Solving ME/CFS · evidence atlas · Disease model · Data catalogue

ME/CFS evidence atlas

A living review of what research has found in ME/CFS and overlapping post-infectious illness, including long COVID. It is organised by body system, graded by how well each finding holds up, and linked to the hypotheses it bears on. Written for someone new to the field and for researchers who know one corner of it well.

Last built 2026-10-28. 228 findings, 670 study entries, 31 hypotheses. Entries marked "unverified" were not re-checked against the paper. Download everything as JSON.

New to this? Read the overview, then the hypotheses, then the body system that interests you. Researcher? Use the search box, the grade filters or the study index.

Replicated Found in several independent cohorts (27) Several studies More than one study, not yet independently settled (50) Early One or two studies, usually small (82) Contested Conflicting results or method doubts (51) Refuted Well-designed work found no effect (18)

Overview by body system

Blood flow, vessels and blood

The most consistent vascular findings in ME/CFS concern circulation when upright: cerebral blood flow falls much more than normal on tilt, stroke volume is low, red-cell and blood volume are below predicted in a large minority to about half, and during exercise the heart is under-filled (preload failure). These are reported by several groups, though the cerebral blood flow and invasive exercise data each come mostly from one centre, and a 2026 study found blood volume did not predict who has orthostatic abnormalities. Reduced endothelial function (FMD, PORH) has been found in three independent cohorts, and newer work shows fewer muscle capillaries than bed rest would explain and narrowed retinal vessels in post-COVID ME/CFS. Complement activation is replicated in long COVID but inconsistent in ME/CFS. Microclots remain contested on method, and pooled blood proteomics do not show raised coagulation, platelet or endothelial-activation proteins.

23 findings · 10 replicated or multi-study

Muscle, exertion and energy

People with ME/CFS reliably have lower aerobic capacity and a low peak heart rate on exercise testing, and they report delayed, prolonged PEM that took about two weeks to resolve after a 2-day test. Whether performance measurably drops on a second test 24 hours later is now contested: several linked groups and two meta-analyses say yes, mainly at the ventilatory threshold, while a 2026 replication found no change. Invasive testing points to poor cardiac filling and poor oxygen extraction, and newer biopsy work from one Amsterdam group finds fewer oxidative fibres, fewer capillaries and lower mitochondrial capacity in ME/CFS that bed rest does not reproduce, plus post-exertion muscle damage and amyloid-containing deposits in long COVID. Mitochondrial studies in blood cells disagree in direction, and the idea that patient serum alters muscle-cell mitochondria failed a large pre-registered replication (Ryback 2026). Post-exertion omics (metabolomics, proteomics, EVs) show resting differences and recovery-phase changes, but within-person exercise effects are modest and mostly unreplicated.

22 findings · 7 replicated or multi-study

Nervous system and brain

The most consistent nervous-system findings in ME/CFS are physiological rather than anatomical. Brain blood flow drops more than normal when patients are upright, often with over-breathing (low CO2), and resting heart-rate variability is lower; both are seen across several independent labs. Small-fibre nerve loss on skin biopsy is common in referral clinics (roughly 30-55%), but the main controlled NIH study found no difference, and nobody has yet shown that nerve loss explains symptoms. Brain imaging gives many positive but small and mostly unreplicated results: raised ventricular lactate (mostly one group), raised brain temperature (one study), TSPO-PET neuroinflammation (one positive and one negative ME/CFS study), and newer diffusion MRI signals. Structural MRI, routine sleep studies and markers of nerve injury are mostly normal or inconsistent. The NIH 'effort preference' interpretation has been seriously challenged, because patients in that study often could not complete the hard task. Claims about craniocervical instability rest on uncontrolled clinic series and case reports.

26 findings · 10 replicated or multi-study

Immune system

Immunology is the oldest and most heavily worked area of ME/CFS research, and the one with the widest gap between positive reports and independent replication. Reduced natural-killer-cell killing is the most repeated finding (meta-analytic Hedges g about 0.96 across 28 papers), yet the largest standardised multi-site attempt (MCAM, 174 vs 86) and two careful phenotyping studies found nothing, so the honest grade is 'contested and not usable as a test'. Several newer cell-level findings are more consistent in direction than in detail: CD8 T-cell exhaustion programmes, shifts in B-cell maturation (more naive, fewer switched-memory), monocyte dysregulation, and an exaggerated innate response to microbial stimulation that worsens after exercise. Cytokines look disturbed in aggregate (TNF, IL-2, IL-4, TGF-beta, CRP in meta-analysis) but no single cytokine survives as a marker, TGF-beta turned out to be largely a blood-handling artefact, and the clearest pattern is that abnormalities depend on illness duration (Hornig 2015) and severity (Montoya 2017) rather than on diagnosis alone. The autoimmunity strand has had the clearest negative resolution: GPCR autoantibodies against beta2/M3/M4 receptors are not reproduced by unbiased antigen profiling, removing them by immunoadsorption fails against sham, rituximab failed in a 151-patient RCT, and BC007's larger trial failed; what survives is narrower and interesting - patient IgG does things to endothelial mitochondria and, in long COVID, transfers pain and fatigue-like phenotypes to mice, and plasma-cell depletion with daratumumab produced striking but uncontrolled responses in 6 of 10 patients. Common-variant genetics (DecodeME) shows no enrichment in any blood immune cell type, which pushes most of these immune findings toward being consequences or state markers rather than inherited causes.

31 findings · 9 replicated or multi-study

Infection and viruses

The most solid finding in this area is about onset: about 1 in 10 people meet ME/CFS criteria six months after several unrelated infections (EBV, Q fever, Ross River virus, Giardia, SARS-CoV-2), and how severe the first illness was predicts this better than psychological factors do. Post-COVID cohorts confirm that infection raises the rate of new ME/CFS several-fold. Whether any virus is still active in people who already have ME/CFS is far less clear. Blood antibody and PCR studies of herpesviruses give mixed results. Tissue findings for enterovirus, HHV-6 and SARS-CoV-2 are small or unreplicated, though some of them are striking. SARS-CoV-2 antigen and RNA do persist in a minority of people for months, but three randomised trials of nirmatrelvir in established long COVID (about 1,200 people) were negative, and antiviral trials in ME/CFS are tiny and unreplicated. The XMRV episode, a retrovirus finding that turned out to be lab contamination, is a reminder that one striking virus finding needs blinded multi-lab replication.

24 findings · 10 replicated or multi-study

Metabolism and blood omics

Blood metabolomics and proteomics consistently find that ME/CFS patients differ from controls on average, but the differences are small (mostly 0.2-0.5 SD), partly cohort-specific, and confounded by medication, caffeine, diet and sample handling. The better-supported signals are modest: lower plasmalogens and some phosphatidylcholines, slightly lower carnitines, lower gut-derived IPA, lower aromatic lactates (a new cross-cohort re-analysis finding), higher methionine sulfoxide, and lower muscle-derived proteins in blood that are not explained by activity. Bigger claims, such as a uniform hypometabolic 'cell danger' state, an IDO metabolic trap, divergent metabolic or proteomic responses to exercise, reproducible methylation signatures, and diagnostic classifiers with near-perfect accuracy, have not held up in independent data or remain untested outside the original lab. No omics test is ready for diagnosis. The largest gaps are severe patients, sampling during the 48-72 h PEM window, disease controls, and pairing omics with fresh-cell function.

27 findings · 9 replicated or multi-study

Hormones

The most consistent hormone finding in ME/CFS is a mildly under-active stress-hormone (HPA) axis: on average cortisol is slightly low in saliva, urine and hair, and feedback suppression is exaggerated. A 2026 meta-analysis of 46 case-control studies confirms this pattern. The shifts are small, most patients fall inside normal reference ranges, and they may be a consequence of illness, sleep disruption or inactivity rather than a cause. Low-dose hydrocortisone gave at best a small, short-lived benefit with side effects, and fludrocortisone failed in two large trials. Newer leads concern the hormones that hold water and salt in the body: low vasopressin and a 'renin-aldosterone paradox' that would fit the low blood volume seen in many patients. These come from small or uncontrolled studies, and the vasopressin finding is not matched by copeptin, which is the more reliable test. The female preponderance (about 1.5 to 3 times more women) is solid, but genetics shows no sex-specific risk variants, and the role of sex hormones, menopause and gynaecological conditions is still only descriptive.

15 findings · 4 replicated or multi-study

Gut and microbiome

Gut bacteria differ between people with ME/CFS and healthy controls more consistently than almost any blood measure. Fewer butyrate-producing species (Faecalibacterium, Roseburia, Coprococcus, Anaerostipes, Eubacterium rectale) and more Ruminococcus gnavus-type species appear in at least four independent cohorts across three labs, including a pre-specified test in the NIH cohort. The same pattern appears in long COVID. Blood levels of a gut-made metabolite, indole-3-propionate (IPA), are lower in four of four plasma cohorts, and re-analysis traces this to the R. gnavus-type overgrowth. What is not established is whether any of this causes symptoms. Dysbiosis is milder in long-standing illness, does not track severity reliably, and genetically predicted gut metabolite levels do not change ME/CFS risk. Markers of a 'leaky gut' (LPS, sCD14) come from a handful of studies, and the only randomised faecal transplant trial (n=11) was negative. Larger trials are under way.

14 findings · 6 replicated or multi-study

Genetics

ME/CFS runs in families, and twin, genealogy and three-generation cohort studies put familial or heritable risk at roughly 20 to 50%. In 2025 DecodeME (15,579 cases) found the first eight robust genome-wide loci, with SNP heritability of about 8 to 10%. Each variant shifts risk only slightly (odds ratios 1.07 to 1.10), so no gene test can diagnose or predict ME/CFS. The overall genetic signal replicates in independent biobanks (MVP, FinnGen, UK Biobank), but the individual loci mostly do not yet, partly because those biobanks use loose case definitions. Across several analyses the signal points to neurons in the brain (and possibly the pituitary), not blood immune cells. It overlaps strongly with chronic pain, fibromyalgia, IBS and migraine, and partly with depression, though the specific ME/CFS loci do not share causal variants with depression. Earlier small candidate-gene and HLA findings mostly failed to replicate. Several 2025–2026 follow-ups (Mendelian randomisation, rare variants, a UK Biobank/All of Us GWAS) are preprints that still need confirmation.

17 findings · 6 replicated or multi-study

Treatment trials and natural history

No treatment has replicated, placebo-controlled evidence of benefit in ME/CFS or post-COVID ME/CFS. Since 2019 the large, well-controlled trials have been negative: rituximab, Paxlovid (three trials), immunoadsorption against sham (two trials), ivabradine for long-COVID POTS, temelimab, and low-dose naltrexone for fatigue. The positive signals (daratumumab, cyclophosphamide, low-dose rapamycin, low-dose aripiprazole, single-dose pyridostigmine, 40-session HBOT) come from small, open-label or single-centre studies, and they need to be read against placebo response rates of 35-40% or more. Graded exercise is no longer recommended in England (NICE 2021), and patient surveys and a German rehab study report frequent worsening, while pacing and volume/compression measures are the most highly rated but untested by trials. Natural history data show low full recovery in adults (median ~5%), prevalence of ~0.2-1%, a female excess of 3-4:1, no clear rise in all-cause mortality, and a marked rise in suicide.

29 findings · 6 replicated or multi-study

Hypotheses

The main explanations proposed for ME/CFS, with the evidence for and against each. Several can be true at once, and many describe different links in the same chain. Tap one to open it.

Endothelial dysfunctionThe lining of blood vessels fails to dilate and regulate flow, so tissues are under-supplied during effort. active

Endothelial cells normally release nitric oxide and other signals so arteries widen when tissue needs more blood. In ME/CFS this response is reported to be blunted, possibly through autoantibodies, inflammation or oxidative stress. Poor flow regulation would leave muscle and brain short of oxygen during exertion and could set off PEM. It is a mechanism layer rather than a root cause, and it could result from several upstream triggers.

Proponents: Scheibenbogen and Wirth (Charité), Fluge and Mella (Bergen), Sandvik/Sørland (Bergen)

Evidence for

Flow-mediated dilation 5.1% vs 8.2% (Sandvik 2023, n=39); reduced endothelial function in CycloME substudy (Sørland 2021)
Patient IgG fragments endothelial mitochondria vs MS and healthy IgG (Prusty 2026, single lab)
Lower vessel-wall/connective-tissue proteins (ITGAV, NOTCH3, TGF-beta set) in 4 cohorts and long COVID (Solving ME/CFS re-analysis)

Evidence against

Effects are average differences with large overlap; deconditioning not always excluded
BC007 and immunoadsorption (aimed at vasoactive GPCR antibodies) failed controlled trials
Vericiguat (VERI-LONG) results not reported

Predicts

Lower flow-mediated dilation and reactive hyperaemia in patients
Endothelial markers or patient IgG effects on endothelial cells in vitro
Vasoactive treatments improving exercise tolerance

Tests that would settle it

Within-person FMD before and during PEM versus sedentary controls
Placebo-controlled trial of a vasoactive drug with 2-day CPET endpoint

Notes from each body system

Blood flow, vessels and blood: for: FMD/PORH/EndoPAT reduced in 3 independent cohorts; patient plasma/IgG impairs endothelial cells in vitro in 3 labs; retinal vessel narrowing in post-COVID ME/CFS.; against: Inactivity alone lowers FMD; blood endothelial-activation proteins are not raised in pooled proteomics; no in vitro assay replicated across labs.

Linked findings

Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, including in patients with normal heart rate and blood pressure responses. Several studies
Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia are reduced in ME/CFS in two independent Norwegian cohorts and by peripheral arterial tonometry in Berlin. Replicated
Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitric oxide production, more mitochondrial fragmentation, and anti-endothelial antibody binding. Early
Retinal imaging shows narrower arterioles, lower arteriolar-to-venular ratio, reduced flicker-induced venular dilation and lower capillary vessel density in post-COVID patients, most pronounced in those meeting ME/CFS criteria. Several studies
Sublingual video-microscopy shows reduced perfused capillary density and glycocalyx changes in long COVID, persisting many months after infection. Early
Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per fibre size in both ME/CFS and long COVID, unlike the pattern after 60 days of bed rest; thickened capillary basement membrane has been reported but is not yet peer-reviewed. Early
Pericytes and vascular smooth-muscle cells have not been examined directly in ME/CFS tissue, but blood proteins from mural cells and vessel matrix (NOTCH3, ITGAV, TGF-beta set, LUM) are lower across several ME/CFS cohorts and in long COVID. Early
Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoactive proteins in ME/CFS; most named candidates are null and the endothelial-activation set trends down. Contested
VEGF-A has been reported lower, not higher, in ME/CFS plasma, and proteomic re-analysis finds no consistent change; long COVID cohorts more often report raised VEGF. Contested
Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (preload failure) and, in a high-flow subgroup, poor systemic oxygen extraction by tissue. Several studies
23Na-MRI in 6 women with ME/CFS and 6 controls found higher resting muscle sodium in all lower-leg compartments and a larger sodium rise after exercise, inversely related to grip strength. Early
ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy, lower mitochondrial capacity and fewer capillaries per fibre, a pattern that 60 days of strict bed rest in healthy people did not reproduce. Early
Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 26% vs 7% in controls), including in patients whose heart rate and blood pressure stay normal. Replicated
In long COVID, one [11C]PBR28 PET study found widespread raised TSPO signal that correlated with vascular markers, and a [11C]PK11195 study found no group difference. Contested
Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS studies, but the results are inconsistent and often confounded by unmeasured orthostatic intolerance. Contested
Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30% of patients by ELISA, is not reproduced by unbiased high-throughput antigen profiling. Refuted
BC007 (rovunaptabin), an aptamer that neutralises GPCR autoantibodies, showed secondary-endpoint fatigue improvement in a 30-patient exploratory trial but failed in the larger company-run phase II. Contested
Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human endothelial cells, an effect absent with multiple-sclerosis and healthy IgG and lost on Fab/Fc cleavage. Early
Newly reported elevations of serum IL-11 and MMP-9 in female ME/CFS patients are unreplicated single-cohort findings. Early
Most of the ~845 FDR-significant serum proteins in Hoel 2026 appear driven by a sample-handling (cell-leak) signal that is absent from other cohorts; about 13 survive adjustment. Contested
Hepatocyte enzymes (ADH4, GSTA1, AGXT, BCHE) are modestly higher, and vascular mural/matrix proteins (ITGAV, NOTCH3, LUM, THBS1, TGF-beta family) modestly lower, in more than one cohort. Early
The aptamer BC007 (rovunaptabin) failed to beat placebo in the company's 114-person long-COVID trial; a 30-person investigator-led trial reported secondary-endpoint gains. Contested

Capillary and microvascular damageCapillaries in muscle and elsewhere are structurally damaged or rarefied, limiting oxygen delivery. active

Muscle biopsies show thickened capillary basement membranes and fewer capillaries, and long-COVID studies show capillary loss under the tongue and in the retina. If blood reaches the tissue but bypasses capillary beds or cannot exchange oxygen, extraction falls, as seen in the 'high-flow' iCPET phenotype. This would explain exertional intolerance unrelated to fitness. Cause could be immune, complement or viral damage to pericytes.

Proponents: Wüst group (VU Amsterdam), Appelman/van Vugt (Amsterdam UMC), Systrom (high-flow preload failure)

Evidence for

Capillary basement membrane thickening and lower capillarisation in ME/CFS and post-COVID biopsies (Wüst group 2025 preprint)
Muscle damage, amyloid deposits and mitochondrial drop after exercise in long COVID (Appelman 2024)
Reduced oxygen extraction in preload failure (Joseph 2021)
40-session HBOT improved symptoms in one RCT (Zilberman-Itskovich 2022)

Evidence against

Biopsy studies are small and largely from one network
10-session HBOT trial negative (HOT-LoCO 2025)

Predicts

Low oxygen extraction on iCPET
Basement membrane thickening and capillary rarefaction on biopsy
Improvement with treatments that repair microvasculature

Tests that would settle it

Blinded biopsy morphometry in large ME/CFS vs deconditioned controls
Correlation of capillary measures with 2-day CPET in the same patients

Notes from each body system

Blood flow, vessels and blood: for: Lower muscle capillarisation unlike bed rest (Charlton 2026); basement membrane thickening (poster); retinal and sublingual capillary loss in long COVID; poor O2 extraction subgroup.; against: Mostly long COVID or small samples; basement-membrane data not peer-reviewed; no severe patients biopsied.
Muscle, exertion and energy: for: Lower capillary-to-fibre ratio in ME/CFS muscle not reproduced by bed rest (Charlton 2026); impaired O2 extraction in high-flow iCPET subgroup (Joseph 2021); slower NIRS reoxygenation (McCully 1999).; against: Capillary finding from one lab; long COVID patients in the same study did not show it as strongly; extraction data largely single-centre.
Metabolism and blood omics: for: Lower NOTCH3, ITGAV, TGF-beta and vascular smooth-muscle protein sets.; against: Endothelial activation and platelet/microclot protein sets are lower, not higher.
Treatment trials and natural history: for: HBOT 40 sessions positive in one long-COVID RCT.; against: HBOT 10 sessions negative in HOT-LoCO.

Linked findings

During upright invasive exercise testing most ME/CFS patients have abnormally low right-atrial filling pressure (preload failure), and a high-flow subgroup shows poor systemic oxygen extraction. Several studies
A subset of patients with unexplained exertional intolerance cannot extract oxygen normally from blood during exercise, associated with hypocapnia and possible arteriovenous shunting. Early
Retinal imaging shows narrower arterioles, lower arteriolar-to-venular ratio, reduced flicker-induced venular dilation and lower capillary vessel density in post-COVID patients, most pronounced in those meeting ME/CFS criteria. Several studies
Sublingual video-microscopy shows reduced perfused capillary density and glycocalyx changes in long COVID, persisting many months after infection. Early
Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per fibre size in both ME/CFS and long COVID, unlike the pattern after 60 days of bed rest; thickened capillary basement membrane has been reported but is not yet peer-reviewed. Early
Pericytes and vascular smooth-muscle cells have not been examined directly in ME/CFS tissue, but blood proteins from mural cells and vessel matrix (NOTCH3, ITGAV, TGF-beta set, LUM) are lower across several ME/CFS cohorts and in long COVID. Early
VEGF-A has been reported lower, not higher, in ME/CFS plasma, and proteomic re-analysis finds no consistent change; long COVID cohorts more often report raised VEGF. Contested
Red blood cells from ME/CFS patients were stiffer in one microfluidic study, and real-time deformability cytometry found altered red-cell deformation persisting after COVID-19. Early
Some patients produce more lactate for a given workload, and this worsens on a repeat test, but plasma studies after one maximal test show a smaller total lactate rise that mainly reflects lower achieved workloads. Several studies
Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (preload failure) and, in a high-flow subgroup, poor systemic oxygen extraction by tissue. Several studies
31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower pH recovery after repeated exercise in a subset of patients, but other MRS studies found no consistent group abnormality. Contested
23Na-MRI in 6 women with ME/CFS and 6 controls found higher resting muscle sodium in all lower-leg compartments and a larger sodium rise after exercise, inversely related to grip strength. Early
In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed lower mitochondrial enzyme activity, more fibre necrosis, more T-cell infiltration and more amyloid-containing deposits than before exercise and than in controls. Early
ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy, lower mitochondrial capacity and fewer capillaries per fibre, a pattern that 60 days of strict bed rest in healthy people did not reproduce. Early
Muscle-derived proteins (myoglobin, troponins, CA3 and similar) are lower in ME/CFS blood at rest across three SomaScan cohorts and UK Biobank, not explained by step counts or walking pace. Several studies
Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison groups, and a 60-day bed-rest model) indicate that deconditioning alone does not explain the exercise and muscle abnormalities in ME/CFS. Several studies
Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in three cohorts from one group, and thalamic or regional lactate in two other groups. It is also raised in depression, so it is not specific. Several studies
Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS studies, but the results are inconsistent and often confounded by unmeasured orthostatic intolerance. Contested
Germain 2022 reported growing metabolic divergence after a 2-day CPET; within-person re-analysis finds the exercise response itself barely differs (blunted lactate and glucose rise aside). Contested
Collagen breakdown products (hydroxyproline, galactosyl-hydroxylysine and related) are higher in ME/CFS plasma (+0.52 SD, 4/4 Metabolon cohorts; hydroxyproline also up on 3 other platforms). Early
Muscle-derived intracellular proteins (myoglobin, troponins, CA3 and similar) are lower in resting ME/CFS blood across three SomaScan cohorts and UK Biobank Olink, and this is not explained by activity measures. Several studies
Hepatocyte enzymes (ADH4, GSTA1, AGXT, BCHE) are modestly higher, and vascular mural/matrix proteins (ITGAV, NOTCH3, LUM, THBS1, TGF-beta family) modestly lower, in more than one cohort. Early
Forty daily sessions of hyperbaric oxygen at 2.0 ATA improved cognition and symptoms versus sham in a 73-person long-COVID trial, but ten sessions at 2.4 bar showed no benefit in an 80-person sham-controlled trial. Contested

Low blood volume and preload failureToo little blood returns to the heart, especially upright and during exercise, reducing cardiac output and brain flow. active

Isotope and CO-rebreathing studies show reduced plasma and red-cell volume, and invasive CPET shows low right-heart filling pressure during exercise in many patients. Upright, this lowers stroke volume and brain blood flow (seen on tilt even without POTS). It could stem from small-fibre neuropathy, low ADH/renin response, or reduced red-cell production. It explains orthostatic intolerance well; whether it explains PEM is less clear.

Proponents: Systrom/Joseph (Harvard), van Campen/Visser (Netherlands), Levine/Fu (POTS), Rowe (Johns Hopkins)

Evidence for

Preload failure common in iCPET series (Joseph 2021, n=160)
Cerebral blood flow falls on tilt in ~90% of patients (van Campen 2020)
Single-dose pyridostigmine maintained cardiac output on repeat iCPET (Joseph 2022)
Patients rate fluids and compression among the most helpful (Eckey 2025)
Low reticulocytes with adequate iron in postviral fatigue (Solving ME/CFS re-analysis, FinnGen; unreplicated)

Evidence against

Fludrocortisone negative in RCT (Rowe 2001)
Ivabradine reduced HR without symptom benefit (RECOVER-AUTONOMIC 2026)
No sham-controlled trial of IV saline
Blood-volume hormones normal in proteomics (Solving ME/CFS re-analysis)

Predicts

Low measured blood volume
Low RAP on iCPET
CBF drop on tilt
Symptom relief with fluids, compression and pyridostigmine

Tests that would settle it

Sham-controlled IV saline crossover with function endpoints
Results of LIFT (pyridostigmine x LDN)
Blood volume measurement before/after illness onset in prospective post-COVID cohorts

Notes from each body system

Blood flow, vessels and blood: for: Red-cell volume deficits in 5 groups; small, under-filled heart explained by volume; preload failure on iCPET in most tested; CBF falls on tilt in about 90%; compression and vasoconstrictors reduce CBF falls.; against: Blood-volume category did not predict orthostatic abnormalities (Yamazaki 2026); iCPET cohorts pre-selected for low RAP; one Dutch clinic supplies most CBF data; volume-regulating hormones not consistently abnormal.
Muscle, exertion and energy: for: Low right atrial pressure during invasive CPET in 160 patients (Joseph 2021); low peak HR across CPET studies.; against: Preload failure explains low peak output but not why muscle pathology worsens a day later or why PEM is delayed.
Nervous system and brain: for: Orthostatic CBF fall even without tachycardia; narrow pulse pressure on lean test in recent-onset patients; preload failure in 92% of iCPET-tested ME/CFS (Novak 2026).; against: Yamazaki 2026: measured total and red-cell volume did not predict lean-test abnormalities; hypocapnia, not volume, was the most common abnormality.
Hormones: for: The renin-aldosterone paradox in POTS and ME, plus low ADH, would explain failure to restore low plasma volume.; against: Resting proteomic renin, ACE and BNP are normal. Fludrocortisone, which replaces aldosterone action, failed in RCTs.
Treatment trials and natural history: for: Pyridostigmine single-dose iCPET effect; patients rate fluids and compression highly.; against: Fludrocortisone RCT negative; ivabradine lowered HR without symptom benefit; no sham-controlled saline trial.

Linked findings

Red cell volume and total blood volume are below predicted in a large minority to about half of ME/CFS patients measured by isotope or CO-rebreathing methods. Replicated
The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not raised in ME/CFS or POTS despite low volume, and blood proteomics shows no consistent change in them. Several studies
Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more than normal on tilt, largely in proportion to low blood volume rather than to weak heart muscle. Replicated
During upright invasive exercise testing most ME/CFS patients have abnormally low right-atrial filling pressure (preload failure), and a high-flow subgroup shows poor systemic oxygen extraction. Several studies
Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, including in patients with normal heart rate and blood pressure responses. Several studies
Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exercise hyperventilation and dysfunctional breathing are common. Replicated
Raising blood pressure with phenylephrine during tilt blocked hyperventilation, restored cerebral blood velocity and improved working memory in ME/CFS, whereas CO2 or acetazolamide alone did not. Early
POTS is present in roughly a fifth of ME/CFS patients on formal testing (prevalence varies 6-43%), but cerebral hypoperfusion and preload failure are found in most patients regardless of whether they meet POTS heart-rate criteria. Replicated
On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled about 5 ml/kg/min below controls) with a low peak heart rate (chronotropic incompetence). Replicated
ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most clearly in workload and VO2 at the ventilatory threshold, but the largest independent replication found no day-2 change. Contested
Some patients produce more lactate for a given workload, and this worsens on a repeat test, but plasma studies after one maximal test show a smaller total lactate rise that mainly reflects lower achieved workloads. Several studies
Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (preload failure) and, in a high-flow subgroup, poor systemic oxygen extraction by tissue. Several studies
Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 26% vs 7% in controls), including in patients whose heart rate and blood pressure stay normal. Replicated
Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of the fall in brain blood flow, and correcting it with phenylephrine restored blood flow and working memory in one small trial. Several studies
POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%) with the setting, the age group and the test protocol. Contested
On the 10-minute NASA lean test, a subgroup of recent-onset patients shows a rising heart rate with an abnormally narrow pulse pressure. Autonomic specialists say the test has not been validated against tilt or active stand. Several studies
Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS studies, but the results are inconsistent and often confounded by unmeasured orthostatic intolerance. Contested
Meta-analysis shows moderate-to-large deficits in information-processing speed, attention, working memory and reaction time in ME/CFS, with intact vocabulary and reasoning. Performance worsens with upright posture and after exertion. Replicated
Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against sham control, and antibody removal does not predict who improves. Refuted
Fludrocortisone, a synthetic salt-retaining steroid, did not improve CFS with neurally mediated hypotension in a 100-person RCT, and failed again in combination with hydrocortisone. Refuted
After overnight fluid deprivation, plasma vasopressin was undetectable in 82% of 111 Swedish ME/CFS patients despite high plasma osmolality in 57%, suggesting partial central diabetes insipidus-like physiology. Early
Patients with POTS and with ME/CFS show normal or low renin and low aldosterone despite reduced blood volume (the 'renin-aldosterone paradox'). Several studies
Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4 in 60% of children with CFS vs 24% of controls; 50% in a Swedish severe cohort; EDS about 100x expected in Finnish registries). Several studies
A single 60 mg pyridostigmine dose raised peak oxygen uptake relative to placebo on a repeat invasive exercise test in 45 ME/CFS patients with preload failure, largely because the placebo group got worse; longer trials are negative or pending. Early
Ivabradine lowered standing heart rate but did not improve symptoms in 181 people with long-COVID POTS (RECOVER-AUTONOMIC), despite an earlier positive 22-person crossover in hyperadrenergic POTS. Refuted

Autonomic dysfunction and small-fibre neuropathyDamage to small autonomic and sensory nerves impairs vascular tone, heart rate control and pain processing. active

A sizeable minority of ME/CFS and POTS patients have reduced intraepidermal nerve fibre density on skin biopsy. Loss of sympathetic fibres to veins would cause venous pooling and preload failure; sensory fibre loss would add pain. The neuropathy may be immune-mediated after infection. It is a likely contributor in a subgroup rather than a universal mechanism.

Proponents: Oaklander (MGH), Systrom, Novak, Blitshteyn

Evidence for

Small-fibre neuropathy frequent in biopsied iCPET patients (Joseph 2021)
POTS and low HRV common in ME/CFS

Evidence against

IVIG negative vs albumin in autoimmune POTS (Vernino 2024)
Prevalence estimates vary widely between centres
Ivabradine trial negative for symptoms

Predicts

Low IENFD on skin biopsy
Abnormal QSART/sudomotor tests
Response to IVIG in immune-mediated subgroup

Tests that would settle it

Population-representative skin biopsy study with blinded counting
IVIG trial restricted to biopsy-proven SFN with function endpoints

Notes from each body system

Blood flow, vessels and blood: for: SFN in about a third to half of iCPET/autonomic cohorts; shunting proposed to explain poor extraction; hyperadrenergic POTS profile.; against: Denervation severity did not correlate with exertional measures (Joseph 2021).
Muscle, exertion and energy: for: 31% SFN and shunting physiology in iCPET (Joseph 2021); loss of normal pH-efflux coupling linked to autonomic function (Jones 2010); adrenergic receptor mRNA changes after exercise (Light 2012).; against: Denervation severity did not correlate with exertional measures (Joseph 2021).
Nervous system and brain: for: Reduced vagal HRV (meta-analysis), orthostatic CBF fall and hypocapnia across labs, SFN in 31-53% of referral cohorts, autonomic-fibre loss in long COVID, POTS in a minority.; against: SFN not found in the controlled NIH cohort; SFN severity does not correlate with exertional limits or symptom scores; HRV changes are non-specific and confounded by deconditioning.

Linked findings

Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more than normal on tilt, largely in proportion to low blood volume rather than to weak heart muscle. Replicated
During upright invasive exercise testing most ME/CFS patients have abnormally low right-atrial filling pressure (preload failure), and a high-flow subgroup shows poor systemic oxygen extraction. Several studies
A subset of patients with unexplained exertional intolerance cannot extract oxygen normally from blood during exercise, associated with hypocapnia and possible arteriovenous shunting. Early
Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, including in patients with normal heart rate and blood pressure responses. Several studies
Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exercise hyperventilation and dysfunctional breathing are common. Replicated
Raising blood pressure with phenylephrine during tilt blocked hyperventilation, restored cerebral blood velocity and improved working memory in ME/CFS, whereas CO2 or acetazolamide alone did not. Early
POTS is present in roughly a fifth of ME/CFS patients on formal testing (prevalence varies 6-43%), but cerebral hypoperfusion and preload failure are found in most patients regardless of whether they meet POTS heart-rate criteria. Replicated
On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled about 5 ml/kg/min below controls) with a low peak heart rate (chronotropic incompetence). Replicated
ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most clearly in workload and VO2 at the ventilatory threshold, but the largest independent replication found no day-2 change. Contested
Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (preload failure) and, in a high-flow subgroup, poor systemic oxygen extraction by tissue. Several studies
31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower pH recovery after repeated exercise in a subset of patients, but other MRS studies found no consistent group abnormality. Contested
After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), adrenergic and immune genes rose for up to 48 h in most CFS patients but not in controls or MS patients. Early
Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 26% vs 7% in controls), including in patients whose heart rate and blood pressure stay normal. Replicated
Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of the fall in brain blood flow, and correcting it with phenylephrine restored blood flow and working memory in one small trial. Several studies
POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%) with the setting, the age group and the test protocol. Contested
Resting heart rate is higher and high-frequency heart-rate variability lower in ME/CFS. This fits reduced vagal (parasympathetic) tone. Replicated
On the 10-minute NASA lean test, a subgroup of recent-onset patients shows a rising heart rate with an abnormally narrow pulse pressure. Autonomic specialists say the test has not been validated against tilt or active stand. Several studies
Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients from specialist referral cohorts, but the controlled NIH study found no difference from healthy volunteers. Contested
Long COVID patients referred for neuropathy evaluation often have small-fibre neuropathy on skin biopsy, including loss of autonomic fibres to sweat glands and hair muscles. Several studies
Brainstem differences (smaller volume, altered signal or connectivity) are among the more often repeated imaging findings in ME/CFS, though the methods differ between studies. Several studies
In the NIH cohort, CSF levels of the catechol metabolites DOPA, DOPAC and DHPG were lower in post-infectious ME/CFS. Noradrenaline and dopamine did not differ. Early
Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep onset and lower sleep efficiency. Standard polysomnography usually finds no clinical disorder that explains non-restorative sleep. Several studies
Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been reported repeatedly, including a two-site study (36 vs 42), but only ever by one research group. Early
Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30% of patients by ELISA, is not reproduced by unbiased high-throughput antigen profiling. Refuted
IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve damage to mice in three independent 2026 studies; no equivalent published ME/CFS transfer experiment exists. Several studies

Viral persistenceVirus or viral antigen persists in tissue reservoirs and drives ongoing immune activation. weakened

After SARS-CoV-2 and possibly other infections, viral RNA or protein may persist in gut, lymphoid or nervous tissue. Persistent antigen would keep the immune system activated and could explain chronic symptoms. Evidence is strongest in long COVID, where viral proteins are detected months later. Whether it applies to classic ME/CFS is unknown.

Proponents: Proal and VanElzakker (PolyBio), Peluso and Deeks (UCSF), Iwasaki (Yale)

Evidence for

SARS-CoV-2 antigen detected in tissue and plasma months after infection (reviewed Proal 2023)
ME/CFS incidence ~5x higher after SARS-CoV-2 (Vernon 2025)

Evidence against

Three nirmatrelvir-ritonavir RCTs null (RECOVER-VITAL 2026, STOP-PASC 2024, PAX LC 2025)
Temelimab (HERV-W) null
Antigen detected also in recovered people in some studies

Predicts

Viral RNA/protein in tissue biopsies
Antivirals or monoclonals improving symptoms
Antigen levels tracking symptoms

Tests that would settle it

Trials of longer or combination antivirals and monoclonals with tissue antigen as a selection marker
Tissue antigen load vs symptom severity in matched recovered controls

Notes from each body system

Muscle, exertion and energy: for: Enterovirus RNA in muscle in older studies (Bowles 1993).; against: Not reproduced with modern methods; Appelman 2024 did not report persistent virus as the driver of post-exertion pathology.
Immune system: for: Chronic antigenic stimulation interpretation of the naive/switched-memory B-cell shift (Walitt 2024); exhaustion programmes in CD8+ T cells; early-immunosenescence phenotype in mild/moderate patients 'indicative of persistent viral infection' (Lee 2025); exaggerated response to microbial antigens.; against: No persisting pathogen has been identified in ME/CFS blood; TCR and BCR repertoires show no clonal expansion or antigen-driven maturation; herpesvirus seroprevalence is normal; immune cell types are not genetically enriched.
Infection and viruses: for: SARS-CoV-2 antigen/RNA persists in plasma and tissues of a minority for months to years, and tissue RNA associated with LC (Zuo 2024 OR 5.17). Enterovirus protein in stomach biopsies (Chia 2008). Early nirmatrelvir may prevent LC (small RCT).; against: Three RCTs of nirmatrelvir in established LC (about 1,200 people) were null, and spike levels did not fall with treatment. Enterovirus findings are unreplicated with blinded methods. XMRV was contamination. The Giardia and Q fever triggers and failed doxycycline in QFS fit a 'hit-and-run' rather than persistent infection.
Genetics: for: Infection-response genes at 3 loci, a larger OLFM4 effect in infection onset, and suggestive EBV-load MR (MVP p=0.012).; against: Infection and non-infection onset are genetically similar, EBV MR control viruses align too, and immune cell types are not enriched.
Treatment trials and natural history: for: Post-COVID ME/CFS incidence HR 4.93 (Vernon 2025); rintatolimod modest effect.; against: Three nirmatrelvir RCTs null; temelimab null.

Linked findings

Active long COVID shows persistent complement dysregulation (terminal complex, alternative and classical pathway activation) with thromboinflammation markers, replicated across several cohorts. Replicated
Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without the fibre atrophy and type 2 shift expected from inactivity, with abnormalities confined to subgroups. Contested
In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed lower mitochondrial enzyme activity, more fibre necrosis, more T-cell infiltration and more amyloid-containing deposits than before exercise and than in controls. Early
Long COVID patients referred for neuropathy evaluation often have small-fibre neuropathy on skin biopsy, including loss of autonomic fibres to sweat glands and hair muscles. Several studies
In long COVID, one [11C]PBR28 PET study found widespread raised TSPO signal that correlated with vascular markers, and a [11C]PK11195 study found no group difference. Contested
Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older literature (Hedges g ~0.96), but the largest standardised multi-site study and two deep-phenotyping studies found no difference. Contested
CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exhaustion, with supporting flow cytometry and an independent cell-free RNA signature. Several studies
T-cell receptor repertoires do not distinguish ME/CFS from controls, arguing against a dominant shared antigen-driven clonal expansion. Refuted
The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-infectious ME/CFS, consistent with chronic antigenic stimulation; supporting signals exist but are weak. Early
Reported skewing of the B-cell receptor repertoire (IGHV3-30/3-33 overuse) partially replicates in mild/moderate patients but shows no clonality, somatic-hypermutation or autoimmunity signature. Contested
A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficiency, with IgG3 and IgG4 the commonest phenotypes, and registry data show strong enrichment of coded antibody deficiency. Several studies
In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after SARS-CoV-2 infection vs 0.93 in uninfected participants (HR 4.93), and 4.5% of infected participants met criteria. Replicated
Detectable cell-free EBV DNA in plasma has been reported more often in ME/CFS (24% vs 4% in one Bulgarian study), but EBV DNA in blood cells is equally common in patients and controls. Early
Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20% of controls, with EV RNA in 37% vs 5%, but this has not been independently replicated with blinded methods. Contested
SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVID than in pre-pandemic samples (excess +11% at 3-6 months, +5.4% at 10-14 months), with more antigenemia after more severe acute illness. Several studies
SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a minority of people, and in a Beijing tissue study, viral RNA in tissue was associated with long COVID symptoms (OR 5.17). Several studies
Three randomised placebo-controlled trials of 15-25 days of nirmatrelvir-ritonavir in established long COVID (STOP-PASC n=155, PAX LC n=100, RECOVER-VITAL n=959; about 1,214 people in total) found no benefit on primary or secondary outcomes. Replicated
Treating acute COVID-19 with nirmatrelvir may reduce later long COVID: a small terminated Norwegian RCT found 26% vs 43% at 3 months (RR 0.60), echoing observational VA data, but larger prospective evidence is lacking. Early
The 2009 report of the mouse retrovirus XMRV in 67% of CFS patients was a laboratory contamination artefact: blinded multi-lab and original-investigator studies found no XMRV/pMLV, and the Science paper was retracted. Refuted
In Finnish registry data, ME/CFS (G93.3) is enriched 3-6-fold for prior viral infections (HSV, EBV, influenza, COVID, viral CNS infection) but not for bacterial pneumonia or erysipelas. Early
Quadram Institute studies of severe ME/CFS with household-matched controls report altered gut phage communities and reduced serum IgG reactivity to gut microbes, in very small samples. Early
In long COVID, the gut microbiome at hospital admission predicted persistent symptoms, and at 6 months patients had more R. gnavus and fewer butyrate producers, mirroring ME/CFS. Several studies
DecodeME (15,579 PEM-defined cases vs 259,909 controls) identified 8 genome-wide significant loci, each with a small effect (OR about 1.07-1.10), and SNP heritability of about 8-10%. Several studies
Three DecodeME loci lie near infection-response genes (OLFM4, RABGAP1L, BTN2A2/TRIM38), and the OLFM4 effect is larger in infection-onset cases, but infection- and non-infection-onset cases are genetically highly correlated. Early
Rintatolimod (Ampligen) gave a modest placebo-adjusted gain in treadmill time in a 234-person phase III trial, but the result was borderline and the FDA declined approval. Contested

Herpesvirus reactivation (EBV, HHV-6)Latent herpesviruses reactivate after a trigger and drive immune and neurological dysfunction. active

EBV and HHV-6 are carried latently by most adults. In ME/CFS, abortive reactivation or high antibody responses are reported, and EBV is a common trigger of onset. Reactivated virus or its proteins (e.g. dUTPase) could activate innate immunity. Antivirals have shown modest benefit in small selected groups.

Proponents: Montoya, Lerner, Ariza, Prusty, Peluso (EBV in long COVID)

Evidence for

Glandular fever is a strong trigger of ME/CFS in prospective cohorts
Valganciclovir RCT positive on secondary scales (Montoya 2013, n=30)
Genetically higher EBV DNA load suggestive risk factor (OR 1.59, MVP; Solving ME/CFS re-analysis)
Elevated EBV gp42 reactivity in antigen profiling (Germain 2025)

Evidence against

STOP-PASC found no EBV reactivation differences (Maestri 2025 preprint)
Antibody titres poor marker of active replication
No replicated antiviral RCT

Predicts

Viral DNA or early-gene antibodies elevated
Antivirals help selected patients
Genetic predisposition to EBV control affects risk

Tests that would settle it

Antiviral RCT selected by viral PCR/early antigen, not IgG
Prospective serial viral load from acute infection through ME/CFS onset

Notes from each body system

Immune system: for: Higher anti-EBV antibody responses in long COVID (Klein 2023); incidental elevated EBV gp42 reactivity in the otherwise negative antigen-profiling study; IgG to arginine-rich EBNA4/EBNA6 sequences and homologous human peptides (Hoheisel 2025); genetically higher EBV DNA load suggestive in MVP (OR 1.59); exploratory re-analysis signals - fewer CMV-driven adaptive NK expansions in patients (1/30 vs 9/28, post-hoc p=0.005), more KLRG1+CX3CR1+ effector CD8 (d +0.60), higher EBV-specific TCR-beta share in severe patients' CD8 (0.88% vs 0.45%, p=0.06).; against: Cliff 2019 found no difference in seroprevalence for six human herpesviruses in 251 patients; the re-analysis signals are exploratory and do not survive multiple-testing correction (q 0.28); valganciclovir evidence rests on one 30-patient trial selected on antibody titres, which are a poor marker of active infection.
Infection and viruses: for: Lytic-marker serology (EA-D) associates with LC fatigue (Peluso 2023). dUTPase antibodies higher in ME/CFS (one lab, repeated). HHV-6 miRNA in ME/CFS brain (n=3). Genetic EBV-load MR suggestive (MVP OR 1.59). Plasma EBV DNA more frequent in one study. Herpes titres track complement activation in LC.; against: Large serology studies (UK Biobank, Swedish cohorts) show no difference. PBMC viral DNA is equally common. The valganciclovir RCT missed its primary outcome and was never replicated. MR signal is not EBV-specific on negative controls. Most positive findings come from single labs.
Treatment trials and natural history: for: Valganciclovir small RCT positive on secondary endpoints (Montoya 2013).; against: Unreplicated; STOP-PASC found no EBV reactivation difference.

Linked findings

Active long COVID shows persistent complement dysregulation (terminal complex, alternative and classical pathway activation) with thromboinflammation markers, replicated across several cohorts. Replicated
Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older literature (Hedges g ~0.96), but the largest standardised multi-site study and two deep-phenotyping studies found no difference. Contested
CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exhaustion, with supporting flow cytometry and an independent cell-free RNA signature. Several studies
Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to non-SARS-CoV-2 viruses especially EBV, and low cortisol, giving a reference phenotype for post-infectious illness. Several studies
Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type, implying most immune findings are downstream consequences rather than inherited causes. Replicated
After infectious mononucleosis in adolescents, 13%, 7% and 4% met paediatric CFS criteria at 6, 12 and 24 months; all 13 cases still ill at 24 months were female. Replicated
In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and overall IgG reactivity do not differ between ME/CFS patients and controls. Several studies
Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (absent or low EBNA-1/VCA antibody-secreting memory B cells in up to 76% of patients, fewer polyfunctional EBNA-1-specific T cells), while CMV responses were normal. Early
Detectable cell-free EBV DNA in plasma has been reported more often in ME/CFS (24% vs 4% in one Bulgarian study), but EBV DNA in blood cells is equally common in patients and controls. Early
Genetic instruments for higher EBV DNA load in blood (Kamitaki 2026) are associated with ME/CFS risk in the Million Veteran Program (OR 1.59, p=0.012) but not significantly in DecodeME (OR 1.13, p=0.25). Early
Antibodies against herpesvirus dUTPases (EBV, HHV-6, VZV), proposed as a marker of lytic or abortive-lytic reactivation, are higher in ME/CFS in studies from one lab, including a 2026 longitudinal study (873 serum samples, 40 patients vs 16 controls) where levels tracked fatigue severity. Early
Serologic evidence of recent EBV reactivation (EA-D IgG) is associated with long COVID fatigue (OR 2.12) in a 280-person cohort, consistent with an earlier small report. Several studies
In active long COVID, classical complement pathway activation tracked raised antibody titres against several herpesviruses, alongside markers of haemolysis, platelet activation and monocyte-platelet aggregates. Early
HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord tissue from 3 ME/CFS post-mortem donors but not in 24 controls (Kasimir 2022). Early
Inherited chromosomally integrated HHV-6 (iciHHV-6, ~1% of people) has been reported at roughly twice the population rate in a neurological/fatigue clinic cohort, but this has not been confirmed in a large ME/CFS sample. Early
In a 30-patient RCT of 6 months of valganciclovir in CFS patients selected for high HHV-6 and EBV IgG, the primary MFI-20 outcome was not significant, though some secondary fatigue and cognitive trajectories favoured the drug. Early
Human endogenous retrovirus (HERV) expression in blood cells differs between ME/CFS, fibromyalgia and controls in small studies (HERV-K raised in moderate ME/CFS; HERV microarray profiles clustered n=43 participants perfectly by diagnosis). Early
In Finnish registry data, ME/CFS (G93.3) is enriched 3-6-fold for prior viral infections (HSV, EBV, influenza, COVID, viral CNS infection) but not for bacterial pneumonia or erysipelas. Early
Blood transcriptome and single-cell RNA studies report monocyte and platelet abnormalities, but no gene or cell subtype difference replicates across independent cohorts. Contested
Three DecodeME loci lie near infection-response genes (OLFM4, RABGAP1L, BTN2A2/TRIM38), and the OLFM4 effect is larger in infection-onset cases, but infection- and non-infection-onset cases are genetically highly correlated. Early
Reported HLA associations are inconsistent: Lande 2020's HLA-C*07:04 and DQB1*03:03 risk alleles are not among DecodeME's hits, and DecodeME's protective HLA-DQA1*05:01 signal is barely visible at SNP level. Contested
Specific inhibitory KIR alleles (e.g. KIR3DL1*020, KIR3DL2*009) were more or less common in 418 Norwegian patients than 473 controls, though KIR gene content did not differ. Early
Valganciclovir in 30 patients selected for high HHV-6 and EBV antibody levels improved some secondary fatigue and cognition scores in a small RCT; it has not been replicated. Early

Autoimmunity (functional autoantibodies)Post-infectious autoantibodies, for example against adrenergic or muscarinic receptors, cause vascular and autonomic dysfunction. weakened

A subset of patients may develop antibodies that disturb receptor signalling or nerve function after infection, similar to other post-infectious autoimmune conditions. Female predominance, infection onset and immune genetics fit. Antibody removal or B-/plasma-cell depletion should then help. Recent controlled trials have narrowed the GPCR version, while plasma-cell depletion and IgG transfer keep a broader version alive.

Proponents: Scheibenbogen (Charité), Fluge and Mella (Bergen), Wallukat (GPCR antibodies), Iwasaki (IgG transfer)

Evidence for

Daratumumab 6/10 sustained responders with IgG fall (Fluge 2025, open-label)
Long-COVID IgG transfers pain hypersensitivity to mice (Chen 2026; Mignolet 2026)
Cyclophosphamide 55% open-label response with HLA association (Rekeland 2020)

Evidence against

Rituximab phase III negative (Fluge 2019)
Immunoadsorption no better than sham (IAMPOCO, IA-PACS-CFS 2026)
No GPCR autoantibody elevation on REAP/Luminex, q>0.68 (Germain 2025)
BC007 company trial negative (2024)

Predicts

Elevated specific autoantibodies
Benefit from antibody removal or plasma-cell depletion
Symptom transfer with patient IgG

Tests that would settle it

Randomised placebo-controlled daratumumab trial
IgG transfer from ME/CFS (non-COVID) patients with blinded readouts
Pre-specified antibody-selected sham-controlled trial

Notes from each body system

Immune system: for: Infection-triggered onset; female predominance; open-label responses to B-cell depletion (early phase II), cyclophosphamide and especially daratumumab (6/10, 5 sustained, serum IgG fell more in responders); patient IgG does things control and MS IgG do not in endothelial cells (Liu/Prusty 2026, 39 ME vs 20 MS vs 41 HC); long COVID IgG transfers pain, fatigue-like behaviour and small-fibre damage to mice in three independent 2026 studies; IgG subclass abnormalities common.; against: Rituximab failed in a 151-patient RCT (26% vs 35% placebo); immunoadsorption removed GPCR antibodies without clinical benefit against sham (IAMPOCO n=40, all endpoints null; IA-PACS-CFS n=65 failed); BC007's larger phase II failed; unbiased profiling of 7,542 antibody-antigen interactions in 172 people found no differential reactivity (no q below 0.68), including beta-adrenergic receptors; BCR repertoires show no clonality, no somatic-hypermutation shift and no autoimmunity signature; no ME/CFS-specific autoantigen has been identified; common-variant genetics shows no immune cell enrichment and HLA associations do not fit an autoimmune pattern.
Genetics: for: HLA associations reported in Norway (Lande 2020) and DQA1*05:01 in DecodeME.; against: The HLA findings disagree with each other and the DecodeME signal is weak at SNP level. Common-variant genetics shows no immune-cell enrichment.
Treatment trials and natural history: for: Daratumumab 6/10 sustained responders with IgG fall (Fluge 2025); cyclophosphamide 55% open-label response; long-COVID IgG transfers pain-like symptoms to mice (Chen 2026).; against: Rituximab phase III negative; immunoadsorption failed against sham twice; BC007 company trial negative; GPCR autoantibodies not elevated on REAP/Luminex (Germain 2025).

Linked findings

Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia are reduced in ME/CFS in two independent Norwegian cohorts and by peripheral arterial tonometry in Berlin. Replicated
Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitric oxide production, more mitochondrial fragmentation, and anti-endothelial antibody binding. Early
The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pre-registered replication of 67 patients vs 53 controls, though a small 3D engineered-muscle study reported weaker contraction with patient serum. Refuted
POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%) with the setting, the age group and the test protocol. Contested
Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients from specialist referral cohorts, but the controlled NIH study found no difference from healthy volunteers. Contested
Long COVID patients referred for neuropathy evaluation often have small-fibre neuropathy on skin biopsy, including loss of autonomic fibres to sweat glands and hair muscles. Several studies
T-cell receptor repertoires do not distinguish ME/CFS from controls, arguing against a dominant shared antigen-driven clonal expansion. Refuted
The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-infectious ME/CFS, consistent with chronic antigenic stimulation; supporting signals exist but are weak. Early
Reported skewing of the B-cell receptor repertoire (IGHV3-30/3-33 overuse) partially replicates in mild/moderate patients but shows no clonality, somatic-hypermutation or autoimmunity signature. Contested
B-cell depletion with rituximab failed in a 151-patient placebo-controlled trial, after positive small phase II results. Refuted
Plasma-cell depletion with daratumumab produced marked, in some cases sustained, improvement in 6 of 10 patients in an open-label pilot, with low baseline NK count predicting non-response. Early
Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30% of patients by ELISA, is not reproduced by unbiased high-throughput antigen profiling. Refuted
Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against sham control, and antibody removal does not predict who improves. Refuted
BC007 (rovunaptabin), an aptamer that neutralises GPCR autoantibodies, showed secondary-endpoint fatigue improvement in a 30-patient exploratory trial but failed in the larger company-run phase II. Contested
IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve damage to mice in three independent 2026 studies; no equivalent published ME/CFS transfer experiment exists. Several studies
Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human endothelial cells, an effect absent with multiple-sclerosis and healthy IgG and lost on Fab/Fc cleavage. Early
A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficiency, with IgG3 and IgG4 the commonest phenotypes, and registry data show strong enrichment of coded antibody deficiency. Several studies
Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type, implying most immune findings are downstream consequences rather than inherited causes. Replicated
In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after SARS-CoV-2 infection vs 0.93 in uninfected participants (HR 4.93), and 4.5% of infected participants met criteria. Replicated
In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and overall IgG reactivity do not differ between ME/CFS patients and controls. Several studies
Genetic instruments for higher EBV DNA load in blood (Kamitaki 2026) are associated with ME/CFS risk in the Million Veteran Program (OR 1.59, p=0.012) but not significantly in DecodeME (OR 1.13, p=0.25). Early
Human endogenous retrovirus (HERV) expression in blood cells differs between ME/CFS, fibromyalgia and controls in small studies (HERV-K raised in moderate ME/CFS; HERV microarray profiles clustered n=43 participants perfectly by diagnosis). Early
ME/CFS and long COVID are more common in women (ME/CFS prevalence about 1.4% vs 0.9%; DecodeME 82% female), but genetic risk variants are not sex-specific. Replicated
Reported HLA associations are inconsistent: Lande 2020's HLA-C*07:04 and DQB1*03:03 risk alleles are not among DecodeME's hits, and DecodeME's protective HLA-DQA1*05:01 signal is barely visible at SNP level. Contested
B-cell depletion with rituximab did not improve fatigue or function versus placebo in a 151-person phase III trial (RituxME). Refuted

Immune (T-cell) exhaustionChronic antigen stimulation leaves CD8 T cells and NK cells exhausted and less able to clear infection. active

Exhausted T cells express inhibitory receptors and lose function after prolonged stimulation. In ME/CFS, single-cell and epigenetic data suggest CD8 T cells are primed toward exhaustion, and NK cytotoxicity is reduced. This fits with a persistent antigen or reactivating virus. It may be consequence rather than cause.

Proponents: Hanson and Grimson (Cornell), Walitt/NIH intramural, Unutmaz (JAX)

Evidence for

CD8 T cells transcriptionally primed toward exhaustion (Iu 2024, PNAS)
T-cell exhaustion signals in NIH deep phenotyping (Walitt 2024)
Low NK count predicted non-response to daratumumab (Fluge 2025)

Evidence against

Immune cell, methylation and repertoire data give few reproducible signals across cohorts (Solving ME/CFS re-analysis)
NK cytotoxicity findings vary by assay

Predicts

Exhaustion markers (PD-1, TOX) on T cells
Reduced NK cytotoxicity
Changes after exercise provocation

Tests that would settle it

Replication of exhaustion signatures in independent cohorts
Longitudinal tracking from infection to ME/CFS onset

Notes from each body system

Immune system: for: CD8+ effector-memory exhaustion transcription factors, chromatin and metabolic reprogramming (Iu 2024); independent cfRNA exhaustion and cytokine-signalling signatures (Gardella 2025); reduced CD8+ mitochondrial membrane potential and glycolysis (Mandarano 2020); exhausted virus-specific CD8+ T cells in long COVID (Yin 2024); early-immunosenescence CD28-CD57- cells in mild/moderate patients (Lee 2025).; against: Re-analysis of the same public scRNA data found exhausted-CD8 gating null and no differentially expressed genes at FDR in any of 15 cell types; TCR repertoires show no clonal expansion, which is hard to square with chronic antigen-driven exhaustion; no immune cell type is genetically enriched; no antigen has been identified.
Infection and viruses: for: Deficient EBV-specific memory B/T responses (Loebel 2014). Exhausted, terminally differentiated virus-specific CD8 T cells in LC (Ma 2026).; against: Single-cell re-analysis found no exhausted CD8 signature in ME/CFS, and EBV-specific clonotypes trended higher, not lower, in severe cases.
Metabolism and blood omics: for: cfRNA T-cell exhaustion signature (Gardella 2025); Germain 2025 suppressed T/B signalling after exertion; plasma-cell RNA dip.; against: No replicated single-cell or bulk transcript difference.

Linked findings

Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal respiration, T-cell glycolysis), higher (lymphoblast maximal respiration with Complex V inefficiency) or normal, depending on lab and cell type. Contested
Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the largest patient-specific changes in the 24 h recovery phase, with suppressed T/B-cell and IL-17 signalling and raised glycolysis proteins. Contested
Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older literature (Hedges g ~0.96), but the largest standardised multi-site study and two deep-phenotyping studies found no difference. Contested
Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been reported repeatedly, including a two-site study (36 vs 42), but only ever by one research group. Early
CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exhaustion, with supporting flow cytometry and an independent cell-free RNA signature. Several studies
T-cell abnormalities differ by severity: mild/moderate patients show cytotoxic/immunosenescence features, severe patients show activated lymphocytes and higher stimulated pro-inflammatory cytokines. Early
Increased mucosal-associated invariant T (MAIT) cell proportions and a shift in CD8+ memory subsets were found in the largest single ME/CFS flow-cytometry cohort, but the finding has not been reproduced in independent data. Contested
The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-infectious ME/CFS, consistent with chronic antigenic stimulation; supporting signals exist but are weak. Early
Plasma cytokine abnormalities depend strongly on illness duration: a pro- and anti-inflammatory activation pattern early (under 3 years) that is absent in long-established illness. Several studies
Plasma cell-free RNA carries an immune-weighted ME/CFS signature that classifies cases with held-out AUC 0.81, with raised plasmacytoid-dendritic, monocyte and T-cell-derived transcripts. Early
Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to non-SARS-CoV-2 viruses especially EBV, and low cortisol, giving a reference phenotype for post-infectious illness. Several studies
Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type, implying most immune findings are downstream consequences rather than inherited causes. Replicated
After infectious mononucleosis in adolescents, 13%, 7% and 4% met paediatric CFS criteria at 6, 12 and 24 months; all 13 cases still ill at 24 months were female. Replicated
Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (absent or low EBNA-1/VCA antibody-secreting memory B cells in up to 76% of patients, fewer polyfunctional EBNA-1-specific T cells), while CMV responses were normal. Early
Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPET, but a within-person re-analysis of the same data finds no protein whose exercise response differs at FDR. Contested
Blood transcriptome and single-cell RNA studies report monocyte and platelet abnormalities, but no gene or cell subtype difference replicates across independent cohorts. Contested
Reported genome-wide blood DNA methylation differences have not replicated across cohorts, at least one large set is explained by cell-composition differences, and epigenetic clocks are not accelerated. Contested
Plasma cell-free RNA classified ME/CFS vs sedentary controls with test-set AUC 0.81 and suggested platelet, pDC and monocyte-derived differences; not yet externally validated. Early
Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-validation but drop to AUC ~0.6-0.77 on external cohorts, so none is usable for diagnosis. Contested
Quadram Institute studies of severe ME/CFS with household-matched controls report altered gut phage communities and reduced serum IgG reactivity to gut microbes, in very small samples. Early
Specific inhibitory KIR alleles (e.g. KIR3DL1*020, KIR3DL2*009) were more or less common in 418 Norwegian patients than 473 controls, though KIR gene content did not differ. Early
Whole-genome sequencing of 464 cases and 611 controls with a deep-learning rare-variant model (HEAL2) proposes 115 loss-of-function-intolerant risk genes, partly overlapping the common-variant synaptic signal. Early
TRPM3 ion channel dysfunction in NK cells and TRPM3 SNPs are reported repeatedly by one research group, but the SNP studies are tiny and no TRPM3 association appears in GWAS. Contested
In an open-label phase II study, 55% of 40 patients responded to six IV cyclophosphamide infusions, and many responders stayed improved at 6 years, but there was no control arm. Early
Low-dose rapamycin (6 mg weekly) was followed by self-reported improvement in about three-quarters of completers in an uncontrolled decentralised study; randomised trials are pending. Early

Innate immune hyperactivationOngoing activation of monocytes, interferon and cytokine pathways sustains symptoms. active

Cytokine profiles differ early in ME/CFS, with an activated pattern in the first three years that later fades. Persistent interferon or inflammasome activity could cause sickness symptoms and metabolic shifts. Effects in blood are inconsistent across studies.

Proponents: Hornig and Lipkin (Columbia), Klimas, Iwasaki

Evidence for

Distinct cytokine signatures early in illness (Hornig 2015, n=298)
Complement and myeloid activation in long COVID (Cervia-Hasler 2024)

Evidence against

Cytokine findings poorly replicated across ME/CFS cohorts
Methylprednisolone trial in long COVID stopped early without benefit (PoCoVIT 2026)
Prednisolone/losartan negative (Myoflame-19 2026)

Predicts

Raised cytokines early in illness
Interferon signature
Benefit from anti-inflammatory treatment

Tests that would settle it

Standardised longitudinal cytokine/interferon panels from onset
Targeted anti-cytokine RCT in a biomarker-defined subgroup

Notes from each body system

Muscle, exertion and energy: for: T-cell and macrophage infiltration in long COVID muscle after exertion (Appelman 2024); exaggerated innate responses and complement after exercise (Che 2025); complement up in male EVs (Glass 2025).; against: Germain 2025 found suppressed rather than activated lymphocyte signalling in recovery; single-cell RNA after exercise showed minimal immune changes (Vu 2024).
Immune system: for: Exaggerated whole-blood responses to microbial antigens, worse 24 h after exercise and correlated with symptom intensity (Che/Lipkin 2025, 56 vs 52); Q fever fatigue syndrome PBMCs over-produce IL-6, TNF and IL-1beta on antigen stimulation; monocyte dysregulation in scRNA (Vu 2024); meta-analytic small elevations of TNF, IL-2, IL-4, CRP; 17 cytokines rising with severity (Montoya 2017); UK Biobank inflammation and liver/insulin markers shifted in both sexes.; against: IL-1 blockade with anakinra gave no clinical benefit and changed no cytokine but IL-1Ra; the best-replicated single cytokine (TGF-beta) is a platelet/centrifugation artefact; re-analysis of five blood RNA cohorts found no gene set passing FDR and interferon only nominally up (meta p=0.12); monocyte scRNA findings do not survive independent re-analysis of the same data; no immune cell type is genetically enriched.
Infection and viruses: for: Severity of acute illness, not pathogen type, predicts post-infective fatigue (Dubbo). The stereotyped syndrome after diverse pathogens suggests a common host response.; against: Few studies measure the acute immune response prospectively, and baseline cytokines have predicted outcome inconsistently.
Gut and microbiome: for: Raised LPS/LBP/sCD14 in some studies. In BioMapAI, the microbiome links to MAIT/γδT inflammatory cells.; against: Re-analysis found 0/622 immune-module associations at q<0.1. The Quadram feasibility study found reduced anti-microbial IgG.

Linked findings

Retinal imaging shows narrower arterioles, lower arteriolar-to-venular ratio, reduced flicker-induced venular dilation and lower capillary vessel density in post-COVID patients, most pronounced in those meeting ME/CFS criteria. Several studies
Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoactive proteins in ME/CFS; most named candidates are null and the endothelial-activation set trends down. Contested
Active long COVID shows persistent complement dysregulation (terminal complex, alternative and classical pathway activation) with thromboinflammation markers, replicated across several cohorts. Replicated
In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetically defined alternative-pathway subgroup, and proteomic signals that point in different directions across cohorts. Contested
In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed lower mitochondrial enzyme activity, more fibre necrosis, more T-cell infiltration and more amyloid-containing deposits than before exercise and than in controls. Early
WASF3 protein is raised in ME/CFS muscle, and in cells and mice it disrupts respiratory supercomplex assembly and lowers exercise capacity via ER stress. Early
After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), adrenergic and immune genes rose for up to 48 h in most CFS patients but not in controls or MS patients. Early
Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but once each person is compared with themselves the exercise response differs little from controls. Contested
Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the largest patient-specific changes in the 24 h recovery phase, with suppressed T/B-cell and IL-17 signalling and raised glycolysis proteins. Contested
CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct immunotypes), but no CSF protein signature has replicated, and ME/CFS and fibromyalgia CSF proteomes were indistinguishable. Contested
T-cell abnormalities differ by severity: mild/moderate patients show cytotoxic/immunosenescence features, severe patients show activated lymphocytes and higher stimulated pro-inflammatory cytokines. Early
Plasma cytokine abnormalities depend strongly on illness duration: a pro- and anti-inflammatory activation pattern early (under 3 years) that is absent in long-established illness. Several studies
In the largest single cytokine study, 17 cytokines rose in step with disease severity although only TGF-beta (up) and resistin (down) differed between all patients and controls. Several studies
Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF-beta and CRP, with 12 other proteins showing no difference. Several studies
The most-replicated single cytokine finding, elevated TGF-beta, is largely a blood-handling artefact driven by platelet activation during centrifugation. Refuted
Blocking IL-1 with anakinra for four weeks did not improve fatigue in CFS, and did not change circulating cytokines beyond IL-1Ra itself. Refuted
Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen after exercise, alongside complement activation and metabolic abnormalities. Early
Single-cell transcriptomics showed classical-monocyte dysregulation at baseline and platelet-activation signatures after exercise, but a re-analysis of the same data found no reproducible composition or expression differences. Contested
Plasma cell-free RNA carries an immune-weighted ME/CFS signature that classifies cases with held-out AUC 0.81, with raised plasmacytoid-dendritic, monocyte and T-cell-derived transcripts. Early
Clinically relevant mast cell activation is reported in up to a quarter of ME/CFS patients, associates with orthostatic intolerance and POTS, and those patients report more benefit from mast-cell-directed treatment. Early
In the interferon-alpha fatigue model, higher baseline IL-6 and IL-10 and a larger early cytokine and fatigue surge predict fatigue that persists after the immune trigger is gone. Early
Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes behind it, predict who develops post-infective fatigue, while microbiological and psychological factors do not. Replicated
Newly reported elevations of serum IL-11 and MMP-9 in female ME/CFS patients are unreplicated single-cohort findings. Early
About 11-12% of people with confirmed EBV, Q fever or Ross River virus infection meet CFS criteria at 6 months, at a similar rate for all three pathogens, and the severity of the acute illness predicts it. Replicated
A waterborne Giardia outbreak in Bergen (2004) left chronic fatigue in 26% of exposed people vs 11% of matched controls 10 years later (adjusted OR 3.0), and more than half of those referred with lasting fatigue met CFS criteria. Several studies

Complement activation and thromboinflammationChronic complement activation damages vessel lining and platelets, so clotting and inflammation feed each other. active

The terminal complement complex is persistently activated in active long COVID, with signs of haemolysis, platelet activation and monocyte-platelet aggregates. This could injure endothelium and capillaries. It is the best-replicated blood abnormality in long COVID but is less studied in classic ME/CFS.

Proponents: Boyman and Cervia-Hasler (Zurich), Baillie (Oxford), Klein/Iwasaki

Evidence for

Cervia-Hasler 2024 (Science), replicated in outline by Baillie 2024, Klein 2023, Barratt-Due 2024
Higher C3 and complement variants reported in ME/CFS (CDC; Solving ME/CFS re-analysis)

Evidence against

One adjusted null replication (Farztdinov preprint)
No complement inhibitor trial yet
Limited ME/CFS data

Predicts

Raised sC5b-9 and complement split products
Endothelial injury markers
Benefit from complement inhibitors

Tests that would settle it

Complement inhibitor RCT in biomarker-positive long COVID
Complement measures in non-COVID ME/CFS cohorts

Notes from each body system

Blood flow, vessels and blood: for: Replicated complement activation in active long COVID; exercise C4a rise and genetic alternative-pathway subgroup in ME/CFS.; against: ME/CFS proteomic complement direction inconsistent across cohorts; no complement-inhibitor trial.
Immune system: for: Complement activation reported in the Che/Lipkin 2025 multi-omics with worsening after exercise; complement dysregulation in active long COVID (Cervia-Hasler 2024) is one of the few independently replicated post-viral blood abnormalities; a CDC study reports higher C3 in ME/CFS.; against: The strong complement evidence is long COVID rather than ME/CFS; one failed replication after age/BMI adjustment; platelet/microclot markers were flat in a harmonised four-cohort proteomic re-analysis.
Infection and viruses: for: Classical complement activation associated with herpesvirus antibody titres in active LC (Cervia-Hasler 2024).; against: Not yet replicated in non-COVID ME/CFS, and causal direction unknown.
Metabolism and blood omics: for: Complement up in Hoel and UKB; EV complement rise post-exercise in men.; against: Complement direction inconsistent across cohorts (down in Walitt); platelet set lower.

Linked findings

Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be increased in platelet-poor plasma in ME/CFS and long COVID, but the method is not standardised and haemostasis experts dispute the inference that they cause symptoms. Contested
Platelet activation findings in ME/CFS are inconsistent: one network reports hyperactivated platelets, while an earlier controlled study found no platelet activation or hypercoagulability. Contested
Routine coagulation markers (D-dimer, fibrinogen, PT/APTT) are generally normal in ME/CFS; claims of hypercoagulability rest on small 1990s studies and on thromboelastography from one network. Contested
Active long COVID shows persistent complement dysregulation (terminal complex, alternative and classical pathway activation) with thromboinflammation markers, replicated across several cohorts. Replicated
In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetically defined alternative-pathway subgroup, and proteomic signals that point in different directions across cohorts. Contested
Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the largest patient-specific changes in the 24 h recovery phase, with suppressed T/B-cell and IL-17 signalling and raised glycolysis proteins. Contested
Extracellular-vesicle protein cargo responds to maximal exercise differently in ME/CFS (blunted in women; TCA proteins down and complement up in men), and recovery-phase changes correlate with PEM severity. Early
CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct immunotypes), but no CSF protein signature has replicated, and ME/CFS and fibromyalgia CSF proteomes were indistinguishable. Contested
Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen after exercise, alongside complement activation and metabolic abnormalities. Early
Single-cell transcriptomics showed classical-monocyte dysregulation at baseline and platelet-activation signatures after exercise, but a re-analysis of the same data found no reproducible composition or expression differences. Contested
Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to non-SARS-CoV-2 viruses especially EBV, and low cortisol, giving a reference phenotype for post-infectious illness. Several studies
Serologic evidence of recent EBV reactivation (EA-D IgG) is associated with long COVID fatigue (OR 2.12) in a 280-person cohort, consistent with an earlier small report. Several studies
In active long COVID, classical complement pathway activation tracked raised antibody titres against several herpesviruses, alongside markers of haemolysis, platelet activation and monocyte-platelet aggregates. Early
SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVID than in pre-pandemic samples (excess +11% at 3-6 months, +5.4% at 10-14 months), with more antigenemia after more severe acute illness. Several studies
Hepatocyte enzymes (ADH4, GSTA1, AGXT, BCHE) are modestly higher, and vascular mural/matrix proteins (ITGAV, NOTCH3, LUM, THBS1, TGF-beta family) modestly lower, in more than one cohort. Early
Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPET, but a within-person re-analysis of the same data finds no protein whose exercise response differs at FDR. Contested
Plasma extracellular vesicle (EV) counts may be higher in ME/CFS, but no EV protein or microRNA difference has survived multiple-testing correction in a replicated cohort. Early
Plasma cell-free RNA classified ME/CFS vs sedentary controls with test-set AUC 0.81 and suggested platelet, pDC and monocyte-derived differences; not yet externally validated. Early
MR of DecodeME suggests higher mitochondrial DNA copy number is protective and higher GlycA and mean platelet volume raise risk, while genetically predicted metabolite leads (IPA, kynurenine, ketones, carnitine, bile acids) and vitamin D, BMI and blood pressure are null or unreplicated. Early

Fibrin amyloid microclotsFibrinolysis-resistant amyloid fibrin(ogen) microclots block capillaries and trap inflammatory molecules. weakened

Fluorescence microscopy of platelet-poor plasma shows amyloid microclots, reported in long COVID and ME/CFS. They could obstruct capillaries and reduce oxygen delivery. Anticoagulant 'triple therapy' was promoted on this basis. The method is difficult to standardise and independent labs disagree.

Proponents: Pretorius (Stellenbosch), Kell (Liverpool)

Evidence for

Pretorius 2021 and later papers from the same network

Evidence against

STOP-PASC found no microclot differences vs controls (Maestri 2025 preprint)
Assay not standardised; few independent replications
No RCT of anticoagulation for microclots; bleeding risk

Predicts

Microclot counts higher in patients
Clearance associated with improvement

Tests that would settle it

Blinded multi-lab microclot quantification on shared samples
Placebo-controlled trial with microclot endpoints

Notes from each body system

Blood flow, vessels and blood: for: Pretorius/Kell network reports raised microclot load in ME/CFS and LC; Dalton 2024 independent count in LC higher on average.; against: Method unstandardised and visually scored; wide overlap with controls; no in vivo proof; haemostasis critiques (Hunt 2024); platelet/coagulation proteins not raised in pooled proteomics; Kennedy 2006 no platelet activation.
Muscle, exertion and energy: for: Amyloid-containing deposits in long COVID muscle increased after PEM induction (Appelman 2024).; against: Not yet shown in ME/CFS muscle; single lab; the composition of the deposits is not established.

Linked findings

Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be increased in platelet-poor plasma in ME/CFS and long COVID, but the method is not standardised and haemostasis experts dispute the inference that they cause symptoms. Contested
Platelet activation findings in ME/CFS are inconsistent: one network reports hyperactivated platelets, while an earlier controlled study found no platelet activation or hypercoagulability. Contested
Routine coagulation markers (D-dimer, fibrinogen, PT/APTT) are generally normal in ME/CFS; claims of hypercoagulability rest on small 1990s studies and on thromboelastography from one network. Contested
In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed lower mitochondrial enzyme activity, more fibre necrosis, more T-cell infiltration and more amyloid-containing deposits than before exercise and than in controls. Early
The most-replicated single cytokine finding, elevated TGF-beta, is largely a blood-handling artefact driven by platelet activation during centrifugation. Refuted
Single-cell transcriptomics showed classical-monocyte dysregulation at baseline and platelet-activation signatures after exercise, but a re-analysis of the same data found no reproducible composition or expression differences. Contested
In active long COVID, classical complement pathway activation tracked raised antibody titres against several herpesviruses, alongside markers of haemolysis, platelet activation and monocyte-platelet aggregates. Early
SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVID than in pre-pandemic samples (excess +11% at 3-6 months, +5.4% at 10-14 months), with more antigenemia after more severe acute illness. Several studies

Cell danger response (Naviaux)Cells stuck in a protective low-energy 'danger' state after a threat fail to complete the healing cycle. speculative

The cell danger response (CDR) is a conserved metabolic reaction to threat in which mitochondria shift from energy production to defence, with extracellular ATP (purinergic) signalling. Naviaux proposes that ME/CFS is a persistent CDR, a hypometabolic state like dauer. His 2016 metabolomics found most altered metabolites decreased, consistent with hypometabolism. Antipurinergic drugs such as suramin are the proposed test.

Proponents: Robert Naviaux (UCSD)

Evidence for

Naviaux 2016 PNAS: 80% of altered metabolites decreased (n=45 vs 39)
Rapamycin responders show purine metabolism changes (Gile 2026)

Evidence against

Metabolomic findings vary across cohorts; Hoel/Germain datasets do not show uniform hypometabolism
No ME/CFS trial of suramin

Predicts

Hypometabolic metabolomic profile
Raised extracellular ATP/purinergic signalling
Response to antipurinergic therapy

Tests that would settle it

Low-dose suramin RCT in ME/CFS
Replication of hypometabolic signature with standardised fasting samples

Notes from each body system

Muscle, exertion and energy: for: Delayed, prolonged PEM (Moore 2023) and recovery-phase molecular changes (Germain 2025) fit a stress response that persists after the stimulus.; against: No study has measured purinergic/cell-danger signalling directly in muscle during PEM.
Metabolism and blood omics: for: Naviaux 2016: 80% of discriminating metabolites lower; lipid/purine changes; Hoel 2021 energy strain.; against: No peer-reviewed replication of a homogeneous hypometabolic profile; pooled cohorts show small, mixed shifts.

