Why is epilepsy so often genetic? A seizure is fundamentally a failure of the balance between excitation and inhibition in cortical networks. Most of the molecular machinery that sets that balance — voltage-gated sodium, potassium, and calcium channels, ligand-gated GABA and glutamate receptors, synaptic-vesicle and neurotransmitter-transporter proteins — is encoded by genes that are individually vulnerable to mutation. A single defective channel subunit can tip an entire network toward hyperexcitability. This is why genetic factors are implicated in up to 70% of all epilepsy even though epilepsy affects only 1–2% of the population.

The ILAE 2017 classification lists 'genetic' as one of the six etiologic categories — epilepsy resulting directly from a known or presumed genetic cause. Critically, 'genetic' is not synonymous with 'inherited': most severe early-onset epilepsies arise from de novo variants that are absent in both parents, so a negative family history does nothing to lower suspicion.

The genetic architecture is best understood as a spectrum of effect size and frequency:

• Rare, highly penetrant single-gene variants — large effect, often de novo, producing the severe developmental and epileptic encephalopathies (DEEs) where one mutation is sufficient to cause disease.
• Common polygenic variants — each of small effect, summing across the genome to set an individual's seizure threshold and underlying the common 'idiopathic' generalized and focal epilepsies.

The practical reason to pursue a precise genetic diagnosis is that it moves a patient from a syndromic label to a mechanism — and mechanism is what tells you which drug will help and which will harm. The landmark demonstration that a single gene could underlie a defined epilepsy syndrome came when de novo SCN1A variants were shown to cause severe myoclonic epilepsy of infancy (Dravet syndrome) Claes et al. 2001, opening the modern era of channelopathy-based epilepsy genetics.

The neonatal and early-infantile brain produces a distinct and often devastating set of encephalopathies because the developing cortex is still tuning its excitatory/inhibitory machinery — and the direction of a channel's defect, not merely its identity, dictates both the phenotype and the correct drug. This is the central reasoning challenge of neonatal epilepsy genetics: the same gene can cause opposite diseases requiring opposite treatments.

The channelopathies illustrate this vividly. KCNQ2 encodes Kv7.2, a potassium channel that carries the M-current — a slow, non-inactivating current that acts as the neuron's standing 'brake', stabilizing the resting potential and limiting repetitive firing. Loss-of-function KCNQ2 variants release that brake, producing day-1-to-3 tonic seizures with a burst-suppression EEG. Because the lesion is too little inhibition rather than too much sodium current, sodium-channel blockers (carbamazepine, phenytoin) are paradoxically effective: they damp the excessive firing through a complementary route, and this is one of the cleanest examples of genotype-guided therapy in all of medicine.

SCN2A (Nav1.2) is the cautionary mirror image and demonstrates why functional direction must be inferred. Gain-of-function variants increase sodium current and present early (<3 months) — here sodium-channel blockers help, because they oppose the excess current. Loss-of-function variants present later (>3 months) with epilepsy and autism, and sodium-channel blockers worsen seizures by further suppressing already-deficient channels. Age of onset is therefore used clinically as a practical proxy for the underlying biophysical defect.

Not all neonatal epilepsy is channel-based. Cortical malformation genes (LIS1, DCX, ARX, the tubulinopathies) disrupt neuronal migration, so the seizure is structural and MRI is essential. And a crucial, time-sensitive subset are the inborn errors of metabolism — biotinidase deficiency, pyridoxine-dependent epilepsy, GLUT1 deficiency, molybdenum cofactor deficiency — where the seizures are a downstream symptom of a treatable biochemical block. The urgency here is therapeutic: every hour of uncontrolled neonatal seizures carries developmental cost, and several of these diagnoses respond dramatically to a specific vitamin, cofactor, or diet rather than to conventional antiseizure drugs.

Inborn errors of metabolism cause epilepsy through three recurring mechanisms, and recognizing which one is in play tells you what to look for and how to treat. First, accumulation of a neurotoxic substrate behind an enzymatic block — leucine in MSUD, glycine in nonketotic hyperglycinemia (which acts as an NMDA co-agonist, driving excitation), ammonia in the urea-cycle and organic-acidemia disorders. Second, deficiency of an essential brain fuel or cofactor — the brain starved of glucose in GLUT1 deficiency, or of the active vitamin B6 cofactor in the pyridoxine-responsive epilepsies. Third, deficient neurotransmitter synthesis, as in the BH4 and biogenic-amine disorders. The seizure is the visible tip of an invisible biochemical derangement.

This framing is what makes metabolic epilepsy a treatment emergency rather than a labeling exercise: if you remove the toxic substrate (dietary restriction, sodium benzoate scavenging), supply the missing fuel (ketogenic diet bypassing GLUT1), or replace the depleted cofactor (pyridoxine, P5P, biotin, BH4), seizures that were refractory to every conventional antiseizure drug can stop. The corollary is that the standard antiseizure drugs often fail entirely here, because they target membrane excitability rather than the metabolic root cause — and a child who 'fails' three drugs may simply have an undiagnosed treatable IEM.

