Preprint Central overload: disrupted interoceptive multi-timescale inference in post-infectious ME/CFS, 2026, Strube

[Dolphin]

Central overload: disrupted interoceptive multi-timescale inference in post-infectious ME/CFS

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is marked by fluctuating, state-dependent symptoms, including post-exertional malaise (PEM), orthostatic intolerance, cognitive dysfunction ("brain fog"), and sensory hypersensitivity. Despite growing evidence for persistent...

sciety.org

Central overload: disrupted interoceptive multi-timescale inference in post-infectious ME/CFS​

1. Andreas Strube

Abstract​

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is marked by fluctuating, state-dependent symptoms, including post-exertional malaise (PEM), orthostatic intolerance, cognitive dysfunction ("brain fog"), and sensory hypersensitivity.

Despite growing evidence for persistent post-infectious immune, autonomic, metabolic, and neurobiological abnormalities, no unifying mechanistic framework yet explains how these disturbances become translated into the characteristic temporal dynamics of the illness.

Recent neuroimaging and neurochemical findings remain largely descriptive and do not explain why PEM emerges with a delay of hours to days, why symptoms worsen nonlinearly, or why autonomic, cognitive, and sensory domains deteriorate in concert.

Here, I propose a biologically plausible and explicitly testable framework in which heterogeneous post-infectious disturbances converge on a disorder of interoceptive inference across nested physiological timescales.

The central claim is that ME/CFS may be understood as a state of central overload: post-infectious neuroimmune and neuromodulatory processes bias precision control, such that expected physiological variability is insufficiently attenuated and bodily deviations acquire disproportionate inferential weight.

The framework is formalised within a hierarchical Bayesian architecture in which a slowly updating capacity hyperprior constrains moment-to-moment interoceptive inference, and in which neuroinflammation-biased precision amplification at lower levels drives excessive updating at higher levels.

Exertion may thereby trigger slow peripheral perturbations whose delayed consequences progressively update higher-order capacity estimates, coupling symptoms across domains and producing PEM.

The framework is distinguished from central sensitisation, functional neurological disorder, and conditioned fear accounts by specific, falsifiable predictions regarding the temporal dynamics, computational parameters, and neurophysiological signatures of the proposed inferential disturbance.

 

[Dolphin]

On the list I'm on, someone posted this information for what it is worth:

According to,
https://cdasr.mclean.harvard.edu/strube/
Andreas Strube earned his Ph.D. degree from the University of Hamburg,
Germany, in 2023 and this paper/report is dated April 10, 2026. He now
holds a postdoctoral position at Harvard University.

See also Google Scholar,

https://scholar.google.com/scholar?oi=bibs&hl=en&cluster=16157472541360913191
https://scholar.google.com/citations?user=FSeU-mIAAAAJ&hl=en

His Ph.D. thesis (2022, not 2023) is available at Hamburg University,
https://ediss.sub.uni-hamburg.de/handle/ediss/10612
See also Google Scholar,

https://scholar.google.com/citation...J&citation_for_view=FSeU-mIAAAAJ:qjMakFHDy7sC

 

[Jonathan Edwards]

I can see what he thinks he is trying to do but I don't personally see a theory cast in this sort of language as being useful. To test it properly you would need to be able to predict exactly what was wrong at the neuron level. Since we are a million miles away from understanding how global brain function relates to events at the individual neuron level we are stuck. I think you have to approach things from a molecular angle, not what is in effect a psychologist's angle. A molecular theory can be tested with drugs, directly, without needing to have a theory of the global mechanisms.

 

[InitialConditions]

Well, he seems to be asking the right questions — notably there is a focus on PEM and why it seems so 'nonlinear' and why different bodily symptoms all seem to fail together, regardless of the form of exertion. I have no idea if his answers are correct, but I do think we need much more research on the ME/CFS brain, especially regarding its role in orchestrating symptoms.

