OUP user menu

fMRI activation during spike and wave discharges in idiopathic generalized epilepsy

Y. Aghakhani, A. P. Bagshaw, C. G. Bénar, C. Hawco, F. Andermann, F. Dubeau, J. Gotman
DOI: http://dx.doi.org/10.1093/brain/awh136 1127-1144 First published online: 19 March 2004


The objectives of this study were to evaluate the haemodynamic response of the cerebral cortex and thalamus during generalized spike and wave or polyspike and wave (GSW) bursts in patients with idiopathic generalized epilepsy (IGE). The haemodynamic response is measured by fMRI [blood oxygenation level‐dependent (BOLD) effect]. We used combined EEG–functional MRI, a method that allows the unambiguous measurement of the BOLD effect during bursts, compared with measurements during the inter‐burst interval. Fifteen patients with IGE had GSW bursts during scanning and technically acceptable studies. fMRI cortical changes as a result of GSW activity were present in 14 patients (93%). Changes in the form of activation (increased BOLD) or deactivation (decreased BOLD) occurred symmetrically in the cortex of both hemispheres, involved anterior as much as posterior head regions, but were variable across patients. Bilateral thalamic changes were also found in 12 patients (80%). Activation predominated over deactivation in the thalamus, whereas the opposite was seen in the cerebral cortex. These results bring a new light to the pathophysiolocal mechanisms generating GSW. The spatial distribution of BOLD responses to GSW was unexpected: it involved as many posterior as anterior head regions, contrary to the usual fronto‐central predominance seen in EEG. The presence of a thalamic BOLD response in most patients provided, for the first time in a group of human patients, confirmation of the evidence of thalamic involvement seen in animal models. The possible mechanisms underlying these phenomena are discussed.

  • EEG–fMRI; idiopathic generalized epilepsy; generalized spike and wave; BOLD response
  • BOLD = blood oxygenation level‐dependent; CBF = cerebral blood flow; fMRI = functional MRI; GSW = generalized spike and wave; HRF = haemodynamic response function; IGE = idiopathic generalized epilepsy


Idiopathic generalized epilepsy (IGE) is characterized by absence seizures, tonic–clonic seizures, myoclonic jerks and paroxysmal generalized spike and wave (GSW) discharges in the EEG. There is a considerable literature on the possible mechanism of generation of three cycles per second GSW discharges, starting with the centrencephalic hypothesis (Penfield and Jasper, 1954) through to the most commonly accepted ‘cortico‐reticular’ theory (Gloor, 1969). Gloor proposed that GSW bursts are generated by an interplay between thalamus and cortex by which normal thalamic discharges are sent to a slightly hyperexcitable cortex, which responds with spike and wave activity. In the many animal models of GSW, whether genetic or pharmacological, the respective roles of thalamo‐cortical cells and of cortical cells are highly variable (Coenen et al., 1992; Avoli, 1995; Danober et al., 1998; Steriade and Contreras, 1998).

The above hypotheses come largely from work on experimental animal models of GSW activity. In humans, apart from the classical scalp EEG recording and a few observations with intracerebral electrodes (Williams, 1953; Niedermeyer et al., 1969; Velasco et al., 1989), it has only been possible to measure the functional consequences of neuronal activity, primarily with PET. PET studies of metabolism, with a typical temporal resolution of several minutes, include periods with GSW bursts as well as background activity. They have shown variable results, including increases in metabolism, decreases or mixed results (Engel et al., 1985; Theodore et al., 1985; Ochs et al., 1987). One study that used H215O to measure cerebral blood flow (CBF) consistently showed a generalized increase in CBF as well as an additional increase in the thalamus (Prevett et al., 1995). This study noted a remarkable lack of correlation between the number or duration of GSW bursts and the level of CBF changes, and suggested that CBF changes could be caused by the hyperventilation required to trigger GSW bursts during the necessarily time‐constrained studies.

