to spectra fulfilling the quality criteria. In principle, the removal of outliers would be problematic. However, for the retest dataset, only 2 of 120 spectra were excluded.

Because we found a considerable variation of GM contents across our ROIs, the correction for voxel composition should be considered crucial. However, the only difference in our results was that after correction, trend-wise effects reached significance. Because there is no gold-standard method allowing the correction of metabolite concentrations for tissue compositions, to avoid false assumptions on the actual variance contribution, we decided for a linear regression method to correct metabolite concentrations based on the estimates of the influence of GM variations. Although we expected a potential increase in measurement reliability for the MRS values after GM correction, the a priori high values however were not changed. If another potentially better correction method would be applied in the future, similar inclusion of reliability assessment might be considered to prove its superiority accordingly.

Correction of metabolite concentrations becomes more difficult when modeling the effects of GM, WM, and CSF together. However, metabolites are found at much lower concentrations in CSF (Lynch et al., 1993; Stephenson et al., 2011). Additionally, the mean contribution of CSF segments to the voxel content was no more than 5%. Therefore, we can exclude significant CSF contributions to metabolite signal in this study.

A critical limitation of our study is the distinction of anatomical boundaries in the PCC. Several relatively small subregions with opposing receptor fingerprints lie in direct vicinity, making it difficult to measure them individually with our voxel size. Therefore, a careful comparison of our metabolite profiles with previously reported receptor variations would lead to expect more variation between the posterior voxels. Next to integration of different regions based on mere vicinity, one may further accuse the interindividual variation in gyrification, which of course could not be sufficiently accounted for by visual voxel prescription in the posterior part, although this is less problematic for the more distant anterior subunits (Vogt et al., 1995). A systematic variation of individual voxel placement could however not be found to support such a confounder. This variability should however not be confused with functional heterogeneity of the PCC region. Increased GABA/Glu ratios together with highest Gln cycling in pgACC would also be expected in the PCC, if the actual location of the ROI was more ventral and caudal, toward the retrosplenial–precuneus portion. In fact, our caudal PCC voxel was located quite dorsally, not overlapping with the posterior default mode node in the PCC (Yu et al., 2011). Although we rejected an influence of daytime on the systematic variations of regional metabolite measures, one should cautiously acknowledge that at larger sample sizes, such subtle mechanisms, which e.g., have been reported on other physiological assessments as well as for severe sleeping disorders (Saude et al., 2007; O'Donoghue et al., 2012) may exist and would have better been avoided using a more rigid study design with fixed acquisition times.

In conclusion, we demonstrated the regional covariation of local concentrations of GABA, Glu, and Gln with previously reported receptor variations using a robust and efficient 7T MRS protocol. Gradual decline of Gln concentration paralleled the distribution of AMPA receptor densities, providing evidence from cross-regional variances which converges with findings of longitudinal changes after AMPA modulating drugs. Strongly increased GABA concentrations in pgACC, as a representative region of the default mode network were paralleled by lowest excitation/inhibition balance. Thus, high GABA concentrations should be representative of the regions potential to deactivate under certain task conditions. More work will be needed to directly link the here observed cross-regional variance to within- and between-subject metabolite variations as seen in health and disease.