The initiation of epileptiform discharges (EDs) is accompanied by intracellular paroxysmal depolarization shifts, and the termination is attributed to a process of recurrent inhibition (1). However, the factors and mechanisms responsible for the generation and termination of EDs and seizures are not well understood. Some researchers have postulated that interictal–ictal transitions depend on a Poisson process with a fixed probability of occurrence between the two states or time-dependent mechanisms characterized by a variable probability of transition based on the length of time elapsed in the current state (2).

Genetic generalized epilepsies (GGEs) are electrographically characterized by bilateral, symmetric, and generalized EDs (3). The distinction between interictal and ictal activity depends on the ED duration and associated clinical features (4). ED duration is a more objective measure, whereas whether symptoms are manifest depends on the precision of testing, particularly in relation to absence seizures (4). There is no consensus on the threshold value of ED duration to distinguish interictal from ictal activity, and 2 or 3 s is often used by researchers as the cut-off (5, 6). Although ED duration is a continuum, it is biologically plausible that there are two distinct populations of ED with relatively shorter and longer durations representing interictal and ictal activities, respectively.

Interictal EDs and high-frequency oscillations are among the important biomarkers of epilepsy (7, 8). The relationship between interictal abnormalities and seizures remains poorly understood. Studying the characteristics of seizures and EDs will help us understand the underlying rhythmicity of the complex process of epileptogenesis. Previous research has demonstrated the circadian rhythmicity of seizures and interictal ED in both focal and generalized epilepsies (9–11). The rhythmicity and patterns of seizure duration, interseizure interval, and seizure bursts in focal epilepsy have also been described (12, 13). Many researchers have previously shown the impact of the sleep–wake cycle on epileptogenicity (10, 14). A study involving two patients diagnosed with “petit mal epilepsy” reported a unimodal or weakly bimodal distribution of durations of EDs. The same study found that discharges were relatively longer during wakefulness and shorter in sleep (15). The characteristics of ED durations and interdischarge interval (IDI) in GGE have not been well studied. A study of ED duration and IDI in relation to sleep–wake cycle will reinforce the findings of previous research by showing the modulation of ED generation. The overall assessment of multiple facets in epileptogenicity (temporal distribution of seizures, the temporal distribution of ED, ED duration, and IDI) in relation to the sleep–wake cycle is likely to provide a more thorough understanding of this complex process.

Against this backdrop, we sought to investigate (1) the characteristics of ED duration and IDI and (2) the influence of vigilance state on ED duration and IDI. We hypothesized that ED duration and IDI are of bimodal distribution, and the characteristics of those populations are influenced by the sleep–wake cycle.

We performed 24-h ambulatory EEGs on 120 patients diagnosed with GGE, and 107 recordings were abnormal showing EDs. We analyzed data from 23 subjects who had 100 or more generalized EDs during the 24-h period. The clinical and demographic characteristics of this cohort are summarized in Table 1. The mean sleep duration was 9.4 h (SD = 1.5), whereas the mean sleep onset and offset times were 2249 and 0735 h, respectively. No EEGs had to be excluded from the analysis due to poor recording quality.

No one had clinical seizures based on the event diary and the “event button” press during the recording. We also did not detect any generalized tonic–clonic seizures or myoclonic seizures based on the EEG. As the EEGs were ambulatory, recorded in patients’ own environment, we had to depend on the diary maintained by patients and families to correlate the EEG with symptoms. In this small cohort, no one reported seizure symptoms. However, the patient-reported symptoms can be unreliable particularly in relation to myoclonic and absence seizures. With ambulatory EEG, there was no opportunity to obtain corroborative evidence of seizure activity by an experienced observer such as a doctor, a nurse, or an EEG technologist. Hence, we did not make a distinction between “ictal” and “interictal” EEG activity. We opted to refer to ED in general without making that clinical distinction.

Epileptiform discharges were significantly more frequent during sleep with the mean ED rate of 15.1/h in sleep versus 4.6 in wakefulness (P < 0.001).

K-means cluster analysis and FMM yielded concordant results. Therefore, we report only the outcome of FMM analysis here. The individual results on ED duration from the 23 subjects are summarized in Table 2. Figure 1 illustrates the histograms and kernel density plots of ED duration and IDI from the pooled data of all 23 patients. Two populations of relatively short and long durations are evident in each histogram (Figures 1A,B). The characteristics of populations revealed by FMM are illustrated in Figures 1C,D. The analysis describes two populations of ED durations: short (population 1) and long (population 2). In the majority of cases (except for subjects 3, 12, and 19), the population 1 had a higher prevalence. For example, in subject 1, the prevalence of populations 1 and 2 were 76 and 24%, respectively.

