Differences in paracingulate connectivity associated with epileptiform discharges and uncontrolled seizures in genetic generalized epilepsy
Benjamin P. Kay,
Neuroscience Graduate Program, University of Cincinnati, Cincinnati, Ohio, U.S.A
Pediatric Neuroimaging Research Consortium, Cincinnati Children's Hospital Medical Center, Cincinnati, Ohio, U.S.A
Address correspondence to Benjamin P. Kay, Graduate Program in Neuroscience, University of Cincinnati Academic Health Center, 260 Stetson Street, Suite 2300, Cincinnati, OH 45267-0525, U.S.A. E-mail: firstname.lastname@example.org
Patients with genetic generalized epilepsy (GGE) frequently continue to have seizures despite appropriate clinical management. GGE is associated with changes in the resting-state networks modulated by clinical factors such as duration of disease and response to treatment. However, the effect of generalized spike and wave discharges (GSWDs) and/or seizures on resting-state functional connectivity (RSFC) is not well understood.
We investigated the effects of GSWD frequency (in GGE patients), GGE (patients vs. healthy controls), and seizures (uncontrolled vs. controlled) on RSFC using seed-based voxel correlation in simultaneous electroencephalography (EEG) and resting-state functional magnetic resonance imaging (fMRI) (EEG/fMRI) data from 72 GGE patients (23 with uncontrolled seizures) and 38 healthy controls. We used seeds in paracingulate cortex, thalamus, cerebellum, and posterior cingulate cortex to examine changes in cortical-subcortical resting-state networks and the default mode network (DMN). We excluded from analyses time points surrounding GSWDs to avoid possible contamination of the resting state.
(1) Higher frequency of GSWDs was associated with an increase in seed-based voxel correlation with cortical and subcortical brain regions associated with executive function, attention, and the DMN; (2) RSFC in patients with GGE, when compared to healthy controls, was increased between paracingulate cortex and anterior, but not posterior, thalamus; and (3) GGE patients with uncontrolled seizures exhibited decreased cerebellar RSFC.
Our findings in this large sample of patients with GGE (1) demonstrate an effect of interictal GSWDs on resting-state networks, (2) provide evidence that different thalamic nuclei may be affected differently by GGE, and (3) suggest that cerebellum is a modulator of ictogenic circuits.
Dr. Benjamin Kay is a Medical Scientist Training Program student at the University of Cincinnati.
Genetic generalized epilepsy (GGE; formerly known as idiopathic generalized epilepsy, IGE) is a seizure disorder of presumed genetic etiology that affects all age groups and accounts for 15–30% of all epilepsy cases. More than 20% of GGE patients experience ongoing or “uncontrolled” seizures despite adequate clinical management.[3, 4] GGE is typically associated with normal intelligence, but many patients exhibit specific, frontal-lobe cognitive deficits[5-8] that are compounded by the negative effects of antiepileptic drugs (AEDs) on cognition.[9-11] Finally, psychiatric symptoms are common in patients with GGE, who often present with comorbid attention deficit, mood, and personality disorders.[13-15] Resting-state functional connectivity (RSFC) reflects anatomic (structural) connectivity in the brain and may be more sensitive than diffusion tensor imaging (DTI) at detecting changes associated with GGE. A growing body of evidence supports the notion of an interaction between functional/structural connectivity and neurologic and psychiatric disorders, including evidence for the effects of epilepsy on cognitive and emotive brain networks.[11, 12, 17-22]
Generalized spike and wave discharges (GSWDs) are a ubiquitous electroencephalographic hallmark of GGE as seen interictally and during absence and generalized tonic–clonic seizures (GTCS).[23, 24] The prevailing theory of GSWD etiology posits that synchronized neuronal activity in reentrant thalamocortical circuits gives rise to generalized seizures,[25-27] possibly via a mechanism shared with sleep spindles.[28, 29] This theory is supported by animal models of generalized epilepsies. Subdural and depth electrodes provide a gold standard of evidence in humans, but these surgical recording techniques are highly invasive and not performed routinely in patients with GGE. Simultaneous electroencephalography and functional magnetic resonance imaging (EEG/fMRI) provides a noninvasive imaging modality with high spatial resolution suitable for clinical and experimental human use. Studies of GGE have shown thalamic and widespread cortical GSWD-related activation consistent with the cortical-subcortical hypothesis of GSWD genesis.[33-39]
EEG/fMRI studies have also demonstrated cortical deactivation associated with GSWDs, specifically in regions corresponding to the default mode network (DMN),[33-38] a resting-state network thought to support consciousness.[40-43] GSWD-related deactivation of the DMN has been hypothesized as an explanation for absence seizure semiology, consistent with the network inhibition hypothesis.[44, 45] Previous RSFC studies of GGE have observed reduced DMN connectivity in patients versus controls, even when EEG/fMRI is used to exclude the effect of GSWDs.[18, 19] This relationship is dependent on response to treatment and the length of time (i.e., duration) for which a patient has had epilepsy. The effects of GSWD frequency on RSFC and the DMN have not, to our knowledge, been previously investigated. Therefore, in this EEG/fMRI study we use seed-based voxel correlations to investigate the effects of GSWDs, GGE, and treatment-resistance on RSFC in cortical-subcortical networks and the DMN.