Linked findings

PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took ME/CFS patients a mean of about 13 days to recover versus about 2 days in sedentary controls. Several studies
Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but once each person is compared with themselves the exercise response differs little from controls. Contested
Extracellular-vesicle protein cargo responds to maximal exercise differently in ME/CFS (blunted in women; TCA proteins down and complement up in men), and recovery-phase changes correlate with PEM severity. Early
Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen after exercise, alongside complement activation and metabolic abnormalities. Early
About 11-12% of people with confirmed EBV, Q fever or Ross River virus infection meet CFS criteria at 6 months, at a similar rate for all three pathogens, and the severity of the acute illness predicts it. Replicated
HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord tissue from 3 ME/CFS post-mortem donors but not in 24 controls (Kasimir 2022). Early
Blood metabolome differences between ME/CFS and controls are real at group level but small (mostly 0.2-0.5 SD) and partly cohort-specific; early near-perfect classification did not carry over to new cohorts. Replicated
Naviaux 2016's picture of a homogeneous 'hypometabolic' dauer-like state (80% of discriminating metabolites decreased) has not been confirmed as a uniform pattern in later independent cohorts. Contested
Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across several cohorts and platforms, with modest effect sizes. Several studies

Mitochondrial / energy metabolism impairmentCells cannot produce energy normally, especially during exertion, causing early lactate and PEM. active

Several lines suggest impaired pyruvate dehydrogenase, Complex V inefficiency or reduced oxidative capacity. This would force early reliance on glycolysis and slow recovery. It may be intrinsic or secondary to poor perfusion or circulating factors.

Proponents: Fluge and Mella (PDH), Missailidis and Fisher (La Trobe), Myhill/Booth, Hanson (Cornell)

Evidence for

Amino-acid pattern implying impaired PDH (Fluge 2016, n=200 vs 102)
Complex V inefficiency in lymphoblasts (Missailidis 2020)
Post-exercise mitochondrial enzyme drop in long COVID muscle (Appelman 2024)
Aromatic lactate deficit across 5 cohorts (Solving ME/CFS re-analysis)

Evidence against

Serum effect on myoblast respiration failed replication at n=120 (Ryback/Ponting 2026)
CoQ10 alone negative in long COVID
Supplement trials small and sponsor-linked

Predicts

Low PDH flux markers
Abnormal mitochondrial respiration in patient cells
Benefit from metabolic supplements

Tests that would settle it

Muscle 31P-MRS and biopsy respiration before/after PEM vs deconditioned controls
Independent replication of lymphoblast Complex V defect

Notes from each body system

Muscle, exertion and energy: for: Lower SDH/OXPHOS and type I atrophy in ME/CFS muscle (Charlton 2026); post-exertion SDH drop in long COVID (Appelman 2024); WASF3 (Wang 2023); 31P-MRS subgroups (Jones 2010/2012); early lactate per watt (Lien 2019); AMPK defect in cultured myotubes (Brown 2015).; against: Ryback 2026 refutes the serum-myoblast respiration effect; blood-cell mitochondrial findings conflict in direction (Tomas vs Missailidis) or are null (Vermeulen 2010, Tomas 2019); OXPHOS does not correlate with VO2max in patients (Charlton 2026), so mitochondria may not be the limiting step; several MRS studies null (Barnes 1993, Godlewska 2025).
Nervous system and brain: for: Raised brain lactate at rest and failure to raise it further under hypoxia (Bader 2026); lower cortical glutathione; post-exertion cognitive decline.; against: Brain lactate also raised in depression; no 31P-MRS energetics replicated; samples small.
Metabolism and blood omics: for: Fluge 2016 PDH pattern; lower carnitines/acylcarnitines; low muscle proteins in blood; Germain 2022 blunted glucose rise; Hoel 2021 metabotypes.; against: Lactate/pyruvate and other energy ratios null in pooled data; exercise metabolome response mostly normal within person.
Genetics: for: MR suggests mtDNA copy number is protective, and the DARS2 gene sits at the chr1 lead.; against: There is no muscle tissue enrichment, mtDNA haplogroups are not associated, and the FBXL4 assignment is distant.
Treatment trials and natural history: for: Small CoQ10+NADH and oxaloacetate RCTs; rapamycin open-label with autophagy marker changes.; against: CoQ10 alone negative in long COVID; positive supplement trials sponsor-linked and unreplicated.

Linked findings

A subset of patients with unexplained exertional intolerance cannot extract oxygen normally from blood during exercise, associated with hypocapnia and possible arteriovenous shunting. Early
Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per fibre size in both ME/CFS and long COVID, unlike the pattern after 60 days of bed rest; thickened capillary basement membrane has been reported but is not yet peer-reviewed. Early
PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took ME/CFS patients a mean of about 13 days to recover versus about 2 days in sedentary controls. Several studies
On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled about 5 ml/kg/min below controls) with a low peak heart rate (chronotropic incompetence). Replicated
ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most clearly in workload and VO2 at the ventilatory threshold, but the largest independent replication found no day-2 change. Contested
Some patients produce more lactate for a given workload, and this worsens on a repeat test, but plasma studies after one maximal test show a smaller total lactate rise that mainly reflects lower achieved workloads. Several studies
Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (preload failure) and, in a high-flow subgroup, poor systemic oxygen extraction by tissue. Several studies
31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower pH recovery after repeated exercise in a subset of patients, but other MRS studies found no consistent group abnormality. Contested
Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without the fibre atrophy and type 2 shift expected from inactivity, with abnormalities confined to subgroups. Contested
In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed lower mitochondrial enzyme activity, more fibre necrosis, more T-cell infiltration and more amyloid-containing deposits than before exercise and than in controls. Early
ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy, lower mitochondrial capacity and fewer capillaries per fibre, a pattern that 60 days of strict bed rest in healthy people did not reproduce. Early
WASF3 protein is raised in ME/CFS muscle, and in cells and mice it disrupts respiratory supercomplex assembly and lowers exercise capacity via ER stress. Early
Muscle cells grown from ME/CFS patients failed to activate AMPK and raise glucose uptake after electrical pulse stimulation, and secreted less IL-6, suggesting a defect that persists outside the body. Early
Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal respiration, T-cell glycolysis), higher (lymphoblast maximal respiration with Complex V inefficiency) or normal, depending on lab and cell type. Contested
Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling amino acids and raised PDK mRNA in blood cells; this has not been independently confirmed. Early
The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pre-registered replication of 67 patients vs 53 controls, though a small 3D engineered-muscle study reported weaker contraction with patient serum. Refuted
Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but once each person is compared with themselves the exercise response differs little from controls. Contested
Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the largest patient-specific changes in the 24 h recovery phase, with suppressed T/B-cell and IL-17 signalling and raised glycolysis proteins. Contested
Extracellular-vesicle protein cargo responds to maximal exercise differently in ME/CFS (blunted in women; TCA proteins down and complement up in men), and recovery-phase changes correlate with PEM severity. Early
Repeated maximal handgrip testing shows lower strength, faster decline across repetitions and poorer recovery after 60 minutes in ME/CFS, correlating with severity and PEM. Several studies
Muscle-derived proteins (myoglobin, troponins, CA3 and similar) are lower in ME/CFS blood at rest across three SomaScan cohorts and UK Biobank, not explained by step counts or walking pace. Several studies
Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison groups, and a 60-day bed-rest model) indicate that deconditioning alone does not explain the exercise and muscle abnormalities in ME/CFS. Several studies
After a single 30-minute exercise bout, women with ME/CFS performed worse on a demanding cognitive task and recruited more parietal and cingulate cortex. Controls improved. Early
During a sustained grip task, ME/CFS patients fatigue earlier while their brain motor-network activity fails to ramp up as it does in controls (NIH; reduced temporoparietal-junction activation). Early
Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in three cohorts from one group, and thalamic or regional lactate in two other groups. It is also raised in depression, so it is not specific. Several studies

IDO metabolic trap (Phair, Davis)A bistable tryptophan-kynurenine circuit, set by IDO1/IDO2 variants and substrate inhibition, flips into a persistent low-tryptophan state. speculative

IDO1 converts tryptophan to kynurenine and is inhibited by high tryptophan. Combined with a deleterious IDO2 variant, a mathematical model predicts two stable states, one of which traps cells with low tryptophan and serotonin and high kynurenine after an inflammatory trigger. Exiting the trap would need a perturbation. It is a mathematical model with limited direct data.

Proponents: Robert Phair, Ron Davis (OMF / Stanford), Alex Kashi

Evidence for

Preliminary OMF data on IDO2 variants (unpublished/limited)
Altered tryptophan metabolites in some metabolomic studies

Evidence against

DecodeME GWAS did not highlight IDO loci
Mendelian randomisation finds no causal effect of kynurenine on ME/CFS risk (Solving ME/CFS re-analysis)
Plasma kynurenine findings inconsistent

Predicts

Enrichment of IDO2 deleterious variants
Altered intracellular tryptophan/kynurenine
Bistable behaviour in patient cells

Tests that would settle it

Intracellular tryptophan/kynurenine measurement in patient cells with IDO genotypes
Test of predicted bistability in culture

Notes from each body system

Muscle, exertion and energy: for: Fluge 2016 amino-acid pattern implies altered fuel handling.; against: Kynurenine pathway metabolites show no patient-specific exercise response in the Solving ME/CFS re-analysis of Germain 2022.
Metabolism and blood omics: for: Model plausibility (Kashi 2019); Kavyani 2024 higher kynurenine tracking severity.; against: Pooled data show slightly lower kynurenine; kyn/trp ratio null; IDO1/KYNU protein null; plasma cannot test intracellular kinetics.

Linked findings

Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling amino acids and raised PDK mRNA in blood cells; this has not been independently confirmed. Early
Targeted CSF metabolomics found raised serine and phospholipids and reduced 5-methyltetrahydrofolate in ME/CFS, with different metabolite responses to exercise than in controls. Early
In the interferon-alpha fatigue model, higher baseline IL-6 and IL-10 and a larger early cytokine and fatigue surge predict fatigue that persists after the immune trigger is gone. Early
The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood data are inconsistent: one cohort found higher kynurenine, others found no difference or slightly lower kynurenine. Contested
NAD+ depletion is a proposed mechanism but has not been directly established in ME/CFS cohorts; NAD precursor trials raise NAD+ without clear clinical benefit. Early
Plasma indole-3-propionate (IPA), a gut-bacteria product of tryptophan, is lower in ME/CFS in 4 of 4 plasma cohorts (pooled −0.33 SD), and the deficit is fully explained by stool composition, mainly the R. gnavus-type overgrowth. Replicated
Oxaloacetate 2 g/day reduced Chalder fatigue more than control (−25% vs −10%) in an 82-person company-run RCT in ME/CFS; a 69-person long-COVID trial missed its primary endpoint. Early

HPA axis bluntingThe stress-hormone axis is under-active, with low cortisol and enhanced feedback. weakened

Mild hypocortisolism, a flattened diurnal rhythm and enhanced negative feedback are reported in ME/CFS. This is probably a consequence of illness, inactivity or sleep disturbance rather than a cause, but it could amplify symptoms. Low-dose hydrocortisone had small effects and suppressed the adrenal axis.

Proponents: Cleare (King's), Demitrack (early work)

Evidence for

Repeated reports of mild hypocortisolism (2026 meta-analysis per atlas)
Cleare 1999 hydrocortisone 5-10 mg: 28% vs 9% improved

Evidence against

McKenzie 1998 hydrocortisone RCT: adrenal suppression, little benefit
Changes small and nonspecific

Predicts

Low salivary awakening cortisol
Enhanced dexamethasone suppression
Response to low-dose hydrocortisone

Tests that would settle it

Longitudinal cortisol from pre-illness in prospective cohorts

Notes from each body system

Hormones: for: A meta-analysis of 46 studies shows mildly low free cortisol and enhanced feedback. The genetic signal is enriched in the pituitary. Adrenal/pituitary diagnoses are enriched in Finnish registries.; against: The effects are small and overlap normal ranges. A similar pattern appears in other fatigue and pain conditions. Hydrocortisone and fludrocortisone trials were weak or null. Cortisol MR is null. In long COVID, low cortisol did not replicate in timed samples.

Linked findings

The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not raised in ME/CFS or POTS despite low volume, and blood proteomics shows no consistent change in them. Several studies
Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep onset and lower sleep efficiency. Standard polysomnography usually finds no clinical disorder that explains non-restorative sleep. Several studies
Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to non-SARS-CoV-2 viruses especially EBV, and low cortisol, giving a reference phenotype for post-infectious illness. Several studies
In UK Biobank, ME/CFS cases show many small blood-trait differences (inflammation, insulin resistance, liver markers) that replicate partly in All of Us and are not explained by low physical activity. Several studies
On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and morning cortisol, 24-h urinary cortisol and hair cortisol) than healthy controls. Replicated
Challenge tests show a blunted cortisol response to ACTH and exaggerated cortisol suppression after a glucocorticoid dose, consistent with enhanced negative feedback. Several studies
Lower cortisol in long COVID, reported in a 2023 Nature cohort, has not held up in later clinic and cohort studies that controlled for sampling time. Contested
Low-dose hydrocortisone produced at most small, short-term fatigue benefits in two RCTs, at the cost of adrenal suppression and bone loss at higher doses. Contested
Fludrocortisone, a synthetic salt-retaining steroid, did not improve CFS with neurally mediated hypotension in a 100-person RCT, and failed again in combination with hydrocortisone. Refuted
Diagnosed adrenal insufficiency and hypopituitarism are 7 to 13 times more common than expected among Finnish patients coded with postviral fatigue syndrome. Early
In one Dutch case-control study, CFS patients had normal TSH but lower total T3, T4 and free T3, with a higher reverse-T3 fraction ('low T3 syndrome' pattern in 16% vs 7%). Early
Reports of low IGF-1 and reduced growth hormone responses in CFS conflict, and the largest controlled study found an intact GH/IGF system. Contested
ME/CFS and long COVID are more common in women (ME/CFS prevalence about 1.4% vs 0.9%; DecodeME 82% female), but genetic risk variants are not sex-specific. Replicated
Women with CFS report more endometriosis, heavy bleeding, pelvic pain, hysterectomy and earlier menopause than controls in US population-based case-control studies. Early
Lower plasma androgenic steroids in ME/CFS (Cornell metabolomics) is mostly an age and sex artefact and does not replicate in BioMapAI. Contested
Melatonin may help the subgroup with delayed melatonin onset, and melatonin plus zinc modestly reduced physical fatigue in one small RCT. Early
ME/CFS genetic risk is enriched in brain tissues and the pituitary, but genetically predicted morning cortisol does not change risk. Early

Pituitary / vasopressin deficiencyLow ADH (vasopressin) secretion despite concentrated blood leads to fluid loss and low blood volume. speculative

A 2025 study reported undetectable vasopressin in most patients despite high plasma osmolality. If true, this would explain polyuria, thirst and low blood volume, and suggest desmopressin for a subgroup. DecodeME tissue enrichment pointing to pituitary gives indirect support.

Proponents: Hanson-adjacent clinical groups, Bragée/Huhmar/Polo (2025 report)

Evidence for

Vasopressin undetectable in 82% in an uncontrolled 2025 series
DecodeME genetic signal enriched in brain and pituitary tissues

Evidence against

Uncontrolled, assay concerns
Desmopressin evidence only acute in POTS (Coffin 2012)

Predicts

Low copeptin relative to osmolality
Response to desmopressin

Tests that would settle it

Controlled copeptin study with water deprivation/hypertonic saline testing

Notes from each body system

Hormones: for: One large uncontrolled series found undetectable vasopressin despite high osmolality. Miwa found low ADH and partial response to desmopressin.; against: Copeptin, the more reliable marker, was normal or low-normal. There is no controlled water-deprivation study and no pituitary lesions on MRI.

Linked findings

Red cell volume and total blood volume are below predicted in a large minority to about half of ME/CFS patients measured by isotope or CO-rebreathing methods. Replicated
The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not raised in ME/CFS or POTS despite low volume, and blood proteomics shows no consistent change in them. Several studies
Diagnosed adrenal insufficiency and hypopituitarism are 7 to 13 times more common than expected among Finnish patients coded with postviral fatigue syndrome. Early
After overnight fluid deprivation, plasma vasopressin was undetectable in 82% of 111 Swedish ME/CFS patients despite high plasma osmolality in 57%, suggesting partial central diabetes insipidus-like physiology. Early
ME/CFS genetic risk is enriched in brain tissues and the pituitary, but genetically predicted morning cortisol does not change risk. Early

Neuroinflammation (microglial activation)Activated microglia and astrocytes in the brain produce fatigue, cognitive problems and pain. active

PET and MR spectroscopy studies suggest brain inflammation in ME/CFS and long COVID. Inflammation in brainstem or limbic regions could generate sickness behaviour and PEM. Results conflict and tracer studies are small.

Proponents: Nakatomi/Watanabe (Japan), VanElzakker, Younger

Evidence for

Nakatomi 2014 TSPO PET positive (n=9)
Long-COVID TSPO increase (VanElzakker 2024)
Raised ventricular lactate (Shungu/Mathew)

Evidence against

Raijmakers 2022 TSPO PET negative
James (Stanford) 2026: no brain TSPO signal, strong muscle signal
LDN fatigue RCT negative (2026)

Predicts

Raised TSPO PET signal
Raised brain lactate/choline
Benefit from glial modulators (LDN, aripiprazole)

Tests that would settle it

Large multi-tracer PET studies with standardised analysis
CSF inflammatory markers vs matched controls

Notes from each body system

Nervous system and brain: for: Nakatomi TSPO-PET; MRS raised lactate, choline and brain temperature; diffusion NII signals; VanElzakker PBR28 in long COVID; CSF cytokine changes (Hornig).; against: Raijmakers TSPO-PET null; Tuomaala long COVID TSPO null; NfL/GFAP normal; CSF proteomes indistinguishable from fibromyalgia and no replicated CSF signature.
Treatment trials and natural history: for: Low-dose aripiprazole, LDN rationales; retrospective reports.; against: LDN RCT negative for fatigue; no controlled aripiprazole data.

Linked findings

Resting heart rate is higher and high-frequency heart-rate variability lower in ME/CFS. This fits reduced vagal (parasympathetic) tone. Replicated
Conventional structural MRI (grey matter volume, routine clinical reads) shows no consistent abnormality in ME/CFS. Reported regional volume differences rarely replicate across groups. Contested
Advanced diffusion MRI models report widespread white-matter differences in ME/CFS that conventional DTI misses, and the pattern may differ between post-infectious and gradual-onset cases. Early
Brainstem differences (smaller volume, altered signal or connectivity) are among the more often repeated imaging findings in ME/CFS, though the methods differ between studies. Several studies
After a single 30-minute exercise bout, women with ME/CFS performed worse on a demanding cognitive task and recruited more parietal and cingulate cortex. Controls improved. Early
During a sustained grip task, ME/CFS patients fatigue earlier while their brain motor-network activity fails to ramp up as it does in controls (NIH; reduced temporoparietal-junction activation). Early
Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in three cohorts from one group, and thalamic or regional lactate in two other groups. It is also raised in depression, so it is not specific. Several studies
MRS thermometry found raised temperature in several brain regions (insula, putamen, frontal cortex, thalamus, cerebellum) in women with ME/CFS, not explained by body temperature. Early
TSPO-PET evidence of microglial activation in ME/CFS is contested: one Japanese study found 45-199% higher binding in limbic and brainstem regions, and a Dutch study using the same tracer found none. Contested
In long COVID, one [11C]PBR28 PET study found widespread raised TSPO signal that correlated with vascular markers, and a [11C]PK11195 study found no group difference. Contested
In the NIH cohort, CSF levels of the catechol metabolites DOPA, DOPAC and DHPG were lower in post-infectious ME/CFS. Noradrenaline and dopamine did not differ. Early
CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct immunotypes), but no CSF protein signature has replicated, and ME/CFS and fibromyalgia CSF proteomes were indistinguishable. Contested
Markers of neuronal or glial injury (NfL, GFAP, tau, UCH-L1) are not raised in ME/CFS blood or CSF. Several studies
Meta-analysis shows moderate-to-large deficits in information-processing speed, attention, working memory and reaction time in ME/CFS, with intact vocabulary and reasoning. Performance worsens with upright posture and after exertion. Replicated
Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep onset and lower sleep efficiency. Standard polysomnography usually finds no clinical disorder that explains non-restorative sleep. Several studies
Self-reported light and sound sensitivity is common in ME/CFS. One ERP study found normal early sensory gating (P50) but a reduced P300, pointing to higher-order processing rather than early filtering. Early
Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and disturbed IL-1 signalling, but CSF studies are small and partly contradicted by later proteomics. Early
Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes behind it, predict who develops post-infective fatigue, while microbiological and psychological factors do not. Replicated
HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord tissue from 3 ME/CFS post-mortem donors but not in 24 controls (Kasimir 2022). Early
SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a minority of people, and in a Beijing tissue study, viral RNA in tissue was associated with long COVID symptoms (OR 5.17). Several studies
CSF proteomics in ME/CFS is limited to small cohorts and has not produced a replicated protein difference; the largest recent study (31 patients) had no controls. Early
Genome-wide ME/CFS heritability is enriched in brain tissues and neurons (including glutamatergic synapse genes and subcortical/cerebellar neuron types), with no enrichment in blood immune cells. Several studies
The first sizeable placebo-controlled trial of low-dose naltrexone in post-COVID ME/CFS (66 vs 71, 16 weeks, up to 4.5 mg) found no fatigue benefit, with over 40% improving on placebo. Refuted
In a retrospective single-clinic review, 74% of 101 patients on very low-dose aripiprazole (mostly 0.25-2 mg) reported improvement; there is no randomised trial. Early
Forty daily sessions of hyperbaric oxygen at 2.0 ATA improved cognition and symptoms versus sham in a 73-person long-COVID trial, but ten sessions at 2.4 bar showed no benefit in an 80-person sham-controlled trial. Contested

Brainstem and craniocervical instabilityBrainstem dysfunction, sometimes from craniocervical instability or Chiari, disrupts autonomic and arousal control. speculative

Some patients, often with hypermobility, are reported to have craniocervical instability or intracranial hypertension compressing the brainstem. Surgery has been followed by improvement in case series. Imaging studies of brainstem volume and connectivity are small. This may explain a minority subgroup at most.

Proponents: Henderson/Bolognese (surgeons), Bragée (Stockholm), Shan (brainstem MRI)

Evidence for

Case series of fusion surgery improvement
Bragée 2020: CCI and intracranial hypertension common in a referral sample

Evidence against

No controlled trial; surgery high risk
Imaging criteria not standardised; selection bias

Predicts

Upright MRI abnormalities more common
Surgical benefit in controlled studies

Tests that would settle it

Blinded imaging comparing ME/CFS with matched hypermobile controls
Prospective registry of surgical outcomes with controls

Notes from each body system

Nervous system and brain: for: Brainstem imaging differences repeatedly reported (volume, connectivity); uncontrolled series with high rates of craniocervical findings and raised opening pressure in a subset.; against: No controlled or blinded imaging of craniocervical anatomy; no trials of surgery; CSF-leak codes not specific to ME/CFS in registry data; imaging brainstem changes need not be mechanical.

Linked findings

Brainstem differences (smaller volume, altered signal or connectivity) are among the more often repeated imaging findings in ME/CFS, though the methods differ between studies. Several studies
Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying tonsils and signs of raised intracranial pressure in severe ME/CFS, and surgical case reports describe improvement. There are no controlled imaging studies or trials. Early
CSF proteomics in ME/CFS is limited to small cohorts and has not produced a replicated protein difference; the largest recent study (31 patients) had no controls. Early
Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4 in 60% of children with CFS vs 24% of controls; 50% in a Swedish severe cohort; EDS about 100x expected in Finnish registries). Several studies

Persistent sickness behaviour / vagal signallingThe brain's sickness response, normally triggered by infection signals (e.g. via the vagus), stays switched on. speculative

During infection, cytokines signal to the brain and cause fatigue, malaise and withdrawal. If a local infection or inflammation near the vagus nerve persists, or the brain's set point changes, sickness behaviour could persist. This frames many symptoms but is hard to test directly.

Proponents: VanElzakker (vagus nerve infection hypothesis), Dantzer (sickness behaviour biology)

Evidence for

Symptoms closely resemble sickness behaviour
Post-infectious onset common

Evidence against

Anti-inflammatory trials negative in long COVID
Direct evidence of vagal infection absent

Predicts

Vagal or brainstem markers of immune signalling
Symptoms reduced by blocking cytokine signalling

Tests that would settle it

Vagal nerve imaging/biopsy studies; vagal stimulation RCTs with sham

Notes from each body system

Muscle, exertion and energy: for: Walitt 2024 framed grip and effort findings as altered effort preference.; against: Effort criteria are met in most CPETs; objective muscle, capillary and post-exertion biopsy changes (Appelman, Charlton) are not explained by behaviour; the effort-preference interpretation has been formally criticised.
Nervous system and brain: for: Central fatigability (brain fails to ramp up motor drive), lower CSF catechol metabolites, altered task activation after exertion: consistent with a brain-level 'sickness' state.; against: The 'effort preference' reading is confounded by inability to complete hard tasks; central changes could equally be downstream of peripheral pathology.
Immune system: for: Dubbo cohort: 11% met CFS criteria six months after three different pathogens with a stereotyped phenotype predicted by acute illness severity, not by pathogen or psychological factors; PBMC IL-1beta and IL-6 release tracks acute fever, malaise, pain, fatigue, mood and concentration; IFN-gamma and IL-10 polymorphisms synergistically set severity and duration; interferon-alpha reliably produces neurovegetative fatigue within two weeks (Capuron 2002) and persistent fatigue in a third of patients, predicted by higher baseline IL-10 and a bigger early surge (Russell 2019).; against: The model explains onset far better than persistence: cytokines are not persistently abnormal in long-duration illness (Hornig 2015), cytokine inhibition does not help, and the acute cytokine signal fades while the illness does not.
Infection and viruses: for: Stereotyped phenotype across pathogens, with no ongoing pathogen required (Dubbo, QFS doxycycline failure).; against: Does not explain why only ~10% persist, or tissue-specific viral findings.
Hormones: for: Hypocortisolism with enhanced feedback fits a chronic post-infectious or inflammatory state.; against: This cannot be separated from deconditioning or sleep effects in cross-sectional data.
Treatment trials and natural history: for: Large placebo responses show symptoms are modifiable by expectation and context (not specific to this hypothesis).; against: PACE effects confined to self-report and not durable; GET rated harmful by patients; rehab worsened 42/97 in CFS_CARE.

Linked findings

PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took ME/CFS patients a mean of about 13 days to recover versus about 2 days in sedentary controls. Several studies
After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), adrenergic and immune genes rose for up to 48 h in most CFS patients but not in controls or MS patients. Early
Repeated maximal handgrip testing shows lower strength, faster decline across repetitions and poorer recovery after 60 minutes in ME/CFS, correlating with severity and PEM. Several studies
Resting heart rate is higher and high-frequency heart-rate variability lower in ME/CFS. This fits reduced vagal (parasympathetic) tone. Replicated
After a single 30-minute exercise bout, women with ME/CFS performed worse on a demanding cognitive task and recruited more parietal and cingulate cortex. Controls improved. Early
During a sustained grip task, ME/CFS patients fatigue earlier while their brain motor-network activity fails to ramp up as it does in controls (NIH; reduced temporoparietal-junction activation). Early
TSPO-PET evidence of microglial activation in ME/CFS is contested: one Japanese study found 45-199% higher binding in limbic and brainstem regions, and a Dutch study using the same tracer found none. Contested
In the NIH cohort, CSF levels of the catechol metabolites DOPA, DOPAC and DHPG were lower in post-infectious ME/CFS. Noradrenaline and dopamine did not differ. Early
The NIH claim that altered 'effort preference' defines post-infectious ME/CFS rests on the EEfRT (OR 1.65 for choosing hard tasks). Patients completed only 65% of the hard tasks they chose vs 96% for controls, which confounds preference with ability. Contested
Meta-analysis shows moderate-to-large deficits in information-processing speed, attention, working memory and reaction time in ME/CFS, with intact vocabulary and reasoning. Performance worsens with upright posture and after exertion. Replicated
Self-reported light and sound sensitivity is common in ME/CFS. One ERP study found normal early sensory gating (P50) but a reduced P300, pointing to higher-order processing rather than early filtering. Early
T-cell abnormalities differ by severity: mild/moderate patients show cytotoxic/immunosenescence features, severe patients show activated lymphocytes and higher stimulated pro-inflammatory cytokines. Early
Plasma cytokine abnormalities depend strongly on illness duration: a pro- and anti-inflammatory activation pattern early (under 3 years) that is absent in long-established illness. Several studies
In the largest single cytokine study, 17 cytokines rose in step with disease severity although only TGF-beta (up) and resistin (down) differed between all patients and controls. Several studies
Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF-beta and CRP, with 12 other proteins showing no difference. Several studies
Blocking IL-1 with anakinra for four weeks did not improve fatigue in CFS, and did not change circulating cytokines beyond IL-1Ra itself. Refuted
Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and disturbed IL-1 signalling, but CSF studies are small and partly contradicted by later proteomics. Early
In the interferon-alpha fatigue model, higher baseline IL-6 and IL-10 and a larger early cytokine and fatigue surge predict fatigue that persists after the immune trigger is gone. Early
Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes behind it, predict who develops post-infective fatigue, while microbiological and psychological factors do not. Replicated
About 11-12% of people with confirmed EBV, Q fever or Ross River virus infection meet CFS criteria at 6 months, at a similar rate for all three pathogens, and the severity of the acute illness predicts it. Replicated
A waterborne Giardia outbreak in Bergen (2004) left chronic fatigue in 26% of exposed people vs 11% of matched controls 10 years later (adjusted OR 3.0), and more than half of those referred with lasting fatigue met CFS criteria. Several studies
Q fever fatigue syndrome (QFS) follows roughly 20% of acute Q fever and persists more than 10 years after infection in many patients, almost all of whom report post-exertional malaise; doxycycline did not help. Several studies
On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and morning cortisol, 24-h urinary cortisol and hair cortisol) than healthy controls. Replicated
ME/CFS genetics overlaps most with chronic pain, fibromyalgia, IBS and migraine, and substantially with depression, but none of the 8 DecodeME loci share a causal variant with depression. Several studies
PACE (n=641) reported modest gains in self-rated fatigue and function with CBT and graded exercise, but these were not matched by objective measures, used Oxford criteria, and vanished between groups by 2.5 years. Contested

Gut dysbiosisAltered gut microbiota and barrier function drive immune activation and metabolic changes. active

Butyrate producers such as Faecalibacterium prausnitzii are reduced in ME/CFS, and short-duration patients show stronger dysbiosis. Microbial products may affect immunity and metabolism. A synbiotic improved several long-COVID symptoms in one trial. Dysbiosis could also be a consequence of illness, diet and inactivity.

Proponents: Lipkin and Hornig (Columbia), Unutmaz (JAX), Hong Kong SIM01 group

Evidence for

Reduced butyrate producers (Guo 2023, n=106 vs 91)
Dysbiosis stronger early (Xiong 2023)
R. gnavus/F. plautii overgrowth replicates and explains low IPA (Solving ME/CFS re-analysis)
SIM01 synbiotic positive in long COVID (Lau 2024, n=463)

Evidence against

MR finds no causal effect of IPA and related metabolites (Solving ME/CFS re-analysis)
Aromatic lactate deficit not explained by gut bacteria (Solving ME/CFS re-analysis)
Faecal transplant data anecdotal

Predicts

Reduced butyrate producers
Microbial metabolites tracking symptoms
Benefit from microbiome interventions

Tests that would settle it

Replication of SIM01-type trial in ME/CFS
Prospective microbiome sampling before onset

Notes from each body system

Infection and viruses: for: Giardia outbreak produced lasting fatigue + IBS. SARS-CoV-2 gut persistence. Enterovirus in stomach (contested).; against: Gut reservoir data are mostly in LC and convenience samples.
Metabolism and blood omics: for: Low IPA in 4/4 cohorts tied to butyrate producers; Xiong 2023 low plasma butyrate; BioMapAI connectivity.; against: Aromatic lactates, the strongest metabolite signal, are not stool-predicted.
Gut and microbiome: for: Loss of butyrate producers and an R. gnavus-type bloom are seen in at least 4 cohorts and 3 labs, and a pre-specified NIH test gave AUC 0.87. Low IPA in 4/4 cohorts is explained by stool composition. The same pattern occurs in long COVID, where the admission microbiome predicted persistence. There is a genetic overlap with IBS and enteric neurons.; against: Dysbiosis fades in long-standing illness while symptoms persist and does not track severity at FDR. MR shows gut metabolites are not causal. IBS confounds the comparison. The only FMT RCT was null (n=11). The translocation markers are small, lab-specific and use contamination-prone assays.

Linked findings

Increased mucosal-associated invariant T (MAIT) cell proportions and a shift in CD8+ memory subsets were found in the largest single ME/CFS flow-cytometry cohort, but the finding has not been reproduced in independent data. Contested
Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen after exercise, alongside complement activation and metabolic abnormalities. Early
A waterborne Giardia outbreak in Bergen (2004) left chronic fatigue in 26% of exposed people vs 11% of matched controls 10 years later (adjusted OR 3.0), and more than half of those referred with lasting fatigue met CFS criteria. Several studies
Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20% of controls, with EV RNA in 37% vs 5%, but this has not been independently replicated with blinded methods. Contested
SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a minority of people, and in a Beijing tissue study, viral RNA in tissue was associated with long COVID symptoms (OR 5.17). Several studies
Blood metabolome differences between ME/CFS and controls are real at group level but small (mostly 0.2-0.5 SD) and partly cohort-specific; early near-perfect classification did not carry over to new cohorts. Replicated
Four aromatic 2-hydroxy acids (phenyllactate, 4-hydroxyphenyllactate, indolelactate, imidazole lactate) are lower in ME/CFS in 5 of 5 cohorts (combined -0.53 SD, I²=0), robust to drug, caffeine, diet and kidney-function adjustment. Several studies
Indole-3-propionate (IPA), a gut-bacterial tryptophan product, is lower in ME/CFS plasma in 4 of 4 cohorts (-0.33 SD) and tracks butyrate-producing gut species. Several studies
Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-validation but drop to AUC ~0.6-0.77 on external cohorts, so none is usable for diagnosis. Contested
Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium prausnitzii, Eubacterium rectale, Roseburia, Coprococcus) and reduced microbial capacity to make butyrate. Replicated
Ruminococcus gnavus, Flavonifractor plautii, Ruthenibacterium lactatiformans and Enterocloster species are increased in ME/CFS, consistently across BioMapAI, Guo 2023 and NIH cohorts and in long COVID. Several studies
Plasma indole-3-propionate (IPA), a gut-bacteria product of tryptophan, is lower in ME/CFS in 4 of 4 plasma cohorts (pooled −0.33 SD), and the deficit is fully explained by stool composition, mainly the R. gnavus-type overgrowth. Replicated
The replicated deficit of aromatic lactates (indolelactate, phenyllactate, 4-hydroxyphenyllactate) in ME/CFS blood is not predicted by stool bacteria or bacterial genes, so it is probably host-side or due to absorption, not dysbiosis. Early
Gut dysbiosis is strongest in the first years of illness and largely resolves in patients ill for more than 10 years, while their metabolic and clinical abnormalities persist. Early
Links between specific gut bacteria and symptom severity are inconsistent: F. prausnitzii tracked fatigue in one cohort, but nothing survived correction for multiple testing in a 143-patient re-analysis. Contested
IBS is about seven times more common in ME/CFS (pooled OR 7.20; prevalence about 37%), and IBS status drives much of the measured microbiome difference. Replicated
Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to enterobacterial LPS) are raised in several small ME/CFS studies, but standardised replication is lacking. Contested
Quadram Institute studies of severe ME/CFS with household-matched controls report altered gut phage communities and reduced serum IgG reactivity to gut microbes, in very small samples. Early
The only published randomised faecal microbiota transplant trial in CFS (n=11) found no benefit, and a larger Norwegian RCT (n=80) has not yet reported. Early
Probiotic trials in CFS are small; one 39-person RCT of Lactobacillus casei Shirota reduced anxiety but did not target fatigue. Early
ME/CFS genetic risk overlaps with IBS and is enriched in enteric neurons, but genetically predicted gut-derived metabolites (IPA, indolelactate, kynurenine, bile acids) do not change ME/CFS risk. Early
In long COVID, the gut microbiome at hospital admission predicted persistent symptoms, and at 6 months patients had more R. gnavus and fewer butyrate producers, mirroring ME/CFS. Several studies
Microbiome-based machine-learning classifiers distinguish ME/CFS from controls in held-out and external cohorts (BioMapAI, Guo 2023), but they are research tools with modest real-world specificity. Several studies

Genetic susceptibility acting through neuronsInherited variants affecting nervous-system genes raise risk, pointing to brain-based mechanisms. active

DecodeME (15,579 cases) found genome-wide significant loci, with tissue enrichment in brain and pituitary rather than immune tissues. The overall genetic signal replicates in the Million Veteran Program and FinnGen. Candidate genes include pain-sensitisation and immune-response genes. Genetics marks susceptibility, not necessarily the ongoing mechanism.

Proponents: Ponting and DecodeME consortium (Edinburgh)

Evidence for

DecodeME 2025 loci; tissue enrichment brain/pituitary
Genetic correlation rg 0.73 with MVP; EPHB1 and CLEC3B replicate by MR (Solving ME/CFS re-analysis)

Evidence against

Most individual loci not yet replicated
Small effect sizes; heritability modest

Predicts

Loci near neuronal genes
Replication in independent cohorts
Mendelian randomisation links to specific proteins

Tests that would settle it

Independent GWAS replication of individual loci
Functional studies of implicated genes

Notes from each body system

Nervous system and brain: for: Solving ME/CFS re-analysis: DecodeME heritability enriched in brain tissues and neurons, with shared architecture with chronic pain and migraine.; against: Structural MRI largely normal; no neuronal injury markers; no locus is an autonomic gene.
Gut and microbiome: for: Heritability is enriched in enteric neurons and the genetic correlation with IBS is about 0.57, suggesting a gut-brain neural link.; against: These are preprints, and the NOS1 region shared with IBS does not colocalise.
Genetics: for: Brain-only tissue enrichment (DecodeME and independent re-analyses), glutamatergic synapse and specific neuron types (Maccallini 2026), CA10 shared with chronic pain, EPHB1 MR, and GRIN2A in Slaughter 2026.; against: Most results are preprints. Individual loci do not yet replicate. Neuronal enrichment is common to many brain-related traits, and the overlap with pain and depression may reflect shared symptom liability.

Linked findings

In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetically defined alternative-pathway subgroup, and proteomic signals that point in different directions across cohorts. Contested
Muscle cells grown from ME/CFS patients failed to activate AMPK and raise glucose uptake after electrical pulse stimulation, and secreted less IL-6, suggesting a defect that persists outside the body. Early
Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal respiration, T-cell glycolysis), higher (lymphoblast maximal respiration with Complex V inefficiency) or normal, depending on lab and cell type. Contested
Conventional structural MRI (grey matter volume, routine clinical reads) shows no consistent abnormality in ME/CFS. Reported regional volume differences rarely replicate across groups. Contested
Markers of neuronal or glial injury (NfL, GFAP, tau, UCH-L1) are not raised in ME/CFS blood or CSF. Several studies
Self-reported light and sound sensitivity is common in ME/CFS. One ERP study found normal early sensory gating (P50) but a reduced P300, pointing to higher-order processing rather than early filtering. Early
Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type, implying most immune findings are downstream consequences rather than inherited causes. Replicated
Genetic instruments for higher EBV DNA load in blood (Kamitaki 2026) are associated with ME/CFS risk in the Million Veteran Program (OR 1.59, p=0.012) but not significantly in DecodeME (OR 1.13, p=0.25). Early
Inherited chromosomally integrated HHV-6 (iciHHV-6, ~1% of people) has been reported at roughly twice the population rate in a neurological/fatigue clinic cohort, but this has not been confirmed in a large ME/CFS sample. Early
Reported genome-wide blood DNA methylation differences have not replicated across cohorts, at least one large set is explained by cell-composition differences, and epigenetic clocks are not accelerated. Contested
ME/CFS genetic risk is enriched in brain tissues and the pituitary, but genetically predicted morning cortisol does not change risk. Early
IBS is about seven times more common in ME/CFS (pooled OR 7.20; prevalence about 37%), and IBS status drives much of the measured microbiome difference. Replicated
ME/CFS genetic risk overlaps with IBS and is enriched in enteric neurons, but genetically predicted gut-derived metabolites (IPA, indolelactate, kynurenine, bile acids) do not change ME/CFS risk. Early
ME/CFS aggregates in families: relatives of cases have about 2-3x the risk, MZ twins are more concordant than DZ twins, and familiality in a 153,803-person Dutch cohort was 42%. Replicated
DecodeME (15,579 PEM-defined cases vs 259,909 controls) identified 8 genome-wide significant loci, each with a small effect (OR about 1.07-1.10), and SNP heritability of about 8-10%. Several studies
The genome-wide DecodeME signal replicates in independent cohorts: genetic correlation 0.65-0.73 with the Million Veteran Program CFS GWAS and 0.80-0.90 with UK Biobank CFS, plus concordant directions for sub-threshold SNPs in FinnGen. Replicated
Genome-wide ME/CFS heritability is enriched in brain tissues and neurons (including glutamatergic synapse genes and subcortical/cerebellar neuron types), with no enrichment in blood immune cells. Several studies
ME/CFS genetics overlaps most with chronic pain, fibromyalgia, IBS and migraine, and substantially with depression, but none of the 8 DecodeME loci share a causal variant with depression. Several studies
Before DecodeME, small GWAS and candidate-gene studies (GRIK2, NPAS2, TPPP, TRA locus and many others) produced no association that replicated in UK Biobank or later cohorts. Refuted
A 2026 UK Biobank GWAS with stricter, multi-source case definitions (1,268 cases) reports 7 loci that replicate in disjoint UK Biobank or All of Us cohorts (near BICD1, GRIN2A, CSMD1, RORA, CLYBL), though none replicate in all three. Early
Proteome-wide cis-MR finds genetically higher plasma EPHB1 (a pain-sensitisation receptor) and CLEC3B (tetranectin) associated with ME/CFS risk in DecodeME, replicating in the independent MVP cohort. Early
Whole-genome sequencing of 464 cases and 611 controls with a deep-learning rare-variant model (HEAL2) proposes 115 loss-of-function-intolerant risk genes, partly overlapping the common-variant synaptic signal. Early
Proprietary combinatorial SNP analyses (PrecisionLife) report thousands of multi-SNP signatures in UK Biobank and DecodeME, with top vs bottom decile of signature count giving OR 1.64. Contested
ME/CFS affects roughly 0.2-1% of adults depending on criteria (US 2021-22: 1.3% self-reported diagnosis; England: 0.16% with a hospital-recorded diagnosis), with women affected 3-4 times as often as men. Replicated

Circulating (serum) factorSomething in patient blood, such as IgG or exosomes, impairs healthy cells. weakened

Several experiments put patient plasma or serum onto healthy cells and observed altered impedance, respiration, mitochondrial morphology or muscle force. If replicated, this would point to a transferable factor and allow bioassays. The best-known muscle mitochondrial version failed a well-powered replication in 2026.

Proponents: Davis/Esfandyarpour (nanoneedle), Fluge and Mella, Prusty (Würzburg), Ramon-Azcon (IBEC)

Evidence for

Plasma-swap nanoneedle impedance (Esfandyarpour 2019, n=20 vs 20)
IgG fragments endothelial mitochondria vs MS and healthy IgG (Prusty 2026)
Long-COVID IgG transfers symptoms to mice (Chen 2026)

Evidence against

Ryback/Ponting 2026: no serum effect on myoblast respiration (n=67 vs 53, d≥0.52 excluded)
IBEC 3D muscle study n=4/5/4 with pseudoreplication risk
Nanoneedle unreplicated in peer review

Predicts

Patient plasma alters healthy cells reproducibly
Effect removed by IgG depletion
Effect disappears with recovery

Tests that would settle it

Multi-lab blinded serum bioassay with disease controls and PEM-timed samples
IgG depletion and rescue experiments

Notes from each body system

Blood flow, vessels and blood: for: Patient plasma lowers endothelial NO; IgG fragments endothelial mitochondria (MS-controlled).; against: Single-lab assays, no cross-lab replication.
Muscle, exertion and energy: for: Fluge 2016 myoblast effect; Mughal 2025 3D muscle force loss (n=4); Prusty 2026 IgG fragments endothelial mitochondria (non-muscle).; against: Ryback 2026 pre-registered replication (67 vs 53) null, excluding the original effect size.
Immune system: for: Patient IgG fragments endothelial mitochondria and alters energetics with no effect from MS or healthy IgG, lost on Fab/Fc cleavage (Liu/Prusty 2026, 106 individuals); long COVID IgG transfer reproduces symptoms in mice; daratumumab responses track falling serum IgG.; against: The flagship serum-factor result (patient serum raising myoblast respiration) failed a preregistered, blinded, well-powered replication (67 vs 53, maximal respiration estimate 2.50 pmol/min, 95% CI -1.62 to 6.62, p=0.23, CI excluding the original d=1.32); no serum-to-cell assay has separated patients from controls in more than one laboratory; removing circulating IgG from patients by immunoadsorption produced no clinical benefit against sham.
Metabolism and blood omics: for: Nanoneedle plasma-swap effect; Fluge 2016 serum effects on myoblasts.; against: No independent replication of the nanoneedle assay.

Linked findings

Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia are reduced in ME/CFS in two independent Norwegian cohorts and by peripheral arterial tonometry in Berlin. Replicated
Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitric oxide production, more mitochondrial fragmentation, and anti-endothelial antibody binding. Early
Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling amino acids and raised PDK mRNA in blood cells; this has not been independently confirmed. Early
The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pre-registered replication of 67 patients vs 53 controls, though a small 3D engineered-muscle study reported weaker contraction with patient serum. Refuted
Plasma-cell depletion with daratumumab produced marked, in some cases sustained, improvement in 6 of 10 patients in an open-label pilot, with low baseline NK count predicting non-response. Early
IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve damage to mice in three independent 2026 studies; no equivalent published ME/CFS transfer experiment exists. Several studies
Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human endothelial cells, an effect absent with multiple-sclerosis and healthy IgG and lost on Fab/Fc cleavage. Early
Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, TCA/urea cycle shifts), but simple blood ratios (lactate/pyruvate, citrulline/arginine) do not differ across pooled cohorts. Contested
Plasma extracellular vesicle (EV) counts may be higher in ME/CFS, but no EV protein or microRNA difference has survived multiple-testing correction in a replicated cohort. Early
A nanoelectronic impedance assay separated 20 ME/CFS from 20 controls perfectly after hyperosmotic stress, but there is no published independent replication. Early
B-cell depletion with rituximab did not improve fatigue or function versus placebo in a 151-person phase III trial (RituxME). Refuted
In an open-label pilot, 6 of 10 women with moderate-to-severe ME/CFS improved markedly after the anti-CD38 antibody daratumumab, and low baseline NK-cell count predicted non-response. Early
Removing antibodies by immunoadsorption was no better than a sham procedure in two controlled trials (IAMPOCO, n=40; IA-PACS-CFS, n=65), although open-label series had reported 50-70% responders. Refuted

Sodium-calcium overload in muscle (Wirth & Scheibenbogen)Under-perfused muscle accumulates sodium and then calcium via the sodium-calcium exchanger, damaging mitochondria and causing PEM. active

The model proposes that beta2-adrenergic dysfunction (perhaps via autoantibodies) and vasoconstriction reduce muscle blood flow. Hypoxic muscle relies on glycolysis; protons are exported by NHE1, loading the cell with sodium, which reverses the sodium-calcium exchanger and raises intracellular calcium, injuring mitochondria. This cycle explains delayed, prolonged PEM and suggests drugs that block NHE1 or reduce sodium loading. Elements are testable with 23Na-MRI.

Proponents: Klaus Wirth, Carmen Scheibenbogen (Charité)

Evidence for

Higher muscle sodium on 23Na-MRI (Petter 2022, 6 vs 6)
Post-exertional muscle damage and mitochondrial drop (Appelman 2024)
Endothelial dysfunction findings

Evidence against

Upstream GPCR-antibody step weakened by negative antigen profiling and failed IA/BC007 trials
Sodium MRI sample tiny
No interventional test yet

Predicts

Raised muscle sodium on 23Na-MRI
Calcium overload and mitochondrial damage after exertion
Benefit from NHE1 inhibitors or vasodilators

Tests that would settle it

Larger 23Na-MRI studies before and after exertion vs deconditioned controls
Pilot trial of NHE1 inhibition

Notes from each body system

Muscle, exertion and energy: for: Higher muscle sodium on 23Na-MRI, inversely correlated with grip (Petter 2022); grip fatigability (Jäkel 2021); review by Scheibenbogen & Wirth 2025.; against: Sodium MRI n=6 vs 6, unreplicated; no direct muscle calcium measurement in patients.

Linked findings

31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower pH recovery after repeated exercise in a subset of patients, but other MRS studies found no consistent group abnormality. Contested
23Na-MRI in 6 women with ME/CFS and 6 controls found higher resting muscle sodium in all lower-leg compartments and a larger sodium rise after exercise, inversely related to grip strength. Early
Repeated maximal handgrip testing shows lower strength, faster decline across repetitions and poorer recovery after 60 minutes in ME/CFS, correlating with severity and PEM. Several studies
Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been reported repeatedly, including a two-site study (36 vs 42), but only ever by one research group. Early
A nanoelectronic impedance assay separated 20 ME/CFS from 20 controls perfectly after hyperosmotic stress, but there is no published independent replication. Early
TRPM3 ion channel dysfunction in NK cells and TRPM3 SNPs are reported repeatedly by one research group, but the SNP studies are tiny and no TRPM3 association appears in GWAS. Contested

Mast cell activationInappropriate mast-cell mediator release causes flushing, tachycardia, GI and cognitive symptoms. speculative

Mast cell activation syndrome overlaps clinically with POTS, hypermobility and ME/CFS. Histamine and other mediators could cause vasodilation and orthostatic symptoms. Treatment with H1/H2 antihistamines is common. Diagnostic criteria are contested and data are observational.

Proponents: Afrin, Weinstock, Glynne (long COVID)

Evidence for

Observational improvement with H1+H2 blockers in long COVID (Glynne 2022)
Clinical overlap with POTS/hEDS

Evidence against

No RCTs; mediator tests rarely positive
Diagnostic criteria inconsistent

Predicts

Raised tryptase/histamine metabolites during flares
Response to mast-cell stabilisers

Tests that would settle it

RCT of antihistamines with mediator-defined selection

Notes from each body system

Immune system: for: Up to 25.3% of ME/CFS patients meet criteria for clinically relevant mast cell activation, with prevalence rising over the disease course and a significant association with orthostatic intolerance and POTS; those patients report more benefit from mast-cell-directed treatment; small flow-cytometry studies find altered mast cell progenitor phenotypes; CSF eotaxin (an eosinophil/allergic chemokine) elevated.; against: Almost all the human evidence is self-reported or retrospective clinic data with no controls; MCAS criteria are themselves contested; no replicated objective mediator (tryptase, urinary metabolites) finding exists in ME/CFS; blood proteomics across four cohorts shows no allergic or mast-cell signature; no controlled trial of mast-cell stabilisers in ME/CFS.

Linked findings

In the largest single cytokine study, 17 cytokines rose in step with disease severity although only TGF-beta (up) and resistin (down) differed between all patients and controls. Several studies
Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and disturbed IL-1 signalling, but CSF studies are small and partly contradicted by later proteomics. Early
Clinically relevant mast cell activation is reported in up to a quarter of ME/CFS patients, associates with orthostatic intolerance and POTS, and those patients report more benefit from mast-cell-directed treatment. Early

Connective tissue / hypermobilityHypermobility and connective-tissue weakness predispose to venous pooling, instability and dysautonomia. speculative

Joint hypermobility is more common in ME/CFS and POTS. Lax connective tissue in veins could worsen pooling, and in the neck could contribute to craniocervical problems. Lower blood levels of vessel-wall/extracellular-matrix proteins have been reported.

Proponents: Rowe, Eccles

Evidence for

Hypermobility overrepresented in clinic samples
Lower ITGAV/NOTCH3/TGF-beta set proteins (Solving ME/CFS re-analysis)

Evidence against

Clinic-based selection; population data lacking
No mechanism-specific treatment

Predicts

Higher hypermobility prevalence
Matrix protein changes
Venous compliance increased

Tests that would settle it

Population-based hypermobility prevalence in ME/CFS
Venous compliance measurement

Notes from each body system

Nervous system and brain: for: Hypermobility in 50% of a severe-clinic series; hEDS cohorts show more severe neurodegeneration and share the autonomic phenotype.; against: Clinic series are uncontrolled; overlap may reflect referral patterns.
Metabolism and blood omics: for: Higher collagen breakdown products; lower ECM/fibroblast and mural proteins.; against: Re-analysis only; diet/supplements not excluded.
Genetics: for: Hypermobility and EDS are enriched clinically and in registries, collagen breakdown markers are raised, and hypermobility is a risk factor for long COVID with PEM.; against: No connective-tissue genes appear among GWAS loci, the hEDS gene is unknown, and the studies are mostly clinic- or survey-based with selection bias.

Linked findings

Pericytes and vascular smooth-muscle cells have not been examined directly in ME/CFS tissue, but blood proteins from mural cells and vessel matrix (NOTCH3, ITGAV, TGF-beta set, LUM) are lower across several ME/CFS cohorts and in long COVID. Early
Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients from specialist referral cohorts, but the controlled NIH study found no difference from healthy volunteers. Contested
Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying tonsils and signs of raised intracranial pressure in severe ME/CFS, and surgical case reports describe improvement. There are no controlled imaging studies or trials. Early
A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficiency, with IgG3 and IgG4 the commonest phenotypes, and registry data show strong enrichment of coded antibody deficiency. Several studies
Collagen breakdown products (hydroxyproline, galactosyl-hydroxylysine and related) are higher in ME/CFS plasma (+0.52 SD, 4/4 Metabolon cohorts; hydroxyproline also up on 3 other platforms). Early
Most of the ~845 FDR-significant serum proteins in Hoel 2026 appear driven by a sample-handling (cell-leak) signal that is absent from other cohorts; about 13 survive adjustment. Contested
Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4 in 60% of children with CFS vs 24% of controls; 50% in a Swedish severe cohort; EDS about 100x expected in Finnish registries). Several studies

Cornell integrated multi-omics model (Hanson)PEM reflects a maladaptive multi-system response to exertion visible across immune, metabolic and proteomic layers. active

Using 2-day CPET with blood sampling, the Cornell group reports that patients fail to mount and resolve normal exercise responses in plasma proteins, metabolites and immune cells, with CD8 T-cell exhaustion features. They frame ME/CFS as an integrated failure of recovery across systems rather than a single lesion. The model is descriptive and generates leads rather than one causal claim.

Proponents: Maureen Hanson, Arnaud Germain, Cornell ME/CFS Collaborative Research Center

Evidence for

Plasma proteome maladaptation after exertion in 79 vs 53 (Germain 2025, Mol Cell Proteomics)
CD8 exhaustion priming (Iu 2024)

Evidence against

Re-analysis of the 2022 metabolomics: within-person exercise responses barely differ; 'growing divergence' does not hold (Solving ME/CFS re-analysis, PEM)
Clean negative autoantibody profiling from the same group (Germain 2025) narrows immune explanations

Predicts

Divergent post-exertion trajectories vs sedentary controls
Changes larger on day 2

Tests that would settle it

Pre-registered within-person analyses in a new 2-day CPET cohort
Matching deconditioned and disease controls

Open Medicine Foundation models (Davis, Xiao, Systrom)A cluster of OMF-funded models: a circulating plasma factor, the IDO metabolic trap, and neurovascular preload failure. active

OMF has backed several ideas rather than one: plasma-swap nanoneedle evidence for a blood factor; the IDO metabolic trap; and Systrom's neurovascular dysregulation with preload failure, tested by pyridostigmine. Its data centre and the 2025 treatment survey aim to link subgroups to treatments. The common thread is a measurable, possibly reversible physiological state.

Proponents: Ron Davis, Wenzhong Xiao, David Systrom, Robert Phair

Evidence for

Nanoneedle plasma-swap (Esfandyarpour 2019)
Pyridostigmine single-dose iCPET RCT (Joseph 2022)
Patient survey subgroups (Eckey 2025)

Evidence against

No peer-reviewed nanoneedle replication
IDO trap not supported by GWAS/MR
Longer pyridostigmine trial in long COVID negative (2025 abstract)

Predicts

Plasma factor in bioassays
IDO2 variants/trap state
Pyridostigmine benefit in preload failure

Tests that would settle it

LIFT trial results
Independent replication of nanoneedle assay

Multisystem post-infectious model (Komaroff & Lipkin reviews)ME/CFS and long COVID share a set of biological abnormalities across brain, immune, energy and vascular systems, triggered by infection. active

Two influential reviews catalogue overlapping findings: neuroinflammation, immune activation and exhaustion, autoantibodies, impaired energy metabolism, endothelial and autonomic dysfunction, and gut dysbiosis. They argue the two illnesses are variants of the same post-infectious process and that research on each informs the other. It is a synthesis, not a single falsifiable mechanism.

Proponents: Anthony Komaroff (Harvard), W. Ian Lipkin (Columbia)

Evidence for

Treatment ratings correlate between ME/CFS and long COVID (R²=0.68; Eckey 2025)
Post-COVID ME/CFS incidence data (Vernon 2025)

Evidence against

Many catalogued abnormalities are unreplicated or contested individually
Breadth makes the model hard to falsify

Predicts

Shared abnormalities between ME/CFS and long COVID
Similar treatment responses

Tests that would settle it

Head-to-head cohorts with identical assays in ME/CFS, long COVID and recovered controls

Altered 'effort preference' (NIH)Patients choose fewer hard-effort options in a decision task, which the NIH team interpreted as a defining feature rather than fatigue or weakness. weakened

In deep phenotyping of 17 post-infectious ME/CFS patients, participants chose the hard task less often in the Effort-Expenditure for Rewards Task, and the authors proposed altered effort preference linked to low CSF catecholamines and motor-network changes. They framed fatigue as arising from brain integrative circuits. Critics argue the task was not validated in ME/CFS and that choosing less effort is a rational response to PEM, not a preference.

Proponents: Walitt, Nath and the NIH intramural team

Evidence for

Fewer hard-task choices (Walitt 2024, n=17 vs 21)
Low CSF catecholamines and altered motor-cortex activity in the same study

Evidence against

Tiny sample; task not validated; PEM-avoidance alternative explanation
Patients and many researchers objected to the framing; findings not replicated

Predicts

Reduced hard-task choices independent of capacity
Brain activity differences during effort tasks

Tests that would settle it

Replication with tasks that separate anticipated PEM cost from motivation
Compare with other disabling illnesses with exertional limits

Psychosomatic / cognitive-behavioural (fear-avoidance) modelAn initial illness is perpetuated by unhelpful beliefs, activity avoidance and deconditioning, so CBT and graded exercise should reverse it. largely refuted

The model holds that after a trigger, beliefs that activity is harmful lead to avoidance, deconditioning and symptom focus, maintaining disability. It predicted that CBT and graded exercise would produce recovery. PACE and Dutch trials reported modest self-reported benefits. The model is now widely disputed: objective outcomes were null and PEM is not explained by beliefs.

Proponents: Wessely, Sharpe, Chalder, White (UK), Knoop and Bleijenberg (Netherlands)

Evidence for

PACE: modest self-reported fatigue and function gains with CBT/GET at 52 weeks (White 2011)
Belief/attribution associated with prognosis in older cohorts (Cairns & Hotopf 2005)
Large placebo responses show context and expectation affect symptom reports

Evidence against

Effects confined to self-report and not durable at 2.5 years (Sharpe 2015; Wilshire 2018)
NICE 2021 rated evidence low/very low and advised against GET
Patients rate GET most harmful (Eckey 2025); rehab worsened 42/97 (CFS_CARE 2026)
Reproducible biological differences (2-day CPET, genetics, muscle pathology) not explained by beliefs
Critique of the model's evidence base (Geraghty 2019)

Predicts

CBT/GET produce recovery including on objective measures
Beliefs predict outcome
No biological abnormality beyond deconditioning

Tests that would settle it

Blinded-assessor trial with objective endpoints in PEM-defined patients
Prospective test of whether beliefs precede and predict onset

DeconditioningExercise intolerance is simply loss of fitness from inactivity. largely refuted

If symptoms reflect detraining, patients should look like sedentary controls on exercise testing, recover normally between tests, and improve with graded training. Deconditioning certainly coexists in severe illness. The question is whether it explains the disease.

Proponents: Implicit in rehabilitation models; early long-COVID exercise studies

Evidence for

Lower peak VO2 in patients is partly explained by fitness
Some long-COVID CPET studies attributed limitation to deconditioning

Evidence against

Day-2 CPET decline in many patients, not seen in sedentary controls (Keller 2024, n=84; memory)
Post-exertional muscle damage and mitochondrial drop not explained by fitness (Appelman 2024)
iCPET shows preload failure and poor extraction atypical of detraining (Joseph 2021, 2023 review)
Lower muscle proteins in blood not explained by steps or lying time (Solving ME/CFS re-analysis)
GET harmful in surveys

Predicts

Normal reproducibility on repeat CPET
Muscle findings match sedentary controls
Training improves capacity

Tests that would settle it

Matched-activity controls (bedrest or immobilised volunteers) on 2-day CPET and biopsy

Notes from each body system

Muscle, exertion and energy: for: (New id: exercise and muscle abnormalities are secondary to inactivity.) Low peak VO2 overlaps with sedentary controls; lower muscle proteins in blood could reflect reduced mass; Mancini 2026 found no day-2 decline.; against: VO2-matched pairs still declined on day 2 (Keller 2024); bed rest does not reproduce fibre-type shift, capillary loss or OXPHOS uncoupling (Charlton 2026); disease controls improve on day 2 (van Campen 2021, Hodges 2018).

Linked findings

On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled about 5 ml/kg/min below controls) with a low peak heart rate (chronotropic incompetence). Replicated
Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without the fibre atrophy and type 2 shift expected from inactivity, with abnormalities confined to subgroups. Contested
ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy, lower mitochondrial capacity and fewer capillaries per fibre, a pattern that 60 days of strict bed rest in healthy people did not reproduce. Early
Muscle-derived proteins (myoglobin, troponins, CA3 and similar) are lower in ME/CFS blood at rest across three SomaScan cohorts and UK Biobank, not explained by step counts or walking pace. Several studies
Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison groups, and a 60-day bed-rest model) indicate that deconditioning alone does not explain the exercise and muscle abnormalities in ME/CFS. Several studies

Findings by body system

Search findings

Blood flow, vessels and blood

Blood volume

Red cell volume and total blood volume are below predicted in a large minority to about half of ME/CFS patients measured by isotope or CO-rebreathing methods. Replicated

Many patients have less blood than expected for their size, mainly because they have fewer red cells. Less blood means less returns to the heart when upright.

Hurwitz 2009 (Clin Sci): total blood volume -15%, plasma volume -13%, red cell volume -19% in CFS, and the blood-volume deficit explained 91-94% of group differences in cardiac volume indices. Newton 2016 linked lower plasma volume to smaller cardiac MRI volumes and to fatigue severity but not to illness duration (argues against pure deconditioning). Five independent groups report a red-cell-volume deficit (Streeten 2000, Hurwitz, Newton, van Campen 2018, Yamazaki 2026). Measurement methods differ (I-131 albumin, Cr-51, CO-rebreathing) and predicted values depend on body-size formulas, so prevalence figures are not comparable across studies.

Evidence

Study	n	Design	Result
Hurwitz 2009, Clin Sci	56 CFS (30 severe, 26 non-severe) vs 90 controls	case-control, isotope blood volume + echo	TBV -15.4%, PV -13.2%, RBCV -19.1%; deficit explained most of the lower stroke index
Newton 2016, Open Heart	CFS vs controls (cardiac MRI + red cell mass)	case-control	Lower cardiac volumes tracked plasma and red cell volume; PV correlated with fatigue severity; no link to illness duration
Yamazaki 2026, Physiol Rep	49 ME/CFS	cross-sectional, blood volume + 10-min lean test	RBC volume deficient in 49%; TBV hypovolaemic 35%, normal 43%, hypervolaemic 22%
Streeten & Bell 2000, Am J Med Sci unverified	15 CFS	case series, Cr-51 red cell mass	Low red cell mass in most patients; orthostatic symptoms

Against or not replicated

Study	n	Design	Result
Yamazaki 2026, Physiol Rep	49 ME/CFS	cross-sectional	Blood-volume category did not predict any lean-test abnormality (POTS, hypocapnia, hypotension); 22% were hypervolaemic

Open questions

Why does red cell mass fall: reduced marrow output (Solving ME/CFS re-analysis of FinnGen labs found lower reticulocytes and soluble transferrin receptor with normal iron), inactivity, or altered EPO signalling?
Does blood volume change with severity or after crashes?
If low blood volume does not predict orthostatic intolerance (Yamazaki 2026), how much of the symptom burden does it explain?

Bears on: Low blood volume and preload failure Pituitary / vasopressin deficiency · Test: Haemoglobin mass by CO-rebreathing (some sports-science and research labs); isotope blood volume (Daxor BVA-100, a few US centres). Reticulocytes, ferritin, transferrin saturation are routine.

The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not raised in ME/CFS or POTS despite low volume, and blood proteomics shows no consistent change in them. Several studies

If blood volume is low the body should release hormones to retain salt and water. In many patients these hormones stay normal or low, which suggests the volume set-point itself is off.

In POTS, low aldosterone and an inappropriately low renin for the plasma-volume deficit (the renin-aldosterone paradox) were described by Raj et al. 2005; adrenal responsiveness to ACTH is intact (Laurin 2023), so the defect sits upstream. Older ME/CFS work reported low renin in tilt-positive patients and normal EPO. A Solving ME/CFS re-analysis of SomaScan and UK Biobank Olink data (about 150 patients plus UKB) found renin, ACE, BNP, EPO and hepcidin flat across cohorts.

Evidence

Study	n	Design	Result
Raj 2005, Circulation unverified	POTS vs controls	case-control	Plasma volume deficit with low renin and aldosterone for the degree of hypovolaemia
Laurin 2023, Auton Neurosci	POTS vs HC	ACTH stimulation	Adrenal aldosterone and cortisol responses to ACTH intact in POTS
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	Hoel, Germain, Walitt SomaScan + UKB Olink	secondary analysis	RAAS/natriuretic, EPO, hepcidin sets null (q>0.7)

Open questions

Is the low-renin pattern present in ME/CFS without POTS?
Does ADH/copeptin respond normally to upright posture?

Bears on: Low blood volume and preload failure Pituitary / vasopressin deficiency HPA axis blunting · Test: Supine and upright renin/aldosterone; copeptin. Routine labs, but interpretation needs posture and salt-intake control.

Cardiac size and stroke volume

Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more than normal on tilt, largely in proportion to low blood volume rather than to weak heart muscle. Replicated

The heart is not damaged, but it is under-filled, so each beat pumps less blood, especially when upright.

Hurwitz 2009: severe CFS had about 10% lower stroke index and end-diastolic volume; adjusting for blood volume removed most of the difference. Hollingsworth 2012 (cardiac MRI) reported markedly lower end-diastolic volume and stroke volume. Miwa reports a 'small heart' (cardiothoracic ratio <0.42) subgroup with orthostatic intolerance. van Campen 2024 (233 POTS, many more normal HR-BP response) showed larger stroke-volume-index falls on tilt in all ME/CFS groups than in controls, with two patterns: venous pooling (HR rise 30-39) and a hyperadrenergic pattern (HR rise >=40) with relatively preserved SVI.

Evidence

Study	n	Design	Result
van Campen 2024, J Clin Med	233 ME/CFS with POTS + normal-response patients vs HC	retrospective tilt with echocardiographic SVI	SVI fall larger than HC in all groups; two POTS haemodynamic profiles
Hollingsworth 2012, Eur J Clin Invest unverified	CFS vs controls	cardiac MRI	Lower LV end-diastolic volume and stroke volume
Miwa 2026, Front Med	review of Japanese series	narrative review	Small heart and downregulated renin-aldosterone/ADH systems in OI subgroup

Against or not replicated

Study	n	Design	Result
Olimulder 2016, Neth Heart J unverified	ME/CFS vs controls	cardiac MRI	Cardiac dimensions within normal range after indexing in a referral cohort

Open questions

Do stroke-volume deficits normalise after volume loading, and does that predict symptomatic benefit?
How much of the van Campen echo data has been reproduced in another centre with blinded readers?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Echocardiogram (routine, but usually reported normal at rest); tilt with stroke-volume monitoring (Finapres/bioimpedance at autonomic units).

Exercise haemodynamics

During upright invasive exercise testing most ME/CFS patients have abnormally low right-atrial filling pressure (preload failure), and a high-flow subgroup shows poor systemic oxygen extraction. Several studies

When patients exercise, too little blood returns to the heart. In some, blood also bypasses the muscles, so oxygen goes unused.

Joseph/Systrom 2021 (Chest): of 1,516 iCPETs, 160 met ME/CFS criteria with RAP <6.5 mmHg; peak RAP 1.9 vs 8.3 mmHg, peak VO2 80% vs 101% predicted. Low-flow tertile had lower peak cardiac output; high-flow tertile had lower systemic O2 extraction (0.74 vs 0.86-0.88). Selection by low RAP is built into the cohort, so the prevalence of preload failure in ME/CFS generally is not established. Novak 2026 reports preload failure in 92% of the 66 ME/CFS patients who had iCPET (same Harvard group). Long COVID iCPET shows the same pattern (Singh 2022; Joseph 2023 review). Independent replication outside the Systrom network is lacking.

Evidence

Study	n	Design	Result
Joseph 2021, Chest	160 ME/CFS vs 36 controls	retrospective iCPET + skin biopsy	Peak RAP 1.9 vs 8.3 mmHg; impaired O2 extraction in high-flow tertile; SFN in 31%
Novak 2026, PLOS One	170 ME/CFS, 143 long COVID, 73 HC, 290 hEDS	retrospective autonomic + iCPET	Preload failure 92% ME/CFS (of 66 tested), 96% LC (of 25)
Joseph 2022, Chest (pyridostigmine RCT)	45 ME/CFS	RCT, single-dose crossover-style iCPET	Pyridostigmine improved peak VO2 vs placebo via higher cardiac output (acute effect)

Open questions

What is the preload-failure prevalence in an unselected ME/CFS cohort?
Can non-invasive tests (pulse pressure narrowing on stand, Day 2023; CO-rebreathing cardiac output) stand in for iCPET?
Does preload failure precede or follow illness onset?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy Capillary and microvascular damage · Test: Invasive CPET at a few US centres (Brigham, Yale, Mount Sinai); not routine. Two-day CPET is non-invasive but does not measure preload.

A subset of patients with unexplained exertional intolerance cannot extract oxygen normally from blood during exercise, associated with hypocapnia and possible arteriovenous shunting. Early

Blood arrives at the muscles but leaves with too much oxygen still in it, so the muscles get less than they need.

Melamed 2019 (Systrom group) described a poor-extraction phenotype with relative hypocapnia (PaCO2 29.9 vs 37.5 mmHg) and high mixed-venous O2 saturation at peak (44% vs 31%). Proposed causes: microvascular shunting from small-fibre neuropathy, mitochondrial limitation, or a left-shifted oxyhaemoglobin curve from hyperventilation. Muscle-oxygenation work by the Wüst group (NIRS post-occlusion) points in the same direction. All iCPET data come from one network.

Evidence

Study	n	Design	Result
Melamed 2019, Eur J Appl Physiol	unexplained exertional intolerance cohort	retrospective iCPET	Poor systemic O2 extraction with hypocapnia; high peak SvO2
Joseph 2021, Chest	160 ME/CFS	iCPET	High-flow tertile extraction 0.74 vs 0.86-0.88

Open questions

Which contributes most: shunting, mitochondria, or CO2-related Bohr shift?
Is impaired extraction present at rest or only at peak?

Bears on: Capillary and microvascular damage Autonomic dysfunction and small-fibre neuropathy Mitochondrial / energy metabolism impairment · Test: Research only (iCPET); muscle NIRS in research labs.

Cerebral blood flow

Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, including in patients with normal heart rate and blood pressure responses. Several studies

Standing up reduces blood flow to the brain far more than normal, even when pulse and blood pressure look fine, and the drop tracks orthostatic symptoms.

van Campen 2020 (429 patients, 44 HC, extracranial carotid+vertebral Doppler): 82% of normal-response, 98% of delayed-OH and 100% of POTS patients exceeded the control lower limit (13%). Flow recovery after tilt is incomplete and slowest in severe patients (van Campen 2021). The CO-CBF relation is roughly 1:1 in most patients, unlike the 3:1 seen in health (van Campen 2024, 2025). Transcranial Doppler work from other groups (Medow/Stewart; Novak 2026: reduced orthostatic CBF velocity in 88% ME/CFS) agrees. Caveats: most volume-flow data come from one Dutch clinic, operators are not blinded to diagnosis, controls are few relative to patients, and the 2025 POTS paper found CBF reduction did not differ between its two haemodynamic subgroups.

Evidence

Study	n	Design	Result
van Campen 2020, Clin Neurophysiol Pract	429 ME/CFS vs 44 HC	case-control, tilt with extracranial Doppler	End-tilt CBF -26% vs -7%; ~90% abnormal
van Campen 2021, Clin Neurophysiol Pract	ME/CFS patients	tilt with post-tilt recovery	CBF reduced after tilt, slowest recovery in most severe
Novak 2026, PLOS One	170 ME/CFS, 143 LC, 73 HC	retrospective TCD tilt	Reduced orthostatic CBFv in 88% ME/CFS, 92% LC
van Campen 2025, J Clin Med	ME/CFS with POTS	retrospective tilt CO + CBF	Two-thirds ~1:1 CO-CBF reduction; one-third limited CO fall with high HR (hyperadrenergic)

Open questions

Independent replication of extracranial volume-flow method with blinded readers
Does CBF reduction predict response to fluids, compression, or vasoconstrictors?
What happens to CBF in sitting (the posture patients actually spend time in)?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy Endothelial dysfunction · Test: Extracranial Doppler on tilt (Stichting CardioZorg, Netherlands); transcranial Doppler on tilt at some autonomic labs.

Raising blood pressure with phenylephrine during tilt blocked hyperventilation, restored cerebral blood velocity and improved working memory in ME/CFS, whereas CO2 or acetazolamide alone did not. Early

In a lab test, a drug that tightens blood vessels improved both brain blood flow and thinking speed when patients were upright.

Medow 2014 and Medow & Stewart 2024 (same New York group) used acute infusions with n-back testing; numbers are small (single-digit to low teens per condition). These are mechanistic studies, not treatment trials. Compression of legs and abdomen reduced cardiac output fall (-15% vs -27%) and CBF fall (-14% vs -25%) on tilt in van Campen 2021 (n=16).

Evidence

Study	n	Design	Result
Medow & Stewart 2024, Am J Physiol Regul	ME/CFS with OI vs controls	acute intervention crossover	Only phenylephrine improved upright n-back and phase synchronisation
van Campen 2021, Front Neurol (compression) unverified	16 ME/CFS	within-subject tilt	Compression reduced CO and CBF falls

Open questions

Do oral analogues (midodrine) reproduce the cognitive effect in daily life?
Placebo-controlled trials of compression garments in ME/CFS

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Research only; midodrine and compression are used clinically for orthostatic intolerance.

Breathing and CO2

Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exercise hyperventilation and dysfunctional breathing are common. Replicated

When standing, some patients breathe off too much CO2. Low CO2 narrows brain arteries, which can add to dizziness and brain fog.

End-tidal CO2 falls on tilt in several groups (Naschitz 2000, Razumovsky 2003, Natelson 2007, Medow 2014, Novak's 'hypocapnic cerebral hypoperfusion'). Yamazaki 2026: postural orthostatic syndrome of hypocapnia (POSH) in 41% of 49 on a lean test. Mancini 2025: hyperventilation and/or dysfunctional breathing during CPET common, with VE/VCO2 34.7 vs 28.1 in hyperventilators. But van Campen 2023 found CO2 reactivity normal and the PETCO2 fall explained only a modest share of the CBF fall, so hypocapnia is probably a contributor, not the main driver. Whether over-breathing is primary (chemoreflex/autonomic) or compensatory (e.g. to aid venous return via thoracic pump) is unresolved.

Evidence

Study	n	Design	Result
Medow 2014, J Appl Physiol	CFS with OI vs controls	tilt + n-back	Upright hyperventilation and CBFv fall; phenylephrine restored n-back performance
Yamazaki 2026, Physiol Rep	49 ME/CFS	lean test with capnography	POSH 40.8%, most common orthostatic abnormality
Mancini 2025, Front Med	58 ME/CFS + sedentary controls	CPET breathing analysis	18 hyperventilators; elevated VE/VCO2 or dysfunctional breathing in 83% of them

Against or not replicated

Study	n	Design	Result
van Campen 2023, Physiol Rep	female ME/CFS vs HC	tilt with PETCO2 and CBF	CO2 reactivity normal; PETCO2 explains only a modest part of CBF fall

Open questions

Is upright hypocapnia a cause, a compensation, or a marker of sympathetic activation?
Does breathing retraining or added CO2 change orthostatic symptoms?

Bears on: Autonomic dysfunction and small-fibre neuropathy Low blood volume and preload failure · Test: Capnography during a 10-minute stand or tilt (cheap; available in some respiratory/autonomic labs); CPET with breathing-pattern analysis.

Endothelium

Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia are reduced in ME/CFS in two independent Norwegian cohorts and by peripheral arterial tonometry in Berlin. Replicated

The lining of blood vessels does not widen properly when it should. This has been found in several groups with different methods.

Sandvik 2023 (RituxME substudy, 39 patients): FMD 5.1% vs 8.2% (p<0.0001, adjusted for diameter and sex), PORH 1354 vs 2208 PU; rituximab had no effect; PORH but not FMD improved over 18 months and PORH correlated with daily steps. Sørland 2021 (cyclophosphamide cohort, 35 patients): FMD 5.9% vs 7.7%, PORH 1331 vs 1886 PU. Scherbakov 2020 (EndoPAT): endothelial dysfunction (RHI <=1.81) in 51% of 35 vs 20% of 20 controls, linked to severity. Haffke 2022: reduced RHI in 5/14 post-COVID ME/CFS. Caveats: inactivity lowers FMD, controls were mostly not activity-matched, and FMD and PORH did not correlate with each other.

