Neurostimulation for generalized epilepsy: should therapy be syndrome-specific?

Introduction

One-third of people with epilepsy have generalized onset seizures,^1,2 which appear to begin in both hemispheres simultaneously on scalp EEG. This contrasts with focal onset seizures, which originate in one hemisphere and have a unilateral EEG appearance.² Although the borders between these seizure types are not always clear, the distinction between focal and generalized epilepsy has strong implications for diagnosis and treatment.

The clinical manifestations of a generalized seizure vary widely, from motor (e.g., tonic stiffening, atonia, clonic or myoclonic jerking) to non-motor symptoms (e.g., impaired awareness), either alone or in combination. The EEG appearance can also show various seizure onset patterns, including repetitive spike-wave discharges, paroxysmal fast activity, rhythmic activity, and diffuse electrodecrement, each suggestive of a unique neuronal firing pattern (Figure-1).

Figure 1: Characteristic EEG features of LGS versus IGE.

LGS and IGE are associated with different generalized interictal epileptiform discharges (IEDs) and ictal patterns on scalp EEG. (A) Interictally, LGS is characterized by generalized ‘slow’ spike-wave (SSW) discharges occurring at a frequency of ≤2.5 Hz, typically on an abnormal background of diffuse theta-delta slowing. (B) Bursts of interictal >10 Hz generalized paroxysmal fast activity (GPFA) are also seen, particularly during sleep. (C) Ictally, tonic seizures of LGS consist of a burst of bilateral >10 Hz fast activity (resembling GPFA) followed by diffuse decrement evolving to higher-amplitude spike and slow-wave discharges. (D) Interictally, patients with IGE show a well-organized background with superimposed ‘faster’ ~3–6 Hz generalized spike-wave (GSW) and/or (E) generalized polyspike-wave (GPSW) activity. (F) Ictally, generalized tonic-clonic seizures of IGE typically demonstrate an onset of bilateral polyspikes and/or evolving spike/polyspike-wave discharges.

Combined with other electroclinical features, patients with generalized seizures may be categorized into distinct epilepsy syndromes,³ each having a unique profile with respect to typical onset age, etiology, comorbidities, quality of life, and impact on families and caregivers, all of which are important to consider when designing treatments and predicting outcomes.

Neurostimulation is rapidly changing the treatment landscape for generalized epilepsy. Clinically available options include continuous and duty-cycling deep brain stimulation (DBS), responsive neurostimulation (RNS), and vagus nerve stimulation (VNS).⁴ However, current applications are largely agnostic to the syndrome and seizure type(s) being treated, and only carry Food and Drug Administration (FDA) approvals in the United States for focal epilepsy, despite increasing off-label use.

A salient example is DBS, where the same anatomical site, the thalamic centromedian nucleus, has been targeted for Lennox-Gastaut syndrome (LGS)^5–9 and idiopathic (genetic) generalized epilepsies (IGE),^7,9–11 among others (e.g., patients with focal, multi-focal, and combined generalized and focal epilepsy).^8,9,12–17 Many retrospective DBS studies have used broad inclusion criteria, with outcomes collapsed across syndromes and seizure types. This likely reflects the still-emerging state of the field and the small number of prospective and controlled studies performed relative to other DBS indications such as movement disorders.

Applications of neurostimulation have progressed in parallel with—but, we would argue, often independent of—a growing understanding of the syndrome- and seizure-specific brain networks underlying generalized epilepsy^18–20 and its associated cognitive and behavioral comorbidities.²¹ The concept of generalized epilepsy networks has been advanced through neuroimaging studies utilizing anatomical^22–27 and diffusion-weighted MRI,^28–32 simultaneous EEG with functional MRI (EEG-fMRI),^33–44 resting-state fMRI,^45–51 positron emission tomography (PET),^52–56 single photon emission computed tomography (SPECT),^57–60 and magnetoencephalography (MEG),^61–65 among others.

A commonly stated goal of neurostimulation is to reduce seizures by modulating the specific brain area(s) thought to generate them or their pathways of propagation. For example, in patients with focal seizures who are not candidates for resective neurosurgery, RNS can be delivered directly to one or more cortical seizure foci.^66,67 However, this goal is more difficult to define in the context of generalized epilepsy, for several reasons. Generalized seizures are seemingly expressed synchronously across widespread brain regions,² which impedes defining a discrete seizure focus and, consequently, stimulation target. Patients with generalized epilepsy also commonly experience multiple seizure types, which, as reviewed below, engage different brain networks. Additionally, generalized epilepsy can occur in patients with focal lesions,^33,68,69 complicating the treatment target: is it the generalized seizure network, the lesion, or both?

Scope of this review

Recent reviews of neurostimulation for generalized epilepsy have focused on clinical efficacy,^4,70–73 targets,^18,74,75 and surgical techniques.⁷⁶ The present review compares the two main syndromic ‘archetypes’ of generalized epilepsy— LGS and IGE—and focuses on the two neurostimulation approaches most actively being studied in recent and ongoing clinical trials: DBS and RNS.

First, we review evidence from neuroimaging studies describing differences in the brain networks underlying LGS and IGE. Second, we discuss potential mechanisms mediating neurostimulation efficacy, how they may differ by syndrome and seizure type, and how these differences may inform selection of appropriate targets and paradigms for DBS/RNS. Third, we consider how LGS and IGE evolve over short (e.g., between wakefulness and sleep) and long (e.g., childhood to adulthood) timescales, and how stimulation programming may be dynamically adjusted to respond to these changes.