This study examined a previously described cohort of 100 patients with epilepsy and 40 healthy controls.[18, 38, 46] Epilepsy patients who satisfied published criteria for the diagnosis of GGE were enrolled after evaluation at the Cincinnati Epilepsy Center. Diagnosis and treatment were directed by an epilepsy specialist with specific inclusion and exclusion criteria published previously. All participants in the study provided written informed consent for a protocol approved by the institutional review boards of the University of Cincinnati and the Cincinnati Children's Hospital Medical Center. Each GGE patient underwent 1–3 consecutive 20-min EEG/fMRI scans, and each healthy control subject underwent 1–2 consecutive scans. All patients and 20/40 control subjects listened to self-selected music during scanning to increase comfort and compliance. The effect of music-listening on resting-state data is discussed elsewhere and was included as a covariate in analyses. As previously, seizure freedom (Seizures−) was defined as no seizures in the 3 months preceding the scanning session,[24, 38] whereas patients with any seizures (absence, tonic–clonic, or myoclonic) were included in the uncontrolled group (Seizures+). Evidence for seizures was obtained from personal interviews with the patient(s) by the primary neurologist/epilepsy specialist, by an interview at the time of EEG/fMRI scanning, and, in a majority of patients by 24–72 h ambulatory EEG.
Eleven GGE patients failed to complete at least one scan due to claustrophobia (n = 3), metallic artifacts (n = 1), or not wanting to continue the procedure (n = 7). Of the 89 GGE patients who completed scanning, 15 patients were excluded due to poor quality data and an additional two were excluded because a high number of GSWDs led to an ill-conditioned design matrix. All control subjects completed scanning, but two were excluded due to poor quality of the data. Thirty scans (28 GGE and two healthy controls) were excluded in total. Resting-state analysis was carried out on 231 scans from 72 GGE patients (152 scans) and 38 healthy controls (67 scans; Table 1). There was no significant difference in root mean square displacement or average motion cost between GGE patients and healthy control subjects (accounting for age and music-listening, displacement p = 0.20, cost p = 0.56), the Seizures+ and Seizures− patient groups (accounting for age, p = 0.15, p = 0.19), or patients who were and were not taking valproate, which is associated with tremor (p = 0.884, p = 0.978), and these motion measures were not correlated with GSWD frequency (p = 0.62, p = 0.24).
Table 1. Demographics of generalized epilepsy (GGE) patients, subdivided by clinical feature, and healthy controls
No. of runs
No. of males
No. of females
Age (years ± SD)
Duration (years ± SD)
No. of current drugs
No. of failed drugs
Seizures−, epilepsy patients who were seizure-free during the 3 months leading up to scanning; Seizures+, GGE patients who experienced at least one seizure during the 3 months leading up to scanning; JME, juvenile myoclonic epilepsy.
Mean age and duration of epilepsy at scanning (±standard deviation) are given in years.
31.3 ± 11.5
14.5 ± 11.4
32.3 ± 10.9
14.9 ± 10.7
29.1 ± 12.6
13.6 ± 13.1
31.9 ± 11.3
14.9 ± 10.5
29.9 ± 12.1
13.6 ± 13.4
27.8 ± 8.8
13.2 ± 9.0
34.8 ± 12.9
15.8 ± 13.4
30.9 ± 10.2
EEG acquisition and processing
Acquisition and processing of EEG data simultaneous with fMRI was carried out as described previously[18, 38] using Scan 4.3.5 software (Compumedics USA, Ltd., El Paso, TX, U.S.A.). Briefly, subjects were fitted with an MRI-compatible EEG cap with electrodes arranged according to the international 10/20 system (Compumedics USA, Ltd.). Conductive gel (Quik-Gel; Compumedics Neuromedia Supplies, Charlotte, NC, U.S.A.) was used to establish low impedance (confirmed as <20 kΩ) between each electrode and the scalp. Sixty-four channels of data, including an electrocardiography (ECG) channel, were recorded at 10 kHz concurrent with fMRI using an MRI compatible system. Time marks generated by the scanner at the onset of each volume acquisition were used to reduce gradient-related artifacts via an average artifact subtraction method. Heartbeat timings generated from the ECG channel were used to reduce the ballistocardiographic artifact via a linear spatial filtering method. All EEG data were reviewed by a board certified epilepsy specialist (JPS), and GSWD timings were marked to within 10 msec precision.
