Chronic negative cognition keeps the HPA axis switched on, and the HPA axis sits upstream of a great deal. That much is not controversial. What happens next is where the popular version runs a long way ahead of the evidence — so this piece walks the chain link by link and grades each one.
A client sits down and tells me, before we have looked at a single marker, that nothing has ever worked for them and this will not work either.
I have learned to treat that sentence as clinical information. Not as pessimism to be talked out of, and not as a character flaw — as a variable, like ferritin or morning cortisol, that will influence what happens over the next six months.
The reason is that a sustained appraisal of threat is not a purely mental event. It has an endocrine output. And once you have an endocrine output you have a chain of downstream consequences, some of which are measurable in blood.
The question is not whether thought has biochemistry. It is how far down the chain the evidence actually reaches — and the honest answer is: further than sceptics allow, and considerably less far than the mind-body literature tends to claim.
Chronic psychological stress activates the hypothalamic-pituitary-adrenal axis. Perceived threat drives CRH release from the hypothalamus, ACTH from the pituitary, cortisol from the adrenal cortex. Alongside it, the sympathetic branch releases adrenaline and noradrenaline.
Crucially, the trigger is appraisal, not events. The axis responds to what the nervous system judges to be happening. Which is why two people in identical circumstances can produce entirely different cortisol curves, and why rumination about a threat that has already passed can maintain activation as effectively as the threat itself.
That is established physiology. Everything after it needs handling with more care.
Established in illness · extrapolated to daily stress
Most circulating thyroid hormone is T4, which is comparatively inactive. It becomes T3 through the deiodinase enzymes in peripheral tissue. Under physiological stress, type 1 deiodinase activity falls and type 3 rises, shunting T4 towards reverse T3 instead of active T3.
This is real and well characterised — as non-thyroidal illness syndrome. It is documented in critical illness, sepsis, surgery, starvation and severe caloric restriction. Glucocorticoids do inhibit deiodinase activity, so the mechanism connecting cortisol to reduced T3 is not invented.
The extrapolation to ordinary chronic psychological stress in an otherwise healthy person is where I would put a marker down. Whether a demanding job and poor sleep produce a clinically meaningful shift in conversion is not established with anything like the confidence usually implied.
The functional medicine convention of treating TSH between 2.5 and 4.5 as a subclinical problem requiring intervention does not survive scrutiny. The TRUST trial, along with several others, found no meaningful symptomatic benefit from treating mild subclinical elevation in older adults.
I taught the narrower range for years. I no longer do. Whatever stress does to thyroid physiology, it does not licence treating a number inside the reference range as a diagnosis.
Directionally supported · not established
The magnesium claim is usually stated far too confidently. What actually exists is a hypothesis — the “vicious circle” concept, revisited by Pickering and colleagues in 2020 — that stress increases magnesium loss and low magnesium increases stress susceptibility, each feeding the other. Supporting it: pre-clinical work, mechanistic plausibility, and observational studies finding low magnesium status in stressed populations.
What is missing is the demonstration that psychological stress causes clinically meaningful depletion in humans. Lopresti's 2019 review in Advances in Nutrition put it fairly: the bulk of evidence suggests stress can affect micronutrient concentrations, often downward, but the effects of stressor type, severity, and acute versus chronic exposure all remain to be worked out.
There is also a measurement trap here. Serum magnesium is a poor status marker — roughly one percent of body magnesium is extracellular, and the body defends serum concentration at the expense of tissue stores. The trials that found benefit used erythrocyte magnesium. If you are going to test, test the right compartment.
Zinc needs a more precise correction. Acute stress and inflammation genuinely do lower serum zinc — but through metallothionein-mediated redistribution into the liver, not through depletion. The zinc has moved, not gone. Reading a low serum zinc during an inflammatory episode as a nutritional deficiency is a straightforward misreading.
Demonstrated for acute stress · unresolved for chronic
This is the link with the best human data, and it is more interesting than the usual telling.
CRH acts peripherally as well as centrally. It activates mucosal mast cells, which degranulate and increase permeability. And the human demonstration of this is unusually complete.