Evidence

Study	n	Design	Result
Sandvik 2023, PLOS One	39 ME/CFS vs healthy controls	case-control within RCT	FMD 5.1% vs 8.2%; PORH 1354 vs 2208 PU
Sørland 2021, Front Med	35 ME/CFS vs 66 HC	case-control within open-label trial	FMD 5.9% vs 7.7%; PORH lower; no change with clinical response
Scherbakov 2020, ESC Heart Fail	35 ME/CFS vs 20 HC	case-control, EndoPAT	ED in 51% vs 20%; associated with severity
Haffke 2022, J Transl Med	14 PCS-ME/CFS, 16 PCS, HC	case-control, EndoPAT + biomarkers	ED in 5/14 and 5/16, none of HC; ET-1 raised

Open questions

Is endothelial dysfunction cause, consequence of inactivity, or a marker of an immune/serum factor?
Does it track PEM or orthostatic severity over time?
Activity-matched control studies

Bears on: Endothelial dysfunction Circulating (serum) factor Autoimmunity (functional autoantibodies) · Test: FMD ultrasound and EndoPAT in vascular research labs; not standard clinical tests.

Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitric oxide production, more mitochondrial fragmentation, and anti-endothelial antibody binding. Early

Something in patients' blood seems to make lab-grown vessel-lining cells work less well. This has been seen by three labs using different tests, none yet repeated by another lab using the same test.

Bertinat 2022: HUVECs exposed to ME/CFS plasma (11 vs 12) produced less NO with higher inhibitory eNOS Thr495 phosphorylation. Flaskamp 2022 (Charité): more anti-endothelial antibody binding in post-COVID ME/CFS, and sera lacked the pro-angiogenic effect seen in non-ME post-COVID sera. Prusty 2026 (BBIH): purified IgG from 39 ME/CFS and 15 post-COVID ME/CFS fragmented endothelial mitochondria versus MS and healthy IgG; single lab, nominal p-values over many readouts (summarised in Solving ME/CFS serum-factors review). No lab has reproduced another's assay.

Evidence

Study	n	Design	Result
Bertinat 2022, Vascul Pharmacol	11 ME/CFS vs 12 HC plasma	in vitro HUVEC	Lower NO production, higher eNOS Thr495 phosphorylation
Flaskamp 2022, Cells	PCS-ME/CFS, PCS, HC sera	in vitro HMEC-1	Higher AECA binding in PCS/CFS; no pro-angiogenic effect
Liu/Prusty 2026, Brain Behav Immun Health unverified	39 ME, 15 PCS-ME, 20 MS, 41 HC IgG	in vitro	Endothelial mitochondrial fragmentation with ME IgG, not MS or HC IgG

Open questions

Blinded multi-lab exchange of the same coded sera
What is the antigen, and is the effect IgG-dependent in every assay?

Bears on: Circulating (serum) factor Autoimmunity (functional autoantibodies) Endothelial dysfunction · Test: Research only.

Microvasculature

Retinal imaging shows narrower arterioles, lower arteriolar-to-venular ratio, reduced flicker-induced venular dilation and lower capillary vessel density in post-COVID patients, most pronounced in those meeting ME/CFS criteria. Several studies

The tiny vessels at the back of the eye, which can be photographed without needles, look narrower and respond less in post-COVID ME/CFS.

Wallraven 2026 (BMC Med, 'All Eyes on PCS'): 102 PCS (62 with ME/CFS) vs 102 recovered and 96 never-infected; CRAE 178 vs 186 um, AVR 0.83 vs 0.87, venular flicker dilation 3.7% vs 4.8%; ME/CFS subgroup most affected (AUC 0.79); robust to CV-risk adjustment except AVR attenuated. Schlick 2022 (Erlangen, OCT-A): lower vessel density in superficial and intermediate capillary plexus in PCS, lower still with chronic fatigue. Data are almost entirely post-COVID; pre-2020 ME/CFS cohorts have not been imaged with these methods.

Evidence

Study	n	Design	Result
Wallraven 2026, BMC Med	102 PCS (62 ME/CFS), 102 recovered, 96 never-infected	prospective case-control, retinal vessel analysis	Narrower arterioles, lower AVR and flicker dilation; ME/CFS most affected; AVR tied to IL-6, ICAM-1, VCAM-1
Schlick 2022, Int J Mol Sci	PCS vs controls	case-control, OCT-A	Lower retinal capillary vessel density; worse with chronic fatigue

Open questions

Is the same seen in pre-pandemic, non-COVID ME/CFS?
Do retinal measures change with PEM or with treatment?

Bears on: Capillary and microvascular damage Endothelial dysfunction Innate immune hyperactivation · Test: OCT-A at ophthalmology clinics (widely available, though reference values for vessel density vary by device); dynamic vessel analysis research only.

Sublingual video-microscopy shows reduced perfused capillary density and glycocalyx changes in long COVID, persisting many months after infection. Early

Under the tongue, fewer small vessels carry blood in long COVID patients.

Osiaevi/Rovas/Kümpers (Münster) used sideStream dark-field microscopy with automated analysis (GlycoCheck) and reported persistent capillary rarefaction in long COVID (Nephrol Dial Transplant 2024; Angiogenesis 2023). No ME/CFS-specific cohort has been published with this method; whether rarefaction correlates with ME/CFS-defined PEM is unknown.

Evidence

Study	n	Design	Result
Osiaevi 2024, Nephrol Dial Transplant	long COVID vs controls	case-control, sublingual video-microscopy	Capillary rarefaction proposed as possible cause of long COVID

Open questions

Apply sublingual microscopy to pre-pandemic ME/CFS
Does capillary density recover over time?

Bears on: Capillary and microvascular damage Endothelial dysfunction · Test: Research only (GlycoCheck/SDF microscopy).

Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per fibre size in both ME/CFS and long COVID, unlike the pattern after 60 days of bed rest; thickened capillary basement membrane has been reported but is not yet peer-reviewed. Early

Muscles of patients have fewer and possibly thicker-walled small blood vessels, and this is different from what happens to healthy people kept in bed.

Charlton 2026 (Nat Commun, Wüst group): ME/CFS (n=23-24) had lower capillary-to-fibre ratio than controls (p<0.001, n=30) and long COVID; both patient groups had lower capillary density and lower capillary supply for a given fibre area; healthy bed rest (n=17-24) instead raised capillary density because fibres shrank. This argues against inactivity as the full explanation. Slaghekke 2025 (conference poster, electron microscopy, about 11 patients per group) reported basement membrane thickening (p<0.0001) and smaller capillary lumens in ME/CFS and post-COVID; the composite CF x collagen IV measure that correlated with CPET has been questioned. Appelman 2024 found amyloid-containing deposits in long COVID muscle that increased after induced PEM.

Evidence

Study	n	Design	Result
Charlton 2026, Nat Commun	~25 LC, ~24 ME/CFS, 30 HC, 24 bed rest	case-control + bed-rest comparison, vastus lateralis biopsy	ME/CFS lower capillary-to-fibre ratio and density; patterns differ from bed rest
Slaghekke 2025, ME/CFS Research Conference poster unverified	~11 per group	electron microscopy	Thicker capillary basement membrane, smaller lumen
Appelman 2024, Nat Commun data	25 LC vs 21 recovered	longitudinal case-control, biopsy pre/post exercise	Amyloid-containing deposits and myopathy worse after PEM

Open questions

Independent replication of basement membrane thickening with blinded EM
Is thickening caused by repeated ischaemia, immune deposition, or pericyte loss?
Severe and housebound patients have not been biopsied

Bears on: Capillary and microvascular damage Endothelial dysfunction Mitochondrial / energy metabolism impairment · Test: Research only (needle biopsy at a research centre).

Pericytes and vascular smooth-muscle cells have not been examined directly in ME/CFS tissue, but blood proteins from mural cells and vessel matrix (NOTCH3, ITGAV, TGF-beta set, LUM) are lower across several ME/CFS cohorts and in long COVID. Early

Pericytes are the cells wrapped around capillaries that control their width. Nobody has looked at them in ME/CFS tissue yet; indirect blood clues suggest they may be fewer or less active.

Solving ME/CFS re-analysis of SomaScan (Hoel, Germain, Walitt) and UK Biobank Olink data: ITGAV d -0.34 (UKB q=0.002), NOTCH3 -0.32, THBS1 -0.50, LUM -0.92; vascular smooth-muscle tissue score d -0.48 (p=0.001) but flat in UKB competitive tests. The long COVID UK Explore set also showed lower endothelial/TGF-beta proteins (ANGPT2 -0.84, EDN1 -0.65). Lower circulating NOTCH3 could reflect less shedding rather than fewer cells. The basement-membrane thickening reported in muscle (where pericytes help build the membrane) is a possible link.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	3 SomaScan cohorts + UKB Olink + LC UK	secondary meta-analysis	Lower mural/ECM proteins in 3-4 cohorts and long COVID

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	UKB	competitive gene-set test	Vascular smooth-muscle set flat in UKB

Open questions

Stain pericyte markers (PDGFRB, NG2, alpha-SMA) in existing muscle and skin biopsies
Relate NOTCH3/ITGAV levels to orthostatic stroke volume

Bears on: Capillary and microvascular damage Endothelial dysfunction Connective tissue / hypermobility · Test: Research only.

Blood biomarkers

Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoactive proteins in ME/CFS; most named candidates are null and the endothelial-activation set trends down. Contested

Blood tests for vessel damage markers mostly come back normal in ME/CFS when large datasets are pooled, which differs from what small studies suggested.

Solving ME/CFS re-analysis: VWF, SELP, SELE, VCAM1, ANGPT1/2, VEGFA, EDN1, ADM, ACE, REN, NPPB null; ICAM1 +0.20 (driven by UKB), ACE2 +0.21, NPPC +0.23; endothelial-activation set d -0.30 (q=0.09, down). By contrast, smaller ELISA studies report raised endothelin-1 (Haffke 2022) and Wallraven 2026 found raised IL-6, ICAM-1, VCAM-1 in PCS and ME/CFS. Post-COVID ME/CFS may differ from pre-pandemic cohorts.

Evidence

Study	n	Design	Result
Haffke 2022, J Transl Med	14 PCS-ME/CFS, 16 PCS, HC	case-control ELISA	ET-1 raised in both patient groups
Wallraven 2026, BMC Med	102 PCS (62 ME/CFS) vs 198 controls	case-control	IL-6, ICAM-1, VCAM-1 raised

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	~150 ME/CFS SomaScan + UKB Olink	secondary analysis	Named endothelial/vasoactive proteins null; activation set lower
Scherbakov 2020, ESC Heart Fail	35 ME/CFS	case-control	sVCAM and sICAM did not differ between patients with and without ED

Open questions

Are post-COVID cohorts more inflammatory/endothelial than pre-pandemic ME/CFS?
Do any of these markers change after exertion?

Bears on: Endothelial dysfunction Innate immune hyperactivation · Test: Endothelin-1, ICAM-1 and VCAM-1 are available from some specialist labs but not validated for ME/CFS.

VEGF-A has been reported lower, not higher, in ME/CFS plasma, and proteomic re-analysis finds no consistent change; long COVID cohorts more often report raised VEGF. Contested

VEGF is a signal that makes new small vessels. Results in ME/CFS are inconsistent, and they differ from long COVID.

Landi 2016 reported clustered reductions of IL-7, IL-16 and VEGF-A in ME/CFS that were not seen in liver-disease fatigue controls. Large cytokine panels (e.g. Montoya 2017) did not find VEGF among robust differences; the Solving ME/CFS proteome re-analysis found VEGFA null. In long COVID several cohorts report raised VEGF (e.g. Patel 2022). Flaskamp 2022 found post-COVID non-ME sera pro-angiogenic but post-COVID ME/CFS sera not, consistent with a missing compensatory angiogenic response in ME/CFS.

Evidence

Study	n	Design	Result
Landi 2016, Cytokine	ME/CFS vs controls and liver-disease controls	case-control multiplex	Lower VEGF-A, IL-7, IL-16
Flaskamp 2022, Cells	PCS-ME/CFS vs PCS vs HC sera	in vitro tube formation	PCS sera pro-angiogenic; PCS-ME/CFS sera not

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	SomaScan + Olink cohorts	secondary analysis	VEGFA null

Open questions

Is there a tissue-level angiogenic deficit (e.g. muscle capillary rarefaction) that blood VEGF misses?

Bears on: Capillary and microvascular damage Endothelial dysfunction · Test: Research only.

Red blood cells

Red blood cells from ME/CFS patients were stiffer in one microfluidic study, and real-time deformability cytometry found altered red-cell deformation persisting after COVID-19. Early

Red cells need to bend to squeeze through capillaries. In a small study, patients' red cells were stiffer.

Saha 2019 (Stanford/OMF, custom microfluidic channels, small paired sample, about 12-14 patients) reported slower entry and reduced elongation of ME/CFS red cells. Kubánková 2021 (Biophys J, real-time deformability cytometry) found altered erythrocyte deformation and size heterogeneity persisting months after COVID-19. No independent ME/CFS replication; older reports of abnormal red-cell shape (stomatocytes) are from the 1990s and used subjective scoring.

Evidence

Study	n	Design	Result
Saha 2019, Clin Hemorheol Microcirc	small ME/CFS vs age-matched HC sample	case-control microfluidics	ME/CFS RBCs significantly stiffer
Kubánková 2021, Biophys J	17 COVID-19, 14 recovered, 24 HC	RT-DC	Erythrocyte deformation changes persisted after recovery

Open questions

Replication with RT-DC or ektacytometry in a larger ME/CFS cohort
Is stiffness linked to oxidative stress or membrane lipids?

Bears on: Capillary and microvascular damage · Test: Research only (ektacytometry exists in some haematology labs).

Coagulation

Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be increased in platelet-poor plasma in ME/CFS and long COVID, but the method is not standardised and haemostasis experts dispute the inference that they cause symptoms. Contested

One research network sees tiny clot-like particles in patients' plasma. Other experts say the method is unreliable and the particles may not exist in circulating blood. Anticoagulant treatment based on them is not supported.

Nunes 2022 (25 ME/CFS vs 15 HC): microclot image area commonly >10-fold higher, platelet spreading score 2.72 vs 1.00, hypercoagulable TEG; load lower than in long COVID. Kruger 2022 found entrapped inflammatory proteins in long COVID microclots. Dalton 2024 (Sheffield, independent group, semi-automated counting) found higher counts in long COVID as a cohort, with wide overlap and high counts also after recent COVID. Critiques: visual scoring, uncertain blinding, ex vivo artefact possible, no in vivo demonstration; Hunt 2024 (RPTH) concluded the hypothesis is not evidence-based and unlicensed antithrombotics or apheresis are not justified outside trials; Kell 2024 rebutted. Solving ME/CFS proteome re-analysis found a platelet/microclot protein set slightly lower, not higher.

Evidence

Study	n	Design	Result
Nunes 2022, Pharmaceuticals	25 ME/CFS vs 15 HC	case-control, fluorescence microscopy + TEG	Microclot area >10x; platelet hyperactivation
Dalton 2024, medRxiv	long COVID vs controls	case-control, semi-automated counting	Higher counts in LC as a group, broad overlap
Kruger 2022, Cardiovasc Diabetol	long COVID	proteomics of microclots	Entrapped inflammatory molecules, alpha-2 antiplasmin

Against or not replicated

Study	n	Design	Result
Hunt 2024, Res Pract Thromb Haemost	review	critical appraisal	Microclot inferences not evidence-based; antithrombotics not justified
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	SomaScan + Olink cohorts	secondary analysis	Platelet/microclot protein set d -0.33 (down)

Open questions

Blinded, multi-lab, standardised counting in ME/CFS with disease controls
Do microclots exist in vivo (e.g. capillaroscopy, biopsy)?
Any controlled trial of antithrombotic therapy

Bears on: Fibrin amyloid microclots Complement activation and thromboinflammation · Test: Not reliably: some private labs offer microclot imaging; no validated clinical test.

Platelet activation findings in ME/CFS are inconsistent: one network reports hyperactivated platelets, while an earlier controlled study found no platelet activation or hypercoagulability. Contested

Whether platelets are overactive in ME/CFS depends on who measured and how.

Nunes 2022 reported PAC-1 and CD62P staining and spreading consistent with activation. Kennedy 2006 (Dundee) measured platelet aggregation, P-selectin and coagulation markers and found no evidence of activation or hypercoagulability, explicitly arguing against antiplatelet therapy. In long COVID, Cervia-Hasler 2024 found monocyte-platelet aggregates and platelet-activation markers raised. No ME/CFS-specific flow cytometry replication.

Evidence

Study	n	Design	Result
Nunes 2022, Pharmaceuticals	25 vs 15	case-control	Platelet hyperactivation by microscopy
Cervia-Hasler 2024, Science	113 COVID-19 (40 LC) + 39 controls	longitudinal proteomics	Platelet activation and monocyte-platelet aggregates in active LC

Against or not replicated

Study	n	Design	Result
Kennedy 2006, Blood Coagul Fibrinolysis	CFS vs controls	case-control	No platelet activation or hypercoagulable state

Open questions

Standardised whole-blood flow cytometry in ME/CFS with exercise challenge

Bears on: Fibrin amyloid microclots Complement activation and thromboinflammation · Test: Research only.

Routine coagulation markers (D-dimer, fibrinogen, PT/APTT) are generally normal in ME/CFS; claims of hypercoagulability rest on small 1990s studies and on thromboelastography from one network. Contested

Standard clotting blood tests are usually normal in ME/CFS.

Berg 1999 reported a 'hypercoagulable state' (soluble fibrin monomer, thrombin-antithrombin) in a small sample; Kennedy 2006 found no coagulation differences. Proteomic re-analysis found the coagulation set null. Thromboelastography abnormalities (Nunes 2022; Saleem 2026 review) are not validated as ME/CFS markers.

Evidence

Study	n	Design	Result
Berg 1999, Blood Coagul Fibrinolysis unverified	small CFS sample	case-control	Reported hypercoagulable markers

Against or not replicated

Study	n	Design	Result
Kennedy 2006, Blood Coagul Fibrinolysis	CFS vs controls	case-control	No coagulation differences
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	SomaScan + Olink cohorts	secondary analysis	Coagulation set null

Open questions

Do markers change transiently after exertion?

Bears on: Fibrin amyloid microclots Complement activation and thromboinflammation · Test: D-dimer, fibrinogen, PT/APTT are routine; expected normal.

Complement and thromboinflammation

Active long COVID shows persistent complement dysregulation (terminal complex, alternative and classical pathway activation) with thromboinflammation markers, replicated across several cohorts. Replicated

Complement is part of the innate immune system. In long COVID it stays switched on months after infection, together with signs of platelet and vessel injury.

Cervia-Hasler 2024 (Science): >6,500 proteins in 268 longitudinal samples; terminal complement dysregulation, classical activation linked to herpesvirus antibody titres, raised haemolysis, tissue-injury and platelet markers. Baillie 2024 (Med, 166 LC vs 79): iC3b, TCC, Ba and C5a predicted LC (AUC 0.785). Further partial replication in Klein 2023 and Barratt-Due 2024; one adjusted null (Farztdinov preprint). Long COVID only; ME/CFS evidence is separate.

Evidence

Study	n	Design	Result
Cervia-Hasler 2024, Science	113 COVID-19 (40 LC) + 39 controls	prospective longitudinal proteomics	Terminal and alternative/classical complement activation; thromboinflammation
Baillie 2024, Med	166 LC vs 79 recovered	case-control	Complement activation fragments raised; 4-marker AUC 0.785

Against or not replicated

Study	n	Design	Result
Farztdinov 2024, preprint unverified	29 vs 56	case-control	Complement differences not significant after age/BMI adjustment

Open questions

Does complement activation persist in long-duration LC that meets ME/CFS criteria?
Trials of complement inhibitors (none completed in LC or ME/CFS as of late 2026 to our knowledge)

Bears on: Complement activation and thromboinflammation Viral persistence Herpesvirus reactivation (EBV, HHV-6) Innate immune hyperactivation · Test: C3, C4, CH50, sC5b-9 available in some immunology labs; not validated for long COVID or ME/CFS.

In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetically defined alternative-pathway subgroup, and proteomic signals that point in different directions across cohorts. Contested

Complement may be abnormal in some ME/CFS patients, but studies disagree on which part and in which direction.

Sorensen 2003/2009 (CDC): C4a rose after exercise in CFS, with lectin-pathway MASP2 transcript differences. Maya 2026 (IJMS, CDC; 50 ME/CFS vs 121 controls from population sample): complement pQTL carriers showed alternative-pathway dysregulation (high C3/low Bb), six pQTLs linked to fatigue phenotypes in UK Biobank; SC5b-9 association possibly genetic. Solving ME/CFS re-analysis: alternative/terminal complement up in Hoel and UKB, null in Germain, down in Walitt (I2 0.7-0.9); CSF complement/coagulation sets lower in ME/CFS with POTS (Bragée 2026, 10 vs 21, exploratory).

Evidence

Study	n	Design	Result
Sorensen 2009, Mol Med	CFS vs controls, exercise	case-control, exercise challenge	Lectin-pathway changes underlying post-exercise C4a rise
Maya 2026, Int J Mol Sci	50 ME/CFS vs 121 non-fatigued	population-based case-control + pQTL	Alternative-pathway subgroup (high C3/low Bb) linked to pQTLs

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome), 2026 unverified data	4 cohorts	secondary meta-analysis	Complement direction inconsistent between cohorts

Open questions

Measure activation fragments (C5a, sC5b-9, Ba) with careful pre-analytics before and after exertion
Is the long COVID complement signal present in pre-pandemic ME/CFS?

Bears on: Complement activation and thromboinflammation Innate immune hyperactivation Genetic susceptibility acting through neurons · Test: Research only for activation products; C3/C4 routine but non-specific.

POTS overlap

POTS is present in roughly a fifth of ME/CFS patients on formal testing (prevalence varies 6-43%), but cerebral hypoperfusion and preload failure are found in most patients regardless of whether they meet POTS heart-rate criteria. Replicated

POTS is a fast heart rate on standing. Only some ME/CFS patients have it, but most have reduced brain blood flow when upright anyway.

Novak 2026: POTS 19% ME/CFS and 22% long COVID, but reduced orthostatic CBFv 88-92%. van Campen 2020: CBF abnormal in 82% of patients with normal HR/BP response. van Campen 2024/2025 split ME/CFS-POTS into a venous-pooling profile and a hyperadrenergic profile (HR rise >=40) with different CO-CBF relations, which may have different treatment implications. Yamazaki 2026: hypocapnia more common than POTS on a lean test. Low plasma volume and the renin-aldosterone paradox are well described in POTS itself.

Evidence

Study	n	Design	Result
Novak 2026, PLOS One	170 ME/CFS, 143 LC	retrospective	POTS 19%/22%; orthostatic CBFv reduction 88%/92%
van Campen 2024, J Clin Med	233 ME/CFS-POTS	retrospective tilt	Two haemodynamic POTS profiles

Open questions

Prospective trial stratified by POTS profile (venous pooling vs hyperadrenergic)
Standardised orthostatic testing duration and posture across studies

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Tilt table or 10-minute NASA lean test (widely available); add capnography and stroke-volume monitoring where possible.

Key reviews

Joseph 2021, Chest: Defines preload failure and impaired extraction phenotypes in the largest iCPET series.
van Campen 2020, Clin Neurophysiol Pract: Largest cerebral blood flow on tilt study (429 patients).
Nunes, Kell, Slaghekke, Wüst 2026, Cell Death Dis: Recent synthesis proposing endothelial senescence as a unifying model (hypothesis paper, not evidence).
Hunt 2024, Res Pract Thromb Haemost: Haemostasis-expert critique of the microclot hypothesis.
Novak 2026, PLOS One: Shared autonomic/cerebrovascular phenotype of long COVID and ME/CFS in 313 patients.
Westermeier, Pretorius, Untersmayr 2026, Cardiovasc Diabetol: Cardiometabolic and vascular perspective on PEM.

Muscle, exertion and energy

PEM definition and timing

PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took ME/CFS patients a mean of about 13 days to recover versus about 2 days in sedentary controls. Several studies

After overdoing it, people with ME/CFS often feel worse later rather than at once, and the crash can last much longer than ordinary tiredness after exercise.

Chu 2018 (n=150 survey): 84% reported PEM lasting 24 h or more; onset varied, and only 11% had a consistent delay of at least 24 h. Moore 2023 (80 vs 64, Ithaca 2-day CPET): judged recovery 12.7 ± 1.2 d vs 2.1 ± 0.2 d, range 1-64 d; one patient had not recovered at one year. NIH qualitative work (Stussman 2024, 10 vs 9) found every patient's PEM had a different onset, trajectory and worst symptom, and VAS scales missed changes that interviews caught. Self-report designs and one centre for the recovery data are the main limits.

Evidence

Study	n	Design	Result
Chu 2018, PLOS ONE	150 Fukuda CFS	patient-centred cross-sectional survey	90% PEM with physical, cognitive and emotional triggers; 84% PEM >=24 h; 11% consistent >=24 h delay
Moore 2023, Medicina	80 ME/CFS vs 64 controls	prospective symptom tracking for 10 days after 2-day CPET	Recovery 12.7 d vs 2.1 d (p<0.0001); <10% of patients took >3 weeks
Stussman 2024, BMJ Neurol Open	10 ME/CFS vs 9 controls	mixed-methods (VAS + qualitative interviews) over 72 h after CPET	Each patient had a unique PEM course; no controls had PEM; interviews outperformed VAS at peak PEM

Against or not replicated

Study	n	Design	Result
Stussman 2025, Front Neurol	34 long COVID exercise cohort, 9 ME/CFS, 9 controls; 244 questionnaire cohort	CPET with serial VAS and interviews over 72 h	Self-reported PEM 67% in long COVID questionnaires, but only 2 of 34 long COVID patients showed observable PEM after CPET; ME/CFS PEM was more severe and prolonged

Open questions

Is there an objective marker that tracks PEM onset and recovery within a person?
Why does self-reported PEM in long COVID often not reproduce after a laboratory stressor?
Do cognitive and orthostatic triggers produce the same biology as physical exertion?

Bears on: Persistent sickness behaviour / vagal signalling Cell danger response (Naviaux) Mitochondrial / energy metabolism impairment · Test: Clinical history and symptom diaries (DSQ-PEM questionnaire); no objective clinical test

Exercise capacity (single CPET)

On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled about 5 ml/kg/min below controls) with a low peak heart rate (chronotropic incompetence). Replicated

Patients can take in and use less oxygen at maximal effort than healthy people, and their heart rate does not rise as high.

Large clinical series (VanNess 2003, n=189) and case-control studies (Keller 2024, Walitt 2024 NIH, Charlton 2026) agree. Effort criteria (RER, perceived exertion) are usually met, which argues against low effort. The main confounder is deconditioning; Keller 2024 used VO2-matched pairs and Charlton 2026 compared with 60-day bed rest, and both found patterns that inactivity did not reproduce. A single CPET's peak VO2 overlaps heavily with sedentary controls, so it is not diagnostic on its own.

Evidence

Study	n	Design	Result
VanNess 2003, Med Sci Sports Exerc	189 CFS	clinical series, single CPET	Peak VO2 below predicted in all impairment groups; peak HR and SBP lower in more impaired groups
Franklin 2018/2021 meta-analysis (Teesside) unverified	pooled studies	systematic review/meta-analysis	Pooled VO2peak 5.2 (95% CI 3.8-6.6) ml/kg/min lower in ME/CFS vs controls
Walitt 2024, Nat Commun (NIH) nih-walitt-2024 mwb-st003178	about 17 PI-ME/CFS vs 21 controls	deep phenotyping	Peak VO2 lower (p=0.004), peak HR lower (p=0.07), about 3.3 METs difference; patients performed at a lower % of predicted
Keller 2024, J Transl Med	84 ME/CFS vs 71 sedentary controls (55 VO2-matched pairs)	2-day CPET, cycle ergometer, 24 h apart	ME/CFS failed to reproduce day-1 values: e.g. ~6% fall in peak VO2/kg and 6.7% fall in VO2 at ventilatory threshold in matched pairs; controls stable; both groups met maximal-effort criteria
Charlton 2026, Nat Commun	19-23 ME/CFS, 24-25 long COVID, 29-30 matched controls, 24 healthy people after 60-day bed rest	case-control biopsies plus bed-rest comparison	ME/CFS: lower type I fibre proportion and type I-specific atrophy, lower SDH activity, lower OXPHOS capacity, lower capillary-to-fibre ratio; bed rest caused atrophy but no fibre-type shift; OXPHOS tracked VO2max only after bed rest

Open questions

How much of low VO2 is preload/venous return versus muscle oxygen use?
Does peak VO2 track severity in severe and housebound patients, who are rarely tested?

Bears on: Low blood volume and preload failure Mitochondrial / energy metabolism impairment Autonomic dysfunction and small-fibre neuropathy Deconditioning · Test: Cardiopulmonary exercise test (CPET) at cardiology/sports-medicine labs; carries PEM risk

2-day CPET

ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most clearly in workload and VO2 at the ventilatory threshold, but the largest independent replication found no day-2 change. Contested

Several labs found that patients cannot repeat their performance a day after a hard test, while healthy people can. A 2026 study using the same design found no drop, so the test's reliability as a PEM marker is in doubt.

Positive studies: Snell 2013 (51 vs 10 women; 95% classification), van Campen 2020/2021 (82 women by severity; idiopathic chronic fatigue improved on day 2 while ME/CFS declined), Keller 2024 (84 vs 71; declines of ~5-7% in peak VO2 and VO2 at threshold), Hodges 2018 (MS controls improved), Davenport 2026 preprint (84 ME/CFS, 79 long COVID, 71 controls). Meta-analyses: Lim 2020 (workload at VT difference -10.8 day 1 vs -33.0 day 2) and Franklin & Graham 2022 (work at anaerobic threshold -21 W, 78% predictive probability of exceeding a 10 W MCID). Most positive data come from the Workwell/Pacific/Ithaca network and the van Campen clinic, so they are not fully independent. Mancini/Natelson 2026 (58 vs 25) and Lien 2019 (VO2peak day-2 fall not different between groups) did not reproduce the VO2 effect. Effect sizes are small relative to test-retest noise, and threshold detection is operator-dependent.

Evidence

Study	n	Design	Result
Snell 2013, Phys Ther	51 CFS vs 10 controls, women	2-day CPET	No difference on test 1; on test 2 lower VO2 and workload at peak and threshold; 95.1% classification
Keller 2024, J Transl Med	84 ME/CFS vs 71 sedentary controls (55 VO2-matched pairs)	2-day CPET, cycle ergometer, 24 h apart	ME/CFS failed to reproduce day-1 values: e.g. ~6% fall in peak VO2/kg and 6.7% fall in VO2 at ventilatory threshold in matched pairs; controls stable; both groups met maximal-effort criteria
van Campen 2020, Healthcare	82 women (31 mild, 31 moderate, 20 severe)	2-day CPET by severity	All four parameters fell on day 2 (p 0.003 to <0.0001); peak workload fall largest in severe (-19%)
van Campen 2021, Healthcare	50 ME/CFS vs 51 idiopathic chronic fatigue (women); 26 vs 25 (men)	2-day CPET, disease-control comparison	ME/CFS worsened on day 2; idiopathic chronic fatigue improved
Hodges 2018, Clin Physiol Funct Imaging	10 CFS, 7 MS, 17 controls	pilot 2-day CPET with disease control	CFS workload fell on day 2; MS and controls improved
Lim 2020, J Clin Med	pooled 2-day CPET studies	meta-analysis	All parameters fell on day 2 in ME/CFS, rose in controls; workload at VT difference grew from -10.8 to -33.0
Franklin & Graham 2022, Fatigue: Biomed Health Behav unverified	4-5 studies per outcome	systematic review and meta-analysis	Day-2 drop in peak work -8.55 W and in work at anaerobic threshold -21 W vs controls
Davenport 2026, Research Square preprint	84 ME/CFS, 79 long COVID, 71 controls	retrospective 2-day CPET	Both patient groups had reduced VO2 and workload at threshold, larger in women; groups similar

Against or not replicated

Study	n	Design	Result
Mancini 2026, Front Physiol	58 ME/CFS vs 25 sedentary controls	2-day CPET replication attempt	No day-2 change in peak VO2 (22.3 to 22.5 ml/kg/min) or VO2 at threshold in either group; patients had lower peak HR and higher perceived exertion; authors say 2-day CPET should not be used to define PEM
Lien 2019, Physiol Rep	18 vs 15 women	2-day CPET with arterial lactate	VO2peak fell on day 2 in patients, but the change did not differ significantly between groups

Open questions

Why did Mancini 2026 find no day-2 effect: cohort selection, threshold method, or sample from a different clinic?
What is the test-retest variability of workload at VT in sedentary people, and does the day-2 effect exceed it?
Can a gentler, submaximal provocation give the same information with less harm (Moore 2023: about two weeks to recover)?

Bears on: Low blood volume and preload failure Mitochondrial / energy metabolism impairment Autonomic dysfunction and small-fibre neuropathy · Test: Offered by a few specialist labs (e.g. Workwell-linked) mainly for disability documentation; provokes PEM and is not advised for moderate-severe patients

Lactate and glycolysis

Some patients produce more lactate for a given workload, and this worsens on a repeat test, but plasma studies after one maximal test show a smaller total lactate rise that mainly reflects lower achieved workloads. Several studies

Muscles in some patients seem to lean on the quick, lactate-producing fuel route earlier than expected, especially the day after hard effort.

Lien 2019 (18 vs 15 women): arterial lactate per power output was higher in patients on both tests and rose on day 2 while it fell in controls (interaction p<0.001). Lane 1998: about a third of patients had an abnormal lactate rise on a sub-anaerobic-threshold test, and this subgroup had fewer type 1 fibres (n=105 biopsies). Wong 1992 (31P-MRS) inferred accelerated glycolysis. The Solving ME/CFS re-analysis of Germain 2022 found the post-exercise lactate rise was smaller in patients (log2 +1.02 vs +1.40), which disappears as a signal once effort is considered. Lactate per watt, not peak lactate, is the informative measure.

Evidence

Study	n	Design	Result
Lien 2019, Physiol Rep	18 vs 15 women	2-day CPET with arterial lactate every 30 s	Higher lactate per power output in patients; lactate increased on day 2 in patients and fell in controls
Lane 1998, J Neurol Neurosurg Psychiatry	105 CFS biopsies	biopsy plus sub-anaerobic threshold exercise test	Abnormal lactate responders had fewer type 1 fibres; no ultrastructural abnormality
Wong 1992, Chest	22 vs 21	31P-MRS during exercise to exhaustion	Similar pH/PCr patterns but faster changes and lower ATP at exhaustion in CFS; authors infer accelerated glycolysis

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis of Germain 2022 data	60 vs 45 (404 samples)	within-person change scores, mixed models, permutation	Resting differences real (~150 of 780 metabolites at p<0.05 vs ~35 expected, perm p=0.002); exercise-response differences at chance (perm p=0.12-0.61); survivors: smaller lactate rise (explained by effort), blunted post-exercise glucose rise, rise in oxidised cys-gly

Open questions

Is early lactate due to poor oxygen delivery, fibre-type shift, or a mitochondrial/PDH block?
Does lactate per watt separate ME/CFS from deconditioned controls prospectively?

Bears on: Mitochondrial / energy metabolism impairment Capillary and microvascular damage Low blood volume and preload failure · Test: Research only (lactate measured during CPET in some labs)

Oxygen delivery and extraction

Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (preload failure) and, in a high-flow subgroup, poor systemic oxygen extraction by tissue. Several studies

During exercise, either not enough blood returns to the heart, or the blood that reaches muscle passes through without giving up enough oxygen.

Joseph 2021 (Systrom lab, 160 ME/CFS vs 36 controls from 1,516 iCPETs): right atrial pressure 1.9 vs 8.3 mmHg, peak VO2 80% vs 101% predicted; the high-flow tertile had impaired extraction (0.74 vs 0.86-0.88), consistent with peripheral shunting; 31% had small-fibre neuropathy on skin biopsy. Melamed 2019: impaired extraction was the sole limit in 12.5% of unexplained exertional intolerance. Vermeulen 2014 (non-invasive, 178 CFS women) estimated lower peak extraction and a steep cardiac output/VO2 slope. McCully 1999 (NIRS) found slower muscle oxygen delivery. Invasive data come largely from one centre; capillary rarefaction in Charlton 2026 offers a structural explanation for low extraction.

Evidence

Study	n	Design	Result
Joseph 2021, Chest	160 ME/CFS vs 36 controls	invasive CPET plus skin biopsy, retrospective	Low RAP (preload failure); impaired O2 extraction in high-flow tertile; 31% SFN
Melamed 2019, Eur J Appl Physiol	313 iCPETs for unexplained exertional intolerance	retrospective iCPET	Impaired systemic O2 extraction sole limit in 12.5%; higher mixed-venous saturation at peak
Vermeulen 2014, J Transl Med	178 CFS women, 172 idiopathic fatigue, 11 healthy women (plus men)	CPET with non-invasive cardiac output (Nexfin)	Peak extraction 10.8 vs 13.5 ml/100 ml in women; dQ/dVO2 >6 L/L in 70% of patients
McCully 1999, Clin Sci	20 vs 12	NIRS plus 31P-MRS	Slower post-exercise and post-ischaemia muscle oxygen resaturation in CFS
Appelman 2024, Nat Commun data	25 vs 21	NIRS during CPET	Muscle deoxygenation responses lower in long COVID

Open questions

Is poor extraction due to microvascular shunting, capillary loss, or mitochondria that cannot use oxygen?
Independent invasive replication outside the Systrom lab is lacking.
Do extraction subgroups predict response to pyridostigmine or volume expansion?

Bears on: Low blood volume and preload failure Capillary and microvascular damage Autonomic dysfunction and small-fibre neuropathy Endothelial dysfunction Mitochondrial / energy metabolism impairment · Test: Invasive CPET (a few centres, e.g. Brigham/Harvard); NIRS research only

31P-MRS muscle energetics

31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower pH recovery after repeated exercise in a subset of patients, but other MRS studies found no consistent group abnormality. Contested

Magnetic scans of calf muscle during exercise suggest that some patients' muscles become more acidic and clear acid slowly, but not all studies see this.

Jones 2010 (16 vs 8): proton efflux immediately post-exercise 1.1 vs 3.6 mmol/L/min, time to maximal efflux 25.6 vs 3.8 s, and the normal link between efflux and pH nadir was lost, possibly via autonomic dysfunction. Jones 2012 (18 vs 12): 10 of 18 had low PCr depletion; the other 8 showed more acidosis over three bouts, no adaptation and about 4-fold longer pH recovery. Lane 1998 found lower end-exercise pH and slower ATP synthesis only in the lactate-responder subgroup. Barnes 1993 (46 patients) found no consistent abnormality overall, with 12 outliers; Wong 1992 found similar end-points; a 2025 7 T study found no resting muscle differences. Small samples, subgroup analyses and heterogeneous criteria dominate.

Evidence

Study	n	Design	Result
Jones 2010, J Intern Med	16 vs 8	31P-MRS plantar flexion at 35% MVC plus HRV	Suppressed proton efflux and delayed peak efflux; correlation with pH nadir lost in CFS
Jones 2012, Eur J Clin Invest	18 vs 12	31P-MRS over three exercise bouts	Subgroup with normal PCr depletion had more acidosis, no adaptation, ~4-fold slower pH recovery
Lane 1998, Neuromuscul Disord	10 SATET+ vs 9 SATET- vs 13 controls	forearm 31P-MRS	Lower end-exercise pH and slower ATP resynthesis in lactate-responder subgroup only

Against or not replicated

Study	n	Design	Result
Barnes 1993, J Neurol Neurosurg Psychiatry	46 CFS vs controls and mitochondrial myopathy	31P-MRS	No consistent abnormality of glycolysis, mitochondrial metabolism or pH regulation; 12 of 46 outside normal range
Wong 1992, Chest	22 vs 21	31P-MRS gastrocnemius	No group differences in pH, PCr or Pi at rest, exhaustion or recovery; lower ATP at exhaustion
Godlewska 2025, Mol Psychiatry	24 ME/CFS, 25 long COVID, 24 controls	7 T MRS brain and calf muscle at rest	Resting muscle metabolites did not differ (brain lactate raised in ME/CFS)

Open questions

Would a pre-registered, larger repeat-bout 31P-MRS study with matched activity levels reproduce the Newcastle subgroup?
Is slow pH recovery a perfusion problem or a muscle-intrinsic one?
Does the MRS phenotype change during PEM?

Bears on: Mitochondrial / energy metabolism impairment Autonomic dysfunction and small-fibre neuropathy Sodium-calcium overload in muscle (Wirth & Scheibenbogen) Capillary and microvascular damage · Test: Research MRI only

Muscle sodium MRI

23Na-MRI in 6 women with ME/CFS and 6 controls found higher resting muscle sodium in all lower-leg compartments and a larger sodium rise after exercise, inversely related to grip strength. Early

A special MRI scan found more salt inside leg muscles of patients, which in theory could disturb muscle energy and calcium handling. The study was very small.

Petter 2022: anterior compartment 12.20 ± 1.66 vs 9.38 ± 0.71 mM (p=0.0034); medial gastrocnemius rose 30% vs 17% at 12 min after plantar flexion. Sodium correlated with lower handgrip strength (R2=0.38). This fits the Wirth/Scheibenbogen model in which hypoperfusion overloads the Na+/H+ exchanger, then Na+/Ca2+ exchange, loading muscle with calcium and damaging mitochondria. No independent replication; n=12 total.

Evidence

Study	n	Design	Result
Petter 2022, J Transl Med	6 vs 6 women	23Na-MRI lower leg before and after plantar flexion	Higher baseline sodium in all compartments; larger exercise rise; correlates with weaker grip

Open questions

Replication in a larger, sex-balanced cohort with deconditioned and disease controls.
Does muscle sodium change during PEM or with treatments that improve perfusion?

Bears on: Sodium-calcium overload in muscle (Wirth & Scheibenbogen) Capillary and microvascular damage Endothelial dysfunction · Test: Research MRI only

Muscle biopsies (pre-2020)

Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without the fibre atrophy and type 2 shift expected from inactivity, with abnormalities confined to subgroups. Contested

Muscle samples taken in the 1990s mostly looked normal or only mildly abnormal under the microscope.

Lane 1998 (105 quadriceps biopsies): type 1 predominance (23%) was more common than type 2 predominance (3%); atrophy in 10%; no ultrastructural abnormality in 41 EM samples; the abnormal-lactate subgroup had fewer type 1 fibres. Grau 1992 (n=20): only minor nonspecific changes in 9. Wassif 1994 (n=10): normal ischaemic lactate responses. Behan 1991 reported mitochondrial structural changes and Bowles 1993 enteroviral RNA in muscle (41/158 postviral fatigue vs 2/152 controls), neither reproduced with modern methods. Criteria, techniques and selection varied widely.

Evidence

Study	n	Design	Result
Lane 1998, J Neurol Neurosurg Psychiatry	105	quadriceps needle biopsy, histometry, EM	Changes not those of inactivity; no ultrastructural abnormality; lactate responders fewer type 1 fibres
Grau 1992, Clin Neuropathol	20	histopathology	Minor nonspecific changes in 9/20
Bowles 1993, J Med	158 postviral fatigue vs 152 control muscles	molecular hybridisation	Enterovirus RNA in 41/158 vs 2/152; histology otherwise normal
Behan 1991, Acta Neuropathol unverified	about 50	electron microscopy	Mitochondrial structural abnormalities reported

Against or not replicated

Study	n	Design	Result
Wassif 1994, Ann Clin Biochem	10 CFS vs 10 alcohol misuse	ischaemic lactate/ammonia test and biopsy histomorphometry	No distinctive CFS abnormality

Open questions

Re-examine archived biopsies with modern immunostaining (capillaries, immune cells, amyloid).

Bears on: Viral persistence Mitochondrial / energy metabolism impairment Deconditioning · Test: Diagnostic muscle biopsy (neuromuscular clinics) mainly to exclude myopathy

Muscle biopsies after exertion

In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed lower mitochondrial enzyme activity, more fibre necrosis, more T-cell infiltration and more amyloid-containing deposits than before exercise and than in controls. Early

When patients with long COVID were biopsied before and after a hard exercise test, their muscle looked more damaged and inflamed afterwards, and had more abnormal protein clumps.

Appelman 2024 (25 vs 21, Amsterdam): lower VO2max and gas exchange threshold; more glycolytic fibres; SDH activity fell after exercise in patients; large necrotic areas in 36% of patients after exercise (group p=0.09, so not firmly significant); CD3+ T cells present in patients but absent in controls; CD68 macrophages higher. Biopsies were scored as present/absent because sizes varied, and a first biopsy can itself cause regeneration at the second site. It is long COVID, not ME/CFS, and the amyloid-deposit finding has not been reproduced by another lab. The study prompted published criticism and replies in Nat Commun 2025.

Evidence

Study	n	Design	Result
Appelman 2024, Nat Commun data	25 long COVID vs 21 controls	longitudinal case-control; vastus lateralis biopsies before and 1 day after a maximal CPET	Lower exercise capacity; after exercise, lower SDH activity, more atrophic fibres and focal necrosis, more T-cell infiltration, more amyloid-containing deposits in patients
Charlton 2025, Trends Endocrinol Metab	review	opinion article from the same group	Argue for intrinsic mitochondrial dysfunction, endothelial abnormalities and glycolytic fibre shift as drivers; PEM trigger unknown

Open questions

Does the same post-exertion pathology occur in ME/CFS (registered study NCT05225688 includes ME/CFS)?
Are the amyloid-containing deposits fibrin(ogen)-derived and do they occlude capillaries?
Is necrosis specific to PEM or a nonspecific response to unaccustomed maximal effort?

Bears on: Fibrin amyloid microclots Mitochondrial / energy metabolism impairment Innate immune hyperactivation Capillary and microvascular damage Viral persistence · Test: Research only

Fibre type and capillaries

ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy, lower mitochondrial capacity and fewer capillaries per fibre, a pattern that 60 days of strict bed rest in healthy people did not reproduce. Early

Patients' muscles have fewer endurance-type fibres and fewer tiny blood vessels. Healthy volunteers kept in bed for two months did not develop the same pattern, so lack of activity alone does not seem to explain it.

Charlton 2026 (Amsterdam group; 19-23 ME/CFS, 24-25 long COVID, 29-30 controls, 24 bed rest): type I proportion lower (p<0.001), type IIa/IIx higher; ME/CFS-only lower SDH; both patient groups lower OXPHOS; capillary-to-fibre ratio lower in ME/CFS than in long COVID and controls (p<0.001), whereas bed rest caused atrophy with unchanged capillary-to-fibre ratio and fibre type. OXPHOS correlated with VO2max in controls and after bed rest but not in patients. Same lab as Appelman; bed-rest cohort was a different (younger, male) population studied separately. Lane 1998 reported type 1 predominance more often than type 2 in older ME/CFS biopsies, which does not fit; differences in criteria and methods may explain this.

Evidence

Study	n	Design	Result
Charlton 2026, Nat Commun	19-23 ME/CFS, 24-25 long COVID, 29-30 matched controls, 24 healthy people after 60-day bed rest	case-control biopsies plus bed-rest comparison	ME/CFS: lower type I fibre proportion and type I-specific atrophy, lower SDH activity, lower OXPHOS capacity, lower capillary-to-fibre ratio; bed rest caused atrophy but no fibre-type shift; OXPHOS tracked VO2max only after bed rest
Appelman 2024, Nat Commun data	25 long COVID vs 21 controls	longitudinal case-control; vastus lateralis biopsies before and 1 day after a maximal CPET	Lower exercise capacity; after exercise, lower SDH activity, more atrophic fibres and focal necrosis, more T-cell infiltration, more amyloid-containing deposits in patients
Bizjak 2024, Int J Mol Sci	small CFS, post-COVID and control groups	high-resolution respirometry and EM of muscle	Complex I OXPHOS lower in post-COVID vs controls; CFS showed more advanced mitochondrial structural changes than post-COVID

Against or not replicated

Study	n	Design	Result
Lane 1998, J Neurol Neurosurg Psychiatry	105	biopsy histometry	Type 1 predominance (23%) more common than type 2 predominance (3%)

Open questions

Independent replication of capillary rarefaction in ME/CFS muscle.
Is the glycolytic shift a cause of exercise intolerance or an adaptation to poor oxygen delivery?
Do severe, bedbound patients show the same profile?

Bears on: Capillary and microvascular damage Mitochondrial / energy metabolism impairment Endothelial dysfunction Deconditioning · Test: Research only

Muscle mitochondria

WASF3 protein is raised in ME/CFS muscle, and in cells and mice it disrupts respiratory supercomplex assembly and lowers exercise capacity via ER stress. Early

One protein, WASF3, was found at high levels in patients' muscle; in experiments it jammed the parts of mitochondria that make energy.

Wang/Hwang 2023 (NIH): started from one patient with high WASF3; transgenic mice overexpressing WASF3 ran less, with lower complex IV; endotoxin-induced ER stress raised WASF3; the ER-stress inhibitor 4-PBA lowered WASF3 and improved the patient's cell respiration. A cohort of ME/CFS muscle biopsies showed raised WASF3 and ER stress markers (small cohort; group n not given in abstract). Glass 2025 also found ER-stress proteins in post-exercise EVs tracking PEM severity. No independent replication of the WASF3 biopsy finding.

Evidence

Study	n	Design	Result
Wang 2023, PNAS	index patient; transgenic mice; small ME/CFS biopsy cohort	mechanistic case study plus cohort biopsies	WASF3 disrupts supercomplexes; raised WASF3 and ER stress in ME/CFS muscle

Open questions

Replication of raised muscle WASF3 in a larger, activity-matched cohort.
Does patient serum induce WASF3 in healthy muscle cells?
Would ER-stress modulators help in a trial?

Bears on: Mitochondrial / energy metabolism impairment er-stress Innate immune hyperactivation · Test: Research only

Cultured patient muscle cells

Muscle cells grown from ME/CFS patients failed to activate AMPK and raise glucose uptake after electrical pulse stimulation, and secreted less IL-6, suggesting a defect that persists outside the body. Early

Patients' muscle cells grown in a dish did not respond to simulated exercise the normal way, which hints at a built-in problem rather than one caused only by something in the blood.

Brown 2015 (Newcastle): control cultures increased AMPK phosphorylation and glucose uptake after 16 h EPS; CFS cultures did not, though insulin-stimulated uptake was preserved; IL-6 secretion lower. Sample sizes small. Tomas 2019 (same group) found no differences in individual complex activity in patient myotubes (n=6), so the defect, if real, is in signalling rather than respiratory chain capacity.

Evidence

Study	n	Design	Result
Brown 2015, PLOS ONE	small (CFS vs control myoblast donors)	in vitro EPS of cultured myotubes	No AMPK activation or glucose uptake rise after EPS; lower IL-6

Against or not replicated

Study	n	Design	Result
Tomas 2019, PeerJ	myotubes n=6; PBMC 13 CFS vs 6 controls	permeabilised-cell extracellular flux	No differences in complex I, II, IV activity or FAO/glutaminolysis-supported respiration

Open questions

Replication with more donors and blinded processing.
Is the defect epigenetic or genetic?

Bears on: Mitochondrial / energy metabolism impairment Genetic susceptibility acting through neurons · Test: Research only

Mitochondria in blood cells

Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal respiration, T-cell glycolysis), higher (lymphoblast maximal respiration with Complex V inefficiency) or normal, depending on lab and cell type. Contested

Studies of energy production in white blood cells disagree with each other, so it is unclear whether blood cells show an energy defect.

Tomas 2017: lower PBMC maximal respiration (p<=0.003), but control groups were small (fresh n=3, frozen n=12, vs 25 and 38 patients). Mandarano 2020 (53 vs 45, Hanson lab): reduced glycolysis in CD4 and CD8 T cells at rest and reduced CD8 membrane potential, with normal mitochondrial respiration in most measures. Missailidis 2020/2021 (51 CCC patients; immortalised lymphoblasts): higher maximal respiration and spare capacity but lower ATP synthesis fraction (Complex V inefficiency), with TCA, PPP and beta-oxidation enzymes up; claimed high discrimination, unreplicated. Vermeulen 2010 and Tomas 2019 found normal PBMC OXPHOS or complex activity. Directions conflict between studies, freezing affects results, and blood cells may not reflect muscle.

Evidence

Study	n	Design	Result
Tomas 2017, PLOS ONE	CFS fresh 25/frozen 38 vs control fresh 3/frozen 12	Seahorse PBMC	Lower maximal respiration and other OXPHOS parameters
Mandarano 2020, J Clin Invest	53 vs 45	Seahorse on isolated CD4/CD8 T cells	Reduced T-cell glycolysis; reduced CD8 membrane potential
Missailidis 2020, Int J Mol Sci	51 ME/CFS lymphoblast lines vs matched controls	Seahorse, proteomics	Complex V inefficiency; elevated maximal OCR and spare capacity; mTORC1 hyperactivation
Missailidis 2021, Int J Mol Sci	up to 34 vs 31 lines	transcriptomics/proteomics	Elevated TCA, PPP, FAO enzymes: dysregulated substrate supply

Against or not replicated

Study	n	Design	Result
Vermeulen 2010, J Transl Med	15 vs 15 women	PBMC OXPHOS plus 2-day CPET	PBMC oxidative phosphorylation normal despite worse day-2 performance
Tomas 2019, PeerJ	13 CFS vs 6 controls (PBMC)	permeabilised cells	No complex activity differences
Fernandez-Guerra 2021, Biomolecules	6 vs 6	PBMC bioenergetics and proteomics	Exploratory; small

Open questions

A multi-lab, blinded, pre-registered PBMC Seahorse study with fresh cells and adequate controls.
Do blood-cell findings correlate with muscle mitochondrial measures in the same people?

Bears on: Mitochondrial / energy metabolism impairment Immune (T-cell) exhaustion Genetic susceptibility acting through neurons · Test: Research only; commercial 'mitochondrial function' blood tests are not validated

Pyruvate dehydrogenase

Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling amino acids and raised PDK mRNA in blood cells; this has not been independently confirmed. Early

One group suggested that the step that lets sugar enter mitochondria to be burned cleanly is partly blocked, pushing cells toward lactate. Others have not yet confirmed it.

Fluge 2016 (200 vs 102): lower serum amino acids feeding the TCA cycle (mainly women), raised 3-methylhistidine in men, PDK1/2/4, SIRT4 and PPARdelta mRNA up in PBMC. Mughal 2025 found PDK3/PDK4 up in 3D muscle exposed to patient serum (n=4 ME/CFS). The serum-transfer part of Fluge 2016 failed replication (Ryback 2026). Solving ME/CFS re-analysis of resting plasma metabolomics found low acylcarnitines and fatty acids (lower fat-fuel supply) rather than a pile-up pattern, and a 5-cohort aromatic-lactate deficit, which point to related but different steps. No study has measured PDH activity directly in ME/CFS muscle.

Evidence

Study	n	Design	Result
Fluge 2016, JCI Insight	200 ME/CFS vs 102 controls (serum amino acids); 12 vs 12 (myoblast serum assay)	case-control metabolomics, PBMC qPCR, serum-transfer cell assay	Lower TCA-fuelling amino acids (mainly women); PDK1/2/4, SIRT4, PPARdelta mRNA up in PBMC; severe-patient serum raised myoblast respiration and lactate secretion
Mughal 2025, Biofabrication	4 ME/CFS, 5 long COVID, 4 controls	3D engineered muscle exposed to serum	PDK3/PDK4 up by qPCR; glycolysis/TCA genes up

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (resting metabolomics, several cohorts)	Germain 2022 60 vs 45 plus 5 cohorts for aromatic lactates	re-analysis of public metabolomics	Resting pattern fits low fat-derived fuel supply, not a PDH/fat-oxidation block with acylcarnitine build-up; aromatic lactates -0.53 SD

Open questions

Direct PDH activity and phosphorylation in muscle biopsies.
Replication of the amino-acid pattern in an independent cohort with diet control.

Bears on: Mitochondrial / energy metabolism impairment IDO metabolic trap (Phair, Davis) Circulating (serum) factor · Test: Research only

Serum factors and muscle cells

The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pre-registered replication of 67 patients vs 53 controls, though a small 3D engineered-muscle study reported weaker contraction with patient serum. Refuted

An influential idea was that something in patients' blood harms muscle cells' energy production. A large, careful repeat of the original experiment found no difference.

Ryback 2026: blinded, randomised, 1,926 wells, primary maximal OCR +2.50 pmol/min (95% CI -1.62 to 6.62); recomputed participant-level d=0.14 (CI -0.22 to 0.50), excluding the original d=1.32. Limits: mostly mild-moderate (only 6 severe), resting samples, batch partly collinear with group, one myoblast donor, single readout. Mughal 2025 (4/5/4 donors, IBEC) found reduced contractile force at 48 h and mitochondrial fragmentation later, with pseudoreplication risk. Prusty 2026 found patient IgG fragments endothelial (not muscle) mitochondria, with MS controls. The refutation applies to the 2D myoblast respiration readout, not to all serum-factor ideas.

Evidence

Study	n	Design	Result
Fluge 2016, JCI Insight	12 vs 12 (9 severe)	serum on healthy myoblasts	Higher respiration and lactate secretion with patient serum
Mughal 2025, Biofabrication	4 ME/CFS, 5 long COVID, 4 controls	3D muscle tissue force under electrical stimulation	Lower force after 48 h patient serum; fragmented mitochondria by 96-144 h

Against or not replicated

Study	n	Design	Result
Ryback 2026, PLOS ONE data	67 ME vs 53 controls, all female	pre-registered, blinded replication of Fluge 2016; healthy myoblasts in 20% serum 6 days, Seahorse	No difference in maximal respiration (+2.50 pmol/min, 95% CI -1.62 to 6.62, p=0.23) or any secondary measure; powered for d=0.52 vs original d=1.32
Solving ME/CFS re-analysis of Ryback 2026 supplementary data	67 vs 53	participant-level recomputation	d=0.14 overall; severe subgroup (n=6) d=0.002

Open questions

Do sera drawn during PEM or from severe patients behave differently?
Would differentiated myotubes, force readouts or IgG fractions show an effect that whole serum in 2D did not?
Can a blinded two-lab sample exchange reproduce the 3D force result?

Bears on: Circulating (serum) factor Autoimmunity (functional autoantibodies) Mitochondrial / energy metabolism impairment · Test: Research only

Exercise-induced gene expression

After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), adrenergic and immune genes rose for up to 48 h in most CFS patients but not in controls or MS patients. Early

In one lab's studies, a 25-minute bout of moderate exercise switched on 'muscle-fatigue sensing' genes in patients' white blood cells for two days, and this tracked how tired and sore they felt.

Light 2012 (48 CFS, 18 FM, 49 controls; 25 min at 70% age-predicted HR, blood at 0.5, 8, 24, 48 h, blinded qPCR of 13 genes): 71% of CFS had increases correlating with fatigue and pain; the other 29% showed decreased alpha-2A receptor mRNA and more orthostatic intolerance. White 2012 (22 CFS, 20 MS, 23 controls) added a disease control. All from the Utah Light lab with a small targeted gene panel; no independent replication. Vu 2024 single-cell RNA-seq after exercise found minimal immune changes other than platelets.

Evidence

Study	n	Design	Result
Light 2012, J Intern Med	48 CFS, 18 FM, 49 controls	pre/post moderate exercise qPCR	Post-exercise rise in sensory/adrenergic/cytokine gene mRNA for 48 h in 71% of CFS; none in controls
White 2012, Psychosom Med	22 CFS, 20 MS, 23 controls	pre/post moderate exercise qPCR	Larger rises in P2X4, TRPV1, CD14 and adrenergic receptors in CFS than controls; MS differed
Light 2009, J Pain unverified	small pilot (n from memory)	pilot	First report of post-exercise gene rises

Against or not replicated

Study	n	Design	Result
Vu 2024, Cell Rep Med data	ME/CFS vs controls, before and after exercise	single-cell RNA-seq of PBMC	Minimal post-exercise changes outside platelets (improper platelet activation)

Open questions

Independent replication with a pre-registered gene panel.
Do leukocyte receptor transcripts reflect muscle afferent signalling at all?

Bears on: Autonomic dysfunction and small-fibre neuropathy Innate immune hyperactivation Persistent sickness behaviour / vagal signalling · Test: Research only

Post-exertion metabolomics

Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but once each person is compared with themselves the exercise response differs little from controls. Contested

Blood chemistry differs between patients and healthy people even at rest. Whether hard exercise changes it in a disease-specific way is less clear.

Germain 2022 reported growing patient-control divergence (women: 6 to 54 metabolites at q<0.05 across four timepoints) and glutamate/TCA-linked recovery pathways. The Solving ME/CFS re-analysis reproduced those counts but found day-1 and day-2 differences correlate at 0.85 and the within-person response is at chance; alpha-ketoglutarate did not reproduce. Surviving modest signals: blunted post-exercise glucose rise (p=0.018, survives lactate adjustment), oxidised cys-gly rise in patients (q=0.033), lower day-2 N-acetylarginine and C6-DC acylcarnitine. Che 2025 (56 vs 52, Lipkin) reported TCA, beta-oxidation and urea-cycle abnormalities worse 24 h after exercise, but with only pre/24 h timepoints. Five of the most impaired male patients lacked post-exercise samples in Germain 2022.

Evidence

Study	n	Design	Result
Germain 2022, JCI Insight data	60 ME/CFS vs 45 controls, 4 timepoints	2-day CPET plasma metabolomics (1,157 metabolites)	Reported growing patient-control disparity over the two days, with lipid, energy and glutamate-linked pathways in recovery
Che 2025, npj Metab Health Dis	56 ME/CFS vs 52 controls (47 matched pairs)	plasma metabolomics/proteomics and stimulated immune responses before and 24 h after exercise	Exaggerated innate responses; impaired energy pathways; many abnormalities worse after exercise and correlated with symptoms

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis of Germain 2022 data	60 vs 45 (404 samples)	within-person change scores, mixed models, permutation	Resting differences real (~150 of 780 metabolites at p<0.05 vs ~35 expected, perm p=0.002); exercise-response differences at chance (perm p=0.12-0.61); survivors: smaller lactate rise (explained by effort), blunted post-exercise glucose rise, rise in oxidised cys-gly

Open questions

Does a within-person analysis of Che 2025 hold up?
Plasma may be the wrong compartment; paired muscle microdialysis or biopsy metabolomics are lacking.

Bears on: Mitochondrial / energy metabolism impairment Innate immune hyperactivation Cell danger response (Naviaux) · Test: Research only

Post-exertion proteomics

Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the largest patient-specific changes in the 24 h recovery phase, with suppressed T/B-cell and IL-17 signalling and raised glycolysis proteins. Contested

The biggest differences in blood proteins appeared during the recovery day, when PEM begins, rather than straight after exercise.

Germain 2025 (SomaScan 7K, 6,361 proteins; 96 F, 36 M): persistent dysregulation of immune, metabolic and neuromuscular pathways; protein-VO2max/AT relationships present in controls were disrupted in patients; sex-specific responses. Che 2025 proteomics added complement activation and extracellular-matrix ligands after exercise. The Solving ME/CFS proteome re-analysis found resting muscle-derived proteins lower in 3 SomaScan cohorts (including Germain) and UK Biobank (d -0.50). Pathway enrichment across thousands of aptamers is vulnerable to multiplicity; no independent post-exertion proteomics cohort yet.

Evidence

Study	n	Design	Result
Germain 2025, Mol Cell Proteomics data	79 ME/CFS vs 53 controls	longitudinal SomaScan 7K across 2-day CPET	Recovery-phase maladaptation; suppressed lymphocyte and IL-17 signalling; glycolysis up; links to muscle pain and sore throat
Che 2025, npj Metab Health Dis	56 vs 52	multi-omics pre/24 h post exercise	Complement activation, ECM disruption, redox imbalance worse post-exercise

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (proteome) data	Germain 79/51 per-person	within-person change	0 proteins with a patient-specific exercise response at FDR; pathway claims not reproduced per person

Open questions

Replication of recovery-phase protein changes in an independent cohort.
Which changes are effort-dependent (adjust for achieved workload)?

Bears on: Immune (T-cell) exhaustion Innate immune hyperactivation Complement activation and thromboinflammation Mitochondrial / energy metabolism impairment · Test: Research only

Extracellular vesicles after exercise

Extracellular-vesicle protein cargo responds to maximal exercise differently in ME/CFS (blunted in women; TCA proteins down and complement up in men), and recovery-phase changes correlate with PEM severity. Early

Tiny packages that cells release into the blood carry different proteins after exercise in patients than in healthy people.

Giloteaux 2024 (18 vs 17 women; pre, 15 min, 24 h; TMT proteomics): patients' EV changes differed and correlated with symptom severity, involving coagulation, muscle contraction, cytoskeleton, immune and brain proteins. Glass 2025 (10 vs 12 men): TCA-cycle proteins down and complement up at 15 min; ER-stress and protein-folding changes in recovery correlated with PEM; EV-VO2 correlations seen in controls were absent in patients. Both from the Hanson lab, small, many comparisons; no transfer experiment shows patient EVs act on cells.

Evidence

Study	n	Design	Result
Giloteaux 2024, J Extracell Vesicles data	18 vs 17 women	EV proteomics pre/15 min/24 h post CPET	Different EV response in patients; correlates with severity
Glass 2025, Clin Transl Med	10 vs 12 men	EV proteomics pre/15 min/24 h post CPET	TCA down, complement up post-exercise; ER-stress changes track PEM

Open questions

Independent replication, and functional tests of patient EVs on muscle or endothelial cells.
Tissue of origin of the EVs (muscle, platelet, endothelium).

Bears on: Complement activation and thromboinflammation er-stress Mitochondrial / energy metabolism impairment Cell danger response (Naviaux) · Test: Research only

Muscle strength and fatigability

Repeated maximal handgrip testing shows lower strength, faster decline across repetitions and poorer recovery after 60 minutes in ME/CFS, correlating with severity and PEM. Several studies

Squeezing a grip meter ten times in a row, twice an hour apart, shows patients' strength fading faster and recovering less than healthy people's. It is cheap and quick.

Jäkel 2021 (105 ME/CFS, 18 cancer-related fatigue, 66 controls; Charité): lower Fmax and Fmean (p<0.0001), higher fatigue ratio, lower recovery ratio (Fmean2/Fmean1); lower grip linked to severity, PEM, muscle pain and post-exertion CK/LDH. Paffrath 2024 (144 women with post-COVID, 78 meeting CCC): grip impaired across post-COVID, but linked to key symptoms only in ME/CFS. Walitt 2024 (NIH) found normal MVC but shorter time to failure at 50% MVC (p=0.0002) and faster decline on repetitive grip, which the authors attributed to 'effort preference', an interpretation that has been disputed. Paul 1999 (10 vs 10) showed slower quadriceps strength recovery up to 24 h. Most data come from one centre; effort and motivation cannot be fully excluded from voluntary tests.

Evidence

Study	n	Design	Result
Jäkel 2021, J Transl Med	105 ME/CFS, 18 CRF, 66 controls	repeat maximal HGS (10 pulls x 2, 60 min apart)	Lower strength, higher fatigue ratio, lower recovery ratio; correlates with severity and PEM
Paffrath 2024, J Clin Med	144 post-COVID women (78 ME/CFS)	repeat HGS	Impaired HGS in all; associations with PEM, fatigue, cognition only in ME/CFS
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	16-20 per group	grip MVC, time to failure, repetitive grip	Normal MVC; shorter time to failure; faster decline on repetitive grip
Paul 1999, Eur J Neurol	10 vs 10	quadriceps fatiguing protocol with recovery to 24 h	Similar decline during exercise; recovery prolonged in CFS at 200 min and 24 h

Open questions

Independent validation outside Charité with deconditioned and depressed controls.
Can repeat HGS track PEM day by day or serve as a trial endpoint?
How much reflects central drive versus peripheral muscle fatigue (Walitt found no central or peripheral fatigue markers)?

Bears on: Sodium-calcium overload in muscle (Wirth & Scheibenbogen) Mitochondrial / energy metabolism impairment Persistent sickness behaviour / vagal signalling · Test: Yes: handgrip dynamometer (cheap; offered by some ME/CFS clinics, e.g. Charité protocol)

Muscle proteins in blood

Muscle-derived proteins (myoglobin, troponins, CA3 and similar) are lower in ME/CFS blood at rest across three SomaScan cohorts and UK Biobank, not explained by step counts or walking pace. Several studies

Proteins that leak from muscle into blood are slightly lower in patients, which suggests less muscle mass or slower muscle turnover rather than ongoing damage.

Solving ME/CFS re-analysis: muscle-damage set d -0.50 (95% CI -0.76 to -0.24, q=0.002, I2 0.07) across Hoel (-0.85), Germain (-0.43) and Walitt (-0.34); UK Biobank competitive shift negative; not mediated by walking pace; unrelated to steps, Bell score or illness duration within patients. Lean mass was not available, so reduced muscle mass from inactivity cannot be excluded. Resting CK and myoglobin are expected to be low-normal.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (proteome) data	Hoel 50/24, Germain 78/51, Walitt 21/21, plus UK Biobank	meta-analysis of public proteomics	Muscle protein set d -0.50, consistent across cohorts

Open questions

Does the signal survive adjustment for DXA lean mass?
Does it relate to type I fibre atrophy or capillary loss in biopsies from the same people?

Bears on: Deconditioning Mitochondrial / energy metabolism impairment Capillary and microvascular damage · Test: Not clinically meaningful; DXA lean mass is available

Deconditioning versus disease

Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison groups, and a 60-day bed-rest model) indicate that deconditioning alone does not explain the exercise and muscle abnormalities in ME/CFS. Several studies

It is often said patients are just unfit from resting. Studies that compared them with equally unfit people or with healthy volunteers kept in bed found differences that inactivity did not produce.

Keller 2024: day-2 declines persisted in 55 pairs matched for peak VO2. van Campen 2021 and Hodges 2018: idiopathic chronic fatigue and MS patients improved on day 2. Lane 1998: biopsy changes were not those of inactivity. Charlton 2026: bed rest caused atrophy with preserved fibre type and capillary-to-fibre ratio; patients showed the opposite pattern; OXPHOS-VO2max coupling present after bed rest but absent in patients. Vermeulen 2014: a high cardiac output/VO2 slope argues against central deconditioning. These are mostly cross-sectional and from enthusiast centres; Mancini 2026 found no day-2 decline at all.

Evidence

Study	n	Design	Result
Keller 2024, J Transl Med	84 ME/CFS vs 71 sedentary controls (55 VO2-matched pairs)	2-day CPET, cycle ergometer, 24 h apart	ME/CFS failed to reproduce day-1 values: e.g. ~6% fall in peak VO2/kg and 6.7% fall in VO2 at ventilatory threshold in matched pairs; controls stable; both groups met maximal-effort criteria
Charlton 2026, Nat Commun	19-23 ME/CFS, 24-25 long COVID, 29-30 matched controls, 24 healthy people after 60-day bed rest	case-control biopsies plus bed-rest comparison	ME/CFS: lower type I fibre proportion and type I-specific atrophy, lower SDH activity, lower OXPHOS capacity, lower capillary-to-fibre ratio; bed rest caused atrophy but no fibre-type shift; OXPHOS tracked VO2max only after bed rest
van Campen 2021, Healthcare	50 vs 51 women	2-day CPET, idiopathic chronic fatigue control	ICF improved on day 2; ME/CFS declined
Charlton 2026, Br J Sports Med	narrative	review/editorial	Argues cardiac deconditioning is a myth in long COVID PEM

Against or not replicated

Study	n	Design	Result
Mancini 2026, Front Physiol	58 ME/CFS vs 25 sedentary controls	2-day CPET replication attempt	No day-2 change in peak VO2 (22.3 to 22.5 ml/kg/min) or VO2 at threshold in either group; patients had lower peak HR and higher perceived exertion; authors say 2-day CPET should not be used to define PEM

Open questions

A prospective study following new post-infectious patients from onset, measuring muscle before inactivity accumulates.

Bears on: Deconditioning Capillary and microvascular damage Mitochondrial / energy metabolism impairment · Test: n/a

Key reviews

Scheibenbogen & Wirth 2025, J Cachexia Sarcopenia Muscle: Collects the evidence for skeletal muscle disturbance and the sodium/calcium overload model (advocacy-leaning, from the model's authors).
Charlton 2025, Trends Endocrinol Metab: Opinion piece from the Amsterdam biopsy group on muscle adaptations and PEM in long COVID.
Lim 2020, J Clin Med: Meta-analysis of 2-day CPET; workload at ventilatory threshold is the most discriminating measure.
Franklin & Graham 2022, Fatigue: Biomed Health Behav: Meta-analysis estimating clinically meaningful day-2 drop in work at anaerobic threshold (-21 W).
Stevens 2018, Front Pediatr: Methods paper for administering 2-day CPET in ME/CFS.
Fanò-Illic 2026, Diagnostics: Recent review of skeletal muscle involvement in ME/CFS (title/abstract seen only).
Solving ME/CFS re-analyses (PEM, serum-check, proteome): Within-person re-analysis of Germain 2022 metabolomics; participant-level check of Ryback 2026; muscle-protein meta-analysis.

Nervous system and brain

Cerebral perfusion / orthostatic

Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 26% vs 7% in controls), including in patients whose heart rate and blood pressure stay normal. Replicated

When patients are tilted upright, less blood reaches the brain than in healthy people, even if their pulse and blood pressure look normal. That can explain light-headedness and brain fog when standing.

van Campen 2020 measured extracranial carotid plus vertebral flow by Doppler during a 30-min tilt in 429 patients vs 44 controls. End-tilt drop was 7% in controls and 26% in patients overall: 24% with a normal HR/BP response, 28% with delayed orthostatic hypotension and 29% with POTS. About 90% exceeded the control range. Transcranial Doppler studies (Medow 2014; Novak 2026, 88% of 170 ME/CFS) agree. A 2025 systematic review (118 studies) found reduced CBF in 12 of 26 ME/CFS-only studies and a larger fall when ME/CFS and orthostatic intolerance coexist. Caveats: most data come from referral clinics; controls are few; and CBF velocity (transcranial Doppler) is not the same as volume flow.

Evidence

Study	n	Design	Result
van Campen 2020, Clin Neurophysiol Pract	429 vs 44	case-control, tilt with extracranial Doppler	End-tilt CBF -26% vs -7%; ~90% of patients abnormal, incl. normal HR/BP responders
Novak 2026, PLoS One	170 ME/CFS, 143 long COVID, 73 HC, 290 hEDS	retrospective autonomic-lab cohort, TCD + capnography	Reduced orthostatic CBF velocity in 88% ME/CFS and 92% long COVID
Medow 2014, J Appl Physiol	15 vs 11	case-control tilt with TCD	CBFv fell 22.5% vs 8.7% in controls
Christopoulos 2025, J Transl Med	118 studies, 9185 participants	systematic review	CBF reduced in ME/CFS and in OI; larger reduction when both present (4/11 studies)

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	deep phenotyping incl. 40-min tilt	No group differences in orthostatic hypotension, tachycardia or tilt-related stopping (CBF not measured)

Open questions

Does the CBF fall cause the cognitive symptoms, or only accompany them?
How much is driven by hypocapnia versus low stroke volume?
Is it specific to ME/CFS versus deconditioning, hEDS or other dysautonomias?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy Endothelial dysfunction · Test: Tilt with extracranial Doppler (few centres, e.g. Netherlands) or transcranial Doppler in some autonomic labs; mostly research or specialist only

Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of the fall in brain blood flow, and correcting it with phenylephrine restored blood flow and working memory in one small trial. Several studies

Standing makes some patients breathe off too much carbon dioxide. Low CO2 narrows brain blood vessels, so less blood reaches the brain. A drug that tightens blood vessels prevented this and improved a memory test in a small study.

Medow 2014: on tilt, ETCO2 was 33.9 vs 42.8 Torr in 15 CFS vs 11 controls. Phenylephrine prevented the hypocapnia, cut the CBFv fall from 22.5% to 8.1%, and returned n-back reaction time to supine levels. Yamazaki 2026: 40.8% of 49 adults met the criteria for postural orthostatic syndrome of hypocapnia (POSH) on a lean test with capnography, more often than POTS. Novak 2026 defines a hypocapnic cerebral hypoperfusion phenotype. In long COVID (O'Brien 2026, n=50), 64% had supine or orthostatic hypocapnia. Solving ME/CFS re-analysis: hypocapnia on tilt is reported by 5 groups and the CBF fall by 4. Treatment evidence is still single-group and acute.

Evidence

Study	n	Design	Result
Medow 2014, J Appl Physiol	15 vs 11	case-control + acute phenylephrine crossover	ETCO2 33.9 vs 42.8 Torr on tilt; phenylephrine normalised CBFv and n-back
Yamazaki 2026, Physiol Rep	49	cross-sectional, lean test with capnography + Daxor blood volume	POSH in 40.8%; lean-test abnormalities not related to blood volume
O'Brien 2026, J Clin Med	50 long COVID	clinic cohort, NASA lean test + ETCO2	Hypocapnia 64%; POTS 32%; breathing retraining improved symptoms, not ETCO2
Medow & Stewart 2024, Am J Physiol Regul Integr Comp Physiol unverified	small	physiology	Phenylephrine alters CBFv-BP phase synchronisation in ME/CFS with OI

Open questions

Is hypocapnia primary (respiratory/chemoreflex) or compensation for low venous return?
Would capnography-guided treatment (compression, midodrine, breathing retraining) improve function over weeks?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Capnography during a 10-min stand or lean test: cheap; available in some respiratory/autonomic clinics

Autonomic

POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%) with the setting, the age group and the test protocol. Contested

Some, but not most, people with ME/CFS meet the heart-rate criteria for POTS. How many depends a lot on who was tested and how.

van Campen 2020: 120 of 429 tilt-tested patients (28%) had POTS. Novak 2026: POTS in 19% of ME/CFS and 22% of long COVID, but widespread mild autonomic failure in 89-95%. Adolescents (Leone 2026): PoTS in 7/18 ME/CFS vs 0/18 controls on a 10-min stand, although 83% reported OI symptoms. In the NIH cohort (Walitt 2024, n=17), tilt showed no excess tachycardia. Solving ME/CFS re-analysis of the literature: POTS prevalence is inconsistent (6-43%) and lists it among the contradictory findings. Orthostatic intolerance symptoms are far more common than POTS defined by heart rate.

