This diversity of responses was puzzling, and we discussed how neuronal excitation and inhibition could contribute to this heterogeneity, in the context of the known thalamocortical interactions characterizing these discharges.

Subsequent to that publication, we became aware of the similarity between the pattern of deactivation seen in some of our patients and the pattern of deactivation seen in normal subjects when the resting awake state is disturbed by external stimuli or cognitive tasks, leading to the concept of a “default” state of brain function (4). This pattern includes deactivation in the parietal regions, the posterior cingulate gyrus, and in the anterior frontal regions, symmetrically. If a link between the spike-and-wave discharges and the suspension of the default state of the brain could be demonstrated, this would explain the distribution of many of the deactivations we found and shed light on the effect of spike-and-wave on brain function. In addition, it could possibly strengthen the concept of default state of brain function itself.

We therefore performed a group analysis on the data from the 15 patients of our previous study (3) to assess the activations and deactivations that were consistent across the patients. Such group analyses have rarely been performed in the study of epileptic patients with EEG/fMRI because of the heterogeneity of the clinical and EEG data, as well as of the fMRI responses, particularly in patients with focal epilepsy. The patients with IGE, who can present generalized seizures of different types, are nevertheless considered a homogeneous epilepsy group because of clinical characteristics and natural history. Furthermore, we are studying their widespread epileptic discharges, and these discharges are quite similar, justifying the group analysis approach.

We postprocessed the EEG acquired inside the scanner, for offline MRI artifact removal and filtering by using femr software (Schwarzer, Munich). An experienced neurophysiologist reviewed filtered EEGs and marked bursts of spike-and-wave.

The functional images were motion-corrected and smoothed [6-mm full-width at half-maximum (FWHM)] by using the software package from the Brain Imaging Center of the Montreal Neurological Institute. Models and signals were prewhitened with an autoregressive filter of order 1, and low frequency drifts in the signal were modeled with a third-order polynomial fitted to each run. Maps of the T statistic (T-maps) were created by using the timing of each spike-and-wave burst as an event in the fMRI analysis [fmristat software (5)]. Maps were obtained for each of four hemodynamic response functions (HRFs), modeled by γ functions peaking at 3, 5, 7, and 9 s after the discharge as described by Bagshaw et al. (6). The analysis described below was repeated for each HRF.

For each subject i and each scanning run j, the method provided an estimate of the effect of interest E_ij and its standard deviation S_ij, and thus a T statistic T_ij = E_ij/S_ij. For each subject, the data of each run were registered and resampled to the position of the first run by using a rigid transformation (three rotations and three translations). Results for all runs of a subject were combined by using a fixed-effects model, leading to an estimation of the global statistics E_i, S_i, and T_i for each subject. To perform an intersubject group analysis, the anatomical MRI scans of the 15 patients were spatially normalized in the stereotactic space of the averaged brain template provided by the Montreal Neurological Institute (7), and the resulting geometrical transformations were used to resample the results of individual fMRI analyses E_i and S_i. The group analysis was performed by using a random-effects linear model with the effects, E_i (as data), and fixed-effects standard deviations, S_i, taken from the previous analysis. This model was fitted by using restricted maximum likelihood and provided notably an estimate of the random-effects variance (5). This last estimate was particularly noisy because of the low degrees of freedom (14 in the present study). To increase the sensitivity of this group analysis, we used the method proposed by Worsley et al. (5), which consists in estimating the ratio of the random-effects variance to the fixed-effects variance, and then regularizing this ratio by spatial smoothing. The variance of the effect is finally estimated by the smoothed ratio multiplied by the fixed-effects variance. The amount of smoothing was chosen to achieve 100 effective degrees of freedom. Note that by using such an approach only the variance ratio is smoothed but not the effects, E_i. This group analysis, to which we will refer as a mixed-effects group analysis, provided the statistics E, S, and T.

