An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Noebels JL, Avoli M, Rogawski MA, et al., editors. Jasper's Basic Mechanisms of the Epilepsies. 5th edition. New York: Oxford University Press; 2024. doi: 10.1093/med/9780197549469.003.0015
Epilepsy is among the most dynamic disorders in neurology. A canonical view holds that seizures, the characteristic sign of epilepsy, occur at random, but for centuries humans have looked for patterns of temporal organization in seizure occurrence. Observations that seizures are cyclical date back to antiquity, but recent technological advances have, for the first time, enabled cycles of seizure occurrence to be quantitatively characterized with direct brain recordings. Chronic recordings of brain activity in humans and in animals have yielded converging evidence for the existence of cycles of epileptic brain activity that operate over diverse timescales: daily (circadian), multiday (multidien), and yearly (circannual). This chapter reviews this evidence, synthesizing data from historical observational studies, modern implanted devices, electronic seizure diaries, and laboratory-based animal neurophysiology. The chapter discusses advances in our understanding of the mechanistic underpinnings of these cycles and highlights the knowledge gaps that remain. The potential clinical applications of a knowledge of cycles in epilepsy, including seizure forecasting and chronotherapy, are discussed in the context of the emerging concept of seizure risk. In essence, this review addresses the broad question of why seizures occur when they occur.
The apparently random occurrence of seizures is a core feature of epilepsy and one of the most debilitating aspects of the disorder. For millennia, humans have used correlative observations to try and explain the timing of seizures (Reynolds, 1990). Modern neurologists still observe conspicuous regularities in seizure occurrence in some individuals with epilepsy, whose diaries show seizures repeatedly striking at similar times of day, days of the month, or even seasons of the year. Since the early 2000s, advances in miniaturization and telecommunication have enabled the production of fully implantable electroencephalogram (EEG) systems that can chronically monitor epileptic brain activity. This approach strongly contrasts with the classical use of noninvasive scalp EEG, which can only be used over several days. Now, EEG data from the same individual can be accumulated over months, or even years, opening the possibility of identifying robust cyclical patterns in seizure occurrence that manifest over long timescales (hereafter, “seizure cycles”).
The statistical analysis of large EEG datasets has matured over the last decade and has now provided convincing evidence that cycles in epilepsy exist at multiple timescales—circadian, multidien, and circannual—and are highly prevalent across human and animal epilepsies.
In this chapter, we review the literature concerning cyclical influences on seizure occurrence and compare the findings from animal and human datasets, highlighting a striking cross-species convergence in cycles of epileptic brain activity. First, we discuss the knowledge gained over time about temporal patterns in epilepsy, from ancient beliefs about seizure recurrence documented since antiquity to the seminal observational studies in the epilepsy colonies, and modern-day quantitative analyses. Second, we describe the cyclical organization of seizures at three timescales: circadian, multidien, and circannual. We highlight statistical advances supporting these observations and synthesize putative mechanisms of epileptic cycles, although mechanistic understanding is currently limited. We synthesize the existing evidence into the concept of cycles of seizure likelihood that condition the stochastic timing of seizures over multiple timescales. We conclude on the practical implications of these discoveries for fundamental, translational, and clinical epilepsy research, including the feasibility to forecast seizures.
The consistent observation of organized cycles of seizure recurrence in historical records, seizure diaries, laboratory animals, and human chronic EEG recordings is remarkable and of great clinical significance.
Documentation of cycles in epilepsy dates back to the first writings in cuneiform in the Babylonian era (Reynolds, 1990). From time immemorial, the striking periodicity of some epilepsies has captured the attention of families of patients suffering from the disorder, healers, and later physicians and scientists (Bercel, 1964). Throughout history, the periodic nature of epilepsy has been associated with divine, diabolic, celestial, or environmental influences. The proposed link between lunar cycles and seizure occurrence is rooted in mythologies that predate Hippocrates (Cule, 1973). European scholars of the Enlightenment debated the influence of celestial movements in conjunction with epilepsy. Sir Richard Mead, a British physician in the eighteenth century, argued for a lunar influence on seizures based on predictive power: “The Moon’s influence was so visible on the generality of them [people with epilepsy] at the new and full, that I have often predicted the times of the fits with tolerable certainty” (Mead, 1748). In contrast, Samuel Auguste Tissot, a Swiss medical authority at the French court fiercely combated the idea: “[These beliefs] originate from an antique superstition, that, ignoring the true cause of the disorder [epilepsy], attribute it to a specific act of celestial anger” (translated from the original French in Tissot, 1784). With the introduction of the first antiseizure medications in the nineteenth century, most authors refuted the view of a lunar connection to seizures and favored the endogenous influence of specific brain states (Leuret, 1843; Moreau, 1854; Echeverria, 1879; Fere, 1890). Meticulous records made a link with the sleep-wake cycle more obvious and led Sir William Richard Gowers to classify epilepsies by circadian seizure chronotype: “nocturnal,” “diurnal,” and “mixed,” with a prevalence of about 22%, 45%, and 33%, respectively (Gowers, 1881). This 150-year-old observation still awaits an explanation.
At the turn of the twentieth century, many epilepsy patients lived in colonies outside of major cities and worked in the fields, an activity believed to be favorable to control seizures. Intendant physicians of the epilepsy colonies had the scientific advantage of studying individuals with epilepsy in an institutionalized setting, where activities such as timing of medications, meals, and sleep were strictly identical for many years in an attempt to improve health (Langdon-Down and Brain, 1929). This controlled “quasi-experimental” environment enabled the observation of major characteristics of the chronobiology of epilepsy: (1) within days, seizure timing is governed both by the time of the day and vigilance states (asleep vs. awake) in individuals; (2) over days, single seizures or clusters of seizures recur at periodic intervals of one to several weeks in many people with epilepsy.
Building on the observations made by Gowers in the nineteenth century, intendant physicians of the epilepsy colonies in the United Kingdom, the United States, and Scandinavia made a series of longitudinal observations on their patients to establish circadian seizure chronotypes (Reynolds, 1861; Gowers, 1881; Patry, 1931; Magnussen, 1936; Griffiths and Fox, 1938; Hodge et al., 2019). A preferential time for occurrence of seizures was found in 65% of individuals in the epilepsy colonies studied, revealing shared peak times in seizure occurrences that were related to times of sleep and wake (Reynolds, 1861; Gowers, 1881; Langdon-Down and Brain, 1929; Griffiths and Fox, 1938).
A few years later, other intendants of the Lingfield colony, G. F. Griffiths and J. T. Fox, identified multidien cycles of recurrent seizures ranging from ~1 week to ~3 months in the same institutionalized individuals (Griffiths and Fox, 1938). In their study, seizure periodicity was highly individual, but shared patterns such as monthly recurrence were found in men and women, adults and children alike. These historical studies also showed that it was possible to shift the identified circadian and multidien peaks of seizure occurrence by modifying times of sleep, wake, meals, or medication (Langdon-Down and Brain, 1929; Patry, 1931; Griffiths and Fox, 1938).
A few years after the discovery of large-timescale “rhythms in epilepsy” through purely observational studies, the invention of EEG (Berger, 1931) shifted the focus to the subsecond timescale and increased the interest for the use of technology in neurology. Indeed, EEG can detect and record the abnormal brain electrical activity underlying seizures with millisecond precision (Gibbs et al., 1935), which has advanced our understanding of epilepsy and continues to elucidate new aspects of ictogenesis and epileptogenesis. One fundamental insight from EEG studies in individuals with epilepsy is that abnormal brain activity correlates with seizures but also exists between seizures in the form of interictal epileptiform activity (IEA); that is, transient discharges visible on the EEG that do not produce overt symptoms (Jasper, 1949). Interictal discharges have great diagnostic value for the initial diagnosis of epilepsy by conventional, brief (~20 minutes) EEG recordings. However, the quantification of IEA for assessing disease severity is not yet well established. The rate of IEA can be expressed as counts of discharges per unit of time, for example, per hour or per day, and is typically thought to reflect different levels of “brain irritability,” although, until recently, this clinical notion lacked an evidence-based definition. Long-term EEG recordings in the hospital have repeatedly shown that both IEA and seizure occurrence exhibit circadian organization (Herman et al., 2001; Hofstra et al., 2009). However, hospital-based studies necessarily adopt cross-sectional designs and cannot assess cycles that span beyond a few days, highlighting the value of tracking seizures in ambulatory settings.