Evidence

Study	n	Design	Result
van Campen 2020, Clin Neurophysiol Pract	429	tilt cohort	POTS 28%, delayed OH 14%, normal HR/BP 58%
Novak 2026, PLoS One	170 ME/CFS	retrospective autonomic-lab cohort	POTS 19%, neurogenic OH 15%, mild-moderate autonomic failure 89%
Leone 2026, Front Pediatr	18 vs 18 adolescents	case-control, 10-min passive stand	PoTS 39% vs 0%; OI history 83% vs 17%

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control tilt	No excess orthostatic tachycardia vs controls

Open questions

Do POTS and non-POTS ME/CFS differ in mechanism (e.g. hyperadrenergic vs low-flow)?
How much do medication, test length and time since onset change the rate?

Bears on: Autonomic dysfunction and small-fibre neuropathy Low blood volume and preload failure Autoimmunity (functional autoantibodies) · Test: 10-min active stand or tilt table test: widely available clinically

Resting heart rate is higher and high-frequency heart-rate variability lower in ME/CFS. This fits reduced vagal (parasympathetic) tone. Replicated

The 'rest and digest' nerve signal to the heart looks weaker. The heart beats a little faster at rest, and its beat-to-beat variation is reduced.

Nelson 2019 meta-analysis (64 case-control studies): resting HR +4.1 bpm, HR on tilt SMD 0.92, resting LF/HF SMD 0.39 higher, HF-HRV SMD -0.34, and peak HR 13.8 bpm lower. Walitt 2024: 24-h Holter SDNNi, rMSSD and pNN50 were lower, HF and LF power reduced, and Valsalva baroreflex-cardiovagal function was lower. Effects are moderate. They are not specific to ME/CFS, and deconditioning, medications and sleep loss confound them. HRV is not a diagnostic test.

Evidence

Study	n	Design	Result
Nelson 2019, Medicine (Baltimore)	64 studies	systematic review and meta-analysis	RHR +4.1 bpm; HF-HRV SMD -0.34; HRtilt SMD 0.92; HRmax -13.8 bpm
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	14 vs 19 (Holter)	case-control	Lower 24-h HRV time and frequency domains; reduced baroreflex-cardiovagal function

Open questions

Is low HRV cause, consequence or a marker of deconditioning/sleep disruption?
Can wearable HRV usefully track PEM risk in individuals?

Bears on: Autonomic dysfunction and small-fibre neuropathy Neuroinflammation (microglial activation) Persistent sickness behaviour / vagal signalling · Test: 24-h Holter HRV (clinical); consumer wearables give rough trends

On the 10-minute NASA lean test, a subgroup of recent-onset patients shows a rising heart rate with an abnormally narrow pulse pressure. Autonomic specialists say the test has not been validated against tilt or active stand. Several studies

The lean test (standing against a wall for 10 minutes while pulse and blood pressure are recorded) is a cheap way to see orthostatic problems. It is useful for spotting a pattern, but it is not an approved test for diagnosing POTS.

Lee 2020 (Bateman Horne Center): 150 ME/CFS (75 ill <4 y, 75 ill >10 y) vs 75 controls. At least 5 min was needed to see changes. The <4-y group had higher HR, a narrower pulse pressure and more OI symptoms, interpreted as inadequate ventricular filling. Yamazaki 2026 found lean-test abnormalities (most often hypocapnia) in 55% of 49 patients, unrelated to measured blood volume. Teuschl 2025 (European autonomic experts) argue that NLT criteria were borrowed from 1970s NASA physiology, that the leaning posture mixes active stand and passive tilt, and that the NLT 'may currently not be used as a diagnostic tool'.

Evidence

Study	n	Design	Result
Lee 2020, J Transl Med	150 vs 75	case-control	<4-y patients: higher HR and narrowed pulse pressure vs >10-y and HC
Yamazaki 2026, Physiol Rep	49	cross-sectional	55% had >=1 lean-test abnormality (POSH 41%); not related to blood volume

Against or not replicated

Study	n	Design	Result
Teuschl 2025, Clin Auton Res	n/a	expert commentary	No validation vs tilt/active stand; should not be used for autonomic diagnosis

Open questions

Head-to-head diagnostic accuracy of NLT vs tilt in ME/CFS
Does narrow pulse pressure on NLT predict response to volume or compression?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: NASA lean test: any clinic or at home with a BP cuff (for monitoring, not formal diagnosis)

Small-fibre neuropathy

Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients from specialist referral cohorts, but the controlled NIH study found no difference from healthy volunteers. Contested

In some clinics, a third to a half of patients have lost some of the thinnest nerves in the skin. Those nerves control sweating and blood vessels. But a carefully matched NIH study did not find this, so how common and how important it is remains uncertain.

Joseph 2021 (Systrom/Oaklander, Harvard): 31% of 160 ME/CFS biopsies were consistent with SFN (epidermal innervation <=5th percentile of predicted). Denervation severity did not correlate with exertional measures. Novak 2026 (same network): SFN in 53% of ME/CFS and 67% of long COVID; hEDS had more severe neurodegeneration. Azcue 2025 (30 ME/CFS, 30 long COVID, 30 HC) found corneal fibre tortuosity and heat-detection changes rather than a clear length-dependent pattern. Walitt 2024: no difference in small-fibre density. Major caveats: the positive cohorts were referred for exertional intolerance or dysautonomia; biopsy normative cut-offs and patchy, non-length-dependent SFN make rates sensitive to method; and ME/CFS-specific matched-control biopsy studies are few. Solving ME/CFS re-analysis of Finnish registry data: an SFN diagnosis is about 8-10 times more common than expected in post-viral fatigue (crude, shared with fibromyalgia).

Evidence

Study	n	Design	Result
Joseph 2021, Chest	160 ME/CFS (36 iCPET controls)	retrospective iCPET + skin biopsy	31% SFN; severity not correlated with exertional limits
Novak 2026, PLoS One	170 ME/CFS, 143 LC	retrospective autonomic-lab cohort	SFN 53% ME/CFS, 67% LC
Azcue 2025, Eur J Neurol	30 ME/CFS, 30 LC, 30 HC	case-control, QST, Sudoscan, corneal confocal microscopy	Corneal fibre tortuosity higher (AUC 0.72); heat detection impaired

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control, thigh + leg biopsies	Small nerve fibre density not different

Open questions

What is the SFN rate in unselected, community ME/CFS with blinded, matched controls?
Does fibre loss explain any symptom (OI, pain, exertional intolerance)?
Is it autoimmune, and does IVIG change it? (no ME/CFS RCT)

Bears on: Autonomic dysfunction and small-fibre neuropathy Autoimmunity (functional autoantibodies) Connective tissue / hypermobility · Test: 3 mm skin punch biopsy for IENFD via neurology (clinically available); corneal confocal microscopy in some centres

Long COVID patients referred for neuropathy evaluation often have small-fibre neuropathy on skin biopsy, including loss of autonomic fibres to sweat glands and hair muscles. Several studies

Similar nerve damage turns up in long COVID, especially in patients with pain or nerve-type symptoms.

Oaklander 2022 (NIH/MGH): 10 of 16 skin biopsies (63%) in referred long COVID were abnormal; 52% average improvement over 1.4 years, often after immunotherapy (uncontrolled). Falco 2026: piloerector and sweat-gland nerve density reduced in 31 painful long COVID vs 42 controls, even with normal IENFD, but not correlated with COMPASS-31 symptom scores. These are selected samples with neuropathic complaints, so the results do not give a general prevalence.

Evidence

Study	n	Design	Result
Oaklander 2022, Neurol Neuroimmunol Neuroinflamm	17	case series	63% of biopsies confirm SFN
Falco 2026, Front Hum Neurosci	31 vs 42	retrospective case-control histology	Autonomic fibre density reduced; no correlation with symptoms

Open questions

Prevalence in unselected long COVID
Does SFN predict who develops ME/CFS after COVID?

Bears on: Autonomic dysfunction and small-fibre neuropathy Autoimmunity (functional autoantibodies) Viral persistence · Test: Skin biopsy (IENFD, sweat-gland nerve density in specialist labs)

Brain structure

Conventional structural MRI (grey matter volume, routine clinical reads) shows no consistent abnormality in ME/CFS. Reported regional volume differences rarely replicate across groups. Contested

Standard brain scans usually look normal. Studies that report bigger or smaller brain regions mostly disagree with each other.

Shan 2020 systematic review (63 imaging studies to 2018): the only findings reported by multiple teams were extra brain-area recruitment during cognitive tasks (11 studies), brainstem abnormalities (9), sluggish BOLD responses, reduced serotonin transporter and regional hypometabolism. Volumetric findings were inconsistent. Lee 2023 meta-analysis (65 studies, 1529 patients) reported insular and thalamic hypoactivity but noted poor inter-rater reliability. Walitt 2024: brain MRI unremarkable. Kimura 2026 (38 ME/CFS, 27 LC, 47 HC): amygdala subfield differences did not survive correction. For contrast, the UK Biobank pre/post scans (Douaud 2022) show that SARS-CoV-2 infection itself produces small grey-matter changes, although that sample was not long COVID or ME/CFS.

Evidence

Study	n	Design	Result
Shan 2020, J Transl Med	63 studies	systematic review	Few consistent findings; brainstem and task recruitment most repeated
Lee 2023, Autoimmun Rev	65 studies; 1529 vs 1715	systematic review and meta-analysis	Insula and thalamus hypoactivity; frontal cortex most often reported

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control 3T MRI incl. DTI	Brain imaging unremarkable
Kimura 2026, Front Neurol	38 ME/CFS, 27 LC, 47 HC	case-control volumetry	No amygdala subfield difference after correction

Open questions

Large, multi-site, pre-registered MRI with harmonised protocols has not been done
Do subtle changes track severity or duration longitudinally?

Bears on: Neuroinflammation (microglial activation) Genetic susceptibility acting through neurons · Test: Clinical MRI (useful only to exclude other diagnoses)

Advanced diffusion MRI models report widespread white-matter differences in ME/CFS that conventional DTI misses, and the pattern may differ between post-infectious and gradual-onset cases. Early

Newer scan analyses suggest subtle changes in the brain's wiring, possibly swelling or inflammation. Standard analyses of the same scans find little.

Yu 2026 (Shan group, 67 vs 67 matched): a neuroinflammation-imaging (NII) model gave lower hindered and restricted water fractions and higher fibre fraction across association, commissural and projection tracts. The authors read this as oedema, cellular infiltration and axonal reorganisation, while conventional DTI showed minimal differences. Yu 2025 (same group): post-infectious cases had higher axial diffusivity, while gradual-onset cases had lower AD in the corpus callosum. Singh 2026 (Griffith, 37 ME/CFS, 19 LC, 27 HC) found DTI/DKI alterations in the cingulum, SMA and corpus callosum. The model-based metrics are not validated against histology in ME/CFS, the analyses carry many degrees of freedom, and independent replication is missing.

Evidence

Study	n	Design	Result
Yu 2026, Hum Brain Mapp	67 vs 67	case-control diffusion MRI (NII model)	Widespread NII differences; conventional DTI minimal
Yu 2025, Sci Rep	PI and GO subgroups vs sedentary HC	case-control diffusion MRI	Opposite axial diffusivity patterns by onset type
Singh 2026, Front Med	37 ME/CFS, 19 LC, 27 HC	case-control DTI/DKI	Microstructural changes in cingulum, SMA, corpus callosum

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	15-direction DTI	No reported imaging abnormality

Open questions

Independent replication with shared analysis code
Correlation with TSPO-PET or MRS markers in the same people

Bears on: Neuroinflammation (microglial activation) · Test: research only

Brainstem differences (smaller volume, altered signal or connectivity) are among the more often repeated imaging findings in ME/CFS, though the methods differ between studies. Several studies

The brainstem controls breathing, heart rate and arousal. Several groups using different scans have reported differences there.

Shan 2020 counted brainstem abnormalities in 9 studies from different teams and modalities. Bader 2026 (preprint, 26 vs 27): brainstem volume 1.46 vs 1.55% of eTIV (FDR p=0.039), with deep grey matter and whole-brain volume unchanged. Inderyas 2026 (7T fMRI, 32 ME/CFS): altered cuneiform-nucleus to medulla connectivity during Stroop. The Griffith (Barnden/Thapaliya) group produced several of the brainstem reports, so the evidence is not fully independent. This imaging work does not support the mechanical 'craniocervical compression' model, which is a separate claim (see neuro-cci).

Evidence

Study	n	Design	Result
Bader 2026, medRxiv (preprint)	26 vs 27	case-control volumetry + ASL + MRS under hypoxia	Brainstem volume reduced
Inderyas 2026, J Transl Med	32 ME/CFS, 19 LC, 27 HC	7T task fMRI	Altered brainstem functional connectivity during cognitive fatigue
Shan 2020, J Transl Med	9 studies	systematic review	Brainstem abnormalities repeatedly reported

Open questions

Which brainstem nuclei, and do they map onto autonomic or respiratory findings (hypocapnia, low HRV)?

Bears on: Brainstem and craniocervical instability Neuroinflammation (microglial activation) Autonomic dysfunction and small-fibre neuropathy · Test: research only

Functional MRI

After a single 30-minute exercise bout, women with ME/CFS performed worse on a demanding cognitive task and recruited more parietal and cingulate cortex. Controls improved. Early

Exercise made patients' thinking worse the next day, and their brains had to work harder at the same task. This is one of the few direct brain measurements of post-exertional malaise.

Cook 2017: 15 vs 15 women, scanned 1 week before and 24 h after 30 min of cycling at 70% peak HR. Group x time interactions showed worse PASAT performance in patients and increased activation in bilateral inferior/superior parietal and cingulate cortex, correlated with symptom change. Finger tapping did not differ. Patients cycled at lower watts, so absolute workload differed. Related work: Schönberg 2024 (34 vs 34) found that BOLD signals in ME/CFS did not adapt across repeated task blocks as they did in controls, and Shan 2020 counted extra task recruitment as a finding repeated by several teams. A direct replication of the exercise-provocation design has not been published.

Evidence

Study	n	Design	Result
Cook 2017, Brain Behav Immun	15 vs 15	pre/post exercise fMRI case-control	Post-exercise cognitive decline and increased parietal/cingulate activation in patients
Schönberg 2024, J Cereb Blood Flow Metab	34 vs 34	task fMRI case-control	Absent BOLD adaptation across repeated SDMT blocks

Open questions

Replication with workload matched relative to individual capacity
Time course over 24-72 h of PEM

Bears on: Neuroinflammation (microglial activation) Mitochondrial / energy metabolism impairment Persistent sickness behaviour / vagal signalling · Test: research only

During a sustained grip task, ME/CFS patients fatigue earlier while their brain motor-network activity fails to ramp up as it does in controls (NIH; reduced temporoparietal-junction activation). Early

When squeezing repeatedly, patients tire sooner. Their brains do not step up the drive to the muscle the way healthy brains do.

Walitt 2024 reported decreased right temporoparietal junction activation during repeated 50% MVC grip in fMRI and a faster decline in force. The authors linked it to 'effort preference', which is disputed. Bedard 2026 (NIH, 15 vs 19, EEG+EMG+fMRI) found that maximum force was equal, but patients reached fatigue much earlier and showed minimal change in brain and muscle activity across blocks. Healthy controls increased activity until fatigue. Interpretation: a central contribution to fatigability. Both studies come from the same NIH programme and have small samples.

Evidence

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control task fMRI	Reduced TPJ activation; earlier force decline
Bedard 2026, Neuroimage Clin	15 vs 19	case-control multimodal EEG/EMG/fMRI	Earlier fatigue; brain output did not increase to drive muscle

Open questions

Is the lack of central ramp-up primary or protective (driven by peripheral afferent feedback)?
Independent replication outside NIH

Bears on: Persistent sickness behaviour / vagal signalling Neuroinflammation (microglial activation) Mitochondrial / energy metabolism impairment · Test: research only

MR spectroscopy

Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in three cohorts from one group, and thalamic or regional lactate in two other groups. It is also raised in depression, so it is not specific. Several studies

Brain scans that measure chemicals show more lactate, a by-product of energy production without enough oxygen, in patients' brains. The same change appears in depression.

Mathew 2009: ventricular lactate +348% vs healthy and +297% vs anxiety disorder (16 vs 15 vs 14). Shungu 2012 (third cohort, 15 vs 13): raised lactate and lower cortical glutathione, not different from major depression. Natelson 2017 replicated higher lactate, lower GSH and lower CBF in pooled CFS. Mueller 2020 (Younger group, whole-brain EPSI, 15 vs 15): raised LAC/CR in some regions, overlapping with raised temperature. Bader 2026 (Vienna, 26 vs 27, preprint): thalamic Lac/tCr 0.171 vs 0.135 at normoxia, with no further rise under hypoxia, unlike controls. Bradley 2016 found raised ventricular lactate in adolescent depression. Small samples, MRS quantification at 3T near detection limits, and the lack of specificity limit interpretation.

Evidence

Study	n	Design	Result
Mathew 2009, NMR Biomed	16 CFS, 14 GAD, 15 HC	case-control 1H MRSI	Ventricular lactate +348% vs HC
Shungu 2012, NMR Biomed	15 CFS, 15 MDD, 13 HC	case-control MRSI + ASL + 31P	Raised lactate, lower GSH, lower rCBF; CFS ~ MDD
Natelson 2017, J Neurol Sci	CFS +/- psychiatric comorbidity vs HC	case-control multimodal	Replicated raised lactate, lower GSH and CBF; 13/26 had >=2 abnormal brain measures
Bader 2026, medRxiv (preprint)	26 vs 27	case-control MRS with hypoxic challenge	Thalamic lactate raised at rest, no rise with hypoxia

Against or not replicated

Study	n	Design	Result
Bradley 2016, Eur Psychiatry	23 MDD vs 29 HC	case-control MRSI	Raised ventricular lactate in adolescent depression (shows non-specificity)

Open questions

Does brain lactate rise further after exertion (PEM)?
Is it from hypoperfusion, mitochondrial limitation or glial activation?

Bears on: Mitochondrial / energy metabolism impairment Neuroinflammation (microglial activation) Capillary and microvascular damage · Test: research only

MRS thermometry found raised temperature in several brain regions (insula, putamen, frontal cortex, thalamus, cerebellum) in women with ME/CFS, not explained by body temperature. Early

One study found parts of the brain running slightly warmer in patients, which could signal low-grade inflammation. It has not been repeated.

Mueller 2020: 15 vs 15 women, whole-brain echo-planar spectroscopic imaging. Temperature was raised in the right insula, putamen, frontal cortex, thalamus and cerebellum (p<0.05, many regions tested), along with raised choline/creatine in the left anterior cingulate. The method responded to an experimental inflammatory challenge (typhoid vaccine) in 44 of 47 regions, but whole-brain temperature did not change (Plank 2022), so its sensitivity is modest. No independent ME/CFS replication was found.

Evidence

Study	n	Design	Result
Mueller 2020, Brain Imaging Behav	15 vs 15	case-control whole-brain MRSI	Raised regional brain temperature; raised CHO/CR

Against or not replicated

Study	n	Design	Result
Plank 2022, Neuroimage Clin	20 healthy	RCT crossover, typhoid vaccine	Regional but not whole-brain temperature rose (method validity context)

Open questions

Independent replication
Correlation with TSPO-PET in the same people

Bears on: Neuroinflammation (microglial activation) · Test: research only

PET neuroinflammation

TSPO-PET evidence of microglial activation in ME/CFS is contested: one Japanese study found 45-199% higher binding in limbic and brainstem regions, and a Dutch study using the same tracer found none. Contested

A special scan for brain immune cells showed inflammation in one small study and none in another of the same size. The question is open.

Nakatomi 2014: [11C]-(R)-PK11195, 9 vs 10. BPND was higher in the cingulate, hippocampus, amygdala, thalamus, midbrain and pons, and correlated with cognitive impairment and pain. Raijmakers 2021: same tracer, 9 CFS, 10 Q-fever fatigue and 9 matched women, stricter exclusion. No difference, and a negative correlation with severity in CFS. Both used the cerebellum as reference region, which can bias results if the reference is also affected. PK11195 has a low signal-to-noise ratio, and second-generation tracers need TSPO genotyping. Stanford DPA-713 data exist only as a conference abstract (not verified). A Belgian [18F]DPA714 study is in protocol (Dooms 2026).

Evidence

Study	n	Design	Result
Nakatomi 2014, J Nucl Med	9 vs 10	case-control PET	BPND +45-199% in limbic/brainstem regions

Against or not replicated

Study	n	Design	Result
Raijmakers 2021, Neurol Neuroimmunol Neuroinflamm	9 CFS, 10 QFS, 9 HC	case-control PET	No difference in BPND

Open questions

Adequately powered study with a second- or third-generation tracer, arterial input and genotyping
Does TSPO signal change during PEM?

Bears on: Neuroinflammation (microglial activation) Persistent sickness behaviour / vagal signalling · Test: research only

In long COVID, one [11C]PBR28 PET study found widespread raised TSPO signal that correlated with vascular markers, and a [11C]PK11195 study found no group difference. Contested

Long COVID brain-inflammation scans also disagree. One found widespread signal and another found none.

VanElzakker 2024: 12 PASC vs 43 normative controls. Raised PBR28 in the mid and anterior cingulate, corpus callosum, thalamus, basal ganglia and periventricular regions, correlated with plasma vascular-dysfunction analytes. The comparison against a pooled normative control set and the small n are limits. Tuomaala 2026: 14 LC vs 11 HC vs 13 MS. TSPO DVR was not raised in LC, though higher in people imaged within 16 months, and limbic DVR correlated with depression and anxiety. An earlier Braga 2023 report in long COVID with depression/cognitive symptoms was not verified this session.

Evidence

Study	n	Design	Result
VanElzakker 2024, Brain Behav Immun	12 vs 43	case-control PET	Widespread raised PBR28; correlates with vascular markers

Against or not replicated

Study	n	Design	Result
Tuomaala 2026, J Neurol	14 LC, 11 HC, 13 MS	case-control PET	No TSPO elevation vs HC; time-since-infection effect

Open questions

Does neuroinflammation fade with time since infection?
Is it seen in long COVID that meets ME/CFS criteria specifically?

Bears on: Neuroinflammation (microglial activation) Viral persistence Endothelial dysfunction · Test: research only

Cerebral perfusion

Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS studies, but the results are inconsistent and often confounded by unmeasured orthostatic intolerance. Contested

Lying-down brain blood flow is sometimes lower, but not consistently. The clearest effect appears when patients are upright.

Shungu 2012 (ASL) found lower rCBF in the left anterior cingulate and right lingual gyrus vs controls, but not vs depression. Natelson 2017 pooled CFS had lower CBF. Christopoulos 2025 review: only 12 of 26 ME/CFS-only studies found reduced CBF, and most did not record OI status. Bader 2026: whole-brain CBF responses to hypoxia were similar in mean but far more variable in patients (variance ratio up to 6.9).

Evidence

Study	n	Design	Result
Shungu 2012, NMR Biomed	15 vs 13	ASL case-control	Lower regional CBF in 2 regions
Christopoulos 2025, J Transl Med	26 ME/CFS-only studies	systematic review	12/26 reduced

Against or not replicated

Study	n	Design	Result
Bader 2026, medRxiv (preprint)	26 vs 27	ASL under normoxia/hypoxia	No mean group difference; higher variability in patients

Open questions

Resting CBF stratified by OI and hypocapnia status

Bears on: Capillary and microvascular damage Endothelial dysfunction Low blood volume and preload failure · Test: research only (ASL MRI)

Cerebrospinal fluid

In the NIH cohort, CSF levels of the catechol metabolites DOPA, DOPAC and DHPG were lower in post-infectious ME/CFS. Noradrenaline and dopamine did not differ. Early

Spinal fluid showed lower breakdown products of dopamine and noradrenaline, which hints at reduced activity of these brain chemical systems. It comes from one small study.

Walitt 2024: CSF DOPA, DOPAC and DHPG were reduced; noradrenaline, cys-DOPA and dopamine were unchanged. Excluding participants on central-acting drugs did not change the result. Catechols correlated with perceived symptoms and, in patients, with only one cognitive test (WCST). CSF metabolomics also showed lower glutamate, tryptophan, spermidine and citrate. n was about 17 vs 21, with many outcomes and no correction across domains. Solving ME/CFS re-analysis: plasma VMA showed nothing, which fits a central rather than peripheral signal but is weak support.

Evidence

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	~17 vs 21	case-control CSF	Lower DOPA, DOPAC, DHPG

Open questions

Replication in an independent CSF cohort
Relation to orthostatic phenotype and HRV

Bears on: Neuroinflammation (microglial activation) Persistent sickness behaviour / vagal signalling Autonomic dysfunction and small-fibre neuropathy · Test: research only (lumbar puncture with catechol panel)

CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct immunotypes), but no CSF protein signature has replicated, and ME/CFS and fibromyalgia CSF proteomes were indistinguishable. Contested

Spinal-fluid tests have hinted at immune changes, but different studies point in different directions, and none gives a reliable marker.

Baraniuk 2005: pooled and individual CSF proteomics (about 10-20 per group) gave a 5-protein set with 80% concordance; pilot only. Hornig 2016 (online 2015): 32 cases vs 40 MS vs 19 controls, 51-plex. CCL11 was raised and the IL-1 network disturbed, with interpretation hindered by the small control group. Schutzer 2023: 2083 proteins quantified; no protein differed between ME/CFS with and without fibromyalgia. Bastos 2025 (40 cases, CSF in cases only) found two MMP-defined clusters. Bragée 2026 (31 cases, no CSF controls) found complement/coagulation pathways associated with severity and POTS. Solving ME/CFS re-analysis: the Walitt CSF SomaScan set gives 0 FDR hits, and CSF neuronal markers do not replicate.

Evidence

Study	n	Design	Result
Hornig 2016, Mol Psychiatry	32 vs 19 HC (40 MS)	case-control 51-plex	CCL11 up; IL-1 network disturbance
Baraniuk 2005, BMC Neurol	10 CFS/10 GWI/10 HC pooled; 9 vs 12	pilot proteomics	5-protein set, 80% concordance
Bastos 2025, J Immunol	40 cases (CSF), 41 HC (plasma)	cases-only CSF clustering	Two MMP-based CSF immunotypes
Bragée 2026, Sci Rep data	31 cases	cases-only CSF proteomics	Complement/coagulation pathways linked to severity; neutrophil degranulation with POTS

Against or not replicated

Study	n	Design	Result
Schutzer 2023, Ann Med	15 vs 15 (ME/CFS +/- FM)	CSF proteomics, 2083 proteins	No proteins differed: indistinguishable
Solving ME/CFS re-analysis (Walitt 2024 CSF SomaScan, GEO GSE251790)	~17 vs 21	re-analysis	0 FDR hits; no case-control leak-score difference

Open questions

A CSF study with adequate healthy controls (n>50) and pre-registered analysis
Paired pre/post-exertion CSF

Bears on: Neuroinflammation (microglial activation) Innate immune hyperactivation Complement activation and thromboinflammation · Test: research only

Targeted CSF metabolomics found raised serine and phospholipids and reduced 5-methyltetrahydrofolate in ME/CFS, with different metabolite responses to exercise than in controls. Early

Spinal fluid chemistry suggests changes in a vitamin/amino-acid pathway (folate and serine), possibly affecting brain fats. These are single-group results.

Baraniuk 2025 (Sci Rep; separate non-exercise and post-exercise LP cohorts): ME/CFS showed raised serine, sarcosine, creatine, purines and sphingomyelins and lower 5MTHF, dimethylglycine and choline. After exercise, metabolites were consumed in ME/CFS but generated in controls. The design is cross-sectional with separate cohorts per condition, not paired. Earlier Baraniuk 2021 work found little non-exercise ME/CFS vs GWI difference. NIH found lower CSF glutamate, tryptophan and spermidine (Walitt 2024), so the two sets point in different directions.

Evidence

Study	n	Design	Result
Baraniuk 2025, Sci Rep data	ME/CFS vs sedentary controls, rest and post-exercise cohorts	case-control targeted metabolomics	Serine up, 5MTHF down; divergent exercise response

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	~17 vs 21	CSF metabolomics	Different pattern: lower glutamate, tryptophan, spermidine, citrate

Open questions

Replication; paired pre/post-exercise sampling in the same people

Bears on: Mitochondrial / energy metabolism impairment IDO metabolic trap (Phair, Davis) · Test: research only

Markers of neuronal or glial injury (NfL, GFAP, tau, UCH-L1) are not raised in ME/CFS blood or CSF. Several studies

Blood and spinal-fluid tests that pick up nerve-cell damage in diseases like MS or dementia are normal in ME/CFS. That argues against ongoing destruction of brain tissue.

Walitt 2024 measured Simoa Neurology 4-Plex (NfL, tau, GFAP, UCH-L1) in plasma and CSF: no group difference. Solving ME/CFS re-analysis of SomaScan/Olink proteomics across about 150 patients: NEFL and GFAP flat or inconsistent (neuronal_glial set -0.25, p=0.07). Long COVID TSPO-PET (Tuomaala 2026) also measured serum NfL/GFAP, without a clear LC effect. A normal result does not exclude functional or neuroinflammatory changes without cell death.

Evidence

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	~17 vs 21	case-control Simoa	No difference in plasma or CSF NfL/GFAP/tau/UCH-L1
Solving ME/CFS re-analysis (proteomics) data	~150 patients across cohorts	re-analysis	NEFL/GFAP not different

Open questions

Do levels rise transiently during PEM or in severe/bedbound patients?

Bears on: Neuroinflammation (microglial activation) Genetic susceptibility acting through neurons · Test: Serum NfL/GFAP increasingly available clinically (mainly to exclude other disease)

Craniocervical / intracranial pressure

Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying tonsils and signs of raised intracranial pressure in severe ME/CFS, and surgical case reports describe improvement. There are no controlled imaging studies or trials. Early

Some clinics report that many severe patients have crowding where the skull meets the neck, or high spinal-fluid pressure, and a few patients improved after neck surgery. Without comparison groups or controlled trials, it is unknown whether these findings are more common than in healthy people or whether surgery helps.

Bragée 2020 (Swedish referral clinic, 229 CCC patients, retrospective): hypermobility 50%, enlarged optic-nerve-sheath ratio 55%, possible IH signs 83%, tonsils below McRae line 56%, and cervical 'craniocervical obstructions' 80% of 125 scanned. No controls; MRI reads were not blinded and the radiological thresholds are not validated. Jolley 2026 (34 cases, no controls): LP opening pressure >=20 cmH2O in 38%; that group reported more relief after LP (80% vs 19%, unblinded). Rowe/Edwards case reports (3+1 patients) describe improvement after cervical decompression. Solving ME/CFS re-analysis of Finnish registry data: CSF-leak codes are not specific (36% of LPs in ME/CFS vs 32% in fibromyalgia, 45% in hypermobility). Neurosurgical series (Ruhoy 2025, n=717) are in connective-tissue-disorder cohorts, not ME/CFS. The evidence quality is low, and surgery carries serious risk.

Evidence

Study	n	Design	Result
Bragée 2020, Front Neurol	229 (205 brain MRI, 125 c-spine MRI)	retrospective uncontrolled series	80% craniocervical obstruction; 83% possible IH signs; 56% low tonsils
Jolley 2026, Front Med	34	uncontrolled case series	38% opening pressure >=20 cmH2O; more post-LP relief in high-OP group
Edwards 2023, Front Neurol	1 (follow-up of 3-case series)	case report	Symptoms returned with recurrent stenosis and improved after re-operation

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (FinnGen/FinRegistry) data	~1,600 G93.3 cases	registry co-occurrence	CSF-leak codes equally common in fibromyalgia/hypermobility; not ME/CFS-specific

Open questions

Blinded, controlled upright/dynamic MRI comparing ME/CFS, hypermobile and healthy controls
Any randomised or sham-controlled evaluation of surgery or CSF drainage

Bears on: Brainstem and craniocervical instability Connective tissue / hypermobility · Test: Clinical cervical MRI is available; upright/flexion MRI and CCI surgery are offered privately with no controlled evidence

Effort and motivation

The NIH claim that altered 'effort preference' defines post-infectious ME/CFS rests on the EEfRT (OR 1.65 for choosing hard tasks). Patients completed only 65% of the hard tasks they chose vs 96% for controls, which confounds preference with ability. Contested

The NIH team said patients chose the harder button-pressing task less often and called this 'effort preference'. Critics showed that many patients could not physically finish the hard task, so the choice more likely reflects ability than wanting to avoid effort.

Walitt 2024: 15 PI-ME/CFS vs 16 HV on EEfRT. Controls were 1.65x more likely to choose hard (95% CI 1.03-2.65, p=0.04), and no group difference in task-related fatigue response was reported. Kirvin-Quamme 2025 re-analysed mapMECFS data. Hard-task completion was 65% (SD 37) in patients vs 96% (SD 8) in controls; 7/15 patients completed fewer hard tasks than any control. Choice correlated with completion (rs 0.38). This fails the EEfRT's own ability check (Treadway 2009). Davenport 2025 (Nat Commun Matters Arising) also criticised the single-day CPET, matching for deconditioning and generalisation from n=17. The NIH reply defended single CPET on safety grounds. The effort result also sits on p=0.04 with no multiplicity correction.

Evidence

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	15 vs 16 (EEfRT)	case-control	OR 1.65 for hard-task choice in controls
Walitt 2025 reply, Nat Commun	n/a	reply	Defends single CPET and interpretation

Against or not replicated

Study	n	Design	Result
Kirvin-Quamme 2025, Front Psychol	15 vs 16 (re-analysis)	secondary analysis of public data	Hard-task completion 65% vs 96%; choice confounded with ability
Davenport 2025, Nat Commun (Matters Arising)	n/a	critique	Single CPET cannot assess PEM; deconditioning/effort conclusions not supported

Open questions

An effort-based decision task calibrated to each participant's ability
Relation of motor fatigability to peripheral muscle findings

Bears on: Persistent sickness behaviour / vagal signalling · Test: research only

Cognition

Meta-analysis shows moderate-to-large deficits in information-processing speed, attention, working memory and reaction time in ME/CFS, with intact vocabulary and reasoning. Performance worsens with upright posture and after exertion. Replicated

'Brain fog' shows up on tests as slower processing and difficulty holding information, especially over sustained effort, rather than loss of intelligence or knowledge.

Cockshell & Mathias 2010: 50 studies, deficits mainly in attention, memory and reaction time, with fine motor speed, vocabulary and reasoning spared. On tilt, n-back reaction time rose from 96 to 148 ms in CFS but not controls (Medow 2014). Post-exercise, PASAT performance fell (Cook 2017). The NIH study found no differences on any of 15 neuropsychological tests despite large subjective complaints (n=17), which shows that small samples and unprovoked testing can miss deficits. Reaction-time variability and sustained-attention tasks are more sensitive than standard batteries.

Evidence

Study	n	Design	Result
Cockshell & Mathias 2010, Psychol Med	50 studies	meta-analysis	Moderate-large deficits in processing speed, attention, memory, RT
Medow 2014, J Appl Physiol	15 vs 11	case-control tilt	Orthostatic worsening of n-back; reversed by phenylephrine

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control, 15 tests	No group differences in objective performance

Open questions

A standard provocation protocol (upright, post-exertion) for clinical cognitive testing
Contribution of sleep, medication and orthostatic hypoperfusion

Bears on: Low blood volume and preload failure Neuroinflammation (microglial activation) Persistent sickness behaviour / vagal signalling · Test: Clinical neuropsychology; computerised reaction-time tests; tests done upright or after exertion are research only

Sleep

Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep onset and lower sleep efficiency. Standard polysomnography usually finds no clinical disorder that explains non-restorative sleep. Several studies

Sleep studies show patients take longer to fall asleep and wake more, but the differences are small. They do not explain why sleep feels so unrefreshing.

Mohamed 2023 meta-analysis: 20 adult studies (426 vs 375) showed more time in bed, longer SOL and WASO, lower efficiency, less N2, more N3 and longer REM latency. In 4 adolescent studies (242 vs 235), total sleep was longer. Heterogeneity was high, and the stage-level findings are inconsistent with older reviews (Jackson & Bruck 2012: few PSG differences; microstructure, such as instability and alpha intrusion, more promising). Walitt 2024: PSG without clinically relevant findings. Solving ME/CFS re-analysis of Finnish registries: hypersomnia codes are enriched about 20-50x and sleep apnoea precedes diagnosis (HR 15), probably reflecting diagnostic work-up as well as biology. Primary sleep disorders still need excluding clinically.

Evidence

Study	n	Design	Result
Mohamed 2023, Sleep Med Rev	24 studies; 668 vs 610	systematic review and meta-analysis	Lower efficiency, longer SOL/WASO in adults
Jackson & Bruck 2012, J Clin Sleep Med	review	narrative review	Few PSG differences; microstructure/HRV in sleep promising

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17	clinical PSG	No clinically relevant findings

Open questions

What makes sleep non-restorative: autonomic arousal, microstructure, or glymphatic function?
No ME/CFS trials of orexin antagonists or low-dose sedating agents

Bears on: Autonomic dysfunction and small-fibre neuropathy Neuroinflammation (microglial activation) HPA axis blunting · Test: Clinical polysomnography (to exclude apnoea, PLMs, narcolepsy)

Sensory sensitivity

Self-reported light and sound sensitivity is common in ME/CFS. One ERP study found normal early sensory gating (P50) but a reduced P300, pointing to higher-order processing rather than early filtering. Early

Many patients find light and noise hard to tolerate. The single brain-wave study suggests that the basic filtering of sounds is normal and that later attention steps are affected.

Kumar 2026: 31 vs 30, auditory paired-click and oddball paradigms. Patients reported more sensory problems; the P50 suppression index did not differ, while P300 amplitude was reduced. Severe and very severe patients report extreme sensitivity (Vink 2026 review), but they are largely absent from laboratory studies. There are no controlled studies of photophobia thresholds, hyperacusis audiometry or migraine-type central sensitisation specific to ME/CFS. The genetic overlap with chronic pain and migraine (Solving ME/CFS genetics re-analysis) is consistent with shared central sensitisation.

Evidence

Study	n	Design	Result
Kumar 2026, Front Med	31 vs 30	case-control ERP	Normal P50 gating; reduced P300
Vink & Vink-Niese 2026, J Clin Med	review	narrative review of severe ME	Extreme light/sound sensitivity described in severe ME

Open questions

Quantitative sensory thresholds in severe patients (home-based measurement)
Overlap with migraine and fibromyalgia central sensitisation

Bears on: Neuroinflammation (microglial activation) Genetic susceptibility acting through neurons Persistent sickness behaviour / vagal signalling · Test: research only

Key reviews

Shan 2020, J Transl Med: Neuroimaging characteristics of ME/CFS, systematic review: Sorts which brain-imaging findings are repeated across teams and which are single reports.
Christopoulos 2025, J Transl Med: Mapping cerebral blood flow in ME/CFS and OI: Up-to-date systematic review of cerebral blood flow, including the OI confound.
Nelson 2019, Medicine: altered cardiac autonomic regulation in ME/CFS, meta-analysis: Quantifies resting HR and HRV differences across 64 studies.
Mohamed 2023, Sleep Med Rev: Objective sleep measures in CFS, meta-analysis: Best summary of PSG/actigraphy differences, adults vs adolescents.
Cockshell & Mathias 2010, Psychol Med: Cognitive functioning in CFS, meta-analysis: Defines the cognitive deficit profile.
VanElzakker 2019, Front Neurol: Neuroinflammation and cytokines in ME/CFS, critical review of methods: Explains the pitfalls of TSPO-PET, MRS and cytokine studies of 'neuroinflammation'.
Kirvin-Quamme 2025, Front Psychol: Unwilling or unable?: Clearest data-based critique of the NIH effort-preference claim.

Immune system

NK cells

Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older literature (Hedges g ~0.96), but the largest standardised multi-site study and two deep-phenotyping studies found no difference. Contested

The most-repeated immune finding in ME/CFS is that one type of white blood cell kills target cells less well. When labs tried to turn it into a standardised test, the difference disappeared.

Baraniuk, Eaton-Fitch and Marshall-Gradisnik 2024 pooled 28 papers and 55 effector:target data points: cytotoxicity about half of control levels, overall Hedges g 0.96 (0.75-1.18), heterogeneity high and explained by E:T ratio and method. Against this: the CDC Multi-Site Clinical Assessment (MCAM) sub-study used an assay validated for overnight-shipped blood in 174 ME/CFS vs 86 healthy controls and found 34.1% vs 33.6% cytotoxicity (p=0.79), with no association with symptom domains; Theorell 2017 found no reproducible difference in two cohorts (48 patients) across granule content, exocytosis, killing or adrenaline sensitivity; Cliff 2019 found no difference in NK numbers, subsets or in vitro responsiveness in 251 patients (54 severe) vs 107 controls. The assay requires fresh blood and is highly method-sensitive, so publication-era and logistics effects are plausible explanations for the discrepancy. Treat reduced NK function as a real signal in some cohorts with some assays, not as a disease marker.

Evidence

Study	n	Design	Result
Baraniuk, Eaton-Fitch & Marshall-Gradisnik 2024, Frontiers in Immunology	28 papers, 55 E:T data points	meta-analysis	Hedges g 0.96 (0.75-1.18); cytotoxicity ~50% of controls
Klimas 1990, J Clin Microbiol	30 patients vs controls	case-control	Low NK cytotoxicity per CD56 cell despite raised CD56 numbers; reduced mitogen responses and IFN-gamma
Brenu 2011, J Transl Med	95 vs 50	case-control	NK and CD8 cytotoxicity reduced, CD56bright reduced, granzyme A/K down, perforin up
Eaton-Fitch 2019, Systematic Reviews	17 studies	systematic review	Cytotoxicity consistently reduced; other NK outcomes equivocal

Against or not replicated

Study	n	Design	Result
Querec 2023 (MCAM), J Transl Med	174 ME/CFS vs 86 HC vs 10 ill controls	multi-site case-control, shipped-sample validated assay	34.1% vs 33.6% cytotoxicity, p=0.79; no symptom association; 'not ready for clinical implementation'
Theorell 2017, Frontiers in Immunology	48 patients, two cohorts, matched HC	case-control, deep cytotoxic-lymphocyte phenotyping	No reproducible difference in number, granules, exocytosis, killing, cytokines or adaptive NK expansion
Cliff 2019, Frontiers in Immunology	251 ME/CFS (54 severe) vs 107 HC vs 46 MS	case-control, UK ME/CFS Biobank	No differences in NK numbers, subsets or in vitro responsiveness

Open questions

Does the deficit appear only in fresh same-day assays, i.e. is it partly a pre-analytic artefact?
Is reduced cytotoxicity confined to a severity or herpesvirus-exposure subgroup?
Does NK function change during PEM (no paired rest/post-exertion study exists)?

Bears on: Immune (T-cell) exhaustion Herpesvirus reactivation (EBV, HHV-6) Viral persistence · Test: Private NK-function assays exist but need a same-day fresh sample; not validated for diagnosis (MCAM explicitly says so)

Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been reported repeatedly, including a two-site study (36 vs 42), but only ever by one research group. Early

One Australian group finds that a calcium channel on immune cells works poorly in patients, which they argue explains weak NK killing. No other lab has tested it.

Sasso 2026 (Front Med) ran whole-cell patch-clamp on isolated NK cells at two laboratory sites with independent investigators: 36 CCC-defined ME/CFS vs 42 healthy controls, significant reduction in all TRPM3 parameters with no site effect. Earlier work in the same programme: Eaton-Fitch 2022 (impaired TRPM3-dependent calcium influx, restored in vitro by naltrexone), Eaton-Fitch 2021 (TRPM3/PIP2 co-localisation), Magawa 2026 (deficient TRPM3-linked mitochondrial calcium influx), plus a post-COVID extension (Magawa 2026, Eur J Immunol) and LDN restoration in long COVID NK cells (Sasso 2025). The 'two sites' are both within the same group's network, so this is internal consistency rather than independent replication; patch-clamp of isolated NK cells is a demanding, low-throughput method available in very few labs. Against it indirectly: DecodeME common-variant genetics shows no enrichment in any blood immune cell type, and the low-dose naltrexone RCT in post-COVID ME/CFS (n=137) was negative, which weakens the therapeutic corollary.

Evidence

Study	n	Design	Result
Sasso 2026, Frontiers in Medicine	36 vs 42	two-site case-control patch-clamp	Significant reduction in all TRPM3 current parameters; no effect of laboratory site
Eaton-Fitch 2022, J Transl Med	small case-control	calcium imaging	Impaired TRPM3-dependent Ca2+ influx; restored in vitro by naltrexone
Magawa 2026, BMC Immunology	case-control	mitochondrial Ca2+ imaging in NK cells	Deficient TRPM3-linked mitochondrial Ca2+ influx

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	503,739 PBMCs, 58 people; 5 cohorts	re-analysis of public scRNA, methylation, repertoires	No immune-cell subtype, gene or methylation site separates cases from controls reproducibly; nothing passed FDR in more than one dataset
Solving ME/CFS re-analysis (downloads, DecodeME MAGMA) data	15,579 cases vs 259,909 controls	GWAS enrichment	No enrichment in any of 29 immune cell types; signal is brain/neuronal

Open questions

Can a lab with no stake in the hypothesis reproduce the patch-clamp result blind?
Is reduced TRPM3 current a cause of poor cytotoxicity or a downstream consequence of cell state/handling?
Does TRPM3 function track symptoms within a person over time?

Bears on: Sodium-calcium overload in muscle (Wirth & Scheibenbogen) Immune (T-cell) exhaustion Autonomic dysfunction and small-fibre neuropathy · Test: research only

T cells

CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exhaustion, with supporting flow cytometry and an independent cell-free RNA signature. Several studies

Killer T cells in patients look 'worn out', like cells that have been fighting an infection for a long time.

Iu 2024 (PNAS) integrated scRNA-seq, bulk RNA-seq and ATAC-seq across sorted T-cell subsets (336,269 T-lineage cells from 28 patients and 30 sedentary controls at baseline and after symptom provocation), finding upregulated exhaustion transcription factors, altered chromatin accessibility and metabolic reprogramming most marked in CD8+ effector-memory and innate-like T cells, validated by flow cytometry in an extended cohort (15 patients, 11 controls, female, age/BMI matched). Gardella 2025 independently found T-cell-exhaustion and cytokine-signalling signatures in plasma cell-free RNA (93 vs 75). Supporting but not identical: Mandarano 2020 found reduced CD8+ mitochondrial membrane potential and reduced glycolysis at rest in 53 vs 45. The main caveat is that the Cornell scRNA cohorts are not independent of each other and a re-analysis of the same public scRNA data found no exhausted-CD8 composition difference once cell typing was redone, so the effect may live in gene programmes rather than cell counts.

Evidence

Study	n	Design	Result
Iu 2024, PNAS	28 vs 30 (scRNA/ATAC); 15 vs 11 flow validation	multiomic case-control with exercise provocation	Exhaustion TFs up in CD8+ Tem, altered chromatin and metabolism; higher frequency of exhaustion markers by flow
Gardella 2025, PNAS data	93 vs 75	plasma cfRNA sequencing + ML	Cytokine-signalling and T-cell-exhaustion network signatures; classifier AUC 0.81
Mandarano 2020, J Clin Invest	53 vs 45	case-control immunometabolism	Reduced CD8+ mitochondrial membrane potential; reduced resting glycolysis in CD4+ and CD8+
Yin 2024, Nature Immunology (long COVID)	deep immune profiling, 8 months post-infection	case-control omics + serology	Exhausted SARS-CoV-2-specific CD8+ T cells, systemic inflammation, uncoordinated T/B responses

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/singlecell_methyl, GSE214283)	30 vs 28, 503,739 PBMCs	re-analysis with CellTypist + marker gating	Exhausted-CD8 gating null; 0 genes at q<0.05 at baseline in all 15 cell types

Open questions

Is the exhaustion programme driven by a persisting antigen, and if so which?
Does checkpoint blockade or metabolic rescue change anything in vivo (never tried)?
Why do composition-level and programme-level analyses of the same data disagree?

Bears on: Immune (T-cell) exhaustion Viral persistence Herpesvirus reactivation (EBV, HHV-6) · Test: research only (specialist flow cytometry)

T-cell abnormalities differ by severity: mild/moderate patients show cytotoxic/immunosenescence features, severe patients show activated lymphocytes and higher stimulated pro-inflammatory cytokines. Early

Mildly and severely affected patients show different immune pictures, which may mean different disease mechanisms rather than one disease of different intensity.

Lee, Cliff and colleagues (UK ME/CFS Biobank; medRxiv 2025, J Transl Med Dec 2025) compared 43 mild/moderate with 53 severe patients by flow cytometry. Mild/moderate: increased cytotoxic effector molecules and more CD28-CD57- early-immunosenescent cells, a pattern associated with persistent viral infection. Severe: more CD69+ and CD38+ activated circulating lymphocytes and more IFN-gamma, TNF and IL-17 after in vitro stimulation, consistent with non-specific ongoing inflammation. Changes spanned CD8+ T cells, MAIT cells and NK cells. This is a within-patient-group comparison without a healthy-control contrast for the severity axis, and severe patients differ in sampling conditions (home visits, deconditioning, medication), so confounding by illness state is not excluded.

Evidence

Study	n	Design	Result
Lee JS 2025, J Transl Med (also medRxiv 2025.01.02.24319359)	43 mild/moderate vs 53 severe	case-case flow cytometry, UK ME/CFS Biobank	Immunosenescence/cytotoxic markers in mild-moderate; activation (CD69+, CD38+) and higher stimulated IFN-gamma/TNF/IL-17 in severe

Open questions

Do these two patterns define stable subgroups or stages of one illness?
Does the severe-patient activation pattern reflect being bedbound rather than disease mechanism?

Bears on: Immune (T-cell) exhaustion Innate immune hyperactivation Persistent sickness behaviour / vagal signalling · Test: research only

Increased mucosal-associated invariant T (MAIT) cell proportions and a shift in CD8+ memory subsets were found in the largest single ME/CFS flow-cytometry cohort, but the finding has not been reproduced in independent data. Contested

One large study found more of a gut-associated T-cell type in patients, especially severe ones. Re-analyses of other datasets do not show it.

Cliff 2019 (251 patients incl. 54 severe, 107 controls, 46 MS) reported increased effector-memory CD8+ and decreased terminally differentiated effector CD8+ T cells, plus a significantly increased proportion of MAIT cells, most marked in severe patients, with no differences in herpesvirus seroprevalence. Iu 2024 also flagged 'certain innate T cells' as among the most dysregulated. Against: a re-analysis across five cohorts and three data types (scRNA of 503,739 PBMCs, four methylation cohorts, B- and T-cell repertoires) found MAIT, gamma-delta T, Th17 and exhausted-CD8 gating all null, and blood RNA cohorts do not show reduced Th17 (meta p=0.39).

Evidence

Study	n	Design	Result
Cliff 2019, Frontiers in Immunology	251 vs 107 vs 46 MS	case-control flow cytometry	MAIT proportion increased (most in severe); CD8+ Tem up, TEMRA down

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/singlecell_methyl; downloads)	5 cohorts; 58-people scRNA; 5 blood RNA cohorts	re-analysis	MAIT, gamma-delta, Th17, ABC B cells and pDC interferon all null; Th17 meta p=0.39

Open questions

Is the MAIT signal a gut-mucosa readout, and does it track gut findings in the same people?

Bears on: Gut dysbiosis Immune (T-cell) exhaustion · Test: research only

T-cell receptor repertoires do not distinguish ME/CFS from controls, arguing against a dominant shared antigen-driven clonal expansion. Refuted

If one infection or self-target were driving the disease, you would expect to see armies of identical T cells. Sequencing says they are not there.

Dibble 2024 (BMC Research Notes) sequenced alpha/delta, beta and gamma TCR loci in 40 samples each from severe ME, mild/moderate ME, MS disease controls and healthy controls, and applied P-SVM classification: the method worked on simulated data but could not partition the real samples significantly. A re-analysis of UK ME/CFS Biobank T-cell repertoires (304 samples) likewise found nothing passing FDR in more than one dataset, with only exploratory herpesvirus-related trends (higher share of EBV-specific TCR-beta in severe patients' CD8, 0.88% vs 0.45%, p=0.06). A near-genome-wide candidate, the TRA locus (Ueland 2022), has not replicated either.

Evidence

Study	n	Design	Result
Dibble 2024, BMC Research Notes	4 groups x 40	bulk TCR repertoire sequencing + machine learning	No statistically significant partitioning; 'findings do not support altered T-cell receptor diversity'
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	304 T-cell repertoire samples	re-analysis, UK ME/CFS Biobank	Nothing reproducible; EBV-specific TCR-beta share higher in severe CD8 at p=0.06 only

Open questions

Would antigen-specific sorting (rather than bulk repertoire) reveal expansions invisible in bulk?
Could tissue (muscle, gut, brainstem) repertoires differ while blood looks normal?

Bears on: Autoimmunity (functional autoantibodies) Viral persistence · Test: research only

B cells

The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-infectious ME/CFS, consistent with chronic antigenic stimulation; supporting signals exist but are weak. Early

Patients seem to have more 'untrained' B cells and fewer that have already matured, as if the immune system keeps being prodded by something.

Walitt 2024 (Nat Commun) did intensive inpatient phenotyping of a small, rigorously selected post-infectious ME/CFS cohort (17 PI-ME/CFS vs 21 healthy volunteers for immune measures) and reported increased naive and decreased switched-memory B cells, interpreted as chronic antigenic stimulation and linked to STAT4-TLR9 B-cell pathway signals; PBMC gene-expression and metabolic differences were sex-dependent. A Solving ME/CFS re-analysis found plasma-cell/immunoglobulin transcripts lower in 2 of 4 blood RNA cohorts (meta d -0.42, p=0.06), which points the same way without reaching significance. Earlier B-cell phenotype work (Mensah 2016, 2018) found no reproducible subset differences but raised CD24 expression on naive and memory B cells. The cohort is tiny and much of Walitt 2024's framing (effort preference) is contested, so this should be read as a lead, not an established feature.

Evidence

Study	n	Design	Result
Walitt 2024, Nature Communications nih-walitt-2024 mwb-st003178	17 PI-ME/CFS vs 21 HV (immune measures; 16 vs 20 for switched memory)	deep phenotyping case-control	Naive B cells increased, switched-memory B cells decreased; interpreted as chronic antigenic stimulation
Solving ME/CFS re-analysis (deep2/proteome) data	4 blood RNA cohorts	meta-analysis of public data	Plasma-cell/immunoglobulin transcripts lower in 2 of 4 cohorts, meta d -0.42, p=0.06
Mensah 2016, Clin Exp Immunol	cross-sectional case-control	extended B-cell immunophenotyping	No reproducible subset differences; higher CD24 expression within naive and memory B cells

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	58-person scRNA; 123-person BCR repertoire	re-analysis	No B-cell composition difference survived; ABC B-cell call lower only at p=0.03, not replicated

Open questions

Does the naive/switched-memory shift replicate in a larger, less selected cohort?
Is it explained by medication, inactivity or infection history?

Bears on: Autoimmunity (functional autoantibodies) Viral persistence Immune (T-cell) exhaustion · Test: Lymphocyte subset flow cytometry is widely available, but the naive/memory B-cell split is not a validated marker

Reported skewing of the B-cell receptor repertoire (IGHV3-30/3-33 overuse) partially replicates in mild/moderate patients but shows no clonality, somatic-hypermutation or autoimmunity signature. Contested

An early report said patients over-use certain antibody genes. A careful follow-up found a trace of that but none of the signs you would expect if antibodies were driving the disease.

Sato 2021 (Brain Behav Immun) reported biased IGHV/IGHD/IGHJ usage in ME/CFS peripheral B cells with ROC separation from controls. Ryback and Cowan 2025 (Front Immunol) sequenced BCR heavy-chain repertoires in 25 mild/moderate ME/CFS, 36 severe, 21 healthy controls and 28 MS: no increased clonality and no differential somatic hypermutation versus healthy controls or MS; IGHV3-30 usage was replicated in mild/moderate patients only, but with no evidence of ongoing antigen-driven responses within IGHV3-30 repertoires and no infection or autoimmunity signature; a skewed IgM:IgG ratio in mild/moderate patients was the one new preliminary observation. A Solving ME/CFS re-analysis of UK ME/CFS Biobank repertoires (123 people) found IgM IGHV3-30/3-33 usage essentially identical (9.34% vs 9.37%, p=0.96).

Evidence

Study	n	Design	Result
Ryback & Cowan 2025, Frontiers in Immunology	25 mild/mod + 36 severe vs 21 HC vs 28 MS	AIRR-seq case-control with disease control	No increased clonality or SHM; IGHV3-30 replicated only in mild/moderate; IgM:IgG ratio skewed
Sato 2021, Brain Behav Immun	case-control	BCR repertoire sequencing	Biased IGHV gene usage; ROC separation suggested

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	123 people, UK ME/CFS Biobank	re-analysis of B-cell repertoires	IGHV3-30/3-33 IgM usage 9.34% vs 9.37%, p=0.96 - claim fails

Open questions

Is the IgM:IgG skew in mild/moderate disease real and does it predict anything?

Bears on: Autoimmunity (functional autoantibodies) Viral persistence · Test: research only

B-cell depletion with rituximab failed in a 151-patient placebo-controlled trial, after positive small phase II results. Refuted

A drug that wipes out antibody-making B cells looked promising in small studies, then failed completely in a proper large trial.

Fluge 2011 (PLOS ONE, n=30 randomised) reported major or moderate response in 10/15 on rituximab vs 2/15 on saline (p=0.003), with delayed onset 2-7 months. RituxME (Fluge 2019, Ann Intern Med) randomised 151 CCC patients to induction plus 4 maintenance infusions over 12 months or placebo: response rates 26.0% rituximab vs 35.1% placebo (difference -9.2 points, 95% CI -5.5 to 23.3 for placebo advantage, p=0.22), no difference in fatigue score over 24 months or any secondary endpoint, with more serious adverse events on drug (26.0% vs 18.9%). The high placebo response rate (35%) is the key number for reading every uncontrolled immunotherapy report in this field. A 6-year follow-up showed SF-36 physical function 45.5 vs 43.1, i.e. both groups improved similarly.

Evidence

Study	n	Design	Result
Fluge 2019 (RituxME), Annals of Internal Medicine	151 randomised (77 vs 74)	multicentre double-blind placebo-controlled RCT	Response 26.0% vs 35.1% placebo, p=0.22; all secondary endpoints null
Fluge 2011, PLOS ONE	30 randomised	double-blind placebo-controlled phase II	10/15 vs 2/15 responders (p=0.003); primary 3-month endpoint negative

Open questions

Does rituximab's failure exclude autoantibodies, or only short-lived B-cell-dependent ones (long-lived plasma cells are CD20-negative)?

Bears on: Autoimmunity (functional autoantibodies) · Test: Not indicated

Plasma cells

Plasma-cell depletion with daratumumab produced marked, in some cases sustained, improvement in 6 of 10 patients in an open-label pilot, with low baseline NK count predicting non-response. Early

A myeloma drug that kills the cells making long-lived antibodies helped six of ten patients, some of them a lot and for years. It was a small study with no placebo group.

Fluge 2025 (Front Med, corrected 2025; EudraCT 2022-000281-18): 10 female moderate-to-severe patients, 12-week run-in, then subcutaneous daratumumab 1800 mg at weeks 0/2/4/6 (last four patients received three additional injections from week 14). No serious adverse events. All 10: mean SF-36 physical function 25.9 to 55.0 at 8-9 months (p=0.002), DSQ-SF 72.3 to 43.1 (p=0.002). Six responders: SF-36 PF 32.2 to 78.3, DSQ-SF 71.1 to 24.3, and five of the six sustained improvement with mean SF-36 PF 88 toward end of follow-up and step counts above 10,000/day for periods. Serum IgG fell 54% in improvers vs 40% in non-improvers. Low baseline blood NK-cell count was significantly associated with lack of response (responders 238 vs 97 cells/microlitre, rho 0.77). The design cannot separate drug effect from the 35%+ open-label/placebo response documented in RituxME, and n=10 with a post-hoc predictor is hypothesis-generating; an RCT is planned.

Evidence

Study	n	Design	Result
Fluge 2025, Frontiers in Medicine	10 (open-label)	prospective open-label pilot, anti-CD38	6/10 marked improvement; SF-36 PF 25.9 to 55.0 (p=0.002); 5 sustained 12-24 months; NK count predicted response
Rekeland 2020 (CycloME), Frontiers in Medicine	40 (open-label)	open-label phase II cyclophosphamide	55% response; SF-36 PF 33.0 to 51.5 at 18 months; 68% of responders still in remission at 4 years; HLA-DQB103:03/C07:04 83% vs 43% response; 11 SAEs in 8 patients

Against or not replicated

Study	n	Design	Result
Fluge 2019 (RituxME), Annals of Internal Medicine	151	RCT	35% response on placebo - the benchmark any open-label 'response rate' must beat

Open questions

Will the planned randomised trial reproduce it?
If long-lived plasma cells matter, why did immunoadsorption (which removes the antibodies themselves) fail against sham?
Is the NK-count predictor real or a chance correlation in n=10?

Bears on: Autoimmunity (functional autoantibodies) Circulating (serum) factor · Test: Lymphocyte subsets (NK count) and serum IgG are routine; the treatment is research/trial only

Cytokines

Plasma cytokine abnormalities depend strongly on illness duration: a pro- and anti-inflammatory activation pattern early (under 3 years) that is absent in long-established illness. Several studies

Blood inflammation markers look disturbed in people ill for a short time and normal in people ill for years, which may be why studies disagree.

Hornig 2015 (Sci Adv) used two multicentre cohorts with controls frequency-matched on season, site, age and sex: 52 cases ill under 3 years, 246 ill longer, 348 controls. Early cases showed prominent activation of both pro- and anti-inflammatory cytokines and dissociation of inter-cytokine regulatory networks; this was not present in the longer-duration group. Cytokine alterations correlated more strongly with duration than with severity measures. This duration-dependence is the single most useful explanation for the field's inconsistency, and it has been widely cited but not formally replicated with a comparably large early-illness sample; it also cannot distinguish a biological evolution from survivorship/sampling differences between recent-onset and long-ill cohorts.

Evidence

Study	n	Design	Result
Hornig 2015, Science Advances	52 early vs 246 long-duration vs 348 controls	multicentre case-control, 51-plex immunoassay	Distinct early cytokine activation and network dissociation, absent later; duration a stronger correlate than severity

Against or not replicated

Study	n	Design	Result
Corbitt 2019, BMC Neurology	15 included studies	systematic review of cytokines	Findings inconclusive; cytokines would not serve as reliable biomarkers

Open questions

Would a prospective inception cohort (post-infection) reproduce the early signature and show it fading?
Does early immunomodulation change outcome - never tested

Bears on: Innate immune hyperactivation Persistent sickness behaviour / vagal signalling Immune (T-cell) exhaustion · Test: research only (multiplex cytokine panels are not clinically interpretable)

In the largest single cytokine study, 17 cytokines rose in step with disease severity although only TGF-beta (up) and resistin (down) differed between all patients and controls. Several studies

Comparing all patients with all controls shows almost nothing, but the sicker the patient, the higher a group of inflammatory signals.

Montoya 2017 (PNAS) measured 51 cytokines by Luminex in 192 ME/CFS patients and 392 healthy controls, regressing each on severity with covariates for age, sex, race and non-specific binding. On average only TGF-beta was elevated (p=0.0052) and resistin lower (p=0.0052). Seventeen cytokines showed significant upward linear trends with severity (CCL11, CXCL1, CXCL10, IFN-gamma, IL-4, IL-5, IL-7, IL-12p70, IL-13, IL-17F, leptin, G-CSF, GM-CSF, LIF, NGF, SCF, TGF-alpha), 13 of them pro-inflammatory; CXCL9 correlated inversely with fatigue duration. Caveats: severity is self-reported and correlates with inactivity, BMI and sleep; multiplex platforms have substantial non-specific binding (which this paper explicitly modelled, unlike most); and the TGF-beta result is now suspect on pre-analytic grounds (see immune-tgfbeta-artefact).

Evidence

Study	n	Design	Result
Montoya 2017, PNAS data	192 vs 392	case-control, 51-plex Luminex with severity regression	TGF-beta up, resistin down overall; 17 cytokines rising with severity

Against or not replicated

Study	n	Design	Result
Roerink 2017, PNAS (letter) and Roerink 2017, J Transl Med	50 vs 48	critique + independent case-control with 92-marker PEA	Only IL-12p40 and CSF-1 raised; TGF-beta no different between patients and controls

Open questions

Are severity-cytokine gradients mediated by deconditioning, adiposity and sleep loss rather than disease?
Do any of the 17 track within-person symptom change?

Bears on: Innate immune hyperactivation Persistent sickness behaviour / vagal signalling Mast cell activation · Test: research only

Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF-beta and CRP, with 12 other proteins showing no difference. Several studies

Pooling all the cytokine studies shows a little extra inflammation on average, but the effects are small and inconsistent, so no blood test comes out of it.

Strawbridge 2019 (Neurosci Biobehav Rev) meta-analysed 42 studies of circulating inflammatory proteins in CFS versus unaffected controls: TNF ES 0.274 (p<0.001), IL-2 0.203 (p=0.006), IL-4 0.373 (p=0.004), TGF-beta 0.967 (p<0.001), CRP 0.622 (p=0.019); 12 proteins did not differ. The authors conclude inflammation is unlikely to be a primary feature in all patients and that subgroups may exist. Blundell 2015 reviewed 38 papers and found elevated TGF-beta in 5 of 8 studies and no other consistent cytokine, with no case-control differences after exercise. Corbitt 2019 reached an explicitly inconclusive verdict. Independent support for low-grade inflammation comes from UK Biobank: Beentjes/Ponting 2025 found inflammation, insulin-resistance and liver markers shifted in both sexes among 1,455 cases vs 131,303 controls, not explained by inactivity, with small effect sizes.

Evidence

Study	n	Design	Result
Strawbridge 2019, Neuroscience & Biobehavioral Reviews	42 studies	systematic review and meta-analysis	TNF, IL-2, IL-4, TGF-beta, CRP elevated (ES 0.20-0.97); 12 proteins null
Blundell 2015, Brain Behavior and Immunity	38 papers, 77 cytokines	systematic review	Only TGF-beta consistently elevated (5/8 studies); no post-exercise differences
Beentjes/Ponting 2025, EMBO Molecular Medicine (via Solving ME/CFS re-analysis)	1,455 cases vs 131,303 controls	UK Biobank cross-sectional, 3,237 traits incl. 2,923 proteins	116 traits differ in both sexes incl. inflammation markers; effects small; not explained by inactivity

Against or not replicated

Study	n	Design	Result
Corbitt 2019, BMC Neurology	15 studies	systematic review	Inconclusive; cytokines not reliable biomarkers
Solving ME/CFS re-analysis (deep2/proteome) data	~150 cases/105 controls across 4 proteomic cohorts + UK Biobank	harmonised meta-analysis	Interferon proteins and most immune protein sets flat or inconsistent; no single protein replicates strongly

Open questions

Is there a genuinely inflamed subgroup identifiable prospectively, and does it respond differently to anything?

Bears on: Innate immune hyperactivation Persistent sickness behaviour / vagal signalling · Test: CRP and ESR are routine but usually normal or only slightly raised; not diagnostic

The most-replicated single cytokine finding, elevated TGF-beta, is largely a blood-handling artefact driven by platelet activation during centrifugation. Refuted

The one cytokine that kept showing up high turned out to depend on how fast the blood was spun, because platelets release it.

Roerink 2018 (Neth J Med) measured plasma TGF-beta1 in two female CFS cohorts (50 patients/48 controls; 90/29) with controls recruited by patients as age-matched acquaintances. Within each cohort there was no patient-control difference, but between cohorts values differed greatly, traced to the centrifuge g-force: the lower force (1361 g) produced more platelet activation and higher P-selectin and TGF-beta. This is the clearest documented pre-analytic trap in ME/CFS immunology and it retrospectively undercuts Blundell's 5-of-8 TGF-beta tally and Montoya 2017's overall TGF-beta elevation. Roerink 2017 (J Transl Med), using a 92-marker proximity extension assay in 50 patients vs 48 neighbourhood controls, found TGF-beta no different and only IL-12p40 and CSF-1 raised.

Evidence

Study	n	Design	Result
Roerink 2018, Netherlands Journal of Medicine	50 vs 48 and 90 vs 29	two case-control cohorts with pre-analytic comparison	No within-cohort difference; between-cohort discrepancy explained by centrifuge g-force and platelet activation
Roerink 2017, J Translational Medicine	50 vs 48	case-control, 92-marker PEA + ELISA, plus anakinra RCT samples	IL-12p40 and CSF-1 higher; TGF-beta not different; anakinra had no effect on circulating cytokines other than IL-1Ra

Open questions

How many other reported cytokine and proteomic differences are pre-analytic? (A parallel leak artefact was found in the Hoel/Fluge SomaScan serum data.)

Bears on: Innate immune hyperactivation Fibrin amyloid microclots · Test: Not indicated

Blocking IL-1 with anakinra for four weeks did not improve fatigue in CFS, and did not change circulating cytokines beyond IL-1Ra itself. Refuted

A drug that blocks one of the main inflammation signals did not help, which argues against simple ongoing inflammation driving symptoms.

Roerink and colleagues ran a randomised placebo-controlled trial of daily anakinra 100 mg for four weeks in 50 severely fatigued female CFS patients (reported in Ann Intern Med 2017 and in the accompanying cytokine paper). After correction for multiple testing, anakinra had no effect on circulating cytokines other than IL-1Ra, and the trial did not produce clinical benefit. A companion endocrine analysis (Roerink 2018, Horm Behav) found a blunted cortisol awakening response in patients (4.2 vs 6.1 nmol/L, p=0.036) and a small anakinra effect on hair cortisol, indicating the drug was biologically active. The trial was short (4 weeks) and in a mostly moderate, female, Dutch cohort, so it cannot exclude benefit from longer or different immunomodulation.

Evidence

Study	n	Design	Result
Roerink 2017, Annals of Internal Medicine / J Transl Med (cytokine arm)	50 randomised	RCT, anakinra vs placebo, 4 weeks	No clinical benefit; no cytokine change beyond IL-1Ra

Open questions

Would IL-6 or TNF blockade, or a longer course, do anything? No adequately powered trial exists.

Bears on: Innate immune hyperactivation Persistent sickness behaviour / vagal signalling · Test: Not indicated

Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and disturbed IL-1 signalling, but CSF studies are small and partly contradicted by later proteomics. Early

Spinal fluid from patients looks immunologically different from healthy people's, hinting at inflammation inside the nervous system. The studies are small.

Hornig 2015 (Mol Psychiatry) analysed CSF from 32 ME/CFS cases, 40 MS patients and 19 normal subjects with a 51-plex assay: raised CCL11 (eotaxin, an eosinophil chemokine) in cases and an inverse relationship between IL-1 receptor antagonist and CSF-1, CSF-2 and IL-17F without changes in IL-1 alpha or beta, interpreted as disturbed IL-1 signalling and a shift toward a Th2/allergic pattern. Peterson/Brenu 2015 found CSF cytokine differences in 18 patients. Against: a Solving ME/CFS re-analysis of NIH CSF proteomics found IL-17 proteins null in both serum and CSF, and neuronal injury markers (NEFL, GFAP) flat. Lumbar puncture cohorts are inevitably small and self-selected, and MS is an imperfect comparator.

Evidence

Study	n	Design	Result
Hornig 2015, Molecular Psychiatry	32 cases vs 40 MS vs 19 normal	case-control CSF 51-plex	CCL11 raised; IL-1Ra vs CSF-1/CSF-2/IL-17F network inversion
Peterson/Brenu 2015, Mediators of Inflammation	18 patients	pilot CSF cytokine study	CSF cytokine differences reported

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/proteome; NIH Walitt CSF) data	NIH CSF and serum	re-analysis	IL-17 proteins null in serum and CSF; neuronal injury markers flat

Open questions

Does CSF CCL11 replicate in an independent cohort with matched non-inflammatory controls?

Bears on: Neuroinflammation (microglial activation) Mast cell activation Persistent sickness behaviour / vagal signalling · Test: research only

Newly reported elevations of serum IL-11 and MMP-9 in female ME/CFS patients are unreplicated single-cohort findings. Early

A 2026 study found two new inflammation-related proteins raised in patients. Nobody has checked it yet.

Chinnappan 2026 (Front Immunol) quantified IL-11 and matrix metalloproteinase-9 in serum from 40 female ME/CFS patients and controls (Nova Southeastern/Institute for Neuro-Immune Medicine), reporting both elevated, with an EBV framing. IL-11 is a fibrosis- and inflammation-associated cytokine and MMP-9 a matrix-remodelling enzyme, so the finding is thematically compatible with the extracellular-matrix disturbance in Che/Lipkin 2025 and the lower vascular mural/matrix proteins found across four proteomic cohorts. But single-cohort, single-lab, two-analyte studies in this field have a poor replication record, and MMP-9 in particular is sensitive to sample handling and neutrophil contamination. Listed here as the kind of new lead a living review should carry with an explicit 'unreplicated' label.

Evidence

Study	n	Design	Result
Chinnappan 2026, Frontiers in Immunology	40 female patients vs controls	case-control serum immunoassay	IL-11 and MMP-9 elevated

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/proteome) data	4 blood protein cohorts	meta-analysis	No single protein replicates strongly; effects that repeat are d 0.25-0.5

Open questions

Does it replicate in an independent cohort with handling controls?

Bears on: Innate immune hyperactivation Endothelial dysfunction · Test: research only

Innate immunity

Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen after exercise, alongside complement activation and metabolic abnormalities. Early

When patients' blood cells are shown bits of bacteria, they over-react, and the over-reaction is worse the day after exercise.

Che, Lipkin and colleagues 2025 (npj Metabolic Health and Disease; medRxiv July 2025) studied 56 ME/CFS cases and 52 healthy controls recruited in New York and California with 47 matched pairs, combining plasma metabolomics, proteomics and immune responses to microbial stimulation before and 24 h after exercise. They report exaggerated innate responses to microbial antigens, complement activation, impaired TCA/beta-oxidation/urea-cycle energy production, disrupted extracellular-matrix homeostasis with release of pro-inflammatory endogenous ligands, reduced cell-cell adhesion with gut dysbiosis, copper-dependent redox disturbance and tryptophan-kynurenine dysregulation; many abnormalities worsened after exercise in patients but not controls and correlated with symptom intensity. This is a single-group study with very many readouts and no pre-registered primary endpoint; the stimulation assay is not standardised across labs. It is nonetheless one of the few designs that captures a post-exertional immune change at all.

Evidence

Study	n	Design	Result
Che/Lipkin 2025, npj Metabolic Health and Disease	56 vs 52 (47 matched pairs)	case-control multi-omics with exercise challenge and ex vivo microbial stimulation	Exaggerated innate response to microbial antigens, complement activation, energy-pathway and ECM abnormalities; worse post-exercise and correlated with symptoms
Raijmakers 2019, Journal of Infection (Q fever fatigue syndrome) data	20 QFS, 19 CFS, 19 seropositive controls, 25 HC	case-control ex vivo stimulation	QFS PBMCs produced more IL-6, TNF-alpha and IL-1beta on Coxiella antigen stimulation; higher circulating IL-6 and IFN-gamma

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/proteome; Germain 2-day CPET) data	SomaScan, 2-day CPET	re-analysis	Nothing changes differently after exercise at FDR in the plasma proteome (0 FDR hits)

Open questions

Does the stimulation hyper-response replicate in an independent lab with blinded samples?
Is it a property of monocytes, of plasma factors, or of both?
Can it be used as a PEM readout?

Bears on: Innate immune hyperactivation Complement activation and thromboinflammation Gut dysbiosis Cell danger response (Naviaux) · Test: research only

Single-cell transcriptomics showed classical-monocyte dysregulation at baseline and platelet-activation signatures after exercise, but a re-analysis of the same data found no reproducible composition or expression differences. Contested

One study found patients' monocytes looked abnormal, and platelets reacted oddly to exercise. Reworking the same data did not confirm it.

Vu 2024 (Cell Rep Med) applied scRNA-seq to PBMCs from patients and controls at baseline and after exercise, reporting classical-monocyte dysregulation suggestive of inappropriate differentiation and tissue migration, with the fraction of 'diseased' monocytes correlating with severity, and post-exercise patterns indicating improper platelet activation with minimal change elsewhere. A Solving ME/CFS re-analysis of the same public dataset (GSE214283; 30 vs 28, 503,739 cells) with CellTypist majority voting and person-level pseudobulk found no composition differences at baseline (lowest q=0.41) or in the exercise change, and 0 genes at q<0.05 at baseline across 15 cell types (only NDUFA10 in non-classical monocytes in the 24 h change, q=0.037); platelet-RNA+ monocyte scores were null. There is no independent ME/CFS PBMC scRNA dataset from another group, which is the central problem. Older monocyte claims (Prieto 1989, naloxone-reversible monocyte dysfunction) were never replicated.

Evidence

Study	n	Design	Result
Vu 2024, Cell Reports Medicine data	case-control, paired pre/post exercise	scRNA-seq of PBMCs	Classical-monocyte dysregulation correlating with severity; post-exercise platelet activation signature

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/singlecell_methyl, GSE214283)	30 vs 28, 503,739 PBMCs	independent re-analysis	No composition difference (lowest q=0.41); 0 DE genes at q<0.05 at baseline in 15 cell types; platelet+ monocyte score null

Open questions

Does any group outside Cornell have ME/CFS single-cell blood data to test this?
Are monocyte findings driven by handling time and cell isolation?

Bears on: Innate immune hyperactivation Fibrin amyloid microclots Complement activation and thromboinflammation · Test: research only

Cell-free RNA

Plasma cell-free RNA carries an immune-weighted ME/CFS signature that classifies cases with held-out AUC 0.81, with raised plasmacytoid-dendritic, monocyte and T-cell-derived transcripts. Early

Fragments of RNA floating in blood plasma, mostly from immune cells, differ enough between patients and controls to build a reasonably accurate classifier.

Gardella 2025 (PNAS) sequenced plasma cfRNA in 93 ME/CFS cases and 75 healthy sedentary controls. A LASSO generalised linear model trained on condition-specific signatures reached test-set AUC 0.81 and accuracy 77%. Deconvolution showed differences in platelet-derived cfRNA and elevated plasmacytoid dendritic, monocyte and T-cell-derived cfRNA; network analysis implicated cytokine signalling and T-cell exhaustion, converging with Iu 2024. Caveats: single cohort with internal hold-out rather than external validation; sedentary controls help but do not remove deconditioning confounding; cfRNA is sensitive to pre-analytic handling. A separate LLM-benchmarking paper using the same cohort (Gaudio 2026, Nat Commun) confirms the ME/CFS cohort is the hardest of three to classify, which is a useful calibration on effect size.