We structure the review as a series of five clinical questions that are encountered when assessing a patient with generalized epilepsy for neurostimulation therapy.

What epilepsy syndrome does the patient have?

Differentiation of LGS and IGE can be traced as far back as the first EEG studies of epilepsy performed by Gibbs, Gibbs, and Lennox in the 1930s^77–79 and 1940s.^80,81 Although the syndromes were not known as such at that time, Gibbs and colleagues described EEG differences in patients with generalized epilepsy, which were later elaborated upon by Gastaut^{82, 83} and Dravet.⁸⁴ They described a ‘fast’ (≥3 Hz) generalized spike-wave (GSW) pattern and a ‘slow’ (≤2.5 Hz) spike-wave (SSW) pattern (Figure-1), and found that the two tended to be associated with different clinical profiles.⁸¹ Patients with SSW often had severe cognitive impairment, greater treatment resistance, tonic seizures, and other EEG patterns including >10 Hz generalized paroxysmal fast activity (GPFA). In contrast, patients with GSW tended to show more preserved cognitive function, usually responded better to anti-seizure medications, and did not have tonic seizures.

These two phenotypes are similar to what we now recognize as LGS and IGE (Table-1). The 2022 International League Against Epilepsy (ILAE) classification⁸⁵ defines LGS by 3 key features: (i) multiple drug-resistant seizure types with onset <18 years, including tonic seizures; (ii) cognitive and/or behavioral impairments (typically mild to profound intellectual disability); and (iii) SSW and GPFA (Figure-1). Although often considered ‘rare’ (0.24–0.28 per 1,000 births),^86–88 the intractability of seizures in LGS leads to a disproportionately high prevalence, with the syndrome accounting for ~4% of children with epilepsy⁸⁹ and up to 17% of epilepsy patients with intellectual disability.⁹⁰

Table 1: Typical electroclinical characteristics of LGS and IGE.

LGS and IGE (and it sub-syndromes) differ with respect to typical interictal and ictal EEG features, predominant seizure type(s), cognitive profiles, onset ages, and etiologies. There is often electroclinical overlap between the IGE sub-syndromes, and one sub-syndrome can evolve to another as patients age.

		Idiopathic generalized epilepsy (IGE) sub-syndromes
	Lennox-Gastaut syndrome (LGS)	Childhood absence epilepsy (CAE)	Juvenile absence epilepsy (JAE)	Juvenile myoclonic epilepsy (JME)	Epilepsy with generalized tonic-clonic seizures alone (GTCA)
EEG background	Diffuse theta-delta slowing and disorganization	Occipital intermittent rhythmic delta activity in 21–30%	Normal	Normal	Normal
Interictal EEG patterns	Awake: Generalized slow ≤2.5 Hz spike-wave Asleep: Generalized paroxysmal fast ≥10 Hz activity	Awake: 2.5–4 Hz generalized spike-wave Asleep: polyspike-wave primarily in drowsiness/sleep	Awake: 3–5.5 Hz generalized spike-wave Asleep: polyspike-wave primarily in drowsiness/sleep	Generalized 3–5.5 Hz spike-wave and polyspike-wave	Generalized 3–5.5 Hz spike-wave and polyspike-wave
Ictal EEG patterns	Tonic seizures: bilateral ≥10 Hz bursts of fast activity with recruiting rhythm	Regular 2.5–4 Hz generalized spike-wave, infrequent disorganized discharges	Regular 3–5.5 Hz generalized spike-wave; 8x more frequent disorganized discharges than CAE	Absences: 3.5–6 Hz generalized spike-wave/polyspike-wave (more disorganized discharges than CAE) Myoclonic jerks: generalized polyspike-wave GTCs: generalized spikes (tonic phase), spike-wave (clonic phase)	Generalized spikes (tonic phase), spike-wave (clonic phase)
Seizure type(s)	Tonic seizures + at least one additional seizure type (e.g., atypical absence, atonic, myoclonic, focal impaired awareness, GTC, epileptic spasms)	Daily to multiple daily 3–20 sec typical absence seizures	Less than daily 5–30 sec typical absence seizures; commonly have GTCs during periods of frequent absence seizures	Myoclonic (mostly upon awakening); GTCs common, preceded by myoclonic jerks; 1/3 have brief 3–8 sec typical absence seizures (infrequent)	Infrequent GTCs (usually within 2 hours of awakening; occurs yearly or less)
Cognitive profile	Mild to severe intellectual disability that often worsens over time	Typically, normal development; ADHD, mood disorders, or learning disabilities can occur	Typically, normal development; ADHD, mood disorders, or learning disabilities can occur	Typically, normal development; ADHD, mood disorders, or learning disabilities can occur	Typically, normal development; ADHD, mood disorders, or learning disabilities can occur
Typical seizure onset age	18 months-8 years (rarely 8–18 years)	4–10 years	9–13 years	10–24 years	10–25 years
Presumed/known etiologies	Structural (most common), genetic (usually de novo), infectious, metabolic, and/or immune causes	Genetic susceptibility (polygenic > monogenic)	Genetic susceptibility (polygenic > monogenic)	Genetic susceptibility (polygenic > monogenic)	Genetic susceptibility (polygenic > monogenic)

Abbreviations: ADHD = attention-deficit/hyperactivity disorder; GTC = generalized tonic-clonic.