Objective monitoring of epileptic brain activity with chronic EEG recording over multiple weeks has been performed in mice and rats for decades for the study of epilepsy (Ashida et al., 1965). Substantial advances in our understanding of epilepsy, including seizure timing, have come from extensive work with animal models (Grone and Baraban, 2015). Although no single animal model fully recapitulates the disorder, the diversity of available models—from zebrafish to rodents and canines—parallels the diversity of human epilepsies, and some models show electrophysiological and pathological features that are hallmarks of human epilepsy syndromes. Despite their lissencephalic brains, rats and mice are often the favored species for epilepsy modeling, given the available armamentarium of genetically modified lines (Frankel, 2009; Noebels, 2015) and the possibility of implanting intracranial and/or cranial EEG electrodes for long-term recording. Canine epilepsy is the most common neurological disorder in the domesticated dog (Canis familiaris; Heske et al., 2014) and is therefore studied in its own right, using EEG recording techniques similar to those used in humans. Canine epilepsy shares many features with human epilepsy (Berendt et al., 1999), including the existence of seizure cycles (Figs. 15–1 and 15–2; Gregg et al., 2020).
To this day, seizure diaries maintained by people with epilepsy or their caregivers are the gold standard for monitoring seizure burden both in clinical practice and clinical trials. On the basis of such records from >1,000 patients, in the 1960s, Nicholas Bercel identified regular seizure cycles in ~10% of people living with epilepsy in the community (Bercel, 1964). However, self-reported seizure diaries are frequently unreliable—they can either underestimate or overestimate seizure rates—and often show limited correlation with the true underlying rate of seizure occurrence as determined by EEG (Blum et al., 1996; Hoppe et al., 2007; Elger and Hoppe, 2018). These inaccuracies could obscure rhythmicity in self-reported diaries. Conversely, self-reports of seizures are subject to recall and reporting biases, which could also be cyclical, a concern that highlights the need for objective data in epilepsy.
With the advent of the internet, online diaries and mobile applications have been developed to facilitate seizure charting (Fisher et al., 2012; Chiang et al., 2020). Although these digital seizure diaries have similar potential biases as their pen and paper counterpart, their availability anywhere and at any time may improve the quality of the collected data. One clear advantage for research into seizure cycles is the uniformity of data format, which enables rapid analysis of seizure patterns across millions of pooled seizures (Ferastraoaru et al., 2018; Goldenholz et al., 2020). Seizures occur in repeating patterns within individuals (and not necessarily cross-sectionally), and a powerful statistical approach to study such periodic signals is to analyze long individual datasets (Karoly et al., 2018).
Recent developments in wearable sensors of physiological parameters (e.g., heart rate, oxygen level, movement) and in algorithms for seizure detection (Johnson and Picard, 2020; Nasseri et al., 2020) have resulted in devices (e.g., Brain Sentinel, Halford et al., 2017; Empatica, Onorati et al., 2017) for seizure detection using peripheral signals. These devices have the potential to improve the identification of seizure cycles, although confirmatory studies are currently lacking. Dry electrodes (Hinrichs et al., 2020) and stick-on scalp EEG patches (Frankel et al., 2021) are emerging technologies aimed at expanding the use of scalp EEG to the ambulatory setting, but they are not currently in widespread use. Over the past decade, chronic EEG monitoring has become possible with the advent of implantable systems with electrodes placed in the brain (stereotactic or depth EEG, Morrell, 2011; Baud and Rao, 2018; Baldassano et al., 2019) on the surface of the brain (subdural EEG, Cook et al., 2013), or between scalp and skull (sub-scalp EEG); the latter is still in trial phase for application in epilepsy (Duun-Henriksen et al., 2020). To date, all chronic EEG recordings (months to years) have been obtained with intracranial electrodes connected to implanted devices used in clinical trials or approved for epilepsy therapy. Chronic EEG enables the quantification of clinical and electrographic seizures and reveals how IEA relates to seizures over longer periods than can be studied in the hospital.
In the past decade, new technology for tracking seizures and IEA, combined with novel analyses, has greatly contributed to our understanding of seizure cycles, which have now been identified in multiple independent datasets from humans (Baud et al., 2018; Karoly et al., 2018; Maturana et al., 2020; Leguia et al., 2021) and animals (Baud et al., 2019; Gregg et al., 2020).
In addition to advances in EEG hardware, the toolbox for analysis of epilepsy EEG data has greatly expanded in the last decades, borrowing methodology from the fields of physics, engineering, and computational neuroscience. With the availability of longitudinal data in epilepsy, the methodology to study cycles can take advantage of the periodic recurrence of a phenomenon within the same individual (i.e., individual cycles) to elucidate the three characteristics of a cycles: its amplitude (or power), period length, and phase (Fig. 15–1). Studies based on short hospital stays have necessarily favored cross-sectional analyses, which cannot uncover individual cycle, and could actually mask the existence of cycles by averaging out individual cycles that are out of phase. The identification of individual temporal patterns is a prerequisite for the elucidation of seizure chronotypes that might be shared across a group of individuals. Additionally, quantitative methodology borrowed from the fields of weather forecasting and machine-learning have supported advances in the pursuit of the holy grail of forecasting seizures (Cook et al., 2013; Karoly et al., 2017; Proix et al., 2020).
Circular data and statistics. Cycles are best described as circular data measured in degrees or radians on a circle, to account for the continuity between 0° and 360°, as opposed to a linear representation. Circular histograms provide (more...)
Evidence indicates that the momentary likelihood of a seizure is co-modulated by cycles operating at various timescales: from the ultradian sleep cycle to circannual influences and possibly even longer changes throughout life. In any given individual, a combination of cyclical modulations at specific timescales and phases may give rise to a unique temporal pattern of seizure occurrence. Shared characteristics at the level of groups of people with epilepsy led to the notion of seizure chronotypes, originally described in a handful of landmark reports at the turn of the twentieth century (Gowers, 1881; Langdon-Down and Brain, 1929; Griffiths and Fox, 1938).
Highlighting the importance of cycles of epileptic brain activity as clinical phenomena, modern cEEG data show a ~90%, ~60%, and ~10% prevalence of circadian (Karoly et al., 2018; Leguia et al., 2021), multidien ( Leguia et al., 2021), and circannual (Leguia et al., 2021) seizure cycles, respectively. Historical reports complete this picture with a documented prevalence of 20%–30% for sleep-related seizures (Gowers, 1881; Langdon-Down and Brain, 1929; Magnussen, 1936; Griffiths and Fox, 1938). In this chapter, we use the term circadian cycle to indicate a ~24-hour rhythm without reference to a specific mechanism, and the term sleep-wake cycle when specifically referring to the effects of vigilance state. We discuss the influence of cycles on seizure occurrence, by order of timescale, from ultradian to circannual.
Although the modulation of seizures by the sleep-wake cycle and the circadian cycle are often conflated in the epilepsy literature, they should be treated as distinct but intertwined modulators of epileptic brain activity (Janz, 1962; Khan et al., 2018). Formally disentangling these modulators can be challenging and might require the use of “forced desynchrony” protocols, possible in animal models and humans, in which behaviors (e.g., sleeping and eating) are evenly scheduled across all phases of the circadian cycle and environmental influences (e.g., light and temperature) are kept constant (Pavlova et al., 2009). However, a long-standing clinical tradition attributed a preponderant role of sleep in seizures occurring during the night, and we here discuss these entities.
