Epilepsy threshold levers — the over-synchronisation threshold-raise, decomposed into three DNA-grounded levers

The previous chapter's over-synchronisation threshold-raise is decomposed into a DNA-grounded three-lever target map: reduce the inward excitatory current (L1), increase the outward potassium current (L2, the dominant M-current axis), or remove the upstream mTOR drive (L3). Sixteen epilepsy genes are placed by their own promoter switch stiffness. Targets are ranked, never drugs or doses; efficacy=0; not medical advice.

The epilepsy chapter read a seizure as the network crossing an over-synchronisation threshold on the global order parameter R: when coupling pushes the population past a critical point, the micro-eddies that normally run in parallel lock into one runaway rhythm. That chapter proved the pole and the corrective sign — the therapeutic push is whatever raises the over-sync threshold — but it left that push abstract: it said raise the threshold without saying which genes or channels actually realise the raise. This chapter fills exactly that gap, with the same piece of inherited technology that the bipolar levers chapter used: the threshold-shift intervention logic from the analgesic reproducibility package (Zenodo 10.5281/zenodo.20733420). Its premise is that a symptom is a firing-threshold crossing and an intervention is a controlled upward shift of that threshold, reachable exactly three ways: L1 reduce the inward (excitatory) current, L2 increase the outward (potassium) current, L3 remove the up-stream sensitising drive. Applied to epilepsy on the same R19 substrate the whole atlas rests on, the abstract threshold-raise resolves into a three-lever target map over sixteen epilepsy excitability genes: the voltage-gated Na⁺/Ca²⁺/NMDA set on L1 (SCN1A, SCN2A, SCN8A, CACNA1A, CACNA1H, GRIN2A), the KCNQ / KCNB / KCNA / KCNT potassium set on L2 — the dominant lever here, because the KCNQ2 / KCNQ3 M-current is the textbook anticonvulsant axis (the retigabine target) — with the inhibitory GABA‑A pair GABRG2 / GABRA1 riding L2-adjacent, and the mTOR set on L3 (DEPDC5, TSC1, TSC2). Each gene is placed by reading its own promoter switch stiffness — γ = −mean nearest-neighbour stacking free energy (SantaLucia 1998) over its promoter window, turned into |h_sp| = spinodal(γ) with the frozen engine read-only, and five of the reads (KCNQ2, KCNQ3, KCNB1, SCN2A, GRIN2A) are carried over verbatim from the bipolar cache. Two honest caveats are recorded, not hidden: KCNT1 is sign-inverted (its gain-of-function is the pathology), and the L1 sodium-block direction is contraindicated in Dravet / SCN1A loss-of-function. Three fail-closed disciplines ride along: every L3 link stays graded [O] cited biology, a forbidden-claim scanner rejects any dose / efficacy / safety / synthesis statement (including seizure-freedom language), and a burden-weighted ranking orders targets, never drugs or doses. The firewall is absolute: the promoter |h_sp| is a gene's own switch stiffness, never the §25 network over-sync threshold on R, a channel voltage, a potency, a dose, or a clinical effect. efficacy = 0; not medical advice; the hard problem stays open.

