Parcellations of the pgACC improve prediction of local glutamate from whole-brain connectivity

Poster No:


Submission Type:

Abstract Submission 


Louise Martens1, Nils Kroemer2, Vanessa Teckentrup2, Lejla Colic3,4, Nicola Palomero-Gallagher5,6, Meng Li2,1,4, Martin Walter2,4,3,1


1Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 2University of Tübingen, Tübingen, Germany, 3Leibniz Institute for Neurobiology, Magdeburg, Germany, 4Clinical Affective Neuroimaging Laboratory, Magdeburg, Germany, 5Institute of Neuroscience and Medicine (INM-1), Research Centre Jülich, Jülich, Germany, 6Department of Psychiatry, Psychotherapy and Psychosomatics, Medical Faculty, RWTH Aachen, Aachen, Germany


Local measures of neurotransmitters such as glutamate (Glu) and GABA provide insights into possible neurobiological changes underlying altered functional connectivity (FC) in mental disorders (e.g., Duncan et al., 2013). However, as the signal-to-noise ratio of conventional magnetic resonance spectroscopy (MRS) is low, a single MRS voxel may cover regions with distinct cyto- and receptorarchitecture , and, therefore, distinct FC profiles. Here, we propose a novel, multi-modal approach offering a more nuanced prediction of Glu and GABA in an MRS voxel. To this end, we employed voxel-wise connectivity-based parcellation (CBP) of a pregenual anterior cingulate (pgACC) MRS voxel and a cytoarchitectonic parcellation (Palomero-Gallagher et al., 2018). We then used two complementary data-driven methods to predict Glu and GABA from cluster-wise connectivity.


88 healthy participants underwent a 7 Tesla MRI protocol. The MRS voxel (20x15x10 mm3) was placed in the pgACC using anatomical landmarks. MRS data were fitted using LCModel, expressed as ratio to total Creatine (tCr), and residualized for voxel gray matter. Resting-state data were preprocessed using the default CONN pipeline (Whitfield-Gabrieli & Nieto-Castanon, 2012), without spatial smoothing. Z-scored timeseries were further denoised by despiking, quadratic detrending and regressing out of 6 motion parameters and mean white matter signal using MATLAB.

For the CBP analysis, we created a composite ROI based on participants' MRS masks (threshold: covered in >1N). FC to the 132 CONN atlas nodes was calculated for each seed voxel within the CBP and the cytoarchitectonic ROIs (i.e. areas p32 and p24). We parcellated the MRS ROI based on FC profiles using hierarchical clustering and cluster-wise FC differences were compared using a paired t-test. To related CBP and cytoarchitectonic ROIs, we computed Dice overlap (DC).

To predict Glu/tCr and GABA/tCr from FC, we used partial least squares regression (PLSR) and elastic net (EN). While PLSR identifies common factors in predictors and outcomes to optimize prediction, EN drops redundant regression coefficients, resulting in sparser models. To statistically test model fit (residual sum of squares), we performed permutation tests (1000 permutations).


Hierarchical clustering of voxels into two clusters reduced within-cluster and increased between-cluster distance approximately twofold (Fig. 1A-D). Cluster 1 overlapped with cytoarchitectonically-defined area p32, but not with area p24. Cluster 2 overlapped with cytoarchitectonically-defined p24, but not with p32. The hierarchical clusters corresponding to p32 and p24 had markedly different FC profiles, indicating differential links to the default mode network and the salience network, respectively (Fig. 1E-G).

Glutamate was predicted better than chance from cluster 1 using EN (p < .001, Fig. 2B). Results were comparable using PSLR, yet not significant (Fig. 2A). In contrast, cluster 2 FC explained less variance in Glu compared to cluster 1, both using PLSR and EN (Fig. 2A-B). Notably, FC from both clusters together could not successfully explain Glu using either method. Predictions using cytoarchitectonic ROIs showed a similar pattern (Fig. 2C-D). GABA/tCr could not be predicted using EN models (all ps >.99). Using PLSR, p24 explained more variance than p32 or both together, but only cluster 2 predicted GABA/tCr better than chance (p < .05).
Supporting Image: 1_Fig1_Martens_OHBM2019_3.png
Supporting Image: 2_Fig2_Martens_OHBM2019.png


Connectivity-based parcellation of a pgACC MRS voxel recovered known histological subregions of the pgACC, with distinct functional connectivity patterns that differentially predict Glu, and outperforms prediction using the unparcellated voxel. Collectively, our results show that multimodal imaging may help to overcome the fundamental limitations of a single method as fMRI can improve the spatial specificity of local neurometabolites assessed with conventional MRS.

Imaging Methods:

MR Spectroscopy 2
Multi-Modal Imaging 1

Modeling and Analysis Methods:

Classification and Predictive Modeling
Multivariate modeling


Transmitter Systems


Magnetic Resonance Spectroscopy (MRS)
Other - Resting-state fMRI; Partial least squares regression; Elastic net; Parcellation; Cytoarchitecture

1|2Indicates the priority used for review

My abstract is being submitted as a Software Demonstration.


Please indicate below if your study was a "resting state" or "task-activation” study.

Resting state

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Healthy subjects

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.


Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Not applicable

Please indicate which methods were used in your research:

Functional MRI

For human MRI, what field strength scanner do you use?


Which processing packages did you use for your study?

Other, Please list  -   Matlab custom code, CONN toolbox

Provide references using author date format

Duncan, N. W. (2013). Glutamate Concentration in the Medial Prefrontal Cortex Predicts Resting-State Cortical-Subcortical Functional Connectivity in Humans. PLoS ONE, 8(4).

Palomero-Gallagher, N. (2018). Human Pregenual Anterior Cingulate Cortex: Structural, Functional, and Connectional Heterogeneity. Cerebral Cortex.

Whitfield-Gabrieli, S. (2012). Conn: A Functional Connectivity Toolbox for Correlated and Anticorrelated Brain Networks. Brain Connectivity, 2(3), 125–141.