MRI acquisition and processing
Acquisition of MRI and functional MRI (fMRI) data was carried out on a 4 Tesla, 61.5 cm bore Varian Unity INOVA system (Varian, Inc., Palo Alto, CA, U.S.A.) equipped with a standard head coil.[18, 38] T1-weighted structural images were acquired for use as an anatomic reference. A modified driven equilibrium Fourier transform (MDEFT) method was used with an 1,100 msec inversion delay, 256 × 196 × 196 mm field of view, 256 × 196 × 196 voxel matrix, 22-degree flip angle, and repetition/echo time (TR/TE) = 13.1/6.0 msec. T2*-weighted echo-planar functional images with blood oxygenation level–dependent (BOLD) contrast, were acquired with an 256 × 256 mm field of view, 64 × 64 voxel matrix, 90-degree flip angle, and 5 mm slice thickness in axial orientation without gap. Four hundred volumes consisting of 30 slices each and repetition/acquisition time (TR/TA) = 3,000/2,000 msec were collected during each scan.
Data were reconstructed and corrected for geometric distortion and Nyquist ghosting with the aid of multi-echo reference scans (MERS). Functional scans underwent slice timing correction, motion correction, rigid-body registration to a high-resolution anatomic scan, and nonlinear registration to a Montreal Neurological Institute MNI152 standard using the FMRIB Software Library (FSL). Data were spatially blurred in-mask with a gaussian kernel of full width at half maximum (FWHM) = 6 mm using Analysis of Functional NeuroImages (AFNI). The quality of functional to anatomic registration was measured using the mutual information cost function. Twelve functional scans with an outlying cost indicating unsatisfactory registration were excluded from the study. The quality of motion correction was measured using the normalized correlation ratio cost function of each time point to the reference volume. Thirteen additional scans with outlying costs indicating excessive motion were excluded from the study.
Seed-based voxel correlation
Seed selection was guided by a previous study of the same patient cohort in which we identified a set of brain regions with GSWD-related activation that was increased in patients who were resistant to treatment with valproate. These included a cluster of voxels in midline paracingulate cortex, located functionally between “sensorimotor” and “executive-control” networks. This region was used as an a priori seed with its centroid at Montreal Neurological Institute (MNI) coordinates X = 2.0, Y = 13.6, and Z = 45.9, and a volume of 38 voxels. Two a posteriori seeds were manually generated from regions exhibiting high functional connectivity with the a priori paracingulate seed. These were bilateral seeds located in dorsal anterior thalamus (X = ±9.2, Y = −15.6, Z = 13.6, five voxels each side, 10 voxels total) and cerebellum (X = ±31.6, Y = −56.4, Z = −28.8, six voxels each side, 12 voxels total). A second a priori seed in the posterior cingulate cortex (PCC, X = 2.0, Y = −58.0, Z = 24.0, 19 voxels), a region regarded as a default mode network “hub,” was used to investigate DMN connectivity.[18, 20]
The mean time course of voxels within each seed region was extracted prior to spatial blurring and was then used as the regressor of interest in a general linear model of the spatially blurred fMRI data for each subject (3dDeconvolve tool in AFNI). The output of 3dDeconvolve was submitted to the 3dREMLfit tool in AFNI to achieve temporal prewhitening via an autoregressive (AR) model. Baseline drift was modeled using a first-order polynomial because no physiologic regressors were available. No global or tissue regressors were used because these may introduce an unwanted bias. However, motion has been shown to have an artifactual effect on resting-state connectivity.[55, 56] Therefore, we included the six-rigid body motion parameters generated by FSL as nuisance regressors in the model. In addition, time points associated with high motion measured as the normalized correlation ratio cost function >0.00185 to the reference volume were excluded from analysis. Three time points were excluded: those preceding, including, and following each high-motion volume.