Historical studies of different epilepsy syndromes indicated that around twice as many individuals with epilepsy have predominant diurnal as compared to nocturnal seizures (Fig. 15–2i,j, Gowers, 1881; Langdon-Down and Brain, 1929; Patry, 1931; Magnussen, 1936; Griffiths and Fox, 1938). Although these studies did not collect information on the sleep-wake cycle, nocturnal seizures are likely to be sleep-related. Among more recent studies that did include EEG-based staging of nocturnal sleep, a meta-analysis of 1,990 seizures in 542 people with epilepsy suggested that seizure rates were 50–90 times higher in NREM sleep than in REM sleep (Ng and Pavlova, 2013). Bringing together knowledge of anatomical and temporal aspects of epilepsy, evidence indicates that individuals with focal epilepsy are more likely to experience seizures during sleep than individuals with generalized epilepsy (Janz, 1962). In focal epilepsy, analyses of seizure localization suggest that frontal lobe seizures are often sleep-related (Provini et al., 1999), whereas temporal lobe seizures variably occur during sleep and wakefulness; data on occipital and parietal seizures were insufficient for firm conclusions to be drawn (van Campen et al., 2015). Canonical examples of specific epilepsy syndromes that are linked to vigilance states include sleep-related hypermotor epilepsy (Tinuper et al., 2016; Licchetta et al., 2017) and seizures after awakening, a form of generalized epilepsy (Janz, 1962; Xu et al., 2018). Evidence of concordance of sleep-wake seizure timing among family members of individuals with specific epilepsy syndromes suggests a possible genetic basis for the influence of vigilance states on seizures (Winawer et al., 2016). Although disruptions to sleep homeostasis are widely acknowledged to influence seizure frequency and timing, causal evidence for this link is lacking (Rossi et al., 2020).
Circadian and multidien seizure cycles. a: Historical perspective on the discovery of circadian and multidien cycles of seizures and the variety of methods used to log these events. b: Reproduction of the meticulous charting of seizures in one subject (more...)
In addition to modulating the timing of overt seizures, the sleep-wake cycle also profoundly modifies electrographic epileptic brain activity. With the advent of EEG in the 1940s, Frederic and Erna Gibbs soon showed that IEA was higher during sleep than during wakefulness, even when individuals presented mostly with wake-related seizures (Gibbs et al., 1936; White et al., 1962). Sleep is now thought to promote IEA by inducing state-dependent neuronal synchrony (Ng and Pavlova, 2013; Frauscher et al., 2015; Frauscher and Gotman, 2019). Several studies using scalp EEG found that synchronous EEG delta activity or K-complexes are associated with the occurrence of sleep-related ictal episodes (Sforza et al., 1993; Provini et al., 1999; Manni et al., 2008). To increase the diagnostic yield of EEG, this sleep-related EEG activation is routinely assessed in clinical practice by encouraging naps during recordings (Schwarz and Zangemeister, 1978; Rossi et al., 2020). Indeed, IEA, including epileptic spikes and high-frequency oscillations, is more prevalent and involves more extended brain networks during NREM sleep, but it is less prevalent and more restricted during wakefulness and REM sleep (Lambert et al., 2018; Frauscher and Gotman, 2019; Kang et al., 2020). In a small, innovative study, five individuals with genetic epilepsies were kept under a constant dim-light protocol over 3 days with equal time for sleep and wakefulness (Pavlova et al., 2009). IEA increased dramatically during NREM sleep in all individuals regardless of the circadian time, highlighting the permissive effect of sleep on IEA.
Beyond the presumed permissive role of sleep in promoting seizures in nocturnal epilepsy, clinical evidence indicates that the majority of individuals (~90%) with any epilepsy syndrome, nocturnal or diurnal, demonstrate peak seizure incidence at particular times of day, with substantial variability, but also shared patterns across individuals, a “seizure chronotype” (Langdon-Down and Brain, 1929; Patry, 1931; Karoly et al., 2018; Leguia et al., 2021).
Historically, Gowers divided the specific timing of seizures into three chronotypes: nocturnal, diurnal, and diffused, the latter showing no particular preference for a time of day. This typology was soon complemented by the addition of a “rising” seizure chronotype, in which seizures occur shortly after awakening (Hopkins, 1935; Griffiths and Fox, 1938).
On the basis of observational and experimental data, variability in the timing of peak seizure incidence across individuals is likely to be driven by endogenous circadian factors with additional contribution of variation in the timing of specific behaviors, such as sleep, eating meals, or taking medication (Langdon-Down and Brain, 1929; Patry, 1931; Griffiths and Fox, 1938). Seizures are more common around sleep-wake transitions, so seizure chronotypes with late evening and/or early morning peaks have been observed at the population level (Fig. 15–2, Hofstra et al., 2009; Anderson et al., 2015; Leguia et al., 2021). However, evidence indicates that some individuals with mostly diurnal seizures have peak seizure incidence in the afternoon (Langdon-Down and Brain, 1929; Leguia et al., 2021). Individuals with sleep-related epilepsy can have a peak seizure time soon after falling asleep (typically before midnight) or in late sleep (typically at dawn, Langdon-Down and Brain, 1929; Patry, 1931), which are not likely to be fully explained by the differential effects of NREM and REM sleep on seizures.
Analysis of chronic EEG data indicates that individual seizure chronotypes are fairly stable over months to years, although some people with epilepsy have a weakening of or “switches” in preferential seizure time (Rao et al., 2020). For example, some individuals who previously experienced only sleep-related seizures can begin to experience wake-related seizures, a spontaneous change that can have dramatic practical consequences for day-to-day life (Rao et al., 2020). Conversely, some individuals with awakening epilepsies experience an increasing frequency of sleep-related seizures after years with epilepsy (Janz, 1962). Some evidence suggests that pediatric and adult populations show different seizure chronotypes (Hofstra et al., 2009; Goldenholz, Rakesh, et al., 2018), which could be a result of age-dependent differences in sleep-wake cycles or hormonal and behavioral factors, but they could also reflect the development of intrinsic circadian regulatory mechanisms in the brain.
In addition to timing seizures, the circadian rhythm may also influence seizure propagation in variable brain networks (Quigg and Straume, 2000; Lagarde et al., 2019; Salami et al., 2020; Schroeder et al., 2020), potentially resulting in different symptoms and/or severity at different times (Loddenkemper et al., 2011; Sánchez Fernández et al., 2013; Goldenholz, Rakesh, et al., 2018). For example, clinical data show that secondary generalization of focal seizures is more likely during the night and during sleep as compared to wakefulness, as if the sleeping brain was prone to amplify seizures. Quantitative studies characterizing ictal and interictal networks confirm that propagation pathways differ at different times (Burns et al., 2014; Ung et al., 2016; Gliske et al., 2018; Saggio et al., 2020; Schroeder et al., 2020) and some seizure patterns recur with circadian rhythmicity (Schroeder et al., 2020).
Paralleling the high prevalence of circadian modulation of seizures in humans, epilepsy models in diurnal (dogs) and nocturnal animals (rodents) all show preferential seizure timing, although at different times in different models. In some dogs with naturally occurring epilepsy, seizures occur with individual preferential timing in the evening or morning, similarly to humans (Gregg et al., 2020). In rats, seizures tend to occur in the resting phase (day), independent of the model chosen (Quigg et al., 1998, 2000; Hellier and Dudek, 1999; Raedt et al., 2009; Tchekalarova et al., 2010; Baud et al., 2019). In mice, seizures tend to occur at the transition to the active phase (night) in the kainate (Pitsch et al., 2017), electroshock (Purnell et al., 2021), and audiogenic seizures models (Schreiber and Schlesinger, 1971; Purnell et al., 2021) and in the resting phase (day), and more specifically sleep in different transgenic models (Fenoglio-Simeone et al., 2009; Wright et al., 2016; Li et al., 2017).
