When Oxygen Convulses: The Nervous System's Own Dose Limit

Stylised editorial illustration of a brain in cross-section inside a hyperbaric chamber: a band of amber electrical-like discharge crossing one hemisphere while teal oxygen molecules drift through the surrounding cerebrovascular network, on a deep navy background

The wellness version of hyperbaric oxygen is a story about fuel: oxygen is what tissue runs on, pressure pushes it deeper in, and more of it must mean more benefit.

The clinical literature tells a different story, and it has two halves. On 11 October this blog examined the lung. This is the other half, and it is the half nobody in a wellness lounge wants to discuss, because it has no barrier protecting it and almost no warning system. At a certain partial pressure of oxygen the central nervous system stops waiting for signals and starts generating them. The result is a generalised convulsion, sometimes preceded by nothing at all — no twitching lip, no visual aura, just a patient or a diver who was fine ten seconds ago.

That event, cerebral oxygen toxicity, is why commercial diving tables have numbers in them, why hyperbaric chambers run air breaks, and why a seizure disorder sits on every hyperbaric consent form as a relative contraindication. In two months, a fresh mouse study, two mechanism papers, a human cohort, a systematic review and a revised international exposure guideline have all landed on the same subject — describing a ceiling that is real, dose-dependent, and apparently lowered by repetition.

A Convulsion That Leaves a Trace, and an Explanation That Does Not Hold

The newest of the papers appeared in Behavioural Brain Research on 14 October. Chenglong L and colleagues at Shanghai University’s School of Medicine and the Naval Medical Center of Naval Medical University placed eighteen male C57 mice into three groups inside a pure-oxygen chamber. Six were held at 500 kPa — roughly five atmospheres absolute, the equivalent of 40 metres of seawater — until they convulsed. Six were held at the same 500 kPa for ten minutes without convulsing, and six at 250 kPa for thirty minutes, a milder exposure that also produced no seizure. Before and after, the animals ran a Y-maze, a standard test of spatial working memory, while blood flow was measured in the brain and in the femoral artery.

The group that convulsed showed significantly impaired spatial memory; the non-convulsing hyperbaric groups showed a mixed picture across memory, locomotion and exploratory behaviour. All of it was transient, with the mice largely recovered within several hours. Cerebral blood flow fell sharply straight after exposure — more sharply at 500 kPa — and then drifted back toward baseline.

The striking result is the one in the negative: no significant association emerged between the post-exposure drop in cerebral blood flow and the behavioural deficits. That matters because the standard explanation for hyperoxic brain effects has long leaned on vasoconstriction — oxygen constricts cerebral vessels, perfusion falls, neurons get less blood, and the electroencephalogram changes. Here the researchers watched the vasoconstriction happen and failed to connect it to the behavioural damage. Whatever the convulsion does to memory, it is not simply starving the brain of blood.

The caveats are substantial: a mouse study with six animals per arm, seizure exposures well above anything used clinically (most cleared indications sit between 2.0 and 2.5 atmospheres absolute), and behavioural tests in rodents rather than cognitive assessments in people. What it provides is a mechanistic correction, not a clinical risk estimate. Two of the same investigators also appear on the Shanghai dose audit covered here on 11 October — neighbouring outputs from one naval-medicine programme working two organ systems, not independent groups converging on an answer.

Glutamate, but Not the Receptor Pharmacology Expected

If oxygen seizures have a molecular trigger, the obvious suspect has always been glutamate, the brain’s main excitatory neurotransmitter, and within that system the NMDA receptor, which is the target most epilepsy research points at. A paper in Biomedicines by Alekseeva and colleagues at the Sechenov Institute of Evolutionary Physiology and Biochemistry in St Petersburg tested that assumption directly and came back with an answer that inverts it.

The team exposed male Wistar rats to 100 percent oxygen at 6 ATA until seizure onset, with 86 animals used for seizure monitoring alone. Some received memantine, which blocks NMDA receptors; others perampanel, which blocks AMPA receptors — the other principal glutamate receptor, and the one generally assigned a supporting role in epileptogenesis. A third arm received both, with methionine sulfoximine controls probing glutamate handling from another direction.

Memantine failed completely. Worse than failed: it significantly accelerated seizure onset, and the blockade made the animals convulse sooner. Perampanel did the opposite, delaying seizures in a dose-dependent fashion and preventing cell death in the hippocampal CA1 and CA3 regions. When both drugs were given together, perampanel cancelled out memantine’s proconvulsant effect.

