Antiseizure Networks

Background

Often, our focus as a field has been on the “focus” - the site of seizure initiation. However, seizures engage broader networks, including both pathways that support seizure propagation and pathways proposed to actively restrain and terminate seizure activity, here coined “antiseizure networks”. The goal of this session was to examine the evidence for antiseizure networks and to bridge mechanistic insights from animal models with findings from human neurosurgical and neuroimaging studies, ultimately addressing a fundamental question in epilepsy research: why are people with epilepsy not continuously having seizures?

The concept of antiseizure networks is not new. Looking beyond the hippocampus and the cortex, inhibitory systems including the extended basal ganglia (BG) ¹ and the cerebellum ² have long been recognized to exert powerful control over seizure initiation and spread. As early as the 1970s, Irving Cooper employed cerebellar stimulation to treat epilepsy. ³ More recently, a renewed focus on cerebellar contributions to seizure control, from Krook-Magnuson and others, has demonstrated remarkable efficacy in rodent models.^4–7 In parallel, foundational studies from Karen Gale in the early 1980s, followed by work from Nico Moshé, Antoine Depaulis, and others established the substantia nigra, and later, broader BG circuitry as potent modulators of seizure expression. Focal inhibition of the substantia nigra robustly inhibits seizures in rodents and in nonhuman primates.^8,9 These experimental findings provided mechanistic grounding for what had become an until-recently “forgotten” neurosurgical approach first employed in animals in the 1950s and later in humans: Dennosuke Jinnai's Forel-H-tomy, in which lesions are placed in pallidothalamic tracts.^10,11 The BG have also gained renewed attention^12–15 with preclinical neuromodulation targeted at substantia nigra, superior colliculus, and striatum, all suppressing seizure in a range of animal models.

What is particularly striking about antiseizure networks is their apparent multipotency to suppress seizures originating in different brain networks. Targeting either the BG or the cerebellum can suppress diverse seizure types, including focal seizures in models of temporal lobe epilepsy, generalized tonic–clonic seizures, and absence seizures. Equally notable is growing evidence that these networks may be engaged before seizure onset. In vivo calcium imaging studies have revealed modulation of cerebellar Purkinje cell activity preceding seizures in mouse models of temporal lobe epilepsy. ¹⁶ Similarly, neurons in the superior colliculus reduce firing rates prior to absence seizures, and absence-like seizures are associated with decreased Gamma-aminobutyric acid (GABA) release in the substantia nigra. ¹⁷ Together, these findings raise the possibility that altered activity within subcortical inhibitory networks reflects a seizure-prone brain state rather than a mere epiphenomenon.

Emerging human data suggest that targeting antiseizure networks is of great relevance to clinical epilepsy. As discussed in the sections below, both lesion-mapping and neurosurgical disconnection studies increasingly implicate subcortical connected targets, while work in nonhuman primate models has begun to clarify how specific seizure patterns engage antiseizure circuitry. Placed within a broader conceptual context, these observations align with the Ictal Suppression Hypothesis (ISH), which posits that large-scale brain networks tonically inhibit seizure onset zones (SOZs) during interictal periods. Together, these converging lines of evidence support a shift from a focus-centric view of epilepsy toward a systems-level understanding of broader networks that restrain seizures.

The Interictal Suppression Hypothesis

The Interictal Suppression Hypothesis posits that broad brain networks establish the interictal period by tonically suppressing SOZ activity. Network-level seizure suppression accommodates multiple forms of SOZ antagonism: classic GABAergic inhibition, network segregation, node desynchronization, and other mechanisms. Intracranial monitoring for patients with drug-resistant epilepsy provides a rare view into the electrographic networks that organize around SOZs. Prior studies of SOZ connectivity show that the SOZ is highly connected to the rest of the brain.^18,19 At face value, this seems paradoxical: why should a volatile region be so integrated into the broader network? Moreover, why is most of this integration directed into the SOZ?

Multimodal analyses using intracranial single-pulse stimulation, resting-state recordings, and diffusion imaging suggest that inward connectivity to the SOZ is a hallmark of the ISH. ³ To fully evaluate the ISH, neuroscientists must collaborate across scales and modalities. Nonetheless, current intracranial evidence demonstrates two key points: (1) the ISH motif of inward connectivity is reliably observed, and (2) it is likely related to seizure suppression. Multiple studies using directed connectivity analyses corroborate this motif across distinct cohorts.^20,21 Notably, these studies used similar methods to ascertain connectivity. All three groups found that directed analysis distinguishes the SOZ and that these networks appear tuned for SOZ suppression. For example, Gunnarsdottir et al. and Alamoudi et al. employed dynamical network models to define transition matrices encoding the directed influence of one intracranial channel on all others. In our analysis, we similarly fit a linear time-varying model to derive a transition matrix, which—like in Alamoudi et al.'s work—was then spectrally decomposed. ²² The ISH motif reliably distinguishes SOZ channels.^22,23 Recent data-driven study of electrographic networks found that SOZ inward connectivity was present in nearly all patients examined and was the dominant feature in 79% of the cohort. ²⁴ Single-pulse stimulation studies further suggest that the ISH motif may reflect inhibition. For instance, augmentation of local fast activity after non-SOZ stimulation may indicate GABAergic inhibition, consistent with models of parvalbumin interneuron activity.^25,26

