Unifying Mechanisms Across Systems
Several mechanisms operate across multiple systems to maintain chronic illness:
Vicious cycles: Metabolic dysfunction impairs immune clearance; chronic inflammation consumes metabolic resources; autonomic dysfunction reduces cerebral perfusion needed for repair. Each cycle self-reinforces, creating path dependency.
Multi-lock hypothesis: Multiple independent failure modes must be addressed simultaneously. Breaking one lock (e.g., mitochondrial support) may be insufficient if other locks (autoantibodies, autonomic dysfunction) remain engaged.
Cycle dynamics: Critical transitions explain sudden deteriorations and non-linear progression. Small perturbations may trigger large state changes when near thresholds.
Non-monotonic dose-response (inverted-U / biphasic / hormetic) as a systems principle (triggered by Kevin Lee, personal communication, July 2026; expanded to all medications with non-monotonic dose-response, July 2026): A recurring pattern across ME/CFS pharmacotherapy is the non-monotonic dose-response. This framework synthesizes at least four pharmacologically distinct mechanism categories that share only a U-shaped dose-response curve shape. The term “hormesis” — in which low-dose stress triggers compensatory adaptive upregulation via Nrf2/ARE signaling while higher doses suppress the same pathway — applies specifically to the Nrf2 cluster (Calabrese corpus (Calabrese 2002) (Calabrese and Baldwin 2003) (Calabrese 2010) (Sun et al. 2020)). Other non-monotonic patterns derive from different mechanisms (partial-agonist inverted-U, receptor-isoform selectivity, BBB penetration, biphasic concentration-response at ion channels) and are grouped here because they share the empirical pattern — not the same molecular mechanism. The Calabrese literature establishes hormesis as a common dose-response pattern in toxicology (cell proliferation, enzyme induction), though its status as the “default” in clinical pharmacology is not consensus. In ME/CFS specifically: in healthy humans, exercise activates Nrf2/AMPK/PGC-1α signaling via mitohormesis (Ristow and Zarse 2010). In ME/CFS, where exercise triggers PEM rather than adaptation, this pathway may be broken (inferred from known mitohormesis in health + documented exercise intolerance in ME/CFS — the NRF2–PGC-1α bidirectional circuit has not been directly measured during exercise in ME/CFS patients). The consequence may be that hormetic pathways are either hypercompensatory at low doses or blunted, but direction and magnitude are unknown.
The following medications in this paper exhibit non-monotonic dose-response patterns. This candidate list was generated by pattern-review of this paper’s own content and has not been validated in any external dataset — whether these patterns share a common mechanism or are drug-specific artifacts is the core question the hormetic reserve hypothesis poses:
Nrf2-mediated hormesis (compensatory upregulation): LDN (optimal dose varies from 0.5–4.5 mg; TLR4/Nrf2 hormetic window — low-dose TLR4 blockade triggers compensatory anti-inflammatory priming; higher doses remove basal TLR4 tone, collapsing benefit. At 50 mg, naltrexone becomes a full mu-opioid antagonist — this is dose-dependent target selection, not a second inversion on the response curve (Calabrese and Kozumbo 2021) (Calabrese and Giordano 2021) (Dara et al. 2023) (Toljan and Vrooman 2018)), low-dose lithium (potential benefit at 2–5 mg/day via GSK-3β inhibition → Nrf2 nuclear localization; also acts on IMPase/inositol at higher concentrations — see mTOR cluster below), melatonin (sleep benefit at 0.3–0.5 mg; worsening at 3+ mg — MT1/MT2 receptor desensitization at supraphysiological doses), sulforaphane (theoretical Nrf2 hormetic window; high-dose pro-oxidant effect may reverse low-dose benefit).
Partial-agonist inverted-U (receptor-occupancy pharmacology, not Nrf2-mediated): LDA/aripiprazole (therapeutic at 0.2–2 mg; D2 partial agonism provides net agonism at low occupancy; antagonist dominance at >50% occupancy inverts benefit — U-shaped dose-response (Crosby, Kalantar, and DeRisi 2021) (Sun et al. 2020)). This is a receptor-occupancy property applicable in any population with dopamine deficit, not ME/CFS-specific.
Catecholamine inverted-U (inverted-U dose-response at D1/α2A receptors in prefrontal cortex — canonical neuroscience principle (Arnsten 2011) (Cools and D’Esposito 2011) (Cools and Arnsten 2022)): modafinil (cognitive benefit at 50–100 mg via DAT blockade; 200+ mg triggers histaminergic/orexin activation → insomnia, anxiety — D1 inverted-U exceeded), duloxetine (serotonergic benefit at 20–30 mg; 60+ mg NE reuptake inhibition dominates → sympathetic activation worsens orthostatic intolerance), beta-blockers (β1-selective HR reduction at low dose; β2 blockade at higher dose removes lipolysis → fatigue worsens — Yerkes-Dodson inverted-U arousal optimization (Calabrese 2008)), guanfacine (PFC-selective α2A at 0.5–1 mg; global NE suppression at 2–4 mg → cognition inverts), gabapentinoids (pathological-circuit-selective α2δ block at 100–300 mg; global neurotransmitter suppression at higher doses → sedation).
Receptor-isoform selectivity: rapamycin/sirolimus (mTORC1 inhibition at 1–3 mg/week restores autophagy/mitophagy; higher cumulative exposure suppresses mTORC2 → immunosuppression, insulin resistance — this is an inherent property of rapamycin’s mTORC1/mTORC2 selectivity, not a host-tissue hormetic response (Sarbassov et al. 2006) (Lamming et al. 2012)), corticosteroids (physiological replacement at 5–10 mg prednisone modulates immunity; higher doses suppress HPA axis → rebound crash on taper).
Pharmacokinetic / BBB penetration effects: DORAs (partial orexin blockade at low dose improves sleep; complete blockade at higher dose causes sleep paralysis/daytime hypersomnia), H1 antihistamines (peripheral H1 at 5–10 mg cetirizine relieves MCAS; CNS H1 penetration above 20 mg → sedation — a concentration-gradient phenomenon, not a biphasic receptor-level mechanism).
Biphasic neurotransmitter concentration-response: allopregnanolone (GABA-A PAM — low concentrations 1–5 nM paradoxically increase anxiety; higher concentrations >10 nM produce anxiolysis (Andréen et al. 2009)), NAC (paradoxical worsening at 600 mg that resolves at 1,200 mg — thiol-radical burst exhausts GSH at low dose; high dose shunts toward net GSH synthesis), ketotifen (MCAS benefit at low dose; sedation above H1 threshold — similar to H1 antihistamine BBB penetration phenomenon), quercetin (COMT inhibition → catecholamine inverted-U optimum for cognition), taVNS (hypothesized non-monotonic dose-response — standard parameters may overshoot into vagal overload; low-intensity input remains within functional range).
Time-dependent hormesis — the temporal dimension of the dose-response window (certainty: 0.30): The preceding framework describes dose-response as a function of concentration. But the Calabrese corpus further demonstrates that hormetic features are time-dependent — stimulatory and inhibitory effects integrate across dose AND time simultaneously, producing a 3D response surface rather than a 2D curve (Sun et al. 2018). Many hormetic responses are transient: they appear at one exposure duration and disappear with continued exposure (Mushak 2016). The therapeutic window is a volume (dose × time × response), not a plane, and failing to account for time collapses it into a potentially misleading 2D projection. This has direct clinical implications for ME/CFS: if the hormetic window is time-dependent, then continuous daily dosing may extinguish it — not because the dose is wrong, but because the resensitization interval that the pathway requires has been filled. Pulsed or intermittent dosing would preserve the window by allowing stress-signal recovery and receptor resensitization during the drug-free interval.
