Mechanism of impaired consciousness in absence seizures: a cross-sectional study

Introduction

Consciousness depends on normal large-scale network function in the brain.^1-3 Impairment of large-scale brain networks leads to disorders of consciousness such as coma, vegetative or minimally conscious states with broad deficits in behavioral responsiveness and higher cognitive performance.⁴ In neurological disorders, impairment of consciousness is defined on the basis of altered responses in many different tests together leading to overall dysfunction in alertness, attention, or awareness.^4-6 Absence epilepsy is a relatively common disorder that affects 10 to 17% of children and adolescents with epilepsy⁷ presenting with brief episodes of impaired consciousness and generalized 3-4 Hz spike-wave discharge on electroencephalography (EEG). Absence seizures meet the definition of a disorder of consciousness by causing transient deficits on a broad range of tests including verbal, visuo-motor, auditory-motor, memory and attentional vigilance testing.⁸ Absence epilepsy is not benign and causes significant psychosocial disability;^7,9,10 however, the mechanism for impaired consciousness in absence seizures is not known. Importantly, although all absence seizures involve similar-appearing spike-wave discharges on EEG, some impair consciousness whereas others spare consciousness even in the same patient.^11-12 There are several possible explanations for why some absence seizures impair and others spare consciousness. One possibility is that both seizure types involve generalized brain networks, but that seizures that impair consciousness affect widespread networks more intensely than seizures which spare consciousness. A second possibility is that seizures that impair or spare consciousness may involve different focal regions of the brain.^13-16 Prior work also suggests that longer absence seizures tend to cause more severe behavioral impairment than shorter seizures. Therefore, another question is whether longer duration causes worse impairment simply because there is more time for brain physiology to be altered. Alternatively, seizures with impaired behavior may be physiologically more severe from the outset.

To examine these questions, we tested behavioral responsiveness as an indicator of consciousness in a cohort of patients with typical childhood or juvenile absence epilepsy during EEG and functional magnetic resonance imaging (fMRI) of the brain. We then compared brain network involvement during seizures that impaired versus those that spared responsiveness.

Methods

For additional details see Supplementary Methods (Figures S1-S7) in the Appendix.

Results

We evaluated 218 patients for eligibility between January 1, 2005 and September 1, 2013. 93 patients were enrolled, of which 39 had absence seizures during testing (Figure 1, Table 1; 1032 seizures total). 23 of 39 (59%) subjects were female; mean age (± SD) was 9·9 ± 3·2 years and duration of epilepsy was 3·0 ± 2·3 years, not significantly different from patients without seizures during testing. However, patients with seizures were more commonly off medications prior to the study (Table 1) and had longer total testing time per subject (mean ± SD, 2·3 ± 1·8 versus 0·9 ± 0·7 hours; mean difference 1·4 hours, 95% CI 0·9 to 2·0; P < 0·0001. Of the 1032 seizures, 810 from 34 patients were used for general fMRI timecourse analysis (Figure 3) and fewer met strict exclusion criteria for analysis of behavior in relation to fMRI or EEG (205 seizures in 27 patients for fMRI in Figure 4; 56 seizures in 10 patients for EEG in Figure 5) (see Supplementary Methods p. 5 and 7 for details). No subjects experienced generalized tonic-clonic seizures or other adverse events during the period of time that medications were withheld for the study (see also Supplementary Methods p. 3).

Mean performance on both the CPT and RTT tasks was rapidly impaired at seizure onset (Figure 2A, time = 0; n = 74 seizures in 24 patients for CPT; n = 148 seizures in 32 patients for RTT), reached a trough, and then quickly returned to baseline at seizure end. Mean (± SD) correct response rate dropped from 82·9 ± 17·0% at baseline (interictal) to 58·3 ± 39·7% during seizures for RTT (mean percent difference 24.6, 95% CI 187 to 305, p <00001, calculated across seizures). Mean correct response rate for CPT dropped from 83·7 ± 23·8% interictally to 35·5 ± 44·1% during seizures (mean percent difference 48·2, 95% CI 400 to 56·4, p <00001, calculated across seizures). Subjects showed greater mean impairment during the more attentionally demanding CPT compared to RTT (Figure 2A). Mean ictal correct response rate for CPT was 35·5 ± 44.0%, compared to RTT at 58·3 ± 40.0% (mean percent difference 22·8, 95% CI 11·2 to 34·4, p = 00001, again calculated across seizures). Behavioral timecourses in Figure 2A include baseline data surrounding additional seizures (136 seizures in 28 patients for CPT, 75 seizures in 26 patients for RTT) where task targets occurred immediately before or after but not during seizures.