Taken together, clinical and experimental observations of daily patterns in individuals with epilepsy suggest that hourly IEA is mostly modulated by the sleep-wake cycle, whereas seizures seem to be modulated by both the circadian cycle and the sleep-wake cycle. This modulation results in a few seizure chronotypes, which are sometimes associated with specific epilepsy syndromes. Therefore, two key differences exist between the circadian rhythmicity of seizures and IEA. First, unlike IEA, which has a similar circadian modulation day after day, seizures do not occur during every circadian cycle, suggesting that phasic alterations create permissive but insufficient conditions for seizure occurrence. Second, no general relationship has been identified between the circadian rhythms of seizures and IEA. Evidence indicates that seizures can occur preferentially near the peaks or the troughs of circadian IEA cycles, depending on the individual (Hofstra and de Weerd, 2009; Karoly et al., 2016; Baud et al., 2018), which perhaps explains why IEA has been reported to increase before or after seizures (Gotman and Marciani, 1985; Janszky et al., 2001; Blumenfeld et al., 2004; Spencer et al., 2008; Krishnan et al., 2014; Karoly et al., 2016). Rapid changes in IEA are likely to reflect changes in vigilance states, which means IEA is most informative as a proxy for “brain irritability” when averaged over a full sleep-wake cycle. In some individuals, the circadian processes governing IEA and ictogenesis might converge, whereas in others, a circadian rise in IEA might coincide with an increase in seizure threshold.
Mechanistic investigations of circadian seizure timing started in humans in the 1930s, soon after Langdon-Down’s report. Early investigations included measurements around the clock of the cerebrospinal fluid (CSF) pressure (Denny-Brown and Graeme Robertson, 1934), and chemical factors such as cholesterol or blood pH (Hopkins, 1935). A role of sleep was also suspected (Magnussen, 1936).
The quasi-experimental setting of the epilepsy colonies can still teach us much about chronobiological regulation of the time of seizures, because many experiments, which were part of the routine of the time, could not be reproduced in today’s society. For example, a 1-hour shift in the regular onset of sleep was sufficient to shift the peak seizure time by 1 hour in a person with sleep-related epilepsy (Griffiths and Fox, 1938), suggesting that the sleep-wake cycle, not the circadian cycle, governs seizure timing in such individuals.
Chronobiological demonstrations in animal models are also highly valuable, again because some of the required manipulations could not be done with patients. In the late 1990s, Mark Quigg and colleagues demonstrated the endogenous nature of the modulation of seizure timing as epileptic rats kept in constant darkness for 2 weeks retained strong and unchanged preferential circadian timing of seizures (Fig. 15–2l,m; Quigg et al., 1998, 2000). This observation rules out the direct effect of daylight as the only explanation for preferential circadian timing of seizures, which is important as some seizures are photosensitive (Danesi, 1988). This landmark chronic experiment was never replicated in humans with epilepsy for obvious ethical reasons, but these animal data lend credence to the idea that human seizures also are under endogenous circadian influence. The circadian influence on IEA and seizures is less pronounced in rodent models of epilepsy than in humans with the disorder (Fig. 15–2; Quigg et al., 1998; Pitsch et al., 2017; Baud et al., 2019), likely owing to the presence of a polyphasic sleep-wake cycle in rodents, although studies measuring both sleep and circadian effects in rodents are lacking to confirm this point. Some experimental evidence further suggests that cortical excitability in rodent models of epilepsy, as assessed by measurements of electrical (Gerstner et al., 2014) and chemical (Stewart et al., 2001) seizure thresholds and cortical responses to short electrical pulses (Matzen et al., 2012), varies according to time of day. Importantly, studies using transcranial magnetic stimulation (TMS) in healthy volunteers have also found that cortical excitability is modulated over the course of 24 hours (Huber et al., 2013; Ly et al., 2016). Such circadian changes in cortical excitability might facilitate ictal transitions in epilepsy and result from an interplay between sleep and circadian processes at the cortical level.
Although both epileptic brain activity and the blood levels of some hormones show circadian fluctuations, studies on a potential causal role of hormones in determining seizure circadian timing are currently lacking. Nevertheless, studies of plasma hormone concentration in individuals with epilepsy can provide further insight into circadian cycles of IEA and seizures. In healthy individuals, melatonin production increases in the evening, providing a robust marker of the circadian phase in humans, although production can also be affected by other factors, including diet (Hofstra and de Weerd, 2009). Some studies have found baseline melatonin levels to be higher in individuals with epilepsy than in healthy individuals (Molina-Carballo et al., 1994; Schapel et al., 1995) and other studies have found the inverse (Laakso et al., 1993; Bazil et al., 2000; Yalýn et al., 2006; Dabak et al., 2016). However, compared with baseline levels, consistent increases in melatonin levels after seizure occurrence have been observed (Molina-Carballo et al., 2007; Hofstra and de Weerd, 2009; Dabak et al., 2016), supporting the concept of bidirectional interactions between epilepsy and the circadian system. Interestingly, animals lacking a pineal gland present a more severe course during epileptogenesis (Rudeen et al., 1980; De Lima et al., 2005), whereas animals receiving exogenous melatonin have a milder course (Rudeen et al., 1980; Petkova et al., 2014). After failed clinical trials, the concept of melatonin as an add-on treatment for epilepsy has not translated into clinical practice, in part due to low-level clinical evidence (Brigo and Igwe, 2016).
The levels of cortisol, another circadian hormone, in saliva have been found to parallel epileptic brain activity in a subset of people with “stress-sensitive” seizures (Van Campen et al., 2016; Heijer et al., 2018). The blood levels of several other hormones, including prolactin and growth hormone, are altered after seizures (Pritchard, 1991), possibly contributing to a feedback loop between the endocrine and the nervous system. Again, parallel measurements of variables known to undergo circadian fluctuations are bound to reveal correlations, whether these hormones play a causal role in modulating seizure likelihood is a different question.
Biochemical factors other than hormones, such as neurotransmitters, amino acids, and mediators of inflammation, may play a role in circadian seizure cycling. Brain concentration of some neurotransmitters, including serotonin (Schreiber and Schlesinger, 1971), acetylcholine (Kametani and Kawamura, 1991), glutamate (Volk et al., 2018), GABA (Lang et al., 2011), and their receptors (Kafka et al., 1986), has been shown to undergo circadian regulation. Experiments involving intracerebral microdialysis and concurrent EEG in humans with drug-resistant epilepsy revealed that, during epileptic seizures, extracellular glutamate is elevated in areas of the brain involved by the seizures (During and Spencer, 1993; Gruenbaum et al., 2019). Similarly, in a rodent model of mesial temporal lobe epilepsy, 24-hour oscillations of extracellular glutamate were observed in the epileptogenic hippocampus (Sandhu et al., 2020), though the functional significance of these cycles remains unclear. At the cellular level, neuronal excitability may also be regulated by circadian changes in the redox state (Naseri Kouzehgarani et al., 2020). At the systemic level, blood concentrations of metabolites such as ketone bodies (Buchhalter et al., 2017), glucose (Wright et al., 2019), and branched-chain amino acids (BCAAs; leucine, isoleucine, valine; Gruenbaum et al., 2019) might influence seizure susceptibility, and dietary modifications (e.g., ketogenic diet) can be powerful antiseizure therapies (Allen, 2008). In particular, BCAAs regulate a variety of biological processes and may play a central role in cerebral metabolism, including the synthesis of glutamate and other neurotransmitters. Moreover, supplementation with BCAAs in a small pediatric cohort seemed to reduce seizure rates (Evangeliou et al., 2009). A transcription factor, Krüppel-like factor 15 (KLF15), is a critical regulator of BCAA metabolism and under the control by clock genes (see below), to impart diurnal rhythmicity to the release, uptake, and utilization of BCAAs (Fan et al., 2018). Concentrations of these metabolites in the blood and CSF display substantial, rapid (minutes to hours) fluctuations in animal and human studies, but the existence of underlying longer circadian rhythms remains underexplored.