The GPCR resensitization clock: G protein-coupled receptors require a coordinated cycle of activation → desensitization → resensitization. Desensitized GPCRs internalize, undergo dephosphorylation in endosomes, and recycle to the membrane in a reactivated state — a process that requires a drug-free interval whose length is receptor-specific (Costa-Neto and Parreiras-E-Silva 2025) (Gupta, Mohan, and Naga Prasad 2018) (Kliewer, Reinscheid, and Schulz 2017). Continuous receptor occupancy fills this interval, preventing the resensitization that would otherwise occur. The dephosphorylation half-life of a given receptor sets the minimum off-period for preservation of drug sensitivity. This principle is distinct from receptor-isoform selectivity (Category 4 in the mechanistic clustering below): it operates on the same receptor through temporal kinetics rather than on different receptor complexes through concentration kinetics. Rapamycin provides the cleanest proof of concept: intermittent dosing (e.g., every 5 days in mice) preserves mTORC1 inhibition benefits while sparing mTORC2-mediated adverse effects — the therapeutic and adverse targets have different temporal sensitivities to the same drug molecule (Arriola Apelo et al. 2016) (Konopka et al. 2023). This principle — differential drug-target recovery kinetics — generalizes to any drug whose therapeutic and adverse targets resensitize at different rates.
ME/CFS-specific vulnerability to tachyphylaxis: Oxidative stress amplifies receptor tachyphylaxis through ROS-mediated receptor oxidation (Teyani, Moghaddam, and Moniri 2024). Since ME/CFS is characterized by oxidative stress across multiple studies, the disease state that most requires sustained pharmacology is the one in which receptor tachyphylaxis is most likely. This creates a clinical paradox: patients with the highest allostatic load may be the fastest to lose benefit from continuous dosing and the most in need of pulsed strategies that preserve therapeutic windows. The withdrawal risk counterpart — some drugs with tachyphylaxis also produce withdrawal or rebound when paused (Hodding, Jann, and Ackerman 1980) — constrains this framework: rapid pulsing (short off-periods) may work for drugs with fast resensitization kinetics and low withdrawal liability, while gradual tapering is required for drugs with slow resensitization and high withdrawal risk. A clinical precedent exists in dermatology: psoriasis management has adopted proactive pulse therapy — treating intermittently after clearance to maintain effect while avoiding cumulative corticosteroid exposure (Papp et al. 2021). The medical community already accepts pulse therapy when the risk-benefit calculus shifts in a chronic inflammatory condition. Which strategy applies to which drug is a drug-specific empirical question — not a general principle — and the framework proposed here is a classification scaffold, not a dosing algorithm.
Consequence: The time-dependent dimension of hormesis reclassifies dosing schedules as a mechanistic variable, not just a convenience. If the hormetic window is transient, then continuous daily dosing of Nrf2-activating drugs (sulforaphane, NAC, melatonin) may be self-defeating — the benefit extinguishes not because the drug stopped working but because the resensitization interval was filled. This would make intermittent dosing strategies (e.g., every other day, two days on / one day off) a testable intervention for drugs already in clinical use. Equally important: this framework provides a formal rationale for why LDN’s overnight dosing schedule — a de facto pulsed regimen with a 4–6 hour blockade window followed by daytime washout — is mechanistically appropriate, not just a convenience convention. Certainty: 0.30 — time-dependent hormesis is well-established in the Calabrese corpus and GPCR resensitization kinetics are conserved across receptor families, but no pulsed-vs-continuous dosing trial has been conducted for any drug in ME/CFS. (Translation gap: all evidence is from in vitro systems, animal models, and general-population pharmacology — no ME/CFS-specific data exist.)
Tachyphylaxis risk classification — a 2×2 decision matrix for the 17-drug framework (certainty: 0.20): Each drug in the hormesis framework can be classified on two axes: (a) does its therapeutic mechanism depend on transient receptor perturbation that requires resensitization? (b) does it have significant withdrawal/rebound risk? Drugs scoring high on (a) and low on (b) are candidates for pulsed dosing; drugs low on (a) are indifferent to timing; drugs high on (b) should NOT be pulsed regardless of (a) because withdrawal risk outweighs any resensitization benefit. Projected classification: {(a) high, (b) low} — sulforaphane, NAC, melatonin (Nrf2 activators, low withdrawal), modafinil, guanfacine (catecholamine cluster, low withdrawal); {(a) high, (b) high} — LDN (GPCR target through opioid/TLR4 but withdrawal risk from endorphin dependence cessation), beta-blockers (tachycardia rebound); {(a) low, (b) low} — LDA (partial-agonist inverted-U is receptor-occupancy-dependent, not time-dependent), allopregnanolone (biphasic concentration-response, not tachyphylaxis); {(a) low, (b) high} — corticosteroids (HPA suppression, high withdrawal risk), gabapentinoids (dependence, high withdrawal). This classification is a first-principles scaffold — no empirical validation exists. The resensitization half-lives for the relevant receptors (TLR4, DAT, NET, β1, α2A, D2, μ-opioid) have not been measured in ME/CFS patients. The classification may be wrong for any specific drug — it is a hypothesis-generating framework, not a clinical algorithm. Origin: brainstorm — /integrate-topic pulsed-therapy.
Continuous dosing may be necessary for some mechanisms. The pulsed-therapy framework assumes resensitization intervals are beneficial, but for opioid compensatory upregulation (LDN’s primary mechanism at 1.5–3.0 mg), therapeutic effect depends on sustained receptor blockade followed by circadian compensatory rebound. If the pulsed interval is too short, the compensatory response never develops; if it’s too long, the blockade never accumulates enough to trigger compensation. The 24-hour overnight dosing cycle may be the ONLY interval that works for this mechanism class because it matches the circadian rhythm of opioid receptor expression. For TRPM3-dependent LDN benefit (3.0–4.5 mg), sustained receptor modulation may also be required — a drug holiday could extinguish the TRPM3 restoration signal. The pulsing framework is drug-specific and mechanism-specific, not universal. Certainty: 0.30 — the circadian dependence of endorphin compensation is plausible but untested. Origin: brainstorm — /integrate-topic pulsed-therapy.
Tachyphylaxis may be misattributed to disease fluctuation. ME/CFS is a fluctuating illness with spontaneous remissions and relapses. A patient who reports “the drug stopped working” may simply be in a natural disease trough — symptoms would have worsened regardless of dosing schedule. Attributing all loss of benefit to tachyphylaxis overestimates the framework’s scope. A substantial fraction of “stopped working” reports are disease fluctuation misattributed to pharmacology. The drug holiday restore-test (protocol Drug Holiday Restore-Test — Distinguishing Adaptive Desensitisation from True Tolerance) distinguishes the two mechanisms: benefit restoration after a holiday confirms tachyphylaxis; no change confirms disease fluctuation. But the test itself has not been validated, and a single negative test does not definitively rule out adaptive desensitization — the off-period may have been too short. Certainty: 0.40 — general confound in all chronic illness pharmacology, not specific to ME/CFS. Origin: brainstorm — /integrate-topic pulsed-therapy.
If pulsed dosing is irrelevant — what would need revision. The null hypothesis: time-dependent dosing has zero effect on therapeutic outcomes for any drug in the 17-drug framework, and all observed loss of benefit is dose-dependent only. If null is true: ch15’s time-dependent hormesis paragraph is an overstatement to be removed; the ch34 cascade on pulsed dosing should be reduced or removed; ch33’s GPCR resensitization rationale for LDN overnight dosing should be downgraded to “empirical convention — mechanism rationale unconfirmed.” A pulsed-vs-continuous RCT of any Nrf2 activator showing zero difference in maintenance of benefit at 12 weeks (n≥60, correct pulsing interval) would provide evidence toward the null. A single null trial would not disprove the framework, but ≥3 null trials across mechanistically distinct drugs with receptor-appropriate intervals would. Consequence: if null is true, a substantial section of the pulsed-therapy content would need removal. This would be a positive outcome for the paper’s epistemic integrity — it would demonstrate willingness to correct overstatement. Certainty: n/a — null hypothesis assessment.
