Some people never recover from an illness that should have passed. The infection cleared, the injury healed, the crisis ended — and they stayed ill. There is a theory that explains this beautifully. Fifteen years on, the entire human evidence base for treating it is one trial of ten children.
This is the article I have rewritten most times, because I have been on both sides of it.
For several years I found the Cell Danger Response the most satisfying explanation I had encountered for something I see constantly and cannot otherwise account for: the person who had glandular fever at nineteen and has never been the same; the one who had a bad reaction to something and stayed unwell for four years; the one whose tests are all fine and who is plainly, obviously ill.
I no longer teach it as established. But I do not think it is nonsense, and I would rather set out honestly why it is seductive and where it stops being supportable than pretend I never found it convincing.
Robert Naviaux at UC San Diego proposed that cells have an evolutionarily ancient defensive state. Threatened by infection, toxin or injury, a cell shifts metabolism away from ordinary business toward defence — stiffening its membrane, altering what it takes in, and releasing ATP into the extracellular space as a danger signal to its neighbours through purinergic receptors.
This is not controversial. Extracellular ATP as a damage signal is established cell biology.
Naviaux’s proposal is that healing requires this response to switch off in a staged sequence, and that in some people it does not. The threat resolves; the defensive state persists. Cells stay locked in danger mode, and the resulting symptoms — fatigue, poor recovery, sensitivity to everything, an immune system that behaves as though something is still wrong — are not the disease. They are the unfinished response to a disease that has gone.
If that were true, it would explain the whole category. Post-viral syndromes, ME/CFS, long Covid, some chemical sensitivities, some autism presentations. One mechanism, many labels. That is precisely why it is so attractive.
Here is where I would ask you to notice something about the shape of it.
Naviaux’s group treated a maternal immune activation mouse model of autism with suramin, a purinergic antagonist, and reported correction of sixteen multisystem abnormalities — social behaviour, Purkinje cell loss, synaptic ultrastructure, metabolism, receptor expression.
A follow-up showed a single dose reversed the abnormalities in six-month-old adult mice, normalising seventeen of eighteen disturbed metabolic pathways. A third study replicated the finding in the Fragile X model.
And one detail from that work that rarely gets quoted: most of the improvement was lost after five weeks of drug washout. Even in mice, this was suppression rather than resolution.
The Suramin Autism Treatment-1 trial, published in 2017. Ten boys. Five matched pairs, randomised to a single intravenous infusion of suramin or saline.
The results were striking: ADOS-2 comparison scores improved by 1.6 points in the suramin group with an effect size of 2.9, and no change in placebo. Secondary measures moved in language and social interaction.
That was nine years ago. There is still no published larger trial. The authors themselves described it as showing promise "in this small study."
Ten children, one dose, nine years. That gap is the most informative thing in this article, and no amount of mechanistic elegance fills it.
The other pillar was metabolomics — the finding of a distinctive hypometabolic signature in ME/CFS, a measurable fingerprint of the danger state.
A 2020 systematic review examined eleven metabolomic studies in ME/CFS. Its conclusion was blunt: no specific metabolite was consistently impaired across the studies, urine results were inconsistent, and the lack of consistency in study design "provides little evidence for metabolomics to be clearly defined as a contributing factor to the pathogenesis."
That does not mean nothing metabolic is happening. A careful 2022 study tracking 1,157 plasma metabolites through two maximal exercise challenges found genuine, measurable metabolic disruption in ME/CFS patients, most distinctly during the 24-hour recovery window — which is exactly where post-exertional malaise lives. Something real is going on.
But "something metabolic is disrupted" is a much smaller claim than "here is the signature of a specific universal defensive state," and only the first is currently supported.
I wrote last week about claims that are half right: a true mechanism carrying a conclusion nobody could test. This is the cleanest example I know, and it is a considerably better one than the examples I chose then.
What result would prove the Cell Danger Response wrong?
I have thought about this for a long time and I cannot construct one. A normal test result fits — the cells are in a defensive state, not a damaged one. Failure to improve fits — the response has not been switched off. Improvement fits. Relapse fits, as an incomplete healing cycle. Every possible observation is accommodated.