IGE is one of the most common forms of epilepsy, accounting for 15–20% of all epilepsy diagnoses.⁹¹ IGE comprises four sub-syndromes (Table-1) associated with different onset ages and predominant seizure types⁹² but sharing similar EEG findings including normal background activity, ~3–6 Hz GSW and/or generalized polyspike-wave discharges (Figure-1). Tonic seizures, SSW, GPFA, and intellectual disability are not typically seen; however, some patients with IGE can develop GPFA, which may be associated with greater treatment resistance.^93,94

LGS and IGE are etiologically divergent. Genomic studies show that IGE has a genetic, typically polygenic, basis.^92,95 Monogenic cases are also reported.⁹² On structural MRI, patients with IGE show visually normal anatomy (Figure-2), although altered cortical thickness and regional brain volumes have been described in quantitative analyses.⁹⁶

Figure 2: Radiological MRI findings in LGS versus IGE.

T1-weighted MRI scans in patients with LGS (A–H) and IGE (I–J). (A) Adult with LGS aged in their 20s with a left frontal lobe focal cortical dysplasia (histopathology: balloon cells); (B) Adult with LGS aged in their 20s with diffuse right frontal atrophy and ventricular dilation due to Rasmussen’s encephalitis; (C) Adult with LGS aged in their 30s with bilateral perisylvian polymicrogyria; (D) Adult with LGS aged in their 30s with right temporal atrophy and gliosis due to a perinatal infarction; (E) Adolescent with LGS and bilateral band heterotopia (double cortex syndrome) and a DCX gene mutation; (F) Child with LGS and a right medial temporal lobe dysembryoplastic neuroepithelial tumor (DNET) on the background of a chromosomal abnormality (partial monosomy 18p); (G) Adult with LGS aged in their 20s with a complex malformation of cortical development involving bilateral periventricular nodular heterotopia (PVNH) and left posterior schizencephaly; (H) Adolescent with LGS with normal MRI and unknown cause of epilepsy. (I) Child with IGE (childhood absence epilepsy [CAE]) with normal MRI. (J) Adolescent with IGE (juvenile myoclonic epilepsy [JME]) with normal MRI. Interpretation: LGS develops secondarily to diverse etiologies, with structural causes accounting for 30–50% of patients.^98–100 In contrast, structural abnormalities are not typically seen in IGE, although subtle alterations are described in quantitative analyses.

In contrast, the etiological profile of LGS is heterogeneous, with a variety of structural, genetic, infectious, immune, and metabolic factors.⁹⁷ 30–50%^98–100 have abnormal structural MRI findings, ranging from malformations of cortical development to acquired brain lesions (Figure-2). The shared phenotype of LGS is thought to reflect a common ‘reaction’ of the brain to these causes—i.e., LGS develops secondarily,^83,97 potentially via convergent neurodevelopmental alterations caused by seizures and other risk factors in early life.³⁴

In summary, LGS and IGE differ with respect to their predominant seizure types (and thus also the target seizure types for treatments including neurostimulation), EEG signatures, cognitive comorbidities, and etiological profiles. In the next section, we review neuroimaging evidence showing that these syndromic differences may be underpinned by distinct patterns of brain network pathology.

What brain networks are involved?

Generalized epilepsy is a disorder of bilateral brain networks,^2,101 i.e., spatially distributed regions connected structurally and/or functionally, within which seizures begin and manifest clinically. These networks can be defined and studied at multiple scales, from synaptic connectivity between neurons to large-scale neuronal ensembles spanning lobes and hemispheres. The latter is the level where neuroimaging techniques operate, to which we now turn our attention.

We begin by considering similarities between LGS and IGE. First, although the diffuse EEG appearance of LGS and IGE might suggest equal participation of all brain areas during seizures and IEDs, functional neuroimaging studies indicate otherwise. Combined EEG-fMRI, which measures blood-oxygen-level-dependent (BOLD) responses time-locked to EEG events, reveals that generalized IEDs of LGS^34,38 and IGE^41,42 engage bilateral but select brain regions (Figures-3,4). Furthermore, the response patterns of involved regions vary, with some showing BOLD increases and others decreases,^33,41,42 likely indicating distinct neuronal responses between different areas during the same IEDs.

Figure 3: Simultaneous EEG-fMRI findings in GPFA versus SSW of LGS and involvement of patient-specific lesions.