Importantly, responses were spared during some of the seizures for both tasks (Figure 2B). Ictal performance showed a bimodal distribution for both tasks, with the majority of seizures having either low or high correct response rates. This provided a natural division in the data allowing us to define “impaired” seizures as those with < 25% and “spared” seizures as those with > 75% response rates to behavioral targets. No significant demographic differences were observed between patients with spared or impaired seizures (Table 1); seizures associated with both spared and impaired behavior could occur within the same individual (Table S1). Age was not significantly correlated with ictal performance for either task (CPT r = 0.19, RTT r = -0.34, p > 0.05 for both correlations). To ensure the bimodal distribution in Figure 2B was not entirely due to seizures in which only one target occurred (therefore with 0% or 100% as only possible results), we reanalyzed the data excluding these seizures (see Figure S8). Because the overall timecourse and distribution of impairment were similar on the two tasks (aside from worse average performance during seizures on CPT (see Figure 2A), data from the two tasks were combined for subsequent analyses by sorting seizures into those showing impaired or spared performance regardless of task. Separate results by task are in the Supplementary Appendix p. 6 and 10.

fMRI timecourse analysis of seizures showed a widespread, evolving pattern of activations and deactivations that persisted long after electrical activity stopped on EEG as in prior work (Figure 3C).^19,21,22 K-means clustering was performed on the mean fMRI timecourse across all 810 seizures to determine regions showing similar timecourses based on correlation between voxels over time (See Supplementary Methods, p. 8). This approach provided regions generalizable across seizures and revealed sequential involvement of three well-known networks²⁴: the default-mode network, task-positive network, and primary sensorimotor cortices and thalamus (Figure 3A and 3B). Because the clustering included the entire brain, each network extended somewhat beyond its traditional boundaries. Thus, the default-mode network, in addition to the ventral medial frontal, posterior cingulate/precuneus, lateral parietal, lateral and medial temporal cortex, also extended to the basal ganglia and other cortical/subcortical regions. The task-positive network included top-down attention areas such as the lateral frontal and parietal cortex, as well as salience regions such as the anterior insula/frontal operculum, dorsal medial frontal cortex and supplementary motor area. Finally, the primary sensorimotor network included occipital (visual), dorsolateral temporal (auditory), Rolandic (somatosensory/motor) cortex, and thalamus. These networks showed distinct, sequential fMRI responses during seizures (Figure 3C). The default-mode network showed small early activations followed by large deactivations. Next, the task-positive network showed a more biphasic response, with early activations roughly matched in amplitude by later deactivations. Finally, the primary sensorimotor areas and thalamus showed activations followed by relatively smaller late deactivations (Figure 3C). Clustering the data into more than three networks was possible but yielded less cohesive clusters (Figures S3, S4) so we focused on these three networks for the remaining analyses.

Seizures with impaired or spared behavior showed similar widespread involvement of all three networks on fMRI (Figure 4). However, the intensity of fMRI signals in all three networks was greater with “impaired” seizures compared to “spared” seizures. Mean percent differences were as follows: default-mode network 017% (impaired: 0·57 ± 0·26%; spared: 0·40 ± 0·16%; 95% CI: 0·11 to 0·23%, p < 00001); task-positive network 0·14% (impaired: 053 ± 029%; spared: 039 ± 015%; 95% CI: 008 to 021%, p < 00001); sensorimotor-thalamic network 007% (impaired: 0·41 ± 0·25%; spared: 0·34 ± 0·14%; 95% CI: 001 to 0·13%, p = 0.02)). See also Figure S9 for t-maps directly contrasting seizures with impaired versus spared performance. These results were confirmed in separate subgroup analyses for the two behavioral tasks CPT and RTT (Figures S10, S11) and were robust to different choices for behavioral cutoffs other than <25% and >75% (Figures S12, S13). A more stringent analysis on the smaller sample of patients (n = 11) who had both seizures with impaired and spared performance showed larger amplitude in seizures with behavioral impairment although this did not reach statistical significance (Figure S14).