Falsifiable prediction: If hormetic reserve is a general systems property of ME/CFS physiology (not merely a collection of drug-specific artifacts), patients with a narrow window for one drug should show narrow windows for at least one other mechanistically distinct drug in a within-patient crossover — the inversion-point positions should correlate across drugs targeting different systems. The registry entry hormesis multi drug principle defines the falsification boundaries for 6-drug HIP-B testing: mean pairwise r ≥ 0.4 to confirm cross-system hormetic reserve; mean r < 0.2 to falsify; 0.2 ≤ r < 0.4 is an indeterminate zone requiring larger sample sizes or additional drugs to resolve. The orthogonal-mechanism correlation test par excellence is LDN × modafinil: these drugs operate through completely distinct receptors (TLR4 vs DAT/NET), cell types (microglia vs presynaptic catecholamine terminals), and anatomical targets. If LDN and modafinil inversion points correlate (r ≥ 0.4 with n ≥ 80 — under n=20 the 95% CI on r covers both 0.2 and 0.4, preventing decisive classification), hormetic reserve is a genuine cross-system trait. Which direction a patient falls (hypercompensation vs. blunting) depends on baseline activation state and individual genetic variation in Nrf2, TLR4, opioid receptor genes, COMT, CYP2D6, and other pharmacogenetic factors. (Calabrese et al. 2023)
Mechanistic clustering of non-monotonic dose-response drugs (certainty: 0.25): The drugs group into at least five mechanistic categories. If inversion-point positions correlate within categories (r ≥ 0.4 for LDN ↔︎ sulforaphane, duloxetine ↔︎ modafinil) but not across categories (r < 0.2 for LDN ↔︎ modafinil), hormetic reserve is a multi-component trait with dissociable subsystems. (1) Nrf2 cluster (LDN, low-dose lithium, melatonin, sulforaphane, NAC, quercetin) — drugs converging on Keap1-Nrf2-ARE signaling and GSK-3β-mediated Nrf2 nuclear exclusion. (2) Catecholamine cluster (duloxetine, modafinil, beta-blockers, guanfacine, gabapentinoids) — drugs whose inverted-U converges on prefrontal D1/α2A receptor dynamic range and catecholamine tone. (3) D2 partial-agonist cluster (aripiprazole/LDA) — inverted-U driven by occupancy-dependent net effect: partial agonism dominates at low occupancy (therapeutic at 0.2–2 mg in dopamine-deficient systems), antagonist property dominates above ~50% occupancy (worsening). This is a receptor-occupancy phenomenon distinct from Nrf2 hormesis and PFC catecholamine inverted-U — the curve is identical in any population with dopamine deficit, not ME/CFS-specific. Whether aripiprazole’s inversion point correlates with the catecholamine cluster (common dopamine substrate) or is fully independent is the empirical question its single-member status poses. (4) mTOR/autophagy cluster (rapamycin) — mTORC1/mTORC2 dose selectivity. Lithium spans Nrf2 and mTOR (via GSK-3β → both Nrf2 nuclear localization and TSC2 → mTORC1) and is designated a “bridge” drug (r ~0.2–0.4 with both clusters). Corticosteroids span mTOR (via FKBP5) and GABAergic (via receptor subunit modulation) — similarly a bridge drug. (5) GABAergic/neurosteroid cluster (allopregnanolone, DORAs) — drugs converging on GABA-A subunit composition, neurosteroid sensitivity, and orexin/GABAergic sleep circuitry. Ketotifen and taVNS are unclassified — their non-monotonic patterns derive from BBB penetration and autonomic overshoot respectively, not from the five mechanistic categories above. Conditional on the clustering being confirmed — no clinical decisions should be based on unvalidated clusters at the present certainty (0.25): A patient classified as “Nrf2-compromised” (narrow LDN+sulforaphane windows, normal duloxetine+modafinil windows) would be expected to benefit from Nrf2-targeting drugs more than catecholamine-targeting drugs. A patient classified as “Global-compromised” (narrow windows across categories) would suggest fragile stress-response physiology requiring ultra-low, ultra-slow single-drug sequential testing — polypharmacy would risk paradoxical response at uncharacterized inversion points.
Guanfacine + NAC as an inter-cluster bridge (catecholamine cluster × Nrf2 cluster). The clustering above treats guanfacine (catecholamine cluster) and NAC (Nrf2 cluster) as separate single drugs. A fixed combination, low-dose guanfacine (PFC-selective α2A agonism for executive function) plus N-acetylcysteine (Nrf2 antioxidant support), is reported as an open-label intervention for cognitive deficits in Long COVID (Fesharaki-Zadeh, Lowe, and Arnsten 2023) and after TBI (with and without donepezil) (Khasnavis et al. 2024) (Fesharaki-Zadeh et al. 2025). Mechanistically the two agents address non-overlapping bottlenecks in the same cognitive/energy-availability problem: guanfacine restores PFC α2A signaling when endogenous noradrenergic drive is inadequate (see Guanfacine as Pharmacologically Distinct from Clonidine), while NAC supports the glutathione/Nrf2 redox system against the oxidative stress that accompanies neuroinflammation (Cherneva et al. 2025). The combination therefore occupies the same conceptual role as the existing “bridge” drugs (lithium, corticosteroids); it spans two mechanistically distinct clusters rather than defining a third. Critical caveat: no direct α2A-adrenoceptor → Nrf2 molecular cross-talk has been established (0 PubMed hits); the bridge rationale is parallel non-overlapping targets, not proven synergy. There are no ME/CFS-specific data. This is hypothesis-generating, not a treatment recommendation.
Falsifiable predictions: (a) In ME/CFS patients with documented oxidative stress (elevated F2-isoprostanes or low GSH/GSSG) and PFC-dependent cognitive deficits, guanfacine+NAC improves n-back/Stroop performance more than either agent alone at 8 weeks; falsified if the combination shows no added benefit over monotherapy. (b) If a true α2A→Nrf2 interaction exists, NAC augments guanfacine’s cognitive effect preferentially in patients with the highest oxidative burden; falsified if the augmentation is oxidative-burden-independent. (c) Guanfacine+NAC produces no greater orthostatic adverse effects than guanfacine alone (Okamoto et al. 2024); falsified if the combination worsens orthostatic intolerance beyond guanfacine monotherapy. (Raw certainty: 0.25–0.35, from open-label case series and general-population mechanism reviews → discounted to ≤0.30 for ME/CFS; no ME/CFS data.)
Consequence: If this bridge is confirmed, it would give clinicians a single, well-tolerated combination that targets both the prefrontal noradrenergic deficit and the oxidative-stress burden implicated in ME/CFS brain fog; today it is an untested hypothesis. It also sharpens the ch15 framework: a multi-component bridge would predict that patients who respond to both a catecholamine-cluster drug and an Nrf2-cluster drug (but neither alone fully) have concurrent noradrenergic and redox deficits. (Severity applicability: unknown; study populations were Long COVID and TBI, not stratified by ME/CFS severity. Not a treatment recommendation.)
Research questions: Distinguishing which mechanisms are primary drivers vs. secondary consequences remains challenging. Longitudinal studies tracking biomarker trajectories from acute onset are needed. The hormetic dose-response hypothesis is entirely theoretical in ME/CFS — no prospective dose-response trial has tested the inverted-U prediction for any drug in this population. A within-range LDN dose-response trial (0.5, 1.5, 3.0, 4.5 mg) would simultaneously test the hormesis framework and establish the single most clinically relevant data point for the most widely prescribed off-label medication in ME/CFS. The HIP-B trial (Hormetic Inversion-Point Battery: 6 drugs — LDN, sulforaphane, duloxetine, modafinil, rapamycin, allopregnanolone — each at 4 within-range doses in within-patient crossover, designed as a Phase 0 pilot for effect-size estimation with n ≥ 80 needed for definitive testing; with n=20 the 95% CI on r spans approximately −0.05 to 0.71, preventing decisive classification for most outcomes) would estimate the cross-drug correlation signal needed to determine whether hormetic reserve is a single trait, a multi-component trait, or an artifact of drug-specific pharmacology. The complete trial specification is maintained in the hypothesis registry (entry hormetic inversion point battery hip b) and this chapter — see ME/CFS-Specific Research Methods for discussion of methodological challenges specific to dose-response trials in ME/CFS populations.