If the ISH explains why patients are not constantly having seizures, what does it imply about seizure onset? The ISH presents two possibilities: seizure onset occurs either when tonic inhibition is lost (a relaxation of protective mechanisms) or seizures initiate because SOZ activity overwhelms tonic inhibition (antagonism of the protective mechanism). Peri-ictal analyses support the latter view. SOZs show increased directed connectivity during the first phase of seizures but lose most incoming connections during the second phase.^11,27 These phases align with seizure spread. As the ISH motif collapses, seizure propagation to the rest of the brain peaks. The collapse of this motif also predicts whether seizures generalize with impaired consciousness. In contrast, focal aware seizures maintain high inward connectivity to the SOZ throughout the entire seizure. Additional peri-ictal analysis reveals a complex interplay between high-frequency activity and low-frequency activity within and surrounding the SOZ. Seizure propagation may be constrained by the amount of low-frequency activity the rest of the network can traffic into the SOZ. ²⁸ Further assessment of the ISH requires a demonstration that inward connectivity to the SOZ is associated with successful seizure prevention. The ISH would predict that interictal epileptiform discharges from the SOZ are met with high connectivity to the SOZ, or that successful neurostimulation is concomitant with increased inward connectivity to the SOZ. Further work beyond intracranial electrophysiology will also be instrumental to assessing the validity of the ISH.

Functional and Structural Antiseizure Networks

Schaper et al. ⁴⁴ recently demonstrated that the risk of developing epilepsy after a structural brain lesion does not depend on the lesion's anatomical location alone but on its functional connectivity to a shared subcortical brain network. Across a heterogeneous collection of lesion etiologies, lesions that led to epilepsy were consistently characterized by negative connectivity (“anticorrelation”) to a set of nodes in the BG (substantia nigra, globus pallidus internus) and cerebellum (dentate nuclei, vermis). In other words, regions within and positively functionally connected to the BG and cerebellum had a lower intrinsic susceptibility to seizures when lesioned, whereas regions anticorrelated with these same regions—such as temporal neocortex, central operculum, and hippocampal CA1—–are more susceptible to seizures when lesioned. These findings suggest there is a spatial pattern of intrinsic susceptibility to seizures across the brain that generalizes to multiple lesion etiologies and proposes that lesions increase epilepsy risk partly through disruption of remote brain networks.

Converging evidence comes from stimulation studies. In patients receiving anterior thalamic deep brain stimulation (DBS) for drug-resistant focal onset epilepsy, Schaper et al. found that seizure improvement was associated with more positive functional connectivity between the stimulation site and the same BG–cerebellum network identified in the lesion network mapping analysis. In other words, lesions that disconnect or disrupt this network increase the likelihood of epilepsy, whereas stimulation sites that modulates this network improves seizures. These findings demonstrate that the BG-cerebellar system is involved in the cause and control of seizures in humans, consistent with the longstanding observations from animal studies that these regions exert widespread regulatory influence on cortical excitability and seizures.^13,17,45

Mechanistically, several hypotheses could explain how these subcortical systems may restrain or stop seizures. The BG could suppress seizures through inhibitory output pathways projecting to thalamocortical circuits, acting as a gating system that limits the propagation of hypersynchronous activity. The substantia nigra pars reticulata, in particular, has been shown in multiple animal models to exert strong seizure-suppressive effects when lesioned or inhibited.^8,9,13 The cerebellum, traditionally viewed as a motor structure, is known to have a widespread modulatory role on the cortex. Cerebellar Purkinje cell output to deep cerebellar nuclei provides a powerful inhibitory projection system capable of modulating thalamic and cortical activity; disruption of this modulation may facilitate epileptogenesis, whereas stimulation of cerebellar nuclei can abort seizures.^2,7,46,47 The convergence of cerebellar and BG outputs on the thalamus may suggest a shared final antiseizure pathway that may represent a common mechanism for stopping seizures. ⁴⁴

Altogether, these findings highlight that seizure suppression relies on a distributed brain network. Lesions may lead to epilepsy by disconnecting or de-regulating these networks, while DBS may restore or modulate their function. Recognizing the BG–cerebellum axis as a core antiseizure network not only reframes classical network models of epilepsy but also provides a principled framework for improving neuromodulatory strategies aimed at enhancing endogenous seizure-stopping mechanisms. ¹

Evidence that opposing seizure networks—at least one supporting seizure generation and another stopping seizures—exist in humans has been recently provided. ⁴⁴ Importantly, this could be directly linked to specific white matter pathways. In patients in which the fornix was implanted with a DBS device, either monolaterally or bilaterally, Koubeissi and colleagues ⁴⁸ showed that stimulation at 5 Hz could decrease hippocampal seizures and spikes in patients with temporal lobe epilepsy, but also caused seizures in patients with extratemporal epilepsy, suggesting white matter pathways are implicated in seizure cessation and generation. ⁴⁹ Likewise, in frontal lobe epilepsy, a recent study by Giampiccolo and colleagues ⁵⁰ showed that disconnection of the anterior thalamic radiation and anterior fronto-striatal projections was associated with long-term seizure freedom, with 80% percent of patients remaining seizure free at 5 years. This result was independent of resection volume, suggesting an important role for disconnection of tracts connecting the frontal lobe with the thalamus and BG. Future stimulation mapping experiments of the thalamus, BG and subthalamic structures ⁵¹ will be critical to define how antiseizure networks are modulated in humans.