Representative individual-level event-related statistical parametric mapping (SPM) of blood-oxygen-level-dependent (BOLD) signal changes associated with scalp EEG-recorded discharges in two patients with LGS. (A) EEG-fMRI acquired during natural sleep in a patient aged in their 20s with LGS on a background of tuberous sclerosis complex. Bursts of GPFA were manually marked on the in-scanner EEG. GPFA onsets and durations were convolved with the canonical hemodynamic response function (HRF) in SPM software (http://www.fil.ion.ucl.ac.uk/spm). One-sample t-tests were used to obtain a spatial map of significant GPFA-related BOLD changes (thresholded using a cluster-level, false discovery rate-corrected threshold of p<0.05 with a cluster-defining threshold of p<0.001, uncorrected). Orange/yellow colors indicate regions showing BOLD increases and blue/light blue colors indicate BOLD decreases. Results are overlaid upon the patient’s T1-weighted MRI scan. (B) EEG-fMRI acquired during natural sleep in a patient aged in their 20s with LGS on the background of a complex malformation of cortical development (including bilateral peri-ventricular nodular heterotopia [PVNH] and a left posterior schizencephaly). Bursts of SSW were manually marked on the in-scanner EEG. Zoomed-in views show BOLD activation in the patient’s lesions (peri-ventricular nodules; green arrows) that is seen together with the more distributed cortical and subcortical SSW-related pattern. Interpretation: GPFA and SSW of LGS show distinct patterns of brain network involvement. In patients with lesions, EEG-fMRI involvement of the lesion can occur together with the more distributed patterns seen in patients with or without lesions,^{33, 34, 105} suggesting that epileptiform activity of LGS is expressed via a shared “secondary network”.⁹⁷

Figure 4: Simultaneous EEG-fMRI findings in SSW of LGS versus GSW of IGE.

Group-level event-related independent component analysis (eICA),¹⁰⁸ which uses a flexible modeling approach to estimate time-courses and spatial sources of fMRI activity associated with EEG events, was performed in 11 patients with LGS (mean age=34 years) and 15 patients with IGE/childhood absence epilepsy (mean age=10 years). Each row shows a different brain network and its associated fMRI time-course, over the period −32 to +32 seconds relative to SSW or GSW onset. The vertical line in each plot indicates the time of GSW or SSW onset, and the horizontal line indicates the baseline fMRI signal level. Asterisks (*) indicate where the fMRI signal is significantly different from baseline (p<0.05, uncorrected). Spatial maps (left and right columns) are z-statistic images, thresholded to show significant (p<0.05) clusters of voxels associated with the event time-courses. The color scale of the spatial maps has been adjusted to indicate whether the brain network shows a predominant fMRI signal decrease (maps with blue/light blue colors) or signal increase (maps with orange/yellow colors) in response to SSW or GSW. Interpretation: Sensorimotor cortex (first row) shows decreased fMRI signal in response to SSW of LGS, whereas it shows increased fMRI signal in response to GSW of IGE. However, both SSW and GSW involve increased fMRI signal in supplementary motor cortex/anterior cingulate (second row) and decreased fMRI signal in regions of the “default-mode network” (third row). Figure created using findings from Warren et al.¹⁰⁷

However, beyond these superficial similarities, important brain network differences can be observed between LGS and IGE, and between different epileptiform event types in the same syndrome. During GPFA, which is characteristic of LGS and shows EEG similarities to tonic seizures (Figure-1), BOLD signal increases occur in diffuse frontal and parietal ‘association’ (i.e., non-primary) cortices together with the thalamus, basal ganglia (caudate and putamen), cerebellum, and pontomedullary reticular formation (Figure-3).^33,34 This pattern is similar between children and adults³⁴ and between individual patients with various causes of LGS,^33,34 suggesting it reflects a ‘secondary network’ response to the specific epileptogenic insult.⁹⁷ It is also similar to the cerebral perfusion changes seen during the early phase of tonic seizures, as revealed by SPECT.⁵⁸

In patients with LGS secondary to a focal cortical lesion, EEG-fMRI involvement of the anatomic lesion can be seen together with the recognized group-level functional patterns shared by patients with or without lesions (Figure-3).^33,102 Prompt lesion removal can sometimes have significant benefits,^33,102,103 including seizure control, developmental improvement, and abolition of generalized EEG abnormalities. In such patients, lesionectomy may be the preferred first-line surgical treatment over neurostimulation,¹⁰⁴ particularly when the lesion is identified early after epilepsy onset. These considerations do not occur in the context of IGE, where focal cortical lesions are not seen (Figure-2).⁹²

The pattern of epileptic involvement during GPFA differs from that seen during SSW of LGS. EEG-fMRI of SSW shows a distinct pattern of cortical changes, with BOLD decreases being more prominent than increases (Figure-3).¹⁰⁵ The distribution of BOLD decreases resembles the ‘default-mode network’ in healthy subjects (including posterior cingulate, precuneus, angular gyrus, and medial prefrontal cortex), which is thought to support self-oriented cognitive processes including awareness and autobiographical memory, among other functions.¹⁰⁶ SSW-related inhibition of this network is hypothesized to contribute to the clinical manifestation of blank staring and impaired awareness during prolonged runs of SSW and atypical absence seizures.^33,97 SSW-related BOLD increases are also reported in supplementary motor cortex, anterior cingulate, thalamus, and cerebellum (Figure-3), although these appear to be more variable than the default-mode network inhibition.¹⁰⁵ The shape and timing of the BOLD response to SSW can deviate from typical hemodynamic assumptions employed in EEG-fMRI analysis,³³ which may contribute to some of this variability.

The brain networks underlying GSW of IGE show similarities and differences to SSW of LGS (Figure-4). Like SSW, EEG-fMRI studies of GSW have detected BOLD decreases in the default-mode network,^41,42,44 and this is similarly linked to impaired awareness during typical absence seizures of IGE.⁴⁴ BOLD increases are also reported in supplementary motor cortex, anterior cingulate, thalamus, and cerebellum, but again, these changes are thought to be more variable,⁴⁴ possibly due to hemodynamic response variability.