Activations. The activations were generally symmetrical, involving homologous areas, and close to the midline and most apparent in the analysis with an HRF peaking at 5 s. The clusters found with the 3-s HRF are a subset of those found at 5-s HRF and are therefore not illustrated. The largest cluster includes the thalami and insulae in both hemispheres (Fig. 1B). The highest T values were found in the thalamus in both hemispheres (T = 5.2). Other regions of significant activation included a wide band of mesial frontal cortex (Fig. 1C) and the cerebellum (Fig. 1 A and C). There was also an activation that closely followed the lateral ventricles, only apparent in the analysis with an HRF peaking at 7 s (Fig. 1D).

Activations in the group analysis were most significant in the analysis with an HRF peaking at 5 s, whereas deactivations were most significant with the HRF peaking at 9 s. This finding is in agreement with the observation of our original publication (3) and of another study (6) in patients with focal epilepsy, that deactivations tended to peak later than activations. It is also clear that, in this patient group, deactivations and activations were spatially separated, not even adjacent, and therefore the deactivations are very unlikely to be the result of an undershoot after an activation. Although we cannot formally exclude the possibility of a vascular steal phenomenon, the distance between regions of activation and deactivation, as well as their difference in timing, make this unlikely.

The most significant activation was in the thalamus, bilaterally, a structure that has long been suspected to be involved in the generation of generalized epileptic discharges, in interaction with the cortex (9). This result confirms the pattern seen in the majority of subjects, and its significance in the context of generalized epileptic discharges was discussed extensively by Aghakhani et al. (3). The group analysis also revealed, however, large midline frontal, bilateral insular, and cerebellar activations. Midline frontal regions have been implicated in the spreading and generalization of discharges of focal origin (10) but not specifically in the generation of generalized discharges. This finding is in agreement with the frequent midfrontal amplitude predominance in the EEG of these discharges. It is also in midfrontal regions that another thalamocortical phenomenon, the sleep spindle, is most prominent, and the link between the two has been established (11, 12).

More surprising was the finding of bilateral activation in the insulae. The insula is a region of convergence of multisensory inputs that has recently been implicated in focal seizures (13) but not in generalized discharges. The insula has widespread thalamic connections, however (14), which may explain its activation during generalized spike-and-wave. A clear activation was seen in the cerebellum. Cerebellar ictal hyperperfusion has been observed in patients with partial seizures, contralateral to the seizure focus (15), and it is therefore not so surprising to see bilateral involvement during bilateral discharges. Intense involvement of the cerebellum during spike-and-wave discharges has been observed in an experimental model, independently of movement, with the suggestion that cerebellar neurons may contribute to spike-and-wave rhythmicity (16). The role of the cerebellum during generalized seizures or discharges has been reviewed by Norden and Blumenfeld (17), but they do not propose a specific hypothesis regarding its function during seizures.

The positive response found in the lateral ventricles is surprising and of course cannot be explained by the usual mechanisms thought to underlie an increase in the BOLD signal. We conjecture that this is an artefactual finding, although we do not have an explanation for it. The ventricles change volume with respiration and the cardiac cycle, but our results originate in a change correlated to the spike-and-wave discharges, and such discharges are not known to be related to these physiological rhythms. Our patients did not, at the time of the study, manifest myoclonic jerks with their spike-and-wave discharges, and only three have a history of myoclonic jerks. We have verified that movement was minimal and movement correction was always performed. This response is therefore unlikely to be related to movement. Could it be that the widespread increase in blood volume resulting from the discharges results in a transitory reduction of volumes of the ventricles and hence a change in the magnetic characteristics of the border of the ventricles?

The group analysis revealed deactivations, including bilateral parietal and anterior frontal regions and the posterior cingulate gyrus. Deactivation in the posterior cingulate gyrus was also reported by Archer et al. (18), after group analysis of an EEG/fMRI study of five patients with spike-and-wave discharges. The authors interpreted this deactivation as possibly being related to the generation of the epileptic discharge itself or to unspecific attention mechanisms. This study, however, used a spike-triggered fMRI paradigm, less powerful than the continuous recording performed in our study, and this may explain why it found only a subset of the deactivated regions found in our study. A small cluster of deactivation was also found in our study in the posterior left temporal region. It is noteworthy that a symmetrical right temporal cluster appeared when we slightly lowered the statistical threshold, indicating that temporal deactivation may be bilateral but more consistent on the left.