Evidence

Study	n	Design	Result
Gardella 2025, PNAS data	93 vs 75	case-control cfRNA sequencing with machine learning	Held-out AUC 0.81, accuracy 77%; raised pDC, monocyte and T-cell cfRNA; cytokine-signalling and exhaustion networks

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (downloads)	5 blood RNA cohorts + cfRNA + SomaScan	re-analysis	No gene set passes FDR in blood gene expression; interferon slightly up in all 5 (meta p=0.12)

Open questions

Does the classifier hold in an independent cohort and against other fatiguing illnesses?
Does the cfRNA signature move with PEM or treatment?

Bears on: Immune (T-cell) exhaustion Innate immune hyperactivation · Test: research only

Autoantibodies

Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30% of patients by ELISA, is not reproduced by unbiased high-throughput antigen profiling. Refuted

A commercial antibody test suggested many patients have antibodies against nerve-signal receptors. When measured with better, unbiased methods, the difference disappears.

Loebel 2015 (Brain Behav Immun) tested a Berlin cohort (n=268) plus pre/post samples from 25 rituximab-trial patients against a panel of GPCRs by ELISA versus 108 healthy controls: beta2, M3 and M4 antibodies significantly elevated, 29.5% of patients above threshold, correlating with IgG1-3 subclass levels and with activated HLA-DR+ T cells, TPO and ANA positivity; in rituximab responders with prolonged B-cell depletion beta2 and M4 antibodies fell, in non-responders they did not. Bynke 2020 partially validated this in two Swedish cohorts (M3 and M4 in both, beta1/beta2 in one; n=24+24 vs 24) with no correlation to severity and nothing detectable in CSF. The decisive negative is Germain, Jaycox, Ring and Hanson 2025 (IJMS 26:2799): 7,542 antibody-antigen interactions across 172 participants using REAP (6,183 extracellular human proteins, 225 viral antigens) and a 1,134-plex Luminex panel found no significant difference in any autoantibody reactivity, including beta-adrenergic receptors (no q below 0.68), with ADRB2, ADRA1A and GRK2 specifically not elevated. In POTS the same conclusion was reached with standard methodology: Hall 2022 (Circulation) found no difference across 11 receptors in 116 POTS vs 81 controls, with 98.3% of patients and 100% of controls above the manufacturer's alpha1 seropositivity threshold, i.e. the commercial cut-offs are not meaningful.

Evidence

Study	n	Design	Result
Loebel 2015, Brain Behavior and Immunity	268 patients vs 108 controls (+25 trial patients)	case-control ELISA panel	beta2, M3, M4 IgG elevated; 29.5% above threshold; fell in rituximab responders only
Bynke 2020, Brain Behav Immun Health	48 patients vs 24 controls (CSF in 24/6)	two-cohort validation, plasma and CSF ELISA	M3 and M4 raised in both cohorts, beta1/beta2 in one; no correlation with severity; nothing in CSF

Against or not replicated

Study	n	Design	Result
Germain, Jaycox, Ring & Hanson 2025, IJMS	172 participants (164 REAP, 103 Luminex)	case-control unbiased antigen profiling (REAP + 1,134-plex)	No autoantibody reactivity differed; no q below 0.68; ADRB2/ADRA1A/GRK2 not elevated; incidental higher EBV gp42 reactivity
Hall 2022, Circulation (POTS)	116 POTS vs 81 HC	case-control, commercial ELISA, standard methodology	No difference for any of 11 receptors; ROC poor; manufacturer thresholds met by essentially everyone

Open questions

Could functional (bioassay) activity exist without differential binding, as the Wallukat-type cardiomyocyte assays claim? Those assays have never been blinded across labs.
Is the ELISA signal an artefact of IgG concentration or non-specific binding (it correlated with total IgG1-3)?

Bears on: Autoimmunity (functional autoantibodies) Endothelial dysfunction Autonomic dysfunction and small-fibre neuropathy · Test: CellTrend GPCR antibody panels are purchasable privately; the evidence does not support using them for diagnosis or treatment decisions

Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against sham control, and antibody removal does not predict who improves. Refuted

Filtering antibodies out of the blood made patients feel better in studies without a comparison group, and did nothing once a fake treatment was used for comparison.

Open-label positives: Scheibenbogen 2018 (10 patients, 5 sessions, beta2-selected) and Stein 2024 (Lancet Reg Health Eur; 20 post-COVID ME/CFS with elevated beta2-AR antibodies; total IgG down 79%, beta2 antibodies down 77%, mean SF-36 PF +17.75, 14/20 responders at six months). The controlled trials are negative: IAMPOCO (Stortz 2026, Lancet Reg Health Eur) randomised 40 post-COVID patients in a patient-blinded sham-controlled crossover - no difference in symptom severity (Post-COVID Functional Scale OR 1.17, 95% CI 0.41-3.36, p=0.771; MFI-20 difference 2.4, p=0.437; Chalder 0.09, p=0.970; Bell -2.6, p=0.246; MoCA -0.01, p=0.993; handgrip 1.3, p=0.234), with more adverse events on immunoadsorption (24 vs 10), and GPCR antibodies depleted by immunoadsorption but not sham - i.e. the mechanism worked and the patients did not improve. IA-PACS-CFS (Charite, NCT05710770, 65 randomised 2:1) also failed its primary endpoint per conference report; the paper is not yet published. Wossidlo 2026 (Infection; 18 patients) found significant antibody reduction, return to baseline within 3-6 months, no improvement in objective performance, only slight self-reported pain/fatigue change and no correlation between antibody change and symptoms. Anft 2025 (Mol Ther; 12 patients) found antibody and cytokine reduction with improved neuropsychological testing and handgrip but no improvement in self-reported symptoms or ME/CFS questionnaires, and rebound within a month.

Evidence

Study	n	Design	Result
Stortz 2026 (IAMPOCO), Lancet Regional Health Europe	40	randomised patient-blinded sham-controlled crossover	No difference on any endpoint; antibodies depleted by IA not sham; more AEs on IA
Wossidlo 2026, Infection	18	prospective exploratory, 5 sessions	Antibodies fell then rebounded; no improvement in objective performance; no antibody-symptom correlation
Anft 2025, Molecular Therapy	12	observational, pre/post	Antibodies and pro-inflammatory cytokines fell, cognition and handgrip improved; questionnaires did not; rebound within a month

Against or not replicated

Study	n	Design	Result
Stein 2024, Lancet Regional Health Europe	20 (open-label)	prospective pre-post cohort	70% responders; SF-36 PF +17.75 sustained to 6 months
Scheibenbogen 2018, PLOS ONE	10 (open-label)	uncontrolled pilot	Symptom improvement after 5 days of IgG-binding-column IA

Open questions

Is any benefit in open-label apheresis due to volume/haemodynamic effects rather than antibody removal?
Would a longer or repeated-cycle regimen beat sham (IAMPOCO was a single crossover)?

Bears on: Autoimmunity (functional autoantibodies) Low blood volume and preload failure · Test: Offered privately in some clinics; controlled evidence is negative

BC007 (rovunaptabin), an aptamer that neutralises GPCR autoantibodies, showed secondary-endpoint fatigue improvement in a 30-patient exploratory trial but failed in the larger company-run phase II. Contested

A drug designed to mop up the suspect antibodies looked promising in a tiny trial and failed in the bigger one. The company ran out of money.

reCOVer (Hohberger 2025, eClinicalMedicine; EudraCT 2022-001781-35) was a prospective exploratory placebo-controlled double-blind randomised phase IIa trial with crossover at Erlangen: 30 post-COVID patients with GPCR functional autoantibodies and fatigue as major symptom, 1350 mg rovunaptabin or placebo at day 0 and day 48. Primary endpoint was safety (9 vs 5 treatment-emergent adverse events, p=0.13). Secondary endpoints improved: FACIT-Fatigue effect size 2.10 (p=0.0378), Bell score 3.64 (p=0.0004), Fatigue Severity Scale -2.66 (p=0.0088), and 4 of 8 quality-of-life items, with confirmed neutralisation of the autoantibodies. However the separately run Berlin Cures phase II (BLOC, NCT05911009; about 114 participants across 19 European centres, results announced November 2024) reported the drug well tolerated but with no superiority over placebo; the company subsequently halted work for lack of funds and the full data are not published. A 30-patient crossover trial with safety as its primary endpoint and multiple positive secondary endpoints is weak evidence against a failed larger trial.

Evidence

Study	n	Design	Result
Hohberger 2025 (reCOVer), eClinicalMedicine	30 randomised	exploratory placebo-controlled double-blind RCT with crossover	Safety primary endpoint met; FACIT-F, Bell, FSS improved as secondary endpoints; antibodies neutralised

Against or not replicated

Study	n	Design	Result
Berlin Cures BLOC phase II (NCT05911009), company announcement November 2024 unverified	~114 across 19 centres	double-blind placebo-controlled phase II	Well tolerated, no superiority over placebo; data unpublished; company halted development

Open questions

Will the BLOC data ever be published, and was the placebo response inflated?
Does any patient subgroup defined by functional (not binding) autoantibody activity respond?

Bears on: Autoimmunity (functional autoantibodies) Endothelial dysfunction · Test: Not available (development halted)

IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve damage to mice in three independent 2026 studies; no equivalent published ME/CFS transfer experiment exists. Several studies

Inject antibodies from long COVID patients into mice and the mice become more pain-sensitive and less active, which is strong evidence those antibodies do something. Nobody has published the same experiment with ME/CFS antibodies.

Chen, Appelman and colleagues 2026 (Cell Rep Med) transferred pooled total IgG from 34 WHO-defined long COVID patients (31 analysed; 15 post-infection asymptomatic controls) into C57BL/6 mice at 260 mg/kg (about one third of the mouse's circulating IgG), 8 mice per group, stratified into three subgroups by plasma GFAP, NFL and interferon-beta: pronounced and persistent mechanical hypersensitivity, reproduced by IgG drawn from the same still-symptomatic patients two years later, with subgroup-linked autoreactivities validated by independent assays. De Sa, Iwasaki and colleagues 2026 (Cell) found CNS- and PNS-reactive autoantibodies correlating with neurological symptoms, IgG reacting with human locus coeruleus, thalamus, adrenal and thyroid and cross-reacting with mouse sciatic nerve and meninges, and passive transfer inducing fatigue-like behaviour, loss of balance, thermal hyperalgesia, small-fibre nerve damage and increased pain-related neuronal activity. Mignolet 2026 (Acta Neuropathologica) transferred IgG from neurological long COVID patients into mice: transient mechanical allodynia and thermal hypersensitivity in week 1 with no anxiety, depression or memory effect; abolished by IgG-depleted serum or papain-digested IgG; IgG accumulated in lumbar dorsal root ganglia colocalising with proprioceptive and nociceptive neurons without local neuroinflammation. These are the template experiments from the fibromyalgia field (Goebel 2021, JCI). The honest caveats: pooled IgG at high dose, pain readouts dominate, the fatigue-like readouts are indirect, and none of this is ME/CFS.

Evidence

Study	n	Design	Result
Chen 2026, Cell Reports Medicine	31 LC patients, 15 HC; 8 mice/group	passive IgG transfer to mice + proteome-wide autoantibody profiling	Persistent mechanical hypersensitivity; reproduced with IgG taken 2 years later
de Sa / Iwasaki 2026, Cell	LC cohort + mouse transfer	autoantibody mapping + passive transfer	Fatigue-like behaviour, balance loss, thermal hyperalgesia, small-fibre damage; IgG binds locus coeruleus, thalamus, sciatic nerve, meninges
Mignolet 2026, Acta Neuropathologica	neurological LC vs asymptomatic matched controls; C57Bl6/J mice	passive IgG transfer, 4 daily i.p. injections	Transient allodynia and thermal hypersensitivity; abolished by IgG depletion or papain digestion; IgG in lumbar DRG

Against or not replicated

Study	n	Design	Result
Mignolet 2026, Acta Neuropathologica	as above	passive transfer	No effect on anxiety, depression or spatial memory, and no local neuroinflammation - so the 'brain fog' part does not transfer
Stortz 2026 (IAMPOCO), Lancet Reg Health Eur	40	sham-controlled crossover	Removing IgG from patients did not improve symptoms - hard to reconcile with a sufficient pathogenic-IgG model

Open questions

Does ME/CFS IgG (non-COVID onset) transfer anything to mice? Presented at conferences, not published.
Why do mouse transfer experiments work while antibody removal in patients does not? Tissue-bound antibody, long-lived plasma cells, or irreversible downstream damage?
Which antigens matter - the studies disagree on targets.

Bears on: Autoimmunity (functional autoantibodies) Autonomic dysfunction and small-fibre neuropathy Circulating (serum) factor · Test: research only

Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human endothelial cells, an effect absent with multiple-sclerosis and healthy IgG and lost on Fab/Fc cleavage. Early

Patient antibodies damage the energy machinery of blood-vessel cells in a dish; antibodies from healthy people and from MS patients do not.

Liu, Hollmann, Prusty and colleagues (medRxiv Aug 2025; Brain Behav Immun Health Feb 2026, 101187) isolated immunoglobulin complexes by protein G from 106 individuals: 39 ME/CFS, 15 post-COVID ME/CFS, 20 MS disease controls and 41 healthy controls, applied them to HUVEC endothelial cells, fibroblasts and PBMCs, and assayed mitochondrial morphology by confocal microscopy and energetics by Seahorse. Endothelial mitochondrial fragmentation occurred with female ME/CFS IgG (p=0.0018 vs controls) with increased spare respiratory capacity and raised IL-1beta (p=0.046), and no effect from MS or healthy IgG; the effect was lost on cleaving IgG into Fab and Fc, with the Fab fragment reproducing the energetic change and Fc alone inducing a hypometabolic phenotype; mass spectrometry found excess extracellular-matrix proteins and SPEG in the immune complexes. Authors' own limits: large inter-plate Seahorse variability forcing split experiments, IgM not captured, unexplained sex difference, autoantigen array arm only 12 vs 3, and nominal p-values across many readouts. Single lab, no independent replication. The same group previously reported ME/CFS serum fragmenting mitochondria in cell lines (Schreiner 2020, ImmunoHorizons, n=10). Context matters: the headline serum-factor claim in muscle cells failed a well-powered replication (Ryback 2026, 67 vs 53, maximal respiration +2.50 pmol/min, 95% CI -1.62 to 6.62, p=0.23).

Evidence

Study	n	Design	Result
Liu / Prusty 2026, Brain Behavior & Immunity - Health	39 ME/CFS, 15 PCS-CFS, 20 MS, 41 HC	in vitro IgG application to endothelial cells with disease control	Mitochondrial fragmentation with female ME/CFS IgG (p=0.0018); IL-1beta up; effect lost on Fab/Fc cleavage
Hackel / Sotzny 2025, medRxiv unverified	post-COVID with and without ME/CFS	IgG fractions applied to monocyte cell lines	Distinct patient-IgG-induced cytokine responses, differing between post-COVID with and without ME/CFS

Against or not replicated

Study	n	Design	Result
Ryback 2026, PLOS ONE (via Solving ME/CFS re-analysis serum-check)	67 ME vs 53 HC	preregistered blinded serum-exposure replication in myoblasts	Maximal respiration +2.50 pmol/min (95% CI -1.62 to 6.62), p=0.23; all secondary endpoints null; CI excludes Fluge's d=1.32

Open questions

Can another lab reproduce the IgG endothelial assay blind, with a detection-floor spike-in control?
Is the signal driven by IgG specificity or by glycosylation/immune-complex composition?
Does the effect size track patient severity or PEM state?

Bears on: Autoimmunity (functional autoantibodies) Circulating (serum) factor Endothelial dysfunction Mitochondrial / energy metabolism impairment · Test: research only

Humoral deficiency

A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficiency, with IgG3 and IgG4 the commonest phenotypes, and registry data show strong enrichment of coded antibody deficiency. Several studies

Some patients are mildly short of particular antibody types, which fits a history of frequent infections. Others have the opposite, with raised antibody levels.

Guenther 2015 (Hum Immunol) retrospectively analysed 300 CFS patients: 25% had decreased levels of at least one antibody class or subclass, IgG3 and IgG4 deficiency commonest, while another 25% had elevated immunoglobulins with excess IgM and IgG2; MBL deficiency in 15% versus 6% in a historical control group; similar frequencies in a second cohort of 168. Lutz 2021 (Biomolecules) found in 262 Austrian patients that 64.9% had a reduction or deficiency in at least one immune parameter and 26.3% showed activation or inflammation, with 17.6% having unclassified antibody deficiency (IgG3/IgG4 most common) and MBL reduced in 32%, deficient in 7%. Finnish national registry data (FinnGen/FinRegistry via Risteys, as used in the Solving ME/CFS re-analysis) show postviral fatigue syndrome enriched for coded IgG subclass antibody deficiency by 20-90x observed/expected, alongside EDS/hypermobility (about 100x) - but healthcare-use bias inflates such registry enrichments, and the published literature on immunoglobulins in ME/CFS is explicitly contradictory (12 studies for, 7 against in that review tally). Both clinic series lack contemporaneous matched controls.

Evidence

Study	n	Design	Result
Guenther 2015, Human Immunology	300 + 168 patients	retrospective clinic series vs historical controls	25% with reduced class/subclass (IgG3, IgG4 commonest); 25% elevated; MBL deficiency 15% vs 6%
Lutz 2021, Biomolecules	262 patients	retrospective clinic series	64.9% with at least one reduced immune parameter; 17.6% unclassified antibody deficiency; MBL reduced in 32%
Solving ME/CFS re-analysis (deep2/epi_literature, FinnGen/FinRegistry) data	~1,622 FinRegistry / 348 FinnGen postviral fatigue cases	registry enrichment	IgG subclass antibody deficiency O/E 44 [87]; crude and healthcare-use inflated; literature contradictory

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/epi_literature literature tally) data	19 studies	literature contradiction count	Immunoglobulins/IgG subclasses contradictory across studies (12 for / 7 against)

Open questions

Does the deficient subgroup differ clinically (infection frequency, severity, onset)?
Is MBL deficiency anything more than a common population variant here?
Does immunoglobulin replacement help the deficient subgroup? Old IVIG trials were small and conflicting; iSTAND in autoimmune POTS (n=30) was negative with both arms improving.

Bears on: Autoimmunity (functional autoantibodies) Viral persistence Connective tissue / hypermobility · Test: Clinical: total IgG/IgA/IgM plus IgG1-4 subclasses; MBL is available in some centres. Cheap and interpretable by an immunologist.

Mast cells

Clinically relevant mast cell activation is reported in up to a quarter of ME/CFS patients, associates with orthostatic intolerance and POTS, and those patients report more benefit from mast-cell-directed treatment. Early

A quarter of patients have symptoms of over-reactive mast cells (flushing, rashes, food and smell sensitivity), and those patients more often have the fast-heart-rate-on-standing problem.

Rohrhofer 2025 (Diagnostics) analysed an Austrian patient-centred survey (687 completed) plus a retrospective clinical cohort (383 patients): mast cell activation prevalence rose over the disease course, up to 25.3% meeting criteria for clinically relevant MCA, and patients with MCAS plus orthostatic intolerance reported symptom alleviation after mast-cell-targeted treatment significantly more often than those without MCAS, while beta-blocker response did not differ (p=0.637); MCA involvement was significantly associated with orthostatic intolerance, especially POTS. Laboratory-level work is thin and small: Nguyen 2017 and Balinas 2018 (both Marshall-Gradisnik group, n=11-18) characterised mast cell progenitor phenotypes and TLR3 responses in PBMCs. There is no replicated serum tryptase or urinary mediator finding in ME/CFS, criteria for MCAS are themselves contested, and self-reported treatment response is not a controlled outcome. One small retrospective case series of continuous high-dose oral cromolyn in ME/CFS with MCAS (5 patients) reported benefit.

Evidence

Study	n	Design	Result
Rohrhofer 2025, Diagnostics	687 survey + 383 retrospective	survey plus retrospective clinical cohort	Up to 25.3% clinically relevant MCA; associated with OI/POTS; better reported response to mast-cell-directed treatment
Nguyen 2017, Asian Pac J Allergy Immunol data	18 patients vs 13 HC	flow cytometry of PBMC mast cell phenotypes	Altered mast cell phenotypes and RAGE/HMGB1 expression

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/proteome) data	4 blood protein cohorts + UK Biobank	harmonised proteomic meta-analysis	No replicated mast-cell or allergic protein signature; immune protein sets flat or inconsistent

Open questions

Do objective mediator measures (serum tryptase, urinary N-methylhistamine, prostaglandin D2) differ in the MCA subgroup?
Is MCA cause, consequence, or a parallel diagnosis of the same autonomic phenotype?
Does a controlled trial of mast-cell stabilisers help ME/CFS with OI?

Bears on: Mast cell activation Autonomic dysfunction and small-fibre neuropathy Innate immune hyperactivation · Test: Clinical: serum tryptase (baseline and during a flare), urinary mediators in specialist centres; antihistamine/cromolyn trials are low-risk

Sickness behaviour

In the interferon-alpha fatigue model, higher baseline IL-6 and IL-10 and a larger early cytokine and fatigue surge predict fatigue that persists after the immune trigger is gone. Early

Giving people interferon as hepatitis treatment causes fatigue; in a third it does not go away afterwards. Those people had a more reactive immune system before treatment started.

Russell 2019 (Psychoneuroendocrinology) followed 55 patients on interferon-alpha for chronic hepatitis C at baseline, during 6-12 months of treatment, and six months after, comparing with 54 CFS patients and 57 healthy volunteers. Eighteen (33%) had persistent fatigue (higher six months post-treatment than baseline). Those who went on to persistent fatigue had a greater fatigue increase in the first four weeks (7.1 +/- 1.5 vs 4.0 +/- 0.8, p=0.046), a trend to higher baseline IL-6, significantly higher baseline IL-10, and larger treatment-induced rises in both (more than doubled by week 4 relative to the others: p=0.011 for IL-6, p=0.001 for IL-10). This is the only human design that measures the immune system before the trigger. Limits: iatrogenic cytokine exposure is not a natural infection, HCV patients differ in many ways, n=18 in the key group, and the model's cytokine differences did not persist to follow-up. It sits alongside the acute sickness-behaviour literature (Capuron 2002: neurovegetative symptoms including fatigue appear within two weeks of interferon-alpha and are less paroxetine-responsive than mood symptoms).

Evidence

Study	n	Design	Result
Russell 2019, Psychoneuroendocrinology	55 HCV patients (18 persistent fatigue) + 54 CFS + 57 HV	prospective pre/during/post immune-trigger cohort	Higher baseline IL-10 (and trend IL-6), larger week-4 cytokine and fatigue surge in those with persistent fatigue
Capuron 2002, Neuropsychopharmacology	40 melanoma patients	randomised paroxetine vs placebo during IFN-alpha	Neurovegetative symptoms (anorexia, fatigue, pain) appear within 2 weeks; mood/cognitive symptoms appear later and are the paroxetine-responsive ones

Open questions

Does a pre-trigger immune profile predict ME/CFS after natural infection (needs a large inception cohort with stored baseline samples)?
Is the persistent-fatigue state after interferon biologically the same as ME/CFS, or only phenotypically similar?

Bears on: Persistent sickness behaviour / vagal signalling Innate immune hyperactivation IDO metabolic trap (Phair, Davis) · Test: research only (prospective design)

Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes behind it, predict who develops post-infective fatigue, while microbiological and psychological factors do not. Replicated

How severe your initial infection is, not which germ you had or your personality, predicts whether you stay ill. Cytokines track how ill you feel at the time.

The Dubbo prospective cohort (Hickie 2006, BMJ) followed 253 patients from acute EBV, Coxiella burnetii or Ross River virus infection: 29 (12%) had a disabling fatigue-pain-cognitive-mood illness at six months and 28 (11%) met CFS criteria, with a stereotyped phenotype at similar incidence after each infection, predicted largely by acute illness severity and not by demographic, psychological or microbiological factors. Vollmer-Conna 2004 (Psychol Med) showed in 21 acute Q fever and 48 acute RRV/EBV patients that spontaneously released IL-1beta and IL-6 from PBMC cultures correlated consistently with fever, malaise, pain, fatigue, mood and concentration, i.e. cytokines index the acute sickness response in humans. Vollmer-Conna 2008 (Clin Infect Dis) found IFN-gamma +874T/A and IL-10 -592C/A polymorphisms acted synergistically on illness severity, cytokine levels and duration across all three infections. This is the most solidly replicated immune-related causal structure in the field, but it explains the onset and early course rather than why illness persists for years, and the same group's later work found no persisting cytokine abnormality.

Evidence

Study	n	Design	Result
Hickie 2006, BMJ	253 followed 12 months	prospective post-infection inception cohort	11% met CFS criteria at 6 months; stereotyped phenotype, equal incidence across three pathogens; predicted by acute severity
Vollmer-Conna 2004, Psychological Medicine	21 + 48 acute infections	prospective symptom-cytokine correlation	PBMC IL-1beta and IL-6 release correlated with fever, malaise, pain, fatigue, mood, concentration
Vollmer-Conna 2008, Clinical Infectious Diseases	Dubbo cohort	genotype-phenotype analysis	IFN-gamma +874T/A and IL-10 -592C/A synergistically affected severity, cytokine levels and illness duration

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (downloads; DecodeME) data	15,579 cases vs 259,909 controls	GWAS	Infection-onset and other-onset ME/CFS are genetically the same (rg 0.98) and no immune cell type is enriched, so inherited immune reactivity is not the main risk axis

Open questions

What converts a severe acute sickness response into years of illness?
Do the cytokine-gene effects replicate in modern post-COVID inception cohorts?

Bears on: Persistent sickness behaviour / vagal signalling Innate immune hyperactivation Neuroinflammation (microglial activation) · Test: research only

Long COVID

Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to non-SARS-CoV-2 viruses especially EBV, and low cortisol, giving a reference phenotype for post-infectious illness. Several studies

Long COVID studies, which are larger and better funded, find a similar picture to ME/CFS: worn-out T cells, signs of old viruses waking up, and low cortisol.

Klein 2023 (Nature) enrolled 275 individuals with or without long COVID for multidimensional immune phenotyping with unbiased machine learning: differences in circulating myeloid and lymphocyte populations, exaggerated humoral responses to SARS-CoV-2, higher antibody responses to non-SARS-CoV-2 pathogens particularly EBV, and lower cortisol in long COVID, with cortisol among the strongest discriminating features. Yin 2024 (Nat Immunol) found systemic inflammation, sex-specific perturbation of cytolytic subsets, more CD4+ T cells poised to migrate to inflamed tissue, exhausted SARS-CoV-2-specific CD8+ T cells and mis-coordination between T and B cell responses. These converge with the ME/CFS exhaustion and herpesvirus-reactivation strands; complement dysregulation in active long COVID (Cervia-Hasler 2024, Science) is, unusually for this space, independently replicated. The caution is that long COVID cohorts are heterogeneous, most participants do not meet ME/CFS criteria, and the low-cortisol finding has not replicated consistently in ME/CFS.

Evidence

Study	n	Design	Result
Klein 2023, Nature data	275	cross-sectional immune phenotyping with ML	Myeloid/lymphocyte shifts, higher anti-EBV and other non-SARS-CoV-2 antibodies, lower cortisol
Yin 2024, Nature Immunology	LC vs non-LC, 8 months post-infection	multi-omic + serology case-control	Exhausted SARS-CoV-2-specific CD8+ T cells; inflammation; uncoordinated adaptive response

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2/proteome; long COVID Olink sets) data	UK Explore 51 vs 51; IRIS 36 vs 9; Bansal 12 vs 44	re-analysis of long COVID proteomics	Only lower vascular mural/matrix proteins overlap with ME/CFS; interferon and immune-activation protein sets do not line up

Open questions

How much of the long COVID immune phenotype is present in non-COVID-onset ME/CFS when measured side by side?
Is the EBV antibody signal reactivation, bystander activation, or cross-reactivity?

Bears on: Immune (T-cell) exhaustion Herpesvirus reactivation (EBV, HHV-6) Complement activation and thromboinflammation HPA axis blunting · Test: EBV and CMV serology and PCR are clinically available; interpretation is limited (IgG shows past infection only)

Genetics vs immunity

Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type, implying most immune findings are downstream consequences rather than inherited causes. Replicated

The largest genetic study points at brain cells, not immune cells. So the immune abnormalities are probably effects of the illness rather than its root cause.

DecodeME (15,579 PEM-required cases vs 259,909 controls, SNP heritability about 9.5% on the liability scale) shows enrichment in brain tissue and neurons with no enrichment in blood immune cell types; the Solving ME/CFS re-analysis reproduced all 8 loci exactly and found MAGMA enrichment in 11 brain regions and pituitary and none of 29 immune cell types (Th17 p=0.52), with S-LDSC enrichment only in CNS tissues and chromatin and no signal in ImmGen or blood-cell ATAC-seq. HLA-DQA1*05:01 is protective (p=1.4e-10) but the earlier ME/CFS HLA candidates (C*07:04, DQB1*03:03 from Lande 2020) are not among DecodeME's hits, and the TRA locus (Ueland 2022) did not replicate. Infection-onset and other-onset cases are genetically indistinguishable (rg 0.98). The qualifier: a GWAS of common variants in 15,000 people can miss rare-variant immune contributions, and the one NK-relevant gene near a lead locus (RAB27A, needed for cytotoxic granule release) is not the authors' preferred candidate but is as plausible as their pick.

Evidence

Study	n	Design	Result
DecodeME 2025 + Solving ME/CFS re-analysis (genetics, downloads) data	15,579 cases vs 259,909 controls	GWAS with MAGMA and S-LDSC enrichment	Brain/neuron and pituitary enrichment; no enrichment in 29 immune cell types; locus-level tissue test null after FDR
Solving ME/CFS re-analysis (deep2/genetics_mr) data	DecodeME + Million Veteran Program	Mendelian randomisation and replication	Genetically higher EBV DNA load suggestive (OR 1.59 in MVP); overall DecodeME signal replicates (rg 0.73) while individual loci mostly do not

Against or not replicated

Study	n	Design	Result
Zhang 2025 (HEAL2), medRxiv	multi-site whole genomes	deep-learning rare-variant analysis	115 candidate risk genes with reduced expression in patients' B and T cells, especially cytotoxic CD4 T cells - rare-variant immune involvement not excluded by common-variant results

Open questions

Do rare immune variants (e.g. perforin-pathway, immunodeficiency panels) contribute in a subgroup?
Could a genuinely causal immune mechanism be environmental (infection timing, dose) and therefore invisible to GWAS?

Bears on: Autoimmunity (functional autoantibodies) Immune (T-cell) exhaustion Genetic susceptibility acting through neurons Herpesvirus reactivation (EBV, HHV-6) · Test: Clinical genetic testing is not indicated for ME/CFS; an immunodeficiency/HLH panel only if red flags (abnormal counts, recurrent severe infection)

Key reviews

Baraniuk, Eaton-Fitch & Marshall-Gradisnik 2024, Frontiers in Immunology: The definitive pooled estimate for the NK cytotoxicity literature (28 papers, g 0.96) - read alongside Querec 2023 for why it has not become a test.
Strawbridge 2019, Neuroscience & Biobehavioral Reviews: The only quantitative meta-analysis of circulating inflammatory proteins (42 studies); gives the honest effect sizes.
Corbitt 2019, BMC Neurology: Systematic review concluding cytokines are not reliable biomarkers - the counterweight to single-cohort cytokine papers.
Mountford 2024, Pain Reports (international workshop position statement): Sets out how passive-transfer evidence should be read in symptom-based disorders, which is the key methodological frame for the 2026 IgG mouse studies.
Fehrer 2026, Autoimmunity Reviews (3rd Charite Fatigue Center conference): Current expert survey of the autoimmunity/post-COVID strand including unpublished trial results.
Wirth & Scheibenbogen 2020, Autoimmunity Reviews: The most-cited mechanistic version of the beta2-adrenergic autoantibody hypothesis; useful to read now that its testable predictions have mostly failed.
Solving ME/CFS re-analysis, deep2/singlecell_methyl and deep2/proteome SUMMARY.md: Systematic re-analysis of public immune-cell, methylation, repertoire and protein data; documents which published immune claims fail, and two specific artefacts (granulocyte contamination, cell-leak) that mislead naive users of these datasets.

Infection and viruses

Post-infectious onset

About 11-12% of people with confirmed EBV, Q fever or Ross River virus infection meet CFS criteria at 6 months, at a similar rate for all three pathogens, and the severity of the acute illness predicts it. Replicated

Three very different germs (a herpesvirus, a bacterium and a mosquito-borne virus) left about 1 in 10 people with the same long-lasting fatigue illness. The sicker someone was at the start, the more likely they were to stay ill.

The Dubbo prospective cohort (n=253, rural Australia) followed people from laboratory-confirmed acute infection: 29/253 (12%) had a post-infective fatigue syndrome at 6 months and 28 (11%) met CFS criteria. The phenotype was 'stereotyped' across pathogens, and acute severity predicted it, whereas demographic, psychological and microbiological factors did not. Because the result does not depend on which pathogen caused the infection, it points to the host response rather than to the particular microbe. The cohort was small, with few cases per pathogen, and was followed to 12 months.

Evidence

Study	n	Design	Result
Hickie 2006, BMJ	253	prospective cohort from acute infection	12% prolonged illness at 6 months, 11% CFS criteria; similar across EBV, Q fever, RRV; predicted by acute severity

Open questions

What immune or metabolic feature of severe acute illness carries forward into the chronic state?
Does early intervention during the acute phase change the 6-month rate?

Bears on: Innate immune hyperactivation Persistent sickness behaviour / vagal signalling Cell danger response (Naviaux) · Test: No predictive test; documentation of the acute infection (serology/PCR at onset) is useful for subtyping.

After infectious mononucleosis in adolescents, 13%, 7% and 4% met paediatric CFS criteria at 6, 12 and 24 months; all 13 cases still ill at 24 months were female. Replicated

After glandular fever, many teenagers are still unwell at six months, but most recover over two years. Those still ill at two years were all girls.

Katz 2009 (n=301 adolescents aged 12-18, Chicago) screened for non-recovery and made diagnoses by blind review. Greater fatigue severity at 12 months predicted persistence, and acute-phase steroids did not raise risk. Jason and colleagues' follow-up of a larger college-student mono cohort (about 238 cases) found that baseline GI symptoms and low IL-13/IL-5 predicted non-recovery, per a secondary summary in the Solving ME/CFS re-analysis. Together with Dubbo, these data show the spontaneous recovery rates any treatment trial in recent-onset illness has to beat.

Evidence

Study	n	Design	Result
Katz 2009, Pediatrics	301	prospective cohort after IM	CFS 13% / 7% / 4% at 6 / 12 / 24 months; all 24-month cases female
Jason 2022 (via Solving ME/CFS re-analysis epi summary) unverified data	238 mono cases	prospective cohort, pre-illness baseline	baseline GI symptoms and low IL-13/IL-5 predicted non-recovery

Open questions

Why are late persisters predominantly female?
Does EBV load or lytic activity during IM predict persistence?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Immune (T-cell) exhaustion · Test: EBV serology (VCA IgM/IgG, EBNA IgG) confirms the IM trigger; no predictive test.

A waterborne Giardia outbreak in Bergen (2004) left chronic fatigue in 26% of exposed people vs 11% of matched controls 10 years later (adjusted OR 3.0), and more than half of those referred with lasting fatigue met CFS criteria. Several studies

A gut parasite outbreak in Norway, which was not a virus, also left a lasting excess of chronic fatigue and IBS a decade later.

Litleskare 2018 followed 1,252 laboratory-confirmed cases and 2,504 age- and sex-matched controls: at 10 years, chronic fatigue was 26% vs 11% (aOR 3.01, 95% CI 2.22-4.08) and IBS 43% vs 14% (aOR 4.74). Response rates at 10 years were around 46% of cases, so selection bias is possible. 'Chronic fatigue' here is a questionnaire measure, not a CFS diagnosis. Naess 2012 clinically evaluated 96 referred patients with persistent post-Giardia fatigue and found 58 (60%) met CDC CFS criteria. A gut-confined protozoan producing the same syndrome is strong evidence against any single-virus explanation.

Evidence

Study	n	Design	Result
Litleskare 2018, Clin Gastroenterol Hepatol	1,252 exposed vs 2,504 controls	prospective matched cohort, 10 years	chronic fatigue 26% vs 11%, aOR 3.01; IBS aOR 4.74
Naess 2012, BMC Gastroenterol	96 referred	clinical case series	58/96 (60%) met CDC CFS criteria; 57% reported a progressive course

Open questions

Is persisting gut barrier/microbiome disturbance the bridge to systemic symptoms?
How many of the 10-year 'chronic fatigue' cases would meet modern ME/CFS criteria with PEM?

Bears on: Gut dysbiosis Innate immune hyperactivation Persistent sickness behaviour / vagal signalling · Test: research only

Q fever fatigue syndrome (QFS) follows roughly 20% of acute Q fever and persists more than 10 years after infection in many patients, almost all of whom report post-exertional malaise; doxycycline did not help. Several studies

After the large Dutch Q fever outbreak (2007-2010), many people developed a lasting ME-like illness. Antibiotics aimed at the bacterium did not improve it.

Spronk 2023 surveyed 368 QFS patients 10 or more years after infection: median EQ-5D index 0.63, and 98.9% reported PEM complaints. Younger and middle-aged patients did worse. The Qure RCT (n=154) found that doxycycline was no better than placebo, and the initial CBT benefit had faded by 1-year follow-up (Raijmakers 2019). The failure of an antimicrobial against the original pathogen is consistent with a post-infectious host process rather than ongoing Coxiella infection. The ~20% figure comes from the Reukers 2019 introduction, a secondary source.

Evidence

Study	n	Design	Result
Spronk 2023, Epidemiol Infect	368	cross-sectional, ≥10 years post-infection	median EQ-5D 0.63; PEM complaints 98.9%
Raijmakers 2019, J Psychosom Res (Qure follow-up) data	154	RCT follow-up (CBT vs doxycycline vs placebo)	neither doxycycline nor CBT differed from placebo at 1 year post-treatment

Open questions

Do QFS cohorts show the same immune/metabolic signatures as EBV- or COVID-onset ME/CFS?

Bears on: Innate immune hyperactivation Persistent sickness behaviour / vagal signalling · Test: Coxiella serology documents the trigger; no test for QFS itself.

In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after SARS-CoV-2 infection vs 0.93 in uninfected participants (HR 4.93), and 4.5% of infected participants met criteria. Replicated

COVID-19 roughly multiplies the chance of developing new ME/CFS by four to five. About 1 in 20 infected adults in a large US study met the diagnostic criteria.

Vernon 2025 (RECOVER-Adult, n=11,785 infected vs 1,439 uninfected) found 531/11,785 (4.5%) vs 9/1,439 (0.6%), and 88.7% of post-COVID ME/CFS cases also met RECOVER long COVID criteria. PEM (24%) was the most common ME/CFS symptom. The criteria were applied from self-report. A meta-analysis of 13 studies (1,973 long COVID patients) found 51% met ME/CFS criteria, but heterogeneity was high and the populations were selected clinic cohorts (Dehlia 2024). Registry signals of the same direction exist for H1N1 influenza (HR 2.04, Norway, Magnus 2015) and for COVID (HR ~1.6, Chen 2024 via the Solving ME/CFS re-analysis). Healthcare-seeking bias inflates some registry estimates (af Geijerstam 2026, via re-analysis), and the INSPIRE cohort found no excess after a COVID-positive vs COVID-negative test (Unger 2024, via re-analysis).

Evidence

Study	n	Design	Result
Vernon 2025, J Gen Intern Med	11,785 infected vs 1,439 uninfected	prospective observational cohort	incidence 2.66 vs 0.93 per 100 PY; HR 4.93 (3.62-6.71)
Dehlia 2024, J Infect	1,973 (13 studies)	meta-analysis	51% (42-60%) of long COVID patients met ME/CFS criteria
Magnus 2015, Vaccine	whole Norwegian population	registry cohort	CFS/ME HR 2.04 after pandemic influenza; 0.97 after adjuvanted vaccine

Against or not replicated

Study	n	Design	Result
Unger 2024 (INSPIRE), via Solving ME/CFS re-analysis unverified	~4,500	prospective cohort, test-positive vs test-negative	no difference in ME/CFS-like illness between COVID+ and COVID- symptomatic participants

Open questions

How much of the registry excess is healthcare-seeking bias?
Does vaccination or variant change the per-infection ME/CFS risk?
Do post-COVID and post-EBV ME/CFS share biology?

Bears on: Viral persistence Innate immune hyperactivation Autoimmunity (functional autoantibodies) · Test: No test; diagnosis is clinical (IOM 2015 criteria).

In Finnish registry data, ME/CFS (G93.3) is enriched 3-6-fold for prior viral infections (HSV, EBV, influenza, COVID, viral CNS infection) but not for bacterial pneumonia or erysipelas. Early

In national health records, people with ME/CFS more often had earlier viral illnesses, but not more bacterial ones, which suggests viruses play a particular role.

Solving ME/CFS re-analysis of FinnGen R12 and FinRegistry (Risteys) found observed/expected ratios of 2-6 for viral infection codes vs ≤1 for bacterial codes, and protozoal intestinal infection was also enriched (18x). Limitations: FinnGen and FinRegistry overlap, so they are not independent. Coding and healthcare-seeking bias can inflate associations, and the timing of infection relative to onset is not always resolved. Infection-onset ME/CFS shares genetic risk with other onsets (DecodeME). The Giardia and Q fever outbreaks show that non-viral triggers also occur, so the viral excess is not absolute.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (epi_literature; FinnGen R12 / FinRegistry) data	1,622 (FinRegistry), 348 (FinnGen) G93.3 cases	registry co-occurrence (O/E, Fisher, BH-FDR)	viral infections O/E 3-6; bacterial pneumonia/erysipelas not enriched

Open questions

Age/sex-matched time-to-event hazard ratios by infection class in UK Biobank or Danish registries

Bears on: Viral persistence Herpesvirus reactivation (EBV, HHV-6) Innate immune hyperactivation · Test: n/a

EBV

In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and overall IgG reactivity do not differ between ME/CFS patients and controls. Several studies

Ordinary EBV antibody tests look the same in people with ME/CFS as in healthy people, so a 'high EBV titre' on a routine panel does not diagnose anything.

UK ME/CFS Biobank (Cliff 2019; 251 ME/CFS including 54 severe, 107 healthy, 46 MS): no seroprevalence differences for six herpesviruses, whereas EBV seroprevalence was higher in MS. Blomberg 2019 (Swedish Canada-criteria cohorts vs blood donors, multiplex assay) found no significant overall differences, only minor relative differences for some antigens. Loebel 2017 peptide array (92 CFS vs 50 HC, 3,054 EBV peptides) found profiles 'quite similar' to controls. The exception was raised IgG to an EBNA-6 repeat, which has homology to human proteins. A meta-analytic tally in the Solving ME/CFS re-analysis lists 10 positive vs 2 negative EBV-antibody studies, so smaller older studies were more positive than the larger recent ones, a pattern typical of publication bias.

Evidence

Study	n	Design	Result
Cliff 2019, Front Immunol	251 vs 107 (+46 MS)	case-control biobank	no difference in seroprevalence for 6 herpesviruses
Blomberg 2019, Front Immunol	several Swedish cohorts vs blood donors	case-control multiplex serology	no significant overall IgG differences to herpesviral antigens
Loebel 2017, PLoS One	92 vs 50	peptide microarray case-control	similar profiles; higher IgG to EBNA-6 repeat peptides

Against or not replicated

Study	n	Design	Result
Shikova 2020, J Med Virol	58 vs 50	case-control serology + nested PCR	IgG/IgM no different, but plasma EBV DNA (active infection) 24.1% vs 4% (p=0.0027)

Open questions

Is there a signal restricted to infection-onset or recent-onset patients?
Would quantitative titres timed to crashes differ when single snapshots don't?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Autoimmunity (functional autoantibodies) · Test: EBV VCA IgM/IgG, EBNA IgG, EA-D IgG (clinical); uninformative as an ME/CFS test.

Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (absent or low EBNA-1/VCA antibody-secreting memory B cells in up to 76% of patients, fewer polyfunctional EBNA-1-specific T cells), while CMV responses were normal. Early

Instead of fighting EBV too hard, some patients' immune memory for EBV seems weak, which could let the virus reactivate more easily.

Loebel 2014 (PLoS One): EBNA-IgG low or absent in 10% of patients, and memory B-cell ELISpot showed diminished EBNA-1/VCA antibody-secreting cells in up to 76%. Ex vivo EBV-induced TNF-α/IFN-γ was lower, and triple-cytokine EBNA-1-specific CD4/CD8 cells were reduced, with no CMV difference. EBER-DNA but not BZLF-1 RNA was more frequent in blood cells, suggesting increased latent, not lytic, load. Sepúlveda 2022 re-analysed the 2017 array and found classification only when restricted to infection-onset patients (two candidate antigens, needing confirmation). The Solving ME/CFS single-cell re-analysis found a non-significant trend toward more EBV-specific TCRβ in severe patients' CD8 (0.88% vs 0.45%, p=0.06), which points the other way, to expansion rather than deficiency. Newer long COVID tetramer work (Ma 2026, Cell Rep Med) finds SARS-CoV-2-, CMV- and EBV-specific CD8 T cells skewed to terminally differentiated, exhausted, cytolytic states in long COVID.

Evidence

Study	n	Design	Result
Loebel 2014, PLoS One	CFS vs HC (tens per group)	case-control cellular immunology	reduced EBV memory B/T responses; normal CMV responses
Sepúlveda 2022, Front Med	92 vs 50	re-analysis of peptide array	no discrimination overall; two antigens discriminated infection-onset patients
Ma 2026, Cell Rep Med	long COVID vs recovered (tetramer cohort)	case-control CD8 tetramer profiling	EBV/CMV/SARS-CoV-2-specific CD8 terminally differentiated, exhausted, cytolytic in LC
Solving ME/CFS re-analysis (TCR, PRJNA940635)	~38 per group per cell type	re-analysis of public TCR repertoire	EBV-matched CD8 clonotypes 0.88% vs 0.45% in severe (p=0.06, q>0.6)

Open questions

Replication of memory B-cell ELISpot deficits outside Charité?
HLA-typed tetramer studies against lytic antigens (BZLF1, BMLF1) in ME/CFS?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Immune (T-cell) exhaustion · Test: research only

Detectable cell-free EBV DNA in plasma has been reported more often in ME/CFS (24% vs 4% in one Bulgarian study), but EBV DNA in blood cells is equally common in patients and controls. Early

Almost everyone carries EBV in their blood cells. Some small studies find free-floating viral DNA, a sign of activity, more often in patients, but this has not been firmly confirmed.

Shikova 2020 (58 vs 50): PBMC EBV DNA 81% vs 84% (no difference), plasma EBV DNA 24.1% vs 4% (p=0.0027). Loebel 2014 found more EBER-DNA but not BZLF-1 RNA in blood cells. In long COVID, EBV viremia at acute infection predicted later PASC (Su 2022, n=309), but EBV viremia at a median of 4 months did not associate with long COVID in Peluso 2023 (n=280), while serologic markers (EA-D IgG) did. Persistently detectable EBV DNA months after IM warrants exclusion of chronic active EBV, which is rare but treatable.

Evidence

Study	n	Design	Result
Shikova 2020, J Med Virol	58 vs 50	case-control nested PCR	plasma EBV DNA 24.1% vs 4%; PBMC DNA no difference
Su 2022, Cell	309	prospective multi-omic cohort (COVID)	EBV viremia at diagnosis one of four PASC-anticipating factors

Against or not replicated

Study	n	Design	Result
Peluso 2023, J Clin Invest	280	post-COVID cohort	LC fatigue associated with EA-D IgG (OR 2.12) and high EBNA IgG, but not with ongoing EBV viremia

Open questions

Does plasma EBV DNA rise during PEM crashes?
Large-cohort quantitative plasma EBV PCR in ME/CFS has not been done.

Bears on: Herpesvirus reactivation (EBV, HHV-6) Viral persistence · Test: EBV DNA PCR (whole blood/plasma) is clinically available; mainly to exclude chronic active EBV.

Genetic instruments for higher EBV DNA load in blood (Kamitaki 2026) are associated with ME/CFS risk in the Million Veteran Program (OR 1.59, p=0.012) but not significantly in DecodeME (OR 1.13, p=0.25). Early

People whose genes make them carry more EBV appear somewhat more likely to get ME/CFS in one large dataset but not clearly in another. This is a weak hint that EBV control matters.

A Solving ME/CFS Mendelian randomisation re-analysis used EBV load instruments from Kamitaki 2026 (n≈655k). It gave a combined OR of about 1.23 (p≈0.02), and HLA instruments at r²<0.01 held (p=0.012, 0.043). HLA patterns that raise antibodies to the lytic EBV proteins ZEBRA and EA-D also align with higher risk, but negative-control serologies (VZV, HHV-6B) align similarly, so the HLA signal is not EBV-specific. Genetic correlation of ME/CFS with infectious mononucleosis was 0.11 ± 0.14, and with EBV load 0.05 ± 0.04, both non-significant. Reverse MR was underpowered. Not yet replicated by an independent group.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (genetics_mr), using Kamitaki 2026 instruments, DecodeME and MVP data	DecodeME ~15k cases; MVP	two-sample Mendelian randomisation	EBV load OR 1.13 (DecodeME, ns), 1.59 (MVP, p=0.012)

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis, negative-control serologies	same	MR specificity check	VZV and HHV-6B HLA instruments also align, undermining EBV specificity

Open questions

Does the signal hold with full Kamitaki summary statistics and the larger DecodeME release?
Is it the MHC driving everything?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Genetic susceptibility acting through neurons Autoimmunity (functional autoantibodies) · Test: research only

Antibodies against herpesvirus dUTPases (EBV, HHV-6, VZV), proposed as a marker of lytic or abortive-lytic reactivation, are higher in ME/CFS in studies from one lab, including a 2026 longitudinal study (873 serum samples, 40 patients vs 16 controls) where levels tracked fatigue severity. Early

One research group measures antibodies to a viral enzyme made when herpesviruses partly wake up. They repeatedly find more of these antibodies in ME/CFS, but other labs have not yet checked.

Halpin 2017: 30.9-52.7% of ME/CFS patients had antibodies to multiple herpesviral and/or human dUTPases vs 17.2% of controls, and anti-EBV dUTPase was higher in ME/CFS. Palomo 2026 (J Med Virol): longitudinal sampling, with EBV/HHV-6/VZV dUTPase IgG significantly higher (p<0.001). 72.5% of patients co-expressed antibodies to multiple herpesvirus and HERV-K dUTPases vs 31% of controls, and EBV and HHV-6 dUTPase IgG correlated with fatigue. The controls were few (n=16). In healthy cohorts positivity ranges from 40-84% and rises with age (Alharshawi 2022), so baseline rates are high. The assays are not commercial, and all data come from the Ariza lab.

Evidence

Study	n	Design	Result
Halpin 2017, J Med Virol	ME/CFS and GWI cohorts vs controls	case-control serology	multiple dUTPase antibodies 31-53% vs 17%; anti-EBV dUTPase higher
Palomo 2026, J Med Virol	40 vs 16 (873 + 378 serial samples)	longitudinal case-control	EBV/HHV-6/VZV dUTPase IgG higher (p<0.001); correlated with fatigue

Against or not replicated

Study	n	Design	Result
Alharshawi 2022, J Med Virol	255 healthy	cross-sectional healthy cohorts	40-84% of asymptomatic people have anti-dUTPase antibodies; higher with age

Open questions

Independent replication with blinded samples?
Does dUTPase protein itself (an immune modulator via TLR2) circulate during crashes?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Innate immune hyperactivation · Test: research only

Serologic evidence of recent EBV reactivation (EA-D IgG) is associated with long COVID fatigue (OR 2.12) in a 280-person cohort, consistent with an earlier small report. Several studies

People with long COVID more often have antibody signs that EBV recently reactivated, which may be one route by which COVID leads to lasting illness.

Peluso 2023 (JCI, n=280): LC fatigue and neurocognitive symptoms were associated with EA-D IgG positivity or high EBNA IgG, independent of covariates, but not with EBV viremia. Prior CMV infection was associated with lower LC risk. Gold 2021 (n=68) reported 66.7% of LC vs 10% of controls positive for EA-D IgG or VCA IgM, a small study with a convenience control group. Cervia-Hasler 2024 found classical complement activation linked to higher herpesvirus antibody titres in active LC. Serology cannot distinguish a cause of LC from a consequence of a general immune disturbance.

Evidence

Study	n	Design	Result
Peluso 2023, J Clin Invest	280	post-COVID cohort, adjusted regression	EA-D IgG associated with fatigue (OR 2.12); no association with viremia; CMV seropositivity protective
Gold 2021, Pathogens	30 LC vs 20 controls (+18)	cross-sectional	EBV reactivation serology 66.7% vs 10%

Open questions

Do people with ME/CFS (non-COVID) show the same EA-D pattern at onset?
Is reactivation sustained or a one-off after acute COVID?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Complement activation and thromboinflammation · Test: EBV EA-D IgG is clinically available; interpretation is non-specific.

Herpes-complement

In active long COVID, classical complement pathway activation tracked raised antibody titres against several herpesviruses, alongside markers of haemolysis, platelet activation and monocyte-platelet aggregates. Early

In long COVID, the blood's complement defence system stays switched on, and this goes along with higher antibodies to herpesviruses. One explanation is that immune complexes against reactivated viruses keep complement firing.

Cervia-Hasler 2024 (Science) ran proteomics on >6,500 proteins in 268 longitudinal samples from Zurich/Basel cohorts followed to 12 months. Active LC showed terminal complement dysregulation and ongoing alternative and classical pathway activation, and the classical arm was associated with increased anti-herpesvirus (e.g., EBV, CMV) titres. Machine learning put complement and thromboinflammation proteins at the top. The cohort was mostly post-hospital-era COVID. No ME/CFS (non-COVID) replication yet. Complement activation fits several mechanisms, and herpes reactivation is only one of them.

Evidence

Study	n	Design	Result
Cervia-Hasler 2024, Science	268 samples (~113 COVID patients + controls)	longitudinal proteomics cohort	complement dysregulation in active LC; classical activation associated with herpesvirus antibody titres

Open questions

Does the herpes-complement association appear in classic ME/CFS?
Would antivirals reduce complement activation in LC?

Bears on: Complement activation and thromboinflammation Herpesvirus reactivation (EBV, HHV-6) Fibrin amyloid microclots · Test: Complement panels (C3, C4, CH50, sC5b-9) are clinically available but not validated for LC/ME/CFS.

HHV-6

HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord tissue from 3 ME/CFS post-mortem donors but not in 24 controls (Kasimir 2022). Early

A small autopsy study found signs of active HHV-6 in the brain and spinal cord of three people with ME/CFS and none in controls. This needs repeating in many more brains.

Kasimir 2022 (Front Mol Biosci) used FISH against an HHV-6 miRNA in frontal lobe, basal ganglia and spinal cord from 3 ME/CFS donors, 3 disease controls and 21 Cambridge Brain Bank controls. The authors also report evidence of EBV activity in tissue. Limitations: n=3 cases, no blinding is described in the abstract, miRNA FISH specificity in post-mortem tissue, and no quantitative PCR or sequencing confirmation. The authors argue that blood negativity in many studies reflects localised tissue reactivation. In vitro, HHV-6 reactivation from integrated genomes fragments mitochondria and induces an antiviral state, and patient serum transferred a similar effect (Schreiner 2020, Immunohorizons).

Evidence

Study	n	Design	Result
Kasimir 2022, Front Mol Biosci	3 ME/CFS vs 24 controls	post-mortem tissue FISH	HHV-6 miR-aU14 in brain/spinal cord only in ME/CFS
Schreiner 2020, Immunohorizons	cell lines + 10 patient sera	in vitro proteomics / serum transfer	HHV-6 reactivation fragments mitochondria, inhibits PDH and FAO; induces antiviral state

Against or not replicated

Study	n	Design	Result
Cliff 2019, Front Immunol	251 vs 107	serology	no HHV-6 seroprevalence difference
Shikova 2020, J Med Virol	58 vs 50	PCR	HHV-6 DNA in PBMC 82.8% vs 82%; plasma 1.7% vs 0%

Open questions

Blinded multi-site replication in brain banks (e.g., with ME/CFS tissue now being collected)?
Is reactivated HHV-6 cause or consequence of illness?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Neuroinflammation (microglial activation) Mitochondrial / energy metabolism impairment Cell danger response (Naviaux) · Test: research only (blood HHV-6 PCR is available but generally uninformative)

Inherited chromosomally integrated HHV-6 (iciHHV-6, ~1% of people) has been reported at roughly twice the population rate in a neurological/fatigue clinic cohort, but this has not been confirmed in a large ME/CFS sample. Early

About 1 in 100 people are born with HHV-6 built into one of their chromosomes. One small study suggested this is more common in ME/CFS, and the large UK DecodeME study is checking.

A 2013 University of South Florida study reported iciHHV-6 in >2% of adults with CFS and neurological symptoms vs ~0.8% population estimates. It was a small, selected clinic cohort, and the information here comes from secondary reports only. UK prevalence is about 1.4%, rising to ~3% in parts of Scotland and Ireland, so ancestry confounding matters. DecodeME has an approved sub-study to measure iciHHV-6 in ~4,500 participants, comparing infection-onset vs other onsets. No published result had been found as of this review. iciHHV-6 can produce false 'high HHV-6 DNA' results on commercial PCR panels, a common source of misinterpretation.

Evidence

Study	n	Design	Result
USF 2013 report (secondary source: press/HHV-6 Foundation) unverified	small clinic cohort	cross-sectional	>2% iciHHV-6 vs ~0.8% population
DecodeME iciHHV-6 sub-study (protocol) unverified	~4,500 planned	genetic cohort sub-study	pending

Open questions

Is iciHHV-6 enriched in DecodeME, and especially in infection-onset cases?
Does integrated-genome reactivation occur in patients' tissues?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Genetic susceptibility acting through neurons · Test: Hair-follicle or nail HHV-6 PCR / ddPCR ratio of viral to cellular genomes can confirm iciHHV-6 (specialist labs).

Enterovirus

Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20% of controls, with EV RNA in 37% vs 5%, but this has not been independently replicated with blinded methods. Contested

One clinician-researcher found enterovirus protein in the stomach lining of most patients he biopsied. Other labs have not confirmed it, and the antibody stain used may cross-react.

Chia 2008 (J Clin Pathol): 135/165 consecutive CFS patients' antrum biopsies stained positive for VP1 in parietal cells vs 7/34 controls (p≤0.001), and EV RNA was found in 9/24 vs 1/21 paraffin samples. Paired samples taken years apart stayed positive. The patients were the author's own referral population. The controls were biopsies taken for other gastric conditions, and the 5D8/1 VP1 antibody has known cross-reactivity concerns. Earlier muscle work found EV RNA in 41/158 postviral fatigue vs 2/152 controls (Bowles 1993), but later muscle studies were inconsistent. O'Neal & Hanson 2021 concluded the methods used were inadequate either to confirm or to rule out persistent EV infection, and a Cornell EV-targeted RNA-seq thesis (2022) detected no EVs in patients or controls.

Evidence

Study	n	Design	Result
Chia 2008, J Clin Pathol	165 vs 34	case series with tissue controls	VP1 82% vs 20%; EV RNA 37% vs 5%
Bowles 1993, J Med	158 PFS vs 152 controls	muscle biopsy hybridisation	EV RNA 41/158 vs 2/152

Against or not replicated

Study	n	Design	Result
O'Neal & Hanson 2021, Front Med (critical review)	review	critical review	prior methods inadequate; evidence neither confirms nor excludes persistent EV
Cornell thesis 2022 (Hanson lab) unverified	small	EV-targeted RNA-seq	no EV detected in ME/CFS or controls

Open questions

Blinded, multi-lab testing of gastric/intestinal tissue with sequencing-based detection of non-cytolytic (5' terminally deleted) EV RNA?
Do historical ME outbreaks (e.g., Royal Free 1955) fit EV epidemiology?

Bears on: Viral persistence Gut dysbiosis · Test: research only (stomach VP1 staining available from one private practice; not validated)

SARS-CoV-2 persistence

SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVID than in pre-pandemic samples (excess +11% at 3-6 months, +5.4% at 10-14 months), with more antigenemia after more severe acute illness. Several studies

In about 1 in 10 people, bits of the virus can still be found in the blood months after COVID, which suggests a hidden reservoir. Whether this causes long COVID is less certain.

Peluso 2024 (Lancet Infect Dis; 171 post-COVID vs 250 pre-pandemic controls, Simoa) found 2% assay positivity in pre-pandemic controls. Antigenemia was associated with hospitalisation and worse acute illness. The association with LC symptoms was modest and not consistent across all symptom groups (detail from secondary reports). Swank 2022/23 (n=63) reported spike in a majority of PASC patients up to 12 months, a small sample later tempered by larger work. In PAX LC, circulating S1/spike levels did not change after 15 days of nirmatrelvir (Bhattacharjee 2026 preprint).

Evidence

Study	n	Design	Result
Peluso 2024, Lancet Infect Dis	171 vs 250 pre-pandemic	cohort with true-negative comparator	excess antigen prevalence +11%, +8.7%, +5.4% at 3-6, 6-10, 10-14 months
Swank 2022, medRxiv (later published in Clin Infect Dis)	63	case-control plasma antigen	spike detected in most PASC patients up to 12 months

Against or not replicated

Study	n	Design	Result
Bhattacharjee 2026, medRxiv (PAX LC exploratory)	82	RCT exploratory biomarker analysis	circulating S1/spike unchanged after nirmatrelvir

Open questions

Is antigenemia from replicating virus, translated persistent RNA, or slow-clearing protein?
Does antigen positivity identify an antiviral-responsive LC subgroup?

Bears on: Viral persistence Complement activation and thromboinflammation Fibrin amyloid microclots · Test: research only (Simoa spike/nucleocapsid assays)

SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a minority of people, and in a Beijing tissue study, viral RNA in tissue was associated with long COVID symptoms (OR 5.17). Several studies

Viral genetic material can hide in the gut and other organs long after the nose test is negative. Several groups have now seen this, and people with long COVID seem to have it more often.

Zollner 2022 (n=46 IBD patients, ~7 months): gut mucosal RNA in 32/46 and nucleocapsid in 24/46, associated with post-acute symptoms. Natarajan 2022 (n=113): fecal RNA in 12.7% at 4 months and 3.8% at 7 months, linked to GI symptoms. Zuo 2024 (Lancet ID; 317 samples from 225 patients): RNA in 30%/27%/11% of solid tissues at 1/2/4 months, with tissue RNA associated with LC symptoms (OR 5.17). Peluso 2024 (Sci Transl Med; n=24 PET) found T-cell activation (F-AraG) in gut wall and spinal cord linked to LC, and RNA in rectosigmoid tissue up to 676 days in some participants. These are mostly convenience tissue samples, and some people without LC also carry RNA.

Evidence

Study	n	Design	Result
Zollner 2022, Gastroenterology	46	endoscopy cohort (IBD)	gut RNA 32/46, nucleocapsid 24/46 at ~7 months
Natarajan 2022, Med	113	longitudinal stool cohort	fecal RNA 12.7% at 4 months, 3.8% at 7 months; linked to GI symptoms
Zuo 2024, Lancet Infect Dis	225 patients, 317 samples	cross-sectional tissue cohort	tissue RNA 30%/27%/11%; associated with LC (OR 5.17)
Peluso 2024, Sci Transl Med	24 (PET) + 5 biopsies	imaging cohort + tissue ISH	T-cell activation in gut/spinal cord with LC; gut RNA up to ~2 years

Open questions

Do tissue reservoirs exist in non-COVID ME/CFS for other viruses?
Why do antivirals not help if reservoirs exist (wrong drug, too short, too late, or reservoir not causal)?

Bears on: Viral persistence Gut dysbiosis Neuroinflammation (microglial activation) · Test: research only

Antiviral trials

Three randomised placebo-controlled trials of 15-25 days of nirmatrelvir-ritonavir in established long COVID (STOP-PASC n=155, PAX LC n=100, RECOVER-VITAL n=959; about 1,214 people in total) found no benefit on primary or secondary outcomes. Replicated

Paxlovid, the COVID antiviral, has been tested several times in people who already have long COVID, and it did not help.

STOP-PASC (Geng 2024, JAMA IM): 15 days, 2:1 randomisation, no difference in pooled severity of 6 symptoms at 10 weeks; mostly vaccinated participants, mean 17.5 months after infection. PAX LC (Sawano 2025, Lancet ID): adjusted PROMIS-29 physical difference -0.55 (p=0.54). RECOVER-VITAL (Baden 2026, Lancet ID; 69 sites): 15- or 25-day regimens, no benefit in cognitive, autonomic or exercise phenotypes. All were safe apart from dysgeusia. Caveats: participants were not selected for antigenemia or reservoir evidence, illness was long-standing, and the drug's tissue penetration is uncertain. Ensitrelvir was also null per a secondary summary.

Evidence

Study	n	Design	Result
Geng 2024, JAMA Intern Med (STOP-PASC)	155	RCT	no difference in core symptom severity at 10 weeks
Sawano 2025, Lancet Infect Dis (PAX LC)	100	decentralised RCT	PROMIS-29 PHSS difference -0.55 (95% CI -2.32 to 1.21)
Baden 2026, Lancet Infect Dis (RECOVER-VITAL)	959	RCT, 3 arms (15d, 25d, placebo)	no benefit in any phenotype

Open questions

Would trials enriched for antigen-positive participants, or using longer/combination antivirals or monoclonal antibodies, differ?

Bears on: Viral persistence · Test: n/a

Treating acute COVID-19 with nirmatrelvir may reduce later long COVID: a small terminated Norwegian RCT found 26% vs 43% at 3 months (RR 0.60), echoing observational VA data, but larger prospective evidence is lacking. Early

Taking the antiviral early, while COVID is still active, might lower the chance of developing long COVID, but the only trial so far was small and stopped early.

PANORAMIC Norway (Oppegaard 2026, Lancet ID): enrolled 144 of a planned 2,000 before stopping for poor recruitment, and found long COVID in 17/66 vs 33/77 (RR 0.60, 95% CI 0.37-0.98, p=0.039). An unblinding effect via dysgeusia (86% vs 18%) is possible because the outcomes were self-reported. VA cohort (Xie 2023, JAMA IM; 35,717 treated vs 246,076) reported lower risk, while Fung 2023 (older Medicare patients) found no reduction. Metformin during acute COVID (COVID-OUT) also reduced later LC, per the Solving ME/CFS treatments summary. Together this suggests the early acute phase may be a window for intervention, which fits the Dubbo finding that acute severity predicts later illness.

Evidence

Study	n	Design	Result
Oppegaard 2026, Lancet Infect Dis (PANORAMIC Norway)	144	RCT (terminated early)	LC 26% vs 43%, RR 0.60
Xie 2023, JAMA Intern Med	35,717 vs 246,076	observational VA cohort	nirmatrelvir associated with reduced PCC risk

Against or not replicated

Study	n	Design	Result
Fung 2023, JAMA Intern Med unverified	Medicare older adults	observational cohort	no reduction in PCC with nirmatrelvir or molnupiravir

Open questions

Would early antivirals for EBV (e.g., during IM) prevent post-IM ME/CFS? No trial exists.

Bears on: Viral persistence Innate immune hyperactivation · Test: n/a

In a 30-patient RCT of 6 months of valganciclovir in CFS patients selected for high HHV-6 and EBV IgG, the primary MFI-20 outcome was not significant, though some secondary fatigue and cognitive trajectories favoured the drug. Early

A strong antiviral helped some tiredness and thinking measures in a small trial, but missed its main goal. It has never been repeated, and it can lower white blood cell counts.

Montoya 2013 (J Med Virol): 2:1 randomisation. The MFI-20 difference at 9 months was non-significant. Trajectories for MFI-20 mental fatigue (p=0.039), FSS (p=0.006) and cognition (p=0.025) were significant, and responder status was 7.4x more likely (p=0.029). Monocytes fell and neutrophils rose. These came from multiple secondary endpoints with no multiplicity correction. Preceding uncontrolled data: Kogelnik 2006 (n=12, 9 responders) and Watt 2012 chart review (n=61, 52% responders, baseline titres did not predict response). Lerner's valacyclovir 'EBV subset' series had small controlled phases and are largely uncontrolled. Antibody titre selection is a weak marker of active infection.

Evidence

Study	n	Design	Result
Montoya 2013, J Med Virol	30 (20 vs 10)	double-blind RCT	primary MFI-20 ns; secondary FSS, mental fatigue, cognition significant
Watt 2012, J Med Virol	61	retrospective uncontrolled chart review	52% responders; titres did not predict response
Lerner 2007, In Vivo	small	small RCT phase + 36-month open follow-up	Energy Index improved more on valacyclovir

Open questions

No adequately powered replication in 13 years
Would selection on lytic markers (plasma DNA, dUTPase, EA-D) rather than IgG identify responders?

Bears on: Herpesvirus reactivation (EBV, HHV-6) · Test: n/a (off-label; requires blood count monitoring)

Endogenous retroviruses

Human endogenous retrovirus (HERV) expression in blood cells differs between ME/CFS, fibromyalgia and controls in small studies (HERV-K raised in moderate ME/CFS; HERV microarray profiles clustered n=43 participants perfectly by diagnosis). Early

Old viral DNA built into our genome can switch back on. Small studies hint that its activity pattern differs in ME/CFS, but nothing is confirmed.

Rodrigues 2019: qPCR found HERV-K overexpressed only in moderately affected patients, with no HERV-W difference. Giménez-Orenga 2025 (eLife; HERV-V3 microarrays, >350k HERV elements, PBMC, n=43 women): hierarchical clustering separated ME/CFS, FM, co-diagnosed and controls 'perfectly'. That sample is very small with very many features, so overfitting is likely without an independent validation set. HERV-K dUTPase antibodies co-occur with herpesvirus dUTPase antibodies in the Ariza lab data (Palomo 2026). Herpesviruses such as EBV can transactivate HERVs, which links this to the herpes hypothesis.

Evidence

Study	n	Design	Result
Rodrigues 2019, Auto Immun Highlights	small (moderate + severe vs controls)	case-control qPCR	HERV-K up in moderate ME/CFS; HERV-W no difference
Giménez-Orenga 2025, eLife	43	case-control microarray	HERV profiles clustered ME/CFS, FM, both and HC separately

Open questions

Replication in an independent, larger, blinded cohort
Are HERV proteins (Env) present in tissues or serum?

Bears on: herv-activation Herpesvirus reactivation (EBV, HHV-6) Autoimmunity (functional autoantibodies) · Test: research only

Temelimab, an antibody against HERV-W envelope protein, did not reduce fatigue versus placebo in a 203-person RCT of HERV-W-positive post-COVID patients. Refuted

A drug designed to block an endogenous retrovirus protein was tested in long COVID patients who had that protein, and it worked no better than placebo.

Nehme 2026 (Brain Behav Immun; GeNeuro-funded, NCT05497089): 24 weeks of monthly IV 54 mg/kg in participants selected for HERV-W ENV positivity. The PROMIS fatigue change was -3.2 vs -3.8 points, with no secondary benefit. Note that temelimab targets HERV-W, not HERV-K, so it does not test the HERV-K findings in ME/CFS. No temelimab trial in non-COVID ME/CFS exists.

Evidence

Study	n	Design	Result
Nehme 2026, Brain Behav Immun	203	double-blind RCT	no difference in fatigue at 24 weeks

Open questions

Is HERV-K (rather than HERV-W) a better target in ME/CFS?

Bears on: herv-activation · Test: n/a

Retroviruses (XMRV)

The 2009 report of the mouse retrovirus XMRV in 67% of CFS patients was a laboratory contamination artefact: blinded multi-lab and original-investigator studies found no XMRV/pMLV, and the Science paper was retracted. Refuted

In 2009 a mouse-derived virus was reported in most CFS patients. Careful blinded testing, including by the original scientists, showed it was contamination, and the paper was withdrawn.

Lombardi 2009 (Science) was partially retracted in 2011 and fully retracted later that year. Simmons 2011 (Blood XMRV Scientific Research Working Group): coded replicate samples sent to 9 labs, and reactivity was similar in CFS and controls. Alter 2012 (mBio; Lipkin-led, 147 CFS vs 146 controls, including the original investigators' labs) found no XMRV or pMLV. Lessons for this field: sensitive PCR contamination (mouse DNA in reagents and cell lines), unblinded testing, and premature clinical use (antiretrovirals were taken by some patients). The episode also set back trust and funding for virus research in ME/CFS.

Evidence

Study	n	Design	Result
Simmons 2011, Science	15 positive vs 15 negative (coded replicates)	blinded multi-laboratory	assays did not reproducibly detect XMRV/MLV; CFS = controls
Alter 2012, mBio	147 vs 146	blinded multicenter	no XMRV or pMLV in any group
Silverman 2011, Science (partial retraction)	-	retraction notice	contamination acknowledged in part of the data

Bears on: Viral persistence · Test: n/a (no valid XMRV test; commercial tests sold in 2010-11 were not meaningful)

Key reviews

O'Neal & Hanson 2021, Front Med: Critical review of enterovirus methods in ME/CFS: why past studies neither proved nor excluded persistent infection.
Proal et al. 2023, Nat Immunol: Review of evidence for a SARS-CoV-2 tissue reservoir in long COVID and research priorities for antiviral trials.
Hickie et al. 2006, BMJ: The foundational prospective cohort showing a stereotyped post-infective syndrome across pathogens.
Alter et al. 2012, mBio: The definitive blinded study closing the XMRV episode, and a model for testing a pathogen claim.
Blomberg et al. 2019, Front Immunol: Careful multiplex herpesvirus serology in Canada-criteria ME/CFS with a discussion of why serology is limited.

Metabolism and blood omics

Metabolomics: overall

Blood metabolome differences between ME/CFS and controls are real at group level but small (mostly 0.2-0.5 SD) and partly cohort-specific; early near-perfect classification did not carry over to new cohorts. Replicated

Patients' blood chemistry differs from healthy people's on average, but only slightly, and the specific list of chemicals changes from study to study.

Single-cohort studies reported AUCs of 0.82-0.96 (Naviaux 2016, Nagy-Szakal 2018, Che 2022), but these were estimated within the training data. A harmonised re-analysis of 10 public cohorts found 74 compounds at q<0.05 with pooled effects mostly 0.2-0.5 SD; two Cornell cohorts agreed only weakly (rho 0.16) and cross-cohort transfer AUCs were 0.56-0.83. Medication exposure (paracetamol, gabapentin, psychoactives, acid suppressants detected in blood) and caffeine are major confounders that most original papers did not adjust for.

Evidence

Study	n	Design	Result
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	20 pathways abnormal; 80% of discriminating metabolites lower; AUC 0.94 (men, 8 metabolites) and 0.96 (women, 13) within the same data
Nagy-Szakal 2018, Sci Rep data	50 vs 50	case-control, untargeted plasma metabolomics + stool metagenomics	altered choline, carnitine and complex lipids; ceramide higher in the ME/CFS+IBS subgroup; cross-validated AUC 0.82 (metabolome), 0.84 combined with stool
Che 2022, Int J Mol Sci data	106 vs 91	case-control, targeted + untargeted plasma metabolomics	lower plasmalogens/ether phospholipids, phosphatidylcholines and sphingomyelins (p<0.001); higher dicarboxylic acids; ML AUC up to 0.873
Solving ME/CFS re-analysis (deep2 metabolome-wide meta-analysis) data	10 cohorts, ~600 cases / ~430 controls	harmonised meta-analysis of public per-person data (Metabolon, Fiehn, LC-MS/MS, NMR)	74 compounds q<0.05; effects mostly 0.2-0.5 SD; panel trained on BioMapAI transfers with AUC 0.60-0.83

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets) data	two Cornell cohorts, 575 shared compounds	cross-cohort concordance	patient-control differences agree only faintly (rho 0.16); model trained on one scores AUC 0.56 on the other; a public GitHub copy of Germain 2020 has case/control labels swapped
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	10-fold CV AUC 0.915 (all omics); external cohorts AUC 0.59-0.77 (Guo, Raijmakers stool; Germain metabolome 0.68, Che 0.61) with 19-79% feature overlap

Open questions

Which differences persist in drug-free, fasting, rapidly processed samples?
Does any metabolite change track severity or PEM within a person?

Bears on: Mitochondrial / energy metabolism impairment Cell danger response (Naviaux) Gut dysbiosis · Test: research only

Metabolomics: cell danger response

Naviaux 2016's picture of a homogeneous 'hypometabolic' dauer-like state (80% of discriminating metabolites decreased) has not been confirmed as a uniform pattern in later independent cohorts. Contested

One influential 2016 study said patients' cells were in a protective 'low-power' mode, with most chemicals reduced. Later studies found more mixed and smaller changes.

Naviaux's 45/39 cohort with targeted LC-MS/MS found lower sphingolipids, phospholipids, purines and other pathways and proposed the cell danger response (CDR). An Open Medicine Foundation validation cohort was reported informally (2018) to reproduce 23 of 30 pathway abnormalities, but no peer-reviewed replication paper was found. Hoel 2021 found heterogeneous metabotypes rather than one state; Che 2022 and Xiong 2023 partly echo the lipid reductions. In the cross-cohort re-analysis, the Naviaux data still carry some replicated signals (e.g. low aromatic lactates on a non-Metabolon platform), but there is no global hypometabolic shift across cohorts.