Table 3 summarizes the results on IDI. Similar to ED durations, we describe two populations of IDI with short and long intervals. In all subjects, the population with short IDI had a higher prevalence.

The influences of vigilance state on ED duration and IDI are summarized in Table 4. In this analysis, we collectively studied all discharges from all subjects dichotomized into two populations of ED durations and IDI. For each population, durations were longer in wakefulness compared with sleep state. These differences were statistically significant (Table 4).

The influences of seizure-free duration and AED therapy are detailed in Tables 5 and 6. The basic population structure of ED duration and IDI remains unchanged in the results with statistically significant differences between populations of short and long durations. This is evident at both group level and individual level. With longer seizure-free durations, ED durations tend to be shorter and IDI tends to be longer although not statistically significant. Similarly, in patients not on AED therapy, ED durations appear longer, but the sample is too skewed to draw any robust conclusions.

We also found that both measures varied throughout the course of the day with similar undulations. Both mean ED duration and mean IDI had two peaks at 1000 and 1800–2000 h, whereas the trough (shortest mean ED duration and ISI) was at 2300–0100 h (Figure 2).

As our sample was heterogeneous, we conducted an additional subgroup analysis of 15 patients diagnosed with juvenile absence epilepsy (JAE). Although parameters changed, the population structure remained intact in this subgroup supporting our hypothesis (Figure 3; Table 7).

Our study shows that ED durations and IDI are clustered into populations rather than demonstrating a unimodal distribution. The populations with short ED duration and short IDI have a higher prevalence. These populations are under the influence of the state of vigilance with both the ED duration and the IDI recording significantly longer durations in wakefulness. Both variables show undulations throughout the course of the day with similar peaks and troughs. In addition, both variables demonstrate individual variations in terms of durations and prevalence of populations, potentially suggesting the individual differences in epileptogenicity. These patterns illustrate different characteristics of ED populations. Our findings strengthen the results of previous research on ED (2, 15).

Our results also highlight the influence of sleep–wake cycle on both the ED duration and the IDI. In a previous study involving patients diagnosed with GGE, we demonstrated that EDs were longer but less frequent during wakefulness while occurring in a time-of-day-dependent rhythm (11). Two previous studies on generalized epilepsy reported that a smaller number of longer EDs occurred during wakefulness, whereas the opposite trend was observed in sleep (15, 21). Stevens et al. studied 5 patients, and 2 were diagnosed with “petit mal epilepsy,” whereas Kellaway et al. studied 19 patients with generalized epilepsy (15, 21). Our current findings, using a rigorous methodology in a larger sample, are concordant with their observations, and we provide more detailed descriptions of cluster characteristics.

Kellaway et al. postulated that ED occurrence in generalized epilepsy was the result of an interaction between two processes: a 24-h circadian cycle and a shorter 100-min cycle related to sleep (21). Our findings of two populations, undulations across 24-h time scale, and changes in different states of vigilance suggest the interactions among intrinsic biorhythms.

There are potential problems linked to the identification and estimation of multipopulation data using common FMM estimation approaches. An important pitfall is the omission of a critical measure affecting the spread of data (22). We adopted a stepwise approach to overcome such pitfalls. We first used graphical aids (histograms and kernel density plots) as a guide to explore the plausibility of our hypothesis, and the plots indicated the existence of two populations with relatively short and long durations. Then, we quantified those populations independently with two approaches: K-means clustering and FMM. Both methods yielded concordant results indicating the robustness of our statistical approach.

Extracellular recordings from the somatosensory cortex of cats have revealed a bimodal distribution of interspike intervals (23). We reiterate that failure to recognize multimodal data will provide a single mean in the analysis, but the single mean will not accurately characterize the population of interest. Our analysis was directed toward an unbiased account of ED durations and IDI, with a view to identifying multimodal data. Our approach was influenced by previous research (15, 23).

Our results indicate that two populations with short and long durations of ED and IDI persist despite the influence of confounders such as AEDs and seizure-free duration. This is seen at both group level and individual level although the population parameters vary. This observation suggests that the generation of ED is modulated by the sleep–wake cycle and external factors such as AED therapy.