We included subjects and scans containing interictal GSWD in our analysis in order to examine the relationship between RSFC and interictal GSWD frequency. To avoid possible contamination of the resting-state by GSWD, time points associated with GSWD were excluded from analysis in a manner analogous to the exclusion of high-motion time points. A total of 19 time points comprising 57 s were excluded for each GSWD: the 9 preceding, 9 following, and one including the GSWD. The exclusion of time points from the general linear model due to GSWD and motion resulted in ill-conditioned design matrices for two GGE subjects (five scans) with very frequent GSWD who were therefore excluded from the study.
The Pearson correlation coefficient of each voxel with the seed time course was converted to a z-value using the Fisher transformation. Voxelwise analysis of the resultant z-values was carried out using R. GGE patients were divided into two groups: those who had experienced at least one seizure during the 3 months leading up to scanning (Seizures+) and those who were seizure-free (Seizures−). T-maps of connectivity for all GGE patients versus controls and for GGE patients who were Seizures+ versus GGE patients who were Seizures− were computed with age and music-listening as covariates. The correlation between connectivity and GSWD frequency (measured as number of GSWD/number of scans) in GGE patients was also computed.
Cluster-based correction for multiple comparisons was carried out using the AlphaSim and 3dmerge tools in AFNI at a significance level of α = 0.05. Ventricular and white matter masks were generated from the Harvard-Oxford subcortical probabilistic atlas (p > 50%) distributed with FSL. Observations within these regions were assumed to be artifactual; therefore, t-maps were masked prior to cluster-based thresholding to avoid inflation of cluster sizes by spurious correlations.
GSWD frequency (number of GSWD/number of scans) was significantly correlated with RSFC in GGE patients for each seed. The most widespread increases in RSFC with GSWD frequency were observed for the paracingulate seed and included the frontal areas of superior frontal, medial frontal, inferior frontal, orbitofrontal, and anterior cingulate cortex (ACC); the posterior areas of precuneus, lingual gyrus, lateral occipital, and posterior cingulate cortex (PCC); and subcortical regions thalamus and basal ganglia. Increased RSFC was also observed in precentral gyrus and posterior cerebellum (Fig. 1A). Similar results were obtained for the PCC seed, except that the posterior regions of precuneus, lingual gyrus, and lateral occipital cortex were unchanged (Fig. 1B). RSFC with the cerebellar seed increased significantly with spike frequency for ACC, left insula, right thalamus, and posterior cerebellum (Fig. 1C). RSFC with the thalamic seed increased significantly with spike frequency for right insula, right thalamus, and left basal ganglia (Fig. 1).
The relationship between GSWD frequency and RSFC was also investigated using a seed in PCC, a default mode network (DMN) hub region. Results were similar to those obtained via the paracingulate seed and included significant positive correlations between RSFC and GSWD frequency in precentral gyrus, insula, thalamus, basal ganglia, superior frontal, medial frontal, inferior frontal, orbitofrontal, and anterior cingulate cortex (Fig. 1D).
GGE patients versus healthy controls
No significant difference in corticothalamic RSFC between GGE patients and healthy controls was detected using the a priori paracingulate seed (Fig. 2A), but significant changes in thalamocortical RSFC were observed using the a posteriori thalamic seed (Fig. 2B). RSFC between thalamus and medial frontal cortex (X = −0.7, Y = 9.1, Z = 55.6, 41 voxels) was significantly greater in GGE patients versus healthy controls, whereas RSFC between thalamus and posterior cingulate cortex (PCC, X = −0.5, Y = −28.3, Z = 30.0, 33 voxels) was significantly reduced.
The a posteriori thalamic seed was manually generated from subthreshold (i.e., not significant) trends in corticothalamic RSFC observed for GGE patients versus healthy controls observed using the paracingulate seed (Fig. 3). GGE patients trended toward greater connectivity than did healthy controls between the paracingulate seed and dorsal anterior thalamus, but they trended toward reduced RSFC between the paracingulate seed and ventral posterior thalamus (see Table 2).
Table 2. Subthreshold (|t| > 0.25) clusters of thalamic connectivity with the paracingulate seed in GGE patients versus healthy controls. The thalamic hemisphere (left vs. right side), sign (increased vs. decreased connectivity), and MNI coordinates (X, Y, Z) are given (see Fig. 3)
Seizures+ versus Seizures− patients
GGE patients with uncontrolled seizures (Seizures+) exhibited significantly reduced RSFC between the paracingulate seed and bilateral cerebellum (X = −32.0, Y = −54.9, Z = −30.3, 36 voxels; X = 34.0, Y = −53.7; Z = −26.9, 27 voxels) when compared to the Seizures− GGE patients (Fig. 4A). A manually created a posteriori seed in the cerebellum exhibited significantly reduced reciprocal RSFC with the paracingulate seed region in Seizures+ versus Seizures− (Fig. 4B). The cerebellar seed also exhibited significantly reduced connectivity with thalamus, basal ganglia, and cortex diffusely.