Evidence

Study	n	Design	Result
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	20 pathways abnormal; 80% of discriminating metabolites lower; AUC 0.94 (men, 8 metabolites) and 0.96 (women, 13) within the same data
OMF metabolomics validation update (MEAction 2018) unverified	not stated	unpublished replication cohort summary	23/30 pathway abnormalities reported as replicated

Against or not replicated

Study	n	Design	Result
Hoel 2021, JCI Insight hoel-2021 hoel-2021-perperson	83 vs 35	case-control, serum metabolomics, lipidomics, hormones	common changes consistent with energy strain and altered fatty-acid/amino-acid fuel use; 3 patient subsets (metabotypes), 2 with distinct energy-metabolism patterns
Solving ME/CFS re-analysis (deep2 metabolome-wide meta-analysis) data	10 cohorts, ~600 cases / ~430 controls	harmonised meta-analysis of public per-person data (Metabolon, Fiehn, LC-MS/MS, NMR)	74 compounds q<0.05; effects mostly 0.2-0.5 SD; panel trained on BioMapAI transfers with AUC 0.60-0.83

Open questions

Has the OMF validation cohort been published with per-person data?
Is any CDR signature (e.g. purinergic signalling) measurable in tissue rather than plasma?

Bears on: Cell danger response (Naviaux) Mitochondrial / energy metabolism impairment · Test: research only

Metabolomics: lipids and sphingolipids

Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across several cohorts and platforms, with modest effect sizes. Several studies

Some fats that form cell membranes, especially a type called plasmalogens, are a little lower in patients' blood in several studies.

Che 2022 (106/91) reported lower plasmalogens/ether phospholipids, PCs and sphingomyelins (p<0.001) and proposed peroxisomal dysfunction. Naviaux 2016 found lower sphingolipids; Xiong 2023 lower sphingomyelins; Nagy-Szakal 2018 higher ceramides only in the IBS subgroup. In the cross-cohort re-analysis PC(14:0/16:0) was -0.42 SD (4 cohorts, 3 non-Metabolon) and plasmalogen P-16:0/18:2 -0.38 SD; dihydroceramide d18:0/16:0 was +0.28. Diet (dairy, fish) and BMI are not well controlled.

Evidence

Study	n	Design	Result
Che 2022, Int J Mol Sci data	106 vs 91	case-control, targeted + untargeted plasma metabolomics	lower plasmalogens/ether phospholipids, phosphatidylcholines and sphingomyelins (p<0.001); higher dicarboxylic acids; ML AUC up to 0.873
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	20 pathways abnormal; 80% of discriminating metabolites lower; AUC 0.94 (men, 8 metabolites) and 0.96 (women, 13) within the same data
Xiong 2023, Cell Host Microbe data	short-term 75, long-term 79 vs 79 controls	case-control, stool metagenomics + plasma metabolomics	plasma butyrate, bile acids, benzoate lower; sphingomyelins depleted; dysbiosis mostly in short-term patients
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	up to 10 cohorts	meta-analysis	PC(14:0/16:0) -0.42 SD; plasmalogen P-16:0/18:2 -0.38 SD; dihydroceramide +0.28 SD

Against or not replicated

Study	n	Design	Result
Nagy-Szakal 2018, Sci Rep data	50 vs 50	case-control, untargeted plasma metabolomics + stool metagenomics	altered choline, carnitine and complex lipids; ceramide higher in the ME/CFS+IBS subgroup; cross-validated AUC 0.82 (metabolome), 0.84 combined with stool

Open questions

Are plasmalogens low in tissue (muscle, red cells) or only plasma?
Do they relate to peroxisomal enzyme activity or to diet?

Bears on: Mitochondrial / energy metabolism impairment Cell danger response (Naviaux) · Test: research only (plasmalogen panels exist in some specialist labs)

Metabolomics: aromatic lactates

Four aromatic 2-hydroxy acids (phenyllactate, 4-hydroxyphenyllactate, indolelactate, imidazole lactate) are lower in ME/CFS in 5 of 5 cohorts (combined -0.53 SD, I²=0), robust to drug, caffeine, diet and kidney-function adjustment. Several studies

A family of small chemicals made from amino acids is consistently lower in patients across five separate studies. It is a group-level difference, not a test.

Solving ME/CFS re-analysis: p=4e-11, holds on Naviaux's targeted LC-MS/MS platform (-0.64), -0.64 in people with no prescription drug detected, -0.73 in men. The precursor keto acids are not lower, so either the reduction step or a dietary source (fermented foods) is implicated. In BioMapAI paired stool, gut species do not predict these compounds (CV R²≈0), unlike IPA. They do not track severity (|r|<0.2). Not yet reported by original study authors; needs targeted, fasting confirmation and D- vs L-isomer separation (bacterial D vs host L).

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	5 cohorts (BioMapAI, Germain 2018/2020/2022, Naviaux)	harmonised meta-analysis with confounder models	combined -0.53 SD, 5/5 cohorts lower, I²=0; not stool-predicted; not severity-linked

Open questions

Is this host lactate-dehydrogenase-type reduction, renal handling, or fermented-food intake?
D- vs L-phenyllactate ratio?
Does it differ from other fatigue conditions?

Bears on: Mitochondrial / energy metabolism impairment Gut dysbiosis · Test: research only

Metabolomics: tryptophan / gut metabolites

Indole-3-propionate (IPA), a gut-bacterial tryptophan product, is lower in ME/CFS plasma in 4 of 4 cohorts (-0.33 SD) and tracks butyrate-producing gut species. Several studies

A chemical made only by certain gut bacteria is a bit lower in patients, which fits the gut-bacteria differences found in several studies.

Pooled -0.33 SD (95% CI -0.49 to -0.16), p=0.0001; individual AUC only 0.57-0.60. In BioMapAI paired stool, IPA correlates with F. prausnitzii and Bifidobacterium and inversely with R. gnavus; adjusting for IPA-linked species reduces the difference from -0.40 to -0.09 (optimistic same-data selection). Xiong 2023 and BioMapAI report reduced microbial butyrate and altered tryptophan metabolism. Interpreted as a gut-ecology marker rather than a host defect.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets / deep2) data	4 plasma cohorts	meta-analysis	IPA -0.33 SD, p=0.0001; AUC 0.57-0.60
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	10-fold CV AUC 0.915 (all omics); external cohorts AUC 0.59-0.77 (Guo, Raijmakers stool; Germain metabolome 0.68, Che 0.61) with 19-79% feature overlap
Xiong 2023, Cell Host Microbe data	short-term 75, long-term 79 vs 79 controls	case-control, stool metagenomics + plasma metabolomics	plasma butyrate, bile acids, benzoate lower; sphingomyelins depleted; dysbiosis mostly in short-term patients

Open questions

Does raising IPA (fibre, polyphenols) change anything clinically?
Is low IPA a consequence of diet changes caused by illness?

Bears on: Gut dysbiosis · Test: research only

Metabolomics: kynurenine pathway / IDO trap

The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood data are inconsistent: one cohort found higher kynurenine, others found no difference or slightly lower kynurenine. Contested

A mathematical theory proposed that a tryptophan-processing enzyme gets stuck in ME/CFS. Blood measurements of that pathway point in different directions in different studies.

Kashi, Davis and Phair 2019 modelled IDO1 substrate inhibition plus common non-functional IDO2 variants producing a bistable 'trap'; it is a hypothesis paper with no patient data. Kavyani 2024 reported higher kynurenine correlating with severity and lower downstream 3-HK, 3-HAA, QUIN, picolinate. Groven 2021 (49 CFS/57 FM/54 HC women) found differences mainly in FM, not CFS vs controls. The cross-cohort re-analysis found kynurenine -0.27 SD (lower), kyn/trp ratio -0.19 (q=0.06), lower kynurenine with >4 years illness, and no change in IDO1/KYNU proteins. Plasma kynurenine says little about intracellular IDO kinetics, which is what the model is about.

Evidence

Study	n	Design	Result
Kashi, Davis, Phair 2019, Diagnostics	none (model)	mathematical model + variant database search	IDO1 substrate inhibition + IDO2 loss + LAT1 asymmetry gives two steady states
Kavyani 2024, Mol Neurobiol	not in abstract	case-control, plasma KP metabolites + cytokines + proteomics	higher kynurenine linked to severity; lower 3-HK, 3-HAA, QUIN, PIC; hepcidin about halved

Against or not replicated

Study	n	Design	Result
Groven 2021, Psychoneuroendocrinology unverified	49 CFS, 57 FM, 54 HC (women)	case-control, plasma KP metabolites	main differences FM vs HC; quinolinic acid differs CFS vs FM; no clear CFS vs HC kynurenine shift
Solving ME/CFS re-analysis (deep2 metabolome-wide meta-analysis) data	10 cohorts, ~600 cases / ~430 controls	harmonised meta-analysis of public per-person data (Metabolon, Fiehn, LC-MS/MS, NMR)	kynurenine -0.27 SD (lower); kyn/trp -0.19, q=0.06; lower in >4 y illness; IDO1/KYNU protein null
Abujrais 2024, ACS Chem Neurosci	38 vs 24	case-control, targeted LC-MS	3-hydroxyanthranilic acid and 3-hydroxykynurenine differ; no headline kynurenine rise

Open questions

Do IDO2 variants differ in frequency in large genetic data (DecodeME)?
Intracellular tryptophan/kynurenine flux in patient cells under stress?

Bears on: IDO metabolic trap (Phair, Davis) Innate immune hyperactivation · Test: Plasma tryptophan/kynurenine available in some commercial labs; not diagnostic

Metabolomics: acylcarnitines and fatty acids

Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts, contrary to the earlier idea of excess circulating fatty acids. Several studies

Carriers that move fats into the cell's power stations are a little lower in patients' blood.

Reported as altered/low in Naviaux 2016, Nagy-Szakal 2018, Germain 2017 and Germain 2022 (carnitines low). Cross-cohort re-analysis: carnitine -0.21 SD, long-chain acylcarnitines -0.22 SD, long-chain fatty acids slightly lower in all 4 cohorts examined; acetylcarnitine/carnitine ratio null. Germain 2020 reported lower acyl cholines. Effect sizes are small and diet/fasting status is rarely recorded. Older acetylcarnitine findings (Reuter and Evans 2011) recalled from memory.

Evidence

Study	n	Design	Result
Nagy-Szakal 2018, Sci Rep data	50 vs 50	case-control, untargeted plasma metabolomics + stool metagenomics	altered choline, carnitine and complex lipids; ceramide higher in the ME/CFS+IBS subgroup; cross-validated AUC 0.82 (metabolome), 0.84 combined with stool
Germain 2017, Mol BioSyst	17 vs 15	pilot case-control, plasma metabolomics	74/361 metabolites p<0.05, 35 at q<0.15; fatty-acid and lipid pathways
Germain 2020, Metabolites data	26 vs 26 (women)	case-control, Metabolon + complex lipids (~1,750 compounds)	acyl cholines lower in two cohorts; steroid changes
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	multiple cohorts	meta-analysis	carnitine -0.21 SD; long-chain acylcarnitines -0.22 SD
Reuter & Evans 2011, J Intern Med unverified	44 patients	case-control serum carnitine profile	lower acetylcarnitine / altered acylcarnitines

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets) data	4 cohorts	meta-analysis	hypothesis of higher fatty acids in patients reversed: long-chain fatty acids and acylcarnitines slightly lower

Open questions

Muscle carnitine and fatty-acid oxidation flux in vivo?
Effect of fasting and dietary restriction on these values?

Bears on: Mitochondrial / energy metabolism impairment · Test: Plasma acylcarnitine profile (clinical metabolic labs); not diagnostic for ME/CFS

Metabolomics: energy metabolism

Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, TCA/urea cycle shifts), but simple blood ratios (lactate/pyruvate, citrulline/arginine) do not differ across pooled cohorts. Contested

Some studies suggest patients' cells burn sugar less efficiently. Blood measures that should show this are mostly normal when many studies are pooled.

Fluge 2016 (200/102) found low TCA-fuelling amino acids in women and higher PDK1/2/4 mRNA in PBMCs; Hoel 2021 described energy strain with altered fuel use in subsets; Yamano 2016 found TCA/urea-cycle ratios (validation AUC 0.75). Armstrong 2015 (NMR, 34/25 women) reported lower blood lactate and glutamate and higher glucose. In the pooled re-analysis, lactate/pyruvate (-0.10), citrulline/arginine, ornithine/arginine and hypoxanthine/xanthine were null; creatine/creatinine +0.19. These are inferences from plasma about tissue metabolism; direct tissue flux measurements are scarce.

Evidence

Study	n	Design	Result
Fluge 2016, JCI Insight	200 vs 102	case-control, serum amino acids + PBMC mRNA	lower TCA-fuelling amino acids mainly in women; PDK1/2/4, SIRT4, PPARδ mRNA higher; suggests impaired pyruvate dehydrogenase
Hoel 2021, JCI Insight hoel-2021 hoel-2021-perperson	83 vs 35	case-control, serum metabolomics, lipidomics, hormones	common changes consistent with energy strain and altered fatty-acid/amino-acid fuel use; 3 patient subsets (metabotypes), 2 with distinct energy-metabolism patterns
Yamano 2016, Sci Rep	133 samples (93 train / 40 validation)	case-control, CE-MS plasma	TCA and urea-cycle intermediates differ; ornithine/citrulline + pyruvate/isocitrate AUC 0.80 train, 0.75 validation
Armstrong 2015, Metabolomics unverified data	34 vs 25 (women)	case-control, 1H NMR blood and urine	lower lactate, acetate, glutamate; higher glucose (via review table)

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	multiple cohorts	pre-specified ratio meta-analysis	lactate/pyruvate, NO/urea-cycle and purine ratios null; no support for an arginine/NO substrate problem

Open questions

Does PDH activity measured in muscle biopsies differ?
Are Hoel's metabotypes stable over time in the same person?

Bears on: Mitochondrial / energy metabolism impairment Circulating (serum) factor · Test: research only

Metabolomics: ketones

Ketone bodies are associated with ME/CFS in UK Biobank NMR data, but 3-hydroxybutyrate shows no consistent difference in pooled case-control cohorts. Contested

One very large data set links ketones (fuel made from fat) to ME/CFS; smaller studies pooled together do not.

Huang 2024 (1,194 ME/CFS in UK Biobank, NMR) reported lipoprotein and ketone-body associations; their ML model (19 baseline characteristics + 9 NMR biomarkers) reached AUC 0.83, with much of the signal from non-omic features. Fasting time and diet differ by group and strongly affect ketones. Pooled re-analysis: 3-hydroxybutyrate null; ketones used as a covariate did not explain other signals.

Evidence

Study	n	Design	Result
Huang 2024, Commun Med	1,194 ME/CFS vs 53,009 controls + comorbid groups	UK Biobank NMR metabolomics, association + ML	168 biomarker associations incl. surface lipids and ketone bodies; AUC 0.83 with 19 baseline characteristics + 9 NMR markers

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets) data	4 cohorts	meta-analysis	3-hydroxybutyrate null

Open questions

Ketones under controlled fasting with orthostatic measures?

Bears on: Mitochondrial / energy metabolism impairment · Test: Blood beta-hydroxybutyrate (routine); not diagnostic

Metabolomics: exercise response

Germain 2022 reported growing metabolic divergence after a 2-day CPET; within-person re-analysis finds the exercise response itself barely differs (blunted lactate and glucose rise aside). Contested

A study took blood before and after two exercise tests and saw bigger differences after. Re-analysed person by person, the changes caused by exercise look mostly the same in patients and controls.

Germain 2022 (60/45, 1,157 metabolites, 4 time points) found female significant counts rising from 6 to 54 across time points. Re-analysis: day-1 and day-2 case-control differences correlate 0.85 and are only ~6% larger on day 2; within-person change scores are at chance (permutation p=0.12-0.61). Surviving signals: smaller lactate rise (likely lower workload), no post-exercise glucose rise in patients, oxidised cys-gly rising only in patients. Five of 8 people with missing post-exercise samples were severely affected men.

Evidence

Study	n	Design	Result
Germain 2022, JCI Insight data	60 vs 45	2-day CPET, 4 time points, Metabolon (1,157 metabolites)	baseline differences; authors report growing patient-control divergence after exercise in lipid and energy pathways

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (PEM) data	60 vs 45, 404 samples	within-person change, mixed models, permutation	exercise response at chance; 3 modest signals; no reproducible response subgroups (ARI 0.01)

Open questions

Samples at 48-72 h, when PEM peaks?
Workload-matched comparisons?

Bears on: Mitochondrial / energy metabolism impairment Capillary and microvascular damage · Test: research only

Metabolomics: collagen turnover

Collagen breakdown products (hydroxyproline, galactosyl-hydroxylysine and related) are higher in ME/CFS plasma (+0.52 SD, 4/4 Metabolon cohorts; hydroxyproline also up on 3 other platforms). Early

Fragments released when collagen (the body's connective-tissue protein) is broken down are a bit higher in patients.

Solving ME/CFS re-analysis only: p=2e-8 for the module; hydroxyproline +0.31 SD across 8 cohorts. Galactosyl-hydroxylysine weakens after drug adjustment. Collagen/gelatin supplements and diet are not excluded. Fits loosely with lower ECM/fibroblast and vascular mural proteins in the proteome re-analysis, but the direction (more breakdown, less matrix protein) needs direct markers (CTX-I, PIIINP).

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	4 Metabolon + 4 other cohorts	meta-analysis	collagen module +0.52 SD; hydroxyproline +0.31 SD

Open questions

Link to hypermobility / POTS subgroups?
Supplement and diet history?

Bears on: Connective tissue / hypermobility Capillary and microvascular damage · Test: research only

Redox / oxidative stress

Markers of oxidative stress (methionine sulfoxide, oxidised glutathione products, lower brain glutathione) are repeatedly reported, but they are non-specific and also seen in depression. Several studies

Signs that cells are under chemical 'rust' stress show up in several studies, but the same signs appear in other illnesses too.

Methionine sulfoxide/methionine +0.35 SD in 7/7 cohorts across 3 platforms (I²=0), though methionine oxidises during storage; 5-oxoproline/glutamate -0.25 SD (Solving ME/CFS re-analysis). Germain 2018 proposed broad redox imbalance; Germain 2022 data show oxidised cys-gly rising after exercise in patients only. Shungu 2012 (MRS, 15 CFS/15 MDD/13 HC) found lower cortical glutathione and higher ventricular lactate in both CFS and depression. Antioxidant trials have not shown clear benefit.

Evidence

Study	n	Design	Result
Germain 2018, Metabolites data	32 vs 19 (women)	case-control, Metabolon (832 metabolites)	14 metabolites differ; authors propose broad redox imbalance
Shungu 2012, NMR Biomed unverified	15 CFS, 15 MDD, 13 HC	1H MRS case-control	higher ventricular lactate, lower cortical glutathione in CFS and MDD; no CFS-MDD difference in GSH
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	7 cohorts	meta-analysis	methionine sulfoxide ratio +0.35 SD, I²=0

Open questions

Does MetSO hold in rapidly processed samples with matched storage time?
Is oxidative stress cause, consequence, or shared with any chronic illness?

Bears on: oxidative-stress Mitochondrial / energy metabolism impairment · Test: research only

NAD+

NAD+ depletion is a proposed mechanism but has not been directly established in ME/CFS cohorts; NAD precursor trials raise NAD+ without clear clinical benefit. Early

NAD+ is a molecule cells need for energy. The idea that patients lack it is mostly theory; supplements raise it but have not clearly helped.

Reviews (Dehhaghi 2022) link the kynurenine pathway to NAD+ synthesis. Castro-Marrero 2015 (n=73) and 2021 (n=207) RCTs of CoQ10 200 mg + NADH 20 mg reported modest improvements in some fatigue/cognitive scores. A long-COVID RCT of nicotinamide riboside 2 g/day (n=58) raised blood NAD+ 2.6-3.1x but showed no between-group difference in cognition or fatigue (p=0.59), with high dropout. Vitamin B3 pathway changes appeared in Abujrais 2024 untargeted data.

Evidence

Study	n	Design	Result
Dehhaghi 2022, Aging Dis	review	narrative review	links KP hyperactivation and NAD+ metabolism in ME/CFS
Castro-Marrero 2021, Nutrients	207 (104 vs 103)	12-week RCT CoQ10 + NADH	reduction in cognitive fatigue and some QoL measures reported
Castro-Marrero 2015, Antioxid Redox Signal	73	8-week RCT CoQ10 + NADH	fatigue impact score reduced; PBMC NAD+/NADH, ATP higher

Against or not replicated

Study	n	Design	Result
Wu 2025, eClinicalMedicine (long COVID)	58	24-week RCT nicotinamide riboside	NAD+ up 2.6-3.1x; no difference in cognition, fatigue, sleep

Open questions

Is intracellular NAD+/NADH measurably low in patient muscle or immune cells?

Bears on: Mitochondrial / energy metabolism impairment IDO metabolic trap (Phair, Davis) · Test: research only (some commercial NAD+ tests exist but are not validated for ME/CFS)

Proteomics and blood traits: UK Biobank

In UK Biobank, ME/CFS cases show many small blood-trait differences (inflammation, insulin resistance, liver markers) that replicate partly in All of Us and are not explained by low physical activity. Several studies

In a huge UK database, people with an ME/CFS diagnosis have many slightly different blood results, and this is not just because they move less.

Beentjes, Ponting and colleagues used semi-parametric (Super Learner, one-step) estimators with activity, walking pace and other mediators. 9 of 14 tested traits replicated in All of Us; people reporting PEM showed stronger differences. Caveats: diagnosis is from records/self-report, so case definition is loose; UK Biobank skews healthier and older; differences are small and single traits do not separate groups. Olink proteomics summary statistics show LEP, ADH4, BCHE, complement factors up and CA14, SOD3 down (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Beentjes 2025, EMBO Mol Med data	up to 1,455 cases vs 131,303 controls (UK Biobank); All of Us replication	population cohort, 3,237 blood traits, mediation analysis	hundreds of traits differ (116 in both sexes): inflammation, insulin resistance, liver; 9/14 replicated in All of Us; activity mediates only 1/3,237; no single trait discriminates

Open questions

Do these survive stricter case definitions (e.g. DecodeME-style questionnaires)?
How much reflects BMI, medication and comorbidity?

Bears on: Innate immune hyperactivation HPA axis blunting · Test: Routine labs (CRP, HbA1c, liver enzymes) exist but are not diagnostic

Proteomics: tissue-of-origin

Muscle-derived intracellular proteins (myoglobin, troponins, CA3 and similar) are lower in resting ME/CFS blood across three SomaScan cohorts and UK Biobank Olink, and this is not explained by activity measures. Several studies

Proteins that normally leak from muscle into blood are a little lower in patients, perhaps meaning less muscle or slower muscle turnover, not ongoing muscle damage.

Hoel 2026 reported reduced intracellular skeletal-muscle proteins. Solving ME/CFS re-analysis: muscle theme d -0.50 (95% CI -0.76 to -0.24), q=0.002, I² 0.07 across Hoel (-0.85), Germain (-0.43), Walitt (-0.34); UKB competitive shift negative; not mediated by walking pace, not related to steps, Bell score or time reclined. Lean mass was not measured, so reduced muscle mass remains a plausible explanation.

Evidence

Study	n	Design	Result
Hoel 2026, Cell Rep Med data	50 vs 29	case-control, SomaScan 7k serum proteomics	1,823 aptamers differ (845 FDR); secreted proteins up, intracellular (skeletal muscle) proteins down; lower activated-neutrophil secretion
Solving ME/CFS re-analysis (deep2 proteome) data	~150 vs ~105 per-person + UKB	random-effects meta-analysis of protein sets	muscle_damage set d -0.50, q=0.002

Open questions

Does it survive adjustment for DXA lean mass?
Relation to muscle biopsy fibre size and capillarisation?

Bears on: Mitochondrial / energy metabolism impairment Capillary and microvascular damage · Test: CK and myoglobin are routine but expected low-normal; not diagnostic

Proteomics: SomaScan artefacts

Most of the ~845 FDR-significant serum proteins in Hoel 2026 appear driven by a sample-handling (cell-leak) signal that is absent from other cohorts; about 13 survive adjustment. Contested

A large protein study found hundreds of differences, but a re-analysis suggests most came from how samples were handled rather than the illness.

Solving ME/CFS re-analysis built a nuclear/cell-leak score: case-control d -1.16 in Hoel, ~0 in Germain, Walitt serum and Walitt CSF. After adjustment Hoel keeps 13 FDR hits (COMP, C6, ANTXR2, LAMA2, FAP and others). Hoel's raw statistics correlate with Germain only through intracellular proteins, and not after adjustment. Hoel authors validated a selection by antibody methods. Across ~6,000 proteins meta-analysed in 4 cohorts, only 32 reach q<0.05 and replicated effects are d 0.25-0.5 (ITGAV, NOTCH3, THBS1, LUM, ADH4, GSTA1, BAFF up).

Evidence

Study	n	Design	Result
Hoel 2026, Cell Rep Med data	50 vs 29	case-control, SomaScan 7k serum proteomics	1,823 aptamers differ (845 FDR); secreted proteins up, intracellular (skeletal muscle) proteins down; lower activated-neutrophil secretion

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome) data	Hoel 50/24 + Germain + Walitt + UKB	artefact score and leak-adjusted meta-analysis	leak score separates only Hoel (d -1.16); 13 hits survive

Open questions

Were case and control samples processed with different delays?
Do the 13 surviving proteins replicate prospectively?

Bears on: Endothelial dysfunction Connective tissue / hypermobility · Test: research only

Proteomics: replicated small shifts

Hepatocyte enzymes (ADH4, GSTA1, AGXT, BCHE) are modestly higher, and vascular mural/matrix proteins (ITGAV, NOTCH3, LUM, THBS1, TGF-beta family) modestly lower, in more than one cohort. Early

A few groups of blood proteins, from the liver and from the walls of small vessels, are slightly shifted in several data sets.

Liver set d +0.40 (q=0.013) but defined after seeing data and tracks BMI; up in Germain and UKB, partly Hoel, not in long COVID. Mural/ECM: ITGAV -0.34 (4 cohorts), NOTCH3 -0.32, TGF-beta set -0.44 (q=0.013); a long COVID UK Olink set also shows lower ECM/endothelial proteins. Endothelial 'activation' markers go down, not up, and platelet/microclot proteins are not raised. Expected candidates (renin, ACE, endothelin, BNP, EPO, hepcidin, NEFL, GFAP, IL-6, CRP) are null.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome) data	4 blood cohorts	set and single-protein meta-analysis	liver +0.40; TGF-beta -0.44; ITGAV -0.34; many named candidates null
Beentjes 2025, EMBO Mol Med data	up to 1,455 cases vs 131,303 controls (UK Biobank); All of Us replication	population cohort, 3,237 blood traits, mediation analysis	hundreds of traits differ (116 in both sexes): inflammation, insulin resistance, liver; 9/14 replicated in All of Us; activity mediates only 1/3,237; no single trait discriminates

Open questions

Do liver proteins survive adjustment for fatty-liver index and ALT/GGT?
Do lower NOTCH3/ITGAV reflect pericyte loss or shedding?

Bears on: Capillary and microvascular damage Endothelial dysfunction Complement activation and thromboinflammation · Test: Routine liver tests; otherwise research only

Proteomics: exertion

Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPET, but a within-person re-analysis of the same data finds no protein whose exercise response differs at FDR. Contested

One large study reported that patients' blood proteins react differently to exercise. When re-analysed person by person, no single protein clearly did.

Germain 2025: 79 ME/CFS vs 53 sedentary controls, 6,361 proteins, 2 CPETs 24 h apart; pathway-level claims of suppressed T/B-cell and IL-17 signalling and raised glycolysis. Re-analysis: 0 FDR hits for differential change; no tissue score changes with exertion. Smaller EV studies after CPET (Giloteaux 2024, 18/17 women; Glass 2025, 10/12 men) report altered EV cargo responses (TCA down, complement up at 15 min; ER-stress proteins correlating with PEM), but n is small.

Evidence

Study	n	Design	Result
Germain 2025, Mol Cell Proteomics data	79 vs 53 (sedentary controls)	2-day CPET, SomaScan 7k, longitudinal	widespread changes after exertion, most in recovery; suppressed T/B signalling and IL-17, higher glycolysis proteins; sex differences
Giloteaux 2024, J Extracell Vesicles data	18 vs 17 (women)	CPET, EV TMT proteomics, 3 time points	EV proteome response differs; changes correlate with severity
Glass 2025, Clin Transl Med	10 vs 12 (men)	CPET, EV proteomics	TCA proteins down, complement up 15 min post-exercise; ER-stress changes track PEM

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome) data	Germain 79/51 per-person	within-person change	0 FDR hits for differential exercise response

Open questions

Sampling at 48-72 h?
Effort-matched comparison and correction for lower workload?

Bears on: Immune (T-cell) exhaustion Mitochondrial / energy metabolism impairment Complement activation and thromboinflammation · Test: research only

Extracellular vesicles

Plasma extracellular vesicle (EV) counts may be higher in ME/CFS, but no EV protein or microRNA difference has survived multiple-testing correction in a replicated cohort. Early

Tiny particles cells release into blood may be more numerous in patients, but their contents have not shown a reliable difference yet.

Rydland 2026 (49/50, SEC + MS proteomics): higher EV concentration; 11 of 424 proteins differ nominally (erythroid, hepatocyte, plasma-cell origin), none after FDR. Re-analysis of Rydland and Giloteaux EV data: no FDR hits. Eguchi 2026 EV-miRNA study had 6 ME/CFS. EV isolation methods vary widely and platelet activation during handling changes EV counts.

Evidence

Study	n	Design	Result
Rydland 2026, Biochem Biophys Rep data	49 vs 50	case-control, EV proteomics	more EVs; 11 nominal protein differences, none FDR
Eguchi 2026, J Transl Med	6 ME/CFS, 6 ICF, 8 depression, 4 HC	EV miRNA profiling	candidate EV-miRNAs; very small n

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome) data	Rydland + Giloteaux EV sets	re-analysis	no FDR hits

Open questions

Standardised EV isolation across sites?
Do EV counts change with PEM?

Bears on: Complement activation and thromboinflammation Circulating (serum) factor · Test: research only

CSF proteomics

CSF proteomics in ME/CFS is limited to small cohorts and has not produced a replicated protein difference; the largest recent study (31 patients) had no controls. Early

Studies of the fluid around the brain and spine are few and small, and have not found a consistent difference.

Bragee 2026 (31 patients, 902 proteins): within-patient associations with POTS (neutrophil degranulation, platelet activation) and severity (complement, coagulation, IGFBP); re-analysis found 0 FDR hits POTS vs no-POTS. Walitt 2024 CSF SomaScan (~17 vs 21) showed no FDR hits in re-analysis, and no rise in neuronal injury markers. Earlier CSF proteome reports (Schutzer 2011, Baraniuk 2005) are recalled from memory and small.

Evidence

Study	n	Design	Result
Bragee 2026, Sci Rep data	31 patients, no controls	CSF LC-MS proteomics	POTS- and severity-associated pathways; hypothesis-generating
Schutzer 2011, PLoS One unverified	~43 CFS vs controls (pooled)	CFS vs neurologic post-Lyme vs controls, CSF MS	distinct CSF proteomes reported

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome) data	Walitt CSF + Bragee	re-analysis	0 FDR hits; NEFL/GFAP null

Open questions

Adequately powered CSF study with controls and severe patients?

Bears on: Neuroinflammation (microglial activation) Brainstem and craniocervical instability · Test: research only

Transcriptomics and single-cell

Blood transcriptome and single-cell RNA studies report monocyte and platelet abnormalities, but no gene or cell subtype difference replicates across independent cohorts. Contested

Studies reading which genes are switched on in blood cells found possible differences, but they have not repeated in other groups of patients.

Vu 2024 (scRNA, ~30/28, before and 24 h after CPET) reported classical monocyte dysregulation correlating with severity and platelet activation signatures post-exercise. Re-analysis of the same data at person-level pseudobulk: 0 genes at q<0.05 at baseline in 15 cell types; composition null. Across 4 bulk RNA cohorts, 0 genome-wide FDR genes; a plasma-cell/immunoglobulin score is lower in 2/4 (meta d -0.42, p=0.06). Higher mitochondrial RNA fraction in lymphocytes (CD8 Tem d +0.85) is a lead that did not replicate in bulk PBMC. Walitt 2024 reported sex-specific PBMC expression differences in 17/21 people.

Evidence

Study	n	Design	Result
Vu 2024, Cell Rep Med data	~30 vs 28 (PBMC, pre/post CPET)	scRNA-seq	monocyte dysregulation tracking severity; platelet activation post-exercise
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 PI-ME/CFS vs 21 HV	deep phenotyping incl. PBMC RNA, flow, metabolomics	sex-specific gene expression and metabolic differences; fewer switched-memory B cells

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 single-cell/methylation, proteome) data	503,739 cells (58 people); 4 bulk RNA cohorts	person-level pseudobulk DE, cross-cohort meta	no replicated gene or cell-type difference

Open questions

An independent PBMC scRNA cohort (only one lab has published so far)?
Severe and housebound patients?

Bears on: Immune (T-cell) exhaustion Innate immune hyperactivation Herpesvirus reactivation (EBV, HHV-6) · Test: research only

Epigenetics / DNA methylation

Reported genome-wide blood DNA methylation differences have not replicated across cohorts, at least one large set is explained by cell-composition differences, and epigenetic clocks are not accelerated. Contested

Chemical tags on DNA were reported to differ in patients, but a closer look shows the differences mostly reflect which blood cells were in the sample.

de Vega 2014/2017/2018, Trivedi 2018 (13/12 discovery, 17,296 CpGs; pyrosequencing in 33/31) and Helliwell 2020 (RRBS, 10/10) each reported many differentially methylated sites with little overlap. Re-analysis of 4 cohorts (177/107): in GSE93266, control 'PBMC' DNA contained much more granulocyte DNA (median 48% vs 14%); 44,794 FDR CpGs fall to 0 after composition adjustment; meta-analysis 0 CpGs at q<0.05. Horvath, Hannum, PhenoAge, GrimAge, DunedinPACE null in T cells and saliva. Commercial epigenetic-age tests are not supported for ME/CFS.

Evidence

Study	n	Design	Result
Trivedi 2018, PLoS One data	13 vs 12 (+33/31 validation)	EPIC array + pyrosequencing	17,296 CpGs in 6,368 genes; 307 promoters
Helliwell 2020, Clin Epigenetics	10 vs 10	RRBS	76 fragments, 394 CpGs; overlap with earlier gene lists
de Vega 2014, PLoS One	PBMC case-control	450K array	differential methylation in immune genes

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 single-cell/methylation) data	4 cohorts, 177 vs 107	composition-adjusted EWAS, clocks	0 replicated CpGs; clocks normal

Open questions

Sorted-cell methylation in a large, age-matched cohort?

Bears on: Immune (T-cell) exhaustion Genetic susceptibility acting through neurons · Test: research only (consumer epigenetic-age tests are not validated here)

Diagnostics: nanoneedle impedance

A nanoelectronic impedance assay separated 20 ME/CFS from 20 controls perfectly after hyperosmotic stress, but there is no published independent replication. Early

A chip measuring how blood cells react to salt stress separated a small group of patients from controls completely. Nobody outside the lab has confirmed it.

Esfandyarpour 2019 (PNAS) used PBMCs incubated in plasma; patients' impedance rose after NaCl stress, controls' did not; plasma swap suggested a plasma factor. The only independent attempt found was a 4 vs 4 bioimpedance preprint (Fernandez 2023) with a different device, which reported higher impedance at one frequency and did not test the original protocol. No disease-control group (e.g. MS, depression) was reported in the original at scale; device-to-device variability is unaddressed in published work.

Evidence

Study	n	Design	Result
Esfandyarpour 2019, PNAS	20 vs 20	case-control, nanoelectronics impedance	complete separation after hyperosmotic stress; plasma swap effect

Against or not replicated

Study	n	Design	Result
Fernandez 2023, Authorea preprint	4 vs 4	bioimpedance spectroscopy, different device	higher impedance at 1.36 kHz; not a replication of the original protocol

Open questions

Blinded multi-site validation with disease controls?
What molecule in plasma drives the effect?

Bears on: Circulating (serum) factor Sodium-calcium overload in muscle (Wirth & Scheibenbogen) · Test: research only

Diagnostics: Raman spectroscopy

Raman spectroscopy of PBMCs or plasma separates ME/CFS from controls in single-centre studies (plasma AUC 0.83-0.85), without external validation. Early

A laser technique that reads the chemical 'fingerprint' of blood looked promising in two labs, but has not been tested in new patients elsewhere.

Xu 2019 (single-cell Raman, PBMCs) found higher phenylalanine-associated bands in patients, similar to mtDNA-depleted rho0 cells; small n. Heidarifard 2026 (115 ME vs 45 sedentary controls, plasma at rest and 90 min after a standardised stress) reported AUC 0.85 and 0.83, specificity 73% and 69%. Models are trained and tested within one cohort; spectral features are hard to map to specific molecules.

Evidence

Study	n	Design	Result
Xu 2019, Analyst	small (PBMCs from patients and controls)	single-cell Raman microspectroscopy	higher phenylalanine bands in CFS PBMCs
Heidarifard 2026, Int J Mol Sci	115 vs 45	plasma Raman + ML, rest and post-stress	AUC 0.85 (rest), 0.83 (post-stress)

Open questions

Independent external test set and disease controls?

Bears on: Mitochondrial / energy metabolism impairment · Test: research only

Diagnostics: cell-free RNA

Plasma cell-free RNA classified ME/CFS vs sedentary controls with test-set AUC 0.81 and suggested platelet, pDC and monocyte-derived differences; not yet externally validated. Early

RNA fragments floating in blood plasma could tell patients from controls about 4 times in 5 in one study.

Gardella 2025 (PNAS; 93 vs 75 sedentary controls): LASSO model, held-out test AUC 0.81, accuracy 77%; deconvolution suggested differences in platelet-derived cfRNA and higher pDC, monocyte and T-cell cfRNA; network analysis pointed to cytokine signalling and T-cell exhaustion. Held-out test within one cohort; the plasma-cell score that dips in PBMC data was null in this cfRNA set in re-analysis.

Evidence

Study	n	Design	Result
Gardella 2025, PNAS data	93 vs 75	case-control, plasma cfRNA-seq + ML	test AUC 0.81; platelet, pDC, monocyte cfRNA differences

Open questions

Performance in a second cohort and against fibromyalgia, depression, long COVID?

Bears on: Immune (T-cell) exhaustion Complement activation and thromboinflammation · Test: research only

Diagnostics: machine-learning classifiers

Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-validation but drop to AUC ~0.6-0.77 on external cohorts, so none is usable for diagnosis. Contested

Computer models trained on many blood and stool measurements do well on the data they learned from but much worse on new patients.

BioMapAI (Xiong 2025, 249 participants, 4-year longitudinal sampling) reached 10-fold CV AUC 0.915 with all omics; on external cohorts AUC was 0.72 and 0.59 (Guo stool), 0.61-0.77 (Raijmakers), 0.68 (Germain metabolome) and 0.61 (Che), with only 19-79% feature overlap. Repeated visits per person raise the question of whether CV folds were split by person; disease controls were not included. A small metabolite panel derived in the Solving ME/CFS re-analysis transferred with AUC 0.60-0.83. Huang 2024 UK Biobank AUC 0.83 relied heavily on 19 baseline (non-omic) characteristics. No multi-omic layer converges on a shared pathway beyond chance (re-analysis).

Evidence

Study	n	Design	Result
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	10-fold CV AUC 0.915 (all omics); external cohorts AUC 0.59-0.77 (Guo, Raijmakers stool; Germain metabolome 0.68, Che 0.61) with 19-79% feature overlap
Huang 2024, Commun Med	1,194 ME/CFS (UK Biobank)	NMR + baseline characteristics ML	AUC 0.83, recall 0.70
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	20 pathways abnormal; 80% of discriminating metabolites lower; AUC 0.94 (men, 8 metabolites) and 0.96 (women, 13) within the same data

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets) data	two Cornell cohorts, 575 shared compounds	cross-cohort concordance	patient-control differences agree only faintly (rho 0.16); model trained on one scores AUC 0.56 on the other; a public GitHub copy of Germain 2020 has case/control labels swapped
Solving ME/CFS re-analysis (deep2 proteome convergence) data	4 omic layers	pathway-score correlation with permutation null	no protein-RNA-metabolite-genetics convergence beyond chance (one modest protein-metabolite rho 0.28)

Open questions

Pre-registered external validation with disease controls (fibromyalgia, depression, MS, long COVID)?
Person-level train/test splits for longitudinal data?

Bears on: Gut dysbiosis Immune (T-cell) exhaustion · Test: research only

Most comprehensive blood studies

Even the deepest blood omics studies sample mainly at rest, in mild-to-moderate patients, in plasma or serum only, and rarely test fresh-cell function or the 48-72 h PEM window. Replicated

The biggest blood studies so far mostly leave out the sickest patients, the days when symptoms flare after effort, and live-cell tests.

Walitt 2024 (NIH, 17/21) is the deepest per person but tiny and excluded many severe patients by design. BioMapAI has 4 years of visits but resting samples only. Germain 2022/2025 and Giloteaux/Glass cover 0, 15 min and 24 h after CPET but not 48-72 h, and the most impaired cannot do CPET (5 of 8 missing post-exercise samples were severe men). UK Biobank is large but one baseline time point with loose case definition. Hoel 2021/2026 are resting serum. Fresh-cell assays (PBMC respiration, NK function, red-cell deformability) are rarely paired with omics; compartments like muscle, CSF and gut tissue are rarely sampled together.

Evidence

Study	n	Design	Result
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	deep phenotyping, multiple compartments	many layers measured, small n
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	10-fold CV AUC 0.915 (all omics); external cohorts AUC 0.59-0.77 (Guo, Raijmakers stool; Germain metabolome 0.68, Che 0.61) with 19-79% feature overlap
Germain 2025, Mol Cell Proteomics data	79 vs 53 (sedentary controls)	2-day CPET, SomaScan 7k, longitudinal	widespread changes after exertion, most in recovery; suppressed T/B signalling and IL-17, higher glycolysis proteins; sex differences
Beentjes 2025, EMBO Mol Med data	up to 1,455 cases vs 131,303 controls (UK Biobank); All of Us replication	population cohort, 3,237 blood traits, mediation analysis	hundreds of traits differ (116 in both sexes): inflammation, insulin resistance, liver; 9/14 replicated in All of Us; activity mediates only 1/3,237; no single trait discriminates

Open questions

Who will run a multi-omic study in severe/housebound patients with home sampling?
Dense sampling through a PEM episode?

Test: n/a

Key reviews

Yamano 2021, Int J Mol Sci: Overview of metabolite biomarker studies in ME/CFS and animal models.
Davis 2025, Front Neurosci: Review of lipid, energy and oxidative-stress omics across ME/CFS, Gulf War illness and fibromyalgia.
Kavyani 2022, Cell Mol Life Sci: Review of the kynurenine pathway case in ME/CFS.
Beentjes 2025, EMBO Mol Med: Largest blood-trait analysis, with explicit handling of inactivity as a confounder.
Xiong 2025 (BioMapAI), Nat Med: Largest longitudinal multi-omic data set; reports external validation numbers worth reading critically.

Hormones

HPA axis

On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and morning cortisol, 24-h urinary cortisol and hair cortisol) than healthy controls. Replicated

Cortisol is the body's main stress and energy-mobilising hormone. On average it runs a little low in ME/CFS, but the gap is modest and most individual results are still normal.

Woo et al. 2026 pooled 46 case-control datasets (1,388 patients, 1,349 controls, 71.9% female) and found lower salivary awakening and morning cortisol, 24-h urinary free cortisol and hair cortisol. Hair cortisol, which reflects about 3 months of exposure, was also lower in a small adolescent study (Vroegindeweij 2024). In the Solving ME/CFS literature tally, 19 groups report lower cortisol and 5 higher. In untargeted plasma metabolomics the signal is weak and inconsistent: Germain 2020 −0.64 SD, Germain 2022 −0.13, BioMapAI −0.05, with men −0.35 (Solving ME/CFS re-analysis). The meta-analysis does not separate cause from effect: sleep disruption, inactivity, illness duration and antidepressant use can all lower cortisol.

Evidence

Study	n	Design	Result
Woo 2026, Molecular Psychiatry	46 studies; 1,388 vs 1,349	systematic review and meta-analysis	Lower salivary awakening/morning, 24-h urinary and hair cortisol
Vroegindeweij 2024, Psychoneuroendocrinology	12 ME/CFS, 20 Q-fever fatigue, 8 post-COVID, 19 JIA vs 57 controls	case-control (LC-MS/MS hair cortisol)	Lower hair cortisol in fatigued patients than controls
Solving ME/CFS re-analysis (epi_literature tally) data	24 study groups	vote count of published studies	19 lower vs 5 higher cortisol

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis of BioMapAI plasma metabolome data	about 236	case-control re-analysis	Plasma cortisol −0.05 SD overall (men −0.35, p=0.04)

Open questions

Is low cortisol present at onset, or does it develop with illness duration and inactivity?
Does cortisol track PEM or severity within a person over time?
How much do medication, sleep timing and sampling time explain?

Bears on: HPA axis blunting Persistent sickness behaviour / vagal signalling · Test: Yes: 9am serum cortisol, salivary awakening cortisol, short Synacthen test (mainly to exclude true adrenal insufficiency)

Challenge tests show a blunted cortisol response to ACTH and exaggerated cortisol suppression after a glucocorticoid dose, consistent with enhanced negative feedback. Several studies

When the stress-hormone system is pushed or braked in a test, it responds less to the push and more to the brake than in healthy people.

Woo 2026 included 12 pharmacological challenge studies: impaired cortisol release after ACTH stimulation and greater suppression after dexamethasone or hydrocortisone. The pattern resembles the 'hypocortisolism with enhanced feedback' described in PTSD, fibromyalgia and atypical depression, so it is not specific to ME/CFS. Insulin stress tests in the Cleare 1999 trial showed the adrenal glands were not suppressed by low-dose hydrocortisone. POTS patients showed normal adrenal responses to ACTH (Laurin 2023, n=8 vs 5).

Evidence

Study	n	Design	Result
Woo 2026, Molecular Psychiatry	12 challenge studies within 46	meta-analysis	Impaired ACTH-stimulated cortisol; exaggerated glucocorticoid suppression

Against or not replicated

Study	n	Design	Result
Laurin 2023, Autonomic Neuroscience	8 POTS vs 5	ACTH stimulation, low-sodium diet	Normal aldosterone and cortisol responses in POTS

Open questions

Is the feedback change central (hypothalamus/pituitary) or adrenal?
Does it differ between infection-onset and gradual-onset cases?

Bears on: HPA axis blunting · Test: Research only (dexamethasone suppression or CRH tests); short Synacthen test is clinical

Lower cortisol in long COVID, reported in a 2023 Nature cohort, has not held up in later clinic and cohort studies that controlled for sampling time. Contested

One high-profile study found low cortisol in long COVID. Later studies that timed the blood draws carefully mostly found normal levels.

Klein 2023 (n=275) reported lower cortisol in long COVID among other immune features. Dalhuisen 2024 (UCSF, n=200, banked 8am–12pm samples) found no overall difference (median 8.9 vs 8.8 µg/dL). Its sub-hour analyses were inconsistent: lower at 8–9am but higher at 9–10am in the ME/CFS-like group. Quach 2025 (Stanford clinic, n=86, 5–10am) found only 1 low value; 36% were above range. Camici 2026 (Rome, n=96) found normal blood cortisol but flattened salivary rhythms. Post-COVID controls without symptoms showed the same salivary pattern, so the change may follow infection rather than long COVID itself.

Evidence

Study	n	Design	Result
Klein 2023, Nature data	275 total	cross-sectional immune profiling	Cortisol lower in long COVID
Camici 2026, Front Cell Infect Microbiol	83 LC, 13 recovered, controls	case-control salivary profile	Reduced morning salivary cortisol, flattened rhythm; blood cortisol not different; ACTH higher

Against or not replicated

Study	n	Design	Result
Dalhuisen 2024, medRxiv	144 LC vs 56 recovered	cross-sectional, banked plasma	No overall difference in morning cortisol
Quach 2025, Biomedicines	86	retrospective clinic series	Low morning cortisol in 1.2%; 36% elevated

Open questions

Do serial or salivary measurements in long COVID with PEM reproduce the ME/CFS pattern?
Is the change post-infectious and non-specific?

Bears on: HPA axis blunting · Test: Yes: timed morning cortisol; salivary diurnal profile (private labs)

HPA axis treatment

Low-dose hydrocortisone produced at most small, short-term fatigue benefits in two RCTs, at the cost of adrenal suppression and bone loss at higher doses. Contested

Replacing some cortisol was tried in the 1990s. The effect was small, side effects were real, and it is not recommended.

McKenzie 1998 (JAMA, n=70, 25–35 mg/day for 12 weeks): global wellness responders 66.7% vs 54.3% (p=0.31). A secondary 5-point improvement was 53% vs 29% (p=0.04), but cosyntropin tests showed adrenal suppression. A sub-study (n=23) found lumbar spine bone density fell 2.0% on hydrocortisone. Cleare 1999 (Lancet, n=32 crossover, 5–10 mg for 1 month) found fatigue fell 4.5 points more than on placebo; 28% vs 9% reached normal-range fatigue. The trial was short, strictly selected (no psychiatric comorbidity) and never replicated. Blockmans 2003 (n=100 crossover, hydrocortisone 5 mg plus fludrocortisone 50 µg for 3 months) was null.

Evidence

Study	n	Design	Result
Cleare 1999, Lancet	32 (crossover)	RCT crossover	Fatigue reduced vs placebo (paired difference 4.5 points, p=0.009)
McKenzie 1998, JAMA	70	RCT	Primary outcome ns (p=0.31); adrenal suppression
McKenzie 2000 (BMD sub-study of the 1998 trial)	23	RCT sub-study	Lateral spine BMD −2.0% on hydrocortisone

Against or not replicated

Study	n	Design	Result
Blockmans 2003, Am J Med	100 (80 completed)	RCT crossover, hydrocortisone + fludrocortisone	No difference in fatigue or well-being

Open questions

Would a subgroup with objectively low cortisol or blunted Synacthen response benefit?
No modern trial has used PEM-based criteria.

Bears on: HPA axis blunting · Test: Not applicable (treatment); not recommended outside proven adrenal insufficiency

Salt-retaining hormones (treatment)

Fludrocortisone, a synthetic salt-retaining steroid, did not improve CFS with neurally mediated hypotension in a 100-person RCT, and failed again in combination with hydrocortisone. Refuted

A drug that makes the kidneys hold salt and water was expected to help low blood pressure on standing. In proper trials it did not help overall.

Rowe 2001 (JAMA, n=100, tilt-positive neurally mediated hypotension, 0.1 mg/day for 9 weeks): ≥15-point wellness improvement in 14% vs 10% (p=0.76). Blockmans 2003 (n=100) found no effect of hydrocortisone plus fludrocortisone. Small open series in POTS still report individual responders, and the trials did not select for measured hypovolaemia or low aldosterone.

Evidence

Study	n	Design	Result
Rowe 2001, JAMA	100	RCT	No benefit (14% vs 10% responders)

Open questions

Would patients selected for low aldosterone or measured low plasma volume respond?

Bears on: Low blood volume and preload failure HPA axis blunting · Test: Not applicable (treatment)

Adrenal and pituitary disease

Diagnosed adrenal insufficiency and hypopituitarism are 7 to 13 times more common than expected among Finnish patients coded with postviral fatigue syndrome. Early

A small minority of people labelled ME/CFS may have a real, treatable gland failure. That makes basic hormone screening worthwhile.

In Finnish registry data (FinRegistry about 1,622 G93.3 cases, FinnGen 348), the crude observed/expected ratio is 9.5–13 for adrenal insufficiency and 6.7–7.6 for hypopituitarism. Hydrocortisone and fludrocortisone prescriptions are enriched 11–62x (Solving ME/CFS re-analysis of Risteys). These figures are not adjusted for age or sex and are inflated by healthcare use. They may reflect misdiagnosis, a true subgroup, or steroid treatment causing adrenal suppression. Case reports describe isolated ACTH and GH deficiency first labelled as CFS, which improved with hydrocortisone (Tokumasu 2021).

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (FinnGen/FinRegistry Risteys) data	about 1,622 and 348 cases	registry co-occurrence (O/E)	Adrenal insufficiency O/E 9.5–13; hypopituitarism 6.7–7.6
Tokumasu 2021, BMJ Case Reports	1	case report	Idiopathic ACTH + GH deficiency mimicking CFS

Open questions

What fraction of a well-phenotyped ME/CFS cohort fails a Synacthen test?
Is this misdiagnosis, a subgroup, or a treatment effect?

Bears on: HPA axis blunting Pituitary / vasopressin deficiency · Test: Yes: 9am cortisol, ACTH, short Synacthen test; pituitary panel if indicated

Water balance (ADH/vasopressin)

After overnight fluid deprivation, plasma vasopressin was undetectable in 82% of 111 Swedish ME/CFS patients despite high plasma osmolality in 57%, suggesting partial central diabetes insipidus-like physiology. Early

Vasopressin (ADH) is the hormone that tells the kidneys to keep water. Some patients appear to make too little of it even when their blood is concentrated, which could worsen low blood volume.

Huhmar 2025 was a consecutive, uncontrolled clinic series (n=124). Urine osmolality was below 750 mOsm/kg in 66%. In the 13 patients tested with copeptin, which is a more stable marker of vasopressin secretion, all values were within the reference range, though at the low end. Brain MRI showed no hypothalamic or pituitary lesions. Miwa 2016 (18 vs 15) found ADH 2.2 vs 3.3 pg/mL (p=0.02) alongside lower aldosterone and small hearts. In open label use, desmopressin improved orthostatic symptoms in 5 of 10 patients. Vasopressin immunoassays are unreliable, and the normal copeptin results argue for caution.

Evidence

Study	n	Design	Result
Huhmar 2025, Endocrine Practice	124 (111 vasopressin, 13 copeptin)	observational, uncontrolled	Vasopressin undetectable in 82%; copeptin normal-low
Miwa 2016, J Cardiology	18 vs 15	case-control	Lower ADH (p=0.02) and aldosterone (p=0.004)

Against or not replicated

Study	n	Design	Result
Huhmar 2025 copeptin subset	13	observational	All copeptin values within reference range

Open questions

Does a controlled water-deprivation or hypertonic saline copeptin test confirm low vasopressin?
Does low vasopressin predict desmopressin response?
Is it central (hypothalamic) or due to altered osmoreceptor input?

Bears on: Pituitary / vasopressin deficiency Low blood volume and preload failure · Test: Yes: copeptin with paired plasma and urine osmolality (fasting); formal water deprivation or hypertonic saline test via endocrinology

Renin-angiotensin-aldosterone

Patients with POTS and with ME/CFS show normal or low renin and low aldosterone despite reduced blood volume (the 'renin-aldosterone paradox'). Several studies

When blood volume is low, the kidneys should release renin and the adrenals aldosterone to retain salt. In POTS and ME/CFS this response seems muted.

Raj 2005 (POTS 15 vs 14, iodine-131 albumin) found a plasma volume deficit of 334 vs 10 mL with unchanged renin activity and aldosterone of 190 vs 380 pmol/L (p=0.017). Miwa 2016 (ME 18 vs 15) found aldosterone 104 vs 157 pg/mL (p=0.004) and renin activity 1.6 vs 2.5 (p=0.06). The adrenal glands respond normally to ACTH in POTS (Laurin 2023), so the defect is upstream of the adrenal. In SomaScan and Olink proteomics, renin, ACE and BNP show no group difference (Solving ME/CFS re-analysis), but those assays measure protein abundance, not posture-stimulated activity.

Evidence

Study	n	Design	Result
Raj 2005, Circulation	15 vs 14	case-control (POTS)	Low aldosterone, unchanged renin despite hypovolaemia
Miwa 2016, J Cardiology	18 vs 15	case-control (ME)	Lower aldosterone and borderline lower renin

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 proteome) data	3 SomaScan cohorts + UKB	re-analysis	Renin, ACE, BNP, EPO, hepcidin not different at rest
Zhang 2025, Vasc Health Risk Manag	65 vs 65 children (POTS)	case-control	Angiotensin II slightly higher in paediatric POTS

Open questions

Are supine/upright renin and aldosterone low in PEM-defined ME/CFS with measured blood volume?
Is the defect renal, sympathetic or central?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Yes: supine and upright renin and aldosterone; blood volume (Daxor or CO-rebreathing) in specialist centres

Thyroid

In one Dutch case-control study, CFS patients had normal TSH but lower total T3, T4 and free T3, with a higher reverse-T3 fraction ('low T3 syndrome' pattern in 16% vs 7%). Early

The thyroid itself looks normal by standard tests, but the active thyroid hormone (T3) runs slightly lower, as seen in many long-term illnesses.

Ruiz-Núñez 2018 (98 vs 99, age/sex matched): TT3 −12.5%, TT4 −11.9%, deiodinase activity −14.4%, rT3 fraction +13.3%. FT3 below range in 16/98 vs 7/99 (OR 2.56, 95% CI 1.00–6.54). The median difference in free T3 was tiny (0.1%), and thyroid values correlated with hsCRP. This pattern ('non-thyroidal illness') appears in many chronic and inflammatory states and is usually adaptive. It has not been independently replicated in ME/CFS. There is no trial of T3 treatment in ME/CFS.

Evidence

Study	n	Design	Result
Ruiz-Núñez 2018, Front Endocrinol	98 vs 99	case-control	Lower TT3/TT4/FT3, higher %rT3, normal TSH; low-T3 OR 2.56

Open questions

Does it replicate in a PEM-defined cohort?
Is it a sign of reduced metabolic rate (hypometabolism) or just non-specific illness?

Bears on: HPA axis blunting Mitochondrial / energy metabolism impairment · Test: Yes: TSH, free T4, free T3; reverse T3 (some labs)

Growth hormone

Reports of low IGF-1 and reduced growth hormone responses in CFS conflict, and the largest controlled study found an intact GH/IGF system. Contested

Growth hormone helps with repair and energy. Early small studies hinted it was low, but a larger careful study found it normal.

Allain 1997 found lower IGF-I (214 vs 263 µg/L) and IGF-II and a reduced GH response to hypoglycaemia. Berwaerts 1998 (20 vs 20) found IGF-I SDS −0.39 vs +0.33 (p=0.02) and a trend to lower nocturnal GH (p=0.06), but normal stimulated GH peaks. Cleare 2000 (37 vs 37, medication-free) found no difference in IGF-I/IGFBPs, urinary GH or GHRH and insulin-stress responses. In a GH treatment RCT (Moorkens 1998, n=20) IGF-1 rose but quality of life did not improve.

Evidence

Study	n	Design	Result
Allain 1997, Biol Psychiatry	small case-control	case-control with insulin tolerance test	Lower IGF-I/IGF-II; blunted GH response
Berwaerts 1998, Growth Horm IGF Res	20 vs 20	case-control	Lower IGF-I SDS; normal stimulated GH

Against or not replicated

Study	n	Design	Result
Cleare 2000, J Clin Endocrinol Metab	37 vs 37	case-control with GHRH and insulin tests	No GH deficiency; IGF system intact
Moorkens 1998, Growth Horm IGF Res	20	RCT of GH	IGF-1 rose; quality of life not improved

Open questions

Is nocturnal GH reduced as a consequence of non-restorative sleep?

Bears on: HPA axis blunting · Test: Yes: IGF-1 (stimulation tests only if pituitary disease suspected)

Sex differences

ME/CFS and long COVID are more common in women (ME/CFS prevalence about 1.4% vs 0.9%; DecodeME 82% female), but genetic risk variants are not sex-specific. Replicated

More women than men get ME/CFS and long COVID. Inherited risk looks the same in both sexes, so the difference probably lies in hormones, immunity or exposure, not in different genes.

Lim 2020 meta-analysis: pooled prevalence 1.36% in women vs 0.86% in men. Clinic cohorts are typically 75–85% female. In RECOVER-Adult (n=12,276), female sex raised long COVID risk (RR 1.31, 95% CI 1.06–1.62), but not at age 18–39 (RR 1.04). At age 40–54 the RR was similar for menopausal (1.42) and non-menopausal (1.45) women. DecodeME: 12,833 female and 2,746 male cases, no sex-biased loci, and two female-discovered signals replicate in males. The mostly male Million Veteran Program replicates the polygenic signal (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Lim 2020, J Transl Med	46 studies, about 1.09 million	meta-analysis of prevalence	Women 1.36% vs men 0.86%
Shah 2025, JAMA Netw Open (RECOVER)	12,276	prospective cohort	Female RR 1.31 for long COVID; age- and pregnancy-dependent
DecodeME 2025, medRxiv preprint data	15,579 cases	GWAS	No sex-biased loci; male replication of 2 signals

Open questions

Which biological factor (sex hormones, X-linked immune genes, autoimmunity propensity) drives the excess?
Why is the excess absent in young adults in RECOVER?

Bears on: HPA axis blunting Autoimmunity (functional autoantibodies) · Test: Not applicable

Sex hormones and menopause

Women with CFS report more endometriosis, heavy bleeding, pelvic pain, hysterectomy and earlier menopause than controls in US population-based case-control studies. Early

Gynaecological problems and early menopause are more common in women with ME/CFS. We do not know which comes first or why.

Boneva 2015 (Georgia, 84 cases vs 73 controls) reported earlier menopause (37.6 vs 48.6 years; adjusted OR 1.22 per year), endometriosis 29.8% vs 12.3% (aOR 3.67), hysterectomy 54.8% vs 19.2% (aOR 3.23) and non-menstrual pelvic pain (aOR 12.0). The Wichita sample (36 vs 48) showed similar but mostly non-significant patterns. Surgical menopause after hysterectomy or oophorectomy confounds the 'early menopause' result, and recall bias is possible. 2026 survey work (Habermann-Horstmeier) describes different symptom structures by sex and a menopausal-status proxy, but it is cross-sectional and self-reported.

Evidence

Study	n	Design	Result
Boneva 2015, Menopause	84 vs 73	population-based case-control	Earlier menopause, more endometriosis and hysterectomy
Boneva 2010, J Womens Health	36 vs 48	population-based case-control	More pelvic pain and endometriosis; menopause about 4 years earlier (ns)
Habermann-Horstmeier 2026, J Transl Med	survey cohort	cross-sectional factor analysis	Immune/GI symptom architecture differs by sex and menopausal proxy

Open questions

Does onset cluster at perimenopause after adjusting for age?
Does HRT change ME/CFS course? No trial exists.
Do symptoms vary with menstrual cycle in a measurable way?

Bears on: HPA axis blunting · Test: Clinical gynaecological assessment; no ME/CFS-specific hormone test

Sex hormones

Lower plasma androgenic steroids in ME/CFS (Cornell metabolomics) is mostly an age and sex artefact and does not replicate in BioMapAI. Contested

An earlier claim that patients have low 'male-type' steroid hormones mostly disappears once age and sex are accounted for.

Across the two Cornell cohorts the androgenic steroid class was −0.42 SD, but in Germain 2022 the effect shrinks after age/sex/BMI adjustment. BioMapAI (n=236) shows −0.01 (p=0.90) and Che testosterone −0.18 (p=0.22) (Solving ME/CFS re-analysis). Cleare 2004 measured DHEA/DHEA-S and their response to CRH as a separate question.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets 03_steroids_crosscheck) data	4 cohorts	re-analysis with covariate adjustment	Androgen deficit not replicated after adjustment

Open questions

Are sex steroids altered within defined subgroups (e.g. men, post-menopausal women)?

Bears on: HPA axis blunting · Test: Yes: testosterone, SHBG, DHEA-S (not diagnostic for ME/CFS)

Melatonin and circadian rhythm

Melatonin may help the subgroup with delayed melatonin onset, and melatonin plus zinc modestly reduced physical fatigue in one small RCT. Early

Melatonin is the night-time hormone. A small trial and an open study suggest it can help some patients, especially those whose body clock runs late.

van Heukelom 2006 (open-label, n=29 with DLMO after 21:30; 5 mg 5 h before DLMO for 3 months) improved fatigue, concentration and activity, more when DLMO was after 22:00. Fatigue normalised in 8/27. Castro-Marrero 2021 (RCT, n=50, melatonin 1 mg plus zinc 10 mg for 16 weeks) reduced physical fatigue at the last visit and improved SF-36 physical summary. Multiple outcomes were tested and the trial was not replicated. An older trial of melatonin and phototherapy reported no benefit (Williams 2002, from memory).

Evidence

Study	n	Design	Result
Castro-Marrero 2021, Antioxidants	50	RCT	Reduced physical fatigue vs placebo (p<0.05)
van Heukelom 2006, Eur J Neurol	29	open-label, DLMO-selected	Fatigue improved, mainly if DLMO >22:00

Against or not replicated

Study	n	Design	Result
Williams 2002, Eur J Clin Invest unverified	about 30	randomised crossover (melatonin, phototherapy)	No benefit (from memory)

Open questions

What proportion of patients have delayed DLMO?
Does timed low-dose melatonin help in a blinded DLMO-selected trial?

Bears on: HPA axis blunting · Test: Research/private: salivary dim-light melatonin onset (DLMO); actigraphy

Genetics of hormone systems

ME/CFS genetic risk is enriched in brain tissues and the pituitary, but genetically predicted morning cortisol does not change risk. Early

Genes linked to ME/CFS are especially active in the brain and the pituitary, the master hormone gland. Genetically high or low cortisol does not seem to cause ME/CFS.

In a DecodeME + Million Veteran Program meta-analysis (Maccallini 2026 preprint, 19,470 cases), tissue enrichment implicated multiple brain regions and the pituitary, with no peripheral tissue. MR using a single morning-cortisol instrument gave OR 1.04 (Solving ME/CFS re-analysis), which is underpowered. The pituitary result could reflect neural gene expression shared with brain rather than endocrine function.

Evidence

Study	n	Design	Result
Maccallini 2026, Research Square preprint	19,470 vs 699,111	GWAS meta-analysis + enrichment	Brain regions and pituitary enriched
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME 15,579 cases	Mendelian randomisation	Morning cortisol (1 SNP) OR 1.04, null

Open questions

Do pituitary-specific eQTLs colocalise with ME/CFS loci?
Can a multi-SNP cortisol instrument test causality properly?

Bears on: HPA axis blunting Pituitary / vasopressin deficiency Genetic susceptibility acting through neurons · Test: Research only

Key reviews

Woo 2026, Molecular Psychiatry: Most comprehensive meta-analysis of HPA findings (46 studies, all sample types and challenge tests).
Zhang & Wang 2024, Front Endocrinol: Review of HPA changes and HPA-targeted treatments in CFS.
Wirth 2026, Preprints.org: Hypothesis review of why vasopressin and RAAS fail to correct hypovolaemia (not peer reviewed).
Miwa 2026, Front Med: Clinical update on small heart, orthostatic intolerance and down-regulated volume hormones.

Gut and microbiome

Dysbiosis

Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium prausnitzii, Eubacterium rectale, Roseburia, Coprococcus) and reduced microbial capacity to make butyrate. Replicated

Some 'good' gut bacteria that feed the gut lining by making butyrate are consistently scarcer in ME/CFS.

Guo 2023 (106 vs 91, shotgun metagenomics, qPCR and stool SCFA) found lower F. prausnitzii and E. rectale and lower butyrate synthesis capacity. Its classifier was robust across US sites and to a validation cohort. Xiong 2023 (75 short-term, 79 long-term vs 79) found reduced butyrate biosynthesis genes and lower plasma butyrate, mostly in short-term patients. Giloteaux 2016 (16S, 48 vs 39) found lower diversity and fewer Firmicutes. Nagy-Szakal 2017 (50 vs 50) found lower Faecalibacterium in the IBS subgroup. In the NIH Walitt cohort (15 vs 17), a pre-specified 14-species signature built on other cohorts separated patients with AUC 0.87; all butyrate producers combined were −0.87 SD (Solving ME/CFS re-analysis). The NIH authors' own per-species tests did not pass FDR (min q 0.088).

Evidence

Study	n	Design	Result
Guo 2023, Cell Host Microbe data	106 vs 91	case-control, shotgun metagenomics + SCFA metabolomics	Reduced F. prausnitzii, E. rectale and butyrate synthesis capacity
Xiong 2023, Cell Host Microbe data	154 vs 79	case-control, metagenomics + plasma metabolomics	Reduced butyrate biosynthesis; dysbiosis strongest in short-term illness
Giloteaux 2016, Microbiome	48 vs 39	case-control, 16S	Lower diversity, fewer Firmicutes
Nagy-Szakal 2017, Microbiome	50 vs 50	case-control, metagenomics	Dysbiosis, with distinct taxa depending on IBS status
Solving ME/CFS re-analysis of Walitt 2024 NIH stool	15 vs 17	pre-specified signature test	AUC 0.87; butyrate producers −0.87 SD (p=0.015)

Against or not replicated

Study	n	Design	Result
Walitt 2024, Nat Commun (authors' own analysis) nih-walitt-2024 mwb-st003178	15 vs 17	case-control, per-species tests	No species significant after FDR
Solving ME/CFS re-analysis (3-cohort species meta) data	NIH, Raijmakers, Columbia	random-effects meta-analysis	Effect sizes agree NIH vs Columbia (rho 0.57) but not NIH vs Netherlands (rho 0.07)

Open questions

Is the change a cause, or a consequence of diet, inactivity, medication or IBS?
Does it appear before or at onset in prospective post-infection cohorts?

Bears on: Gut dysbiosis · Test: Private shotgun stool metagenomics only; no validated clinical test or reference range

Ruminococcus gnavus, Flavonifractor plautii, Ruthenibacterium lactatiformans and Enterocloster species are increased in ME/CFS, consistently across BioMapAI, Guo 2023 and NIH cohorts and in long COVID. Several studies

Alongside the loss of 'good' species, a group of bacteria often linked to gut inflammation is overgrown.

In BioMapAI/JAX data (143 vs 88 people), R. lactatiformans (β +0.69), R. gnavus (+0.68) and F. plautii (+0.60) were raised (all q<1e-3). Guo 2023 agreed in direction for 14/17 shared species (binomial p=0.013; Solving ME/CFS re-analysis), but Guo listed only nominally significant species. In the NIH cohort, R. gnavus was +0.80 SD (p=0.014), E. bolteae +0.91 and C. symbiosum +0.77. Long COVID shows the same direction: Liu 2022 (106 patients followed 6 months) found higher R. gnavus and lower F. prausnitzii in post-acute COVID syndrome.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis of BioMapAI (Xiong 2025) data	143 vs 88	species association, adjusted	R. gnavus, F. plautii, R. lactatiformans raised
Liu 2022, Gut	106 COVID patients, 68 controls	prospective cohort, metagenomics	PACS linked to higher R. gnavus, lower F. prausnitzii

Open questions

Does R. gnavus produce inflammatory polysaccharides or interfere with tryptophan metabolism in these patients?
Is the bloom driven by antibiotics, diet or low gut transit?

Bears on: Gut dysbiosis Innate immune hyperactivation · Test: Private stool metagenomics only

Gut dysbiosis is strongest in the first years of illness and largely resolves in patients ill for more than 10 years, while their metabolic and clinical abnormalities persist. Early

Gut differences are clearest early in the illness and fade later, even though patients stay ill. This argues against gut bacteria alone keeping the illness going.

Xiong 2023: short-term (<4 years, n=75) patients showed significant microbial dysbiosis, while long-term (>10 years, n=79) patients had largely resolved dysbiosis but more severe metabolic and clinical aberrations. This comes from one cross-sectional comparison of two cohorts, not follow-up of the same people. The NIH cohort, which is mostly under 5 years ill, still shows the signature (AUC 0.87).

Evidence

Study	n	Design	Result
Xiong 2023, Cell Host Microbe data	75 short, 79 long vs 79	cross-sectional two-cohort comparison	Dysbiosis mainly in short-term illness

Open questions

Do the same patients' microbiomes normalise over time?
Is early dysbiosis a consequence of the triggering infection or antibiotics?

Bears on: Gut dysbiosis · Test: Research only

Microbial metabolites

Plasma indole-3-propionate (IPA), a gut-bacteria product of tryptophan, is lower in ME/CFS in 4 of 4 plasma cohorts (pooled −0.33 SD), and the deficit is fully explained by stool composition, mainly the R. gnavus-type overgrowth. Replicated

A protective molecule that only gut bacteria make is slightly low in patients' blood. The overgrown bacteria seem to crowd out the microbes that make it.

Pooled across Germain 2020, Germain 2022, Che 2022 and BioMapAI: −0.33 SD (95% CI −0.49 to −0.16), p=0.0001. As a test it is weak (AUC 0.57–0.60). In 231 people with paired stool and plasma, a stool model predicted plasma IPA (CV Spearman 0.44), and adjusting for it removed the patient-control gap (−0.36 → −0.06). R. gnavus, F. plautii and R. lactatiformans were each inversely related to IPA. The IPA-producing C. sporogenes guild was not reduced. Fruit intake tracks IPA. No severity association survived FDR (all Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (datasets 02, deep2 microbiome_function) data	4 plasma cohorts; 231 paired stool-plasma	meta-analysis + cross-validated prediction	IPA −0.33 SD; gap explained by stool composition
Xiong 2025 (BioMapAI), Nat Med data	249	longitudinal multi-omics	Altered microbiome-tryptophan-metabolite links

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (Mendelian randomisation) data	DecodeME 15,579; MVP	two-sample MR	Genetically higher IPA does not lower ME/CFS risk (OR 1.19, 0.93–1.53)

Open questions

Does raising IPA (diet, fibre, polyphenols) change any symptom?
Does low IPA replicate with paired stool in a second cohort?

Bears on: Gut dysbiosis IDO metabolic trap (Phair, Davis) · Test: Research metabolomics only

The replicated deficit of aromatic lactates (indolelactate, phenyllactate, 4-hydroxyphenyllactate) in ME/CFS blood is not predicted by stool bacteria or bacterial genes, so it is probably host-side or due to absorption, not dysbiosis. Early

Another set of low blood molecules looks like gut products, but the gut bacteria data do not explain them, so probiotics aimed at them are unlikely to help.

Aromatic-lactate score −0.55 SD (p 6.5e-6) in BioMapAI, and −0.53 SD across 5 cohorts. Stool species and gene models predict it at zero (CV Spearman 0.01). 0/4,429 KEGG genes and 0/1,696 taxa associate with it, and Bifidobacterium plus Lactobacillus abundance is uncorrelated (rho −0.01). Adjusting for diet, medications, IBS and kidney function makes the deficit larger, not smaller. In Esteban 2026, stool aromatic metabolites were higher in patients, which hints at reduced absorption (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 microbiome_function) data	231 paired; 5 plasma cohorts	cross-validated prediction, hypothesis-free scans	No gut predictor of the lactate deficit
Esteban 2026, MicrobiologyOpen	19 vs 23	case-control, targeted stool metabolomics + AHR assay	Nine tryptophan metabolites elevated in stool; AHR activation not different

Open questions

Is it host transamination/LDH handling, small-intestine microbes (not sampled by stool) or uptake?
D- vs L-phenyllactate would distinguish bacterial from human origin.

Bears on: Gut dysbiosis Mitochondrial / energy metabolism impairment · Test: Research only

Dysbiosis and symptoms

Links between specific gut bacteria and symptom severity are inconsistent: F. prausnitzii tracked fatigue in one cohort, but nothing survived correction for multiple testing in a 143-patient re-analysis. Contested

It is not yet clear that people with more abnormal gut bacteria are sicker.

Guo 2023 found F. prausnitzii abundance inversely associated with fatigue severity. Nagy-Szakal 2017 reported correlations between distinct taxa and pain, fatigue and motivation within subgroups. In BioMapAI (143 patients), no taxon or metabolite module associated with severity, orthostatic intolerance or duration at FDR. The best nominal result was lower IPA with worse GI symptoms (p=0.005, q=0.41) (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Guo 2023, Cell Host Microbe data	106 vs 91	case-control	F. prausnitzii inversely associated with fatigue

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis of BioMapAI data	143 patients	within-patient association	No severity, OI or duration association at FDR

Open questions

Do within-person changes in microbiome track PEM episodes?

Bears on: Gut dysbiosis · Test: Research only

IBS and GI symptoms

IBS is about seven times more common in ME/CFS (pooled OR 7.20; prevalence about 37%), and IBS status drives much of the measured microbiome difference. Replicated

Many people with ME/CFS also have irritable bowel syndrome. Studies need to separate the two, because IBS changes gut bacteria by itself.

Rajabi 2026 meta-analysis: 22 studies, 18 pooled. OR 7.20 (2.77–18.76), I² 72%; pooled IBS prevalence 37% (23–54%), varying by ME/CFS criteria. Brown 2026 (116 vs 80): 53% vs 8% prior IBS diagnosis, with GI symptom burden tracking fatigue, cognitive, pain and flu-like symptoms. In Nagy-Szakal 2017, IBS co-morbidity was the strongest factor separating microbiome networks, and the top marker taxa differed between IBS and non-IBS subgroups. Genetically, ME/CFS correlates with IBS (rg 0.57) and its heritability is enriched in enteric neurons (Lee 2026 medRxiv; from Solving ME/CFS genetics summary).

Evidence

Study	n	Design	Result
Rajabi 2026, Front Med	18 studies pooled	systematic review and meta-analysis	OR 7.20 for IBS; prevalence 37%
Brown 2026, J Transl Med	116 vs 80	case-control	IBS 53% vs 8%; GI burden correlates with CRP and core symptoms
Nagy-Szakal 2017, Microbiome	50 vs 50	case-control	IBS status strongest driver of microbiome separation

Open questions

Is ME/CFS dysbiosis independent of IBS in properly stratified analyses?
Are enteric-neuron genetics a shared gut-brain mechanism?

Bears on: Gut dysbiosis Genetic susceptibility acting through neurons Autonomic dysfunction and small-fibre neuropathy · Test: Clinical: Rome IV IBS criteria; faecal calprotectin to exclude IBD

Gut permeability and translocation

Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to enterobacterial LPS) are raised in several small ME/CFS studies, but standardised replication is lacking. Contested

Some studies suggest bits of gut bacteria leak into the blood ('leaky gut') and trigger inflammation. The evidence is small, uses different tests, and comes mostly from a few labs.

Giloteaux 2016 (48 vs 39) found LPS, LBP and sCD14 elevated and correlated with each other. I-FABP (gut-cell damage) was also measured. Maes 2007 reported higher serum IgA and IgM against LPS of gram-negative enterobacteria, correlated with severity and IBS symptoms. De Meirleir 2018 (140 vs 140, retrospective) found sCD14 among the most discriminating serum markers. In the Solving ME/CFS tally, 5 groups report raised LPS/sCD14 and 2 do not. Brown 2026 found higher IgM to LPS only in the subgroup with more flu-like symptoms. Endotoxin assays are contamination-prone, sCD14 is a general monocyte activation marker, and no study has used a direct permeability test (lactulose/mannitol). Maes' 'leaky gut diet' studies were uncontrolled.

