The Epilepsy Renaissance: New Mechanisms, New Molecules, and the Dawn of Genetic Therapy

Bottom Line: For two decades, a parade of new antiseizure medications improved tolerability without moving the needle on the roughly one-third of patients who remain drug-resistant. The modern era broke that pattern on four fronts:

• Better efficacy: cenobamate sustained a 76.1% median seizure reduction, with 16.4% of observed patients seizure-free at 36–48 months (C017, 2022).
• New mechanisms: the investigational Kv7 opener azetukalner cut focal seizures 52.8% vs 18.2% with placebo (X-TOLE, 2023).
• Beyond ion channels: serotonergic and cannabinoid agents reshaped the developmental and epileptic encephalopathies — though exploring novel targets is exciting, clinical proof of efficacy is still needed (soticlestat failed in SKYWAY, 2026).
• The genetic frontier: zorevunersen (MONARCH/ADMIRAL, 2026), an intrathecal antisense oligonucleotide, raised NaV1.1 expression and cut convulsive seizures 59–91% in early-phase, open-label Dravet testing — the first clinical epilepsy gene-regulation therapy to show disease-modifying signals.

Why This Matters

Epilepsy pharmacology has long suffered from an uncomfortable truth: more drugs did not mean more controlled patients. Each new agent of the 1990s and 2000s promised cleaner pharmacokinetics or fewer interactions, but the ceiling held — about 30% of patients never achieve sustained seizure freedom no matter how many medications are tried. Comparative effectiveness work reinforced the sense of stagnation. In SANAD II (2021), levetiracetam and zonisamide failed to beat lamotrigine in focal epilepsy (levetiracetam vs lamotrigine HR 1.18, 97.5% CI 0.95–1.47), and valproate remained superior to levetiracetam for 12-month remission in generalized epilepsy (HR 0.79, 95% CI 0.65–0.97). The newer drug was rarely the better drug.

Even the timing of treatment offered no escape. MESS (2005) showed that immediate treatment after a first seizure delayed recurrence (HR 1.4, 95% CI 1.2–1.7; p<0.0001) but conferred no long-term advantage — seizure freedom between years 3 and 5 was 76% with immediate treatment versus 77% with deferred, while adverse events rose from 31% to 39%. The lesson of that era was modesty.

What changed was the question. Instead of "another sodium-channel blocker," investigators asked which mechanism, which receptor, and ultimately which gene. The result is a genuine renaissance: better drugs, drugs that reach beyond ion channels, drugs matched to a cause, and — finally — a drug that compensates for the underlying genetic deficit itself.

Better Drugs, Not Just More

Cenobamate (C017, 2022; C013, 2021)

Cenobamate is the clearest signal that a new molecule can outperform its predecessors on the outcome that matters most — seizure freedom. In the open-label extension reported as Cenobamate C017 (2022), 355 adults with uncontrolled focal seizures were followed for a median of 53.9 months. The median seizure reduction climbed progressively from 65.4% at months 1–6 to 76.1% at months 43–48. Among patients observed during months 36–48, 16.4% achieved 100% seizure reduction, 39.1% reached at least 90% reduction, and 76.4% reached at least 50% reduction. Using the more conservative initial mITT population as the denominator, 100% seizure reduction was still seen in 10.2% and at least 90% reduction in 24.3%. Crucially, 18.4% of patients (65/354) had a consecutive stretch of at least 12 months of complete seizure freedom. Retention was high — 83% at 12 months and 62% at 48 months.

This came at a real but manageable tolerability cost: treatment-emergent adverse events occurred in 88.2%, serious events in 20.3%, with dizziness (34.4%), somnolence (24.5%), and headache (15.2%) most common. Discontinuation for adverse events was 8.7%, and the six deaths were judged unrelated to the drug. The companion extension Cenobamate C013 (2021), following 149 patients up to 7.8 years, confirmed durability: 57% remained in the study at cutoff (median 6.8 years), and among the 107 who reached one year, 79.4% were still on cenobamate after a median of 6.8 years. Notably, 53.7% discontinued one or more concomitant ASMs while continuing cenobamate — a real-world signal of efficacy. The principal caveat is design: both are single-arm, open-label extensions, so the seizure-freedom figures lack a contemporaneous control.