The recently developed method of combined EEG and functional MRI (fMRI) allows the limitations of PET studies to be overcome. It is possible to measure specifically the blood oxygenation level‐dependent (BOLD) response of discrete EEG events such as interictal spikes or short GSW bursts, without confusing them with background. In focal epilepsy, this method has shown a good correspondence between the region of fMRI activation and the region presumed to generate the epileptic discharge (Warach et al., 1996; Krakow et al., 1999; Lazeyras et al., 2000; Jager et al., 2002; Al‐Asmi et al., 2003). A recent study reported the use of this method in a single patient with prolonged absence seizures, showing thalamic activation and widespread cortical ‘deactivation’ (reduction in the fMRI signal) (Salek‐Haddadi et al., 2003). We investigated patients with IGE and GSW, using the combined EEG–fMRI method to determine the importance and the consistency of the role of the thalamus and cortical regions in the generation of GSW discharges.

Subjects and methods

We selected 25 patients with IGE who had active interictal generalized spike or polyspike and wave at a frequency of 2–4 Hz during routine EEG. Informed consent was obtained in accordance with regulations of the Research Ethics Board of our institution, The Montreal Neurological Institute of McGill University, which approved the study.

EEG recordings and analysis

Twenty‐one Ag/AgCl electrodes were applied with conductive paste using the 10/20 system. Continuous ECG monitoring was also obtained. Electrodes were connected to an EMR32 amplifier (Schwarzer, Munich, Germany) with a sampling rate of 1000 Hz and a reference consisting of the average of O1 and O2. The amplifier was connected to the recording computer outside the scanner room via a fibre optic cable. The EEG was recorded in a dark room, and no hyperventilation, photic stimulation or sedative–hypnotic medications were used. The EEG was processed off‐line in order to filter out the scanner artefact (FEMR software, Schwarzer) (Hoffmann et al., 2000). We did not use any special procedure to remove the ballistocardiogram artefact, as it did not interfere significantly with the identification of GSW discharges. Filtered EEGs were visually reviewed and GSW timing and duration were marked by an electroencephalographer familiar with EEG recording inside the MRI scanner (Fig. 1). Knowledge of the morphology of the patient’s EEG discharges recorded outside the scanner was sometimes used to help interpret the distorted patterns seen inside the scanner. In each individual, all GSW bursts recorded during the scanning periods were used in the analysis.

Fig. 1 Example of GSW burst recorded in the MRI scanner. (A) While no scanning is taking place, (B) during MRI scanning (each 2.5 s section corresponds to one frame of 25 slices), (C) EEG of (B) after removal of the scanner artefact (patient 1 with childhood absence epilepsy).

fMRI acquisition and analysis

MRIs were obtained with a 1.5 T MRI scanner (Vision, Siemens, Erlangen, Germany). At the start of each study, an anatomical MRI was performed (acquisition time ∼15 min). This was a global T1‐weighted 256 × 256 sagittal image of the whole brain consisting of 160 slices, 1 mm thick, and slice gap of 0.2 mm. The echo time (TE) was 9.2 ms, the relaxation time (TR) was 22 ms and the flip angle was 30°. For fMRI data acquisition, each frame consisted of 25 BOLD EPI 64 × 64 axial slices acquired sequentially from top to bottom, covering the entire brain with a voxel size of 5 × 5 × 5 mm. The TE was 50 ms and the flip angle was 90°. The inter‐slice time interval within each frame was 100 ms and the total inter‐frame interval was 3 s. Each run consisted of 120 frames, lasted 6 min, and the inter‐run gap was 2 min. Study duration varied between 90 and 120 min, including the anatomical scan. fMRI images were motion corrected and smoothed (Gaussian kernel; full width at half maximum: 6 mm) using in‐house software.