Four converging lines of evidence — time-dependent hormesis (Sun et al. 2018) (Mushak 2016) (Calabrese 2016), GPCR resensitization kinetics (Costa-Neto and Parreiras-E-Silva 2025) (Gupta, Mohan, and Naga Prasad 2018) (Kliewer, Reinscheid, and Schulz 2017), intermittent rapamycin dosing as proof of concept (Arriola Apelo et al. 2016), and a clinical precedent in psoriasis pulse therapy (Papp et al. 2021) — collectively argue that the therapeutic window is a 3D volume (dose × time × response), not the 2D curve the existing hormesis framework describes. Continuous daily dosing of drugs whose therapeutic mechanism depends on transient receptor perturbation (Nrf2 activators, GPCR-targeting drugs) fills the resensitization interval, converting a reversible adaptive desensitization into apparent drug failure. The framework provides a molecular rationale for existing clinical practice — LDN’s overnight blockade window and rapamycin’s intermittent weekly dosing are de facto pulsed regimens with mechanistic justification — and a zero-cost clinical tool: the drug holiday restore-test (Drug Holiday Restore-Test — Distinguishing Adaptive Desensitisation from True Tolerance) distinguishes adaptive desensitization from true tolerance. The strongest constraint is the withdrawal risk counterpart (Hodding, Jann, and Ackerman 1980): drugs with high withdrawal liability (corticosteroids, gabapentinoids) should NOT be pulsed regardless of resensitization kinetics. The most important open question is whether oxidative stress accelerates tachyphylaxis in ME/CFS specifically (Teyani, Moghaddam, and Moniri 2024) — if so, the disease state that most needs sustained pharmacology is the one in which receptor desensitization is fastest. All evidence is preclinical or general-population: no pulsed-vs-continuous dosing trial has been conducted for any drug in ME/CFS.
Consequence: Time-dependent dosing is not a dosing convenience — it is a mechanistic variable at the same level of importance as dose selection. For every drug in the 17-drug hormesis framework, the clinical question is now “what dose AND at what interval?” rather than “what dose?” alone. A drug holiday restore-test costs nothing, requires no lab, and could prevent years of taking a medication that stopped working months ago simply because the resensitization interval was never provided.
1 The Unified Post-Infectious Model: Impaired Physiological Resilience
Watton and Prusty (Watton and Prusty 2026) have proposed a comprehensive reframing of ME/CFS as a disorder of impaired physiological resilience within post-infectious disease biology. This unified model integrates findings across immunological, vascular, metabolic, and virological domains into a single overarching framework centered on loss of adaptive capacity.
ME/CFS is best understood as a disorder of impaired physiological adaptability and resilience. The contemporary literature converges on a biologically coherent model in which immune disturbance, metabolic reprogramming, endothelial dysfunction, and impaired physiological adaptability interact across systems, positioning ME/CFS within a broader class of post-infectious, immune-mediated chronic diseases characterised by dynamic, stress-revealed pathology rather than fixed structural lesions. (Watton and Prusty 2026) (Certainty: 0.45 — model is a review synthesis subject to revision; individual component certainties vary)
Key insight — state-dependent pathology. A central conceptual advance is the recognition that some key abnormalities are exaggerated or revealed under physiological stress rather than fully manifest at rest. Cellular and vascular systems show varying degrees of resting dysfunction (e.g., flow-mediated dilation impairment at rest, ~5.1% vs 8.2% in controls, Section Vascular Dysfunction) but often exhibit disproportionate failure under metabolic, oxidative, or immune challenge — unable to up-regulate energy production, maintain redox balance, or preserve barrier integrity. (Watton and Prusty 2026) This pattern implies that resting biomarkers will systematically underestimate, but not misrepresent, disease burden. The principle of stress-revealed pathology does not mean “normal at rest” — resting abnormalities exist and are measurable — but rather that the full magnitude of dysfunction becomes visible only under challenge.
Limitations of the “state-dependent” framing. Several documented ME/CFS abnormalities appear to be structural or constitutive rather than state-dependent: capillary basement membrane thickening (collagen IV deposition, Chapter Cardiovascular Dysfunction) — independently replicated in three countries (Amsterdam (Charlton et al. 2025), Berlin (Aschman et al. 2023), Aarhus (Agergaard et al. 2023)) and definitively distinguished from deconditioning by a 60-day bed rest comparator showing a structurally different phenotype (Charlton et al. 2025); RBC morphological abnormalities and deformability deficits persisting beyond the 120-day RBC lifespan; and ion channel dysfunction (TRPM3) observable in resting NK cells. The state-dependent model must coexist with evidence for constitutive pathology; the two are not mutually exclusive. Some abnormalities may be structural and cumulative (basement membrane thickening, epigenetic marks), while others are dynamically stress-revealed (endothelial adaptive capacity, mitochondrial metabolic reserve, CSF proteomic shifts). A complete model requires both categories.
Operationalising “physiological resilience.” The concept of impaired physiological resilience requires measurable parameters to avoid tautology. Candidate operational definitions include: (a) spare respiratory capacity (the difference between maximal and basal mitochondrial oxygen consumption — a direct measure of metabolic reserve, recently demonstrated to be acutely modifiable by whole-body hyperthermia in ME/CFS PBMCs (Hochecker et al. 2025), and proposed as the bottleneck for thermoregulatory work ch06:Spare Respiratory Capacity as Thermoregulatory Capacity Proxy in ME/CFS); (b) endothelial barrier recovery time after standardised challenge (TNF-\(\alpha\) or H2O2 exposure, measuring time to restoration of transendothelial electrical resistance); (c) heart rate recovery after standardised orthostatic or exercise challenge; (d) the hysteresis magnitude between stress removal and symptom/sign resolution (the temporal lag characterising PEM); and (e) the ratio of resting to stress-provoked biomarker levels (e.g., ATG13, haptoglobin proteoforms, EV cargo profiles). These are individually measurable and collectively define a multidimensional resilience phenotype. The Watton–Prusty model gains testability to the extent that these or analogous parameters are specified. (Overall model certainty: 0.45)
Mechanistic convergence across domains. The unified model identifies convergent abnormalities spanning:
- Immune dysregulation: Persistent innate and adaptive immune activation, selective monocyte/T-cell reprogramming, abortive viral reactivation (herpesvirus dUTPases driving chronic innate immune activation without productive infection), fibronectin–IgG circulating immune complex alterations with homeostatic antibody depletion, and autoantibodies against \(\beta\)-adrenergic and muscarinic receptors (with the strongest quantitative autonomic–autoantibody correlation to date being \(r=0.45\), \(p=0.001\), documented by Azcue et al. 2026 (Azcue et al. 2026))
- Metabolic reprogramming: Hypometabolic signatures, mitochondrial fragmentation induced by circulating factors (IgG fractions), impaired oxidative phosphorylation, and dysregulated autophagy/mitophagy (elevated ATG13)
- Endothelial and microvascular dysfunction: Stress-revealed impairment of endothelial metabolic responsiveness, coagulation abnormalities, fibrinolysis-resistant fibrin(ogen) aggregates, haptoglobin proteoform insufficiency exacerbating haem-mediated oxidative stress
- Lipid and membrane biology: SMPDL3B-mediated sphingolipid imbalance altering lipid raft fluidity and immune receptor organisation, linking omics-level lipid abnormalities to cellular immunophenotypes
- Extracellular vesicle communication: EVs carrying altered proteomic signatures, severity-associated immune markers, and mitochondrial DNA capable of activating innate immune pathways, providing a mechanism for local cellular dysfunction to propagate systemic effects
Replicability at pathway vs. gene level. A systematic reanalysis by Keele et al. found that only 2 genes were consistently replicated across ME/CFS transcriptomic studies, with proteomic convergence largely absent. However, convergence emerged at the pathway level, with mitochondrial dysregulation as a recurrent feature across multiple data modalities. This pattern — limited reproducibility at individual molecular signals but consistency at the systems level — supports the unified model’s emphasis on coordinated multi-system dysfunction rather than single-gene defects.
