Address correspondence to Friederike Moeller, Clinic for Child Neurology, University of Kiel, Schwanenweg 20, 24105 Kiel, Germany. E-mail: firstname.lastname@example.org
Purpose: In patients with idiopathic generalized epilepsy (IGE), blood oxygen level dependent (BOLD) EEG during functional MRI (EEG-fMRI) has been successfully used to link changes in regional neuronal activity to the occurrence of generalized spike-and-wave (GSW) discharges. Most EEG-fMRI studies have been performed on adult patients with long-standing epilepsy who were on antiepileptic medication. Here, we applied EEG-fMRI to investigate BOLD signal changes during absence seizures in children with newly diagnosed childhood absence epilepsy (CAE).
Methods: Ten drug-naive children with newly diagnosed CAE underwent simultaneous EEG-fMRI. BOLD signal changes associated with ictal EEG activity (i.e., periods of three per second GSW) were analyzed in predefined regions-of-interests (ROIs), including the thalamus, the precuneus, and caudate nucleus.
Results: In 6 out of 10 children, EEG recordings showed periods of three per second GSW during fMRI. Three per second GSW were associated with regional BOLD signal decreases in parietal areas, precuneus, and caudate nucleus along with a bilateral increase in the BOLD signal in the medial thalamus. Taking into account the normal delay in the hemodynamic response, temporal analysis showed that the onset of BOLD signal changes coincided with the onset of GSW.
Discussion: In drug-naive individuals with CAE, ictal three per second GSW are associated with BOLD signal changes in the same striato-thalamo-cortical network that changes its regional activity during primary and secondary generalized paroxysms in treated adults. No BOLD signal changes in the striato-thalamo-cortical network preceded the onset of three per second GSW in unmediated children with CAE.
Taken together, these studies suggest that the primary and secondary GSW, whether polyspike-wave discharges or three per second GSW, modify the regional BOLD signal (as an index of regional neuronal activity) in the same striato-thalamo-cortical network, even though the temporal pattern of activation for different paroxysms may differ. We have recently shown that BOLD signal changes in the thalamocortical network precede generalized polyspike-wave discharges by several seconds (Moeller et al., 2008). It is unclear whether three per second GSW paroxysms are also preceded by BOLD signal changes in the thalamocortical network.
At present, case studies have only been published on changes in the striato-thalamo-cortical network associated with absence seizures (Salek-Haddadi et al., 2003; Labate et al., 2005; Laufs et al., 2006). Except for one case report (Labate et al., 2005), fMRI of GSW-related activity was performed in adult patients who had a long-standing history of epilepsy and who had been on antiepileptic mediactions for several years. It has been argued that concurrent medication and the chronic presence of epilepsy might have influenced the results, but this issue remains to be addressed.
In this study, we investigated a homogeneous group of newly diagnosed and untreated children with CAE to assess the consistency and temporal pattern of network modulation associated with ictal, three per second GSW discharges which cause absence seizures.
Subjects and Methods
From May 2005 to November 2007, 10 patients with newly diagnosed, untreated childhood absence epilepsy (CAE) were recruited from the Department of Neuropediatrics at the University Hospital Schleswig-Holstein, Campus Kiel and the Pediatric Department of the Kiel General Hospital. Their mean age was 8 years (age range: 4–12 years) at time of the study. Diagnosis was made according to the ILAE 2001 classification scheme (Commission on Classification and Terminology of the International League Against Epilepsy, 2001). All patients had normal neurodevelopmental milestones and structural MRI. The study was performed according to the Declaration of Helsinki and approved by the local Ethics Committee of the University of Kiel. The parents gave written informed consent prior to the experiment. After participating in the study, all patients were treated with valproate and became seizure free with medication.