the §25 over-synchronisation threshold-raise decomposed into a DNA-grounded three-lever target map = the abstract over-sync-threshold-raising push of §25 is resolved into three concrete levers (L1 reduce inward Na/Ca/NMDA current, L2 increase outward K+ current -- the dominant lever, the retigabine M-current axis, L3 remove the upstream mTOR drive) over 16 epilepsy excitability genes placed by their own promoter switch stiffness; targets are ranked, never drugs or doses — the epilepsy chapter (§25) proved seizures are the network crossing an over-synchronisation threshold on the global order parameter R and that the corrective SIGN is threshold-raising, but it treated that push as ONE undifferentiated operator -- it never said WHICH genes or channels realise the raise. This module applies the SAME threshold-shift intervention logic the bipolar levers chapter used (inherited from the analgesic reproducibility package, Zenodo 10.5281/zenodo.20733420) to DECOMPOSE that single operator into a three-lever target map on the SAME R19 substrate, with NO new mechanism and NO new tuned constant. A symptom is a firing-threshold crossing and an intervention is a controlled UPWARD shift of that threshold, reachable three ways: (L1) reduce the inward, excitatory current -- the voltage-gated Na+/Ca2+/NMDA set SCN1A, SCN2A, SCN8A, CACNA1A, CACNA1H, GRIN2A; (L2, DOMINANT) increase the outward, repolarising K+ current -- the KCNQ2/KCNQ3 M-current pair (the retigabine target), KCNB1, KCNA1 and KCNT1, with the inhibitory GABA-A pair GABRG2/GABRA1 L2-adjacent; (L3) remove the up-stream mTOR drive -- the repressors DEPDC5, TSC1, TSC2 (the everolimus direction). Each gene's gamma = -mean(nearest-neighbour stacking dG, SantaLucia 1998) is read from its OWN promoter window (TSS-2000..+500, Homo sapiens) and turned into that promoter's switch stiffness |h_sp| = spinodal(gamma) and barrier = gamma^2/4 using the frozen engine READ-ONLY -- the identical pipeline the analgesic and bipolar packages ran, and five reads (KCNQ2, KCNQ3, KCNB1, SCN2A, GRIN2A) are carried over VERBATIM from the bipolar promoter cache (gamma is strand-symmetric). Two honest caveats are recorded, not hidden: KCNT1 is sign-INVERTED (its gain-of-function is the pathology) and the L1 Na-block direction is CONTRAINDICATED in Dravet/SCN1A loss-of-function -- exactly why gamma is graded [V] for promoter STRUCTURE only, trait-blind, with clinical direction [O]. Three fail-closed disciplines ride along: an L3-honesty gate keeps every mTOR link graded [O] cited biology (never derived); a forbidden-claim scanner rejects any dose, efficacy, safety or synthesis statement incl. seizure-freedom language (with a planted self-test that must fire); and a burden-weighted prioritisation ranks TARGETS (SCN1A tops the genetic-burden tier but is flagged not-actionable by the generic sign, KCNQ2 and the TSC/GATOR1 mTOR set follow) with the gamma read carried alongside as structural context but NEVER folded into the score -- the stiffest promoter CACNA1H sits LOW on priority while the top-priority SCN1A reads near the softest, the decoupling that makes the firewall visible. FIREWALL (non-negotiable): the promoter |h_sp| is the gene's own switch stiffness, NEVER equated with the sec.25 network over-sync threshold on R, nor with a channel's activation voltage, a drug's potency, a dose, an in-vivo selectivity, or a clinical effect. Registered in run_all_atlas.py as the 9th atlas citizen (EPI-T2a-L), reproducing bit-for-bit with engine byte-unchanged. medium_efficacy_tested=0; ranks targets not drugs; not medical advice; hard problem OPEN. [V map · O links]. this chapter

epilepsy = the over-synchronisation pole; the ictal state = collapse of the selective gate = excitatory bias drives R to the over-sync ceiling and all assemblies ignite; an anticonvulsant reverses it and raises the threshold — epilepsy is the over-synchronisation pole of the engine's synchrony axis (the global Kuramoto order parameter R on the measured ephaptic kernel, plus the M3 R19 ignition gate). An excitatory/disinhibitory E/I bias raises R monotonically toward the over-sync ceiling (0.422 > health 0.390); past a critical bias (+0.3) the selective single-winner gate collapses and EVERY candidate assembly ignites — the ictal state, a loss of GATING (Axis-A firewall: a mechanism boundary, not a claim about ictal experience; consciousness_claim=0). On one axis: autism-T (under-ignited) < health (selective) < schizophrenia (aberrant, some irrelevant) < epilepsy (all ignited). An inhibitory / threshold-raising anticonvulsant-class push moves R back down and restores the gate — it raises the seizure threshold by moving the operating point away from the over-sync ceiling (the same ceiling the θ-cap must stay below, §20). The static susceptibility is characterised; the ictal time-course is OWED to a state-switching layer (E2). efficacy=0; not medical advice. [V mech]. canonical §25