The individual timing of seizures could be linked to the molecular machinery responsible for circadian changes in neuronal dynamics (Cho, 2012). The suprachiasmatic nucleus (SCN) lies in the hypothalamus and acts as the body’s master circadian clock. In the SCN, key molecular relays, including the dimerized transcription factors CLOCK/BMAL1 (positive feedforward), and the negative feedback by the heterodimers PER1/2, CRY1/2, and REV-ERB/ROR, drive local transcription-translation loops with far-reaching influences on almost all organs (Bass and Lazar, 2016), cells (Reppert and Weaver, 2002), and microbiota (Teichman et al., 2020). Beyond the SCN, circadian transcription-translation feedback loops seem to be ubiquitous in the brain, although different genes undergo circadian expression in different brain regions (Mure et al., 2018). Molecular analyses at fine spatial scales in the brains of healthy rodents identified circadian fluctuations in 70% of the synaptic transcriptome, proteome, and phosphorylome in the forebrain (Brüning et al., 2019; Noya et al., 2019). As a result, circadian fluctuations in cortical excitability could result from synaptic changes in neurotransmission (Kafka et al., 1986) and ionic conductance (Pracucci et al., 2021). Sleep and wake states could have a direct effect on cortical excitability through similar molecular pathways, as gene expression differs among vigilance states (Brüning et al., 2019; Noya et al., 2019). Thus, the molecular landscape of the brain undergoes specific cycles as a function of circadian time, brain states, brain regions, and likely neuronal type, forbidding a monolithic and static view of neuronal networks in the healthy (Mure et al., 2018) or epileptic brain (Debski et al., 2020). Furthermore, in the epileptic brain, bidirectional influences are theoretically possible: (1) on one hand, oscillating molecular changes may play a role in seizure timing; (2) on the other hand, seizures themselves could influence gene expression.
Molecular studies of human brain parenchyma obtained from epileptogenic tissue resected under general anesthesia (e.g., in mesio-temporal resections) initially revealed greatest changes in expression of genes related to GABAergic neurotransmission, signal transduction (MAP-kinase pathways), and lipid metabolism relevant to oligodendrocytes (Arion et al., 2006). Subsequent studies focusing on the expression of clock genes (mostly from focal cortical dysplasia and tubers; Li et al., 2017) revealed a decrease in CLOCK, as well as downstream cryptochromes (CRY1/2), periods (PER1/2), and DBP regulator proteins, which are all under modulation by the BMAL-1/CLOCK heterodimer (BMAL-1 levels remained unaltered; Li et al., 2017). Other human studies specifically evaluated the expression of REV-ERBα, which has recently attracted attention as a druggable nuclear receptor at the crossroads of the molecular clock and the regulation of the metabolism and inflammation (Kojetin and Burris, 2014). Results in human epilepsy, however, were contradictory with one study showing an increase (Zhang et al., 2021) and another study showing a decrease (Yue et al., 2020) in expression, highlighting the limitations of single timepoint measurements.
Similar transcriptomic studies carried out in animal models have several advantages over human tissue collection: (1) tissue can be collected after a minimal duration of anesthesia, which is a concern when measuring dynamical changes in transcription, (2) tissue can be collected at different timepoints of the circadian cycle, and (3) control tissue can be collected in different animals or outside of the epileptogenic tissue. In the hippocampus and neocortex of rodents receiving different types of acute electro- and chemo-convulsive seizures, the circadian regulation of many core clock genes was transiently altered with decreases, increases, or lost rhythmicity of individual genes (Table 15–1; Kim et al., 2018; Matos et al., 2018; Rambousek et al., 2020; Yue et al., 2020; Zhang et al., 2021). A large study of the expression of core clock genes at four timepoints in the hippocampus of five different models of TLE in rats and mice revealed altered expression only in the acute status-epilepticus phase, but resumed their rhythmicity thereafter (latent and chronic phase of these models; Rambousek et al., 2020). Another large transcriptomic study of the mouse (chronically) epileptic hippocampus (Debski et al., 2020) showed amplified circadian oscillations in all core clock genes (out of phase for some), as well in other sets of transcripts. Interestingly, some of these transcripts oscillate in epilepsy, but not in healthy animals, and others oscillate in healthy animals but are stable in epilepsy. Although these studies are difficult to reconcile (Table 15–1), two key points can be made: (1) massive acute seizures can alter the expression of clock genes; (2) clock genes do oscillate in chronic epilepsy, albeit sometimes with different phases, and could thus govern seizure timing. “Reprogrammed” local oscillations of the molecular architecture in the epileptic parenchyma may be one key to understanding cyclicity in epilepsy.
Studies of Clock Genes in Epilepsy.
The causal effect of clock genes on seizure timing is best investigated with transgenic models manipulating clock genes. To date, only a few knock-out models for clock genes, including clock, bmal-1, Rev-erbα, and PAR bZip, have been used in the study of epilepsy, but these important studies point to a key role of the molecular clock in timing seizures: (1) mice with a targeted deletion of the clock gene in pyramidal neurons have spontaneous seizures, predominantly during sleep, suggesting that pyramidal neurons act in an uncoordinated way at times when they should sleep and synchronize with neighboring cells (Li et al., 2017); (2) epileptic mice with a constitutive deletion of bmal-1 lose the circadian modulation of seizure susceptibility (Gerstner et al., 2014); (3) similarly, ablation or antagonism of Rev-erbα also leads to a loss of circadian changes in the severity of chemically induced seizures (Zhang et al., 2021). Finally, a triple-knock-out for the murine PAR bZip transcription factor family led to severe audiogenic seizures, although their circadian timing was not investigated (Gachon et al., 2004).
The circadian rhythmicity of IEA and seizures is thus grounded in the oscillatory molecular landscape of the brain (Bernard, 2021), perhaps via large changes in hormonal or local, bidirectional signaling, affecting neurotransmission, and cortical excitability during different circadian phases and vigilance states. This body of work is important from a mechanistic standpoint because it investigates the causality of factors regulating seizure timing. However, it remains incomplete and suffers from weaknesses, such as the use of models of acute seizure inductions (chemically or electrically), which must be completed by the study of the timing of occurrence of spontaneous seizures in chronic epilepsy.
A rapidly growing body of evidence supports the role of inflammation in epileptogenesis (Vezzani et al., 2011), but more work is needed to determine how inflammatory mediators may also influence the timing of individual seizures. The immune and circadian systems interact with each other (Yamakawa et al., 2020), and the immune response, including pro-inflammatory cytokines, can alter circadian rhythms in the brain. Interleukin-1β (IL-1β), a neuroinflammatory cytokine, has been closely linked with epileptogenesis and can disrupt the circadian clock in various tissues (Javeed et al., 2021).
An emerging regulator of neuronal function, in particular cell excitability, is the lymphatic system that controls the flow of CSF in the brain during the sleep-wake cycle (Rasmussen et al., 2021). Importantly, neuromodulators can rapidly alter the concentration of ions in the CSF between wake and sleep, thus directly switching neuronal activity between higher and lower levels of activity (Ding et al., 2016). It will be interesting to test whether a circadian regulation of these modulators is phased to the circadian distribution of seizures.
In summary, the circadian distribution of seizures is embedded in a changing landscape of cortical excitability, which likely represents the final common pathway of a constellation of other influencing factors, some local, such as the interstitial neurochemistry, or the neuronal molecular machinery, while others act a distance, including inflammation, hormones, and other factors.
More recently, building upon seminal observations made in the epilepsy colonies, converging evidence from different trials of chronic implanted EEG system yielded a very intriguing result: the existence of multidien cycles in epilepsy, which are periodic fluctuation in epileptic brain activity, leading up to overt seizures.