The EEG was continuously recorded during fMRI from 30 scalp sites (10–20 system plus FC1, FC2, CP1, CP2, FC5, FC6, CP5, CP6, TP9, TP10) with a reference located between Fz and Cz. Sintered Ag/AgCl ring electrodes with built-in 5 kOhm resistors were attached using the “EasyCap” (Falk-Minow Services, Herrsching-Breitbrunn, Germany), which is part of the MR-compatible EEG recording system “BrainAmp-MR” (Brainproducts Co., Munich, Germany). Electrode impedance was kept below 10 kOhms. Two additional electrodes were placed on the infraorbital ridge of the left eye for recordings of the vertical electrooculogram (EOG) and on the left perivertebral part of the lower back for acquisition of the electrocardiogram (ECG) to control for heartbeat artifacts. Data were transmitted from the high-input impedance amplifier (250 Hz low-pass filter, 10 s time constant, 16-bit resolution, dynamic range 16.38 mV) which was placed directly behind the head coil inside the scanner room and connected to a computer located outside the scanner room via an optic fiber cable. The scanner (10-MHz sampling rate) was synchronized with the EEG amplifier (5-kHz sampling rate). Online correction of gradient artifacts based on the averaged artifact subtraction (AAS) algorithm was performed using RecView software (Brainproducts Co.). This enabled visual inspection of GSW throughout the entire recording time. Foam pads were used to help secure the EEG leads, minimize motion, and improve patient comfort.
BOLD-sensitive MRI was performed with a 3-Tesla MR scanner (Philips Achieva, Philips, Best, The Netherlands) and a standard, 8-channel SENSE head coil. A single-shot T2*-weighted, gradient-echo planar imaging (EPI) sequence was used for fMRI (TR = 2250 ms, TE = 45 ms, 30 slices, 64 × 64 matrix, slice thickness = 3.5 mm, field of view (FOV) = 200 mm, flip angle = 90°). Five hundred forty brain volumes were acquired during the 20-min fMRI session. The first five images were discarded to ensure steady-state longitudinal magnetization. fMRI measurements were complemented by a structural MRI using a T1-weighted, three-dimensional MPR sequence (1-mm slice thickness, 208 × 208 matrix, 150 contiguous slices, FOV = 208 mm, TE = 3.6 ms, TR = 7.8 ms, flip angle = 8°, number of signals averaged (NSA) = 2).
During the fMRI examination, the parents of the patients stayed in the MR scanner suite and continuously monitored their children for any clinical signs of absences such as staring, eye deviation, and eyelid myoclonia. Parents were asked to indicate the onset and duration of each absence seizure by raising their hand. This was noted by one of the examiners (FM, MS) who were located at the MR console outside the MR suite to monitor the online EEG recordings.
EEG data processing
EEG recordings were processed offline using the BrainVision Analyser 1.05 software (Brainproducts Co.). Gradient artifacts as a result of electromagnetic distortion of the EEG due to static and dynamic magnetic field during MR data acquisition were removed using the AAS method described by Allen et al. (2000). Data were corrected relative to the onset of MR-volume gradient artifacts, which was indicated by a trigger received from the MR system and recorded with the EEG. A moving average width of 10 MR volumes (TRs) was used for gradient correction. Corrected EEG data were filtered using a high-pass filter at 0.03 Hz and a low-pass filter at 75 Hz. Data were then down-sampled to 250 Hz. Ballistocardiographic artifacts were corrected using the same AAS method (Allen et al., 1998). In cases of prominent residual artifacts after AAS correction, the independent component analysis (ICA)-based procedures were applied as described by Srivastava et al. (2005). GSW were independently marked by two experienced neurophysiologists (FM and MS). Consensus was achieved by comparing and discussing the results of these independently identified GSW.
MRI data analysis
Individual MRI datasets were analyzed using Statistical Parametric Mapping (SPM5) software (Wellcome Department of Imaging Neurosciences, University College London, U.K., http://www.fil.ion.ucl.ac.uk/spm). All volumes were realigned to the first volume and spatially normalized to the Montreal Neurological Institute (MNI) template supplied by SPM). Then, images were smoothed using an isotropic Gaussian kernel of 6 mM and high-pass filtered at a cutoff of 128 s. The preprocessed fMRI time series were statistically analyzed at an individual level using simple regression analysis in which each GSW of an absence was treated as a single event. Each event (i.e., GSW as revealed by online EEG) was used to produce a stick function that was convolved with a canonical hemodynamic response function (peak at 6 s relative to onset, delay of undershoot = 16 s, length of kernel = 32 s) as implemented in SPM5 (Friston et al., 1995). In each individual, one-tailed t-tests were applied to test for regional increases or decreases in the BOLD signal that were temporally linked to the three per second GSW periods. At the voxel level, the significance level was set at p < 0.05 after correction for multiple comparisons across the whole brain using the family-wise correction method as implemented in SPM5 (Friston et al., 1995). This corresponded to t-values above 4.7. An extent threshold of five contiguous voxels was also applied. Individual statistical parametric t-maps were color-coded and superimposed onto the individual coregistered T1-weighted images. Relative increases and decreases were visualized using the Anatomist software (http://brainvisa.info). Voxels coded in blue-green indicate significant decreases in the BOLD signal while voxels coded in yellow-red correspond to regional increases in the BOLD signal.