What §25 left abstract

The epilepsy chapter placed the disorder at the over-synchronisation pole of the atlas: health is a population of micro-eddies running in parallel at a moderate global coherence, and a seizure is that population crossing a critical coupling threshold into one runaway, fully phase-locked rhythm — the order parameter R climbing above the band health lives in. That chapter proved two things and stopped: the pole (epilepsy is over-synchronisation, the mirror of the under-coordination disorders) and the corrective sign (the therapeutic push is whatever raises the over-sync threshold, so the network has farther to travel before it locks). That sign is forced and honest, but it describes the intervention as a single abstract operator: raise the threshold. It does not say by what physical handle the threshold is raised — which current, which channel, which gene. A mechanism atlas should be able to say more than something raises the threshold; it should enumerate the ways a threshold can be raised and attach real molecular targets to each. The bipolar chapter showed that this enumeration is exactly what the inherited threshold-shift technology supplies — and epilepsy is where that technology fits most naturally of all, because the over-sync threshold is literally an excitability threshold.

The inherited technology, applied a second time

The handle is not invented here, and it is not even adapted here — it is the same logic the previous chapter inherited from the analgesic reproducibility package (Zenodo 10.5281/zenodo.20733420), now turned on a third problem. Its premise is general: if a symptom is a firing-threshold crossing, then an intervention is a controlled upward shift of that threshold, and a threshold is set by a balance of currents, so there are exactly three levers on it. L1 — reduce the inward, excitatory current that drives the system toward the crossing. L2 — increase the outward, repolarising current that pulls it back. L3 — remove the up-stream sensitising drive that lowers the threshold in the first place. The frame applies unchanged because the epilepsy chapter, the bipolar chapter, and the analgesic package all share the same R19 substrate — the engine's supercritical pitchfork ṣ = g·s − s³ + h, whose spinodal fold IS the switching barrier. There is, however, one difference of emphasis that epilepsy makes vivid: of the three levers, L2 is the dominant one here. The single most mechanistically transparent anticonvulsant principle — opening the neuronal M-current to brake excitability — is an L2 move, and it has a named molecular precedent in the KCNQ2/KCNQ3 channels. Where bipolar disorder leaned on L1 (its replicated GWAS loci are calcium channels), epilepsy leans on L2.

The reuse is literal at the level of code, not just analogy. The pipeline reads each gene's promoter on a fixed window and turns it into a switch stiffness through the engine's own spinodal and barrier functions; this chapter runs that identical pipeline on the epilepsy genes, and five of the reads — KCNQ2, KCNQ3, KCNB1, SCN2A, GRIN2A — are carried over verbatim from the bipolar promoter cache: the same gene, the same promoter window, the same number, because γ is a strand-symmetric property of the sequence and does not change between problems. Nothing about the engine is touched; the module re-emerges the frozen tree read-only and confirms it byte-unchanged, and registers as the ninth atlas citizen (EPI-T2a-L).

L1 — reduce the inward sodium / calcium / NMDA current

The first lever lowers the drive toward the over-sync crossing. Epilepsy genetics populates it densely, because the monogenic epilepsies are dominated by voltage-gated ion-channel defects. Three sodium channels sit here: SCN1A (Na_V1.1, the most-replicated monogenic epilepsy gene, central to Dravet syndrome), SCN2A (Na_V1.2) and SCN8A (Na_V1.6, gain-of-function epileptic encephalopathy). Two calcium channels extend it: CACNA1A (Ca_V2.1, the P/Q-type channel implicated in absence epilepsy and episodic ataxia) and CACNA1H (Ca_V3.2, the low-threshold T-type channel of the thalamocortical absence rhythm). GRIN2A carries the NMDA-receptor (glutamatergic, calcium-permeable) side, central to the epilepsy-aphasia spectrum. The mechanism DIRECTION on this lever is the textbook one: agents that reduce voltage-gated inward current — the sodium-channel-blocking anticonvulsants on the Na side, the T-type blockade that defines the absence-seizure drugs on the Ca side — are threshold-raising in the L1 sense. But here the model records a caveat it must not hide. In Dravet syndrome the SCN1A defect is a loss-of-function in inhibitory interneurons, so a blunt sodium-channel blocker — the generic L1 move — worsens the disorder and is clinically contraindicated. This is precisely why the promoter read γ is graded [V] for structure only, trait-blind: it places the gene on the lever by the stiffness of its promoter switch, and says nothing about the sign of the clinical intervention, which is gene-, variant-, and cell-type-specific and stays graded [O]. A direction, never a dose — efficacy = 0.