Long-COVID as convergent evidence — with important caveats. The emergence of Long-COVID did not provide conceptual novelty, but it did provide statistical power, temporal anchoring (known infection date), and institutional legitimacy — enabling larger-scale investigation of biological processes relevant to ME/CFS. Large-scale Long-COVID cohorts have identified abnormalities in immune dysregulation, endothelial function, coagulation, metabolism, and viral persistence that map onto pre-existing ME/CFS research strands. This convergence suggests that some post-infectious biological mechanisms may be shared. (Watton and Prusty 2026)
However, three important qualifications apply. First, several objective measures show substantive divergence between Long-COVID and ME/CFS rather than simple convergence. Long-COVID patients with self-reported PEM show no 2-day CPET deterioration — the single most robust objective biomarker in ME/CFS (see Section Speculative Cross-Disease Connections on CPET divergence). Circulating cell-free mitochondrial DNA dynamics are directionally opposite: reduced ccf-mtDNA in Long-COVID (impaired mitophagy) versus elevated exosome-associated mtDNA post-exercise in ME/CFS (stress-triggered release). Shahbaz et al. found both overlap and divergence in immune cell phenotypes between Long-COVID-associated and idiopathic ME/CFS — shared immune activation but more pronounced T-cell exhaustion, NK-cell alteration, and MAIT/\(\gamma \delta\) T-cell depletion in the Long-COVID-associated group. Second, Long-COVID cohorts are heterogeneous and include many patients whose symptoms resolve within 12 months, whereas ME/CFS by definition requires 6 months and typically persists for years to decades — the comparison conveys partial rather than complete equivalence. Third, the claim that prior ME/CFS uncertainty “reflected insufficient investigative capacity, not the absence of biology” is a plausible inference, not a demonstrated fact. Alternative interpretations are not eliminated: genuine biological heterogeneity, diagnostic criteria so broad they produce irreproducible cohorts, or fundamentally different disease processes in subsets of patients could all contribute to prior null findings independent of funding level. The Keele et al. finding of pathway-level but not gene-level convergence is equally consistent with heterogeneous biology being studied with better tools as it is with a single biology finally illuminated by adequate methods. The question warrants an open mind, not premature closure. (Watton and Prusty 2026)
Key uncertainties. The Watton–Prusty model identifies critical unresolved questions: what upstream events stabilise persistent immune-metabolic dysregulation? Why do only a subset of individuals exposed to the same trigger develop chronic illness? What mechanism governs the transition from acute trigger to sustained dysregulation? Are the identified abnormalities primary drivers or secondary adaptations? Resolving these questions requires large deeply-phenotyped cohorts incorporating physiological challenge paradigms together with integrated multi-omic and functional approaches, including time-resolved single-cell transcriptomics (scSLAM-seq), exercise-challenged CSF proteomics, and plasma-mediated pathophysiology studies. (Watton and Prusty 2026) (Certainty: 0.55)
Do ME/CFS, Long COVID, and other post-infectious syndromes share a common core mechanism (impaired physiological resilience) despite differing initial triggers, or do they represent mechanistically distinct entities with superficial symptom overlap? The Shahbaz et al. comparison of Long-COVID-associated ME/CFS vs. idiopathic ME/CFS found both overlap and divergence: shared immune activation was evident, but Long-COVID-associated cases showed more pronounced T-cell exhaustion, NK-cell alteration, MAIT/\(\gamma \delta\) T-cell depletion, and inflammatory monocyte skewing. These findings support placement within a broader post-infectious framework while arguing against unqualified mechanistic equivalence. Longitudinal studies tracking the emergence and evolution of these signatures from acute infection through chronicity are needed to resolve this question. (Watton and Prusty 2026)
2 The Historical Arc: From Post-Infectious Framing to Biological Re-Framing
The Watton–Prusty review (Watton and Prusty 2026) provides a historical analysis contextualising current biological models within the broader evolution of ME/CFS understanding. This review synthesis should be read alongside primary historical sources and the responses of researchers whose work is critiqued. (Certainty: 0.45 — a review synthesis, not original historical research)
Early post-infectious framing. Descriptions of illness consistent with ME date to mid-20th-century epidemic outbreaks, most notably the Royal Free Hospital outbreak in London (1955). These accounts documented prolonged post-infectious disability with neurological and autonomic features, establishing an initial biomedical framing. The outbreak affected 292 staff members over several months and was investigated by multiple clinicians who concluded it represented an organic encephalomyelitis. However, in 1970, McEvedy and Beard retrospectively reclassified the Royal Free and other outbreaks as “mass psychogenic illness” — a records review, not an epidemiological investigation. This reclassification has itself been contested: physicians involved in the outbreak disputed the psychogenic interpretation, and a 2020 re-examination (Parsons, Medical Humanities) argued the psychogenic diagnosis reflected gendered assumptions about women’s health rather than dispassionate analysis. Nevertheless, the reclassification introduced an enduring scepticism that influenced subsequent research priorities and funding allocations. The episode illustrates how explanatory frameworks, once established in the absence of definitive laboratory tests, can persist beyond the evidence that originally supported them — a pattern relevant to both the psychogenic and the biological framings of ME/CFS. (Watton and Prusty 2026)
The strongest case for the biopsychosocial model. Engel’s biopsychosocial (BPS) model, properly understood, was conceived as an extension of biomedical reasoning — not a replacement. It proposed that health and disease arise from interactions between biological, psychological, and social factors, with biological processes remaining a central component. Applied to ME/CFS, the strongest form of the BPS argument holds: (a) all chronic illnesses, not just ME/CFS, involve interaction between biological pathology and psychological/social adaptation — the question is one of weighting, not category; (b) in the absence of validated biomarkers, a model incorporating reversible perpetuating factors (deconditioning, activity avoidance, illness beliefs) is a reasonable clinical framework rather than an ideological position — it provides clinicians with actionable treatment targets while biological research continues; (c) the original PACE protocol included therapist-guided, patient-negotiated activity targets with explicit instructions not to push beyond tolerated limits, which is more nuanced than simple “graded exercise”; (d) pre-PACE trials (Fulcher and White 1997; Powell et al. 2001) showed benefit on both subjective and some objective measures, and these findings deserve engagement rather than omission; and (e) the 2021 Cochrane review of exercise therapy for CFS (Larun et al.) concluded that exercise therapy probably reduces fatigue — a finding consistent with, not contradicted by, the PACE data, and one that must be weighed against the patient survey evidence of harm for a complete picture. (Watton and Prusty 2026)
The PACE trial: evidence, critique, and contested interpretation. The PACE trial (2011, \(n = 641\)) tested CBT and graded exercise therapy (GET) against specialist medical care and adaptive pacing therapy for CFS/ME, reporting that CBT and GET were “moderately effective.” The trial has since become one of the most heavily scrutinised studies in modern medicine. Subsequent critiques by patients, independent researchers, and investigative journalists identified: outcome switching (post-hoc relaxation of recovery thresholds such that a participant could deteriorate on two primary outcomes and still be classified as “recovered”), reliance on subjective outcomes in an unblinded trial (participants knew which treatment they received), and claims of recovery that were inconsistent with objective measures (the 6-minute walk test showed no clinically significant between-group difference at final follow-up). Freedom of Information tribunal rulings compelled data release; independent reanalyses of the trial data (by Alem Matthees, and by Wilshire et al.) demonstrated that the originally reported treatment effects were substantially attenuated or eliminated when the protocol-specified outcomes and thresholds were used. These critiques led to the 2021 revision of UK NICE guidance, which removed GET as a recommended treatment and reframed CBT as supportive rather than curative. (Watton and Prusty 2026)