The distinction between “interictal” and “ictal” epileptiform activity in GGE remains contentious. In general, interictal EEG abnormalities constitute “epileptiform patterns occurring singly or in bursts lasting at most a few seconds,” whereas ictal rhythms are defined as “repetitive EEG discharges with a relatively abrupt onset and termination and characteristic pattern of evolution lasting at least several seconds” (24). When an EEG seizure pattern is not accompanied by clinical signs and symptoms, the diagnosis of a subclinical seizure is made (24). Myoclonic seizures and generalized tonic–clonic seizures demonstrate well-characterized EEG changes and clinical features; therefore, the distinction between ictal and interictal EEG activities is usually unambiguous (4). Differentiating interictal from ictal EDs can be difficult in absence seizures, as EDs demonstrate little evolution. Therefore, to differentiate ictal from interictal activity, one has to depend on the duration of the activity and accompanying clinical features, particularly impairment of consciousness, during the ED. There is no universal agreement on the duration of the GSW paroxysm constituting an absence seizure. Sadleir et al. diagnosed the absence of seizures based on the two criteria: (1) generalized spike-wave activity of any duration when accompanied by clinical signs, and (2) GSW lasting >2 seconds even if not accompanied by clinical correlates. Discharges of <2-s duration without clinical signs were identified as interictal fragments (6). Other researchers have defined a GSW burst lasting ≥3 seconds, irrespective of clinical signs, as an absence seizure (5, 25). Our study does not define the cut-off duration to distinguish between interictal and ictal activities. We were unable to correlate with clinical features as no one in the cohort reported seizure symptoms in their diaries during the EEG recording period. However, based on our finding of two populations of ED durations, one may speculate that short-duration ED populations represent interictal activity, whereas long-duration populations are ictal in nature. Further research is needed to clarify this uncertainty.

Circadian patterns of epileptogenicity have been described at the population level (14, 26). A recent study based on the long-term intracranial recordings in human focal epilepsy has found that both seizures and interictal EDs follow a circadian pattern of temporal distribution with individual differences in this rhythm (9). There is conflicting evidence on the relationship between interictal ED and seizures with studies indicating that IED may prevent seizures, facilitate seizures, or are simply an epiphenomenon (27–29). It is also unclear whether this process differs between focal and generalized epilepsies. Our study illustrates another dimension of intrinsic epileptogenicity. Understanding these multifaceted temporal patterns and dynamics of epileptogenesis at both the individual and the population levels is likely to enhance our ability of seizure prediction (30).

We acknowledge some study limitations. The sample size was small, and patients were recruited from tertiary centers introducing a bias. Our sample was heterogeneous. However, the subgroup analysis of JAE shows a similar population structure supporting our hypothesis. Most patients were on antiepileptic medications at the time of EEG recording. A drug naive, incident cohort would have been ideal for a study of this nature. Sodium valproate, levetiracetam, and lamotrigine are known to suppress EDs (31–33). Twenty of 23 patients in the cohort were treated with those 3 AEDs as monotherapy or polytherapy. Although we studied the impact of the number of AEDs used, the sample was too small and skewed to do a detailed evaluation of the influence of AED dosage. All EEGs were scored by the same reader, who was not blinded. Signal dropouts and artifacts are potential sources of error in ambulatory EEG. However, having provided written instructions to patients, we were able to obtain high-fidelity recordings.

Due to such limitations, we suggest that findings in this study should be considered preliminary. Future studies should focus on a large sample of drug naive incident cases of GGE allowing separate analysis of different syndromic subgroups with detailed sleep staging. A detailed morphologic analysis of EDs including rhythmic versus non-rhythmic and ictal versus non-ictal patterns in different states of vigilance should also be included.

In summary, distinct populations with relatively short and long durations of ED and IDI can be identified in relation to EDs in GGE. These populations are identifiable at both group and individual levels although the population parameters vary. The population characteristics are heavily influenced by the state of vigilance. These patterns highlight the characteristics of ED and IDI populations in GGE and the influence of the sleep–wake cycle on the generation of EDs.

This study was conducted with approvals from the Human Research Ethics Committees of Monash Health and St. Vincent’s Hospital, Melbourne. All subjects provided written informed consent in keeping with the Declaration of Helsinki.

US contributed to study concept and design, literature search, data collection, data analysis, data interpretation, drafting, and critical revision of manuscript. RB contributed to data analysis and interpretation and critical revision of manuscript. MC contributed to study concept and design, data interpretation, and critical revision of manuscript. WD contributed to study concept and design, data interpretation, and critical revision of manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.