EEG showed broadly distributed 3-4 Hz spike-wave discharges with a frontal amplitude predominance (see Figures S1, S2) as in previous studies¹⁵ and similar overall distribution pattern for seizures with spared or impaired performance (Figure 5 A, B, D, E). However, seizures with impaired task performance demonstrated greater EEG power in widespread regions compared to seizures with spared task performance (Figure 5C, F). Impaired seizures showed greater amplitude in the frequency range of both waves (2·5-4 Hz and spikes (10-125 Hz) in frontal to occipital regions. Mean differences in fractional EEG power were as follows for waves: Frontal leads 25·6 (impaired: 50·4 ± 15·2; spared: 24·8 ± 6·5; 95% CI: 210 to 30·3); Middle leads 22·1 (impaired: 35·4 ± 6·5; spared: 13 3 ± 3·4; 95% CI: 20·0 to 24·1); Posterior leads 170 (impaired: 41·6 ± 5·3; spared: 24·6 ± 8·6; 95% CI: 14·4 to 19·7); and as follows for spikes: Frontal leads 123 (impaired: 198 ± 0·36; spared: 0·75 ± 0·26; 95% CI: 111 to 136); Middle leads 103 (impaired: 1·62 ± 0·24; spared: 0·59 ± 0·14; 95% CI: 0·95 to 111); Posterior leads 0·51 (impaired: 123 ± 015; spared: 0·72 ± 0·27; 95% CI: 0·43 to 0·59), with p < 00001 for all comparisons. See Figure S15 for individual seizure examples. Results were confirmed in separate subgroup analyses for the two behavioral tasks (Supplementary Figures S16, S17), using a different behavioral cutoff (Figure S18), and with the small subset of three patients who had both seizures with impaired and spared performance (Figure S19).

Seizure duration was longer in the behaviorally impaired seizures (79 ± 66 s; mean ± SD) compared to the behaviorally spared seizures (3·8 ± 3·0 s) (p < 0·0001; mean difference 4.1 s; 95% CI 3.0 to 5.3 s). To determine the timing of physiological differences between seizures with impaired versus spared task performance we examined the timecourses of fMRI and EEG changes (Figure 6). fMRI showed larger amplitude from the outset in impaired seizures in all three identified networks (Figure 6 A-C). Interestingly, the larger fMRI amplitude in impaired seizures actually preceded seizure onset by about ten seconds. Later fMRI increases as well as decreases were also larger in seizures with impaired performance. Similarly, EEG measurements showed greater amplitude from the outset of seizures with impaired performance (Figure 6 D-E). EEG amplitude for both waves (2·5-4 Hz) and spikes (10-125 Hz) diverged at or just prior to the time of seizure onset. Thus seizures with impaired performance showed more severe physiological changes on both fMRI and EEG at or even before seizure onset.

Discussion

Using 1032 seizures captured from 39 patients with typical childhood or juvenile absence epilepsy, we found that the overall intensity of seizures correlates with behavioral impairment. This relationship held for seizure amplitudes on both fMRI and EEG in widespread regions of the brain, was present from the outset of seizures, and extended to tasks of varying difficulty. These results represent a fundamental shift from prevailing views of impaired consciousness in absence seizures, which hold that focal brain involvement^8,13,15 or prolonged seizure duration⁸ determine the severity of impaired consciousness. Our data present a very different mechanism for impaired consciousness, demonstrating instead that: seizures with impairment have greater physiological intensity in widespread networks including most of the brain, not only in focal areas; and seizures with impairment are more physiologically intense at seizure initiation, not just at later times during seizures.

Recent theories of normal and disordered consciousness have recognized the importance of large-scale information integration or of a global neuronal workspace necessary for the emergence of consciousness.^1,3 Neuroenergetic considerations also emphasize the overall function of widespread brain networks. However, it has long been recognized that absence seizures do not affect the brain homogenously and that some regions, particularly the midline frontal cortex, are more intensely involved than others on EEG.¹⁵ fMRI studies have also shown regional heterogeneity that varies from one patient to the next.^16,22 Animal models suggest that spike-wave discharges arise from focal regions of enhanced cortical excitability.^14,25 Our present findings demonstrate that although seizures may be more intense in some regions than others, the difference between seizures that do or do not impair consciousness is not confined to local regions, but rather affects virtually the entire brain. This supports and extends previous work which proposed that impaired consciousness in absence seizures is caused by default mode suspension together with reduced perception through sensory inputs, as well as impaired function of the thalamus, frontal regions and the insulae²³. Thus, although seizures may have regional heterogeneity, our findings support a model in which global rather than local differences between seizures are related to impaired consciousness.