A caution on the same logic in reverse: in ALDH5A1 / SSADH deficiency, the problem is impaired GABA catabolism with downstream accumulation of GABA and GHB, so vigabatrin must be avoided — by irreversibly inhibiting GABA-transaminase it raises GABA further and can paradoxically worsen the picture. This is the mechanistic counterpart to the channelopathy 'wrong-drug' traps.

Because the workup is broad and time-sensitive, a systematic metabolic panel is mandatory for any infant with unexplained early seizures: plasma and CSF amino acids (paired, for the glycine ratio), urine organic acids, urine AASA, paired CSF/serum glucose, biotinidase, and lactate. For metabolic disease beyond epilepsy, see the Inborn Errors of Metabolism module.

The hardest part of epilepsy genetics is not generating a result but deciding whether the result actually explains this patient — and in genetic epilepsy the stakes of getting it wrong run in both directions, because a label can commit a child to (or wrongly exclude them from) a specific drug regimen. The discipline is to integrate four threads before acting: variant classification, gene–phenotype fit, inheritance/phase, and functional data.

Gene–phenotype fit is the anchor. A pathogenic-looking SCN1A variant means very different things in a child with temperature-sensitive prolonged seizures beginning at 6 months (classic Dravet) versus a child with isolated simple febrile seizures and normal development. The same molecular finding does not carry the same clinical weight when the phenotype does not match — and treating a poor-fit case as confirmed Dravet, for instance by permanently banning sodium-channel blockers, is itself a harm.

Phase is decisive in recessive disease. Two heterozygous ALDH7A1 variants only cause pyridoxine-dependent epilepsy if they sit on different alleles (in trans). Two variants in cis leave one normal allele intact and do not cause disease, so parental segregation or long-read phasing is required before a recessive diagnosis is made — the difference determines whether the family receives a 25% recurrence-risk counseling or reassurance.

Functional direction changes management even when classification is settled. As SCN2A shows, knowing a variant is pathogenic is insufficient; whether it is gain- or loss-of-function flips the drug choice. Variant interpretation in epilepsy must therefore reach beyond the binary pathogenic/benign call toward mechanism.

Finally, uncertainty is managed, not forced. A VUS must not drive treatment under ACMG guidance; the correct response is to document it, treat by phenotype, and reanalyze — roughly 10–15% of previously non-diagnostic exomes become diagnostic on reanalysis 1–3 years later as ClinVar, GeneMatcher, and the literature catch up. A negative or uncertain result today is a scheduled question for tomorrow, not a closed case.

Pyridoxine-dependent epilepsy (PDE-ALDH7A1) is the model precision-treatment epilepsy because its biochemistry explains every clinical feature — diagnosis, treatment, and treatment's limits — in one continuous chain of reasoning. It is autosomal recessive, caused by biallelic variants in ALDH7A1 encoding antiquitin, an aldehyde dehydrogenase in the cerebral lysine degradation pathway.

The mechanism is a vitamin-trapping reaction. Without antiquitin, its substrate alpha-aminoadipic semialdehyde (AASA) accumulates, in equilibrium with its cyclic form piperideine-6-carboxylate (P6C). P6C reacts with pyridoxal-5-phosphate (PLP) — the active form of vitamin B6 — in a Knoevenagel condensation that chemically inactivates it. PLP is the obligatory cofactor for dozens of brain enzymes, including glutamate decarboxylase, which synthesizes the inhibitory neurotransmitter GABA. So the seizures are, at root, a cerebral PLP deficiency produced not by a lack of vitamin intake but by ongoing chemical consumption of the active cofactor. This was established when antiquitin mutations were identified in affected children with elevated urinary AASA Mills et al. 2006.

Every piece of management follows from that chain. Why pyridoxine works: flooding the system with B6 overwhelms the P6C trapping reaction and restores enough functional PLP — which is why these neonates with refractory seizures stop seizing within minutes of an adequate dose while failing conventional drugs. Why AASA stays diagnostic on treatment: pyridoxine corrects the cofactor deficiency downstream but does nothing to the upstream enzymatic block, so AASA remains elevated and reliable even after therapy has begun. Why diagnosis must be biochemical and genetic, not therapeutic alone: a related disorder, PNPO deficiency, gives a similar phenotype but with a normal AASA and a requirement for P5P rather than pyridoxine — so a pyridoxine response does not by itself pin the gene.

And why precision treatment still falls short completes the lesson: even with prompt, lifelong pyridoxine, roughly 75% of patients have intellectual disability. The likely reason is that the toxic AASA/P6C accumulation and the lysine-pathway disturbance continue independently of cofactor repletion — which is the rationale for triple therapy (pyridoxine plus a lysine-restricted diet plus arginine, which competes with lysine for transport) aimed at lowering substrate accumulation itself. It improves but does not normalize outcomes, the strongest argument for earliest possible — ideally newborn-screened — diagnosis.