(I am aware that this sort of hypothesis is somewhat ripe for psychologisation, even if they put it on a neurobiological footing.)

 

[Jonathan Edwards]

but I do think we need much more research on the ME/CFS brain, especially regarding its role in orchestrating symptoms.

Agreed.

I think the Bayesian theoretical framework probably gets in the way here though.

 

[Utsikt]

The summary of the study on pain makes the basic error of equating reported pain with perception of pain.

 

[Creekside]

One explanation for PEM's delay is immune system timing. For example, exertion triggers an increase in IFN-g ~24 hrs later, which fits my experiences with physically-induced PEM. I think that's a fairly consistent t-cell response. Perhaps that mechanism is testable (how consistent PWME's delays between exertion and symptoms).

My cognitively-induced PEM was much shorter (<1 hr) and much more variable. I attribute that to activating the same PEM mechanism more directly, perhaps without IFN-g involvement.

I don't see washing a window for < 1 minute constitutes "central overload". Muscle cell microtears triggering immune response, triggering IFN-g which triggers glial cells seems more plausible to me.

 

[Jonathan Edwards]

I agree. I don't think concepts like 'central overload' have any useful basis. I don't think we have an neural data to ink it to and the concept itself seems confused. Is it overload if it is a normal input that is being responded to differently? If it is defined by the adverse response then does it assume what it tries to explain?

 

[Jonathan Edwards]

The thesis astract is pretty vague. I am not clear what is meant by a Bayesian framework here. He points out that what he proposes is in contradiction to the active inference prediction and error detection model used by those who study specific senses like vision.

The basic idea seems to be that if you expect to feel pain you are more likely to feel pain if you receive sensory stimuli. That is pretty much a BPS approach I think. It invokes expectations based on ideas and he tests it with placebos.

 

[jnmaciuch]

One explanation for PEM's delay is immune system timing. For example, exertion triggers an increase in IFN-g ~24 hrs later, which fits my experiences with physically-induced PEM. I think that's a fairly consistent t-cell response. Perhaps that mechanism is testable (how consistent PWME's delays between exertion and symptoms).

My cognitively-induced PEM was much shorter (<1 hr) and much more variable. I attribute that to activating the same PEM mechanism more directly, perhaps without IFN-g involvement.

I don't see washing a window for < 1 minute constitutes "central overload". Muscle cell microtears triggering immune response, triggering IFN-g which triggers glial cells seems more plausible to me.

Sorry if this is derailing the thread, but what makes you think specifically interferon gamma as opposed to other immune responses? Muscle cell damage would lead to a robust innate response but I’m not sure about an adaptive one

 

[jnmaciuch]

The thesis astract is pretty vague. I am not clear what is meant by a Bayesian framework here. He points out that what he proposes is in contradiction to the active inference prediction and error detection model used by those who study specific senses like vision.

The basic idea seems to be that if you expect to feel pain you are more likely to feel pain if you receive sensory stimuli. That is pretty much a BPS approach I think. It invokes expectations based on ideas and he tests it with placebos.

And I should think it would be contradicted by many pwME reporting [edit: random fluctuations where they have unexpectedly severe consequences to a level of exertion that they could previously manage without problems]

 

[Jonathan Edwards]

And I should think it would be contradicted by many pwME reporting [edit: random fluctuations where they have unexpectedly severe consequences to a level of exertion that they could previously manage without problems]

Absolutely, not to mention the PACE trial showing that you can make people at least say they are better but they turn out objectively to be the same. If this is just a re-run of ME symptoms just being believing you have ME symptoms we can pass on.

 

[Jonathan Edwards]

Muscle cell damage would lead to a robust innate response but I’m not sure about an adaptive one

Just for fun I asked Google, who said:

Yes, gamma interferon (IFN-g) is produced during innate immune responses. While primarily associated with adaptive immunity (Th1 cells), early production of IFN-γ
is a key part of the innate defense, acting as a bridge between innate and adaptive systems.

!