That is not a criticism of Naviaux, who has done careful mechanistic work and is explicit about what is and is not established. It is a criticism of how the framework travels once it leaves his lab — where it becomes an explanation for everything, and therefore a test of nothing.
And it is seductive for a reason that has nothing to do with evidence. It tells someone who has been ill for years and repeatedly told their tests are normal that they are not imagining it, they are not weak, and there is a mechanism. That relief is real and largely deserved. It is also exactly the moment at which people stop asking for evidence, which is why I am being careful with it here rather than dismissive.
There is a competing account of the same phenomena that is quietly accumulating more independent support, and almost nobody in my field talks about it.
Komaroff and colleagues have argued that the symptoms of long Covid, ME/CFS and other post-acute infection syndromes may be produced by sickness behaviour and torpor — ancient, evolutionarily conserved responses to threat, mediated by specific neural circuits that have recently been identified, and activated by neuroinflammation.
Notice how close that is to the Cell Danger Response conceptually: an ancient defensive programme that fails to switch off. The difference is where it locates the mechanism — in identified neural circuits rather than in a universal cellular state — and that difference makes it testable. You can go and look at the circuits.
Their account also names several concrete abnormalities found in both conditions: autoantibodies against neural targets, endothelial dysfunction, acquired mitochondrial dysfunction, a pro-inflammatory gut microbiome. Each of those is individually measurable, and each could be individually wrong.
Not nothing, and not a protocol.
I use the observation and leave the framework. The observation is that some people do not recover from things they should recover from, and that pushing them harder makes it worse. That is real regardless of mechanism, and it changes how I work: smaller interventions, longer timescales, careful attention to whether anything is making them worse.
In post-exertional malaise, the pacing principle is not optional. This is the single most important practical point in the article. PEM is not deconditioning, does not respond to graded loading, and NICE withdrew its recommendation for graded exercise therapy in ME/CFS in 2021 for good reason. Whatever is driving it, the clinical response is the same.
I measure what is measurable. Inflammatory markers, iron status, thyroid function, B12, the kynurenine pathway on organic acids. Not because these prove or disprove anything about danger states, but because correctable things are found this way and frameworks do not find them.
Suramin. It is a century-old antiparasitic with serious toxicity, tested in ten children, once. Anyone offering it outside a trial is doing something I would want no part of.
Protocols sold as “resetting the cell danger response.” There is no established way to do this, no marker confirming it has happened, and no trial showing it helps.
And the polyvagal layer, which usually arrives attached to this framework. In early 2026 thirty-nine specialists in vagal physiology published a joint evaluation concluding the theory is untenable on its neurophysiological foundations. Importantly, that critique targets the explanation, not the practices — slow breathing does not stop working because the story around it is contested. But it should not be presented as settled science, and I have written about that separately.
I have deliberately not resolved this, because I do not think it resolves.
The Cell Danger Response might be substantially right. The animal work is not trivial, the mechanism is real, and the clinical phenomenon it describes is unmistakable to anyone who has sat with these patients. If a properly powered trial arrives and it holds, I will say so plainly and this article will look overcautious.
But fifteen years of mouse studies and ten treated children is not a foundation for a clinical protocol, and the honest position is that we have an elegant hypothesis and an unmet obligation to test it.
The uncomfortable truth is that a theory can be beautiful, mechanistically coherent, consistent with everything you observe in practice — and still not be knowledge. Those are different things, and the gap between them is where most of medicine’s expensive mistakes have lived.
If you have been ill for years and this framework gave you an explanation that finally made sense of it, I am not trying to take that away. I am saying: hold it loosely, keep looking for the correctable things, and be careful of anyone selling you a way to reset something nobody can yet measure.
Educational content, not medical advice. If you have post-exertional malaise, pacing rather than graded exercise is the current clinical guidance, and any exercise decision should be individual and monitored. Persistent unexplained symptoms deserve proper medical investigation regardless of which framework appeals.
Especially here. Frameworks explain; they do not find the correctable thing that has been sitting there for three years.
See testing options Book a discovery call