The subject’s head, as well as most of the electrode wires near the head, were immobilized with a plastic bag filled with very small polystyrene spheres, in which a vacuum was obtained by air suction (S&S X‐Ray products, Brooklyn, NY). The bag was wrapped around the head to cover the back, the sides and part of the top of the head, as well as the back of the neck. This proved extremely effective in preventing head movement and also rendered the procedure more comfortable, removing the strain of lying directly on electrodes. Subjects stayed in the magnet for ∼90 min and seldom complained of discomfort.

Statistical analysis of fMRI data

Statistical processing of the images was performed using the methods and software of Worsley et al. (1996, 2002) in order to find the areas that were changed in response to the GSW discharges. The basis of the analysis is the comparison between the actual fMRI signal and a model of the expected response given the knowledge of GSW timing. The analysis program took into account the fact that the different slices are not all measured at the same time with respect to the GSW event because it takes 2.5 s to scan the whole head. This correction is possible because the time between slices is known. The onset of GSW was considered as time zero and the regions of responses were found by statistical comparison with a model including five possible haemodynamic response functions (HRFs): a standard HRF, modelled using two gamma functions with the initial positive peaking at 5.4 s and the negative undershoot at 10.8 s (Glover, 1999), and four single gamma functions peaking at 3, 5, 7 and 9 s (Fig 2). The purpose of this approach was to see if the fMRI data contained significant deviations from baseline that were time‐locked to the EEG activity but occurred with a different time course from that assumed in the original analysis using a standard HRF. Such an approach has been used (Buckner et al., 1998) by delaying the onset of a single gamma function. One could also have used HRF models with a shape similar to the ‘Glover’ HRF but different timings of the first peak. The question would then arise as to the width of the positive peak and the relative timing of the second peak. Simple gamma functions were selected to avoid having to deal with too many variables and because we feel they capture the main part of the BOLD response.

Fig. 2 The standard haemodynamic response functions (HRFs) and four single gamma functions peaking at 3, 5, 7 and 9 s. The standard HRF is shown with higher magnitude to avoid confusion, but fMRI analysis is independent of amplitude.

Statistical maps were obtained, indicating at each voxel the level of correlation between the BOLD signal and the model (t stat maps). The first three frames of each run were excluded from the analysis to let the system reach a steady state. Model and signals were pre‐whitened with an AR model of order 1. Slow fluctuations in the fMRI signal were taken into account by including in the linear regression a third order polynomial fitted to each run. For each of the five t stat maps created using the individual HRFs, statistical significance was defined based on the size of a cluster of voxels above a given threshold. A cluster of five contiguous voxels above a t value of 3 corresponds to a P value of 0.01.

The level of significance of the composite analysis is therefore ∼0.05 as a result of the fact that each data set is being analysed five times, once for each of the five HRFs. Among the results from the different models (Glover, peaks 3, 5, 7 and 9 s), the one yielding the highest t stat value was selected. Since the main interest was with the thalamo‐cortical system, responses in the brainstem, basal ganglia and cerebellum were ignored. Responses in the ventricles, cisterns, bone and scalp were also ignored.

Responses with negative t stat scores were labelled deactivation and those with positive t stat scores were labelled activation. We defined the volume of an activation or deactivation region as the number of contiguous voxels above threshold multiplied by the voxel size.


Nine patients were excluded due to lack of GSW during the study and one due to technical problems. Fifteen patients who had GSW activity during the fMRI study were evaluated (Table 1). The mean age of the patients at time of study was 35 years (range 18–66), and the age at seizure onset was 11 years (range 3–24). EEG disclosed spike and wave (five patients) or polyspike and wave (10 patients) activity at a frequency of 2–4 Hz always with anterior head preponderance. The clinical syndromes according to the International Classification (Commission on Classification and Terminology of the International League Against Epilepsy, 1989) were: juvenile absence epilepsy in six patients, childhood absence epilepsy in three, generalized tonic–clonic seizures alone in three, juvenile myoclonic epilepsy in one, childhood absence epilepsy with facial myoclonus in one, and juvenile absence epilepsy with eyelid myoclonus in one. All but one were receiving antiepileptic medications. All patients except five had some residual seizures. The anatomical MRI was normal in all but three patients; non‐specific periventricular white matter changes were present in one, a pineal gland cyst in one, and mild ventricular dilatation in one.