Longer duration of spike-wave discharges on EEG has been associated with more severe behavioral impairment⁸ although exceptions clearly exist.^8,11 In agreement with a recent study²⁶, fMRI amplitude was linearly related to spike-wave duration (Figure S7). An important question is whether longer seizures cause more severe effects simply because there is more time for longer seizures to induce physiological changes in the brain. However, because we incorporated this relation into our model, differences in fMRI signals between seizures with spared versus impaired task performance (Figure 4) were not simply due to spike-wave duration. Alternatively, longer and more behaviourally impaired seizures might be more physiologically intense from the outset. Interestingly, some previous studies have shown that fMRI changes can precede seizures by several seconds,^19,21,22 and our results identify early fMRI changes preceding seizures with behavioural impairment, but not when behaviour is spared. Additionally, we noted early differences in EEG signals between seizures with impaired versus spared task performance. These results suggest that greater behavioural impairment from longer seizures is not simply due to gradual worsening of abnormal physiology during the course of the seizure, but rather that some seizures have more severe physiological abnormalities at or before seizure onset. This raises the intriguing possibility that a vulnerable state occurs at the initiation of some seizures in which an unknown combination of intrinsic brain fluctuations and seizure initiation mechanisms interact to generate more physiologically and behaviorally severe seizures.

Whole-brain parcellation of the fMRI timecourse in absence seizures revealed three well-described networks, each with a unique hemodynamic response. These include sequential involvement of the default-mode, task-positive, and primary sensorimotor-thalamic networks. Therefore even pathological states such as absence seizures may acquiesce to established network boundaries. This result complements evidence from normal subjects whereby established cortical networks persist across states from rest to task processing,²⁴ and in normal loss of consciousness in transitioning from wakefulness to sleep.²⁷ Successful task performance requires coordination of multiple cognitive processes. Primary visual and motor cortices are involved in sensory input and motor output, while attention initiation and maintenance involves a wider collection of areas including dorsal and ventral fronto-parietal circuits and the default-mode network. However, while temporally covarying activity across regions is thought to reflect processing of mutual information under normal task conditions, the hypersynchronous state of the brain in these same networks during absence seizures may paradoxically impair normal functional connections and behavior.

To investigate the full spatio-temporal profile of fMRI changes during seizures, we used a data-driven approach without prior assumptions about the signals. To analyze EEG signals, we likewise avoided models or assumptions and simply examined amplitude mapped over the recording surface. The strong convergence between fMRI and electrophysiological results in our data allays possible concerns that in some circumstances fMRI may not accurately reflect underlying neuronal activity.²⁵ Notably, both fMRI and EEG demonstrated early and widespread increased amplitude in seizures with behavioral impairment. One additional potential factor to consider is the effect of breath-holding during seizures on fMRI amplitude²⁸, which could be addressed by measuring breathing during fMRI in future studies.

An important remaining question is why fMRI and EEG signal increases and decreases are larger in some seizures, namely those with more severe behavioral impairment. The amplitude of fMRI signals under normal conditions^29,30 and during spike-wave seizures^14,25 is related to the number of neurons involved in a given volume of brain and the intensity of their electrical activity. EEG amplitude is also related to the intensity and the synchrony of neuronal electrical activity. We postulate that seizures with more severe behavioral impairment engage a larger number and/or more intense activity in neurons in the seconds preceding obvious EEG onset, which may then cause larger and longer-lasting fMRI and EEG changes throughout the brain during and following seizures. Our study is unable to demonstrate whether particular areas of the networks rather than entire networks cause impairment in consciousness, and the possibility remains that some specific region or regions are more important than others. However, we show widespread brain involvement in absence seizures as well as correlation of behavioral impairments with disruptions in each network. Our cross-sectional study using noninvasive measurements on human subjects is intrinsically limited in testing direct causal relationships between brain states and behavior. The fundamental mechanisms determining seizure severity will require further investigation, perhaps best performed in experimental models where direct neuronal recordings can be obtained.

Absence seizures are not benign, and cause substantial behavioral deficits in many children living with this disorder.^8,9 Our observation that larger amplitude fMRI and EEG changes occur in widespread networks early during initiation of more severe seizures may have broader implications for other seizure types and for brain-state transitions in general. These findings suggest that future work should investigate the physiology of these early, widespread changes in brain networks. Crucially, understanding the mechanisms regulating absence and other seizure severity may ultimately provide new opportunities to intervene therapeutically and protect patients from deficits associated with seizures.

Supplementary Material