Using the parameter estimates obtained by single-subject analyses, we performed a one-sample t-test to test for consistent patterns of BOLD signal changes across patients. Based on previous EEG-FMRI studies on GSW- related BOLD signal changes (Archer et al., 2003; Salek-Haddadi et al., 2003; Aghakhani et al., 2004; Gotman et al., 2005; Labate et al., 2005; Hamandi et al., 2006; Laufs et al., 2006; Moeller et al., 2008), we performed a region-of-interest (ROI) analysis. Spherical ROIs with a radius of 10 mM were defined in the thalamus, precuneus, left and right parietal areas, and the left and right head of the caudate nucleus. The ROIs were centered according to coordinates derived from group analysis done in adult patients with GSW (Hamandi et al., 2006). The statistical images were thresholded at p < 0.05 after correction for the voxels within each ROI (false discovery rate [FDR] corrected within each search volume).
To investigate whether BOLD signal changes occur prior to the absence seizure, we visualized the averaged time course of the BOLD signal relative to the onset of the absence seizure. In light of our a priori hypothesis based on previous EEG-fMRI studies of BOLD signal changes during GSW (Archer et al., 2003; Salek-Haddadi et al., 2003; Aghakhani et al., 2004; Gotman et al., 2005; Labate et al., 2005; Hamandi et al., 2006; Laufs et al., 2006) and polyspike and wave discharges (Moeller et al., 2008), the thalamus, the precuneus, and the head of the caudate nucleus were defined as ROIs. The ROIs with a radius of 10 mM were centered on coordinates given in Table 3, taken from Hamandi et al., 2006. The eigenvariates of BOLD signal intensities of all 540 scans of the predefined ROIs were extracted in each subject. For all absence seizures, the BOLD signal of the 8 scans before and the 15 scans after the onset of the seizure were aligned to the seizure onset. The scan during which the GSW started was defined as scan zero. The time bins were predefined by the sampling rate (TR = 2.25 s). For each subject, signal intensities of scans before and after the onset of GSW were z-transformed to normalize differences in the absolute signal intensity across patients (Cheadle et al., 2003). The z-transformation is an alternative strategy to normalize different baselines of BOLD signals across the patients and can also be used among other methods such as, for example, the calculation of the percent of signal change. In a second step, the z-transformed BOLD signal values of all scans before and after the GSW were averaged across all patients. The mean BOLD signal changes and standard error were plotted for the medial thalamus, the right and left caudate nucleus and the medial precuneus.
Online EEG revealed three per second GSW discharges that were associated with the clinical signs of absence seizures during the fMRI session in 6 out of 10 patients (see, for example, Fig. 1). In these patients, the three per second GSW paroxysms recorded with online EEG coincided with the clinical manifestation of absence as observed by the parents in the MR suite. The number of absences per fMRI session ranged from one to four absences with a duration ranging from 4 to 30 s. In two patients, the examination had to be terminated due to claustrophobia they experienced inside the MRI scanner. A third patient had to be excluded because of pronounced movement artifacts, and one patient had no absences during fMRI scanning.
