To report the structural connectivity results obtained from graph network analysis in schizophrenia patients.

Schizophrenia is a psychiatric illness characterized by a “dysmetric cognitive state”. Several disconnectivity theories have been suggested to explain the pathophysiological mechanism underlying schizophrenia¹. Recently graph theory was introduced to study the brain as a complex network to help map the diverse topological properties of the brain network². These analyses can provide insight about the altered brain connectivity and function seen in schizophrenia.

We analyzed 48 participants with schizophrenia and 46 control participants using a graph theory approach to study the structural connectivity assessed using diffusion weighted imaging. MR images were collected on 3T Siemens Tim Trio equipped with 12-channel head coil. T1-weighted anatomical images were collected using a coronal 3D MP-RAGE sequence: TE=2.8 ms, TR = 2530 ms, TI=909 ms, NEX=1, Flip angle=10°, Field of View=25.6x25.6x25.6 cm2 Matrix=256x256×256. Diffusion-weighted images were gathered with the following parameters from 70 slices: TE = 82ms, TR = 8700ms, Flip angle = 90°, b-value = 1000, gradient directions = 30, Field of View = 25.6cm2, Matrix = 128×128, bandwidth = 1395Hz, slice thickness/gap=2.0/0.0mm. The T1-weighted images were AC-PC aligned and intensity normalized before performing tissue classification and anatomical labeling using fully automated BRAINS software³. Following this, the anatomical data was parcellated into 68 cortical regions using Freesurfer⁴, which were later used for defining the nodes in the diffusion data for the graph-based network analysis. Diffusion data were first processed using DTIprep software⁵ to perform adequate quality control including eddy-current correction, head motion correction, bed vibration and pulsation, venetian blind artifacts, as well as slice-wise and gradient-wise intensity inconsistencies. DTI scalar maps including FA and MD were computed and whole brain fiber tracking was performed using deterministic fiber tracking, based on the FACT (fiber assignment by continuous tracking) algorithm as implemented in DSIStudio software package⁶. Streamlines were started from 8 randomly placed seeds from each of the white matter voxels and tracking proceeded by connecting the primary diffusion direction from one voxel to the next voxel. This procedure reconstructed all possible fiber tracts within the brain. Fiber tracking along a streamline was terminated when a voxel was reached with a FA < 0.1, when the streamline exited the brain or when the fiber tract made a sharp turn > 45°. Following whole brain tracking, the fiber tracts interconnecting pairs of regions in the 68 cortical regions were identified for all possible pairs of cortical regions and a 68 x 68 weighted connectivity matrix was created with streamline counts as the weights. Using the connectivity matrix, basic topological metrics of the network including strength, modularity, clustering co-efficient and global efficiency were computed. The above measures were normalized using 1000 random networks that were generated using similar connectivity distribution using the Brain Connectivity Toolbox⁷. The metrics were generated for each subject and then used to test group differences between the patient and control groups using permutation testing.

As reported previously in literature^8, the most highly connected regions in the schizophrenia as well as control population were bilateral precuneus, superior frontal cortex, superior parietal cortex and superior temporal cortex and insula, which formed the so-called rich-club hubs in the network. However, contrary to previously reported findings, we did not find reduced rich-club connectivity in the participants with schizophrenia as compared to controls in this dataset. Instead, we find that the feeder connections as well as the local connections are reduced in strength in our data (see figure 1). The results of permutation testing on the topological measures of the graph are shown in figure 2. We observed that the participants with schizophrenia had a reduced strength of connectivity, clustering coefficient and modularity compared to the control participants. However, only the reduction in clustering co-efficient was statistically significant (p=0.047 corrected after correcting for multiple comparisons).Disruption in small-world networks in structural connectivity data has been suggested in many previous reports in schizophrenia. However, in our analysis, the only indication of disruption of this network is the significant reduction in clustering co-efficient in the structural connectivity data, while the characteristic path length remains almost the same. This suggests the need for more test-retest reliability studies on these network measures.