Altered FMRI connectivity dynamics in temporal lobe epilepsy might explain seizure semiology

Abstract

Temporal lobe epilepsy (TLE) can be conceptualized as a network disease. The network can be characterized by inter-regional functional connectivity, i.e., blood oxygen level-dependent (BOLD) signal correlations between any two regions. However, functional connectivity is not constant over time, thus computing correlation at a given time and then at some later time could give different results (non-stationarity). We hypothesized (1) that non-stationarities can be induced by epilepsy (e.g., interictal epileptic activity) increasing local signal variance and that (2) these transient events contribute to fluctuations in connectivity leading to pathological functioning, i.e., TLE semiology. We analyzed fMRI data from 27 patients with TLE and 22 healthy controls focusing on EEG-confirmed wake epochs only to protect against sleep-induced connectivity changes. Testing hypothesis (1), we identified brain regions where the BOLD signal variance was significantly greater in TLE than in controls: the temporal pole - including the hippocampus. Taking the latter as the seed region and testing hypothesis (2), we calculated the time-varying inter-regional correlation values (dynamic functional connectivity) to other brain regions and found greater connectivity variance in the TLE than the control group mainly in the precuneus, the supplementary and sensorimotor, and the frontal cortices. We conclude that the highest BOLD signal variance in the hippocampi is highly suggestive of a specific epilepsy-related effect. The altered connectivity dynamics in TLE patients might help to explain the hallmark semiological features of dyscognitive seizures including impaired consciousness (precuneus, frontal cortex), sensory disturbance, and motor automatisms (sensorimotor cortices, supplementary motor cortex). Accounting for the non-stationarity and state-dependence of functional connectivity are a prerequisite in the search for potential connectivity-derived biomarkers in TLE.

Keywords: EEG-fMRI; biomarker; functional connectivity; interictal epileptiform discharges; non-stationarity; seizure; semiology; temporal lobe epilepsy.

Figures

Figure 1

Procedure to study BOLD signal and functional connectivity temporal variability. (A) To compute the voxel-wise map of BOLD signal variability, the time series for every voxel in the EPI data is first extracted. Then, the variance of the signal is computed, resulting in the required map. (B) To compute the voxel-wise map of BOLD functional connectivity temporal variability with a seed, a region in the AAL template is first selected (in this work, the left and right hippocampi are used as seeds) and the average signal in the region is computed. Then, for every voxel in the EPI data, correlations over time are obtained using a sliding window (30 s) and the temporal dynamics of functional connectivity are computed. Note that in this example the dynamics are non-constant with moments of drastic loss of connectivity between regions. Finally, the temporal variance of the functional connectivity time series is computed, resulting in a spatial map (see Figure 4) encoding the variability in the interaction between the seed region and every voxel.

Figure 2

Distribution of percent time spent in different sleep stages (wake, N1 and N2; no N3 sleep was observed) for healthy controls, left and right TLE patients (mean ± SEM).

Figure 3

Spatial map (coronal, sagittal, axial slices) of significantly greater variance of the blood oxygen level-dependent signal in patients with temporal lobe epilepsy (pooled right and left) than in healthy controls. Color bar indicates p-value (thresholded at p < 0.001 for display, cluster survives family-wise error correction at p < 0.05). Left on figure is right in the brain (coronal and axial slices).

Figure 4

Spatial map (coronal, sagittal, axial slices) of significantly greater variance of hippocampal dynamic functional connectivity in patients with temporal lobe epilepsy (here: seed in left hippocampus) than in healthy controls. For differences between seeding in the left hippocampus and in the right, see Tables 3 and 4, of the pooled analysis, only results for the left hippocampus as the seed region are displayed (seed in right hippocampus yielded most similar results). Color bar indicates p-value (thresholded at p < 0.001 for display, cluster survives family-wise error correction at p < 0.05). Left on figure is right in the brain (coronal and axial slices).

Figure 5

Spatial map (coronal, sagittal, axial slices) of significantly greater variance of the blood oxygen level-dependent signal in patients with temporal lobe epilepsy (pooled right and left) than in healthy controls. An additional preprocessing step was performed by erasing volumes associated with large head displacements, as well as surrounding volumes. Color bar indicates p-value (thresholded at p < 0.001 for display, cluster survives family-wise error correction at p < 0.05). Left on figure is right in the brain (coronal and axial slices).

Figure 6

Spatial map (coronal, sagittal, axial slices) of significantly greater variance of hippocampal dynamic functional connectivity in patients with temporal lobe epilepsy (here: seed in left hippocampus) than in healthy controls. An additional preprocessing step was performed by erasing volumes associated with large head displacements, as well as six surrounding volumes. Color bar indicates p-value (thresholded at p < 0.001 for display, cluster survives family-wise error correction at p < 0.05). Left on figure is right in the brain (coronal and axial slices).

Figure 7

Spatial map (coronal, sagittal, axial slices) of significantly greater variance of hippocampal dynamic functional connectivity in patients with temporal lobe epilepsy (here: seed in left hippocampus) than in healthy controls. A sliding window length of 15 s was used for the computation of dynamic functional connectivity time series. Color bar indicates p-value (thresholded at p < 0.001 for display, cluster survives family-wise error correction at p < 0.05). Left on figure is right in the brain (coronal and axial slices).

Figure A1

Spatial map (coronal, sagittal, axial slices) of significantly decreased hippocampal signal linear correlation in patients with temporal lobe epilepsy (here: seed in left hippocampus) than in healthy controls. Color bar indicates p-value (thresholded at p < 0.001 for display, cluster survives family-wise error correction at p < 0.05). Left on figure is right in the brain (coronal and axial slices).

Figure A2

Correlations between BOLD signal variance/variance of dynamic connectivity time series with the left hippocampus and VIQ/PIQ scores for the TLE group (left + right). Statistical significance was determined at p < 0.05 with a cluster threshold of 10 voxels. Maps were masked with the regions where the metrics were significantly different between controls and TLE patients. Left: results for PIQ and VIQ correlation with variance. In all cases positive correlations were found. Right: results for PIQ and VIQ correlation with variance of connectivity time series with the left hippocampus. In all cases negative correlations were found.