Twenty-three healthy volunteers, small intestinal permeability measured by lactulose-mannitol excretion, across four conditions. Public speaking increased permeability (0.059 against a control of 0.030). Anticipation of electric shocks did not.
And the effect appeared only in those whose cortisol actually rose. Subgroup analysis found the permeability change was confined to subjects with a significant cortisol elevation — which is about as direct a link between the endocrine response and the barrier as you could ask for.
Then the mechanism was tested rather than assumed. Intravenous CRH reproduced the effect. Pre-treatment with disodium cromoglycate — a mast cell stabiliser — abolished it, and abolished the effect of the public speech too. Stress raises cortisol, CRH activates mast cells, mast cells open the barrier, and blocking the mast cells blocks the whole thing.
That is not an inference from animal work. It is the chain, demonstrated forwards and then interrupted, in humans. Later work has added detail — Gerdin's 2022 study found the change is specifically paracellular, and Mogilevski's group confirmed acute barrier disruption after intravenous CRH.
So acute stress does this. It is shown, not assumed.
But a 2023 review in Neurobiology of Stress reached a more sobering conclusion about the chronic case: despite strong in vitro and animal evidence, studies have not produced unequivocal evidence of the relationship in healthy humans. And a study of US Marines through survival training — about as severe as combined physical and psychological stress gets — found no evidence of increased intestinal permeability at all, though blood-brain barrier markers did rise.
Which is a useful corrective. Acute laboratory stress opens the barrier. Whether chronic life stress does, and by how much, and in whom, is genuinely open.
Well supported
This is the strongest link in the chain and it has been strong for thirty years. Kiecolt-Glaser and Glaser's work on medical students at exam time, on dementia caregivers, on the recently bereaved and on marital conflict established repeatedly that sustained psychological stress reduces natural killer cell cytotoxicity and impairs several other measures of cell-mediated immunity.
The pattern is directional in a way worth noting: acute stress mobilises immune cells and can transiently enhance certain defences. It is chronic activation that suppresses. The two get collapsed together constantly, and they are not the same phenomenon.
Mechanism certain · magnitude in ordinary stress uncertain
Cortisol is a glucocorticoid; raising blood glucose is its job description. It drives hepatic gluconeogenesis, opposes insulin at peripheral tissue, and promotes visceral fat deposition. Cushing's syndrome makes the endpoint unmistakable.
What is less certain is the dose-response at the levels produced by everyday chronic stress. Observational data links chronic work stress to metabolic syndrome; disentangling the cortisol contribution from the sleep loss, the alcohol, the missed meals and the abandoned exercise that travel with it is much harder. The mechanism is not in doubt. Its independent contribution is.
Certain for drug doses · weaker for endogenous stress
Glucocorticoid-induced osteoporosis from prescribed steroids is one of the best-documented drug adverse effects there is. Glucocorticoids suppress osteoblast function and lengthen osteoclast survival, and bone loss is fastest in the first months of treatment.
Whether endogenous stress-driven cortisol does the same thing is a much weaker claim. Meta-analyses do find lower bone mineral density in people with depression, which is suggestive. But depression brings reduced physical activity, poorer nutrition, higher smoking rates and antidepressant exposure with it. Extrapolating from prednisolone to a difficult year is not warranted on current evidence, and I would not put it to a client as established.
Cognition to HPA activation: solid. HPA to immune suppression: solid. HPA to gut barrier under acute stress: demonstrated in humans. HPA to insulin resistance: mechanism certain, magnitude unclear.
Thyroid conversion, mineral status and bone: plausible, partly extrapolated from populations that are not the person in front of you. A chain is not made stronger by drawing more arrows on it. Four good links are worth more than eight assumed ones.
The placebo effect gets all the attention. Its mirror image is better evidence for the thesis of this article, because it is measurable and it is unwelcome.