Counterpoints that must be weighed. The PACE investigators maintain that: (a) the primary outcomes specified in the published protocol differed from those in the earlier trial registration, but both were published and available for comparison; (b) unblinding is inherent to any behavioural intervention trial and affects all such studies — singling out PACE applies a standard not consistently applied to other unblinded ME/CFS research; (c) self-report outcomes are the standard primary endpoints in most ME/CFS and chronic pain trials, including many studies cited favourably elsewhere in this paper — the epistemological problem of unblinding is universal in behavioural research, not unique to PACE; and (d) the 2021 NICE committee was subject to significant lobbying and parliamentary pressure from patient advocacy groups, raising the question of whether the reversal reflected pure evidential reassessment or also political dynamics — the same question the paper asks of PACE’s original adoption, and one that should be symmetrically applied. Other national guidelines (Australian, Dutch) retained graded exercise recommendations after 2021. The NICE reversal is not the universal consensus the term “reversal” implies. (Watton and Prusty 2026)
The broader lesson — symmetrically stated. ME/CFS illustrates two symmetrical risks in clinical reasoning: the risk of prematurely adopting explanatory models because they are convenient, scalable, or professionally reassuring (applicable to the BPS-as-applied, just as the BPS model might argue that premature adoption of unreplicated biological models carries the same risk), and the risk of dismissing treatment approaches before their evidence is fully and fairly evaluated. The availability of an explanation, whether psychosocial or biological, should not be conflated with its validity. In conditions of uncertainty, explanatory frameworks should remain continuously accountable to evolving evidence, with the same standards of scrutiny applied regardless of whether the framework supports or challenges the reader’s prior beliefs. (Watton and Prusty 2026)
From chronic underfunding to mechanistic investigation. The longstanding absence of definitive biomarkers or targeted therapies reflects, in part, a persistent mismatch between disease burden and biomedical investment. For decades, ME/CFS research lacked the funding to support large, well-phenotyped cohorts or to test systems-level hypotheses. The emergence of Long-COVID changed this: within months, unprecedented funding, infrastructure, and interdisciplinary efforts were mobilised, enabling more detailed investigation of biological processes relevant to ME/CFS. This investment demonstrated that post-infectious biological signals are detectable when studied at adequate scale — a finding that does not, on its own, resolve whether prior null results reflected insufficient methodology, genuine biological heterogeneity, or both. The Long-COVID experience suggests that increased investment in ME/CFS research is warranted, but the specific claims of the unified model must be judged on their own evidence, not on the funding differential. (Watton and Prusty 2026)
Additional historical context and limitations. This historical analysis is a review synthesis, not original historical research. The PACE trial’s original investigators maintain that methodological criticisms are overstated, and their published responses deserve engagement alongside the critiques. The McEvedy and Beard reclassification of the Royal Free outbreak has itself been challenged on methodological and sociological grounds. Engel’s original BPS framework was developed to counter biomedical reductionism, and some ME/CFS researchers have argued that the paper’s own multi-system model is broadly compatible with BPS principles, differing primarily in the weighting assigned to biological versus psychological factors. Readers should weigh the full record — trial protocols, reanalyses, investigator responses, independent critiques, and the 2021 NICE revision — when forming conclusions about this period of ME/CFS history. A steelman of any position is a prerequisite for a durable critique.
3 Neuroimmune Classification Framework
Certainty: 0.70. Blitshteyn, Doherty, and Steinman (Blitshteyn, Doherty, and Steinman 2026) have synthesized a convergent neuroimmune framework across POTS, ME/CFS, and Long COVID — arguing that these three conditions form a spectrum of neuroimmune disorders with shared pathophysiology. The core evidence for convergence rests on four pillars: (1) GPCR autoantibodies (adrenergic α1, β1/β2, muscarinic M2/M4) identified across all three conditions by multiple independent groups (Gunning et al. 2019) (Fedorowski et al. 2017) (Wallukat et al. 2021); (2) brainstem neuroinflammation at the dorsolateral inferior medulla as a shared CNS substrate (Blitshteyn 2025) (Wagoner et al. 2019); (3) vagus nerve / cholinergic anti-inflammatory pathway dysfunction as a common neuroimmune mechanism; and (4) immunomodulatory treatment overlap (IVIG, immunoadsorption, rituximab) with preliminary benefit in all three conditions.
Certainty basis: The GPCR autoantibody evidence is strongest in POTS (0.65–0.70 from functional validation including cell-based assays (Fedorowski et al. 2017) and rabbit passive transfer (Li et al. 2019)), moderate in Long COVID (0.60–0.70 from Wallukat 2021 bioassay evidence (Wallukat et al. 2021)), and moderate in ME/CFS (0.50–0.60 from multiple ELISA studies (Loebel et al. 2016) (Bynke et al. 2020)). The Germain 2025 null result in ME/CFS (n=172, REAP/Luminex) (Germain et al. 2025) introduces an unresolved assay-sensitivity issue. Brainstem neuroinflammation evidence remains modest (0.45–0.60). Clinical neuroimmune overlap between the three conditions is the highest-certainty pillar (0.80) based on epidemiological and clinical convergence (Moen and Iwasaki 2025) (El-Rhermoul et al. 2023).
Replication status: GPCR autoantibodies in POTS replicated by three independent groups (Oklahoma, Lund, CellTrend). In ME/CFS, two independent replications (Loebel 2016, Bynke 2020) but Germain 2025 null challenges reproducibility. The three-condition convergence is a narrative synthesis, not a head-to-head comparison study.
Limitations: Narrative review (not systematic); limited original data; GPCR autoantibody detection not standardized across labs; whether GPCR autoantibodies are causal or epiphenomenal debated; Germain 2025 null unresolved; brainstem neuroinflammation lacks direct human histopathology.
Convergence caveat — shared methodological vulnerability. The convergence described above draws on multiple lines of evidence that are individually weak. However, they do not constitute strong evidence because several key pillars (GPCR autoantibody prevalence, autonomic-autoantibody correlations, treatment response signals) depend on the same detection platform (CellTrend ELISA). If the CellTrend ELISA method is wrong — if it detects non-specific binding, if GPCR epitopes are lost during plate coating, or if the conformational epitopes required for functional activity are not preserved — then the autoantibody pillar collapses, weakening the neuroimmune convergence that depends on it. Convergence of multiple weak lines of evidence that share a single methodological vulnerability does not produce strong evidence. Independent validation on orthogonal platforms (Luminex, functional bioassay, mass spectrometry) for the same cohorts is essential before the convergence claim can be considered robust.
Certainty: 0.60. Functional GPCR autoantibodies targeting α1/β1/β2 adrenergic and M2/M4 muscarinic receptors represent a shared biomarker across the three conditions, with prevalence decreasing from POTS (89% by Gunning 2019 (Gunning et al. 2019), 8/17 by Fedorowski 2017 functional assay (Fedorowski et al. 2017)) through Long COVID (Wallukat 2021 functional bioassay (Wallukat et al. 2021)) to ME/CFS (29.5–91% prevalence depending on assay and cohort (Loebel et al. 2016) (Bynke et al. 2020)). The autoantibodies demonstrate functional activity in cell-based assays (Fedorowski et al. 2017) and animal passive transfer models (Li et al. 2019), supporting pathogenicity.
Replication status: Multiple independent cohorts in POTS (Oklahoma, Lund, Korean). Two independent cohorts in ME/CFS (Würzburg, Uppsala). Germain 2025 (Germain et al. 2025) null in ME/CFS (n=172 using REAP/Luminex) constitutes a substantive non-replication that may reflect assay sensitivity differences (conformational vs. linear epitopes).
Limitations: ELISA-based detection is not standardized across labs; functional bioassays are not scalable to routine clinical use; the Germain 2025 null requires resolution; no longitudinal data exist on titer stability.
Assay artifact as the single biggest threat to the GPCR autoantibody evidence base. Three detection platforms used across studies produce discordant results: CellTrend ELISA (conformation-preserving, full-length GPCRs on HEK cells) finds positive signals across ME/CFS, POTS, and Long COVID; functional cardiomyocyte bioassays (beating-rate assays) confirm activity in POTS and Long COVID but have not been systematically applied to ME/CFS; Luminex/REAP (linear epitope fragments) finds no signal in the largest ME/CFS screen to date (Germain et al. 2025). If the CellTrend ELISA method detects non-specific binding, conformational artifacts, or irrelevant IgG subclasses rather than pathogenic autoantibodies, then the entire GPCR AAb evidence pillar — on which the transdiagnostic biomarker claim, the treatment rationale, and the neuroimmune convergence rest — is compromised. Resolving this discordance through head-to-head platform comparisons in the same patient cohort is the single highest priority for the GPCR autoantibody field.
