The pressure valve: sleep, mitochondria and the alarm clock headache

This is a synthesis, not a discovery. Nobody has run the experiment that would prove these three papers are describing the same biology, and I want to say that before anything else. What follows is a hypothesis built from reading a fly-sleep paper in Nature next to two human genetic studies of cluster headache published within months of it. The excitement is real. So are the caveats, and I have tried to give them equal space.

The clue cluster headache has never fully explained

Cluster headache is famous among headache disorders for behaving like clockwork. Attacks cluster at the same time of day, often the same time of night, frequently enough that patients call it the "alarm clock headache." That is not folklore. In 1970, Dexter and Weitzman monitored sleep architecture during headache attacks and found cluster and migraine attacks disproportionately began around REM sleep onset, one of the earliest hints that whatever drives an attack is entangled with the sleep-wake system itself, not just running on a 24-hour timer next to it.

Brain imaging gave the clockwork idea an anatomical home. In 1998, May and colleagues used PET imaging during active attacks and found activation specifically in the posterior hypothalamic grey matter, on the same side as the pain. That finding is the reason posterior hypothalamic deep brain stimulation exists as a treatment today, and the reason cluster headache is usually described as a hypothalamic disorder with trigeminal-autonomic output, rather than a simple vascular headache.

The obvious next move was to look for the gene behind that clock. In 2004, Rainero and colleagues reported that a polymorphism in the hypocretin (orexin) receptor 2 gene, HCRTR2, was associated with cluster headache risk. Orexin neurons are hypothalamic, wake-promoting, and sit directly inside the circuitry that negotiates sleep and wake. It was a satisfying story: a wake-promoting neuropeptide receptor, in the same hypothalamic neighbourhood already implicated by imaging, associated with a headache disorder famous for its sleep-linked timing. A 2007 meta-analysis by the same group, pooling nearly 600 cases across Europe, found the association held up. But the field kept looking, and a 2021 systematic field synopsis by Cargnin and colleagues, applying stricter statistical correction across all the HCRTR2 variants examined to date, concluded the evidence did not clear the bar for a credible, replicated finding. Twenty years on, the orexin story remains suggestive, not settled.

That is the state cluster headache genetics was in when two new things happened almost at once: a fly-sleep lab published a mechanism for how sleep pressure is actually built at the molecular level, and two independent groups ran large-scale human genetic studies asking what chemistry cluster headache risk is actually made of.

What the fly brain just showed us

In September 2025, Sarnataro, Velasco, Monaco, Kempf and Miesenböck, working out of Oxford's Centre for Neural Circuits and Behaviour, published Mitochondrial origins of the pressure to sleep in Nature. The paper studies a small population of neurons in the fly brain called dorsal fan-shaped body neurons, or dFBNs, which act as the fly's sleep-control switch. These neurons are marked by three things: the dopamine receptor Dop1R2, the serotonin receptor 5-HT2B, and the neuropeptide Allatostatin-A.

The key experiment was simple to describe and hard to do: keep flies awake overnight, then look at what actually changed inside these sleep-control neurons. The transcripts that shifted were not the genes you would guess from a "sleep gene" search. They pointed almost exclusively at two processes: mitochondrial energy metabolism and synaptic transmission. In plain terms, the harder these neurons had to work to keep the fly awake, the more their mitochondria showed signs of strain, and that strain is what the paper argues actually is sleep pressure, not just a downstream symptom of it. Sleep pressure, on this account, is closer to an oxidative exhaust gauge building inside a specific switch than a chemical countdown clock.

Wake, hours awake

Mitochondrial strain in dFB neurons

Dop1R2 / 5-HT2B / AstA

Sleep pressure signal

Sleep switch fires

Pressure discharges

What two new human genetic studies just showed us

Around the same window, two groups asked a different question in humans: not "what does sleep pressure run on," but "what chemistry actually raises or lowers cluster headache risk." Both used Mendelian randomization, a method that treats naturally occurring genetic variants as a kind of built-in randomized trial, letting researchers test for causal direction rather than just correlation.