View this table:
Table 1

Demographic data of 15 IGE patients with GSW activity during the scanning period

PatientAge/age at onset (years)/sexClinical diagnosis*EEG GSW (Hz) Medication at time of testSeizure control
118/8/FCAE3 SW ETHYes
327/3/FCAE+3–4 PSW VAP, ETHNo
446/3/FCAE with FM3 PSW VAP, CBZNo
1124/12/FJAE with EM2.5–3 SWVAP, TPM, CLONo
1226/14/FJME3–4 SWCBZYes
1342/10/FGTCS2.5 PSW None Yes
1436/24/FGTCS 3 PSW CBZYes
1537/4/FGTCS 3–4 SW TPM, PTHNo

*According to Classification of International League Against Epilepsy: CAE = childhood absence epilepsy; CAE with EM = childhood epilepsy with eyelid myoclonus; CAE with FM = childhood absence epilepsy with facial myoclonus; JAE = juvenile absence epilepsy; JME = juvenile myoclonic epilepsy; GTCS = generalized tonic–clonic seizure only. +Uncertain clinical diagnosis. SW = spike and wave; PSW = polyspike and wave. CBZ = carbamazepine; CLOB = clobazam; CLO = clonazepam; ETH = ethosuximide; LTG = lamotrigine; PTH = phenytoin; TPM = topiramate; VAP = sodium valproate.

Fourteen patients (93%) showed significant fMRI responses. They had a mean rate of GSW discharges of 46.4/h of scanning (range 15.7–153), and a mean event duration of 2.5 s (range 1–14). The remaining patient with no fMRI response had 10 GSW events per hour, lasting 1–3 s.

Results regarding cortical responses are presented first, followed by those regarding thalamic responses, and finally those regarding the timing of responses.

Cortical responses

Cortical responses were largely bilaterally symmetrical; they usually had the same extent and distribution in the two hemispheres whether they involved anterior or posterior head regions, or both, or were limited to smaller areas (Table 2, Figs 3 and 4). With respect to spatial extent, responses encompassed diffusely most of the hemispheres in three patients (example in Fig. 3), and involved multiple separate regions in the remaining 11 (Table 3, example in Fig. 4). Whether responses were diffuse or spatially restricted, we examined the antero‐posterior distribution: five patients showed responses almost equally distributed between anterior and posterior head regions, six showed anterior head predominance, and three posterior predominance.

Fig. 3 Patient 2 with childhood absence epilepsy. (A) GSW burst recorded inside the MRI scanner. (B) fMRI results superimposed on anatomical MRI showing diffuse cortical fMRI activation, as well as thalamic activation. The cortical activation involves widespread cortical areas, with maximum t stat scores in superior and mesial frontal regions with the HRF peaking at 5.4 s (Glover). The regions may appear disconnected, but are largely part of a single volume of activation. The right hemisphere is shown on the right.

Fig. 4 Patient 7 with juvenile absence epilepsy. (A) GSW burst recorded inside the MRI scanner. (B) Example of cortical deactivation (note scale) and absence of thalamic response when using the HRF peaking at 7 s. The deactivation is present in frontal and occipital regions, with predominance in the latter.