Falsifiable prediction: Two-way ANOVA (condition × AAb panel) will show significant main effect of condition (POTS → ME/CFS → LC → HC, p < 0.01) with effect size η² ≥ 0.25 for β2-AR and M3 targets. Falsified if no significant condition effect or η² < 0.10.
{{/* H4: GPCR AAb → mast cell sensitization loop (Tier 2, cert 0.35) */}}
Certainty: 0.35. GPCR autoantibodies (β2-adrenergic, M3 muscarinic) may directly sensitize mast cells, which express the cognate receptors. β2-AR autoantibodies can chronically activate mast cell β2-AR, leading to desensitization and removal of the adrenergic brake on degranulation. M3 autoantibodies directly activate muscarinic receptors on mast cells, triggering histamine and tryptase release. This creates a bidirectional amplification loop: GPCR autoantibodies → mast cell degranulation → inflammatory mediators → increased BBB permeability → more autoantibody CNS access → brainstem neuroinflammation → autonomic dysregulation → further immune dysregulation. The loop explains why both MCAS-targeted treatment and immunomodulation show benefit — they interrupt different points of the same autoantibody-mast cell loop. (Blitshteyn, Doherty, and Steinman 2026) (Blitshteyn 2025)
Replication status: Mast cell β2-AR expression and M3-muscarinic expression are established in the mast cell biology literature. The specific GPCR autoantibody-mast cell interaction has not been tested in ME/CFS. This is a mechanistic synthesis from two independently established lines.
Limitations: No direct evidence of GPCR autoantibody binding to mast cells from ME/CFS patients. The direction of β2-AR autoantibody effect (agonistic vs antagonistic) on mast cells is unknown. The loop model is a hypothesis requiring direct cellular testing.
Falsifiable prediction: Plasma tryptase will correlate with β2-AR AAb titer (r > 0.4, p < 0.01) in ME/CFS patients (n ≥ 30). Falsified if r ≤ 0.2 or no significant correlation.
{{/* H6: CAP-autoantibody blockade (Tier 1, cert 0.40) */}}
Certainty: 0.40. GPCR autoantibodies (particularly M2/M4 muscarinic and β2-adrenergic) may block the cholinergic anti-inflammatory pathway (CAP) at multiple points: (a) M2 autoantibodies on cardiac vagal terminals impair vagal efferent signaling before the spleen; (b) β2-AR autoantibodies on splenic T cells block the adrenergic-to-cholinergic conversion step that normally suppresses macrophage TNF-α. The result is the vagus brake on systemic inflammation becoming disengaged, producing the low-grade inflammatory state characteristic of ME/CFS, POTS, and Long COVID. This hypothesis predicts that taVNS (transcutaneous auricular vagus nerve stimulation) efficacy depends on GPCR autoantibody profile: patients with high β2-AR autoantibodies should show blunted taVNS response because the CAP effector step is blocked at the splenic T cell level, regardless of vagal afferent activation. (Blitshteyn, Doherty, and Steinman 2026) (Moen and Iwasaki 2025) (Azcue et al. 2026)
Replication status: CAP blockade by GPCR autoantibodies is a mechanistic hypothesis that has not been directly tested in any condition. The individual components (M2 AAb on vagal terminals, β2-AR on splenic T cells) are established in isolated experiments but not as an integrated pathway.
Limitations: No direct test of CAP function in GPCR autoantibody-positive ME/CFS. The splenic β2-AR T cell step is established in the basic immunology literature but the functional effect of patient-derived GPCR autoantibodies on this pathway has not been assayed. The taVNS stratification prediction is untested.
{{/* H7: Structural vagal cholinergic denervation → CAP disruption (cert 0.35; cross-disease extrapolation from Acanfora/Woo/Lladós Long COVID; vagal-gastric-denervation-longcovid stream) */}}
Certainty: 0.35. Two mechanistically distinct routes could disengage the vagal brake on systemic inflammation in Long COVID (with unknown generalizability to other post-viral conditions, including ME/CFS). This certainty is below the 0.45 assigned to the underlying Long COVID clinical finding (First In Vivo Report of Structural Vagal Cholinergic Denervation in Long COVID (Single Unreplicated Study)) because the present claim — that Acanfora’s finding extrapolates to ME/CFS and constitutes a distinct structural CAP-disengagement route — adds a cross-disease inference step not yet supported by direct evidence. The first, described elsewhere in this framework, is functional: GPCR autoantibodies interfere with cholinergic signaling without destroying nerves (Cholinergic Anti-Inflammatory Pathway Blockade by GPCR Autoantibodies — Potential Mechanism for Vagal Treatment Stratification). This hypothesis proposes a second, structural route: physical loss of vagal cholinergic fibers. Acanfora et al. (Acanfora et al. 2026) reported selective depletion of VIP-positive (cholinergic) fibers in the gastric mucosa of Long COVID patients, correlating with reduced vagal HRV indices — a single unreplicated in vivo observation (\(n=12\)) that autonomic dysfunction may rest on an anatomical lesion, not merely altered signaling. A proximal mechanism is available: Woo et al. (Woo et al. 2023) found SARS-CoV-2 RNA and monocyte infiltration within the vagus nerve itself, and Lladós et al. (Lladós et al. 2024) reported vagus-nerve thickening on ultrasound with reduced gastrointestinal peristalsis post-COVID — providing the first direct in vivo evidence in Long COVID consistent with (not confirmation of) the decade-old ME/CFS-specific prediction of VanElzakker (VanElzakker 2013) that pathogen-driven vagal injury underlies post-viral symptomatology; VanElzakker’s prediction concerned ME/CFS, whereas all three studies are Long COVID/acute COVID, so the prediction is corroborated only by analogy. Because the vagus is the efferent arm of the cholinergic anti-inflammatory pathway (CAP; vagus → \(\alpha\) 7-nicotinic acetylcholine receptor → suppression of macrophage TNF-\(\alpha\); a reflex demonstrated predominantly in animal models with limited direct human clinical validation), structural cholinergic denervation would in principle disengage the CAP (Tracey 2002) (Bonaz, Bazin, and Pellissier 2018), plausibly producing a self-sustaining proinflammatory state that could maintain persistent post-viral symptoms (Acanfora studied Long COVID patients, a population with high but incomplete ME/CFS diagnostic overlap). The two routes are not necessarily complementary: if functional autoantibody blockade alone proves sufficient to disengage the CAP, the structural route may be redundant rather than additive — and both could be unnecessary if neither proves causal in ME/CFS. The extrapolation to ME/CFS is currently indirect and, if the mechanism operates at all, would plausibly be restricted to an autonomic-predominant subset rather than all patients: no ME/CFS-specific vagal-mucosal histology exists, and ME/CFS shows somatic small fiber neuropathy (Peripheral Nervous System) whereas Long COVID (Acanfora) showed selectively visceral loss (Does Post-Viral Small Fiber Pathology Preferentially Target Visceral (Vagal) or Somatic Fibers?), so the two conditions may differ in fiber tropism.
Testable prediction: If structural vagal denervation drives CAP disengagement, then (a) ME/CFS patients undergoing gastric mucosal biopsy should show reduced VIP-positive fiber density relative to controls; (b) mucosal cholinergic fiber density should correlate inversely with circulating TNF-\(\alpha\) / CRP and positively with HF-HRV; and (c) this correlation should hold independently of GPCR autoantibody status, distinguishing the structural route from the functional autoantibody route. A stronger, two-hit version predicts super-additive CAP failure in patients positive for both structural fiber loss and CAP-relevant GPCR autoantibodies (Cholinergic Anti-Inflammatory Pathway Blockade by GPCR Autoantibodies — Potential Mechanism for Vagal Treatment Stratification): such double-positive patients should show the most refractory inflammation and the poorest taVNS response, because the pathway is disrupted at both the fiber and the signaling level. If ME/CFS gastric innervation is normal, if inflammation tracks autoantibody titres but not fiber density, or if double-positive patients are no worse than either single-positive subset, the corresponding claim is not supported for ME/CFS.