Seizures often occur periodically in clusters interleaved with symptom-free intervals (Fig. 15–2b-e); about-monthly cycles in seizure occurrence (Bercel, 1964) have been termed circalunar (Quigg et al., 2008), or, when related to the female menstrual cycle, catamenial (Laidlaw, 1956; Herzog, 2008). However, historical studies in institutionalized cohorts (Fig. 15–2a,b) identified about-monthly cycles of seizure occurrence in men and women, children, and adults, and these cycles were not limited to catamenial epilepsy or an exact relationship to the lunar cycle (Griffiths and Fox, 1938). In subsequent studies based on self-reporting of seizures (Fig. 15–2c), about-monthly cycles of seizures were consistently found (Bercel, 1964), with additional periods of 1 or 2 weeks (Bercel, 1964), including in catamenial epilepsy (Quigg et al., 2008). Other studies of seizure records, using statistical analyses borrowed from the field of physics, identified the presence of long-range dynamics governing the duration of intervals between seizures (Binnie et al., 1984; Milton et al., 1987; Osorio et al., 2009; Cook et al., 2014). Analysis of a very large set of online seizure diaries also identified a weak influence of the day of the week on the frequency of self-reported seizures (Ferastraoaru et al., 2018; Karoly et al., 2018). In the same cohort, approximately one-quarter of individuals reported seizure cycles of longer than 3 weeks, which were not aligned to lunar cycles (Karoly et al., 2018). These self-reported cycles had a similar prevalence across different epilepsy syndromes and seizure types (Karoly et al., 2018).
In addition to the observed periodic intervals between seizures, analyses of cEEG data from individuals with epilepsy found that IEA fluctuates with similar periodicity (Fig. 15–2d,e; Karoly et al., 2016; Baud et al., 2018; Maturana et al., 2020). This multidien periodicity in seizures and IEA has been identified in the cEEG data of most individuals (~60%) that have been studied to date (Karoly et al., 2016; Baud et al., 2018; Maturana et al., 2020; Rao et al., 2020; Leguia et al., 2021). Generally, multidien cycles in epilepsy do not align with calendar days or other cycles of fixed period length; rather, their period length fluctuates around a central frequency (Baud et al., 2018; Rao et al., 2020; Leguia et al., 2021). Moreover, complex rhythms comprising multiple component periods are often observed (Baud et al., 2018). The average periodicity varies individually, but group trends (multidien seizure chronotypes; Leguia et al., 2021) show an about-monthly periodicity of ~20–35 days (Baud et al., 2018), which is often accompanied by more rapid cycles of 14–15 days and 7–10 days (Baud et al., 2018; Leguia et al., 2021). Although periodicity varies across individuals, a given pattern is typically robust over years within an individual (Baud et al., 2018; Rao et al., 2020).
Multidien rhythms of IEA and seizures have also been observed in animal models of epilepsy (Fig. 15–2f,g). The results of a cEEG study in male rats with epilepsy showed multidien cycles of IEA recorded over several weeks without intervention (Fig. 15–2g; Baud et al., 2019). These cycles lasted 6 days on average and were not aligned with calendar days. Importantly, multidien rhythms were not synchronous across animals housed together, suggesting that the cycles were not driven by a shared environment (Baud et al., 2019). Recent work using ambulatory cEEG in canines with naturally occurring epilepsy also found that circadian and multidien (approximately weekly to monthly; Fig. 15–2b-e) seizure periodicities were common (Gregg et al., 2020). Canine seizure cycles occurred independent of antiseizure medication schedules and were asynchronous across animals. Evidence indicates that multidien cycles are present in ~60% of individuals with focal epilepsy and are as strong as circadian influences (Baud et al., 2018; Leguia et al., 2021), supporting the long-standing view that seizures represent the “tip of the iceberg” of the complex and varied activity that occurs in epileptic networks over multiple timescales.
Seizure periodicity is best understood by examining the relationship between seizure timing and cycles of IEA at long timescales: cycle after cycle, seizures tend to occur on the rising phase of multidien cycles of average daily IEA (Fig. 15–2f). The observation of this phasic relationship (see Fig. 15–1 and Fig. 15–2h), which holds true for cycles spanning 6–45 days and is consistent across individuals (Baud et al., 2018), helps settle a long-standing debate about the relationship between IEA and seizures. Previous studies relied on short-timescale recordings and reported that IEA increases, decreases, or remains unchanged before seizures (Gotman and Marciani, 1985; Krishnan et al., 2014; Karoly et al., 2016), and changes in IEA after seizures were also variable across individuals (Gotman and Marciani, 1985; Janszky et al., 2001; Spencer et al., 2008). In our opinion, these apparent discrepancies can be reconciled by considering the timescale of recording and the intermingling of circadian and multidien cycles of IEA: seizures tend to occur when average daily IEA rises over days, so daily IEA tends to increase both before and after seizures (Karoly et al., 2016; Baud et al., 2018). By contrast, seizure timing in relation to circadian cycles of hourly IEA is more variable across individuals, and the preferred phase could be near the peak or the trough of these shorter cycles (Karoly et al., 2016; Baud et al., 2018). Thus, studies examining hour-to-hour IEA trends in individuals with different circadian seizure chronotypes could draw seemingly contradictory conclusions. In sharp contrast, more recent studies that recorded IEA over longer time periods consistently observed single seizures or clusters of seizures when IEA rose over several consecutive days (Baud et al., 2018, 2019; Maturana et al., 2020). This observation was initially made in a cohort of patients implanted with an EEG recording and neurostimulation device (Baud et al., 2018) and later confirmed in data from participants with an implanted EEG recording device (Maturana et al., 2020), as well as in rats (Baud et al., 2019) and dogs (Gregg et al., 2020) with epilepsy. Thus, these independent studies, based on data from different recording devices in different organisms, reached the same conclusion, supporting the veracity of the finding. Nevertheless, these studies defined IEA in several different ways, and the recording techniques used emphasized different epileptiform waveforms (e.g., interictal spikes vs high-frequency activity). Future work should focus on unravelling the intricacies of the spatiotemporal relationship between distinct epileptiform discharges and seizure cycles.
In contrast to the circadian influences on epilepsy, where photic inputs act as zeitgeber for neuronal firing, the statistical characterization of multidien cycles indicates the lack of strong influence of environmental cues (Rao et al., 2020). Multidien cycles of IEA and seizures do not strictly synchronize with days of the week or month, or lunar phases, but can occasionally align with these cycles of fixed period length (Rao et al., 2020; Leguia et al., 2021). Therefore, multidien cycles seem to stem from endogenous mechanisms with a free-running period, which perhaps explains why these cycles were difficult to study before the introduction of chronic EEG. Multidien cycles have been described in other areas of medicine, including psychiatry (Benedetti et al., 1996; Wehr, 2018), oncology (Coventry et al., 2009), and cardiology (Zoghi et al., 2008), suggesting a systemic basis for these cycles. Systemic multidien oscillations are most common at approximate weekly and monthly periodicities (Touitou and Haus, 1992), which is similar to observed cycles of IEA and seizures (Leguia et al., 2021). For instance, limited human studies have identified free-running weekly and monthly cycles in cortisol, ketosteroid, and immune proteins, as well as heart rate and blood pressure (Touitou and Haus, 1992). One study in adults with epilepsy found multidien cycles of resting heart rate were correlated with seizure occurrence (Karoly et al., 2020).
Mechanisms of multidien rhythms in epilepsy and more broadly in medicine are virtually unknown. Large cohort studies collecting biospecimen in conjunction with physiological data are needed and unravelling the physiological drivers of multidien cycles in epilepsy will remain speculative until then. We nevertheless discuss two possible mechanisms underlying long cycles in epilepsy in the following section.