We observed significant functional connectivity changes in GGE patients correlated with GSWD frequency despite excluding a generous amount of fMRI data (19 TRs = 57 s) around each GSWD detected using simultaneous EEG. The most extensive findings were for the paracingulate seed, which is associated with treatment resistance, and the PCC seed, which measures DMN RSFC. Increased correlations of paracingulate RSFC with GSWD frequency (Fig. 1A) were observed in frontal regions associated with attention, anterior insulae and posterior regions associated with the DMN, and subcortical regions. Increased correlations of PCC RSFC with GSWD were also observed in frontal/attentional regions. These findings support the hypothesis that RSFC is altered in brains with frequent GSWD activity. Executive networks, especially the DMN, appear to be the most affected, consistent with reduced DMN connectivity[18-20] and executive function[5-8, 12-15] in GGE patients.
Paracingulate RSFC with precentral gyrus was correlated with GSWD frequency (Fig. 1A). This finding could be an artifact of the analysis, as the paracingulate seed is located functionally on the border of motor cortex. However, paracingulate RSFC was correlated with GSWD frequency for thalamus and basal ganglia as well, subcortical regions that support both motor and executive functions. Therefore, the involvement of precentral gyrus could alternatively be interpreted as evidence for increased crosstalk between motor, default mode, and executive networks in patients with frequent GSWDs.
GGE patients versus healthy controls
Whereas GSWDs are associated with widespread cortical activation, we found evidence for increased thalamocortical RSFC only in medial frontal cortex (Fig. 2B). This is consistent with the so-called “cortical focus” hypothesis that specific cortical regions (i.e., medial frontal cortex) sustain reentrant activity in thalamocortical ictogenic circuits.[38, 60, 61] However, this finding is equivocal because significant changes in thalamocortical RSFC were found using the a posteriori thalamic (Fig. 2B), but not a priori paracingulate (Fig. 2A) seed, and could thus have been biased by manual seed selection (i.e., “seed-fishing”).
Despite considerable evidence for the thalamocortical hypothesis of GSWD, changes in thalamocortical connectivity are not consistently observed using MRI. Significance in fMRI is often determined using cluster-based thresholding, which required at least 26 suprathreshold voxels in our analysis. Anatomically, there are 36 voxels in each thalamic hemisphere (Harvard-Oxford Subcortical Atlas, probability > 90%) at our imaging resolution. Therefore, implicit in our analysis is the assumption that at least 26/36 = 72% of thalamus is affected the same way by GGE. However, in Figure 3 we observe positive and negative clusters of subthreshold changes in RSFC that divide the thalamus roughly in half. Although the findings in Figure 3 are based on subthreshold analyses, they are consistent with previous studies in which thalamic nuclei are affected differently by epilepsy, exhibit different, and even opposite, connectivity changes in epilepsy, and play different roles in the initiation versus maintenance of seizure activity. Therefore, the statistical power to detect thalamic changes in RSFC associated with GGE is limited in our and potentially other fMRI studies.
Seizures+ versus Seizures− patients
At least two previous EEG/fMRI studies have discussed inconclusive cerebellar findings in GGE.[21, 28] The cerebellum shares reciprocal connections with thalamus, cortex, and basal ganglia through which it could, theoretically, modulate ictogenic activity throughout the brain. Paracingulate RSFC in Seizures+ versus Seizures− was significantly reduced exclusively with cerebellum (Fig. 4A); however, the same relationship was not observed for GGE patients versus controls (Fig. 2A). Reciprocal cerebellar RSFC was reduced in thalamus, basal ganglia, and most of cortex for Seizures+ versus Seizures− (Fig. 4B). Although cerebellum is not thought to be a primary cause of ictogenesis in GGE, these data suggest that loss of cerebellar connectivity with thalamus, basal ganglia, and cortex is associated with seizures that are resistant to treatment. Prospective studies would be needed to establish causality and rule out confounding effects of AED treatment.
This study was supported in part by a grant from the National Institute of Neurological Disorders and Stroke (K23 NS052468) and in part by funds from the Department of Neurology at the University of Cincinnati Academic Health Center, Cincinnati, OH, U.S.A. The first author received support from the Medical Scientist Training Program (MSTP) at the University of Cincinnati (T32 GM063483).
None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.