Evidence

Study	n	Design	Result
Giloteaux 2016, Microbiome	48 vs 39	case-control	Raised LPS, LBP, sCD14
Maes 2007, J Affect Disord	CFS, partial CFS and controls	case-control	Raised IgA/IgM to enterobacterial LPS
De Meirleir 2018, J Transl Med	140 vs 140	retrospective case-control	sCD14 among best discriminating markers

Against or not replicated

Study	n	Design	Result
Maes & Leunis 2008, Neuro Endocrinol Lett	41	uncontrolled before-after	Claimed improvement with 'leaky gut' supplements; no control group
Seton 2023, medRxiv (Quadram)	5 severe patients vs 5 household controls	feasibility case-control (IgG-Seq)	Reduced, not increased, serum IgG reactivity to stool microbes

Open questions

Does a direct permeability test (sugar absorption, zonulin is unreliable) show a difference?
Do translocation markers rise after exertion (PEM)?

Bears on: Gut dysbiosis Innate immune hyperactivation · Test: Research only (LPS, LBP, sCD14); commercial 'leaky gut' panels are not validated

Virome and mucosal immunity (Quadram Institute)

Quadram Institute studies of severe ME/CFS with household-matched controls report altered gut phage communities and reduced serum IgG reactivity to gut microbes, in very small samples. Early

A Norwich team is studying the viruses that infect gut bacteria, and the immune response to gut microbes, comparing patients with people in the same household. Their samples are tiny so far.

Hsieh 2023 combined virus-like-particle and whole-metagenome sequencing in a small severe-ME/CFS cohort. It defined a core of tailed Caudoviricetes phages and predicted ME/CFS-specific virus-host links involving Anaerotruncus. Seton 2023 (5 pairs with same-household controls) found reduced serum IgG capacity and reactivity to stool microbes. In a twin pair discordant for ME/CFS, Giloteaux 2016 reported lower diversity and expanded Caudovirales phages in the ill twin. The virome of the Norwegian Comeback FMT trial (Quadram/UEA thesis 2024) and a RESTORE-ME trial virome analysis are in progress, and no results are published.

Evidence

Study	n	Design	Result
Hsieh 2023, Int J Mol Sci	small severe cohort + healthy	virome + metagenome case-control	Core phage virome; predicted disease-associated virus-host interactions
Seton 2023, medRxiv	5 vs 5 household	feasibility case-control	Reduced serum IgG reactivity to gut microbes
Giloteaux 2016, Am J Case Rep	1 MZ twin pair	case report	Ill twin: lower diversity, Caudovirales expansion

Open questions

Do phage shifts replicate in a powered cohort?
Does reduced anti-microbial IgG reflect immune exhaustion or sampling noise?

Bears on: Gut dysbiosis Immune (T-cell) exhaustion Viral persistence · Test: Research only

Treatment: faecal transplant

The only published randomised faecal microbiota transplant trial in CFS (n=11) found no benefit, and a larger Norwegian RCT (n=80) has not yet reported. Early

Transferring stool from a healthy donor has been tried once in a proper small trial and did not help. A bigger trial is under way.

Salonen 2023 (Finland, donor FMT n=5 vs autologous n=6 by colonoscopy) found no change in VAS, MFIS, 15D or EQ-5D at 1 or 6 months, and no adverse events. It was underpowered. Kenyon 2019 reported improvement in a retrospective non-randomised series (from memory). The Comeback Study (NCT03691987, Harstad; 80 participants randomised 1:1:2 to two donors or autologous placebo, primary outcome FSS at 3 months) published its protocol in 2024. As of October 2026 I found no results.

Evidence

Study	n	Design	Result
Salonen 2023, J Transl Med	11	RCT (donor vs autologous FMT)	No improvement in fatigue or HRQOL
Skjevling 2024, BMJ Open (Comeback protocol)	80 planned	RCT protocol	Results pending

Against or not replicated

Study	n	Design	Result
Kenyon 2019, Human Microbiome Journal unverified	42 (21 FMT)	retrospective, non-randomised	Reported improvement with FMT (from memory)

Open questions

Will the Comeback RCT show benefit?
Should FMT target patients with the butyrate-loss / R. gnavus signature?

Bears on: Gut dysbiosis · Test: Not applicable (treatment; trials only)

Treatment: probiotics and diet

Probiotic trials in CFS are small; one 39-person RCT of Lactobacillus casei Shirota reduced anxiety but did not target fatigue. Early

No probiotic has been shown to improve ME/CFS itself.

Rao 2009 (n=39, 24 billion CFU daily for 2 months) raised stool Lactobacillus and Bifidobacteria and reduced Beck anxiety scores (p=0.01). Fatigue was not the primary outcome. Venturini 2019 was an uncontrolled pilot. Re-analysis argues against Bifidobacterium/Lactobacillus products for the indolelactate deficit, because their abundance does not track it. Fibre and polyphenols have a better mechanistic case for IPA, but there are no trials (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Rao 2009, Gut Pathog	39	RCT	Reduced anxiety; stool lactobacilli up
Venturini 2019, Oxid Med Cell Longev	small	uncontrolled pilot	Changes in immune/oxidative markers

Open questions

Would a butyrate-producer consortium or fibre intervention shift symptoms in signature-positive patients?

Bears on: Gut dysbiosis · Test: Not applicable

Genetics of the gut-brain axis

ME/CFS genetic risk overlaps with IBS and is enriched in enteric neurons, but genetically predicted gut-derived metabolites (IPA, indolelactate, kynurenine, bile acids) do not change ME/CFS risk. Early

The genes involved overlap with those for irritable bowel syndrome and gut nerves. But inheriting higher or lower levels of gut metabolites does not change ME/CFS risk, so those metabolites are probably effects, not causes.

Solving ME/CFS MR (DecodeME discovery, MVP replication): IPA OR 1.19 (0.93–1.53), indolelactate 0.99, kynurenine 1.01, glycocholate 1.02, all null. Lee 2026 (medRxiv, from Solving ME/CFS genetics summary): rg with IBS 0.57 and heritability enrichment in enteric neurons (FDR 0.004). Wielscher 2026 (preprint): 'barrier-microbiome' traits show among the strongest genetic overlap, with IBS contributing most pleiotropy. OLFM4, a DecodeME locus, is highly expressed in ileal stem cells and neutrophils.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 genetics_mr) data	15,579 + MVP	two-sample MR	All gut-metabolite exposures null
Wielscher 2026, Research Square preprint	DecodeME 15,579	genetic correlation + MR	Barrier-microbiome traits and IBS share pleiotropy with ME/CFS

Open questions

Is the shared IBS genetics neuronal (enteric nervous system) rather than microbial?

Bears on: Gut dysbiosis Genetic susceptibility acting through neurons · Test: Research only

Long COVID

In long COVID, the gut microbiome at hospital admission predicted persistent symptoms, and at 6 months patients had more R. gnavus and fewer butyrate producers, mirroring ME/CFS. Several studies

Long COVID shows a gut pattern similar to ME/CFS, and in one study the gut profile early in COVID predicted who stayed ill.

Liu 2022 (Hong Kong, 106 patients, 258 serial stool samples; 68 controls) reported that admission microbiota composition was associated with PACS. Patients who recovered had normalised microbiomes, while PACS showed higher R. gnavus and B. vulgatus and lower F. prausnitzii. Butyrate producers (B. pseudocatenulatum, F. prausnitzii) showed the largest inverse correlation with PACS. Antibiotic use and hospital care in severe COVID are confounders.

Evidence

Study	n	Design	Result
Liu 2022, Gut	106 vs 68	prospective cohort	Admission microbiome associated with PACS; R. gnavus up, F. prausnitzii down

Open questions

Does pre-infection microbiome predict long COVID in community (non-hospitalised) cohorts?

Bears on: Gut dysbiosis Viral persistence · Test: Research only

Multi-omics classifiers

Microbiome-based machine-learning classifiers distinguish ME/CFS from controls in held-out and external cohorts (BioMapAI, Guo 2023), but they are research tools with modest real-world specificity. Several studies

Computer models can tell patient and control stool apart better than chance, but this is not a diagnostic test yet.

BioMapAI (Xiong 2025, Nat Med; 249 participants over 4 years) integrated metagenomics, plasma metabolomics, immune profiling and labs to predict severity. It classified held-out and external cohorts and linked SCFA, BCAA, tryptophan and benzoate metabolism to MAIT/γδT inflammation. Guo 2023's classifier generalised across US sites. Classifiers have not been tested against disease controls (IBS, depression, MS, fibromyalgia), so specificity is unknown. Within-cohort cross-validation in the plasma metabolome transferred poorly (AUC 0.56; Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Xiong 2025, Nat Med data	249	longitudinal multi-omics, deep learning	Classifies ME/CFS in held-out and external cohorts
Guo 2023, Cell Host Microbe data	106 vs 91 + validation	case-control classifier	Robust to geography; generalised to validation cohort

Open questions

How do classifiers perform against IBS and other fatigue conditions?

Bears on: Gut dysbiosis · Test: Research only

Key reviews

König 2022, Front Immunol: Broad review of microbiome hypotheses (dysbiosis, permeability, SCFA, D-lactate, tryptophan) and interventions.
Nikolova 2025, Int J Mol Sci: Recent overview of gut microbiome mechanisms in ME/CFS.
Rajabi 2026, Front Med: First meta-analysis quantifying IBS overlap.
Perera 2025, Int J Mol Sci (Quadram): Argues that mucosal viruses are an under-studied piece.

Genetics

Heritability and family studies

ME/CFS aggregates in families: relatives of cases have about 2-3x the risk, MZ twins are more concordant than DZ twins, and familiality in a 153,803-person Dutch cohort was 42%. Replicated

Having a close relative with ME/CFS raises your own risk, which means genes (and perhaps shared environment) play a part.

In the Utah Population Database (Albright 2011), relative risk was 2.70 (1.56–4.66) for first-degree relatives, 2.34 for second-degree and 1.93 for third-degree, with excess distant relatedness (p=0.010). Buchwald 2001 (146 female twin pairs) found 55% MZ vs 19% DZ concordance for idiopathic chronic fatigue and estimated heritability up to 51%. Its confidence intervals are wide, and it measured chronic fatigue, not strict ME/CFS. Bos 2025 (Lifelines, 153,803 people, 90,397 with a first-degree relative) found recurrence risk ratio 1.45–2.23 within disorders and familiality 42% (33–50) for ME/CFS. Familial correlations were 0.97 with fibromyalgia and 0.83 with depression. Familiality includes shared household environment.

Evidence

Study	n	Design	Result
Albright 2011, BMC Neurology	Utah genealogy-linked health records	familial aggregation	First-degree RR 2.70; third-degree RR 1.93
Buchwald 2001, Psychosom Med	146 female twin pairs	classic twin study	MZ 55% vs DZ 19% concordance; h² up to 51%
Bos 2025, Psychol Med	153,803	three-generation cohort familiality	Familiality 42% for ME/CFS; rf 0.97 with fibromyalgia, 0.83 with MDD

Open questions

How much of familiality is genetic vs shared environment or infection exposure?
Does the familial risk differ between infection-onset and gradual-onset cases?

Bears on: Genetic susceptibility acting through neurons · Test: Family history only

Genome-wide association (GWAS)

DecodeME (15,579 PEM-defined cases vs 259,909 controls) identified 8 genome-wide significant loci, each with a small effect (OR about 1.07-1.10), and SNP heritability of about 8-10%. Several studies

The largest DNA study of ME/CFS found eight regions of the genome linked to the illness. Each nudges risk only a tiny amount.

Lead genes as assigned: RABGAP1L, BTN2A2/TRIM38, FBXL4, SUDS3, OLFM4, CCPG1, CA10, ARFGEF2/CSE1L/ZNFX1 (the chr20 signal p=1e-11). Four loci (RABGAP1L, FBXL4, OLFM4, CA10) were nominally associated (p<0.05) with PEM-plus-fatigue cases in UK Biobank and Lifelines, but none passed multiple-testing correction in external sets. There was no sex-biased locus, and ARFGEF2 and CA10 replicated in males. The CA10 signal colocalises with multisite chronic pain. As of October 2026 this is still a medRxiv preprint; I found no peer-reviewed version. Gene assignments are weak at some loci: the 6q16 lead is about 890 kb from FBXL4, and the chr1 lead sits inside DARS2 (Solving ME/CFS re-analysis). An absolute-risk calculation shows one risk copy moves baseline risk from about 0.9% to about 0.97–0.99%.

Evidence

Study	n	Design	Result
DecodeME Genetics Delivery Team 2025, medRxiv data	15,579 vs 259,909	GWAS (European ancestry)	8 loci; no colocalisation with depression/anxiety

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (MVP-only and FinnGen replication) data	3,891 MVP cases; 283 FinnGen	replication of lead SNPs	Lead loci 5/8 same direction in MVP (p=0.13), 4/8 in FinnGen; BTN2A2, OLFM4, CA10 wrong way in MVP

Open questions

Which genes are causal at each locus (fine-mapping, functional work)?
Will loci replicate in a cohort with PEM-based phenotyping?

Bears on: Genetic susceptibility acting through neurons Viral persistence Innate immune hyperactivation · Test: Not useful individually (no clinical genetic test)

A 2026 UK Biobank GWAS with stricter, multi-source case definitions (1,268 cases) reports 7 loci that replicate in disjoint UK Biobank or All of Us cohorts (near BICD1, GRIN2A, CSMD1, RORA, CLYBL), though none replicate in all three. Early

A new study using carefully confirmed cases in two biobanks found seven genetic regions that repeated in a second group. It is a preprint and the groups are small.

Slaughter 2026 (medRxiv) used TarGene with electronic health records plus survey data to confirm case and control status. 176 variants passed FDR<5% in discovery, and replication cohorts had 319 (UKB) and 371 (All of Us) cases. Fine-mapping at one locus colocalised with reduced CLYBL expression in putamen, but the CLYBL stop-gain variant was not associated. No gene-by-sex or gene-by-deprivation interactions survived correction. FDR-based discovery in a small sample and no overlap with DecodeME's 8 loci call for caution. GRIN2A (an NMDA receptor subunit) fits the glutamatergic signal.

Evidence

Study	n	Design	Result
Slaughter 2026, medRxiv	1,268 discovery; 319 + 371 replication	GWAS with two replications	7 replicated loci, none in all three cohorts

Open questions

Do these loci show any signal in DecodeME summary statistics?

Bears on: Genetic susceptibility acting through neurons · Test: Research only

Replication in biobanks

The genome-wide DecodeME signal replicates in independent cohorts: genetic correlation 0.65-0.73 with the Million Veteran Program CFS GWAS and 0.80-0.90 with UK Biobank CFS, plus concordant directions for sub-threshold SNPs in FinnGen. Replicated

Even though single hits do not yet repeat everywhere, the overall genetic fingerprint of ME/CFS shows up again in other large populations, including mostly male US veterans.

Solving ME/CFS re-analysis derived MVP-only statistics (3,891 independent cases) from the public DecodeME+MVP meta. rg 0.73 ± 0.16. DecodeME SNPs at p<1e-5 were concordant 42/63 in MVP (p=0.006) and 41/61 in FinnGen postviral fatigue (p=0.005). Pan-UKB self-report CFS gave rg 0.90, but it shares controls. An independent GitHub re-analysis (trafalmadorian97) reported rg 0.65 with MVP and 0.80 with Neale UKB. The MVP cohort is mostly male, so polygenic risk applies across sexes.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME 15,579; MVP 3,891; FinnGen 283	LDSC rg and sign concordance	rg 0.73; concordance p=0.006 (MVP), 0.005 (FinnGen)
Maccallini 2026, Research Square preprint	19,470 vs 699,111	DecodeME + MVP meta-analysis	5 loci incl. new TSHZ3 region; brain and pituitary enrichment

Open questions

How much does loose biobank phenotyping (billing codes, self-report) dilute single-locus replication?

Bears on: Genetic susceptibility acting through neurons · Test: Research only

Tissue and cell-type enrichment

Genome-wide ME/CFS heritability is enriched in brain tissues and neurons (including glutamatergic synapse genes and subcortical/cerebellar neuron types), with no enrichment in blood immune cells. Several studies

The genes involved are mainly active in nerve cells in the brain, not in blood immune cells. This suggests the brain is central to inherited risk.

DecodeME: MAGMA gene-property enrichment in all 13 GTEx brain tissues and no others. Independent S-LDSC re-analysis (trafalmadorian97 GitHub): CNS tissues and chromatin only. ImmGen and blood-cell ATAC-seq were null, and medium spiny neuron clusters were implicated. Maccallini 2026 (DecodeME+MVP): brain regions and pituitary, glutamatergic synapse gene sets, with 7/12 gene sets replicating in rare-variant genes from HEAL2. Lee 2026 (medRxiv): inhibitory neurons (p=1.2e-7) and enteric neurons. Immune abnormalities may still be downstream of illness. A locus-level permutation with only 8 loci could not place any tissue (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
DecodeME 2025, medRxiv data	15,579 cases	GWAS + MAGMA	All 13 brain tissues enriched; no others
Maccallini 2026, Research Square preprint	19,470 cases	meta-GWAS enrichment + rare-variant replication	Glutamatergic synapse; subcortical and cerebellar neurons; pituitary

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (locus tissue permutation)	8 loci	permutation	No tissue survives FDR at locus level (expected with 8 loci)

Open questions

Which brain circuits (basal ganglia, cerebellum, claustrum) matter, and do imaging studies of genetic-risk extremes show differences?

Bears on: Genetic susceptibility acting through neurons Neuroinflammation (microglial activation) · Test: Research only

Shared genetics with other conditions

ME/CFS genetics overlaps most with chronic pain, fibromyalgia, IBS and migraine, and substantially with depression, but none of the 8 DecodeME loci share a causal variant with depression. Several studies

Genes that raise ME/CFS risk also raise risk of chronic pain and gut-brain conditions. This is shared biology, not evidence that ME/CFS is depression.

The CA10 locus colocalises with multisite chronic pain (PPH4 92.8%). Re-analyses report the highest rg with fibromyalgia, then multisite pain, as well as syncope, migraine and asthma. Lee 2026 medRxiv: rg with major depression 0.60 and with IBS 0.57. MR in Solving ME/CFS re-analysis: depression OR 1.82 (MVP 1.36) and multisite pain OR 3.3, but reverse MR is also positive (ME/CFS → depression OR 1.15) and heterogeneity is high, which indicates shared liability rather than one-way causation. A UK Biobank MR (Song 2025) found no causal effect of depression on ME/CFS.

Evidence

Study	n	Design	Result
DecodeME 2025, medRxiv data	15,579 cases	GWAS colocalisation	CA10 shares causal variant with chronic pain; no colocalisation with depression/anxiety
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME + MVP	bidirectional MR	Depression, pain, insomnia associated both ways; shared liability

Against or not replicated

Study	n	Design	Result
Song 2025, Sci Rep	UK Biobank CFS GWAS	two-sample MR	No causal effect of MDD on ME/CFS

Open questions

Which part of ME/CFS genetics is specific after subtracting pain genetics? A re-analysis suggests the chr6 BTN region and one striatal neuron cluster.

Bears on: Genetic susceptibility acting through neurons Persistent sickness behaviour / vagal signalling · Test: Research only

Infection response genes

Three DecodeME loci lie near infection-response genes (OLFM4, RABGAP1L, BTN2A2/TRIM38), and the OLFM4 effect is larger in infection-onset cases, but infection- and non-infection-onset cases are genetically highly correlated. Early

Some risk genes help the body respond to infections, which fits the common post-viral onset. But people whose illness began without an obvious infection carry similar overall genetic risk.

Infection-onset GWAS (9,738 cases): 3 loci (6p22.2, 20q13.13, OLFM4). The OLFM4 beta was 0.095±0.017 with infection onset vs 0.034±0.022 without. The no-infection GWAS (5,841) found nothing, which is a power issue. Independent re-analysis finds high rg between onset subgroups. OLFM4 is a neutrophil and ileal stem-cell protein. ZNFX1 (a dsRNA sensor) and CSE1L (needed for STAT1 import) sit at the chr20 locus. Genetically higher EBV DNA load gave OR 1.13 (ns) in DecodeME and 1.59 (p=0.012) in MVP (Solving ME/CFS re-analysis). MHC-based negative controls (VZV, HHV-6B) also aligned, so the result is not EBV-specific. The long COVID FOXP4 variant is not associated with ME/CFS (p=0.59).

Evidence

Study	n	Design	Result
DecodeME 2025, medRxiv data	9,738 infection-onset cases	stratified GWAS	OLFM4 effect larger with infection onset
Solving ME/CFS re-analysis (EBV load MR) data	DecodeME; MVP	MR with Kamitaki 2026 instruments	OR 1.13 (ns) / 1.59 (p=0.012)

Against or not replicated

Study	n	Design	Result
Lammi 2025, Nat Genet (long COVID GWAS)	6,450 cases	GWAS	FOXP4 lead variant; no association with ME/CFS in DecodeME

Open questions

Do OLFM4 risk-genotype carriers show different neutrophil or gut responses (recall-by-genotype)?
Is self-reported onset type too noisy to stratify on?

Bears on: Viral persistence Herpesvirus reactivation (EBV, HHV-6) Innate immune hyperactivation · Test: Research only

HLA

Reported HLA associations are inconsistent: Lande 2020's HLA-C*07:04 and DQB1*03:03 risk alleles are not among DecodeME's hits, and DecodeME's protective HLA-DQA1*05:01 signal is barely visible at SNP level. Contested

HLA genes shape immune responses and are linked to autoimmune diseases. Claims of specific HLA types in ME/CFS have not held up consistently.

Lande 2020 (426 Norwegian CCC cases vs 4,511 controls) reported HLA-C*07:04 (OR 2.1) and DQB1*03:03 (OR 1.5); carriers of either were 19.2% vs 12.2%. DecodeME reported DQA1*05:01 as protective (p=1.4e-10; 21.7% vs 23.2%), with no DRB1*03:01 or DQB1*02:01 signal. In Solving ME/CFS re-analysis, the best DQ2.5 tag SNP was OR 0.96 (p=0.025). That gap suggests the imputed-allele p-value is inflated, possibly because HLA was imputed separately for cases and controls, a problem the authors flagged. MHC z-scores correlate only 0.21 between DecodeME and MVP. A purely in-silico 2025 study modelled Lande alleles binding herpes peptides, with no patient data.

Evidence

Study	n	Design	Result
Lande 2020, Sci Rep	426 vs 4,511	HLA NGS case-control	C07:04 OR 2.1; DQB103:03 OR 1.5
DecodeME 2025, medRxiv data	15,579 vs 259,909	HLA imputation	DQA1*05:01 protective p=1.4e-10

Against or not replicated

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 HLA) data	DecodeME + MVP	SNP-level MHC re-analysis	Best DQ2.5 tag OR 0.96; DecodeME-MVP MHC z correlation 0.21

Open questions

Will joint HLA imputation of cases and controls confirm any allele?

Bears on: Autoimmunity (functional autoantibodies) Herpesvirus reactivation (EBV, HHV-6) · Test: HLA typing is available but not useful for ME/CFS

NK-cell receptor genes

Specific inhibitory KIR alleles (e.g. KIR3DL1*020, KIR3DL2*009) were more or less common in 418 Norwegian patients than 473 controls, though KIR gene content did not differ. Early

Variants in genes for natural killer cell 'brake' receptors may differ slightly in ME/CFS. This has not been checked in another cohort.

Ramadan 2025 used high-resolution KIR sequencing with HLA ligand annotation. Patient-enriched alleles were KIR3DL3*002 (OR 1.43), KIR3DL1*020 (OR 2.20) and KIR3DL2*009 (OR 1.56); reduced alleles were KIR3DL3*013 (OR 0.60) and KIR3DL2*010 (OR 0.46, p=0.0005). There was no association with gene content or copy number. Nominal p-values without a replication cohort; KIR is poorly captured by GWAS arrays, so DecodeME cannot test it.

Evidence

Study	n	Design	Result
Ramadan 2025, Brain Behav Immun	418 vs 473	case-control, KIR allele sequencing	Several inhibitory KIR alleles associated (nominal p 0.0005–0.01)

Open questions

Replication in UK or US cohorts; link to NK cytotoxicity measures.

Bears on: Immune (T-cell) exhaustion Herpesvirus reactivation (EBV, HHV-6) · Test: Research only

Earlier GWAS and candidate genes

Before DecodeME, small GWAS and candidate-gene studies (GRIK2, NPAS2, TPPP, TRA locus and many others) produced no association that replicated in UK Biobank or later cohorts. Refuted

Older genetic studies were too small. Their 'hits' did not hold up when tested in larger groups.

Smith 2011 (40 vs 40, 116k SNPs plus expression) flagged GRIK2 and NPAS2. Hajdarevic 2022 (Norwegian 427 discovery, Danish 460, UKB 2,105) found no genome-wide significant loci; the best signal was TPPP (p=8.5e-7) and the top SNP did not replicate. Dibble 2020 concluded that most candidate gene associations are not replicated by the UK Biobank CFS cohort. Ueland 2022's T-cell receptor alpha signal did not replicate (from Solving ME/CFS genetics summary).

Evidence

Study	n	Design	Result
Hajdarevic 2022, Brain Behav Immun	2,532 cases total	GWAS with replication	No genome-wide significant loci; TPPP not replicated
Dibble 2020, Hum Mol Genet	review incl. UKB	critical review	Most candidate gene associations not replicated
Smith 2011, Neuropsychobiology	40 vs 40	convergent SNP + expression	GRIK2, NPAS2 nominal

Bears on: Genetic susceptibility acting through neurons · Test: No: commercial 'ME/CFS gene panels' are not supported

Mendelian randomisation

Proteome-wide cis-MR finds genetically higher plasma EPHB1 (a pain-sensitisation receptor) and CLEC3B (tetranectin) associated with ME/CFS risk in DecodeME, replicating in the independent MVP cohort. Early

Using genes as a natural experiment, two blood proteins look like they may influence ME/CFS risk. One of them is involved in how the nervous system amplifies pain.

Solving ME/CFS re-analysis of about 1,950 UKB-PPP and 1,690 deCODE plasma proteins (Wald ratio). EPHB1: OR 1.18/SD (p=2e-4), MVP 1.29 (p=0.0016), combined FDR 0.002; its pQTL is in LD with the regional DecodeME peak (r² 0.84). CLEC3B: OR 1.21, MVP 1.44 (p=0.0008), but not in LD with the regional peak. PRDX6 (1q25) was protective (0.75; MVP 0.70, p=0.037), and dermatopontin was nominal on both platforms. Only 4/93 nominal hits replicated, versus about 2.3 expected by chance, so these are leads. There was no formal colocalisation because full pQTL statistics need a login, and there are no usable drugs.

Evidence

Study	n	Design	Result
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME 15,579; MVP 3,891	cis-pQTL MR with replication	EPHB1 and CLEC3B replicate; PRDX6 nominal

Open questions

Full colocalisation with deCODE/UKB-PPP region data; CSF/brain pQTLs for EPHB1.

Bears on: Genetic susceptibility acting through neurons · Test: Research only

MR of DecodeME suggests higher mitochondrial DNA copy number is protective and higher GlycA and mean platelet volume raise risk, while genetically predicted metabolite leads (IPA, kynurenine, ketones, carnitine, bile acids) and vitamin D, BMI and blood pressure are null or unreplicated. Early

Genetic natural experiments hint that inflammation and blood platelets may contribute to risk. Many blood molecules that are low in patients look like effects of being ill, not causes.

Wielscher 2026 (Research Square, not peer reviewed) reported mtDNA copy number as protective and GlycA and MPV as risk-raising. It also found the strongest genetic overlap in barrier-microbiome, neurovascular and energetics traits. Solving ME/CFS MR found indolelactate OR 0.99, kynurenine 1.01 and 3-hydroxybutyrate 0.99, all null. Palmitoylcarnitine 0.88 did not replicate. Vitamin D OR 0.88 (p=0.03) was not replicated and had a pleiotropic Egger intercept. Many published two-sample MR papers on CFS use small UK Biobank self-report GWAS; one was retracted in 2025.

Evidence

Study	n	Design	Result
Wielscher 2026, Research Square preprint	DecodeME 15,579	rg, pleiotropy, MR	mtDNA copy number protective; GlycA, MPV risk
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME + MVP	MR with replication	Metabolite and risk-factor exposures null or unreplicated

Against or not replicated

Study	n	Design	Result
Retraction notice 2025, Food Sci Nutr	-	retracted two-mediation MR study	Example of low-quality MR in CFS literature

Open questions

Replicate GlycA/MPV/mtDNA MR in MVP; test platelet activation in patients.

Bears on: Mitochondrial / energy metabolism impairment Complement activation and thromboinflammation Innate immune hyperactivation · Test: Research only

Rare variants and sequencing

Whole-genome sequencing of 464 cases and 611 controls with a deep-learning rare-variant model (HEAL2) proposes 115 loss-of-function-intolerant risk genes, partly overlapping the common-variant synaptic signal. Early

Looking at rare DNA changes, one study suggests a set of genes, many active in nerves and immune cells. The method is new and the sample is small.

Zhang 2025 (medRxiv) reports that a genetic score from personal rare variants tracked PEM, unrefreshing sleep, brain fog and muscle pain. Patient multi-omics showed reduced expression of the risk genes, including in cytotoxic CD4 T cells. Maccallini 2026 found 7 of 12 common-variant gene sets replicate among HEAL2 genes. Deep-learning gene prioritisation in about 1,000 genomes has high overfitting risk, and there has been no independent sequencing replication. No single rare variant has been shown to cause ME/CFS. A verified, specific rare-variant 'Mendelian ME/CFS' report does not exist to my knowledge.

Evidence

Study	n	Design	Result
Zhang 2025, medRxiv (HEAL2)	464 vs 611	WGS + deep learning	115 risk genes; symptom links

Open questions

Independent WGS replication; burden tests in DecodeME exomes when available.

Bears on: Genetic susceptibility acting through neurons Immune (T-cell) exhaustion · Test: Research only; clinical exome only for red-flag mimics (mitochondrial, HLH, immunodeficiency)

Combinatorial analyses

Proprietary combinatorial SNP analyses (PrecisionLife) report thousands of multi-SNP signatures in UK Biobank and DecodeME, with top vs bottom decile of signature count giving OR 1.64. Contested

A company's method finds combinations of DNA variants linked to ME/CFS, but the method is closed and hard for others to check.

Das 2022 (2,382 UKB cases) mapped 84 SNP combinations to 14 genes, including CLOCK, SLC15A4, INSR and KCNB1. Sardell 2026 (J Transl Med, DecodeME vs UKB controls) reported 22,411 signatures from 7,555 SNPs in 2,311 genes, reproducible across three disjoint DecodeME case sets (p=4e-21 for count), 259 core genes, and 76/180 long-COVID genes overlapping. 'Reproducible' here means within related cohorts sharing UK Biobank controls; the method is proprietary and the gene list is very broad.

Evidence

Study	n	Design	Result
Sardell 2026, J Transl Med	DecodeME cohorts vs UKB controls	combinatorial genetics	OR 1.64 top vs bottom decile of signature count
Das 2022, J Transl Med unverified	2,382 cases	combinatorial genetics (UKB)	14 genes; 5 SNPs replicated in a UKB CFS set

Open questions

Do signatures replicate in an external cohort with different controls (MVP, FinnGen, All of Us)?

Bears on: Genetic susceptibility acting through neurons · Test: Research only

TRPM3 ion channel

TRPM3 ion channel dysfunction in NK cells and TRPM3 SNPs are reported repeatedly by one research group, but the SNP studies are tiny and no TRPM3 association appears in GWAS. Contested

A calcium channel called TRPM3 is reported to work poorly in immune cells of ME/CFS patients. Nearly all of this comes from one team, and large DNA studies do not point to the TRPM3 gene.

Marshall-Gradisnik 2016 genotyped B cells from 11 patients vs 11 controls and reported SNPs/genotypes in TRPM3 and acetylcholine receptor genes. An earlier NK-cell SNP study had about 115 vs 90 (from memory). Functional work: Sasso 2026 used patch-clamp at two sites run by the same consortium (36 vs 42) and found reduced TRPM3-like currents in NK cells. Independent-lab replication and blinding details are lacking, and TRPM3 is not among DecodeME, MVP or Slaughter 2026 loci. Functional changes could be acquired rather than inherited.

Evidence

Study	n	Design	Result
Sasso 2026, Front Med	36 vs 42	case-control patch-clamp, two sites	Reduced TRPM3 function in NK cells
Marshall-Gradisnik 2016, J Int Med Res	11 vs 11	SNP/genotype case-control	TRPM3 and AChR gene SNPs reported

Against or not replicated

Study	n	Design	Result
DecodeME 2025, medRxiv data	15,579 cases	GWAS	No TRPM3 locus among genome-wide hits

Open questions

Will an independent lab, blinded to case status, reproduce reduced TRPM3 currents?
Is TRPM3 gene-level association present in DecodeME MAGMA results?

Bears on: Sodium-calcium overload in muscle (Wirth & Scheibenbogen) Immune (T-cell) exhaustion · Test: Research only

Connective tissue

Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4 in 60% of children with CFS vs 24% of controls; 50% in a Swedish severe cohort; EDS about 100x expected in Finnish registries). Several studies

People with very flexible joints, or hypermobile Ehlers-Danlos syndrome, seem more likely to have ME/CFS or long COVID with PEM. No gene test exists for the hypermobile type.

Barron 2002 (58 incident paediatric CFS vs 58 controls): median Beighton 4 vs 1, OR 3.5 (1.6–7.5). Bragée 2020 (229 severe CCC patients, no controls): 50% hypermobile, with high rates of possible intracranial hypertension and craniocervical findings on MRI. Finnish FinnGen/FinRegistry postviral fatigue shows EDS O/E 106–153, which is crude and inflated by healthcare use (Solving ME/CFS re-analysis). Lubell 2026 (online survey, 1,816 people after COVID) found extreme hypermobility OR 3.15 for long COVID with high post-exertional symptom exacerbation, rising to OR 9.46 with a pre-COVID orthostatic/neurocognitive profile. Hypermobile EDS has no known causal gene, and connective-tissue genes are not among GWAS hits. Collagen breakdown markers are raised in 4/4 Metabolon cohorts (Solving ME/CFS re-analysis).

Evidence

Study	n	Design	Result
Barron 2002, J Pediatr	58 vs 58	matched case-control	Hypermobility 60% vs 24%; OR 3.5
Bragée 2020, Front Neurol	229	retrospective cross-sectional, no controls	50% hypermobile
Lubell 2026, medRxiv	1,816	retrospective online survey	Extreme hypermobility OR 3.15 for long COVID with high PESE

Open questions

Prevalence in population-based ME/CFS cohorts with blinded Beighton scoring?
Does hypermobility define a subgroup with distinct orthostatic or craniocervical features?

Bears on: Connective tissue / hypermobility Brainstem and craniocervical instability Low blood volume and preload failure · Test: Yes: Beighton score and 2017 hEDS criteria (clinical); genetic testing only for other EDS types

Mitochondrial DNA

Mitochondrial DNA haplogroups and common mtDNA SNPs are not associated with ME/CFS status, though some variants were associated with particular symptoms. Early

Inherited types of mitochondrial DNA do not differ between patients and controls.

Billing-Ross 2016 (Cornell cohorts) found no haplogroup or SNP association with disease status but reported variant links to symptom severity. Hanson 2016 is the authors' reply clarifying this. MR suggesting higher mtDNA copy number is protective (Wielscher 2026) concerns copy number, not haplogroup. The chr1 DecodeME lead lies in DARS2, a mitochondrial tRNA synthetase, which is a weak hint.

Evidence

Study	n	Design	Result
Hanson 2016, J Transl Med (reply on Billing-Ross 2016)	Cornell cohorts	mtDNA sequencing case-control	No association with disease status; symptom associations

Open questions

Heteroplasmy and somatic mtDNA changes in muscle have not been studied at scale.

Bears on: Mitochondrial / energy metabolism impairment · Test: Research only

Key reviews

Dibble 2020, Hum Mol Genet: Critical review of pre-DecodeME genetics showing candidate genes do not replicate.
DecodeME 2025, medRxiv preprint: Primary report of the 8 loci, HLA, sex and onset analyses; summary statistics public on OSF.
Maccallini 2026, Research Square preprint: DecodeME+MVP meta-analysis with tissue, cell-type and rare-variant cross-replication.

Treatment trials and natural history

Immune: B-cell depletion

B-cell depletion with rituximab did not improve fatigue or function versus placebo in a 151-person phase III trial (RituxME). Refuted

A drug that removes antibody-making B cells looked promising in small Norwegian studies, but a large placebo-controlled trial found it worked no better than a dummy infusion.

RituxME (Fluge 2019): 77 rituximab vs 74 placebo over 24 months; overall response 26.0% vs 35.1% (difference 9.2 points favouring placebo, 95% CI -5.5 to 23.3, p=0.22); no difference on any secondary endpoint; serious adverse events 26% vs 19%. Six-year follow-up (Rekeland 2024): SF-36 physical function 32.9 to 45.5 on rituximab vs 32.3 to 43.1 on placebo. The earlier positive phase II results (Fluge 2011, 2015) were small and partly open-label; the phase III result is the standard example of how large the placebo/natural-course improvement can be in ME/CFS trials.

Evidence

Study	n	Design	Result
Fluge 2019, Ann Intern Med (RituxME)	77 vs 74	phase III double-blind RCT, 5 Norwegian centres	Response 26.0% rituximab vs 35.1% placebo, p=0.22; no secondary benefit
Rekeland 2024, PLoS One	RituxME 151 + CycloME 40 (75.7% / 94.4% followed)	6-year follow-up of two trials	SF-36 PF at 6 yr 45.5 rituximab vs 43.1 placebo; SF-36 PF >=70 in 27.6% vs 20.4%

Against or not replicated

Study	n	Design	Result
Fluge 2015, PLoS One (KTS-2-2010)	29	open-label phase II with maintenance	18/29 reported clinically significant responses (not controlled)

Open questions

Is there a B-cell-dependent subgroup that a whole-population trial would dilute?
Why did both arms improve by ~10 SF-36 points over 18 months?

Bears on: Autoimmunity (functional autoantibodies) Circulating (serum) factor · Test: Not recommended outside trials

Immune: broad immunosuppression

In an open-label phase II study, 55% of 40 patients responded to six IV cyclophosphamide infusions, and many responders stayed improved at 6 years, but there was no control arm. Early

A chemotherapy drug that suppresses the immune system was followed by improvement in about half of a small Norwegian group, but without a comparison group the effect cannot be separated from natural improvement or placebo.

CycloME (Rekeland 2020): 600-700 mg/m2 every 4 weeks x6; 22/40 (55%) responders; SF-36 PF 35.4 to 54.4 at 18 months and 56.7 at 6 years; 44.1% had SF-36 PF >=70 at 6 years versus 20.4% of RituxME placebo patients (different trial, not randomised). Nausea in ~90%; 11 serious adverse events in 8 patients. Response was higher in carriers of HLA-DQB1*03:03 and/or HLA-C*07:04 (83% vs 43%), an unreplicated post hoc finding. The authors themselves advise against use outside trials because of toxicity (infertility, bladder and cancer risk).

Evidence

Study	n	Design	Result
Rekeland 2020, Front Med	40	open-label phase II	55% responders; 68% of responders still improved at 3-4 years
Rekeland 2024, PLoS One	36 of 40 followed	6-year follow-up	SF-36 PF 56.7 at 6 yr; 17.6% at >=90 (normal range)

Open questions

Would a randomised trial against an active-placebo reproduce the effect?
Does the HLA association replicate (HLA-C*07:04 also appears in other ME/CFS genetic reports)?

Bears on: Autoimmunity (functional autoantibodies) Immune (T-cell) exhaustion · Test: Not recommended outside trials

Immune: plasma-cell depletion

In an open-label pilot, 6 of 10 women with moderate-to-severe ME/CFS improved markedly after the anti-CD38 antibody daratumumab, and low baseline NK-cell count predicted non-response. Early

A drug that removes long-lived antibody-producing plasma cells was followed by large, lasting improvements in six of ten patients. This is the most striking recent signal, but it is ten people with no placebo group.

Fluge 2025: 4-7 subcutaneous injections after a 12-week run-in. Mean SF-36 PF 25.9 to 55.0 at 8-9 months (p=0.002); in the 6 responders 32.2 to 78.3; 5 sustained improvement (SF-36 PF ~88) with step counts above 10,000/day for weeks. Mean steps 3,359 to 5,862. IgG fell 54% in responders vs 40% in non-responders. Low baseline NK-cell counts associated with no response. No serious adverse events. A randomised trial is planned in Norway; Charité is pursuing a related approach with inebilizumab. Given RituxME's ~35% placebo response, n=10 open-label cannot establish efficacy.

Evidence

Study	n	Design	Result
Fluge 2025, Front Med	10	open-label pilot (EudraCT 2022-000281-18)	6/10 marked improvement; SF-36 PF 25.9 to 55.0 (p=0.002)
Rucker, The Sick Times 2026 (conference report)	10	follow-up talk, Berlin 2026	Fluge reported 5 of 6 responders still in remission at 2-3 years

Open questions

Does the response survive randomisation and blinding?
Is NK-cell count a genuine predictor or a chance correlation in n=10?
Which antibodies, if any, are being removed?

Bears on: Autoimmunity (functional autoantibodies) Circulating (serum) factor · Test: Research only (trial planned in Norway)

Immune: antibody removal

Removing antibodies by immunoadsorption was no better than a sham procedure in two controlled trials (IAMPOCO, n=40; IA-PACS-CFS, n=65), although open-label series had reported 50-70% responders. Refuted

Filtering antibodies out of the blood seemed to help in uncontrolled studies. When patients were compared with others who had a fake procedure, both groups improved similarly.

Stein 2025 (open-label, post-COVID ME/CFS with raised beta2-adrenergic receptor antibodies, n=20): 70% 'responders', SF-36 PF +17.75 at 4 weeks; antibody fall did not distinguish responders. IAMPOCO (Mainz, tryptophan adsorbers, patient-blinded sham crossover, n=40): no difference on the Post-COVID Functional Scale (OR 1.17, p=0.771) or fatigue, cognition, grip strength; 24 vs 10 adverse events. IA-PACS-CFS (Charité, 44 IA vs 21 sham, 5 sessions): 'no statistically or clinically significant differences' on Chalder fatigue at day 60 (preliminary, conference). Critics note IA-PACS-CFS did not select for autoantibody-positive patients; IAMPOCO removed GPCR antibodies and still showed no benefit. A 2025 antigen-profiling study found no elevated GPCR autoantibodies in ME/CFS at all.

Evidence

Study	n	Design	Result
Stein 2025, Lancet Reg Health Eur	20	prospective open-label cohort (repeat IA, beta2-AR antibody positive)	70% responders at 4 weeks; no control
Stortz/Kraus et al. 2026, IAMPOCO (Mainz repository; Lancet Reg Health Eur correspondence)	40	randomised patient-blinded sham-controlled crossover	No difference: PCFS OR 1.17, p=0.771; no secondary benefit
IA-PACS-CFS (Charité), presented 2026	44 vs 21	double-blind sham-controlled RCT	Failed primary endpoint (Chalder fatigue, day 60)

Against or not replicated

Study	n	Design	Result
Stein, conference case series 2026	15	uncontrolled, autoantibody-positive	7/15 responded
Wossidlo 2026, Infection	18	prospective exploratory, uncontrolled	Reports changes in autoantibodies and performance after 5 sessions

Open questions

Does an autoantibody-selected subgroup benefit in a sham-controlled design?
Are the measured GPCR antibodies meaningful at all, given negative antigen-profiling data?

Bears on: Autoimmunity (functional autoantibodies) Circulating (serum) factor · Test: Not recommended outside trials; GPCR antibody tests (CellTrend) are commercial but of unproven meaning

Immune: GPCR-antibody neutralisation

The aptamer BC007 (rovunaptabin) failed to beat placebo in the company's 114-person long-COVID trial; a 30-person investigator-led trial reported secondary-endpoint gains. Contested

A drug designed to neutralise antibodies against blood-vessel and heart receptors was tested twice in long COVID. The larger trial found no benefit; a smaller exploratory one reported improvements on fatigue scales.

reCOVer (Erlangen, Hohberger 2025): n=30 with GPCR autoantibodies, primary endpoint safety (9 vs 5 adverse events, NS); secondary endpoints improved (FACIT fatigue p=0.038, Bell score p=0.0004, FSS p=0.009) in an exploratory crossover analysis. BLOC (Berlin Cures, NCT05911009, ~114 participants, 19 sites): topline November 2024 showed no difference from placebo; data unpublished; the company stopped operating. The negative result is larger but unpublished; the positive one is small, exploratory and multiplicity-prone. Combined with the immunoadsorption sham trials and the negative antigen-profiling study, the GPCR-autoantibody route has weakened.

Evidence

Study	n	Design	Result
Hohberger 2025, eClinicalMedicine (reCOVer)	30	phase IIa double-blind placebo-controlled crossover RCT	Safety primary met; secondary fatigue/QoL improvements

Against or not replicated

Study	n	Design	Result
Berlin Cures BLOC trial topline, 2024 (NCT05911009) unverified	~114	phase II double-blind RCT	No significant effect vs placebo (company announcement; unpublished)

Open questions

Will the BLOC data be published?
Is there any validated assay for functional GPCR antibodies?

Bears on: Autoimmunity (functional autoantibodies) Endothelial dysfunction · Test: Not available

Immune: immunoglobulin

IVIG has no consistent controlled evidence: 1990s ME/CFS RCTs conflicted, and a 30-person trial in autoimmune POTS found no difference from albumin. Contested

Infusions of pooled antibodies from donors are sometimes used for severe patients. Old trials gave mixed answers and the most recent small trial in POTS found the same improvement with a volume-matched control infusion.

iSTAND (Vernino 2024): 16 IVIG vs 14 albumin, 0.4 g/kg x8 over 12 weeks; COMPASS-31 change -5.5 vs -10.6 (p=0.63); responders 46.7% vs 38.5% (NS). Albumin also expands volume, so the comparator may have masked or matched a volume effect. 1990s ME/CFS RCTs: Lloyd 1990 (n=49) positive, Peterson 1990 (n=30) negative, Vollmer-Conna 1997 (n=99) negative. In the Eckey/Xiao 2025 patient survey IVIG was rated among the most helpful in the severe multisystemic cluster (self-report, selected users). The RECOVER-AUTONOMIC IVIG arm in long-COVID POTS has not yet reported (as of Oct 2026).

Evidence

Study	n	Design	Result
Vernino 2024, Clin Auton Res (iSTAND)	16 vs 14	RCT vs albumin	No difference in COMPASS-31 (p=0.629)
Lloyd 1990, Am J Med unverified	49	RCT	Benefit reported

Against or not replicated

Study	n	Design	Result
Vollmer-Conna 1997, Am J Med unverified	99	RCT, adolescents/adults	No benefit over placebo
Peterson 1990, Am J Med unverified	30	RCT	No benefit

Open questions

What will RECOVER-AUTONOMIC's IVIG arm show?
Is there a small-fibre-neuropathy subgroup that responds?

Bears on: Autoimmunity (functional autoantibodies) Autonomic dysfunction and small-fibre neuropathy · Test: Prescribed in some specialist clinics; costly

Immune: TLR3 agonist

Rintatolimod (Ampligen) gave a modest placebo-adjusted gain in treadmill time in a 234-person phase III trial, but the result was borderline and the FDA declined approval. Contested

An immune-stimulating drug given by drip twice a week helped exercise time a little in one large company trial; regulators did not find this convincing.

AMP-516 (Strayer 2012): 40 weeks IV 400 mg twice weekly; intention-to-treat placebo-adjusted exercise tolerance +21.3% (p=0.047), +28% after compliance correction; reduced use of symptomatic drugs (p=0.048). Lead authors employed by the manufacturer. FDA issued a complete response letter in 2009 and has not approved it. A 2020 post hoc analysis suggested better response in illness of 2-8 years' duration. Not replicated by an independent group.

Evidence

Study	n	Design	Result
Strayer 2012, PLoS One	234	phase III double-blind RCT, 12 sites	Exercise tolerance +21.3% vs placebo (p=0.047)

Against or not replicated

Study	n	Design	Result
Strayer 2020, PLoS One (post hoc) unverified	ITT population	post hoc subgroup	Benefit concentrated in 2-8 years' duration (hypothesis-generating)

Open questions

Would an independent trial with a pre-specified duration subgroup replicate?

Bears on: Innate immune hyperactivation Viral persistence · Test: Expanded access in the US only

Immune/viral: HERV-W

The anti-HERV-W envelope antibody temelimab did not improve fatigue versus placebo in 203 HERV-W-positive long-COVID patients. Refuted

A drug aimed at a dormant virus-like gene product that can be switched on after infection did not help long-COVID fatigue in a proper trial.

GNC-501 (GeNeuro, 2024): IV temelimab vs placebo, primary endpoint 3-point fall in PROMIS Fatigue SF-7a at week 24; no clinically meaningful difference and most secondary endpoints negative; well tolerated. Company suggested shorter disease duration or changes in HERV-W status may matter (post hoc). Company press release; full peer-reviewed paper not seen.

Evidence

Study	n	Design	Result
GeNeuro 2024, GNC-501 topline (company release) unverified	203	phase II double-blind RCT	Primary fatigue endpoint not met

Open questions

Is HERV-W activation a marker without causal role?

Bears on: Viral persistence Innate immune hyperactivation · Test: Not available

Immune/metabolic: mTOR inhibition

Low-dose rapamycin (6 mg weekly) was followed by self-reported improvement in about three-quarters of completers in an uncontrolled decentralised study; randomised trials are pending. Early

A low weekly dose of an immune-modulating drug thought to restart cellular clean-up (autophagy) was followed by improvement in many patients, but no one in the study got a placebo.

Ruan 2025: 86 enrolled, 70 reached day 30; 52 (74.3%) reported improvement in fatigue, PEM and orthostatic intolerance, with MFI and SF-36 gains; autophagy markers (pSer258-ATG13 down, BECLIN-1 up) moved and correlated with outcomes. No serious adverse events. Follow-on metabolomics (Gile 2026) in responders only. Uncontrolled response rates of this size are similar to placebo responses in RCTs (35-40%+) plus attrition bias. The Mount Sinai placebo-controlled trial in long COVID (NCT06960928, ~80-90) ends Nov 2026.

Evidence

Study	n	Design	Result
Ruan 2025, J Transl Med	86 (70 assessed)	decentralised open-label	74.3% improved; autophagy markers changed
Gile 2026, J Transl Med	responder subset	observational mechanistic follow-up	Purine metabolism and mitochondrial changes in responders

Open questions

Will the Mount Sinai RCT confirm benefit?
Can autophagy markers select responders prospectively?

Bears on: Mitochondrial / energy metabolism impairment Immune (T-cell) exhaustion · Test: Off-label prescription in some US clinics; research only in UK

Antiviral: SARS-CoV-2

Three randomised trials of nirmatrelvir-ritonavir (Paxlovid) for 15-25 days in established long COVID (total ~1,200 people) found no benefit on any phenotype. Refuted

If leftover coronavirus were driving long COVID, an antiviral might help. Three good trials found it did not, at least at the doses and durations tested.

RECOVER-VITAL (Baden 2026): 959 randomised to 15 or 25 days vs placebo-ritonavir, cognitive, autonomic and exercise phenotypes; no primary benefit (e.g. autonomic -6.4%, 95% CI -18.5 to 5.7). STOP-PASC (Stanford, n=155, 15 days) and PAX LC (Yale, n=100, 15 days, PROMIS-29) also null. STOP-PASC immune profiling found no EBV reactivation or microclot differences versus controls. The trials do not exclude a role for persistence in tissue reservoirs that this drug fails to reach, or effects in very early or highly selected patients.

Evidence

Study	n	Design	Result
Baden 2026, Lancet Infect Dis (RECOVER-VITAL)	959	phase 2 double-blind RCT, 69 US sites	No benefit for any phenotype at day 90
Sawano 2025, Lancet Infect Dis (PAX LC) unverified	100	phase 2 double-blind RCT	No change in PROMIS-29 physical score vs placebo
Geng 2024, JAMA Intern Med (STOP-PASC)	155	double-blind RCT	No benefit

Open questions

Would longer courses, combination antivirals or monoclonal antibodies (trials under way) reach tissue reservoirs?
Is there a biomarker of persistent antigen that would select responders?

Bears on: Viral persistence · Test: Not recommended for long COVID

Antiviral: herpesviruses

Valganciclovir in 30 patients selected for high HHV-6 and EBV antibody levels improved some secondary fatigue and cognition scores in a small RCT; it has not been replicated. Early

Antivirals against herpesviruses have been tried in patients with high antibody levels, with modest results in one small trial. High antibodies show past infection, not necessarily active virus.

Montoya 2013: 2:1 randomised, 6 months; reported benefit on MFI mental fatigue (p=0.039), Fatigue Severity Scale (p=0.006) and cognition (p=0.025); primary endpoint detail not confirmed this session. Preceded by an open-label series (Kogelnik 2006, 9/12 near-resolution). Neutropenia risk requires monitoring. Lerner's valacyclovir/valganciclovir EBV-subset series are uncontrolled. Selection by IgG titre is a weak marker of active replication.

Evidence

Study	n	Design	Result
Montoya 2013, J Med Virol unverified	30	double-blind RCT 2:1	Secondary fatigue/cognition improvements
Kogelnik 2006, J Clin Virol	12	open-label	9/12 near resolution

Open questions

Would selection by viral DNA (PCR) rather than antibody titre identify responders?

Bears on: Herpesvirus reactivation (EBV, HHV-6) Viral persistence · Test: EBV/CMV/HHV-6 PCR and serology are clinically available; antiviral use off-label

Neuro-immune: low-dose naltrexone

The first sizeable placebo-controlled trial of low-dose naltrexone in post-COVID ME/CFS (66 vs 71, 16 weeks, up to 4.5 mg) found no fatigue benefit, with over 40% improving on placebo. Refuted

Low-dose naltrexone is widely used off-label and many patients report benefit. In a proper trial, the drug did not reduce fatigue more than placebo. Pain and subgroup results are still being analysed.

The trial is Canadian (Nacul, University of British Columbia; NCT05430152), presented at the Berlin International ME/CFS Conference, May 2026; preliminary and unpublished. Earlier evidence was retrospective (Polo 2019, n=218, 74% self-reported benefit) or uncontrolled long-COVID series. I found no separate German LDN RCT; the 'German' attribution likely reflects the Berlin venue. The LIFT trial (pyridostigmine x LDN factorial, n=160, Harvard/OMF, NCT06366724) and a RECOVER paediatric trial are pending. Effects on pain, sleep or specific subgroups remain possible.

Evidence

Study	n	Design	Result
Nacul et al., presented May 2026 (UBC; NCT05430152)	66 vs 71	double-blind placebo-controlled RCT, 16 weeks	Primary (fatigue) not met; >40% placebo responders

Against or not replicated

Study	n	Design	Result
Polo 2019, Fatigue: Biomed Health Behav unverified	218	retrospective chart review	74% reported benefit

Open questions

Does LDN help pain or a pain-predominant subgroup?
What will the LIFT factorial show?

Bears on: Neuroinflammation (microglial activation) Innate immune hyperactivation · Test: Private/off-label prescription

Neuro: low-dose aripiprazole

In a retrospective single-clinic review, 74% of 101 patients on very low-dose aripiprazole (mostly 0.25-2 mg) reported improvement; there is no randomised trial. Early

A tiny dose of an antipsychotic drug, thought to act on dopamine and brain immune cells, was linked to improvement in a clinic's records, but this kind of study cannot rule out placebo or selective reporting.

Crosby 2021 (Stanford): 101 patients; ~74% reported improvement in brain fog, fatigue, unrefreshing sleep or energy; side effects included insomnia, akathisia, vivid dreams. Retrospective, no control, outcomes from clinic notes. Rationale (microglial modulation, dopamine partial agonism) is speculative. No RCT reported as of Oct 2026.

Evidence

Study	n	Design	Result
Crosby 2021, J Transl Med unverified	101	retrospective single-clinic	74% improved

Open questions

Would a placebo-controlled trial confirm benefit?

Bears on: Neuroinflammation (microglial activation) · Test: Off-label prescription

Circulation: cholinergic / preload

A single 60 mg pyridostigmine dose raised peak oxygen uptake relative to placebo on a repeat invasive exercise test in 45 ME/CFS patients with preload failure, largely because the placebo group got worse; longer trials are negative or pending. Early

Pyridostigmine boosts nerve signalling that tightens veins and returns blood to the heart. In a one-day lab study it stopped exercise capacity dropping on a second test. Whether it helps daily life is untested; the factorial LIFT trial should answer that.

Joseph/Systrom 2022: 23 vs 22; peak VO2 +13.3 vs -40.2 mL/min (difference ~54 mL/min), with cardiac output and right atrial pressure maintained instead of falling; oxygen extraction unchanged. Same-day laboratory endpoint. São Paulo 2025 (ATS abstract, long COVID with RV preload dysfunction, n=32, up to 360 mg/day for 6-10 weeks): negative overall (p=0.35), borderline +103 mL in a post hoc subgroup. Kanjwal 2011 POTS retrospective (n=203): 43% improved, 19% stopped for GI effects. LIFT (n=160, 13 weeks, FUNCAP-55 and CPET endpoints) has not reported.

Evidence

Study	n	Design	Result
Joseph 2022, Chest	45 (23 vs 22)	single-dose double-blind RCT between two iCPETs	Peak VO2 +13 vs -40 mL/min

Against or not replicated

Study	n	Design	Result
São Paulo group 2025, ATS abstract unverified	32	RCT 6-10 weeks, long COVID	Negative overall (p=0.35)
Kanjwal 2011, Pacing Clin Electrophysiol unverified	203	retrospective POTS	43% improved; 19% stopped for GI effects

Open questions

Does LIFT show benefit on daily function?
Does preload failure on iCPET predict response?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Licensed drug (myasthenia); off-label use; iCPET research only

Circulation: heart-rate control

Ivabradine lowered standing heart rate but did not improve symptoms in 181 people with long-COVID POTS (RECOVER-AUTONOMIC), despite an earlier positive 22-person crossover in hyperadrenergic POTS. Refuted

Slowing a racing heart is not the same as making people feel better. A large trial in long-COVID POTS showed exactly that.

Taub 2021: 22 hyperadrenergic POTS (noradrenaline >600 pg/mL), 1-month crossover; lower HR and better RAND-36 QoL. RECOVER-AUTONOMIC (2026, ACC late-breaker): 181 adults, 12 weeks; no meaningful change in orthostatic intolerance or long-COVID symptoms vs placebo, though supine-to-standing HR change fell significantly. The factorial coordinated non-pharmacological care arm (diet-mediated volume expansion, rehab, education) is reported separately. Full paper pending. Whether a hyperadrenergic subgroup benefits remains open.

Evidence

Study	n	Design	Result
Taub 2021, J Am Coll Cardiol	22	double-blind crossover RCT	Lower HR and better QoL
RECOVER-AUTONOMIC ivabradine, ACC 2026 (NIH release)	181	double-blind placebo-controlled RCT	No symptom benefit; HR change reduced

Open questions

Does a noradrenaline-defined hyperadrenergic subgroup respond?

Bears on: Autonomic dysfunction and small-fibre neuropathy Low blood volume and preload failure · Test: Licensed drug; off-label for POTS

Circulation: vasoconstrictors and volume

Fludrocortisone did not beat placebo in a 100-person RCT in CFS with neurally mediated hypotension; midodrine reduces orthostatic heart rate acutely in small POTS studies but has no ME/CFS outcome trial. Refuted

Two standard drugs for low blood pressure on standing have been tried. Fludrocortisone, which makes the body hold salt, did not help in a proper trial in CFS. Midodrine tightens blood vessels and helps some POTS patients in the short term, but it has not been properly tested in ME/CFS.

Rowe 2001: fludrocortisone 0.1 mg vs placebo for 9 weeks in CFS + NMH on tilt; >=15-point wellness improvement in 14% vs 10% (p=0.76). Peterson 1998 (n=25) also negative. Midodrine: small acute crossover studies in POTS/orthostatic intolerance show lower standing HR (memory, not re-checked); risks include supine hypertension (relevant for bedbound patients). No ME/CFS RCT with function as endpoint. Grade refers to fludrocortisone; midodrine evidence is early.

Evidence

Study	n	Design	Result
Rowe 2001, JAMA	100	double-blind RCT	14% vs 10% improved (p=0.76)

Against or not replicated

Study	n	Design	Result
Ross 2014, Am J Physiol (midodrine in POTS) unverified	small	acute crossover	Improved orthostatic tolerance acutely

Open questions

Would a midodrine trial in ME/CFS with low-flow physiology show functional benefit?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: Licensed drugs; tilt test available on NHS

Metabolic: supplements

Two Spanish RCTs of CoQ10 200 mg + NADH 20 mg (n=73, n=207) reported modest reductions in fatigue scores; CoQ10 alone failed in a 121-person long-COVID crossover trial. Contested

Supplements meant to support mitochondria produced small improvements in two trials from one group, with supplier involvement. CoQ10 on its own did nothing in long COVID.

Castro-Marrero 2015: 8 weeks, n=73; FIS-40 improved (p<0.05) with higher NAD+/NADH, CoQ10, ATP in PBMCs. Castro-Marrero 2021: 12 weeks, 104 vs 103; cognitive fatigue and FIS-40 total improved (p~0.02), SF-36 improved. Same research group, industry links; effect sizes small and not replicated independently. Hansen 2023 (Denmark, CoQ10 500 mg/day, n=121 crossover) in long COVID: no benefit.

Evidence

Study	n	Design	Result
Castro-Marrero 2015, Antioxid Redox Signal	73	double-blind RCT	FIS-40 improved (p<0.05)
Castro-Marrero 2021, Nutrients	207	double-blind RCT	Cognitive fatigue and FIS-40 improved

Against or not replicated

Study	n	Design	Result
Hansen 2023, eClinicalMedicine unverified	121	crossover RCT, long COVID	CoQ10 alone: no benefit

Open questions

Would an independent group replicate the CoQ10+NADH result?

Bears on: Mitochondrial / energy metabolism impairment · Test: Over-the-counter

Oxaloacetate 2 g/day reduced Chalder fatigue more than control (−25% vs −10%) in an 82-person company-run RCT in ME/CFS; a 69-person long-COVID trial missed its primary endpoint. Early

A supplement that feeds a step in energy metabolism reduced fatigue in one trial run with the manufacturer. A similar trial in long COVID did not meet its main goal.

RESTORE ME (Cash 2024): 90 days, 2,000 mg vs rice-flour control; fatigue −25% vs −10% (between-group p=0.0039); secondary analysis (Vernon 2025) showed cognitive gains at day 60. Sponsor (Terra Biological) involvement. REGAIN (long COVID, n=69, 42 days): primary Chalder negative, secondary cognition gains. No independent replication.

Evidence

Study	n	Design	Result
Cash 2024, Front Neurol (RESTORE ME)	82	double-blind controlled RCT	Fatigue −25% vs −10%, p=0.0039
Vernon 2025, Front Neurol	82	secondary analysis	Cognitive improvement at day 60 (p=0.034)

Against or not replicated

Study	n	Design	Result
REGAIN 2025 (long COVID) unverified	69	RCT	Primary endpoint not met

Open questions

Independent replication with objective endpoints (actigraphy)?

Bears on: Mitochondrial / energy metabolism impairment IDO metabolic trap (Phair, Davis) · Test: Over-the-counter

Physical: hyperbaric oxygen

Forty daily sessions of hyperbaric oxygen at 2.0 ATA improved cognition and symptoms versus sham in a 73-person long-COVID trial, but ten sessions at 2.4 bar showed no benefit in an 80-person sham-controlled trial. Contested

Breathing pure oxygen under pressure helped in an Israeli trial with 40 sessions, but not in a Swedish trial with 10 sessions. Dose may matter, or the positive trial may not hold up. It needs daily travel, which many patients cannot do.

Zilberman-Itskovich 2022 (Tel Aviv): 37 vs 36, 40 sessions, 5/week, 100% O2 at 2.0 ATA for 90 min with air breaks, sham at 1.03 ATA air; global cognition improved (d≈0.5, p=0.038), plus energy, sleep, psychiatric symptoms and pain; benefits persisted at ~1-year follow-up in 31 patients (Hadanny 2024); a cardiac substudy showed improved GLS. HOT-LoCO (Kjellberg 2025, Sweden): 40 vs 40, 10 sessions over 6 weeks, 100% O2 at 2.4 bar 90 min vs air at 1.2-1.34 bar; RAND-36 PF difference 0.63 (95% CI −7.0 to 8.3, p=0.87); both groups improved. Charité (Kedor, 2026 conference): 20 vs 40 sessions; improvements only in the 40-session group (no sham arm reported). The 'sham' pressure differs between trials and blinding of pressure is imperfect. No ME/CFS (non-COVID) RCT.

Evidence

Study	n	Design	Result
Zilberman-Itskovich 2022, Sci Rep	73	double-blind sham RCT, 40 sessions at 2.0 ATA	Cognition, energy, sleep, pain improved
Hadanny 2024, Sci Rep	31	1-year follow-up of RCT HBOT arm	Improvements maintained

Against or not replicated

Study	n	Design	Result
Kjellberg 2025, BMJ Open (HOT-LoCO)	80	double-blind sham RCT, 10 sessions at 2.4 bar	No difference (RAND-36 PF diff 0.63, p=0.87)

Open questions

Is there a dose-response (sessions x pressure) in a single trial with proper sham?
Does it apply to non-COVID ME/CFS?

Bears on: Capillary and microvascular damage Mitochondrial / energy metabolism impairment Neuroinflammation (microglial activation) · Test: Private clinics; not NHS-funded for this use

Rehabilitation: GET and CBT

PACE (n=641) reported modest gains in self-rated fatigue and function with CBT and graded exercise, but these were not matched by objective measures, used Oxford criteria, and vanished between groups by 2.5 years. Contested

The largest trial of talking therapy and gradually increasing exercise said both helped moderately. Re-analysis using the trial's own original plan found the benefits much smaller and limited to questionnaires that unblinded patients filled in.

White 2011: CBT and GET vs specialist medical care: Chalder fatigue −3.4 and −3.2 points, SF-36 PF +7.1 and +9.4 at 52 weeks; adaptive pacing no better than SMC. Long-term follow-up (Sharpe 2015): no between-group differences at median 31 months. Wilshire 2018 re-analysis using protocol-specified outcomes (data via FOI): CBT/GET did not significantly beat control after correction; recovery rates low (single digits to ~10%) and not different; objective measures (6-minute walk modest, step test, employment, benefits) largely null. Criticisms: Oxford criteria (fatigue only; PEM not required), outcome switching, unblinded self-report, mid-trial newsletter. Cochrane exercise review (Larun, re-dated 2024) rates evidence low-moderate and notes limited applicability to ME/CFS defined with PEM; a full update is pending.

Evidence

Study	n	Design	Result
White 2011, Lancet (PACE)	641	4-arm RCT	Modest self-reported gains for CBT/GET; APT no effect
Sharpe 2015, Lancet Psychiatry	481 followed	long-term follow-up	No between-group differences at ~2.5 years

Against or not replicated

Study	n	Design	Result
Wilshire 2018, BMC Psychol	641 (re-analysis)	protocol-based re-analysis	CBT/GET not significantly better after correction; effects confined to self-report
Larun 2019/2024, Cochrane (CD003200) unverified	8 RCTs	systematic review	Exercise may slightly reduce fatigue; low-certainty; applicability to PEM-defined ME/CFS uncertain

Open questions

Would a blinded-assessor trial with objective endpoints in PEM-defined patients show any benefit?

Bears on: Persistent sickness behaviour / vagal signalling · Test: n/a

Rehabilitation: guidance and harms

NICE's 2021 guideline (NG206) told clinicians not to offer graded exercise therapy, and patient surveys plus a 2026 German rehabilitation study report frequent worsening after exercise-based or even pacing-focused rehab. Several studies

England's health guidance body reviewed the evidence and concluded that the old exercise trials were low quality and that GET should not be offered. Large patient surveys and a German study suggest rehab programmes often make people worse.

NICE NG206 (Oct 2021): rated the GET/CBT evidence 'low' or 'very low' quality; recommends against GET or any programme based on fixed incremental increases; CBT may be offered only as supportive therapy, not as cure. Eckey/Xiao 2025 survey (n≈3,925): GET had a net score of about −72% (most harmful item). CFS_CARE (Charité, 2026 conference): a specialised rehab programme including pacing left 42 of 97 participants worse on the Bell scale, few better. Survey data are self-selected; CFS_CARE is uncontrolled and preliminary. Nevertheless the direction is consistent across sources.

Evidence

Study	n	Design	Result
NICE 2021, NG206 unverified	guideline	evidence review	Do not offer GET; CBT only as supportive
Eckey 2025, PNAS	~3,925	patient survey	GET net −72%
Wittke/Scheibenbogen, CFS_CARE, presented 2026	97	uncontrolled rehab study	42/97 worsened

Against or not replicated

Study	n	Design	Result
White 2011, Lancet (PACE)	641	RCT	No excess serious adverse reactions reported with GET

Open questions

What exactly distinguishes harmful from tolerable activity programmes?
How common is lasting deterioration after rehab in a controlled study?

Bears on: Persistent sickness behaviour / vagal signalling Mitochondrial / energy metabolism impairment · Test: n/a

Self-management: pacing

Pacing (staying within one's energy limits) is the top-rated approach in patient surveys and NICE's recommended foundation, but it has never shown benefit in a controlled trial (adaptive pacing was no better than usual care in PACE). Early

Patients consistently rate pacing as the most helpful thing they do, and guidelines recommend it. It has not been properly tested in a trial designed with patients, so its benefit rests on consensus and self-report.

Eckey/Xiao 2025: pacing net score ~75% (highest of >150 treatments). PACE APT arm (n=160): no difference vs SMC. Critics argue APT in PACE was not the envelope-style pacing patients use. Pacing is plausible as harm avoidance (preventing PEM) rather than a treatment. Heart-rate-guided pacing studies are small and uncontrolled.

Evidence

Study	n	Design	Result
Eckey 2025, PNAS	~3,925	patient survey	Pacing net assessment ~75%

Against or not replicated

Study	n	Design	Result
White 2011, Lancet (PACE)	160 APT vs 160 SMC	RCT	APT no better than specialist care

Open questions

Does HR- or wearable-guided pacing reduce PEM frequency in a randomised trial?

Bears on: Mitochondrial / energy metabolism impairment Low blood volume and preload failure · Test: Self-management; NHS ME/CFS services

Patient-reported outcomes

A 3,925-person survey rated >150 treatments; illness severity predicted perceived response more strongly than diagnosis, and symptom clusters (severe multisystemic, POTS-dominant, cognitive/pain, mild) preferred different treatments. Early

Thousands of patients rated what helped. The cheap supportive measures came top (pacing, fluids, compression), graded exercise came bottom, and different kinds of patients reported benefit from different things.

Eckey, Li, Morrison, Davis, Xiao, PNAS 2025: ME/CFS and long COVID ratings correlated (R²=0.68). Top: pacing 75%, fluids/electrolytes ~69%, compression ~62%; GET −72%. Clusters: severe multisystemic (IVIG, lymphatic drainage), POTS-dominant (compression, autonomic modulators), cognitive (CNS stimulants), mild. Self-selected social-media sample, no placebo, recall bias; useful for prioritising trials, not for efficacy claims.

Evidence

Study	n	Design	Result
Eckey 2025, PNAS	~3,925	online patient survey	Pacing, fluids, compression highest; GET lowest; severity shaped response

Open questions

Do survey-defined clusters map onto biological subgroups?

Bears on: Low blood volume and preload failure Autonomic dysfunction and small-fibre neuropathy · Test: n/a

Trial methodology

Placebo and natural-course improvement in ME/CFS and long-COVID trials is large (35% responders in RituxME, >40% in the LDN trial, both arms improving in IVIG, HBOT and immunoadsorption trials), so uncontrolled response rates of 50-75% are uninterpretable. Replicated

In almost every controlled trial, many people on the dummy treatment also improve. This is why open-label studies that report most patients getting better cannot show a treatment works.

RituxME placebo SF-36 PF 32.3 to 45.5 at 18 months; LDN >40% placebo responders; iSTAND albumin COMPASS −10.6; HOT-LoCO both arms improved; IAMPOCO both arms improved; BLOC failure attributed partly to placebo response. Contributing factors: regression to the mean (patients enrol when worse), natural fluctuation, expectation, and post-COVID spontaneous recovery in early cohorts.

Evidence

Study	n	Design	Result
Fluge 2019, Ann Intern Med	151	RCT	35.1% placebo response
Vernino 2024, Clin Auton Res	30	RCT	Albumin arm improved as much as IVIG
Kjellberg 2025, BMJ Open	80	RCT	Both arms improved

Open questions

What run-in periods and objective endpoints best reduce placebo noise?

Test: n/a

Natural history: recovery

Full recovery from established adult ME/CFS is uncommon: a systematic review of 14 criteria-based cohorts found a median full-recovery rate of 5% (range 0-31%) and median improvement of 39.5%. Several studies

Most adults with long-standing ME/CFS do not fully recover, although many improve somewhat over years. Children and people early in the illness do better.

Cairns & Hotopf 2005: 28 studies, 14 with operational CFS criteria; full recovery median 5%, improvement 39.5% (8-63%); return to work 8-30%. Better outcome with lower baseline fatigue. Adolescents have better prognosis (studies report a majority improving or recovering over 5-10 years; memory, not re-checked). RituxME placebo arm at 6 years: 7.4% reached SF-36 PF >=90. Old cohorts used heterogeneous criteria; newer prospective data in post-COVID ME/CFS (Legler 2023) show little improvement to 20 months.

Evidence

Study	n	Design	Result
Cairns & Hotopf 2005, Occup Med	14 CFS cohorts	systematic review	Median full recovery 5%; improvement 39.5%
Rekeland 2024, PLoS One	placebo arm ~54	6-year follow-up	7.4% placebo at SF-36 PF >=90; 14.8% worsened >=20 points

Open questions

What are recovery rates in PEM-defined adult cohorts followed prospectively from onset?

Test: n/a

Natural history: post-COVID ME/CFS

About 4.5% of SARS-CoV-2-infected adults in RECOVER met IOM ME/CFS criteria (vs 0.6% uninfected; HR 4.93), and those meeting Canadian criteria at Charité stayed severely ill to 20 months while other long-COVID patients improved. Several studies

COVID roughly quintupled the rate of new ME/CFS. People whose long COVID meets ME/CFS criteria tend not to recover over the first two years, unlike milder long COVID.

Vernon 2025 (RECOVER-Adult, n=11,785 infected, 1,439 uninfected): incidence 2.66 vs 0.93 per 100 person-years; 88.7% of post-COVID ME/CFS also met long-COVID criteria; criteria based on self-report. Legler 2023 (Charité, n=106, 3-8, 9-16, 17-20 months): PCS-ME/CFS persistently severe; lower initial grip strength predicted persistence. Kedor 2022 (n=42): about half met CCC at 6 months.

Evidence

Study	n	Design	Result
Vernon 2025, J Gen Intern Med	11,785 infected vs 1,439	prospective cohort	4.5% vs 0.6% met ME/CFS criteria; HR 4.93
Legler 2023, eClinicalMedicine	106	prospective cohort	PCS-ME/CFS severity persistent at 20 months
Kedor 2022, Nat Commun	42	prospective observational	~half met CCC ME/CFS at 6 months

Open questions

What fraction of post-COVID ME/CFS recovers by 5 years?

Bears on: Viral persistence Autoimmunity (functional autoantibodies) · Test: n/a

Natural history: prevalence and sex ratio

ME/CFS affects roughly 0.2-1% of adults depending on criteria (US 2021-22: 1.3% self-reported diagnosis; England: 0.16% with a hospital-recorded diagnosis), with women affected 3-4 times as often as men. Replicated

Somewhere between 1 in 100 and 1 in 500 adults have ME/CFS, depending on how it is counted. About three to four in every four patients are women.

Lim 2020 meta-analysis (46 studies): pooled prevalence 0.39%, meta-analytic 0.68% (0.48-0.97); women ~1.5-2x more often in pooled data. Vahratian 2023 (NHIS 2021-22): 1.3% of US adults (women ~1.7% vs men ~0.9%) — self-report of a clinician diagnosis plus current illness. UK: Nacul 2011 primary-care minimum 0.2%; Samms & Ponting 2025 (Hospital Episode Statistics): 100,055 diagnosed in England (0.16%), F:M 3.88:1, peaking ~6:1 in the 40s; white people ~5x more likely diagnosed, suggesting under-diagnosis. DecodeME's cohort is ~83% female. Commonly cited UK total: ~250,000-400,000 (estimate). Long COVID: ONS estimated ~2 million self-reported (~3%) in England/Scotland in 2024 (memory).

Evidence

Study	n	Design	Result
Lim 2020, J Transl Med	46 studies, >1 million participants	meta-analysis	Pooled 0.39%; meta-analytic 0.68%
Vahratian 2023, NCHS Data Brief 488 unverified	NHIS 2021-22	national survey	1.3% of US adults
Samms & Ponting 2025, BMC Public Health unverified	100,055 diagnosed	national hospital records	0.16%; F:M 3.88:1
Nacul 2011, BMC Med unverified	3 English regions	primary care cross-sectional	Minimum prevalence 0.2%

Open questions

Why the female predominance (DecodeME found few sex-specific loci)?
How much is under-diagnosis in men and ethnic minorities?

Bears on: Genetic susceptibility acting through neurons Autoimmunity (functional autoantibodies) · Test: n/a

Natural history: severity

Roughly a quarter of people with ME/CFS are estimated to be housebound or bedbound, and severity, not diagnosis, was the strongest predictor of reported treatment response. Early

A large minority are too ill to leave home. Severely ill patients are under-represented in almost every trial, so trial results may not apply to them.