Azetukalner / XEN1101 (X-TOLE, 2023)

If cenobamate represents incremental excellence, azetukalner represents mechanistic novelty. A selective Kv7.2/Kv7.3 potassium-channel opener with a roughly 10-day half-life allowing once-daily dosing without titration, it revives a target abandoned after ezogabine's tissue-discoloration problems — but, per the data, with no tissue discoloration observed. In the phase 2b X-TOLE (2023) trial, adults with focal-onset seizures on 1–3 ASMs were randomized across three doses and placebo. The 25 mg dose produced a median focal-seizure reduction of 52.8% versus 18.2% with placebo (P<0.001), with a clear dose-response (46.4% at 20 mg, 33.2% at 10 mg). The 50% responder rate was 54.5% at 25 mg versus 14.9% with placebo (OR 7.30, 95% CI 3.77–14.10), and seizure freedom over the 8-week period reached 6.3% at 25 mg versus 1.8% with placebo. Onset was rapid — 53.6% were responders by the end of week 1. Adverse events were CNS-predominant (dizziness 24.6%, somnolence 15.6%) and dose-related discontinuation reached 15.8% at 25 mg; no deaths occurred. The limitation is duration and phase: an 8-week phase 2b window cannot speak to long-term seizure freedom, and confirmatory phase 3 data are what will determine its place.

Beyond Ion Channels

The most striking conceptual shift has been the move away from voltage-gated channels toward serotonergic, cannabinoid, and other non-canonical targets — particularly in the developmental and epileptic encephalopathies (DEEs), where the unmet need is greatest.

Cannabidiol (Cannabidiol in Dravet Syndrome, 2017; Cannabidiol as Adjunct for LGS, 2018)

Pharmaceutical-grade cannabidiol was the first cannabinoid to clear rigorous randomized testing. In Cannabidiol in Dravet Syndrome (GWPCARE1, 2017), 120 children received 20 mg/kg/day or placebo; median convulsive-seizure frequency fell 38.9% versus 13.3% (adjusted median difference −22.8 percentage points, 95% CI −41.1 to −5.4; P=0.01). The 50% responder rate was 43% versus 27% (OR 2.00, P=0.08 — notably not statistically significant), and three CBD patients became seizure-free versus none on placebo. In Cannabidiol as Adjunct for LGS (GWPCARE3, 2018), 225 patients showed median drop-seizure reductions of 41.9% (20 mg/kg) and 37.2% (10 mg/kg) versus 17.2% with placebo (P=0.005 and P=0.002). The dominant safety signal in both was hepatic: elevated aminotransferases above 3× ULN in 12 Dravet patients (all on valproate) and in 14/149 (9%) of LGS patients (79% on valproate) — a drug-interaction effect that demands monitoring.

Fenfluramine (Fenfluramine Dravet Study, 2019; Fenfluramine LGS, 2022)

Fenfluramine — a repurposed serotonergic agent once withdrawn for valvulopathy at appetite-suppressant doses — delivered some of the largest effect sizes in DEE. In the Fenfluramine Dravet Study (2019), 0.7 mg/kg/day reduced convulsive seizures by 62.3% versus 1.2% with placebo (P<0.001), with a 68% responder rate and a longest seizure-free interval of 22 days versus 3.5 days. In Fenfluramine LGS (2022), the 0.7 mg/kg/day dose cut drop seizures a median 26.5% versus 7.6% (estimated median difference −19.9 percentage points, 95% CI −31.0 to −8.7; p=0.001), with generalized tonic-clonic seizures the most responsive subtype (−45.7% vs +3.7%). The reassuring headline across both: no valvular heart disease or pulmonary arterial hypertension at epilepsy doses — though mandatory echocardiographic surveillance remains.