Current evidence of a hormonal basis for seizure cycles is inconclusive. Studies in large cohorts of healthy women of child-bearing age have established that most have a consistent 29-day-long menstrual cycle, but some have a more variable cycle length averaging ~35 days (Li et al., 2020). In many women with epilepsy, seizure timing co-occurs with menstruation (about monthly) or with both menstruation and the follicular phase (about biweekly; Herzog, 2008, 2015). These important clinical observations are concordant with the observation that female sex hormones and/or their derivatives (i.e., the brain-active metabolites) can modulate the seizure threshold (Harden and Pennell, 2013) by acting on neuronal receptors, for example, the GABA-A receptor (Dorota et al., 1986). More specifically, a long-standing hypothesis based largely on experimental evidence from rodent models stipulates that estradiol is proictal and progesterone anti-ictal (Harden and Pennell, 2013). A small number of animal studies have identified an important role for ovarian hormones in seizure susceptibility in female rats (D’Amour et al., 2015) and mice (Maguire et al., 2005). A proposed mechanism involves estrous cycle-dependent fluctuation in the expression of δ subunit-containing GABA-A receptors that mediate tonic inhibition in the hippocampus (Maguire et al., 2005). Conversely, a landmark clinical trial in women with epilepsy found no difference in seizure frequency between those receiving progesterone therapy and those receiving placebo, although subsequent analysis identified a subpopulation of participants (~20% of the cohort) in whom treatment showed some benefit (Herzog et al., 2012; Herzog, 2015). This modest response rate to progesterone therapy suggests that the modulation of seizure occurrence by female sex hormones might be less prevalent than previously thought.
Fluctuations in female sex hormones alone cannot explain clinical observations that about-monthly seizure cycles are present in children (Griffiths and Fox, 1938), men (Griffiths and Fox, 1938; Baud et al., 2018; Leguia et al., 2021), and postmenopausal women with a similar prevalence as in premenopausal women. Whether or not males have multiday cycles in hormone levels remains an open question owing to a paucity of evidence. However, the results of small studies in men suggest that testosterone levels fluctuate with a ~20–30-day periodicity (Celec et al., 2002) and that estradiol fluctuates with a 12-day periodicity (Celec et al., 2006). Evidence also indicates that aldosterone fluctuates with a periodicity of 7 days in men (Rakova et al., 2013) and that corticosterone fluctuates with a similar periodicity in male rodents (Jozsa et al., 2005). On the basis of its effects on neuronal excitability and its correlation with IEA fluctuations in individuals with stress-sensitive seizures, cortisol is a candidate hormonal modulator of the epileptic brain (Van Campen et al., 2016; Heijer et al., 2018). In a survey of 89 individuals with epilepsy, the majority felt that stress increased the frequency of their seizures (Haut et al., 2003); however, stress is typically considered to be a noncyclical precipitating factor in epilepsy.
A brain-centric hypothesis for seizure cycle mechanisms, which provides an alternative to the hormonal hypotheses described above, involves a weak coupling between the epileptic network and the rest of the brain. Indeed, a modern conception holds that epilepsy is a network disorder (Scott et al., 2018; Davis and Morgan, 2020), with seizures resulting from the dynamic topology of interconnected brain regions, including those remote to the classical seizure-onset zone (Rings et al., 2019). Recent analyses of days-long EEG recordings provide support for the possibility that time-varying reconfigurations of large-scale brain networks help determine preictal periods (Fruengel et al., 2020). Furthermore, periodic fluctuations in functional brain networks—of which IEA may be one easily measured manifestation—help determine the timing of seizures (Mitsis et al., 2020). Similarly, markers of global network resilience can be used to track seizure susceptibility and show individual-specific multidien rhythms (Chang et al., 2018; Maturana et al., 2020). Remarkably, within an epileptic network, the spatiotemporal evolution of seizures itself demonstrates cyclical variation over circadian and slower timescales (Schroeder et al., 2020). Collectively, these studies indicate that seizure-generating brain networks have dynamic, multiscale properties that may underlie the cyclical occurrence of seizures and specific seizure pathways.
An interconnection between circadian and multidien cycles in epilepsy is furthermore possible. Modeling studies showed progressive drift in the phase of biological oscillators operating near the circadian entrainment domain, which can lead to the development of longer oscillations, suggesting that multidien cycling might be an emergent property of quasiperiodic shorter (about-daily) oscillations (Leloup and Goldbeter, 2008). Fundamentally, this would mean that epileptic networks are out of pace with other brain networks that operate closer to a physiological regime, resulting in a rhythmic interplay between the two at short (circadian) and long timescales (multidien).
Circannual cycles can drive profound physiological and behavioral changes in animals and humans (Foster and Roenneberg, 2008). Seasonal changes in seizure frequency have been observed in individuals with epilepsy (Motta et al., 2011). Instances of individuals experiencing a single seizure per year, always during the same month, have been described anecdotally (Griffiths and Fox, 1938). Even when seizures occur more than once per month, the seasons could have a modulatory effect resulting in increases in seizure rates during specific months of the year. For example, by tracking seizures as an admission diagnosis, or seizure rates on inpatient epilepsy wards, multiple studies have identified a peak occurrence of seizures in either winter (Baxendale, 2009; Brás et al., 2018) or summer (Alexandratou et al., 2020), although these cross-sectional studies provide limited insight into the long-term effects of seasonal factors in a natural environment at the level of the individual patient. Longitudinal analysis of self-reported seizure diaries from a small cohort of individuals with epilepsy identified a weak increase in seizure frequency during winter (Ünsal et al., 2020). However, similar analyses in two large cohorts have found no consistent effect of season on seizure frequency across individuals (Karoly et al., 2018; Leguia et al., 2021), but some individuals (~10%) nevertheless reported substantially more seizures in a given season compared with the rest of the year (Fig. 15–3; Griffiths and Fox, 1938; Leguia et al., 2021), underscoring the notion that seizure cycles have individual-specific characteristics at all timescales. Overall, circannual cycles are likely to be minor modulators of seizure timing, especially in comparison to other cycles discussed here.
Seasonal seizure cycles. Circular histograms derived from self-reported seizure times of four individuals in the NeuroPace cohort showing different preferential seasons.
Limited data exist on the mechanisms underlying circannual cycles in humans. A recent cross-sectional study showed a weak influence of time of the year on measured hormone concentrations in humans (Tendler et al., 2021). Seasonal variation in human cognition, assessed with neuropsychological testing and cognitive task-based functional magnetic resonance imaging (MRI), has been observed (Meyer et al., 2016; Lim et al., 2018), indicating that brain physiology can change with seasons, although seasonal modulation at the level of the individual participant was not described in these studies.
In addition to possible circannual physiological changes, external influences might also have a role in circannual seizure cycles. Evidence suggests that the weather can influence seizure likelihood (Motta et al., 2011), with one study showing that low atmospheric pressure and high relative air humidity were associated with an increased risk of epileptic seizures, whereas high ambient temperatures were associated with a decreased seizure risk within a population (Rakers et al., 2017). Daylight duration might also have a role in modulating seizure risk, either via an influence on sleep quality, the circadian cycle (Rao et al., 2020), or other effects of sunlight, the latter based on one observation that seizures on inpatient epilepsy wards are less likely to occur on sunny days (Baxendale, 2009). To our knowledge, longitudinal mechanistic studies of circannual cycles in epilepsy have not yet been performed, but at least one study in humans has involved serial scalp EEG recordings during different seasons (Danesi, 1988), and another identified an association between the seasons and the threshold for chemoconvulsant-induced seizures in mice (Löscher and Fiedler, 1996).
The study of chronobiology in epilepsy is undergoing a renaissance. Modern quantitative methods and emerging technologies have validated and extended clinical observations that have been made since antiquity. The concept of time-varying seizure risk provides a framework for understanding the temporal structure of epilepsy and expanding the search for seizure precursor signatures in EEG recordings. Patterns of seizure occurrence are apparent over wide-ranging timescales—including years, seasons, days, and hours—and are increasingly understood to be multifactorial, involving genetic, hormonal, environmental, and behavioral factors. Among neurological disorders, epilepsy is arguably one of the most dynamic, and conventional static therapies are evolving to reflect this aspect. The implications of our increasing knowledge of cycles in epilepsy are far-reaching and include the development of devices that can track cycles efficiently, optimization of clinical trial design, reduction of unpredictability through seizure forecasting, and use of risk-stratified chronotherapy.