However, when this variability is accounted for, differences emerge between SSW and GSW. One EEG-fMRI study¹⁰⁷ used a flexible analysis approach that made fewer assumptions about the shape and timing of hemodynamic responses to IEDs,¹⁰⁸ and found that SSW of LGS showed BOLD decreases in sensorimotor cortex whereas the same areas showed BOLD increases during GSW of IGE (Figure-4). Sensorimotor activation during GSW may contribute to myoclonic jerking seen during seizures of some IGE variants,¹⁰⁹ or may reflect the evolving nature of IGE sub-syndromes across age (e.g., from absence seizure-predominant childhood absence epilepsy to myoclonic seizure-predominant juvenile myoclonic epilepsy).⁹² It is also consistent with early sensorimotor activation seen during GSW in rodent models,^110,111 and with findings from MEG studies.⁶⁵ Hence, while SSW and GSW both involve inhibition of the default-mode network, sensorimotor differences may be one factor contributing to their distinct EEG appearances and clinical correlates.

In addition to these functional changes, there are potential differences in brain network structural alterations between LGS and IGE, although direct comparisons are lacking. In one study of 10 adults with LGS, there was widespread gray- and white-matter atrophy relative to controls, notably in medial frontal cortex and the pons,²⁶ as measured by voxel-based morphometry of T1-weighted MRI. In contrast, a study of 289 adults with mixed forms of IGE found maximal gray matter atrophy in precentral/para-central cortex and the thalamus,²⁵ echoing the pattern of GSW-related sensorimotor activation described above.¹⁰⁷

There is emerging evidence of brain network differences between IGE subsyndromes. For example, differences in dopamine uptake are seen in PET scans of juvenile myoclonic epilepsy versus epilepsy with generalized tonic-clonic seizures alone, the former showing reduced tracer binding in the midbrain and the latter showing reductions in the putamen.⁵⁴ Similarly, differences in diffusion MRI white-matter architecture²⁸ and fMRI connectivity⁵⁰ are found between refractory and non-refractory forms of IGE, and between IGE sub-syndromes,¹¹² suggesting unique patterns of seizure engagement and their related network alterations.

What is the optimal stimulation target?

If LGS and IGE are expressed via different brain networks, the optimal stimulation targets may also differ—if neurostimulation exerts therapeutic effects by acting on the specific neural generators of seizures and not a more seizure type-agnostic mechanism,¹¹³ such as modulating overall cortical arousal.

The thalamic centromedian nucleus is the most widely explored stimulation target for LGS and IGE, or at least the most widely intended target, as different neurosurgical targeting and post-implantation programming strategies likely have differing accuracies¹¹⁴ with respect to electrode trajectories and stimulation field effects on the centromedian nucleus versus surrounding structures.

Evidence for efficacy of centromedian stimulation, specifically duty-cycling or continuous DBS, is more mature for LGS than it is for IGE. Centromedian DBS has been studied in LGS for >30 years, starting with the pioneering work of Velasco,^6,115–118 Fisher,¹⁵ and colleagues. Benefits have also been described in several unblinded studies.^8,9,13,14 Most recently, the efficacy of duty-cycling bilateral centromedian DBS was evaluated in a randomized, double-blind, placebo-controlled trial (Electrical Stimulation of Thalamus for Epilepsy of Lennox-Gastaut phenotype [ESTEL]), showing a significantly greater reduction in EEG-recorded—but not diary-recorded—seizures after 3 months in the treatment versus control groups, with outcomes measured as the sum of multiple seizure types.⁵ There have been no such randomized controlled trials of centromedian DBS for IGE, but a small number of case studies give reason to be optimistic, with seizure reductions of 75–97% reported.^9,11 One single-blind trial including 4 patients with IGE found a 50–100% reduction in absence and generalized tonic-clonic seizures after 3 months of DBS.⁷ Additional evidence comes from unblinded case reports of centromedian RNS for IGE.^10,119–121

The rationale for centromedian stimulation in generalized epilepsy stems from a long-held view implicating this nucleus in modulating diffuse cortical excitability, as supported by the observation of generalized cortical potentials evoked by stimulating it,^74,122 and by a somewhat contentious notion of the centromedian nucleus being a “non-specific” thalamic region with “widespread” cortical and subcortical connections.¹²³ However, more recent work suggests a greater degree of specificity in the functions and connectivity of this nucleus than perhaps previously thought.^124,125 Axonal tracing studies in non-human primates show it is a major source of excitatory input to the striatum (particularly sensorimotor striatal territories including caudal putamen and caudate), whereas its projections to the cortex (which are mostly to central and pre-central regions of sensorimotor cortex) and extra-striatal basal ganglia are comparatively sparse.^126–128 The centromedian nucleus also receives brainstem inputs from the reticular formation, vestibular nucleus, and solitary nucleus.¹²⁵ Similar patterns are seen in human neuroimaging studies,^13,114,129 including an analysis of diffusion MRI data from the Human Connectome Project¹³⁰ performed as part of the current review (Figure-5).

Figure 5: Thalamic anatomy and structural connectivity of the centromedian nucleus.