L2 — increase the outward potassium current (the dominant lever)

The second lever raises the pull back from the crossing by strengthening the outward, repolarising potassium current — and in epilepsy this is the lever with the clearest mechanism of all. Four channels carry it. KCNQ2 and KCNQ3 (K_V7.2 / K_V7.3) are the neuronal M-current pair — the canonical excitability brake, mutated in benign familial neonatal epilepsy and in severe KCNQ2 encephalopathy, and the target of retigabine, the one marketed anticonvulsant whose entire mechanism is opening a potassium channel. That is the named molecular precedent that makes L2 the dominant lever here: an M-current opener is threshold-raising in the L2 sense, the mirror image of the L1 inward-current reduction, and it is the cleanest example in the whole anticonvulsant pharmacopoeia. KCNB1 (K_V2.1) is the major somatic delayed-rectifier; KCNA1 (K_V1.1) carries the episodic-ataxia-with-epilepsy phenotype. The fourth channel forces a second honest caveat: KCNT1 (the sodium-activated potassium channel Slack) is on the potassium lever by biophysics, but its epilepsy variants are gain-of-function — too much of this outward current is the pathology in malignant migrating partial seizures of infancy — so its therapeutic direction is inverted relative to the generic L2 move (the precedent here, quinidine, is a channel blocker). The model carries that inverted sign explicitly rather than smoothing it away. Riding L2-adjacent are the inhibitory GABA‑A subunits GABRG2 (the γ2 subunit of GEFS+ / febrile seizures) and GABRA1 (the α1 subunit of juvenile myoclonic epilepsy): they do not carry a potassium current, but they raise the same threshold by increasing inhibitory chloride conductance — the benzodiazepine / barbiturate axis — which is the L2 logic (strengthen the restoring pull) realised through a different ion. Every claim on this lever is a mechanism direction only; no opener, blocker, dose, or patient is named.

L3 — remove the upstream mTOR drive (the [O] lever)

The third lever does not touch the channel at all; it removes the up-stream drive that keeps the threshold low. In a large and growing class of epilepsies that drive is a single signalling pathway: mTOR. When the mTORC1 complex is disinhibited, neurons and cortical architecture become chronically hyperexcitable — the mTORopathies. Three genes anchor this lever, all of them repressors of mTOR whose loss releases the drive: DEPDC5 (a subunit of the GATOR1 complex, the leading cause of familial focal epilepsy and a frequent driver of focal cortical dysplasia), and TSC1 (hamartin) and TSC2 (tuberin), the two subunits of the tuberous-sclerosis complex, whose loss causes the severe, often drug-resistant epilepsy of tuberous sclerosis. The mechanism DIRECTION here has a named precedent — mTOR inhibition, the everolimus principle — but this is the lever that demands the most discipline, and it gets it. Every L3 link is held at grade [O] — open, cited biology, never derived from the substrate. The model does not claim to compute that mTOR sets the over-sync threshold; it records, as cited upstream biology, that this is the drive whose removal would raise it. A fail-closed L3-honesty gate enforces exactly this: it checks that the declared L3 set is present, that each member is graded [O] and marked not-derived, that no mTOR-pathway gene is mislabelled as a current-carrying ion channel, and that the firewall sentence separating the promoter |h_sp| from the §25 network over-sync threshold is present — and it FAILS the build if any of these slips.