The authors’ conclusion is that fast AMPA receptor activation, not NMDA signalling, is the primary upstream trigger for hyperoxic convulsions — a mechanism that “sharply contrasts with the established priority of NMDA glutamate receptors in epileptogenesis,” as they put it. That is a genuinely surprising claim and needs treating as one: rats at 6 ATA, far beyond clinical practice, with a single exposure model and a limited set of pharmacological probes. Perampanel is a prescription antiepileptic with its own central side effects, and nothing here suggests patients about to enter a chamber should be given it. What the paper does establish is that oxygen seizures are not simply epilepsy with a different cause, and that translating epilepsy pharmacology into the chamber is a mistake.

The Finding That Cuts Against Session Volume

The most commercially awkward result of the past two months comes not from Shanghai but from Duke University’s Center for Hyperbaric Medicine and Environmental Physiology, where Gasier and colleagues published a study in Experimental Neurology on what happens when hyperbaric exposure is repeated rather than performed once.

Mice were exposed to 4.5 ATA oxygen for 60 minutes, then re-exposed at intervals of 48 hours, 72 hours or seven days. Repeated exposure shortened the time to seizure — the animals became more susceptible, not acclimatised. The researchers also tested tiagabine, a drug that increases GABA, the brain’s main inhibitory transmitter, and which delays oxygen-induced seizures in a single exposure. It did not hold: tiagabine lost its antiseizure efficacy in the repeated-exposure setting and failed to preserve presynaptic neurotransmitter content.

Underneath that behavioural result sat changes consistent with a cell running an overdrawn energy budget. Synaptosomal mitochondrial respiration rose early and then fell, with reduced maximal respiration and spare respiratory capacity after repeated exposure, and mitophagy markers shifted in a region- and time-dependent way — reduced mitolysosomal signal at seven days in hippocampal CA3 and the cerebellar granular layer. Spatial transcriptomics showed the response narrowing from broad activation after a single exposure to restricted intracellular and metabolic programmes after repetition. The authors’ summary is that susceptibility to oxygen toxicity is influenced by the recovery interval.

For anyone selling or buying hyperbaric sessions in packages of forty, sixty or more, that sentence deserves to be read twice. The mechanism data points in the direction of longer intervals, not shorter ones, and gives no support whatsoever to the intuitive logic that more frequent exposure accumulates benefit. It does not prove that a 60-session wellness course causes harm. It does mean that the spacing of sessions is a pharmacological variable, not a scheduling convenience.

What Human Data Actually Shows

Human evidence on this question is thin, and the thinness is itself the finding.

The most directly relevant cohort was published in PLOS ONE by Park and colleagues at the University of Toronto and Toronto General Hospital’s Hyperbaric Medicine Unit, who went back through the charts of patients referred for hyperbaric oxygen despite a history of seizures — the group conventionally treated as a relative contraindication. Of 43 patients referred between January 2020 and December 2023, 21 never proceeded to treatment. The remaining 22 completed 634 sessions in monoplace chambers, with five-minute air breaks, averaging 29 sessions each over a range of three to 60, lasting 90 to 120 minutes at pressures of 1.8 ATA, 2.0 ATA or 2.4 ATA. Fifteen were taking oral antiseizure medication during the course.

One seizure occurred across all 634 treatments: an incidence of one in 634. The authors’ conclusion is measured — that seizure history is a genuine risk factor, but that most such patients can be treated safely when predetermined protocols are followed. The limits of that reassurance are the usual ones for a retrospective chart review: a small single-region sample, descriptive statistics, no comparison group, and no way to know what the incidence would have been without antiseizure medication or with different pressures.

What the human literature cannot yet offer is early detection. A systematic review in Diving and Hyperbaric Medicine by Barnes, Hallum and Vrijdag at the University of Auckland went looking for it in the electroencephalogram: 1,025 abstracts screened, 46 papers read in full, ending with 22 articles covering 16 studies of EEG measurement during hyperoxic ventilation in healthy adults. The cohorts were typically small and mostly male non-divers, and the analytical methods scattered across spectral analysis, sensory-evoked potentials, connectivity measures and expert qualitative reading. Severe exposures tended to be reported qualitatively and mild exposures quantitatively — close to a confession that the interesting regime is the hardest to measure. The conclusion is that the field needs a large randomised trial with quantitative measures before real-time monitoring of an impending seizure can be built.