Treatment implication: Structural denervation, unlike functional blockade, may be only partially reversible. If confirmed in ME/CFS, this would raise the question of whether early intervention could preserve surviving fibers — a question that has no empirical basis at present. It provides a mechanistic rationale (not a clinical recommendation) for investigating vagus-nerve stimulation and cholinergic modulation — see the treatment chapters for the (limited) evidence base. This is a mechanistic hypothesis, not a basis for treatment.
Limitations: All direct structural evidence is Long-COVID-specific (Acanfora \(n=12\), single-center, dyspeptic controls; Woo postmortem/acute; Lladós pilot \(n=30\)). The CAP linkage (Tracey, Bonaz) is foundational and well-replicated but predominantly animal-derived and not ME/CFS-specific. The structural and functional routes may coexist, may be redundant (one sufficient, the other unnecessary), or may both be irrelevant to ME/CFS. No study has yet measured gastric cholinergic innervation in ME/CFS. Not independently replicated. The null hypothesis — that these gastric findings reflect Long-COVID-specific pathophysiology, an enteric (not vagal) or methodological artifact, or deconditioning, with no relevance to ME/CFS — remains at least as plausible as the structural hypothesis and should be treated as the default until ME/CFS-specific replication exists (see Five Unresolved Threats to the Structural Vagal Denervation Model).
Consequence: This offers a concrete, biopsy-testable way to ask whether the autonomic problems in ME/CFS come from destroyed vagal nerves versus merely blocked signaling — a distinction that matters because destroyed nerves may need to be protected early rather than un-blocked later, and the answer would tell researchers whether treatments aimed at stimulating the vagus nerve are even targeting an intact structure.
{{/* Critical categories 10-12: three threats to the structural model (vagal-enteric ambiguity, deconditioning, winner’s curse) */}}
Certainty of the critique: moderate. (Origin: brainstorm critical categories 10–12.) Beyond the sample-size caveats noted with Structural Vagal Cholinergic Denervation as an Anatomical Substrate for Post-Viral Dysautonomia and Sustained Inflammation, five specific threats could undermine the structural-denervation interpretation and its extrapolation to ME/CFS. (1) Vagal-versus-enteric ambiguity. The VIP+ fibers Acanfora quantified in the gastric mucosa cannot, by immunohistochemistry alone, be assigned to extrinsic vagal terminals versus intrinsic enteric neurons (Acanfora et al. 2026); the authors state this explicitly. If the lost fibers are predominantly enteric, the central “vagal denervation” label — and the inference to the cholinergic anti-inflammatory reflex, which depends on extrinsic vagal efferents — is weakened. Single-nucleus RNA-seq or retrograde tracing would be required to resolve the source. (2) Deconditioning confound. Reduced HF-HRV and even peripheral fiber changes can arise from prolonged inactivity, which is near-universal in both Long COVID and ME/CFS. Neither Acanfora nor the correlated HRV finding was controlled for physical conditioning, so the structural deficit could be partly a consequence of the illness state rather than an upstream cause. (3) Winner’s curse. With \(n=12\), the reported correlations (\(R=0.50\)–$ 0.61$; \(r^2 \approx 0.25\)–$ 0.37$, i.e. most variance unexplained) have wide confidence intervals and are prone to inflated effect sizes; independent replication in a larger, activity-matched, healthy-controlled cohort is the minimum needed before the effect magnitude can be trusted. (4) Dyspeptic-control confound. Acanfora’s controls were patients undergoing endoscopy for functional dyspepsia, not healthy volunteers; functional-dyspepsia patients may themselves have altered gastric innervation, so the case–control difference could be attenuated, exaggerated, or spurious — the direction of bias is unknown. (5) Functional-route redundancy. The paper’s pre-existing functional model — CAP blockade by GPCR autoantibodies (Cholinergic Anti-Inflammatory Pathway Blockade by GPCR Autoantibodies — Potential Mechanism for Vagal Treatment Stratification) — may by itself be sufficient to disengage the cholinergic anti-inflammatory pathway. If so, structural denervation is not a complementary “second hit” but a potentially redundant explanation, and the two accounts are competing, not additive; neither has been shown causal in ME/CFS.
Consequence: These are the specific results a skeptic would demand before accepting that vagal nerves are physically destroyed (rather than the finding reflecting gut-local nerves, inactivity, an unrepresentative control group, a small-sample fluke, or a mechanism already covered by autoantibodies) — so they define exactly what the next, larger study must measure, and they caution clinicians against acting on the structural model until it is replicated.
{{/* Phase 10a synthesis: vagal structural-functional axis (condenses ach/hyp/spec/lim/oq across ch08/ch10/ch13) */}}
The environments in this cycle assemble a two-route model of how the vagal brake on inflammation might fail in post-viral illness, and how confidently that model transfers to ME/CFS. Three interpretive themes can be drawn — though all three rest largely on Acanfora’s single \(n=12\) Long COVID study and so represent facets of one dataset rather than three independent lines of evidence. First, autonomic dysfunction may have a structural substrate, not only a functional one: Acanfora’s in vivo observation of selective cholinergic (VIP+) gastric-mucosal denervation correlating with vagal HRV (First In Vivo Report of Structural Vagal Cholinergic Denervation in Long COVID (Single Unreplicated Study)), read alongside vagus-nerve viral inflammation and ultrasound thickening and the long-standing (largely animal-model) cholinergic-anti-inflammatory-pathway biology, is consistent with a structural→CAP-disengagement→inflammation chain (Structural Vagal Cholinergic Denervation as an Anatomical Substrate for Post-Viral Dysautonomia and Sustained Inflammation) — though the correlations are modest (\(r^2 \approx 0.25\)–$ 0.37$) and the chain is inferential, not demonstrated. Second, the structural route and the functional autoantibody route (Cholinergic Anti-Inflammatory Pathway Blockade by GPCR Autoantibodies — Potential Mechanism for Vagal Treatment Stratification) reach the same endpoint (a disengaged vagal anti-inflammatory reflex) by different means; they could be complementary (the two-hit prediction that patients with both are worst affected), but one could equally render the other redundant, or neither may prove causal in ME/CFS — current evidence cannot distinguish these. Third, the model is tentatively formalizable as a distinct structural-capacity variable that would impose a ceiling on vagus-stimulation efficacy (Structural Vagal Damage as a Distinct Upstream Node in the Causal DAG), though that node is a pre-data placeholder. Against these themes, the evidence imposes hard constraints: every direct structural finding is Long-COVID-specific with no ME/CFS gastric histology yet obtained, ME/CFS instead shows somatic small fiber neuropathy while Long COVID showed selectively visceral loss (Does Post-Viral Small Fiber Pathology Preferentially Target Visceral (Vagal) or Somatic Fibers?), and the founding study cannot even distinguish vagal from enteric fibers, is uncontrolled for deconditioning, uses dyspeptic controls, and rests on twelve patients (Five Unresolved Threats to the Structural Vagal Denervation Model). The null — that these gastric findings reflect Long-COVID-specific pathophysiology, an enteric or methodological artifact, or deconditioning, with no relevance to ME/CFS — remains at least as plausible and should be the default until ME/CFS-specific replication. The central open question is therefore not whether the vagal anti-inflammatory reflex matters in post-viral illness — the CAP biology is reasonably established (predominantly in animal models; its clinical significance in ME/CFS awaits direct confirmation) — but whether ME/CFS specifically involves structural vagal fiber loss at all, which a single paired gastric-and-skin-biopsy study in ME/CFS could largely resolve.
Consequence: Taken together, these findings raise the possibility that some autonomic problems in post-viral illness reflect physically damaged vagal nerves rather than merely mis-tuned signaling — which would matter for how early treatment must begin and whether nerve-stimulation therapies can work at all — but because the direct evidence is entirely from Long COVID and rests on one small study, the honest current status is a well-motivated hypothesis awaiting a decisive ME/CFS biopsy study, not an established mechanism.