 

[Utsikt]

I’ve read the intro and looked at a couple of the sections.

The syndrome may instead be understood as reflecting a breakdown in how bodily signals are regulated across nested timescales: from rapid autonomic adjustments, to intermediate exertion-related shifts, to delayed post-exertional worsening, and ultimately to higher-order estimates of bodily capacity that shape future anticipation and behaviour.

At the same time, explanatory accounts that drift toward psychologising narratives have often been experienced as delegitimising and stigmatizing.23,24 What is needed, therefore, are explicitly non-blaming models that remain compatible with multi-system physiology and can be tested against objective physiological readouts.

No, the issue isn’t that the models «blame» the patients, it’s that the models have been completely unfounded, and contradictory evidence has been hidden, ignored and lied about by people with vested interests, causing unfathomable harm to the patients.

What is needed is a correct model of the disease, regardless of what that entails.

In the following, “priors”, “predictions”, and “beliefs” refer to neurally implemented statistical expectations within predictive control circuits, not to conscious appraisals or acts of volitional control over symptoms. Rather than contrasting “peripheral” and “central” accounts, the present framework treats them as coupled stages of a single process in which post-infectious perturbations shape the afferent evidence stream while neuroimmune interactions bias the gain and precision control through which that evidence is integrated.

Are they arguing for hard determinism here?

7. Control and pacing as a bidirectional precision lever

Activity pacing - that is, monitoring exertion and staying within individually tolerable limits - is widely recommended and strongly endorsed by patients.83,84 Within the present framework, its effects may extend beyond simple energy conservation.

By reducing physiological volatility and lowering the probability of delayed post-exertional worsening, pacing changes the statistical structure of activity-symptom contingencies.

In predictive terms, this may increase behavioural predictability and perceived controllability, even in a system whose underlying physiology remains only partially controllable.70

Pacing can therefore be understood not merely as a practical coping strategy, but as an intervention on the statistical landscape within which bodily inference unfolds.

If pacing affected the statistical landscape, you’d expect pacing to produce improvements, no? As far as I’m aware, there is no evidence of that happening.

This logic also suggests that control may act as a bidirectional lever. Recent work indicates that control can increase expectation precision.70,71,85 In ME/CFS, such increased precision may be adaptive when pacing reliably prevents crashes: the system learns that a given level of activity is safe, and stable expectations of tolerability can be reinforced.

Under conditions in which exertion repeatedly produces delayed worsening, however - whether due to illness dynamics, external demands, or prescribed escalation - the same learning mechanism may strengthen expectations of deterioration.

In that case, control does not simply reduce uncertainty; it sharpens whichever activity–outcome contingencies the system repeatedly encounters.

So pacing becomes more «effective» the longer you do it? In the sense that you’ve be able to do more and more without experiencing PEM?

This account provides a mechanistic rationale for why patient-controlled activity management may be beneficial and why externally imposed escalation could, under conditions of delayed post-exertional worsening, reinforce maladaptive activity-symptom contingencies.86

More broadly, it suggests that rehabilitation approaches validated in other conditions should not be transferred uncritically to ME/CFS. Agency and control do not operate “outside the illness process”, but potentially interact with altered interoceptive inference itself.70

Because the system learns from the delayed consequences of exertion, the manner in which activity is structured may shape not only short-term symptom burden, but also the longer-term updating of perceived bodily capacity.

This could be used to argue for pacing up..

This is how they propose to test their prediction for pacing:

Readout

Predicted signature

Manipulation / design

Falsification

Predictability (pacing)

Structured pacing reduces PEM vs dose-matched unpredictable exertion schedule.

Cross-over: structured vs jittered schedule (dose-matched).

No predictability effect beyond total activity reduction.

It seems like they want to try to (or at least risk( induce PEM repeatedly (in the jittered schedule). I think they might benefit from repeating the ethics class if this is what they consider to be a realistically testable hypothesis.