View this table:
Table 2

Cortical fMRI responses in 15 IGE patients and GSW

PatientEvent rate/hEvent duration (s)Distribution of cortical activationsMax t statTotal volume (cm3)HRFs model
1432–8Diffuse*, anterior > posterior11.8366.9Glover
–11.0274.5Peak 9
21531–3Diffuse, anterior > posterior14.3530Glover
–6.120.5Peak 9
3341Multiregional+, anterior > posterior7.230.1Peak 5
–5.569.1Peak 9
4151–14Multiregional, anterior =  posterior5.735.6Glover
5161–3Multiregional, anterior > posterior7.043.4Peak 7
–5.214.6Peak 3
6561–6Diffuse, anterior =  posterior11.136.1Peak 7
–17.6522.9Peak 7
7391–11Multiregional, posterior > anterior7.136.1Glover
–8.1215.1Peak 7
8163–7Multiregional, posterior > anterior6.327.1Glover
–8.3110.9Peak 7
9181–2Multiregional, anterior > posterior5.43.6Glover
11361–3Multiregional, anterior > posterior6.8193.4Glover
–5.016.4Peak 9
12691–2Multiregional, posterior  anteriorNoneNone
–9.357.25Peak 5
13751–2Multiregional, anterior = posterior8.820.1Glover
–6.832.1Peak 5
14322–3Multiregional, anterior = posterior6.01.4Glover
–10.1133.7Peak 7
15471–4Multiregional, anterior = posterior7.521.2Glover
–11.068.7Peak 5

*Diffuse = bilateral, symmetrical and widespread cortical responses, and almost connected; +multiregional = cortical responses involved multiple separated regions usually symmetrically.

View this table:
Table 3

Spatial distribution of BOLD responses in 11 patients with multiple regional fMRI responses


Regions are given independent of side because responses were symmetrical. Numbers are number of patients.

We found both activation and deactivation in the cortex. Responses consisted predominantly of deactivations in nine patients (mean volume of deactivation region 135.0 versus 16.2 cm3 for activation), and predominantly of activations in five (mean volume of activation 233.9 versus 65.5 cm3 for deactivation). The maximum t stat scores of cortical activations and deactivations were in a similar range; they varied from 3.5 to 14.3 for activations and from –3.4 to –17.6 for deactivations. The activation usually corresponded to an earlier HRF than the deactivation (see below ‘Timing of responses’) and the peaks, positive and negative, were in different regions (Fig. 5). The fMRI responses were similar among the clinical subtypes of IGE. No difference in responses was seen between spike and wave and polyspike and wave activity.

Fig. 5 Patient 1 with childhood absence epilepsy. This patient shows both activation (A) and deactivation (B). Note that, although there is some overlap, the local positive and negative peaks do not overlap. The activation corresponds to an HRF peaking at 5.4 s (Glover) and the deactivation to an HRF peaking at 9 s.

Thalamic responses

Twelve patients (80%) showed thalamic responses. Eight had bilateral activation (Table 4, example in Fig. 6), two had deactivation (unilateral in one), and two had both activation and deactivation bilaterally but in different regions. We visually analysed the spatial distribution within the thalamus, although the MRI resolution only allowed a general topographic differentiation. We divided thalamus into anterior, posterior, medial, lateral, ventral and dorsal regions. Responses involved the following regions: the whole thalamus in two patients, anterior regions in six, medial in five, dorsal in two, posterolateral in one, and ventral in one. The maximum t stat was 4.3–10.5 with mean volume of 11.8 cm3 for activations, and –3.2 to –5.6 with a volume of 4.2 cm3 for deactivations.

Fig. 6 Patient 13 with history of generalized tonic–clonic seizure only. (A) Short burst of spike and wave. The pattern is distorted by the recording inside the scanner, but is interpreted in the context of the patient’s recordings outside the scanner. (B) Clear activation in the thalamus with the HRF peaking at 5.4 s. The patient also had bilateral posteriorly predominant deactivation (not shown).

View this table:
Table 4

Thalamic response data in 15 IGE patients and GSW

PatientDistribution of thalamic activationMax t‐statTotal volume (cm3)HRFs model
1Bil medial 8.610.9Glover
2Bil whole10.523.1Glover
3Bil medial 5.412.4Glover
4Bil anterior6.510.1Glover
5Bil anterior5.93.0Peak 7
Bil medial –5.32.5Peak 3
6Bil ventral –5.68.1Peak 5
8Bil whole6.538.9Peak 9
11Bil posterolateral and R anteromedial4.31.9Glover
12Bil dorsal7.44.9Glover
Bil anterior–4.35.2Peak 9
13Bil dorsomedial8.811.1Glover
14Bil anterior4.41.1Peak 3
15R anterior–4.31.0Peak 9

Bil =  bilateral; R = right.