Bexicaserin (PACIFIC)

The newest serotonergic entry, bexicaserin, is a 5-HT2C receptor superagonist. The phase 1b/2a PACIFIC trial (topline reported 2024) deliberately enrolled an etiologically mixed DEE population (n=43 bexicaserin, n=9 placebo). Countable motor seizures fell a median 59.8% versus 17.4% with placebo, with consistent effect across subtypes — Dravet syndrome −74.6%, Lennox-Gastaut syndrome −50.8%, and "DEE Other" −65.5%. The responder rate was 60.0% versus 33.3%. Drug-related adverse events were more frequent (65.1% vs 33.3%), with somnolence the leading reason for discontinuation (16.3% during titration, 4.7% during maintenance). The numbers are encouraging but must be read against a very small, early-phase sample with only nine placebo patients; the drug has advanced to phase 3 on this basis.

The honest miss: soticlestat (SKYWAY, 2026)

Not every novel mechanism succeeds, and intellectual honesty requires saying so. Soticlestat, a first-in-class cholesterol 24-hydroxylase (CH24H) inhibitor, was tested as adjunctive therapy in 270 LGS patients in SKYWAY (2026). It failed its primary endpoint: the placebo-adjusted median difference in major motor drop seizure frequency was −1.17% (95% CI −13.02 to 9.99, p=.785) over the full treatment period and 2.43% (p=.778) over maintenance — essentially no effect. Because hierarchical testing stopped at the failed primary, the only positive signals (CGI-I Disruptive Behaviors OR 1.91, nominal p=.032; CGI-I Seizure Intensity and Duration OR 1.67, nominal p=.029) are nominal and unadjusted. The investigators' conclusion was unambiguous: CH24H inhibition should not be pursued for LGS. A clean negative trial is as informative as a positive one — it closes a door and redirects effort.

Precision by Cause

Everolimus for TSC (EXIST-3, 2016)

The conceptual bridge from "treat the seizure" to "treat the cause" was built in tuberous sclerosis complex. TSC arises from TSC1/TSC2 mutations that disinhibit the mTOR pathway, and everolimus is an mTOR inhibitor — meaning EXIST-3 (2016) was, in effect, the first genotype-directed epilepsy therapy tested at scale. Among 366 patients with drug-resistant TSC-associated focal seizures, the high-trough group (9–15 ng/mL) achieved a 40.0% responder rate versus 15.1% with placebo (p<0.001) and a 39.6% median seizure reduction versus 14.9%, with a clear exposure-response relationship.

But EXIST-3 is also a lesson in the limits of mechanism-targeting that stops short of the lesion. Seizure-freedom rates were very low (5.1% low-trough, 3.8% high-trough, 0.8% placebo). There was no improvement in quality of life, adaptive functioning, or cognition despite fewer seizures. And the cost was substantial — grade 3–4 adverse events reached 24% in the high-trough arm, with stomatitis, infections, hypercholesterolemia, and cytopenias reflecting systemic mTOR blockade. Targeting the pathway helped; it did not cure. That gap is precisely what genetic therapy aims to close.

The Genetic Frontier

Zorevunersen (MONARCH/ADMIRAL, 2026)

Dravet syndrome is the proving ground for genetic medicine in epilepsy because its biology is unusually clean. In roughly 80% of cases it stems from a loss-of-function variant in SCN1A, the gene encoding the NaV1.1 sodium channel — a haploinsufficiency in which one functional allele cannot produce enough channel protein, leaving inhibitory interneurons under-excited and the cortex prone to seizures. The therapeutic insight behind zorevunersen is elegant: rather than block a channel or modulate a receptor, use an antisense oligonucleotide to up-regulate expression from the patient's own healthy SCN1A allele, restoring NaV1.1 toward normal levels. This is a targeted augmentation of a deficient gene product — disease modification aimed at the molecular cause.

In MONARCH/ADMIRAL (2026), children and adolescents (ages 2–18) with genetically confirmed Dravet syndrome received intrathecal zorevunersen across single- and multiple-ascending-dose cohorts (10–70 mg), with an extension dosing up to 45 mg every four months. At the 70 mg dose, convulsive-seizure frequency fell by a median 59–91% over 20 months, and — the part that distinguishes disease modification from symptom suppression — improvements in clinical status, quality of life, and adaptive behavior were sustained over 36 months. Seizure drugs reduce seizures; this therapy was associated with developmental and functional gains, the domains that conventional ASMs (and even everolimus in EXIST-3) leave untouched.