Fig. 2 denotes the individual BOLD response patterns of all six patients who had absences during fMRI. All three per second GSW paroxysms led to regional changes in the BOLD signal. All patients showed bilateral increases in the BOLD signal in the medial portion of the thalamus. Individual t-values varied from 5.57 to 9.01 and cluster extent ranged from 48 to 813 voxels. There was a highly significant correlation between mean duration of absences and the number of activated voxels (pearson correlation 0.927, p = 0.008) whereas no correlation was found between mean absence duration and peak t-values (pearson correlation 0.601, p = 0.207). Positive BOLD signal changes were also seen to a variable degree in the cerebellum and at some cortical sites without regional preference in some of the patients. When lowering the statistical threshold to p < 0.001 noncorrected, we observed scattered increases in BOLD signal of the cortex in three patients (patient 1, 2, and 5), but no focal increase in BOLD signal in the somatosensory cortex.
Thalamic increases in the BOLD signal were accompanied by concurrent decreases in the BOLD signal at the cortical level. Bilateral symmetrical decreases in the BOLD signal in the posterior parietal cortex, the precuneus, and frontal cortical areas were present in four of the six patients. In patient 2, there was also a bilateral BOLD signal decrease in the head of the caudate nucleus and the brainstem.
Fig. 3 and Table 1 represent the results of the group analysis. The statistical threshold was set p < 0.05 (FDR corrected within the search volume) for positive and negative BOLD signal changes. The group analysis showed a significant increase in the BOLD signal in the thalamus. A decrease in BOLD signal was found in the ROIs bilaterally in the caudate nucleus and the left parietal cortex. The deactivation in the predefined region of interest in the right parietal cortex and the precuneus were not significant, however nearby deactivations were.
Table 1. Results of region-of-interest analyses (second level group analysis)
Z-scores are reported for activation (thalamus) and deactivation (caudate nucleus, parietal cortex, precuneus) within spherical ROIs (10-mm radius) centered at coordinates taken from (Hamandi et al., 2006); p-values are FDR corrected with small volume correction. Note that although the ROIs centered in the precuneus and the right parietal cortex did not reach significance, nearby areas did (indicated by *).
Using a ROI-based approach, the BOLD signal changes of each absence seizure were pooled together and realigned to the onset of the three per second GSW paroxysm. The mean time course of BOLD signal changes for ROIs in the thalamus, the precuneus, and the right and left head of the caudate nucleus are displayed in Fig. 4. When plotting the time course of the BOLD signal in the predefined ROI of the thalamus, the BOLD signal starts to increase at the onset of GSW. The BOLD signal increase was maximum approximately 6 s after the onset of GSW. Although the duration of absences varied considerably ranging from 4 to 30 s, standard errors of the mean BOLD signal remained constant during the rising and falling phase of thalamic BOLD signal changes during three per second GSW discharges. This suggests that the activation of the thalamus may only be transient at the beginning of the absences and then shows a decay irrespective of the length of the three per second GSW discharges. The BOLD signal change in the ROI of the precuneus decreased approximately 2 s after the onset of GSW and showed a maximum decrease 9 s after the GSW. In the ROI of the right and left head of the caudate nucleus, the BOLD signal decreased after the onset of GSW and showed a maximum of deactivation approximately 6–13 s after the onset of GSW. No BOLD signal changes preceded the onset of GSW in any ROI when taking into account the normal delay between the neuronal event and the hemodynamic response.
We found that drug-naive children with newly diagnosed CAE showed a consistent change in activation of the striato-thalamo-cortical during absence seizures. The presence of three per second GSW discharges was associated with BOLD signal increases in the thalamus and decreases in the parietal and frontal cortex, the precuneus, and the caudate nucleus. Our results extend previous EEG-fMRI studies on GSW-associated BOLD signal changes in adult patients with CAE (Salek-Haddadi et al., 2003; Aghakhani et al., 2004; Gotman et al., 2005; Hamandi et al., 2006; Laufs et al., 2006). Although most of the previously studied patients had a long history of epilepsy and were on antiepileptic treatment, the spatial pattern of BOLD signal changes during three per second GSW discharges was strikingly similar. Together, these studies reveal a common metabolic signature of three per second GSW discharges in the basal ganglia-thalamocortical circuitry of the human brain. Previous studies showed that age (Richter & Richter, 2003; Marcar et al., 2004) and antiepileptic medication (Bell et al., 2005) might influence the BOLD signal. However, the consistency of three per second GSW-associated BOLD signal changes across studies indicates that differences in age, medication, and disease duration do not have a substantial impact on the regional expression of this activity pattern suggesting that the BOLD signal changes in the basal ganglia-thalamocortical network are closely linked to the presence of three per second GSW discharges.