Timing of responses

Data were analysed with the standard model peaking at 5.4 s and with HRFs peaking at 3, 5, 7 and 9 s. We retained only the responses corresponding to the highest t stat score. We summarized, in Fig. 7, which HRF yielded this highest score. Most activations were obtained with the standard model, whereas deactivations were more evenly spread among the different HRFs. It may be noted that a relatively high proportion of late responses were deactivations, underlying the importance of an analysis with HRFs at multiple latencies to discover deactivations.

Fig. 7 Bar graph of timing of cortical activation and deactivation. Whereas activation was relatively well detected with the ‘standard’ Glover response, deactivation more probably corresponded to HRFs with a later peak.

The highest positive thalamic t stat was seen with Glover’s model in six patients, peak 3 in one, and peak 9 in one. The two patients with negative thalamic activation had the highest t stat at peak 9 in one, and at peak 5 in one, and neither showed deactivation using the Glover model. In the remaining two patients, who showed both thalamic activations and deactivations, the highest activation (peak 7) followed the early deactivation (peak 3) in one, and the highest deactivation (peak 9) followed the highest positive activation with Glover’s model in the other. The regions of maximum activation and of maximum deactivation did not overlap.

Duration of EEG bursts

We compared the responses to short and long bursts of GSW to see if they yielded different regions or type of response. We divided bursts into those lasting <3 s and those lasting >3 s (3–14 s). In four patients, we had at least five bursts in each category. In three of these four patients, the patterns of response were the same with short and long bursts. t stat values and volumes varied, but the sign of the response, activation or deactivation, the main regions involved and the general extent were the same. In the fourth patient, there was no significant response to short bursts, whereas the response to long bursts was diffuse.


The EEG–fMRI method ensures that the measured responses are directly linked to the EEG pattern. This is in contrast to most other functional methods, which do not have specific time relationships with epileptic discharges and therefore include the inter‐burst data as much as the bursts. We found fMRI responses in 93% of patients with GSW during the scanning period. This is in contrast to ∼50% in patients having focal epileptic discharges (Warach et al., 1996; Krakow et al., 1999; Lazeyras et al., 2000; Jager et al., 2002). This may be due, in part, to the duration of the discharges, typically longer in GSW than in focal discharges, since our earlier study showed that focal discharge duration was positively correlated with likelihood of activation (Al‐Asmi et al., 2003). Regions and types of fMRI responses were, however, diverse, and we will review the main aspects of our results. It may be noted that our patients were relatively old and that mechanisms could be different in children with the prototypical bursts of spike and wave.

A prominent feature of the regions of fMRI responses was their symmetrical distribution over the cerebral hemispheres. Whether responses were predominant anteriorly or posteriorly, or were diffuse, they were remarkably symmetrical in their right–left extent. This serves to reassure us that the responses are directly linked to the GSW discharges and is illustrated in all the cases shown in the figures. It reflects the bilateral nature of the EEG discharge, and the critical role played by callosal fibres in bilateral synchrony (Erickson, 1940; Musgrave and Gloor, 1980; Gotman, 1981; Oguni et al., 1994).