People told a side effect is likely get that side effect at markedly higher rates. In statin trials, the phenomenon has been quantified: participants report muscle symptoms when they know they might be taking a statin, and those symptoms largely disappear under blinding. Beta-blocker trials show erectile dysfunction rates rising when the effect is disclosed in advance. Warned about nausea, more people vomit.
The mechanisms are partly characterised — anticipatory anxiety, CCK-mediated pain facilitation, conditioned learning — and can be blocked pharmacologically, which is about as direct a demonstration as you could ask for that an expectation is running through a physiological pathway.
The clinical implication is uncomfortable and unavoidable: how a practitioner frames an intervention is part of the intervention. Handing someone a test result with an alarmed expression measurably changes what happens next. This is a reason for accuracy, not for false reassurance — but it is a reason to be deliberate about language.
Three names come up whenever this territory is discussed, and they deserve differentiating rather than bundling.
Bessel van der Kolk did the field a service by insisting trauma has a physiology and is not purely a narrative problem. The autonomic and interoceptive findings he draws on are real, and body-based approaches have moved from fringe to mainstream partly through his advocacy.
But The Body Keeps the Score has now had the systematic examination it went a decade without. Michael Scheeringa, a child psychiatrist at Tulane with a long research career in PTSD, published a review in BJPsych Bulletin in 2026 assessing 122 discrete claims from the book — 42 on neurobiology, 51 on treatment efficacy and 29 on other topics — and found repeated instances where the cited evidence was incomplete, selectively presented, or inconsistent with the wider literature. His summary of the two central propositions is that trauma causes lasting neurobiological damage and that body-based treatments are uniquely effective, and that both have been widely embraced while receiving remarkably little systematic evaluation.
Two things to hold alongside that. The paper sits in the journal's “Against the Stream” section, which exists explicitly for contrarian commentary, and Scheeringa has a self-published book making the same argument — he holds a position, which does not make him wrong but is worth knowing. And van der Kolk's defenders make a fair point: he never literally claimed trauma is stored in muscle tissue, and his actual argument concerns nervous system dysregulation.
Which is precisely the gap worth naming. The distance between what a book argues and what its popular reading becomes is where the clinical misuse happens — and in this case the popular reading has travelled a good deal further than the text.
Babette Rothschild — and it is Babette, not Barbara; the misattribution is common enough to be worth correcting — wrote The Body Remembers, and is notably more careful. She has been explicitly critical of overreach in the trauma field, including of the idea that memory is stored in tissue in any literal sense.
Gabor Maté is the one I would handle with the most caution, and I say that as someone who finds his clinical observations frequently acute. The compassion in his work is real and his account of how adversity shapes physiology is often useful. But the stronger version of the thesis — that particular emotional patterns cause particular diseases, that suppressed anger contributes to cancer or that a specific personality type predisposes to autoimmunity — is not supported. The “type C” cancer-prone personality literature has largely failed to replicate. And there is a cost to this framing that is not hypothetical: it hands sick people responsibility for their illness.
Supported: adverse childhood experiences are associated with worse adult health outcomes across many domains, with a dose-response relationship. Chronic stress alters immune and endocrine function. Trauma affects autonomic regulation, and body-based approaches help some people considerably.
Not supported: that memories are stored in specific tissues, that a given emotional conflict produces a given disease, or that unresolved psychological material is the primary cause of chronic illness. The first set is a reason to take the psychological seriously. The second is a reason not to let it become a theory of everything.
If the sympathetic and HPA arms drive the chain, the parasympathetic arm is the obvious lever — and here there is a piece of human evidence that closes the loop rather neatly.
Mogilevski and colleagues gave healthy volunteers intravenous CRH to induce the barrier dysfunction described above, then applied transcutaneous auricular vagal nerve stimulation or sham. Vagal stimulation cut the lactulose-to-mannitol ratio by roughly half compared with sham. Sixteen participants, crossover design — small, but a direct human demonstration that raising vagal activity protects the gut barrier against a stress signal.
That is a mechanism with a measurable input and a measurable output, which is more than most of this field offers.