The parallel to nusinersen in spinal muscular atrophy is hard to miss and deliberately so: an intrathecally delivered antisense oligonucleotide that addresses a single-gene deficit and changes the trajectory of a devastating pediatric disease. If that comparison holds, zorevunersen is not merely a better seizure drug but a different category of intervention.

The caveats are real and must be stated plainly. These are early, largely open-label dose-escalation data without the placebo-controlled efficacy figures (p-values, confidence intervals) that anchor the symptomatic trials above; the reported 59–91% range reflects this immaturity. Delivery is intrathecal — repeated lumbar punctures, not a pill. The safety profile reflects that route: post-lumbar puncture syndrome in 25% and elevated CSF protein in 45% during extensions. And there were three deaths in the program (two SUDEP, one malnutrition) — a sobering reminder that Dravet carries high baseline mortality and that any disease-modifying claim must ultimately be weighed against hard outcomes in controlled study. Promise, yes; proof, not yet.

Neuromodulation in Parallel

While pharmacology and genetics advanced, device-based therapy matured into a durable option for patients who are not surgical candidates. The RNS System Pivotal Trial (2011) provided Class I evidence for responsive cortical stimulation: a 37.9% seizure reduction versus 17.3% with sham during the blinded period (P=0.012), rising to a 46% responder rate at two years, with no deterioration in mood or cognition. The SANTE (2010) trial of deep brain stimulation of the anterior thalamus showed a 40.4% median reduction versus 14.5% in controls at three months (p=0.002), improving to 56% median reduction and a 54% responder rate at two years. And vagus nerve stimulation, the longest-established option, showed in VNS for Partial Seizures (1998) a 27.9% reduction with high-frequency versus 15.2% with low-frequency stimulation (p=0.04). These modalities palliate rather than cure, but they meaningfully reduce burden in the most refractory patients.

Strengths and Limitations

• Short durations and surrogate endpoints. Most efficacy trials (X-TOLE at 8 weeks; the CBD and fenfluramine DEE trials at 14 weeks) measure percent seizure reduction over weeks, not seizure freedom over years. The cenobamate extensions provide the long view but lack a control arm.
• Pediatric-only and rare-disease data. The DEE and genetic-therapy evidence (Dravet, LGS, TSC) comes largely from children and small populations — zorevunersen's cohorts are dose-escalation tiers, and PACIFIC randomized only nine patients to placebo.
• Negative and non-significant results matter. SKYWAY was flatly negative; the CBD Dravet 50% responder endpoint (OR 2.00, P=0.08) and perampanel's responder rates (p=0.0760 and p=0.0914 in the Perampanel Pivotal trial, 2012) were not statistically significant. Effect on median seizure frequency is not the same as making patients seizure-free.
• Safety and access. Everolimus carries a 24% grade 3–4 adverse-event rate; CBD and fenfluramine require hepatic and cardiac monitoring respectively; zorevunersen requires repeated intrathecal dosing. Genetic and device therapies also raise cost and access questions these data cannot answer.

What This Means for Practice

The trajectory across these trials is unmistakable. We began with broad suppression — sodium-channel blockers and their many cousins, refined for tolerability but stuck against a drug-resistance ceiling that SANAD II and MESS made painfully clear. We moved to better and more mechanistically diverse suppression: cenobamate's durable seizure freedom, azetukalner's Kv7 opening, and the serotonergic and cannabinoid agents that finally gave the DEEs real options. Then came mechanism-by-cause — everolimus treating the mTOR substrate of TSC. And now, with zorevunersen, the field has reached genetically targeted disease modification, aiming not to mask the SCN1A deficit but to compensate for it by boosting expression from the healthy allele.

Epilepsy is being reclassified in front of us: from a disease defined by its seizures to a set of diseases defined by their mechanisms and genes — and, increasingly, treated accordingly.

For the first time, the most exciting epilepsy question is no longer "which drug suppresses the seizure?" but "which gene caused it, and can we fix that?"