The DNA grounding: a promoter's own switch stiffness

What places each of the sixteen genes is not a list but a read. For every gene, the module takes its promoter window (transcription start −2000 to +500 bases, Homo sapiens) and computes γ = −mean of the nearest-neighbour base-stacking free energies along that window (the SantaLucia 1998 nearest-neighbour thermodynamics), then turns that γ into the promoter's switch stiffness through the frozen engine's own functions: |h_sp| = spinodal(γ) = 2(γ/3)^1.5 and barrier = γ²/4. The reads span a real range — the stiffest promoter in the set is CACNA1H at γ ≈ 1.64 (|h_sp| ≈ 0.81, the T-type calcium channel), and the softest is SCN2A at γ ≈ 1.20 (|h_sp| ≈ 0.50) — with the M-current brake KCNQ2 reading stiff (γ ≈ 1.58, |h_sp| ≈ 0.76) and the Dravet gene SCN1A reading near the soft end (γ ≈ 1.25, |h_sp| ≈ 0.53). These are read on the same R19 substrate, with the same engine, that the bipolar and analgesic packages used, which is the whole point of the inheritance: one substrate, one pipeline, three problems. The γ read is a property of the gene's promoter sequence, and that is all it is.

Ranking targets, and the firewall that keeps gamma honest

The last component prioritises, and it prioritises targets, never drugs or doses. A burden-weighted score combines three declared, cited weights — clinical burden (0.40), unmet need (0.35), and genetic-evidence / druggability (0.25) — on cited 1–5 tiers, and ranks the genes by that score alone. SCN1A tops the list (Dravet, high SUDEP burden, the most-replicated monogenic gene), followed by KCNQ2 and the tuberous-sclerosis / GATOR1 mTOR set; SCN1A is also flagged not actionable by the generic lever sign, carrying its Dravet contraindication right in the ranking. Crucially, the γ read is carried alongside each target as structural context but is never folded into the score — and the result is a clean demonstration of the firewall: the priority ranking and the γ / |h_sp| ranking are decoupled. The stiffest promoter read in the whole set, CACNA1H (|h_sp| ≈ 0.81), sits near the bottom of priority; the top-priority target SCN1A has one of the softest reads in the set (|h_sp| ≈ 0.53). If promoter stiffness drove the ranking, neither could sit where it does. That decoupling is the firewall made visible, and it must be stated once more in full: the promoter |h_sp| is a gene's own switch stiffness, and it is never equated with the §25 network over-synchronisation threshold on R, nor with a channel's activation voltage, a compound's potency, a dose, an in-vivo selectivity, or any clinical effect. A fail-closed forbidden-claim scanner guards the whole package: it scans the written results for any dose, efficacy-as-fact, safety-as-fact, or synthesis statement — including seizure-freedom and stops-seizures language — carries a negation guard, and includes a planted self-test that must fire on its own bait, failing the build if it ever does not. This module reproduces bit-for-bit with the engine byte-unchanged.

Everything here is an in-silico reading of promoter sequence and a frame for organising targets, not a clinical measure, a diagnosis, or a prescription. The model asserts mechanism directions and target placements — a threshold can be raised three ways; these sixteen genes populate the three levers; these targets carry the highest genetic burden — and nothing about which agent acts on any lever, at what dose, in whom, whether any real drug raises anyone's seizure threshold, or that anyone should change a treatment. The agents named as directions (the sodium-channel blockers and T-type blockers on L1, retigabine via the M-current on L2, everolimus via mTOR on L3) are illustrations of a sign, never a recommendation — and the two recorded caveats (the Dravet contraindication on SCN1A, the inverted sign on KCNT1) are there precisely because a generic lever sign is not a clinical direction. Real epilepsy is heterogeneous, frequently polygenic, and often drug-resistant, and that heterogeneity is locked. A promoter read and a lever assignment are mechanism boundaries, not a claim about the felt quality of a seizure or its aftermath (Axis-A firewall — consciousness_claim = 0, the hard problem stays open). This is not medical advice, not a diagnosis, not a treatment protocol, and not a cure. medium_efficacy_tested = 0; targets ranked, never drugs or doses.