Cortical involvement was generally widespread and included all cortical areas when looking across all patients. There was a tendency for frontal and parietal regions (lateral and mesial) to be involved slightly more often, but central, occipital and temporal regions were also frequently involved, sometimes with the highest statistical scores. In the patient reported by Salek‐Haddadi et al. (2003) (their Fig. 2), statistical values were generally higher in posterior head regions. Bursts of GSW generally predominate, from the EEG point of view, in frontal and central regions and involve much less the parietal, occipital and temporal regions (Rodin and Ancheta, 1987). There is, therefore, a difference between the spatial distribution as seen in the EEG and in the fMRI response. The EEG signal results from synchronous postsynaptic potentials in cortical pyramidal cells (Creutzfeldt and Houchin, 1974). It is possible to obtain a large‐amplitude EEG even if a relatively small proportion of cells are involved, as long as they are well synchronized (Nunez, 1981). It may be that the amplitude predominance of the EEG in anterior head regions originates from a neuronal population that is more highly synchronized than in posterior head regions, in which cells could be just as actively involved, but less synchronized. Our results, therefore, point to a relatively equal involvement of all brain regions in GSW rather than the traditional fronto‐central predominance. This widespread involvement is compatible with the generalized but mild nature of the behavioural impairment during GSW and absence seizures.

Another important aspect of our results is the involvement of the thalamus in 12 of 15 patients (80%). This involvement again was most often bilateral and symmetrical. The responses, after the spatial smoothing required by our analysis, appeared to involve primarily the anterior and mesial aspects of the thalamus, but the spatial resolution is not sufficient to allow discussion of particular thalamic nuclei. The only PET study of CBF regarding thalamic involvement (Prevett et al., 1995) had raised the possibility that CBF changes could have been caused by hyperventilation. In our EEG–fMRI study, GSW occur spontaneously and this possibility can be eliminated. Our results establish unambiguously for the first time that experimental models that actively involve the thalamus are appropriate models, in this respect, of the human condition.

An unexpected finding of this study is the variability in the type of BOLD response, activation or deactivation, in thalamus and cortex. Overall, we found that thalamic activation was much more frequent than deactivation (eight versus two patients), but we also found two patients having both thalamic activation and deactivation in adjacent regions. The cortical response was more often predominantly a deactivation (nine patients) than activation (five patients). Here again, however, we found 13 of 14 patients with both cortical activation and deactivation. As in the thalamus, these occurred in different regions. In the patient reported by Salek‐Haddadi et al. (2003), there was an increased BOLD signal (activation) in the thalamus and a decreased BOLD signal (deactivation) in the cortex.

Synaptic activity, excitatory or inhibitory, as well as neuronal firing contribute to neuronal oxidative metabolism (Attwell and Laughlin, 2001). Increased synaptic or neuronal activity results in increased deoxyhaemoglobin in blood, and an increase in CBF, which results in a decrease in the concentration of deoxyhaemoglobin because the increase in CBF is large. This causes an increased BOLD signal (Ogawa et al., 1992). In studies combining the measurement of neuronal firing, synaptic activity and BOLD, Logothetis et al. (2001) concluded that synaptic activity is the major contributor to the BOLD signal. During an epileptic discharge, one would expect increased synaptic activity, whether excitatory or inhibitory, hence a BOLD activation. This is what is seen most often in focal spikes. How can we explain the frequent BOLD deactivation that we observe?

We consider four possible explanations. In the first, blood flow can be diffusely reduced during GSW (Klingelhofer et al., 1991; Sperling and Skolnick, 1995; Diehl et al., 1998). For instance, using the 133xenon method to assess CBF, Sperling and Skolnick (1995) suggested that a decrease in CBF during GSW reflects a decrease in cortical metabolic demand and neuronal activity. This hypothesis cannot explain the fact that we found both activation and deactivation in the majority of our patients. In the second explanation, we assume that there is a relative CBF reduction in the deactivated areas caused by a steal phenomenon secondary to the increased CBF in activated regions. The steal phenomenon, however, is unlikely because it does not explain the exclusive (one patient) or prominent deactivations seen in eight patients. A third explanation assumes abnormal coupling between neuronal activity and regional CBF. If increased neuronal activity is not accompanied by the usual increase in blood flow, a decrease in the BOLD signal will be observed. With this hypothesis, we assume that epileptic discharges are accompanied by increased synaptic activity, but some regions have an abnormal neuronal–CBF coupling. However, there is no evidence of abnormal neuro‐vascular coupling in this type of patient without obvious pathological abnormalities. A fourth explanation is that the regions of BOLD deactivation correspond to decreased synaptic activity, such as that caused by reduced neuronal input (Logothetis, 2003). In this context, we interpret our results as indicating that not all GSW are generated by the same mechanisms and that some brain regions become more active (receive more synaptic input) whereas others become less active. It is important to remember that the ‘increase’ and ‘decrease’ in fMRI we have discussed are always relative to the inter‐burst baseline.