Raichle and his colleagues (4, 19) have analyzed many fMRI and positron-emission tomography (PET) studies involving a “resting state” that is contrasted to a state during which a variety of tasks are performed. They found that, during many such experiments, the task condition is accompanied by a deactivation involving the frontoparietal cortical regions, including the posterior cingulate gyrus. They have hypothesized that this deactivation, which is not task-specific and which is relative to the resting state, corresponds in fact to the arrest of the “default state of the brain” during which the subject is not performing any specific task, but is awake and conscious. As soon as a task is presented, this state is suspended, and the subject concentrates on the task at hand. Our group analysis of spike-and-wave discharges showed deactivations that had a strikingly similar distribution to the deactivations found as a result of the interruption of this default state of the brain.

During EEG/fMRI recordings, our epileptic subjects lay with their eyes closed, as during the rest condition of the above-mentioned experiments. Our data suggest that, during a generalized epileptic discharge, the default state may also be momentarily suspended, not as a result of an external requirement to perform a task, but as a result of the epileptic discharge. We can hypothesize that spike-and-wave bursts cause a partial suspension of the default state of the brain. Therefore, this suspension of the normal state of attention, combined with a dampening in the perception of sensory inputs and reduction in the ability to respond resulting from activation in the thalamus, midfrontal regions, and insulae, may explain the state of altered consciousness seen during generalized epileptic discharges. This explanation may provide a more accurate understanding of the mental state of patients during spike-and-wave discharges than the usual “loss of consciousness,” which Gloor (20) thought unspecific and not a very useful description of the manifestation of epileptic seizures. Patients with spike-and-wave do not experience a clinical absence with every burst, but responsiveness is often impaired during such bursts (21), and the level of cognitive impairment increases when discharges become longer (see ref. 22 for a review). Similarly, it has been shown that the default state of the brain can be gradually and actively attenuated in accordance with the difficulty of the task (23, 24).

Frontal and parietal regions have also been involved in various conditions related to more severe alterations of consciousness. Generalized tonic-clonic seizures result in a profound loss of consciousness, and it has been hypothesized from blood flow studies (25) that frontal and parietal regions are primarily involved in the loss of consciousness, although in this case with increases in blood flow. When consciousness is impaired during temporal lobe seizures, frontal and parietal association areas show a decrease in blood flow (26). In the vegetative state, a more sustained and severe cortical impairment, frontoparietal regions show a decrease in glucose consumption (27), although in a somewhat different spatial distribution, with important involvement of premotor areas. The authors attribute impaired awareness in the vegetative state to this decrease and to decreased connectivity between frontal regions and the posterior cingulate area. These results are consistent with the concept of the default state of the brain implicating high metabolism in frontoparietal regions when nonspecific attention is maintained and with our interpretation of their role during generalized epileptic discharges.

We noted that the hemodynamic response functions of deactivations tended to peak later (7-9 s) than those of activations (≈5 s). It is possible that there is a 2- or 3-s delay between the onset of the epileptic burst, which results itself in the activations, and the suspension of the default state, which is represented by deactivations. An alternative explanation is that the mechanisms of generation of activations and deactivations are different, resulting in deactivations occurring later than activations. This hypothesis is supported by previous observations that deactivations also tend to peak later than activations in focal epilepsy (6).

We reexamined our original individual data with the hindsight of these results and found that a deactivation with the pattern of the default state of the brain was clearly present in 4 of 15 subjects, could be recognized but was more diffuse or partial in another 4 subjects, was present only in its posterior regions in 4 subjects, and was clearly absent in 3 subjects.