Most of what you will read about vagal tone is framed through polyvagal theory, so it matters that in early 2026 thirty-nine specialists in vagal physiology and vertebrate evolution — many previously cited in the polyvagal literature as supporting it — published a joint evaluation in Clinical Neuropsychiatry concluding the theory is untenable on existing neurophysiological and evolutionary evidence. Porges replied at length in the same issue.
Two caveats in fairness. Paul Grossman selected the co-authors and has been a critic of the theory since 2007, so this is not a neutral panel. And Porges argues the critique misrepresents what the theory claims.
But here is the part that matters clinically, and it gets lost in the noise. The critique targets the evolutionary and neurophysiological foundations — not the clinical practices. Slow breathing does not stop working because the explanatory story around it is contested. The interventions below have their own evidence, independent of the theory used to justify them. “Polyvagal theory is untenable, therefore breathwork is worthless” is a non-sequitur, and one worth refusing in both directions.
Slow breathing at around six breaths per minute. The best-supported intervention available, and free. It works by aligning breathing with the baroreflex resonance frequency, maximising respiratory sinus arrhythmia. Note that this increases HRV amplitude during practice; whether regular practice raises resting HRV is a separate and less settled question.
Extended exhalation. Vagal outflow to the heart rises during expiration. Making the out-breath longer than the in-breath biases the balance parasympathetically within a few breaths.
Regular aerobic exercise. The most reliable way to raise resting HRV over months, with far better evidence than most of the devices sold for the purpose.
Sleep. Unglamorous and upstream of everything else here. Sleep restriction alone reproduces a substantial part of the stress physiology described above.
Transcutaneous vagal stimulation. Genuine mechanism, accumulating evidence, mostly small trials. Interesting rather than established.
What I would not oversell: cold exposure, humming, gargling and the rest of the popular vagal toolkit. Some produce an acute autonomic shift. Whether that shift accumulates into anything durable is not demonstrated, and the confident claims made for them are not proportionate to the data.
Here is why any of this matters at a desk rather than in a review paper.
I run a five-test panel. It produces a great deal of information, and it produces it regardless of whether the person receiving it has any current capacity to act on it. Handing a comprehensive protocol to someone in the middle of a bereavement, a divorce or a burnout is not thoroughness. It is a way of generating non-adherence and then attributing it to the client.
What has changed in the last twelve months? Not symptoms — circumstances. Loss, upheaval, caring responsibilities, work.
What has been tried, and what happened? Someone with five failed attempts behind them is not starting from neutral, and the sixth protocol inherits that history whether or not anyone names it.
What does this person expect to happen? Given the nocebo literature, this is a physiological question, not a rhetorical one.
What is realistically available — time, money, attention, cooking capacity? A protocol that assumes resources the person does not have is a protocol designed to fail.
And: does this person need a practitioner or a therapist right now? Sometimes the answer is a therapist, and the honest thing is to say so. Testing is not the appropriate first move for everyone who arrives asking for it.
Where readiness is low, the intervention changes shape. Fewer variables, sleep and food timing first, one thing at a time, and an explicit conversation about what we are not doing yet. That is not a lesser protocol. It is the one with a chance of being followed.
Not that thinking positively fixes disease. Not that illness is a message. Not that the mind causes what the body suffers.
Only this: a sustained pattern of threat appraisal produces a sustained endocrine output, that output has measurable downstream consequences, several of those consequences are visible on the same panels I would run anyway, and a person's expectations measurably influence what an intervention does.
Which means psychological state belongs in the assessment rather than in a separate conversation afterwards. Not as the answer — as one of the variables.
The person who told me nothing would work was, in a narrow sense, giving me a result. It sat alongside a flattened cortisol curve, and it was every bit as relevant to what to do next.
Educational content, not medical advice, and not a substitute for mental health care. If you are struggling with your mental health, that deserves proper support in its own right rather than as an adjunct to a nutrition protocol — speak to your GP or a qualified therapist.
The best protocol is the one you are actually in a position to follow. That is a clinical judgement, and it is worth making out loud.
Ask AIdan → Book a discovery call