(A) Thalamic anatomy defined by the Krauth/Morel atlas,¹⁸⁶ including the centromedian nucleus (yellow). (B) Results of a structural connectivity analysis performed using diffusion MRI scans in the “100 unrelated subjects” dataset of healthy young adults (mean age=29 years) from the Human Connectome Project.¹³⁰ Analysis followed our described pipeline,¹³¹ using MRtrix3 software (https://www.mrtrix.org). Briefly, whole-brain probabilistic tractography (20 million streamlines) and spherically informed filtering of tractograms (SIFT2) was performed per HCP subject. The resulting SIFT2-weighted tractograms were used to calculate structural connectivity strength between left and right masks of the thalamic centromedian nucleus¹⁸⁶ and a whole-brain parcellation of grey matter (https://identifiers.org/neurovault.collection:11930).¹³¹ Left and right connectivity values were summed together, then results were averaged across all 100 HCP subjects. For display, these subject-averaged values were converted to percentages (out of the total bilateral connectivity strength across all parcels). Cortical hemisphere views were generated using ggsegGlasser software (https://github.com/ggseg/ggsegGlasser).¹⁸⁷ Positions of subcortical axial views are indicated by the dotted white lines on the sagittal view. Interpretation: Cortically, the centromedian nucleus shows strongest connectivity with primary motor (Brodmann area 4), primary somatosensory (Brodmann area 1), and medial posterior supplementary motor cortex (Brodmann area 6mp). Subcortically, there is strongest connectivity with posterior caudate, posterior putamen, globus pallidus, midbrain reticular formation, and peri-aqueductal grey area. Abbreviations: A = anterior; Ant = anterior nuclear group; BA1 = Brodmann area 1; BA4 = Brodmann area 4; BA6mp = Brodmann area 6 medial posterior; Caud = caudate; CL = central lateral nucleus; CM = centromedian nucleus; GP = globus pallidus; L = left; Lat = lateral; MD = mediodorsal nucleus; Med = medial; MRF = midbrain reticular formation; P = posterior; PAG = peri-aqueductral grey area; Pulv = pulvinar nucleus; Put = putamen; VA = ventral anterior nucleus; VL = ventral lateral nucleus; VM = ventral medial nucleus; VPL = ventral posterior lateral nucleus; VPM = ventral posterior medial nucleus.

How might this connectivity pattern relate to efficacy of centromedian stimulation for LGS and IGE? One hypothesis is that overlaps exist between these connections and the epileptic networks underlying each syndrome. For example, both the centromedian nucleus and the network implicated in GPFA/tonic seizures of LGS connect to the brainstem, putamen, caudate, and cerebellum (Figures-3,5). Similarly, regarding GSW of IGE, a key overlap may be sensorimotor cortex (Figures-4,5). Supporting this hypothesis is the recent finding from the ESTEL trial that DBS efficacy for LGS is positively correlated with structural connectivity between thalamic stimulation sites and areas of GPFA-related BOLD activation.¹³¹

However, there are also areas of non-overlap, which may be relevant to understanding the seizure types for which centromedian stimulation is most efficacious or may speak to therapeutic mechanisms other than direct connections with the specific brain regions driving seizures. For example, connections between the centromedian nucleus and areas of the default-mode network (which shows prominent inhibition during SSW and GSW) are less apparent, for example with the precuneus and posterior cingulate (Figure-5). This contrasts with the reported efficacy, at least in some individuals, of centromedian DBS for absence seizures,^6–8 the ictal correlate of generalized spike-wave. One interpretation is that therapeutic effects are instead mediated via less direct means, such as shifting the brain to a state where absence seizures are less likely to occur.¹³² For example, absence seizures show an inverse correlation with vigilance, being less frequent during periods of high arousal, including cognitive or physical tasks.^133,134 Such states are known to disengage the default-mode network,¹⁰⁶ and there is some evidence that centromedian DBS alters cortical arousal levels.¹³⁵

Another possibility is that stimulation targets other than—or additional to—the centromedian nucleus may be more effective for specific syndromes or seizure types, owing to differences in the brain networks involved. For example, Valentin et al.⁷ found that centromedian DBS was significantly more effective for 6 patients with mixed forms of generalized epilepsy (4 with IGE) compared to 5 with frontal lobe epilepsy, potentially reflecting preferential seizure networks modulated by centromedian stimulation. The optimal stimulation site in the ESTEL trial included the anterior and inferolateral “parvo-cellular” sub-region of the centromedian nucleus, but also extended into the adjacent ventral lateral nucleus,¹³¹ raising the potential strategy of stimulating other and/or multiple thalamic targets for LGS (as may be possible with, for example, current steering¹³⁶ or dual thalamic implant¹³⁷ approaches). Within the centromedian nucleus, Son et al.¹⁴ found that DBS was more effective in different nuclear sub-regions for LGS (optimal in anterior and inferolateral sub-region, like in ESTEL¹³¹) versus multi-lobar epilepsy (optimal in dorsal sub-region). The pulvinar¹³⁸ and central lateral¹³⁹ thalamic nuclei show connectivity with cortical regions of the default-mode network, this being potentially relevant to the inhibition patterns seen during SSW and GSW (Figures-3,4); these nuclei are being targeted in ongoing trials.^140,141 Other studied targets for generalized epilepsy include the caudate¹⁴² and cerebellum,^142–144 both of which appear involved in the epileptic networks of LGS and IGE.