In experimental models of GSW, the general level of firing of cortical neurons during a burst is highly variable from model to model. It has, for instance, been suggested that there is a general reduction of synaptic input (disfacilitation) in the cortex during GSW in the Genetic Absence Epilepsy Rats from Strasbourg (GAERS) model (Charpier et al., 1999). In some models (Lytton et al., 1997), it has also been shown that the majority of thalamo‐cortical cells are silent during GSW, whereas in other models active firing of thalamo‐cortical cells was demonstrated (Steriade and Contreras, 1995). It is remarkable that results very similar to ours were found in an fMRI study of an experimental model of absence seizure (Tenney et al., 2003): activation in the thalamus and both activations and deactivations in the cortex. If we consider that activation predominates in the thalamus and deactivation predominates in the cortex, the following mechanism could be at play, although it is clearly not the complete explanation of our results: a widespread hyperpolarization of most thalamic cells takes place during GSW bursts. This has been observed in the GAERS model (Pinault et al., 1998; Charpier et al., 1999). It could explain the increased BOLD response in the thalamus and result in a relative deafferentation of the cortex, thus explaining the predominance of cortical BOLD deactivations. On the background of the thalamic hyperpolarization, there are also rhythmic discharges, which explain the spike–wave bursts.

Finally, we review the issue of the timing of the BOLD responses. We have performed our analysis with model haemodynamic responses peaking at 3, 5, 7 and 9 s, as well as with a relatively standard biphasic model, which peaks at 5.4 s and has an undershoot at 10.8 s (Glover, 1999). Our results indicate that positive responses (activations) had a peak most often at 5–7 s, although there was inter‐individual variability. The standard analysis with the standard (Glover) model would therefore reveal most of these activations. Negative responses (deactivations), however, frequently peaked later, with eight patients showing the peak deactivation with the 9 s or with the 7 s model. Such deactivations would probably be missed with the standard model. It is worth noting that regions of positive and negative responses did not overlap spatially: the late negative responses could, therefore, not be the undershoot of an earlier positive response. We do not have an explanation for the later occurrence of the negative responses. It would be of interest to determine the actual shape of the HRF in the regions with statistically significant responses. When we did this in focal epileptic spikes (Bénar et al., 2002), we found that there was an important variability in response, even from region to region in the same patient. In future analyses, this variability could be used to enhance the analysis (Kang et al., 2003).

In conclusion, this extensive study of fMRI activation resulting from GSW bursts has demonstrated for the first time in human, without ambiguity, that the thalamus is involved in these discharges. Unexpectedly, posterior and temporal head regions appear almost as frequently involved as anterior head regions, unlike results revealed by the usual EEG observations. Despite the apparent uniformity of the GSW pattern, it appears that several possible mechanisms may be at play in its generation, all involving thalamo‐cortical circuits, but with some regions activated and others deactivated. The frequent cortical deactivations may reflect a relative deafferentation of the cortex during GSW compared with the state between bursts. We hypothesize that this deafferentation is mediated by a widespread hyperpolarization of the thalamus, which has been observed in experimental models and which could also explain the frequent thalamic BOLD activation. Such a deafferentation could be the cause of the behavioural arrest or incomplete perception observed during absences.


We wish to thank Drs E. Andermann, M. Veilleux and B. Zifkin for referring patients for this study, and Dr B. Pike for discussion of the results. This work was supported by grant MOP 38079 of the Canadian Institutes of Health Research.


View Abstract