How cycles of seizures relate to their localization in the brain remains a topic for future research. At the circadian timescale, it is well established that frontal lobe epilepsy tends to generate sleep-related seizures (van Campen et al., 2015), whereas genetic epilepsies generate seizures at the sleep-wake transition (Carreño and Fernández, 2016; Xu et al., 2018). Temporal lobe epilepsies can lead to any circadian preference, and data are scarce for parietal, occipital, and insular epilepsy (van Campen et al., 2015). Although, multidien periodicities did not relate to given focal epilepsy location in a recent study (Leguia et al., 2021), data were insufficient for a definitive conclusion in subgroup analyses. Additionally, all studies employing chronic EEG to capture multidien cycles have so far been done in patients with focal epilepsy; therefore, multidien periodicities in generalized epilepsy are less well-known.
For individuals with epilepsy, many difficulties arise from the apparent unpredictability of seizures, and a fundamental challenge in epilepsy treatment is to forecast seizures (Janse et al., 2019). Methods that can anticipate seizure onset would enable the development of novel preventative treatments and could lead to improvements in quality of life for individuals with epilepsy (Baud and Rao, 2018). The existence of seizure cycles at multiple timescales has now enabled the temporal organization of seizure likelihood to inform probabilistic forecasts (Karoly, Rao, et al., 2021). This approach, akin to weather forecasting, enables information from multiple sources and at multiple timescales to be leveraged to estimate seizure probability with increasing precision over days, hours, and seconds (Karoly, Rao, et al., 2021). Early studies explored the use of individuals’ circadian cycles of seizure risk in predictive algorithms (Schelter et al., 2006) and a seminal trial demonstrated the prospective accuracy of seizure forecasting (Cook et al., 2013) that could be improved by incorporating circadian seizure cycles (Karoly et al., 2017). More recently, a few key studies have validated algorithms that explicitly track the phase of cycles at multiple timescales to extrapolate upcoming periods when seizures are most likely, that is, when the highest-risk phases of multiple cycles align, with promising results (Baud et al., 2018; Maturana et al., 2020; Proix et al., 2020). Notably, the discovery of multidien cycles of epileptic brain activity and its characterization at the individual level holds the promise of being able to forecast seizures over days, a future horizon that was previously unthinkable (Proix et al., 2020). In addition to intracranial EEG, it may be possible to track multiday cycles of seizure likelihood using seizure diaries (Goldenholz et al., 2020; Karoly et al., 2020), wearable sensors, and possibly, minimally invasive subscalp EEG recording devices (Duun-Henriksen et al., 2020). However, the practical utility of probabilistic forecasts based on multiday seizure cycles requires prospective validation within large-scale clinical trials, and widespread uptake within clinical settings is likely to require further conceptual shifts, particularly in the assessment and interpretation of forecasting performance (Chiang et al., 2021).
Meanwhile, for clinicians, cycles in epilepsy have important implications for routine diagnostic procedures. For example, the yield of short-timescale recordings of brain activity, such as brief outpatient EEG or days-long inpatient video-EEG studies, likely depends heavily on the timing of these procedures in relation to the underlying cycles of brain activity (Baud and Rao, 2018; Baud et al., 2021; Karoly, Eden, et al., 2021). The nature and timing of clinical therapies might also be informed by cycles in epilepsy. By definition, effective therapies reduce seizure risk, but whether this reduction involves the lengthening of seizure risk cycles, such that seizures still occur cyclically but with reduced average frequency, or the dampening of seizure risk cycles, such that critical thresholds are more difficult to cross (Baud et al., 2020), remains unknown.
Most of the current treatments for seizures—such as taking the same doses of the same antiseizure medication each day—are relatively static, contrasting sharply with the dynamic nature of epilepsy and potentially exposing people with epilepsy to unnecessary adverse effects when therapies are administered during periods of low seizure risk. Chronotherapy refers to the adjustment of treatment on the basis of temporal changes in seizure risk (Ramgopal et al., 2013; Sánchez Fernández and Loddenkemper, 2018). For example, circadian chronotherapy can involve administering higher doses of medication at bedtime for individuals with epilepsy who have primarily nocturnal seizures (Thome-Souza et al., 2016). Some women with catamenial epilepsy benefit from taking extra antiseizure medication during times of their menstrual cycle when seizures are most likely (Herzog et al., 2012). Treatment of catamenial epilepsy in this way represents a form of multidien chronotherapy, but the feasibility and benefit of using information on multidien cycles of epileptic brain activity to guide chronotherapy, whether with medication, behavioral modifications, or time-varying neurostimulation paradigms, will need to be tested directly in clinical trials. Natural variability in seizure frequency over time explains a portion of the “placebo” and “regression to the mean” effects observed in epilepsy clinical trials, thus contributing to their high cost and inefficiency (Goldenholz, Goldenholz, et al., 2018; Karoly et al., 2019). Preintervention assessments should involve the capture of full cycles of seizure risk, or, at least, appropriate statistical methods should be employed to account for the presence of partial cycles during the trial.
Discoveries about cycles of seizure likelihood in epilepsy raise as many questions as they answer. Although progress has been made in elucidating the physiology of circadian rhythms, much still remains to be learned, in particular the differential contributions of circadian modulation and sleep-wake cycles to IEA and seizures. In addition, the biological mechanisms underlying multidien cycles in epilepsy remain virtually unknown. Animal models are the most feasible system for the investigation of mechanisms of seizure cycles, so the demonstration of multidien seizure cycles, similar to those seen in humans with epilepsy, in rodent and canine models is fortuitous (Baud et al., 2019; Gregg et al., 2020). Efforts are also underway to determine which measurable physiological, cognitive, and behavioral variables fluctuate in concert with circadian and multidien cycles of epileptic brain activity in humans. For instance, identification of noninvasively measured variables (heart rate, electrodermal activity, actigraphy, etc.) that faithfully track multidien cycles provide new avenues to understand seizure cycles (Meisel et al., 2020; Karoly et al., 2021; Vieluf et al., 2021).
Unravelling the physiological drivers of multidien cycles in epilepsy will remain speculative until sufficient multimodal longitudinal data become available. However, owing to a cycle length of many days, collection of such datasets is challenging. Because multidien periodicity in epilepsy can change in length, even in one individual, finding matching periodicities between EEG recordings of IEA, as a ground truth, and biochemical assays will be instrumental in enabling precise monitoring of these cycles. Catamenial epilepsy should be regarded as a special case of epilepsy with multidien cycles, and future efforts must focus on the identification of mechanisms of multidien cycles that generalize to either sex. Ultimately, research into the multidien rhythms of epilepsy should aim to track multimodal and multiscale biomarkers that can explain the striking prevalence of seizure periodicity observed across animals and humans, children and adults, and men and women.
M. O. B. reports personal fees and grants from the Wyss Center for neurotechnology in Geneva, and has a patent application pending under the Patent Cooperation Treaty (62665486). V. R. R. has served as a consultant for NeuroPace, Inc., manufacturer of the RNS System, a device used in some of the studies referenced here, but NeuroPace, Inc. did not provide targeted funding for this work. P. J. K. reports personal fees and a financial interest in Seer Medical Pty.
This is an open access publication, available online and distributed under the terms of a Creative Commons Attribution-Non Commercial-No Derivatives 4.0 International licence (CC BY-NC-ND 4.0), a copy of which is available at https://creativecommons.org/licenses/by-nc-nd/4.0/. Subject to this license, all rights are reserved.
Your browsing activity is empty.
Activity recording is turned off.
See more...