To date, most neurostimulation studies for generalized epilepsy have focused on the subcortex, predominantly the thalamus. However, the affected regions are more widespread, and there is evidence that the cortex may participate earlier than the thalamus during some epileptiform event types in LGS^34,145 and IGE,^110,146,147 at least with respect to EEG onset times. These observations have led to the concept of dual thalamic and cortical neurostimulation,^17,148 including an ongoing single-blind clinical trial of RNS in LGS.¹⁴⁹ In this trial, bilateral neurostimulators are implanted, each with a depth lead targeting the centromedian nucleus and a cortical strip lead targeting a “hotspot” in LGS— premotor frontal cortex—recently identified from a multi-modal synthesis of EEGfMRI,³⁴ PET,⁵⁶ and structural connectivity¹³¹ studies. The goal is to improve the speed and precision of seizure detection (and thus responsive stimulations) and to provide broader modulation of the epileptic network underlying LGS.

There are myriad other factors that likely influence a stimulation target’s efficacy, including those that occur at the individual patient (rather than syndrome) level. In other conditions, such as Parkinson’s disease, there is evidence that individual genotypes can influence DBS response, with superior outcomes for patients carrying certain mutations (e.g., LRRK2) compared to others (e.g., GBA).¹⁵⁰ Similar findings occur in the pharmacological treatment of generalized epilepsy, where outcomes are dependent upon the genetic etiology.^151,152 Genetic factors also influence many properties that govern how brain tissue responds to stimulation, including variability in glutamate signaling,¹⁵³ which can affect clinical response.^154,155 We envision that personalized treatment approaches based on individual etiologies or genotypes will soon be adopted in epilepsy neurostimulation to enable targeted selection of patients most likely to benefit.

What is the optimal stimulation paradigm?

Beyond the target(s), neurostimulation involves decisions regarding parameters including stimulation frequency, amplitude, pulse width, constant voltage or current, bipolar or referential polarity, unilateral or bilateral, and continuous or duty-cycling (on/off) or closed-loop detection thresholds.

This vast parameter space remains largely unexplored in generalized epilepsy. Most DBS and RNS studies employ stimulation settings used in previous trials, likely due to regulatory and safety considerations, as well as extending generalizability by using settings already known to “work”, at the expense of investigating novel ones that may yield superior outcomes. Practical constraints also limit the number of settings that can feasibly be explored, given that outcomes typically take several months to assess; multiple paradigms would require lengthy trial durations. Anecdotally, more exploration happens outside the trial context, with device adjustments occurring in the course of ongoing clinical management, but the insights gained tend to be specific to individual patients.

In centromedian DBS, typical choices^5,8,9 include bilateral, high-frequency (e.g., 130–145 Hz) stimulation, delivered either continuously or in a duty-cycling fashion (e.g., 1 min on/5 min off), with the specific contacts and polarity often tailored to avoid patient side-effects. Although the exact neurophysiological effects are unknown, and likely differ between patients and anatomical targets, it is commonly thought that high-frequency DBS mimics ‘lesioning’ by inhibiting neuronal firing,^156,157 which theoretically reduces excitability. However, more recent hypotheses suggest the mechanism may be more complex, such as ‘overriding’ pathological neuronal oscillations with a more ‘regular’, stimulation-induced pattern,¹⁵⁸ modulating wider dysfunctional circuits connected to the target,¹⁵⁹ or inducing long-term neuroplastic changes¹⁶⁰ (which may contribute to the progressive seizure improvements seen over months and years of DBS)¹⁶¹.

One strategy that may facilitate faster and more systematic identification of optimal paradigms is to develop disease biomarkers against which stimulation can be rapidly titrated. In other indications, for example Parkinson’s disease, there are established electrophysiological biomarkers that are measurable intra- or post-operatively and correlate with disease symptoms and DBS response, including beta-band (~11–30 Hz) frequency power of local field potentials¹⁶² and evoked resonant neural activity (“ERNA”).¹⁶³ These can inform stimulation adjustments, like selecting paradigms that suppress beta-band power¹⁶⁴.

At present, no such biomarkers are clinically used for neurostimulation in generalized epilepsy, although some are proposed.^165–168 For example, the burden of scalp EEG-recorded GPFA discharges correlates with seizure outcomes in DBS for LGS.¹⁶⁹ Other measures may be imported from the broader domain of epilepsy neurosurgery, where several electrophysiological biomarkers have been identified¹⁷⁰ to tailor resections (e.g., high-frequency oscillations); these may hold similar value in optimizing stimulation targeting and paradigms.

Additionally, the choice of stimulation paradigm may depend on the therapeutic goal. Is it to “chase and abort” seizures whilst having minimal impact on normal brain function to avoid potential stimulation side-effects (as may be more effectively achieved by the intermittent stimulations given in closed-loop approaches)? Or is to create a lasting neuroplastic change from a state of abnormal brain organization, of which seizures are a symptom (as may be more effectively achieved by delivering more frequent stimulation, or stimulating during periods when the brain is most capable of such “reorganization”)? Supporting the latter goal is the recent—and somewhat counterintuitive—observation that closed-loop stimulation for focal epilepsy leads to better outcomes when stimulation is provided during periods of less epileptiform activity and lower seizure susceptibility,¹⁷¹ suggesting the interictal period may be when the brain is more amenable to the neuroplastic changes required for prolonged seizure reduction. Whether the same holds true in generalized epilepsy, and whether this differs between LGS and IGE, will be important to confirm in ongoing RNS trials.^149,172

What is the plan for adjusting stimulation over time?