So the physiological prodrome varies, the monitor does not exist, and the protective work therefore falls entirely on fixed exposure limits set decades ago.

The 1991 Limits Were Never Really Tested

Those limits are the subject of the sharpest document in this cluster. In the third-quarter 2025 issue of Diving and Hyperbaric Medicine, Hoyt and colleagues — drawing on the NOAA Diving Program, the Navy Experimental Diving Unit and Université Laval, and including Dr Simon Mitchell — set out to find the experimental evidence behind the inspired-oxygen limits that have governed diving since 1991. They found very little of it. Those limits came from US Navy studies at higher inspired oxygen partial pressures than modern technical divers use, which means the duration limits in the range that actually matters — 1.3 to 1.4 atmospheres — had, in the authors’ words, “relatively little experimental provenance.” Meanwhile contemporary technical dives routinely exceed them during decompression, and anecdotally this happens without a corresponding rise in cerebral oxygen toxicity.

Evidence for a revision turned out to exist for exactly one number: 1.3 atmospheres, the common default setting in constant-partial-pressure rebreathers. The 1991 limit for a single exposure at 1.3 atm was 180 minutes, with a 24-hour maximum of 210 minutes. The committee’s revised recommendation allows dives with up to 240 minutes of working activity followed by up to 240 minutes of resting decompression at that inspired partial pressure, on the basis that recent studies show an acceptably low risk of cerebral oxygen toxicity — a recommendation promulgated and endorsed at a workshop convened by NOAA.

This is where the guideline touches the wellness market. A 1.3 ATA chamber breathing pure oxygen delivers an inspired oxygen partial pressure in the same territory the technical-diving community is debating, and the protective factor documented across this literature is not the word “mild” — it is partial pressure multiplied by time, with air breaks subtracted. A lounge session presented as inherently safe because of its pressure rating rests on a variable the newest evidence suggests is less protective than assumed, in a population of generally healthy adults who have never been studied at all. The revision is also narrower than it may read: a diving recommendation for working technical and scientific divers, not a hyperbaric therapy protocol, and it licenses no longer clinical or commercial sessions.

Where This Leaves the Chamber

None of this changes what hyperbaric oxygen is cleared to treat. The FDA maintains a defined list of indications — decompression sickness, carbon monoxide poisoning, non-healing diabetic wounds, late radiation tissue injury, gas gangrene, severe anaemia, compromised skin grafts and flaps, crush injury, thermal burns, idiopathic sudden sensorineural hearing loss, intracranial abscess — delivered through Class II devices cleared under Product Code CBF. Medicare’s National Coverage Determination 20.29, read directly from the CMS database, limits reimbursement to therapy administered in a chamber, names each covered condition, and requires that diabetic lower-extremity wounds be Wagner grade III or higher and have failed at least 30 days of standard wound therapy.

Cerebral oxygen toxicity appears on no coverage list, because it is not a condition the therapy treats. It is a hazard the therapy produces, and the FDA’s August 2025 letter to health care providers frames the wider safety picture: reports of fires causing serious injuries and deaths, manufacturer instructions to be followed, staff training, continuous patient monitoring, maintenance checks, prohibition of electrical and static-generating items, and clothing restricted to hyperbaric-compatible materials such as cotton.

Nothing yet protects against any of this in humans. The candidate interventions in the literature are preclinical, and both aim at the mitochondrion: an immunosuppressant that pre-activates the clearance of damaged mitochondria in rat neurons, and a caffeine-plus-astaxanthin combination that delayed seizures in mice. Both were tested at inspired oxygen pressures far above clinical practice, both were measured by seizure latency rather than safety, and one carries a disclosed supplement-industry affiliation. The second result in particular will be read as a licence to stack compounds before a session; nothing in it supports that, and there is no human dose to translate.

The honest summary is that under properly run clinical protocols, at 2.0 to 2.5 atmospheres with air breaks and patient screening, oxygen convulsions are uncommon — the Toronto cohort’s one event in 634 sessions is a fair illustration of the order of magnitude. But uncommon is not the same as harmless. The new mouse work suggests the convulsion leaves a transient cognitive bruise that vasoconstriction cannot explain. The Duke work says repeated exposure shortens the fuse. And the exposure limits everyone relies on were, by the admission of the committee that reviewed them, built on thin experimental ground for the very pressure range the commercial hyperbaric market now occupies.

Sources

← Back to blog