The thalamus was the only brain region that showed a consistent increase in BOLD signal during the three per second GSW discharges. This is concordant with previous studies on absence epilepsy which demonstrated thalamic activation during absence seizures in animal models (Blumenfeld, 2005) and human subjects (Salek-Haddadi et al., 2003; Labate et al., 2005; Laufs et al., 2006). These data corroborate the notion that the thalamus plays a crucial role in the pathophysiology of absence seizures, although its involvement in triggering and maintaining thalamocortical synchronization is still a matter of debate (Blumenfeld, 2005; Meeren et al., 2005). At the cortical level, parietal areas and the precuneus showed a decrease in the BOLD signal during the three per second GSW discharges. Similar decreases were reported in previous EEG-fMRI studies (Gotman et al., 2005; Hamandi et al., 2006; Laufs et al., 2006) and interpreted in the context of the “default mode network” (Raichle & Mintun, 2006). According to this hypothesis, the parietal cortex, the precuneus and frontal cortical areas are active at rest and support the state of consciousness in an awake subject (Raichle et al., 2001). The abnormal thalamocortical synchronization during three per second GSW disrupts the thalamo-cortical and cortico-cortical connectivity among cortical regions that constitute the “default mode network.” This reduces the level of cortical activity in these regions and may represent a hemodynamic correlate of impaired consciousness during absence seizures (Gotman et al., 2005; Hamandi et al., 2006; Laufs et al., 2006). However, the loss of consciousness is not always complete during absence seizures (Gloor, 1986) and may correlate with the duration of the absences (van Luijtelaar et al., 1991). Continuous neuropsychological assessment during EEG-fMRI would be necessary to detect any impairment of consciousness during GSW paroxysms and to relate the impairment in consciousness with regional changes in BOLD signal. Neither the present nor previous studies have directly assessed the impairment of consciousness during the absences investigated by EEG-fMRI. Therefore, this important issue remains to be addressed on future studies.
During three per second GSW paroxysms, the deactivation of cortical areas was less consistent relative to the activation of the thalamus, suggesting that the spatial pattern of cortical deactivation is more variable across subjects. When participants are studied during the so-called “resting state,” the magnitude and distribution of BOLD signal decreases may well depend on the relative “activation” of the “default mode” regions at the time of the three per second GSW discharge which may vary within and across subjects. Additionally, we used adult MRI templates to normalize children's EPI images which might have caused some distortions during normalization and thus reduced sensitivity to detect BOLD signal changes at the cortical level. Despite a more variable spatial pattern, a relative deactivation of cortical “default mode” areas was also observed in EEG-fMRI studies on medicated adults with longstanding absence epilepsy (Archer et al., 2003; Salek-Haddadi et al., 2003; Aghakhani et al., 2004; Gotman et al., 2005; Hamandi et al., 2006; Laufs et al., 2006). It should be noted that all studies assessed three per second GSW associated BOLD signal changes at rest. A different cortical deactivation pattern may emerge if three per second GSW occur while participants actively perform a specific task.
Although cortex and thalamus are involved in the generation of three per second GSW discharges, it is still unclear which structure initiates the cascade of processes underlying absence seizures (Blumenfeld, 2005). Animal studies in genetic models of absence epilepsy strongly suggest that GSW are triggered in a restricted region of the cerebral cortex (Meeren et al., 2002, 2005; Manning et al., 2004; Polack et al., 2007), more specifically in the facial region of the somatosensory cortex (Meeren et al., 2002, 2005; Polack et al., 2007). In this study, neither single-subject nor group anaylsis revealed a focal cortical increase in BOLD signal with three per second GSW discharges. The failure to detect a focal change in cortical activity might be attributed to variable locations of the cortical trigger zone. Animal studies also showed a hippocampal involvement in the regulation of absence seizures (Tolmacheva & van Luijtelaar, 2007; Velazquez et al., 2007). Again, the present study showed no consistent changes in BOLD signal in the hippocampal region. Alternative analytic procedures such as ICA (Rodionov et al., 2007) or direct measurements of neuronal currents with MRI (Truong & Song, 2006; Blagoev et al., 2007) might be more sensitive to detect a focal variation in BOLD signal within the cortex or hippocampus during human absence seizures.