 

[jnmaciuch]

Just for fun I asked Google, who said:

Yes, gamma interferon (IFN-g) is produced during innate immune responses. While primarily associated with adaptive immunity (Th1 cells), early production of IFN-γ
is a key part of the innate defense, acting as a bridge between innate and adaptive systems.

!

Do you have the cited study? I assume it’s talking about IFN-g from NK cells which are technically considered innate. Though I think it’s a safe assumption that other facets of the innate response with PAMP/DAMP receptors are going to be the main source of the immune response from muscle damage rather than NKs, [edit: though NKs or adaptive response may play some small part downstream].

 

[duncan]

I'm sorry. Isn't this a silly fancy way of suggesting that how we feel is incorrect? That we only think we are sick, that a faulty sense of feeling shitty ,and its timing, is merely that: misappropriated perception?? We only feel sick; we really aren't. If there is a pathology, it's in the brain fooling itself?

Can't we just collectively say: Fuck you. We ARE sick?

 

[Trish]

I had a go at reading it. I'm not much the wiser.

It seems to be a long and convoluted way of saying that it's not central sensitisation or FND or catastrophising which are all versions of the brain wrongly interpreting normal sensations as abnormal.

Instead it is a more complicated version of our interoception being amplified in the brain so we experience normal sensations as abnormal, or mild symptoms as worse than they are, and that this gets worse the more we push ourselves into PEM because it creates 'overload' in the brain by a sort of feedback loop, so it becomes a spiral downwards the more we push into PEM.

So he actually is in favour of stabilisation by pacing with the amount of exertion determined by the pwME to try to avoid PEM in order not to create this overload. And that externally prescribed increasing exertion is bad because it pushes us into repeated PEM and therefore deterioration.

I hope I've got that somewhere near right.

 

[Jonathan Edwards]

Are they arguing for hard determinism here?

Those quotes suggest the usual sort of confusion over the precise dynamics that you get if you treat the brain as a 'system' and try to make events non-local (as this does). It tries hard to dissociate 'agency' and 'control' from consciousness but is bound to fail if it tries to explain in this non-local way. All physical science is local.

I had never expected my interest in the way minds work to be relevant to ME/CFS but I think it is here. You can only explain ME/CFS in brain terms without ending up with what Duncan is objecting to if you ditch the intuitive model of mind and agency as a system property. If you do that you put ME/CFS in the same category as sciatica or phantom limb pain or epilepsy or even multiple sclerosis. The absence of structural 'damage' of a conventional type needn't be a problem.

 

[Utsikt]

Those quotes suggest the usual sort of confusion over the precise dynamics that you get if you treat the brain as a 'system' and try to make events non-local (as this does). It tries hard to dissociate 'agency' and 'control' from consciousness but is bound to fail if it tries to explain in this non-local way. All physical science is local.

That might be what was tripping me up. Thank you for explaining.

I had never expected my interest in the way minds work to be relevant to ME/CFS but I think it is here. You can only explain ME/CFS in brain terms without ending up with what Duncan is objecting to if you ditch the intuitive model of mind and agency as a system property. If you do that you put ME/CFS in the same category as sciatica or phantom limb pain or epilepsy or even multiple sclerosis. The absence of structural 'damage' of a conventional type needn't be a problem.

So in this model, different neurons would have different «tasks», and the ones concerned with «is this input the same as expected» aren’t the ones that result in what we perceive as our stream of consciousness?

One a side note: isn’t doping in sports evidence against the predictive coding model? Doping lets you do far more than you’d expect, but it doesn’t make them feel sick in any way (unless the doping has side effects). And they don’t get sick like ME/CFS when they stop the doping and experience the reverse.

The same with fractures. If you break you leg you can suddenly so a lot less than you’re used to. That doesn’t cause ME/CFS symptoms. Nor when you take of the whole leg cast and can suddenly do more.

Or the aftermath of strokes.

I guess these might just be examples of why you can’t explain disease with normal rules.