Yu and colleagues ran a bidirectional Mendelian randomization analysis across 338 cerebrospinal fluid metabolites against genetic data for 1,833 cluster headache cases and 498,515 controls, published in the Journal of Pain Research. In the forward direction, eleven CSF metabolites showed significant associations with cluster headache risk, with the strongest effects seen for orotate, betaine and 5-oxoproline. Biologically, those three sit squarely inside oxidative stress, mitochondrial metabolism and lipid signaling pathways, the same broad neighbourhood the fly paper just implicated in building sleep pressure. In the reverse direction, genetic liability to cluster headache was nominally associated with shifts in eight metabolites, including lysine and kynurenine. Kynurenine is what tryptophan becomes when it is shunted away from serotonin synthesis down a separate metabolic branch, the same amino acid fork that supplies the serotonin receptor pathway the fly's sleep neurons depend on.

Xiong and colleagues took a different angle in the Journal of Headache and Pain, running a proteome-wide Mendelian randomization analysis using plasma protein data from over 54,000 UK Biobank participants against genetic data from 4,043 cluster headache cases and 21,729 controls. Eleven proteins showed significant causal associations, clustering mainly in complement activation and immune/inflammatory regulation, with several overlapping the mechanisms of existing preventive drugs such as verapamil, topiramate and valproic acid. Cortical surface area was found to partially mediate the effect, in the range of 3 to 10 percent, depending on the protein.

Where the threads cross

Put the three papers next to each other and a pattern appears that none of them were looking for. A basic neuroscience paper in a fly brain finds that the pressure to sleep is generated by mitochondrial energy metabolism inside a small population of neurons running on dopamine, serotonin and a neuropeptide code. A human genetics paper, using a completely different method on a completely different species, finds that mitochondrial metabolism, oxidative stress and the serotonin-adjacent kynurenine pathway are causally entangled with cluster headache risk. Neither team was testing the other's hypothesis. They arrived at the same biological currency independently.

This gives the old orexin story a second life, not a replacement. HCRTR2 was always a reasonable guess because orexin neurons are wake-promoting and hypothalamic, sitting in the same neighbourhood the imaging studies lit up. It never replicated cleanly, and the 2021 field synopsis is a fair reason to stay sceptical of any single-gene version of that story. But a single gene was probably always too small a unit to explain a whole clockwork disease. A mitochondrial "pressure gauge" operating across a wake-promoting circuit is a systems-level mechanism, not a single receptor polymorphism, and it would be far more resistant to the kind of small-cohort false starts that have dogged HCRTR2. If cluster headache's clock runs on the same currency that builds sleep pressure in the first place, that is a more durable hypothesis than any one gene was ever going to be.

What this doesn't prove

None of this is confirmed, and it is worth being precise about where the gaps are. Dorsal fan-shaped body neurons are a fly circuit. No confirmed one-to-one human homolog has been mapped, and Allatostatin-A itself has no direct human counterpart, only functional analogues in wake-promoting neuropeptide systems such as orexin. Both Mendelian randomization studies describe their own findings as preliminary: modest sample sizes for a rare disease, associations that need replication in larger and more ethnically diverse cohorts, and forward/reverse signals that are suggestive rather than mechanistically proven at the receptor level. Nobody has measured mitochondrial stress markers in the actual sleep-gating circuitry of a person with cluster headache during an active bout. And the HCRTR2 story is a direct warning against assuming a plausible mechanism will survive its own replication.

What would make this testable

A hypothesis earns its keep by making predictions. Here is what I would want to see:

• Bout versus remission chemistry: CSF or blood oxidative-stress and kynurenine-pathway markers measured in the same patients during an active bout and during remission, checked against how closely attacks track sleep stage.
• A fresh look at orexin, this time at GWAS scale rather than candidate-gene scale, now that there is a specific metabolic mechanism to test rather than one polymorphism in isolation.
• Treatment-response biomarkers: whether melatonin and verapamil, both established cluster headache preventives with plausible mitochondrial or circadian mechanisms, measurably shift these same oxidative or kynurenine-pathway markers in people who respond to them.

Cluster headache's clockwork has always felt like the loudest clue nobody has fully explained. A fly-sleep paper that was never about headache at all, read next to two months-old human genetics papers that were never about sleep mechanism at all, converging on the same handful of molecules, does not prove anything by itself. But it is exactly the kind of quiet correlation worth writing down before someone runs the experiment that actually tests it.