In the basal ganglia, we found a bilateral BOLD signal decrease in the head of the caudate nuclei during three per second GSW paroxysms. This is in accordance with a study by Hamandi et al. that reported a deactivation of the caudate nucleus in a group analysis in patients with different subtypes of IGE (Hamandi et al., 2006). However, another group analysis on patients with IGE did not show any deactivation in the caudate nucleus (Gotman et al., 2005). Because the caudate nucleus receives strong cortical input from cortical associative areas, the decrease in the BOLD signal may reflect a reduced corticostriatal drive during three per second GSW discharges. In accordance with this hypothesis, animal studies in GAERS rats, a genetic model for absence epilepsy, showed that the discharges recorded in the caudate nucleus occurred a few seconds after the discharges had started in the thalamus or the cortex (Danober et al., 1998). This implies that the change in the caudate nucleus may be a consequence of changes in activity at the thalamocortical level and is an argument against the caudate nucleus having a critical role in the initiation of absences.
Recent studies on GAERS rats suggest that the basal ganglia act as a remote control system for absence seizures. During GSW, the cortico-subthalamo-pallidal network shows rhythmic bursting, whereas striatal output neurons are silenced (Slaght et al., 2004; Paz et al., 2005). It was suggested that the acute drop in firing rate of striatal neurons results from a feed-forward synaptic inhibition which may also contribute to seizure termination (Paz et al., 2005; 2007). The decrease in striatal neuronal activity during generalized paroxysms in GAERS rats ties in with the deactivation of the caudate nucleus in our patients.
We analyzed the time course of the BOLD signal with respect to the onset of the three per second GSW discharges. The BOLD signal changes in the striato-thalamo-cortical network coincided with the onset of the three per second GSW paroxysm when taking into account the temporal delay between the neuronal event and the hemodynamic response. In a recent study, we examined the transition from normal to pathological neuronal activity in patients with polyspike wave discharges (Moeller et al., 2008) and found that thalamic activation and cortical deactivation started several seconds before the onset of the polyspike wave (Moeller et al., 2008). Based on this finding, we argued that polyspike wave discharges result from a cascade of pathological activity in the striato-thalamo-cortical network that ultimately results in a polyspike wave. If so, this raises the question of why thalamocortical changes in the BOLD signal started with the onset of the three per second GSW. A possible explanation may be that the neuronal cascade that leads to three per second GSW is absence seizures is recruited faster than during the generation of polyspike waves.
When interpreting this discrepant finding, it is important to consider the differences between three per second GSW and polyspike wave. Polyspike wave consists of a brief series of irregular generalized spikes with frequencies of 12–16 Hz, whereas three per second GSW discharges are bilaterally synchronous, symmetrical, and regular with a burst frequency of 2,4-4 Hz. The different onset of BOLD signal changes with respect to the onset of polyspike wave or three per second GSW discharges provides further evidence that these two paroxysmal discharge patterns have a different pathophysiology, even though they may occur in the same patient or in the same form of IGE.
In conclusion, we show that in newly diagnosed, untreated children with CAE, absence seizures are associated with thalamic activation along with a deactivation of “default mode” areas as well as the caudate nucleus. This is consistent with EEG-fMRI results in pharmacologically treated adults with longstanding epilepsy. Unlike polyspike wave discharges which are preceded by thalamocortical BOLD signal changes, thalamic activation as well as cortical deactivation coincide with the onset of the three per second GSW discharge which confirms the differences in the underlying pathophysiology.
We thank the children and their parents for participating in our study and Dr. Klaus Westerbeck for referring patients. This work was supported by grants from the BMBF (Bundesministerium für Bildung und Forschung) to H. R. Siebner (01GO 0511) and M. Siniatchkin and an intramural grant from the Medical Faculty of the University of Kiel to F. Moeller.
Conflict of interest: 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. The authors have no professional or financial affiliations that might be perceived as having biased the presentation.