Topographical reorganization of brain functional connectivity during an early period of epileptogenesis

Abstract

Objective: The current study aims to investigate functional brain network representations during the early period of epileptogenesis.

Methods: Eighteen rats with the intrahippocampal kainate model of mesial temporal lobe epilepsy were used for this experiment. Functional magnetic resonance imaging (fMRI) measurements were made 1 week after status epilepticus, followed by 2-4-month electrophysiological and video monitoring. Animals were identified as having (1) developed epilepsy (E+, n = 9) or (2) not developed epilepsy (E-, n = 6). Nine additional animals served as controls. Graph theory analysis was performed on the fMRI data to quantify the functional brain networks in all animals prior to the development of epilepsy. Spectrum clustering with the network features was performed to estimate their predictability in epileptogenesis.

Results: Our data indicated that E+ animals showed an overall increase in functional connectivity strength compared to E- and control animals. Global network features and small-worldness of E- rats were similar to controls, whereas E+ rats demonstrated increased small-worldness, including increased reorganization degree, clustering coefficient, and global efficiency, with reduced shortest pathlength. A notable classification of the combined brain network parameters was found in E+ and E- animals. For the local network parameters, the E- rats showed increased hubs in sensorimotor cortex, and decreased hubness in hippocampus. The E+ rats showed a complete loss of hippocampal hubs, and the appearance of new hubs in the prefrontal cortex. We also observed that lesion severity was not related to epileptogenesis.

Significance: Our data provide a view of the reorganization of topographical functional brain networks in the early period of epileptogenesis and how it can significantly predict the development of epilepsy. The differences from E- animals offer a potential means for applying noninvasive neuroimaging tools for the early prediction of epilepsy.

Keywords: brain networks; epileptogenesis; fMRI; graph theory; self-cured.

Fig. 1.

The experimental setup. (a) Representative structural MRI images (coronal, T2-weighted contrast) of control (upper left), experimental animals (bottom left) and the brain lesion extractions (bottom right). (b) The time-line of the experimental protocol and the illustration of electrophysiological data for animal group classification. (c) A 152-regions rat brain atlas was applied and (d) the 3D representation of the 40 ROIs selected in this study. The color of spheres represents manually defined brain areas: Frontal (red), Temporal (blue) and Parietal (green).

Fig. 2.

(a-c) 3D topographical representations of the functional brain networks in control, E+, and E− groups. Colors of spheres represent clusters of brain areas: Frontal (red), Temporal (blue) and Parietal (green). Lines between spheres represent the significant connection between two specific brain areas. (d-f) The group averaged adjacency matrices for the three animal groups. The adjacency matrices were mapped by the z-transferred Pearson’s r value, where the red color indicates high connectivity strength, and blue color indicates the lowest connectivity. (g-i) Representative the standard deviation maps computed from the group data. Visual analysis revealed a noticeable increase in functional connectivity in the E+ group, while E− group preserved a similar connectivity pattern compared with the control group. All three groups showed a low standard deviation while the variations in the experimental groups (E+ and E−) are more obvious. (j & k) Histograms of connectivity strength across the whole brain in control, E+, and E− animals. There was a shift towards high connectivity in both experimental groups compared to control, and this was more prominent in the E + group. (l) Quantifications of the average connectivity strength across all brain nodes considered in control, E+, and E− group. A significant difference was found between control and E+.

Fig. 3.

Global characteristics and small-worldness features of resting-state brain networks derived from graph theory analysis. Data are presented with mean and standard deviations for each graphic measurement along all sparsity values. (a-c) The sparsity-dependent clustering coefficient (C_p), the shortest path length (L_p), and global efficiency (E_glob) quantified from control (blue), E+ (red), and E− animals (pink). Significant differences resulting from post-hoc multi-comparisons are presented as grey/black dots along the sparsity range. Specifically, E+ animals have a larger C_p and Eglob over the majority of the sparsity range (0.14<S<1), while E− animals showed no difference to control. (d-f) The normalized characteristic path length (λ), the normalized clustering coefficient (γ), and small-worldness (σ), values quantified from the control (blue), E+ (red), and E− group (pink). Compared to the control group, the E+ group yielded the lowest small-worldness parameters over the entire sparsity range. In contrast, the E− group were not significantly different from the control group at any sparsity level. (g) The 7 selected brain network parameters used for classification. (h) The outcome of the estimation for predictor importance. Results indicated that all 7 parameters have nearly equal weights as predictors. (i) The outcome of the unsupervised learning through a spectral clustering approach. By applying a down-dimension estimation (metafeatures), the E+ and E− groups are clearly discriminated.

Fig. 4.

Topographical representations of local network metrics parameterized by nodal degree (Ki, a-c) and nodal efficiency (Enod, f-h) in control, E+ and E− groups. The spheres plotted in the 3D brain template indicated the brain areas that pass the threshold for hubness in Ki (upper plots) and Enod (bottom plots). The sphere size represents the hubness strength, the larger the size of sphere, the larger the hubness strength. Sphere colors represent clusters of brain areas: Frontal (red), Temporal (blue) and Parietal (green). For both hubness parameters, the prelimbic cortex (Prl) were found to be new hubs in the E+ rats (Fig 4b, 4g). In the E− group, we identified the primary motor cortex (M1) as new hubs compared to the control group (Fig 4c, 4h). It is also evident that in the E+ group, the ventral hippocampal area (HipP) was no longer considered to be a hub. (d) Regional group comparisons of the normalized nodal degree in E+ vs. control and E− vs. control. The red stars represent the brain areas are significantly larger in E+/E− group in comparison of control, and blue stars represent the significantly smaller. The “+” represent the comparison between E+ and E− group. Major differences between E+ vs E−, and E− vs control were showed in (Fig 4e, red arrow: increase; blue arrow: decrease). For the hubness changes under the parameters Enod (f-h), consistent results were observed except the HipPR hub (h, red square) in E− group.