Stimulation programming is typically static in the short-term (hours and days), but often adjusted in the long-term (months and years). However, generalized epilepsy is dynamic and presents differently over time, suggesting stimulation paradigms may also need to flexibly adapt to these changes.

Regarding short-term changes, epileptiform activity of LGS and IGE shows circadian variation. In LGS, seizures and IEDs occur most frequently during non-rapid eye movement (NREM) sleep and are comparatively less frequent during wakefulness and rapid eye movement (REM) sleep.^173,174 A similar pattern is seen in IGE.¹⁷⁵ Sleep is not typically factored into stimulation programming for generalized epilepsy, despite evidence that DBS can alter sleep architecture, including by changing the timing and duration of REM and NREM sleep stages in subthalamic DBS for Parkinson’s disease,¹⁷⁶ or by increasing the number of nighttime arousals in anterior thalamic DBS for focal epilepsy.¹⁷⁷ Hence, stimulation programming for LGS and IGE may benefit from considering impacts on sleep cycles, given their close association with seizure susceptibility.

Seizure occurrence also varies over multi-day cycles (e.g., weekly-monthly),¹⁷⁸ with the exact timing seemingly patient-specific and associated with endogenous cycles in heart rate¹⁷⁹, body temperature,¹⁸⁰ electrodermal activity,¹⁸⁰ and other physiological systems. There is also evidence that seizure occurrence may even be influenced by environmental factors including elevated carbon monoxide concentrations due to ambient air pollution.¹⁸¹ These multi-day cycles have mostly been investigated in the context of focal epilepsy, and it will be important to confirm whether similar patterns are also observed in LGS and IGE. However, they raise the potential of dynamically adjusting stimulation using information from long-term physiological or environmental recording devices, which are already being used to, for example, forecast risk states associated with sudden unexpected death in epilepsy via subcutaneous cardiac monitors.¹⁸²

In the longer term, LGS and IGE show significant evolution over the course of a patient’s life. At onset of LGS (peak age 3–5 years), not all electroclinical features may be present in combination,⁸⁵ and some seizure types can wax and wane over time. For example, tonic and atypical absence seizures tend to persist into adulthood whereas atonic seizures can sometimes diminish.¹⁰⁰ Additionally, the nature of caregiver burden in LGS can change over time, with cognitive, behavioral, and physical impairments (e.g., sleep disturbances) often becoming equally or even more challenging to manage than seizures as patients become older.^183,184 Whether neurostimulation therapies can play a role in managing these non-seizure consequences remains largely unstudied in epilepsy.¹⁸⁵

A similar evolution occurs in IGE, where there is often an age-related shift from one sub-syndrome to another (e.g., from childhood absence epilepsy to juvenile myoclonic epilepsy; Table-1). Given emerging evidence of variability in the brain network substrates of different IGE sub-syndromes,^28,50,54,112 stimulation programming may need to be adjusted to respond to the evolving nature of IGE.

Summary

We are early in the journey towards optimizing neurostimulation therapy for generalized epilepsy. There is accumulating evidence of efficacy for select targets and paradigms, including DBS of the thalamic centromedian nucleus. However, generalized epilepsy is not a uniform disease, suggesting that the optimal forms of neurostimulation may lie beyond the prevailing “one-target-fits-all” approach.

LGS and IGE differ with respect to their EEG signatures, predominant seizure type(s), cognitive comorbidities, etiological profiles, and prognosis. Recent neuroimaging studies demonstrate that these differences may be underpinned by syndrome-specific patterns of brain network pathology. Hence, therapeutic efficacy of neurostimulation may be enhanced by applying knowledge of the underlying brain networks and how they evolve over short and long timescales.

Synopsis.

Current applications of neurostimulation for generalized epilepsy employ a one-target-fits-all approach that is agnostic to the specific epilepsy syndrome and seizure type(s) being treated. Here we describe similarities and differences between the two ‘archetypes’ of generalized epilepsy—Lennox-Gastaut syndrome and Idiopathic Generalized Epilepsy—and review recent neuroimaging evidence for syndrome-specific brain networks underlying seizures. Implications for stimulation targeting and programming are discussed using five clinical questions: What epilepsy syndrome does the patient have? What brain networks are involved? What is the optimal stimulation target? What is the optimal stimulation paradigm? And what is the plan for adjusting stimulation over time?

Key points.

• Lennox-Gastaut syndrome (LGS) and Idiopathic Generalized Epilepsy (IGE) show electroclinical differences including distinct interictal epileptiform discharges and predominant seizure types

• Neurostimulation devices, including deep brain stimulation and responsive neurostimulation, aim to therapeutically modulate epileptic brain networks

• Recent neuroimaging studies suggest that the brain networks underlying LGS and IGE are syndrome-specific

• These findings may have important implications for syndrome-specific optimization of stimulation targeting and programming

Clinical care points.

• LGS and IGE are distinct generalized epilepsy syndromes

• Different brain networks underlie LGS and IGE

• Neurostimulation is not currently tailored to each syndrome

• Optimal stimulation targets and programming may be syndrome-specific