Item

Abstract

Introduction:
Magnetic resonance spectroscopy (MRS) measurements of excitatory and inhibitory neurotransmission may provide valuable insights into the underlying neurobiology of altered functional connectivity in psychiatric disorders. Consequently, local measures of glutamate (Glu) and GABA are often reported to moderate resting-state functional connectivity measures (Duncan et al., 2013; Horn et al., 2010; Kapogiannis et al., 2013). However, the inherently low signal-to-noise ratio of conventional MRS measurements necessitates voxel sizes far exceeding those of fMRI measurements, leading to crude measures of local neurometabolism. Here, we tested the hypothesis that decomposing the pregenual anterior cingulate's (pgACC) functional connectome into a more fine-grained fMRI-based resolution provides incremental information on the neurotransmitters governing its function. To this end, we employed a novel, data-driven approach that aims to predict (pgACC) glutamatergic and GABAergic signatures by assigning weights to seed-voxel connections according to their predictive power.
Methods:
77 healthy participants underwent an MRI protocol consisting of structural, functional, and MRS measurements at 7 Tesla. The MRS voxel (20 x 15 x 10 mm3) was placed in the pgACC according to a previously established protocol (Dou et al., 2013). GABA and Glu levels were fitted using the LCModel software (Provencher, 2001) and expressed relative to total creatine. GABA measures were log transformed. We used the default CONN pipeline (Whitfield-Gabrieli et al., 2012) for fMRI preprocessing without smoothing. Data were then z-scored, despiked, and detrended, after which six motion parameters were regressed out. We created a composite mask of each subject's MRS mask for subsequent analyses. As seed voxels, we selected only those fMRI voxels in the composite MRS mask. The time series of the seed voxels were correlated with mean time series of the 132 CONN atlas nodes. To account for the statistical redundancy of the voxel-based predictor matrix, we employed PLS regression, which projects the predictor variables into a latent space (similar to principal component analysis) while optimizing the prediction of the outcome. We entered the resulting connectivity matrices into two PLS regression models (McIntosh et al., 2004) using 1 component to predict pgACC Glu or GABA. To statistically assess the obtained model fit (residual sum of squares), we performed a permutation test with 1000 permutations of the outcome measure.
Conclusions:
A novel, data-driven approach employing PLS regression allowed us to predict Glu, but not GABA from the functional connectivity profiles of pgACC seed voxels. For Glu, the resulting weight matrix showed that a widespread pattern of functional connectivity contributes to the successful prediction of local glutamate suggesting that glutamate levels might have a diffuse global effect on functional connectivity. Further analyses will be performed to cross-validate the prediction models and explore whether specific seed voxels in the pgACC have higher predictive power than others.