The 25% severe figure (often cited from IOM 2015 and UK charity surveys) rests on clinic and survey estimates, not population sampling (memory). Eckey/Xiao 2025: severity shaped treatment ratings more than ME/CFS vs long COVID. RituxME and CycloME excluded the most severe; Fluge's daratumumab study included moderate-to-severe. The original Fluge 2016 serum effect came from severe patients and the 2026 replication used mostly moderate patients.

Evidence

Study	n	Design	Result
IOM 2015, Beyond ME/CFS unverified	report	evidence review	At least a quarter housebound/bedbound at some point (estimate)
Eckey 2025, PNAS	~3,925	patient survey	Severity strongest correlate of response

Open questions

Population-based severity distribution?
Trials designed for severe, bedbound patients (home-based endpoints)?

Test: n/a

Natural history: mortality

In 2,147 secondary-care CFS patients in London, all-cause and cancer mortality were not raised, but suicide mortality was about seven times expected (SMR 6.85). Several studies

ME/CFS is not, on average, shown to shorten life in registry data, but suicide risk is markedly higher, and rare deaths from severe illness (for example malnutrition) are documented.

Roberts 2016 (SLaM CRIS, 2007-2013): 17 deaths; all-cause SMR 1.14 (0.65-1.85); cancer SMR 1.39 (0.60-2.73); suicide SMR 6.85 (2.22-15.98). Jason 2006 memorial-list analysis reported younger age at death from heart failure, cancer and suicide, but the sample is self-selected and biased. Deaths from severe ME with nutritional failure (e.g. UK coroner cases) are reported but not systematically counted. Small numbers of deaths limit precision.

Evidence

Study	n	Design	Result
Roberts 2016, Lancet	2,147	retrospective cohort	All-cause SMR 1.14 (NS); suicide SMR 6.85

Against or not replicated

Study	n	Design	Result
Jason 2006, Health Care Women Int	166 deaths (memorial list)	case series	Younger mean age at death for several causes; biased sample

Open questions

Mortality in severe/very severe cohorts?
Suicide risk factors (disbelief, lack of care, pain)?

Test: n/a

Key reviews

NICE 2021, NG206 ME/CFS guideline: Most thorough independent appraisal of GET/CBT evidence; current UK standard of care.
Eckey 2025, PNAS: Patient-reported treatment outcomes in ME/CFS and long COVID: Largest dataset on what patients use and rate; maps treatments to symptom clusters.
Wilshire 2018, BMC Psychology: PACE re-analysis: Protocol-based re-analysis of the most influential behavioural trial.
Blitshteyn 2025, Front Cell Infect Microbiol: Immunotherapies for POTS and long COVID: Overview of immune treatments and the trials still needed.
Proal 2025, Lancet Infect Dis: Targeting the SARS-CoV-2 reservoir: Rationale for antiviral/monoclonal trials after the Paxlovid failures.
Cairns & Hotopf 2005, Occup Med: prognosis of CFS: Still the standard systematic review of natural history.

The deepest blood studies, and what they miss

A null result only rules out what was measured, in the people sampled, at the time sampled.

Study	People	Compartments	Time points	Severity	Live-cell tests	Main limits
Walitt 2024 (NIH intramural)	17 PI-ME/CFS vs 21 HV	plasma, CSF, PBMC, stool, muscle physiology, brain imaging	mostly single visit; exercise tests	mild-moderate; strict entry criteria	flow cytometry only	very small n; results by sex underpowered
BioMapAI / Xiong 2025 (JAX, Bateman Horne)	249 participants, 4 years	plasma metabolome, stool metagenome, immune flow, routine labs	up to 4 yearly visits at rest	mixed, mostly ambulatory	stimulated cytokine flow	no disease controls; external AUC 0.59-0.77
Germain 2022 / 2025, Giloteaux 2024, Glass 2025 (Cornell)	60/45 metabolome; 79/53 proteome; EV subsets 18/17 F, 10/12 M	plasma metabolome, SomaScan proteome, EVs	pre/post CPET day 1 and day 2 (15 min, 24 h)	able to complete 2 CPETs	none in these papers	no 48-72 h samples; most impaired excluded or missing
Beentjes 2025 (UK Biobank, All of Us)	up to 1,455 cases vs 131,303	routine biochemistry, counts, NMR, Olink	single baseline	unknown; record/self-report diagnosis	none	loose case definition; healthy-volunteer bias
Hoel 2021 / Hoel 2026 (Bergen)	83/35 metabolome; 50/29 proteome	serum	single resting	mild to severe	none	sample-handling signal in proteome; per-person data not public

Gaps nobody has closed

Blood flow, vessels and blood: No blinded, multi-centre replication of the van Campen cerebral blood flow method.
Blood flow, vessels and blood: Preload failure prevalence in unselected ME/CFS (iCPET cohorts are pre-selected).
Blood flow, vessels and blood: No placebo-controlled trials of volume expansion (saline, desmopressin, fludrocortisone) targeting measured low blood volume in ME/CFS.
Blood flow, vessels and blood: Microclots: no standardised blinded multi-lab assay and no in vivo demonstration.
Blood flow, vessels and blood: Retinal, sublingual and muscle capillary studies are almost all post-COVID; pre-pandemic ME/CFS is rarely studied.
Blood flow, vessels and blood: Severe and very severe patients are almost absent from vascular studies.
Blood flow, vessels and blood: Pericyte markers have never been stained in ME/CFS tissue.
Blood flow, vessels and blood: Complement activation fragments have not been measured with modern pre-analytics before and after exertion in ME/CFS.
Blood flow, vessels and blood: Whether endothelial dysfunction exceeds what activity-matched inactive controls show.
Blood flow, vessels and blood: Mechanism of low red-cell mass (marrow output vs destruction vs EPO signalling) is unstudied.
Muscle, exertion and energy: No ME/CFS (as opposed to long COVID) study has biopsied muscle before and after induced PEM; the registered Amsterdam study NCT05225688 and Systrom's OMF-funded biopsy study have not yet reported.
Muscle, exertion and energy: No independent replication of 2-day CPET from a group outside the Workwell/Ithaca/van Campen network that also reports test-retest variability in sedentary controls; Mancini 2026 null needs explaining.
Muscle, exertion and energy: Severe and housebound patients are almost absent from exercise, biopsy and MRS studies; a provocation-free design (e.g. orthostatic or small-muscle stressor) is needed.
Muscle, exertion and energy: No study pairs muscle biopsy, 31P-MRS, sodium MRI and invasive CPET in the same people to locate the limiting step in the oxygen cascade.
Muscle, exertion and energy: Serum-transfer studies have not used sera drawn during PEM, fractionation (IgG, complement, EVs), or blinded multi-lab sample exchange.
Muscle, exertion and energy: No pre-registered multi-site PBMC or muscle respirometry study with adequate controls exists.
Muscle, exertion and energy: Lean mass has not been measured alongside the low muscle-protein blood signal.
Muscle, exertion and energy: Handgrip fatigability needs external validation against deconditioned, depressed and other-fatigue controls before use as a trial endpoint.
Muscle, exertion and energy: Whether the amyloid-containing deposits in long COVID muscle also occur in ME/CFS, and what they are made of, is untested.
Nervous system and brain: No adequately powered (n>50 per group), multi-site TSPO-PET study in ME/CFS using a modern tracer and arterial input.
Nervous system and brain: No blinded, controlled imaging study of craniocervical anatomy in ME/CFS vs hypermobile and healthy controls, and no controlled trial of CCI surgery or CSF drainage.
Nervous system and brain: No skin-biopsy SFN prevalence study in unselected community ME/CFS with matched controls and blinded reading.
Nervous system and brain: Brain measures before and after exertion (MRS lactate, fMRI, CSF) in the same people across the 24-72 h PEM window are almost absent; Cook 2017 has not been directly replicated.
Nervous system and brain: No CSF study with a large healthy-control group and a pre-registered analysis; most recent CSF cohorts lack controls.
Nervous system and brain: No trial of treatments that target orthostatic hypocapnia or cerebral hypoperfusion (capnography-guided breathing, midodrine, compression) measuring function over weeks.
Nervous system and brain: Severe and very severe patients are largely excluded from imaging, CSF and sensory studies.
Nervous system and brain: Few studies stratify by orthostatic phenotype (POTS vs hypocapnia vs normal HR/BP), which may explain much of the inconsistency in perfusion and imaging results.
Nervous system and brain: No effort-based decision study in ME/CFS that calibrates task difficulty to each person's ability.
Immune system: No immune measurement during PEM: there is no paired rest / post-exertional-crash study of NK function, T-cell phenotype or autoantibodies, despite PEM being the defining feature.
Immune system: No published passive-transfer experiment with ME/CFS IgG (as opposed to long COVID or fibromyalgia IgG), even though that is now a standard, affordable design.
Immune system: No independent laboratory has attempted the TRPM3 patch-clamp or the IgG-endothelial-mitochondria assay; both rest entirely on one group each.
Immune system: No independent ME/CFS single-cell blood dataset exists outside one research group, so single-cell claims cannot be externally replicated.
Immune system: No prospective inception cohort with pre-infection immune samples large enough to test whether a 'primed' immune profile predicts ME/CFS after natural infection (the interferon-alpha model is the only proxy).
Immune system: Severe and very severe patients are almost absent from immune studies that require travel; the few in-home studies are small, and the one well-powered serum replication included only 6 severe patients.
Immune system: No standardised, multi-site-validated functional immune assay exists at all - NK cytotoxicity failed that test, and nothing has replaced it.
Immune system: Why antibody-removal trials fail while antibody-transfer experiments work is unexplained and untested (tissue-bound IgG, long-lived plasma cells, irreversible downstream damage are all untested explanations).
Immune system: The daratumumab signal has no randomised test yet, and no second study of its proposed NK-count predictor.
Immune system: Mast cell activation in ME/CFS has no objective mediator study with controls and no controlled treatment trial, despite being common and treatable in principle.
Immune system: Pre-analytic standardisation is still not routine: after the TGF-beta centrifugation artefact and the SomaScan cell-leak artefact, immune studies in this field still rarely report or adjust for handling variables.
Immune system: Tissue immunology is almost untouched: muscle, gut, dorsal root ganglia and brainstem immune cells have barely been examined in ME/CFS, while the long COVID IgG studies point precisely at peripheral sensory ganglia.
Infection and viruses: No adequately powered, replicated antiviral RCT in ME/CFS (valganciclovir n=30 in 2013 is still the best evidence).
Infection and viruses: No LC antiviral trial has enrolled people selected by biomarker evidence of persistence (antigenemia or tissue RNA).
Infection and viruses: No prospective trial of early antiviral treatment during infectious mononucleosis to prevent post-IM ME/CFS.
Infection and viruses: Blinded, multi-lab, sequencing-based tissue studies (gut, muscle, brain) for enterovirus, herpesvirus and SARS-CoV-2 in non-COVID ME/CFS are missing.
Infection and viruses: Longitudinal viral sampling timed to PEM crashes (plasma EBV/HHV-6 DNA, dUTPase, EA-D) has barely been done.
Infection and viruses: DecodeME iciHHV-6 prevalence results are not yet published.
Infection and viruses: Independent replication of dUTPase antibody findings outside the originating lab.
Infection and viruses: Whether post-COVID, post-EBV, post-Giardia and post-Q-fever ME/CFS share the same downstream biology.
Metabolism and blood omics: No multi-omic study in severe or housebound patients with home sampling.
Metabolism and blood omics: No dense sampling across a PEM episode (48-72 h and beyond); exertion studies stop at 24 h.
Metabolism and blood omics: Very few studies include disease controls (fibromyalgia, depression, MS, deconditioning, long COVID) needed to test specificity.
Metabolism and blood omics: Drug exposure, caffeine, fasting status and sample processing time are rarely recorded or adjusted, despite large effects on metabolomics and proteomics.
Metabolism and blood omics: No pre-registered, multi-site external validation of any diagnostic classifier (nanoneedle, Raman, cfRNA, BioMapAI).
Metabolism and blood omics: Targeted confirmation of low aromatic lactates (with D/L isomers) and high collagen breakdown in fasting, drug-free samples.
Metabolism and blood omics: Lean mass not measured alongside the low muscle-protein signal.
Metabolism and blood omics: Fresh-cell functional assays (PBMC respiration, NK cytotoxicity, red-cell deformability) are not paired with omics in the same people.
Metabolism and blood omics: Only one lab's PBMC single-cell RNA data exist; no independent scRNA cohort.
Metabolism and blood omics: Intracellular NAD+/NADH and tryptophan flux have not been measured directly in patient tissue.
Hormones: No controlled copeptin study with a standard stimulus (water deprivation, hypertonic saline or arginine) in PEM-defined ME/CFS.
Hormones: No study pairs measured blood volume with supine/upright renin, aldosterone and copeptin in the same patients.
Hormones: No longitudinal study shows whether low cortisol precedes illness, as it could in post-infection cohorts.
Hormones: No modern RCT of any hormone therapy using PEM-based criteria or biomarker selection, for example hydrocortisone only in Synacthen low-responders, or desmopressin with copeptin stratification.
Hormones: There are no trials of HRT or menstrual-cycle-linked interventions, and no prospective data on onset around perimenopause.
Hormones: Low T3 has not been replicated and never tested as a treatment target.
Hormones: Prevalence of true adrenal insufficiency or hypopituitarism among clinic ME/CFS patients is unknown.
Gut and microbiome: No prospective study samples stool before or at the triggering infection and follows people into ME/CFS, outside the hospitalised COVID cohort of Liu 2022.
Gut and microbiome: No microbiome study of ME/CFS uses disease controls (IBS without ME/CFS, depression, MS, deconditioned controls) to test specificity.
Gut and microbiome: No direct intestinal permeability measurement (lactulose/mannitol or similar) in a powered ME/CFS cohort; no study of translocation markers before and after exertion.
Gut and microbiome: Results of the 80-person Comeback FMT RCT and the RESTORE-ME virome work are not yet published.
Gut and microbiome: No intervention trial selects patients by microbiome signature (e.g. butyrate-producer loss) or tests fibre/polyphenols to raise IPA.
Gut and microbiome: Small-intestinal microbiota (which stool does not sample) has not been studied, though it could explain the aromatic-lactate deficit.
Gut and microbiome: Within-person longitudinal sampling is needed to test whether microbiome shifts track PEM or relapses.
Genetics: DecodeME is still a preprint; peer review and independent locus-level replication in a PEM-phenotyped cohort are needed.
Genetics: No GWAS of PEM severity or of illness course; DecodeME questionnaires would allow one.
Genetics: No recall-by-genotype studies (e.g. OLFM4 or CA10 risk homozygotes) to link variants to measurable biology.
Genetics: Non-European ancestry is almost absent from ME/CFS genetics.
Genetics: HLA needs joint imputation or sequencing of cases and controls; KIR findings need replication.
Genetics: No independent, blinded replication of TRPM3 functional or SNP findings outside the originating consortium.
Genetics: Hypermobility prevalence in population-based ME/CFS with blinded assessment is unknown, and genetic studies of hypermobility-ME/CFS overlap have not been done.
Genetics: Rare-variant evidence rests on one deep-learning study of about 1,000 genomes; larger sequencing studies are needed.
Treatment trials and natural history: No completed randomised trial of daratumumab, low-dose rapamycin (ME/CFS), aripiprazole or IV saline.
Treatment trials and natural history: No trial designed for severe or very severe (bedbound) patients with home-based objective endpoints.
Treatment trials and natural history: Almost no trial pre-specifies a biological selection marker; subgroup claims (NK cells, HLA, preload failure, illness duration) are all post hoc.
Treatment trials and natural history: Objective outcomes (actigraphy, 2-day CPET, step counts) are rarely primary endpoints, leaving self-report vulnerable to placebo response.
Treatment trials and natural history: Several key trials remain unpublished (BLOC/BC007, IA-PACS-CFS, Nacul LDN, RECOVER-AUTONOMIC full results, LIFT).
Treatment trials and natural history: Long-term natural history of PEM-defined, adult-onset ME/CFS from onset (including post-COVID) beyond two years.
Treatment trials and natural history: Systematic counting of deaths and harms (including rehab-induced deterioration) in ME/CFS.

Study index

Every study cited above. Tap a column to sort.

Study	n	Design	Area	Finding	Direction
Abujrais 2024, ACS Chem Neurosci	38 vs 24	case-control, targeted LC-MS	Metabolism and blood omics	The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood	contradicts
Albright 2011, BMC Neurology	Utah genealogy-linked health records	familial aggregation	Genetics	ME/CFS aggregates in families: relatives of cases have about 2-3x the risk, MZ twins are m	supports
Alharshawi 2022, J Med Virol	255 healthy	cross-sectional healthy cohorts	Infection and viruses	Antibodies against herpesvirus dUTPases (EBV, HHV-6, VZV), proposed as a marker of lytic o	contradicts
Allain 1997, Biol Psychiatry	small case-control	case-control with insulin tolerance test	Hormones	Reports of low IGF-1 and reduced growth hormone responses in CFS conflict, and the largest	supports
Alter 2012, mBio	147 vs 146	blinded multicenter	Infection and viruses	The 2009 report of the mouse retrovirus XMRV in 67% of CFS patients was a laboratory conta	supports
Anft 2025, Molecular Therapy	12	observational, pre/post	Immune system	Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against	supports
Appelman 2024, Nat Commun data	25 LC vs 21 recovered	longitudinal case-control, biopsy pre/post exercise	Blood flow, vessels and blood	Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per f	supports
Appelman 2024, Nat Commun data	25 vs 21	NIRS during CPET	Muscle, exertion and energy	Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (prel	supports
Appelman 2024, Nat Commun data	25 long COVID vs 21 controls	longitudinal case-control; vastus lateralis biopsies before and 1 day after a maximal CPET	Muscle, exertion and energy	In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed	supports
Appelman 2024, Nat Commun data	25 long COVID vs 21 controls	longitudinal case-control; vastus lateralis biopsies before and 1 day after a maximal CPET	Muscle, exertion and energy	ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy	supports
Armstrong 2015, Metabolomics data	34 vs 25 (women)	case-control, 1H NMR blood and urine	Metabolism and blood omics	Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, T	supports
Azcue 2025, Eur J Neurol	30 ME/CFS, 30 LC, 30 HC	case-control, QST, Sudoscan, corneal confocal microscopy	Nervous system and brain	Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients fr	supports
Baden 2026, Lancet Infect Dis (RECOVER-VITAL)	959	RCT, 3 arms (15d, 25d, placebo)	Infection and viruses	Three randomised placebo-controlled trials of 15-25 days of nirmatrelvir-ritonavir in esta	supports
Baden 2026, Lancet Infect Dis (RECOVER-VITAL)	959	phase 2 double-blind RCT, 69 US sites	Treatment trials and natural history	Three randomised trials of nirmatrelvir-ritonavir (Paxlovid) for 15-25 days in established	supports
Bader 2026, medRxiv (preprint)	26 vs 27	case-control volumetry + ASL + MRS under hypoxia	Nervous system and brain	Brainstem differences (smaller volume, altered signal or connectivity) are among the more	supports
Bader 2026, medRxiv (preprint)	26 vs 27	case-control MRS with hypoxic challenge	Nervous system and brain	Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in	supports
Bader 2026, medRxiv (preprint)	26 vs 27	ASL under normoxia/hypoxia	Nervous system and brain	Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS s	contradicts
Baillie 2024, Med	166 LC vs 79 recovered	case-control	Blood flow, vessels and blood	Active long COVID shows persistent complement dysregulation (terminal complex, alternative	supports
Baraniuk 2005, BMC Neurol	10 CFS/10 GWI/10 HC pooled; 9 vs 12	pilot proteomics	Nervous system and brain	CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct	supports
Baraniuk 2025, Sci Rep data	ME/CFS vs sedentary controls, rest and post-exercise cohorts	case-control targeted metabolomics	Nervous system and brain	Targeted CSF metabolomics found raised serine and phospholipids and reduced 5-methyltetrah	supports
Baraniuk, Eaton-Fitch & Marshall-Gradisnik 2024, Frontiers in Immunology	28 papers, 55 E:T data points	meta-analysis	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	supports
Barnes 1993, J Neurol Neurosurg Psychiatry	46 CFS vs controls and mitochondrial myopathy	31P-MRS	Muscle, exertion and energy	31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower	contradicts
Barron 2002, J Pediatr	58 vs 58	matched case-control	Genetics	Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4	supports
Bastos 2025, J Immunol	40 cases (CSF), 41 HC (plasma)	cases-only CSF clustering	Nervous system and brain	CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct	supports
Bedard 2026, Neuroimage Clin	15 vs 19	case-control multimodal EEG/EMG/fMRI	Nervous system and brain	During a sustained grip task, ME/CFS patients fatigue earlier while their brain motor-netw	supports
Beentjes 2025, EMBO Mol Med data	up to 1,455 cases vs 131,303 controls (UK Biobank); All of Us replication	population cohort, 3,237 blood traits, mediation analysis	Metabolism and blood omics	In UK Biobank, ME/CFS cases show many small blood-trait differences (inflammation, insulin	supports
Beentjes 2025, EMBO Mol Med data	up to 1,455 cases vs 131,303 controls (UK Biobank); All of Us replication	population cohort, 3,237 blood traits, mediation analysis	Metabolism and blood omics	Hepatocyte enzymes (ADH4, GSTA1, AGXT, BCHE) are modestly higher, and vascular mural/matri	supports
Beentjes 2025, EMBO Mol Med data	up to 1,455 cases vs 131,303 controls (UK Biobank); All of Us replication	population cohort, 3,237 blood traits, mediation analysis	Metabolism and blood omics	Even the deepest blood omics studies sample mainly at rest, in mild-to-moderate patients,	supports
Beentjes/Ponting 2025, EMBO Molecular Medicine (via Solving ME/CFS re-analysis)	1,455 cases vs 131,303 controls	UK Biobank cross-sectional, 3,237 traits incl. 2,923 proteins	Immune system	Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF	supports
Behan 1991, Acta Neuropathol	about 50	electron microscopy	Muscle, exertion and energy	Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without t	supports
Berg 1999, Blood Coagul Fibrinolysis	small CFS sample	case-control	Blood flow, vessels and blood	Routine coagulation markers (D-dimer, fibrinogen, PT/APTT) are generally normal in ME/CFS;	supports
Berlin Cures BLOC phase II (NCT05911009), company announcement November 2024	~114 across 19 centres	double-blind placebo-controlled phase II	Immune system	BC007 (rovunaptabin), an aptamer that neutralises GPCR autoantibodies, showed secondary-en	contradicts
Berlin Cures BLOC trial topline, 2024 (NCT05911009)	~114	phase II double-blind RCT	Treatment trials and natural history	The aptamer BC007 (rovunaptabin) failed to beat placebo in the company's 114-person long-C	contradicts
Bertinat 2022, Vascul Pharmacol	11 ME/CFS vs 12 HC plasma	in vitro HUVEC	Blood flow, vessels and blood	Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitri	supports
Berwaerts 1998, Growth Horm IGF Res	20 vs 20	case-control	Hormones	Reports of low IGF-1 and reduced growth hormone responses in CFS conflict, and the largest	supports
Bhattacharjee 2026, medRxiv (PAX LC exploratory)	82	RCT exploratory biomarker analysis	Infection and viruses	SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVI	contradicts
Bizjak 2024, Int J Mol Sci	small CFS, post-COVID and control groups	high-resolution respirometry and EM of muscle	Muscle, exertion and energy	ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy	supports
Blockmans 2003, Am J Med	100 (80 completed)	RCT crossover, hydrocortisone + fludrocortisone	Hormones	Low-dose hydrocortisone produced at most small, short-term fatigue benefits in two RCTs, a	contradicts
Blomberg 2019, Front Immunol	several Swedish cohorts vs blood donors	case-control multiplex serology	Infection and viruses	In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and	supports
Blundell 2015, Brain Behavior and Immunity	38 papers, 77 cytokines	systematic review	Immune system	Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF	supports
Boneva 2010, J Womens Health	36 vs 48	population-based case-control	Hormones	Women with CFS report more endometriosis, heavy bleeding, pelvic pain, hysterectomy and ea	supports
Boneva 2015, Menopause	84 vs 73	population-based case-control	Hormones	Women with CFS report more endometriosis, heavy bleeding, pelvic pain, hysterectomy and ea	supports
Bos 2025, Psychol Med	153,803	three-generation cohort familiality	Genetics	ME/CFS aggregates in families: relatives of cases have about 2-3x the risk, MZ twins are m	supports
Bowles 1993, J Med	158 postviral fatigue vs 152 control muscles	molecular hybridisation	Muscle, exertion and energy	Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without t	supports
Bowles 1993, J Med	158 PFS vs 152 controls	muscle biopsy hybridisation	Infection and viruses	Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20%	supports
Bradley 2016, Eur Psychiatry	23 MDD vs 29 HC	case-control MRSI	Nervous system and brain	Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in	contradicts
Bragee 2026, Sci Rep data	31 patients, no controls	CSF LC-MS proteomics	Metabolism and blood omics	CSF proteomics in ME/CFS is limited to small cohorts and has not produced a replicated pro	supports
Bragée 2020, Front Neurol	229 (205 brain MRI, 125 c-spine MRI)	retrospective uncontrolled series	Nervous system and brain	Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying to	supports
Bragée 2020, Front Neurol	229	retrospective cross-sectional, no controls	Genetics	Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4	supports
Bragée 2026, Sci Rep data	31 cases	cases-only CSF proteomics	Nervous system and brain	CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct	supports
Brenu 2011, J Transl Med	95 vs 50	case-control	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	supports
Brown 2015, PLOS ONE	small (CFS vs control myoblast donors)	in vitro EPS of cultured myotubes	Muscle, exertion and energy	Muscle cells grown from ME/CFS patients failed to activate AMPK and raise glucose uptake a	supports
Brown 2026, J Transl Med	116 vs 80	case-control	Gut and microbiome	IBS is about seven times more common in ME/CFS (pooled OR 7.20; prevalence about 37%), and	supports
Buchwald 2001, Psychosom Med	146 female twin pairs	classic twin study	Genetics	ME/CFS aggregates in families: relatives of cases have about 2-3x the risk, MZ twins are m	supports
Bynke 2020, Brain Behav Immun Health	48 patients vs 24 controls (CSF in 24/6)	two-cohort validation, plasma and CSF ELISA	Immune system	Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30	supports
Cairns & Hotopf 2005, Occup Med	14 CFS cohorts	systematic review	Treatment trials and natural history	Full recovery from established adult ME/CFS is uncommon: a systematic review of 14 criteri	supports
Camici 2026, Front Cell Infect Microbiol	83 LC, 13 recovered, controls	case-control salivary profile	Hormones	Lower cortisol in long COVID, reported in a 2023 Nature cohort, has not held up in later c	supports
Capuron 2002, Neuropsychopharmacology	40 melanoma patients	randomised paroxetine vs placebo during IFN-alpha	Immune system	In the interferon-alpha fatigue model, higher baseline IL-6 and IL-10 and a larger early c	supports
Cash 2024, Front Neurol (RESTORE ME)	82	double-blind controlled RCT	Treatment trials and natural history	Oxaloacetate 2 g/day reduced Chalder fatigue more than control (−25% vs −10%) in an 82-per	supports
Castro-Marrero 2015, Antioxid Redox Signal	73	8-week RCT CoQ10 + NADH	Metabolism and blood omics	NAD+ depletion is a proposed mechanism but has not been directly established in ME/CFS coh	supports
Castro-Marrero 2015, Antioxid Redox Signal	73	double-blind RCT	Treatment trials and natural history	Two Spanish RCTs of CoQ10 200 mg + NADH 20 mg (n=73, n=207) reported modest reductions in	supports
Castro-Marrero 2021, Antioxidants	50	RCT	Hormones	Melatonin may help the subgroup with delayed melatonin onset, and melatonin plus zinc mode	supports
Castro-Marrero 2021, Nutrients	207 (104 vs 103)	12-week RCT CoQ10 + NADH	Metabolism and blood omics	NAD+ depletion is a proposed mechanism but has not been directly established in ME/CFS coh	supports
Castro-Marrero 2021, Nutrients	207	double-blind RCT	Treatment trials and natural history	Two Spanish RCTs of CoQ10 200 mg + NADH 20 mg (n=73, n=207) reported modest reductions in	supports
Cervia-Hasler 2024, Science	113 COVID-19 (40 LC) + 39 controls	longitudinal proteomics	Blood flow, vessels and blood	Platelet activation findings in ME/CFS are inconsistent: one network reports hyperactivate	supports
Cervia-Hasler 2024, Science	113 COVID-19 (40 LC) + 39 controls	prospective longitudinal proteomics	Blood flow, vessels and blood	Active long COVID shows persistent complement dysregulation (terminal complex, alternative	supports
Cervia-Hasler 2024, Science	268 samples (~113 COVID patients + controls)	longitudinal proteomics cohort	Infection and viruses	In active long COVID, classical complement pathway activation tracked raised antibody titr	supports
Charlton 2025, Trends Endocrinol Metab	review	opinion article from the same group	Muscle, exertion and energy	In long COVID with PEM, muscle biopsies taken one day after a maximal exercise test showed	supports
Charlton 2026, Br J Sports Med	narrative	review/editorial	Muscle, exertion and energy	Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison g	supports
Charlton 2026, Nat Commun	~25 LC, ~24 ME/CFS, 30 HC, 24 bed rest	case-control + bed-rest comparison, vastus lateralis biopsy	Blood flow, vessels and blood	Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per f	supports
Charlton 2026, Nat Commun	19-23 ME/CFS, 24-25 long COVID, 29-30 matched controls, 24 healthy people after 60-day bed rest	case-control biopsies plus bed-rest comparison	Muscle, exertion and energy	On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled a	supports
Charlton 2026, Nat Commun	19-23 ME/CFS, 24-25 long COVID, 29-30 matched controls, 24 healthy people after 60-day bed rest	case-control biopsies plus bed-rest comparison	Muscle, exertion and energy	ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy	supports
Charlton 2026, Nat Commun	19-23 ME/CFS, 24-25 long COVID, 29-30 matched controls, 24 healthy people after 60-day bed rest	case-control biopsies plus bed-rest comparison	Muscle, exertion and energy	Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison g	supports
Che 2022, Int J Mol Sci data	106 vs 91	case-control, targeted + untargeted plasma metabolomics	Metabolism and blood omics	Blood metabolome differences between ME/CFS and controls are real at group level but small	supports
Che 2022, Int J Mol Sci data	106 vs 91	case-control, targeted + untargeted plasma metabolomics	Metabolism and blood omics	Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across sever	supports
Che 2025, npj Metab Health Dis	56 ME/CFS vs 52 controls (47 matched pairs)	plasma metabolomics/proteomics and stimulated immune responses before and 24 h after exercise	Muscle, exertion and energy	Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but onc	supports
Che 2025, npj Metab Health Dis	56 vs 52	multi-omics pre/24 h post exercise	Muscle, exertion and energy	Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the lar	supports
Che/Lipkin 2025, npj Metabolic Health and Disease	56 vs 52 (47 matched pairs)	case-control multi-omics with exercise challenge and ex vivo microbial stimulation	Immune system	Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen af	supports
Chen 2026, Cell Reports Medicine	31 LC patients, 15 HC; 8 mice/group	passive IgG transfer to mice + proteome-wide autoantibody profiling	Immune system	IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve	supports
Chia 2008, J Clin Pathol	165 vs 34	case series with tissue controls	Infection and viruses	Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20%	supports
Chinnappan 2026, Frontiers in Immunology	40 female patients vs controls	case-control serum immunoassay	Immune system	Newly reported elevations of serum IL-11 and MMP-9 in female ME/CFS patients are unreplica	supports
Christopoulos 2025, J Transl Med	118 studies, 9185 participants	systematic review	Nervous system and brain	Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 2	supports
Christopoulos 2025, J Transl Med	26 ME/CFS-only studies	systematic review	Nervous system and brain	Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS s	supports
Chu 2018, PLOS ONE	150 Fukuda CFS	patient-centred cross-sectional survey	Muscle, exertion and energy	PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took	supports
Cleare 1999, Lancet	32 (crossover)	RCT crossover	Hormones	Low-dose hydrocortisone produced at most small, short-term fatigue benefits in two RCTs, a	supports
Cleare 2000, J Clin Endocrinol Metab	37 vs 37	case-control with GHRH and insulin tests	Hormones	Reports of low IGF-1 and reduced growth hormone responses in CFS conflict, and the largest	contradicts
Cliff 2019, Front Immunol	251 vs 107 (+46 MS)	case-control biobank	Infection and viruses	In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and	supports
Cliff 2019, Front Immunol	251 vs 107	serology	Infection and viruses	HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord t	contradicts
Cliff 2019, Frontiers in Immunology	251 ME/CFS (54 severe) vs 107 HC vs 46 MS	case-control, UK ME/CFS Biobank	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	contradicts
Cliff 2019, Frontiers in Immunology	251 vs 107 vs 46 MS	case-control flow cytometry	Immune system	Increased mucosal-associated invariant T (MAIT) cell proportions and a shift in CD8+ memor	supports
Cockshell & Mathias 2010, Psychol Med	50 studies	meta-analysis	Nervous system and brain	Meta-analysis shows moderate-to-large deficits in information-processing speed, attention,	supports
Cook 2017, Brain Behav Immun	15 vs 15	pre/post exercise fMRI case-control	Nervous system and brain	After a single 30-minute exercise bout, women with ME/CFS performed worse on a demanding c	supports
Corbitt 2019, BMC Neurology	15 included studies	systematic review of cytokines	Immune system	Plasma cytokine abnormalities depend strongly on illness duration: a pro- and anti-inflamm	contradicts
Corbitt 2019, BMC Neurology	15 studies	systematic review	Immune system	Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF	contradicts
Cornell thesis 2022 (Hanson lab)	small	EV-targeted RNA-seq	Infection and viruses	Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20%	contradicts
Crosby 2021, J Transl Med	101	retrospective single-clinic	Treatment trials and natural history	In a retrospective single-clinic review, 74% of 101 patients on very low-dose aripiprazole	supports
Dalhuisen 2024, medRxiv	144 LC vs 56 recovered	cross-sectional, banked plasma	Hormones	Lower cortisol in long COVID, reported in a 2023 Nature cohort, has not held up in later c	contradicts
Dalton 2024, medRxiv	long COVID vs controls	case-control, semi-automated counting	Blood flow, vessels and blood	Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be	supports
Das 2022, J Transl Med	2,382 cases	combinatorial genetics (UKB)	Genetics	Proprietary combinatorial SNP analyses (PrecisionLife) report thousands of multi-SNP signa	supports
Davenport 2025, Nat Commun (Matters Arising)	n/a	critique	Nervous system and brain	The NIH claim that altered 'effort preference' defines post-infectious ME/CFS rests on the	contradicts
Davenport 2026, Research Square preprint	84 ME/CFS, 79 long COVID, 71 controls	retrospective 2-day CPET	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
De Meirleir 2018, J Transl Med	140 vs 140	retrospective case-control	Gut and microbiome	Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to ente	supports
DecodeME 2025 + Solving ME/CFS re-analysis (genetics, downloads) data	15,579 cases vs 259,909 controls	GWAS with MAGMA and S-LDSC enrichment	Immune system	Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type,	supports
DecodeME 2025, medRxiv data	15,579 cases	GWAS + MAGMA	Genetics	Genome-wide ME/CFS heritability is enriched in brain tissues and neurons (including glutam	supports
DecodeME 2025, medRxiv data	15,579 cases	GWAS colocalisation	Genetics	ME/CFS genetics overlaps most with chronic pain, fibromyalgia, IBS and migraine, and subst	supports
DecodeME 2025, medRxiv data	9,738 infection-onset cases	stratified GWAS	Genetics	Three DecodeME loci lie near infection-response genes (OLFM4, RABGAP1L, BTN2A2/TRIM38), an	supports
DecodeME 2025, medRxiv data	15,579 vs 259,909	HLA imputation	Genetics	Reported HLA associations are inconsistent: Lande 2020's HLA-C07:04 and DQB103:03 risk a	supports
DecodeME 2025, medRxiv data	15,579 cases	GWAS	Genetics	TRPM3 ion channel dysfunction in NK cells and TRPM3 SNPs are reported repeatedly by one re	contradicts
DecodeME 2025, medRxiv preprint data	15,579 cases	GWAS	Hormones	ME/CFS and long COVID are more common in women (ME/CFS prevalence about 1.4% vs 0.9%; Deco	supports
DecodeME Genetics Delivery Team 2025, medRxiv data	15,579 vs 259,909	GWAS (European ancestry)	Genetics	DecodeME (15,579 PEM-defined cases vs 259,909 controls) identified 8 genome-wide significa	supports
DecodeME iciHHV-6 sub-study (protocol)	~4,500 planned	genetic cohort sub-study	Infection and viruses	Inherited chromosomally integrated HHV-6 (iciHHV-6, ~1% of people) has been reported at ro	supports
Dehhaghi 2022, Aging Dis	review	narrative review	Metabolism and blood omics	NAD+ depletion is a proposed mechanism but has not been directly established in ME/CFS coh	supports
Dehlia 2024, J Infect	1,973 (13 studies)	meta-analysis	Infection and viruses	In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after S	supports
Dibble 2020, Hum Mol Genet	review incl. UKB	critical review	Genetics	Before DecodeME, small GWAS and candidate-gene studies (GRIK2, NPAS2, TPPP, TRA locus and	supports
Dibble 2024, BMC Research Notes	4 groups x 40	bulk TCR repertoire sequencing + machine learning	Immune system	T-cell receptor repertoires do not distinguish ME/CFS from controls, arguing against a dom	supports
Eaton-Fitch 2019, Systematic Reviews	17 studies	systematic review	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	supports
Eaton-Fitch 2022, J Transl Med	small case-control	calcium imaging	Immune system	Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been report	supports
Eckey 2025, PNAS	~3,925	patient survey	Treatment trials and natural history	NICE's 2021 guideline (NG206) told clinicians not to offer graded exercise therapy, and pa	supports
Eckey 2025, PNAS	~3,925	patient survey	Treatment trials and natural history	Pacing (staying within one's energy limits) is the top-rated approach in patient surveys a	supports
Eckey 2025, PNAS	~3,925	online patient survey	Treatment trials and natural history	A 3,925-person survey rated >150 treatments; illness severity predicted perceived response	supports
Eckey 2025, PNAS	~3,925	patient survey	Treatment trials and natural history	Roughly a quarter of people with ME/CFS are estimated to be housebound or bedbound, and se	supports
Edwards 2023, Front Neurol	1 (follow-up of 3-case series)	case report	Nervous system and brain	Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying to	supports
Eguchi 2026, J Transl Med	6 ME/CFS, 6 ICF, 8 depression, 4 HC	EV miRNA profiling	Metabolism and blood omics	Plasma extracellular vesicle (EV) counts may be higher in ME/CFS, but no EV protein or mic	supports
Esfandyarpour 2019, PNAS	20 vs 20	case-control, nanoelectronics impedance	Metabolism and blood omics	A nanoelectronic impedance assay separated 20 ME/CFS from 20 controls perfectly after hype	supports
Esteban 2026, MicrobiologyOpen	19 vs 23	case-control, targeted stool metabolomics + AHR assay	Gut and microbiome	The replicated deficit of aromatic lactates (indolelactate, phenyllactate, 4-hydroxyphenyl	supports
Falco 2026, Front Hum Neurosci	31 vs 42	retrospective case-control histology	Nervous system and brain	Long COVID patients referred for neuropathy evaluation often have small-fibre neuropathy o	supports
Farztdinov 2024, preprint	29 vs 56	case-control	Blood flow, vessels and blood	Active long COVID shows persistent complement dysregulation (terminal complex, alternative	contradicts
Fernandez 2023, Authorea preprint	4 vs 4	bioimpedance spectroscopy, different device	Metabolism and blood omics	A nanoelectronic impedance assay separated 20 ME/CFS from 20 controls perfectly after hype	contradicts
Fernandez-Guerra 2021, Biomolecules	6 vs 6	PBMC bioenergetics and proteomics	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	contradicts
Flaskamp 2022, Cells	PCS-ME/CFS, PCS, HC sera	in vitro HMEC-1	Blood flow, vessels and blood	Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitri	supports
Flaskamp 2022, Cells	PCS-ME/CFS vs PCS vs HC sera	in vitro tube formation	Blood flow, vessels and blood	VEGF-A has been reported lower, not higher, in ME/CFS plasma, and proteomic re-analysis fi	supports
Fluge 2011, PLOS ONE	30 randomised	double-blind placebo-controlled phase II	Immune system	B-cell depletion with rituximab failed in a 151-patient placebo-controlled trial, after po	supports
Fluge 2015, PLoS One (KTS-2-2010)	29	open-label phase II with maintenance	Treatment trials and natural history	B-cell depletion with rituximab did not improve fatigue or function versus placebo in a 15	contradicts
Fluge 2016, JCI Insight	200 ME/CFS vs 102 controls (serum amino acids); 12 vs 12 (myoblast serum assay)	case-control metabolomics, PBMC qPCR, serum-transfer cell assay	Muscle, exertion and energy	Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling a	supports
Fluge 2016, JCI Insight	12 vs 12 (9 severe)	serum on healthy myoblasts	Muscle, exertion and energy	The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pr	supports
Fluge 2016, JCI Insight	200 vs 102	case-control, serum amino acids + PBMC mRNA	Metabolism and blood omics	Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, T	supports
Fluge 2019 (RituxME), Annals of Internal Medicine	151 randomised (77 vs 74)	multicentre double-blind placebo-controlled RCT	Immune system	B-cell depletion with rituximab failed in a 151-patient placebo-controlled trial, after po	supports
Fluge 2019 (RituxME), Annals of Internal Medicine	151	RCT	Immune system	Plasma-cell depletion with daratumumab produced marked, in some cases sustained, improveme	contradicts
Fluge 2019, Ann Intern Med	151	RCT	Treatment trials and natural history	Placebo and natural-course improvement in ME/CFS and long-COVID trials is large (35% respo	supports
Fluge 2019, Ann Intern Med (RituxME)	77 vs 74	phase III double-blind RCT, 5 Norwegian centres	Treatment trials and natural history	B-cell depletion with rituximab did not improve fatigue or function versus placebo in a 15	supports
Fluge 2025, Front Med	10	open-label pilot (EudraCT 2022-000281-18)	Treatment trials and natural history	In an open-label pilot, 6 of 10 women with moderate-to-severe ME/CFS improved markedly aft	supports
Fluge 2025, Frontiers in Medicine	10 (open-label)	prospective open-label pilot, anti-CD38	Immune system	Plasma-cell depletion with daratumumab produced marked, in some cases sustained, improveme	supports
Franklin & Graham 2022, Fatigue: Biomed Health Behav	4-5 studies per outcome	systematic review and meta-analysis	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
Franklin 2018/2021 meta-analysis (Teesside)	pooled studies	systematic review/meta-analysis	Muscle, exertion and energy	On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled a	supports
Fung 2023, JAMA Intern Med	Medicare older adults	observational cohort	Infection and viruses	Treating acute COVID-19 with nirmatrelvir may reduce later long COVID: a small terminated	contradicts
Gardella 2025, PNAS data	93 vs 75	plasma cfRNA sequencing + ML	Immune system	CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exha	supports
Gardella 2025, PNAS data	93 vs 75	case-control cfRNA sequencing with machine learning	Immune system	Plasma cell-free RNA carries an immune-weighted ME/CFS signature that classifies cases wit	supports
Gardella 2025, PNAS data	93 vs 75	case-control, plasma cfRNA-seq + ML	Metabolism and blood omics	Plasma cell-free RNA classified ME/CFS vs sedentary controls with test-set AUC 0.81 and su	supports
GeNeuro 2024, GNC-501 topline (company release)	203	phase II double-blind RCT	Treatment trials and natural history	The anti-HERV-W envelope antibody temelimab did not improve fatigue versus placebo in 203	supports
Geng 2024, JAMA Intern Med (STOP-PASC)	155	RCT	Infection and viruses	Three randomised placebo-controlled trials of 15-25 days of nirmatrelvir-ritonavir in esta	supports
Geng 2024, JAMA Intern Med (STOP-PASC)	155	double-blind RCT	Treatment trials and natural history	Three randomised trials of nirmatrelvir-ritonavir (Paxlovid) for 15-25 days in established	supports
Germain 2017, Mol BioSyst	17 vs 15	pilot case-control, plasma metabolomics	Metabolism and blood omics	Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts,	supports
Germain 2018, Metabolites data	32 vs 19 (women)	case-control, Metabolon (832 metabolites)	Metabolism and blood omics	Markers of oxidative stress (methionine sulfoxide, oxidised glutathione products, lower br	supports
Germain 2020, Metabolites data	26 vs 26 (women)	case-control, Metabolon + complex lipids (~1,750 compounds)	Metabolism and blood omics	Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts,	supports
Germain 2022, JCI Insight data	60 ME/CFS vs 45 controls, 4 timepoints	2-day CPET plasma metabolomics (1,157 metabolites)	Muscle, exertion and energy	Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but onc	supports
Germain 2022, JCI Insight data	60 vs 45	2-day CPET, 4 time points, Metabolon (1,157 metabolites)	Metabolism and blood omics	Germain 2022 reported growing metabolic divergence after a 2-day CPET; within-person re-an	supports
Germain 2025, Mol Cell Proteomics data	79 ME/CFS vs 53 controls	longitudinal SomaScan 7K across 2-day CPET	Muscle, exertion and energy	Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the lar	supports
Germain 2025, Mol Cell Proteomics data	79 vs 53 (sedentary controls)	2-day CPET, SomaScan 7k, longitudinal	Metabolism and blood omics	Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPE	supports
Germain 2025, Mol Cell Proteomics data	79 vs 53 (sedentary controls)	2-day CPET, SomaScan 7k, longitudinal	Metabolism and blood omics	Even the deepest blood omics studies sample mainly at rest, in mild-to-moderate patients,	supports
Germain, Jaycox, Ring & Hanson 2025, IJMS	172 participants (164 REAP, 103 Luminex)	case-control unbiased antigen profiling (REAP + 1,134-plex)	Immune system	Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30	contradicts
Gile 2026, J Transl Med	responder subset	observational mechanistic follow-up	Treatment trials and natural history	Low-dose rapamycin (6 mg weekly) was followed by self-reported improvement in about three-	supports
Giloteaux 2016, Am J Case Rep	1 MZ twin pair	case report	Gut and microbiome	Quadram Institute studies of severe ME/CFS with household-matched controls report altered	supports
Giloteaux 2016, Microbiome	48 vs 39	case-control, 16S	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	supports
Giloteaux 2016, Microbiome	48 vs 39	case-control	Gut and microbiome	Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to ente	supports
Giloteaux 2024, J Extracell Vesicles data	18 vs 17 women	EV proteomics pre/15 min/24 h post CPET	Muscle, exertion and energy	Extracellular-vesicle protein cargo responds to maximal exercise differently in ME/CFS (bl	supports
Giloteaux 2024, J Extracell Vesicles data	18 vs 17 (women)	CPET, EV TMT proteomics, 3 time points	Metabolism and blood omics	Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPE	supports
Giménez-Orenga 2025, eLife	43	case-control microarray	Infection and viruses	Human endogenous retrovirus (HERV) expression in blood cells differs between ME/CFS, fibro	supports
Glass 2025, Clin Transl Med	10 vs 12 men	EV proteomics pre/15 min/24 h post CPET	Muscle, exertion and energy	Extracellular-vesicle protein cargo responds to maximal exercise differently in ME/CFS (bl	supports
Glass 2025, Clin Transl Med	10 vs 12 (men)	CPET, EV proteomics	Metabolism and blood omics	Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPE	supports
Godlewska 2025, Mol Psychiatry	24 ME/CFS, 25 long COVID, 24 controls	7 T MRS brain and calf muscle at rest	Muscle, exertion and energy	31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower	contradicts
Gold 2021, Pathogens	30 LC vs 20 controls (+18)	cross-sectional	Infection and viruses	Serologic evidence of recent EBV reactivation (EA-D IgG) is associated with long COVID fat	supports
Grau 1992, Clin Neuropathol	20	histopathology	Muscle, exertion and energy	Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without t	supports
Groven 2021, Psychoneuroendocrinology	49 CFS, 57 FM, 54 HC (women)	case-control, plasma KP metabolites	Metabolism and blood omics	The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood	contradicts
Guenther 2015, Human Immunology	300 + 168 patients	retrospective clinic series vs historical controls	Immune system	A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficie	supports
Guo 2023, Cell Host Microbe data	106 vs 91	case-control, shotgun metagenomics + SCFA metabolomics	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	supports
Guo 2023, Cell Host Microbe data	106 vs 91	case-control	Gut and microbiome	Links between specific gut bacteria and symptom severity are inconsistent: F. prausnitzii	supports
Guo 2023, Cell Host Microbe data	106 vs 91 + validation	case-control classifier	Gut and microbiome	Microbiome-based machine-learning classifiers distinguish ME/CFS from controls in held-out	supports
Habermann-Horstmeier 2026, J Transl Med	survey cohort	cross-sectional factor analysis	Hormones	Women with CFS report more endometriosis, heavy bleeding, pelvic pain, hysterectomy and ea	supports
Hackel / Sotzny 2025, medRxiv	post-COVID with and without ME/CFS	IgG fractions applied to monocyte cell lines	Immune system	Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human en	supports
Hadanny 2024, Sci Rep	31	1-year follow-up of RCT HBOT arm	Treatment trials and natural history	Forty daily sessions of hyperbaric oxygen at 2.0 ATA improved cognition and symptoms versu	supports
Haffke 2022, J Transl Med	14 PCS-ME/CFS, 16 PCS, HC	case-control, EndoPAT + biomarkers	Blood flow, vessels and blood	Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia a	supports
Haffke 2022, J Transl Med	14 PCS-ME/CFS, 16 PCS, HC	case-control ELISA	Blood flow, vessels and blood	Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoac	supports
Hajdarevic 2022, Brain Behav Immun	2,532 cases total	GWAS with replication	Genetics	Before DecodeME, small GWAS and candidate-gene studies (GRIK2, NPAS2, TPPP, TRA locus and	supports
Hall 2022, Circulation (POTS)	116 POTS vs 81 HC	case-control, commercial ELISA, standard methodology	Immune system	Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30	contradicts
Halpin 2017, J Med Virol	ME/CFS and GWI cohorts vs controls	case-control serology	Infection and viruses	Antibodies against herpesvirus dUTPases (EBV, HHV-6, VZV), proposed as a marker of lytic o	supports
Hansen 2023, eClinicalMedicine	121	crossover RCT, long COVID	Treatment trials and natural history	Two Spanish RCTs of CoQ10 200 mg + NADH 20 mg (n=73, n=207) reported modest reductions in	contradicts
Hanson 2016, J Transl Med (reply on Billing-Ross 2016)	Cornell cohorts	mtDNA sequencing case-control	Genetics	Mitochondrial DNA haplogroups and common mtDNA SNPs are not associated with ME/CFS status,	supports
Heidarifard 2026, Int J Mol Sci	115 vs 45	plasma Raman + ML, rest and post-stress	Metabolism and blood omics	Raman spectroscopy of PBMCs or plasma separates ME/CFS from controls in single-centre stud	supports
Helliwell 2020, Clin Epigenetics	10 vs 10	RRBS	Metabolism and blood omics	Reported genome-wide blood DNA methylation differences have not replicated across cohorts,	supports
Hickie 2006, BMJ	253 followed 12 months	prospective post-infection inception cohort	Immune system	Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes	supports
Hickie 2006, BMJ	253	prospective cohort from acute infection	Infection and viruses	About 11-12% of people with confirmed EBV, Q fever or Ross River virus infection meet CFS	supports
Hodges 2018, Clin Physiol Funct Imaging	10 CFS, 7 MS, 17 controls	pilot 2-day CPET with disease control	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
Hoel 2021, JCI Insight hoel-2021 hoel-2021-perperson	83 vs 35	case-control, serum metabolomics, lipidomics, hormones	Metabolism and blood omics	Naviaux 2016's picture of a homogeneous 'hypometabolic' dauer-like state (80% of discrimin	contradicts
Hoel 2021, JCI Insight hoel-2021 hoel-2021-perperson	83 vs 35	case-control, serum metabolomics, lipidomics, hormones	Metabolism and blood omics	Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, T	supports
Hoel 2026, Cell Rep Med data	50 vs 29	case-control, SomaScan 7k serum proteomics	Metabolism and blood omics	Muscle-derived intracellular proteins (myoglobin, troponins, CA3 and similar) are lower in	supports
Hoel 2026, Cell Rep Med data	50 vs 29	case-control, SomaScan 7k serum proteomics	Metabolism and blood omics	Most of the ~845 FDR-significant serum proteins in Hoel 2026 appear driven by a sample-han	supports
Hohberger 2025 (reCOVer), eClinicalMedicine	30 randomised	exploratory placebo-controlled double-blind RCT with crossover	Immune system	BC007 (rovunaptabin), an aptamer that neutralises GPCR autoantibodies, showed secondary-en	supports
Hohberger 2025, eClinicalMedicine (reCOVer)	30	phase IIa double-blind placebo-controlled crossover RCT	Treatment trials and natural history	The aptamer BC007 (rovunaptabin) failed to beat placebo in the company's 114-person long-C	supports
Hollingsworth 2012, Eur J Clin Invest	CFS vs controls	cardiac MRI	Blood flow, vessels and blood	Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more tha	supports
Hornig 2015, Molecular Psychiatry	32 cases vs 40 MS vs 19 normal	case-control CSF 51-plex	Immune system	Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and	supports
Hornig 2015, Science Advances	52 early vs 246 long-duration vs 348 controls	multicentre case-control, 51-plex immunoassay	Immune system	Plasma cytokine abnormalities depend strongly on illness duration: a pro- and anti-inflamm	supports
Hornig 2016, Mol Psychiatry	32 vs 19 HC (40 MS)	case-control 51-plex	Nervous system and brain	CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct	supports
Hsieh 2023, Int J Mol Sci	small severe cohort + healthy	virome + metagenome case-control	Gut and microbiome	Quadram Institute studies of severe ME/CFS with household-matched controls report altered	supports
Huang 2024, Commun Med	1,194 ME/CFS vs 53,009 controls + comorbid groups	UK Biobank NMR metabolomics, association + ML	Metabolism and blood omics	Ketone bodies are associated with ME/CFS in UK Biobank NMR data, but 3-hydroxybutyrate sho	supports
Huang 2024, Commun Med	1,194 ME/CFS (UK Biobank)	NMR + baseline characteristics ML	Metabolism and blood omics	Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-valida	supports
Huhmar 2025 copeptin subset	13	observational	Hormones	After overnight fluid deprivation, plasma vasopressin was undetectable in 82% of 111 Swedi	contradicts
Huhmar 2025, Endocrine Practice	124 (111 vasopressin, 13 copeptin)	observational, uncontrolled	Hormones	After overnight fluid deprivation, plasma vasopressin was undetectable in 82% of 111 Swedi	supports
Hunt 2024, Res Pract Thromb Haemost	review	critical appraisal	Blood flow, vessels and blood	Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be	contradicts
Hurwitz 2009, Clin Sci	56 CFS (30 severe, 26 non-severe) vs 90 controls	case-control, isotope blood volume + echo	Blood flow, vessels and blood	Red cell volume and total blood volume are below predicted in a large minority to about ha	supports
IA-PACS-CFS (Charité), presented 2026	44 vs 21	double-blind sham-controlled RCT	Treatment trials and natural history	Removing antibodies by immunoadsorption was no better than a sham procedure in two control	supports
IOM 2015, Beyond ME/CFS	report	evidence review	Treatment trials and natural history	Roughly a quarter of people with ME/CFS are estimated to be housebound or bedbound, and se	supports
Inderyas 2026, J Transl Med	32 ME/CFS, 19 LC, 27 HC	7T task fMRI	Nervous system and brain	Brainstem differences (smaller volume, altered signal or connectivity) are among the more	supports
Iu 2024, PNAS	28 vs 30 (scRNA/ATAC); 15 vs 11 flow validation	multiomic case-control with exercise provocation	Immune system	CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exha	supports
Jackson & Bruck 2012, J Clin Sleep Med	review	narrative review	Nervous system and brain	Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep	supports
Jason 2006, Health Care Women Int	166 deaths (memorial list)	case series	Treatment trials and natural history	In 2,147 secondary-care CFS patients in London, all-cause and cancer mortality were not ra	contradicts
Jason 2022 (via Solving ME/CFS re-analysis epi summary) data	238 mono cases	prospective cohort, pre-illness baseline	Infection and viruses	After infectious mononucleosis in adolescents, 13%, 7% and 4% met paediatric CFS criteria	supports
Jolley 2026, Front Med	34	uncontrolled case series	Nervous system and brain	Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying to	supports
Jones 2010, J Intern Med	16 vs 8	31P-MRS plantar flexion at 35% MVC plus HRV	Muscle, exertion and energy	31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower	supports
Jones 2012, Eur J Clin Invest	18 vs 12	31P-MRS over three exercise bouts	Muscle, exertion and energy	31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower	supports
Joseph 2021, Chest	160 ME/CFS vs 36 controls	retrospective iCPET + skin biopsy	Blood flow, vessels and blood	During upright invasive exercise testing most ME/CFS patients have abnormally low right-at	supports
Joseph 2021, Chest	160 ME/CFS	iCPET	Blood flow, vessels and blood	A subset of patients with unexplained exertional intolerance cannot extract oxygen normall	supports
Joseph 2021, Chest	160 ME/CFS vs 36 controls	invasive CPET plus skin biopsy, retrospective	Muscle, exertion and energy	Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (prel	supports
Joseph 2021, Chest	160 ME/CFS (36 iCPET controls)	retrospective iCPET + skin biopsy	Nervous system and brain	Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients fr	supports
Joseph 2022, Chest	45 (23 vs 22)	single-dose double-blind RCT between two iCPETs	Treatment trials and natural history	A single 60 mg pyridostigmine dose raised peak oxygen uptake relative to placebo on a repe	supports
Joseph 2022, Chest (pyridostigmine RCT)	45 ME/CFS	RCT, single-dose crossover-style iCPET	Blood flow, vessels and blood	During upright invasive exercise testing most ME/CFS patients have abnormally low right-at	supports
Jäkel 2021, J Transl Med	105 ME/CFS, 18 CRF, 66 controls	repeat maximal HGS (10 pulls x 2, 60 min apart)	Muscle, exertion and energy	Repeated maximal handgrip testing shows lower strength, faster decline across repetitions	supports
Kanjwal 2011, Pacing Clin Electrophysiol	203	retrospective POTS	Treatment trials and natural history	A single 60 mg pyridostigmine dose raised peak oxygen uptake relative to placebo on a repe	contradicts
Kashi, Davis, Phair 2019, Diagnostics	none (model)	mathematical model + variant database search	Metabolism and blood omics	The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood	supports
Kasimir 2022, Front Mol Biosci	3 ME/CFS vs 24 controls	post-mortem tissue FISH	Infection and viruses	HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord t	supports
Katz 2009, Pediatrics	301	prospective cohort after IM	Infection and viruses	After infectious mononucleosis in adolescents, 13%, 7% and 4% met paediatric CFS criteria	supports
Kavyani 2024, Mol Neurobiol	not in abstract	case-control, plasma KP metabolites + cytokines + proteomics	Metabolism and blood omics	The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood	supports
Kedor 2022, Nat Commun	42	prospective observational	Treatment trials and natural history	About 4.5% of SARS-CoV-2-infected adults in RECOVER met IOM ME/CFS criteria (vs 0.6% uninf	supports
Keller 2024, J Transl Med	84 ME/CFS vs 71 sedentary controls (55 VO2-matched pairs)	2-day CPET, cycle ergometer, 24 h apart	Muscle, exertion and energy	On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled a	supports
Keller 2024, J Transl Med	84 ME/CFS vs 71 sedentary controls (55 VO2-matched pairs)	2-day CPET, cycle ergometer, 24 h apart	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
Keller 2024, J Transl Med	84 ME/CFS vs 71 sedentary controls (55 VO2-matched pairs)	2-day CPET, cycle ergometer, 24 h apart	Muscle, exertion and energy	Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison g	supports
Kennedy 2006, Blood Coagul Fibrinolysis	CFS vs controls	case-control	Blood flow, vessels and blood	Platelet activation findings in ME/CFS are inconsistent: one network reports hyperactivate	contradicts
Kennedy 2006, Blood Coagul Fibrinolysis	CFS vs controls	case-control	Blood flow, vessels and blood	Routine coagulation markers (D-dimer, fibrinogen, PT/APTT) are generally normal in ME/CFS;	contradicts
Kenyon 2019, Human Microbiome Journal	42 (21 FMT)	retrospective, non-randomised	Gut and microbiome	The only published randomised faecal microbiota transplant trial in CFS (n=11) found no be	contradicts
Kimura 2026, Front Neurol	38 ME/CFS, 27 LC, 47 HC	case-control volumetry	Nervous system and brain	Conventional structural MRI (grey matter volume, routine clinical reads) shows no consiste	contradicts
Kirvin-Quamme 2025, Front Psychol	15 vs 16 (re-analysis)	secondary analysis of public data	Nervous system and brain	The NIH claim that altered 'effort preference' defines post-infectious ME/CFS rests on the	contradicts
Kjellberg 2025, BMJ Open	80	RCT	Treatment trials and natural history	Placebo and natural-course improvement in ME/CFS and long-COVID trials is large (35% respo	supports
Kjellberg 2025, BMJ Open (HOT-LoCO)	80	double-blind sham RCT, 10 sessions at 2.4 bar	Treatment trials and natural history	Forty daily sessions of hyperbaric oxygen at 2.0 ATA improved cognition and symptoms versu	contradicts
Klein 2023, Nature data	275	cross-sectional immune phenotyping with ML	Immune system	Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to	supports
Klein 2023, Nature data	275 total	cross-sectional immune profiling	Hormones	Lower cortisol in long COVID, reported in a 2023 Nature cohort, has not held up in later c	supports
Klimas 1990, J Clin Microbiol	30 patients vs controls	case-control	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	supports
Kogelnik 2006, J Clin Virol	12	open-label	Treatment trials and natural history	Valganciclovir in 30 patients selected for high HHV-6 and EBV antibody levels improved som	supports
Kruger 2022, Cardiovasc Diabetol	long COVID	proteomics of microclots	Blood flow, vessels and blood	Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be	supports
Kubánková 2021, Biophys J	17 COVID-19, 14 recovered, 24 HC	RT-DC	Blood flow, vessels and blood	Red blood cells from ME/CFS patients were stiffer in one microfluidic study, and real-time	supports
Kumar 2026, Front Med	31 vs 30	case-control ERP	Nervous system and brain	Self-reported light and sound sensitivity is common in ME/CFS. One ERP study found normal	supports
Lammi 2025, Nat Genet (long COVID GWAS)	6,450 cases	GWAS	Genetics	Three DecodeME loci lie near infection-response genes (OLFM4, RABGAP1L, BTN2A2/TRIM38), an	contradicts
Lande 2020, Sci Rep	426 vs 4,511	HLA NGS case-control	Genetics	Reported HLA associations are inconsistent: Lande 2020's HLA-C07:04 and DQB103:03 risk a	supports
Landi 2016, Cytokine	ME/CFS vs controls and liver-disease controls	case-control multiplex	Blood flow, vessels and blood	VEGF-A has been reported lower, not higher, in ME/CFS plasma, and proteomic re-analysis fi	supports
Lane 1998, J Neurol Neurosurg Psychiatry	105 CFS biopsies	biopsy plus sub-anaerobic threshold exercise test	Muscle, exertion and energy	Some patients produce more lactate for a given workload, and this worsens on a repeat test	supports
Lane 1998, J Neurol Neurosurg Psychiatry	105	quadriceps needle biopsy, histometry, EM	Muscle, exertion and energy	Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without t	supports
Lane 1998, J Neurol Neurosurg Psychiatry	105	biopsy histometry	Muscle, exertion and energy	ME/CFS muscle has a lower proportion of type I (oxidative) fibres, type I-specific atrophy	contradicts
Lane 1998, Neuromuscul Disord	10 SATET+ vs 9 SATET- vs 13 controls	forearm 31P-MRS	Muscle, exertion and energy	31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower	supports
Larun 2019/2024, Cochrane (CD003200)	8 RCTs	systematic review	Treatment trials and natural history	PACE (n=641) reported modest gains in self-rated fatigue and function with CBT and graded	contradicts
Laurin 2023, Auton Neurosci	POTS vs HC	ACTH stimulation	Blood flow, vessels and blood	The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not ra	supports
Laurin 2023, Autonomic Neuroscience	8 POTS vs 5	ACTH stimulation, low-sodium diet	Hormones	Challenge tests show a blunted cortisol response to ACTH and exaggerated cortisol suppress	contradicts
Lee 2020, J Transl Med	150 vs 75	case-control	Nervous system and brain	On the 10-minute NASA lean test, a subgroup of recent-onset patients shows a rising heart	supports
Lee 2023, Autoimmun Rev	65 studies; 1529 vs 1715	systematic review and meta-analysis	Nervous system and brain	Conventional structural MRI (grey matter volume, routine clinical reads) shows no consiste	supports
Lee JS 2025, J Transl Med (also medRxiv 2025.01.02.24319359)	43 mild/moderate vs 53 severe	case-case flow cytometry, UK ME/CFS Biobank	Immune system	T-cell abnormalities differ by severity: mild/moderate patients show cytotoxic/immunosenes	supports
Legler 2023, eClinicalMedicine	106	prospective cohort	Treatment trials and natural history	About 4.5% of SARS-CoV-2-infected adults in RECOVER met IOM ME/CFS criteria (vs 0.6% uninf	supports
Leone 2026, Front Pediatr	18 vs 18 adolescents	case-control, 10-min passive stand	Nervous system and brain	POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%	supports
Lerner 2007, In Vivo	small	small RCT phase + 36-month open follow-up	Infection and viruses	In a 30-patient RCT of 6 months of valganciclovir in CFS patients selected for high HHV-6	supports
Lien 2019, Physiol Rep	18 vs 15 women	2-day CPET with arterial lactate	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	contradicts
Lien 2019, Physiol Rep	18 vs 15 women	2-day CPET with arterial lactate every 30 s	Muscle, exertion and energy	Some patients produce more lactate for a given workload, and this worsens on a repeat test	supports
Light 2009, J Pain	small pilot (n from memory)	pilot	Muscle, exertion and energy	After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), a	supports
Light 2012, J Intern Med	48 CFS, 18 FM, 49 controls	pre/post moderate exercise qPCR	Muscle, exertion and energy	After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), a	supports
Lim 2020, J Clin Med	pooled 2-day CPET studies	meta-analysis	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
Lim 2020, J Transl Med	46 studies, about 1.09 million	meta-analysis of prevalence	Hormones	ME/CFS and long COVID are more common in women (ME/CFS prevalence about 1.4% vs 0.9%; Deco	supports
Lim 2020, J Transl Med	46 studies, >1 million participants	meta-analysis	Treatment trials and natural history	ME/CFS affects roughly 0.2-1% of adults depending on criteria (US 2021-22: 1.3% self-repor	supports
Litleskare 2018, Clin Gastroenterol Hepatol	1,252 exposed vs 2,504 controls	prospective matched cohort, 10 years	Infection and viruses	A waterborne Giardia outbreak in Bergen (2004) left chronic fatigue in 26% of exposed peop	supports
Liu / Prusty 2026, Brain Behavior & Immunity - Health	39 ME/CFS, 15 PCS-CFS, 20 MS, 41 HC	in vitro IgG application to endothelial cells with disease control	Immune system	Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human en	supports
Liu 2022, Gut	106 COVID patients, 68 controls	prospective cohort, metagenomics	Gut and microbiome	Ruminococcus gnavus, Flavonifractor plautii, Ruthenibacterium lactatiformans and Enteroclo	supports
Liu 2022, Gut	106 vs 68	prospective cohort	Gut and microbiome	In long COVID, the gut microbiome at hospital admission predicted persistent symptoms, and	supports
Liu/Prusty 2026, Brain Behav Immun Health	39 ME, 15 PCS-ME, 20 MS, 41 HC IgG	in vitro	Blood flow, vessels and blood	Plasma or purified IgG from ME/CFS patients alters cultured endothelial cells: lower nitri	supports
Lloyd 1990, Am J Med	49	RCT	Treatment trials and natural history	IVIG has no consistent controlled evidence: 1990s ME/CFS RCTs conflicted, and a 30-person	supports
Loebel 2014, PLoS One	CFS vs HC (tens per group)	case-control cellular immunology	Infection and viruses	Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (abse	supports
Loebel 2015, Brain Behavior and Immunity	268 patients vs 108 controls (+25 trial patients)	case-control ELISA panel	Immune system	Elevated IgG against beta2-adrenergic and M3/M4 muscarinic receptors, reported in about 30	supports
Loebel 2017, PLoS One	92 vs 50	peptide microarray case-control	Infection and viruses	In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and	supports
Lubell 2026, medRxiv	1,816	retrospective online survey	Genetics	Joint hypermobility and Ehlers-Danlos syndrome are over-represented in ME/CFS (Beighton ≥4	supports
Lutz 2021, Biomolecules	262 patients	retrospective clinic series	Immune system	A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficie	supports
Ma 2026, Cell Rep Med	long COVID vs recovered (tetramer cohort)	case-control CD8 tetramer profiling	Infection and viruses	Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (abse	supports
Maccallini 2026, Research Square preprint	19,470 vs 699,111	GWAS meta-analysis + enrichment	Hormones	ME/CFS genetic risk is enriched in brain tissues and the pituitary, but genetically predic	supports
Maccallini 2026, Research Square preprint	19,470 vs 699,111	DecodeME + MVP meta-analysis	Genetics	The genome-wide DecodeME signal replicates in independent cohorts: genetic correlation 0.6	supports
Maccallini 2026, Research Square preprint	19,470 cases	meta-GWAS enrichment + rare-variant replication	Genetics	Genome-wide ME/CFS heritability is enriched in brain tissues and neurons (including glutam	supports
Maes & Leunis 2008, Neuro Endocrinol Lett	41	uncontrolled before-after	Gut and microbiome	Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to ente	contradicts
Maes 2007, J Affect Disord	CFS, partial CFS and controls	case-control	Gut and microbiome	Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to ente	supports
Magawa 2026, BMC Immunology	case-control	mitochondrial Ca2+ imaging in NK cells	Immune system	Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been report	supports
Magnus 2015, Vaccine	whole Norwegian population	registry cohort	Infection and viruses	In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after S	supports
Mancini 2025, Front Med	58 ME/CFS + sedentary controls	CPET breathing analysis	Blood flow, vessels and blood	Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exerci	supports
Mancini 2026, Front Physiol	58 ME/CFS vs 25 sedentary controls	2-day CPET replication attempt	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	contradicts
Mancini 2026, Front Physiol	58 ME/CFS vs 25 sedentary controls	2-day CPET replication attempt	Muscle, exertion and energy	Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison g	contradicts
Mandarano 2020, J Clin Invest	53 vs 45	Seahorse on isolated CD4/CD8 T cells	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	supports
Mandarano 2020, J Clin Invest	53 vs 45	case-control immunometabolism	Immune system	CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exha	supports
Marshall-Gradisnik 2016, J Int Med Res	11 vs 11	SNP/genotype case-control	Genetics	TRPM3 ion channel dysfunction in NK cells and TRPM3 SNPs are reported repeatedly by one re	supports
Mathew 2009, NMR Biomed	16 CFS, 14 GAD, 15 HC	case-control 1H MRSI	Nervous system and brain	Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in	supports
Maya 2026, Int J Mol Sci	50 ME/CFS vs 121 non-fatigued	population-based case-control + pQTL	Blood flow, vessels and blood	In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetica	supports
McCully 1999, Clin Sci	20 vs 12	NIRS plus 31P-MRS	Muscle, exertion and energy	Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (prel	supports
McKenzie 1998, JAMA	70	RCT	Hormones	Low-dose hydrocortisone produced at most small, short-term fatigue benefits in two RCTs, a	supports
McKenzie 2000 (BMD sub-study of the 1998 trial)	23	RCT sub-study	Hormones	Low-dose hydrocortisone produced at most small, short-term fatigue benefits in two RCTs, a	supports
Medow & Stewart 2024, Am J Physiol Regul	ME/CFS with OI vs controls	acute intervention crossover	Blood flow, vessels and blood	Raising blood pressure with phenylephrine during tilt blocked hyperventilation, restored c	supports
Medow & Stewart 2024, Am J Physiol Regul Integr Comp Physiol	small	physiology	Nervous system and brain	Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of	supports
Medow 2014, J Appl Physiol	CFS with OI vs controls	tilt + n-back	Blood flow, vessels and blood	Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exerci	supports
Medow 2014, J Appl Physiol	15 vs 11	case-control tilt with TCD	Nervous system and brain	Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 2	supports
Medow 2014, J Appl Physiol	15 vs 11	case-control + acute phenylephrine crossover	Nervous system and brain	Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of	supports
Medow 2014, J Appl Physiol	15 vs 11	case-control tilt	Nervous system and brain	Meta-analysis shows moderate-to-large deficits in information-processing speed, attention,	supports
Melamed 2019, Eur J Appl Physiol	unexplained exertional intolerance cohort	retrospective iCPET	Blood flow, vessels and blood	A subset of patients with unexplained exertional intolerance cannot extract oxygen normall	supports
Melamed 2019, Eur J Appl Physiol	313 iCPETs for unexplained exertional intolerance	retrospective iCPET	Muscle, exertion and energy	Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (prel	supports
Mensah 2016, Clin Exp Immunol	cross-sectional case-control	extended B-cell immunophenotyping	Immune system	The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-	supports
Mignolet 2026, Acta Neuropathologica	neurological LC vs asymptomatic matched controls; C57Bl6/J mice	passive IgG transfer, 4 daily i.p. injections	Immune system	IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve	supports
Mignolet 2026, Acta Neuropathologica	as above	passive transfer	Immune system	IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve	contradicts
Missailidis 2020, Int J Mol Sci	51 ME/CFS lymphoblast lines vs matched controls	Seahorse, proteomics	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	supports
Missailidis 2021, Int J Mol Sci	up to 34 vs 31 lines	transcriptomics/proteomics	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	supports
Miwa 2016, J Cardiology	18 vs 15	case-control	Hormones	After overnight fluid deprivation, plasma vasopressin was undetectable in 82% of 111 Swedi	supports
Miwa 2016, J Cardiology	18 vs 15	case-control (ME)	Hormones	Patients with POTS and with ME/CFS show normal or low renin and low aldosterone despite re	supports
Miwa 2026, Front Med	review of Japanese series	narrative review	Blood flow, vessels and blood	Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more tha	supports
Mohamed 2023, Sleep Med Rev	24 studies; 668 vs 610	systematic review and meta-analysis	Nervous system and brain	Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep	supports
Montoya 2013, J Med Virol	30 (20 vs 10)	double-blind RCT	Infection and viruses	In a 30-patient RCT of 6 months of valganciclovir in CFS patients selected for high HHV-6	supports
Montoya 2013, J Med Virol	30	double-blind RCT 2:1	Treatment trials and natural history	Valganciclovir in 30 patients selected for high HHV-6 and EBV antibody levels improved som	supports
Montoya 2017, PNAS data	192 vs 392	case-control, 51-plex Luminex with severity regression	Immune system	In the largest single cytokine study, 17 cytokines rose in step with disease severity alth	supports
Moore 2023, Medicina	80 ME/CFS vs 64 controls	prospective symptom tracking for 10 days after 2-day CPET	Muscle, exertion and energy	PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took	supports
Moorkens 1998, Growth Horm IGF Res	20	RCT of GH	Hormones	Reports of low IGF-1 and reduced growth hormone responses in CFS conflict, and the largest	contradicts
Mueller 2020, Brain Imaging Behav	15 vs 15	case-control whole-brain MRSI	Nervous system and brain	MRS thermometry found raised temperature in several brain regions (insula, putamen, fronta	supports
Mughal 2025, Biofabrication	4 ME/CFS, 5 long COVID, 4 controls	3D engineered muscle exposed to serum	Muscle, exertion and energy	Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling a	supports
Mughal 2025, Biofabrication	4 ME/CFS, 5 long COVID, 4 controls	3D muscle tissue force under electrical stimulation	Muscle, exertion and energy	The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pr	supports
NICE 2021, NG206	guideline	evidence review	Treatment trials and natural history	NICE's 2021 guideline (NG206) told clinicians not to offer graded exercise therapy, and pa	supports
Nacul 2011, BMC Med	3 English regions	primary care cross-sectional	Treatment trials and natural history	ME/CFS affects roughly 0.2-1% of adults depending on criteria (US 2021-22: 1.3% self-repor	supports
Nacul et al., presented May 2026 (UBC; NCT05430152)	66 vs 71	double-blind placebo-controlled RCT, 16 weeks	Treatment trials and natural history	The first sizeable placebo-controlled trial of low-dose naltrexone in post-COVID ME/CFS (6	supports
Naess 2012, BMC Gastroenterol	96 referred	clinical case series	Infection and viruses	A waterborne Giardia outbreak in Bergen (2004) left chronic fatigue in 26% of exposed peop	supports
Nagy-Szakal 2017, Microbiome	50 vs 50	case-control, metagenomics	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	supports
Nagy-Szakal 2017, Microbiome	50 vs 50	case-control	Gut and microbiome	IBS is about seven times more common in ME/CFS (pooled OR 7.20; prevalence about 37%), and	supports
Nagy-Szakal 2018, Sci Rep data	50 vs 50	case-control, untargeted plasma metabolomics + stool metagenomics	Metabolism and blood omics	Blood metabolome differences between ME/CFS and controls are real at group level but small	supports
Nagy-Szakal 2018, Sci Rep data	50 vs 50	case-control, untargeted plasma metabolomics + stool metagenomics	Metabolism and blood omics	Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across sever	contradicts
Nagy-Szakal 2018, Sci Rep data	50 vs 50	case-control, untargeted plasma metabolomics + stool metagenomics	Metabolism and blood omics	Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts,	supports
Nakatomi 2014, J Nucl Med	9 vs 10	case-control PET	Nervous system and brain	TSPO-PET evidence of microglial activation in ME/CFS is contested: one Japanese study foun	supports
Natarajan 2022, Med	113	longitudinal stool cohort	Infection and viruses	SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a m	supports
Natelson 2017, J Neurol Sci	CFS +/- psychiatric comorbidity vs HC	case-control multimodal	Nervous system and brain	Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in	supports
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	Metabolism and blood omics	Blood metabolome differences between ME/CFS and controls are real at group level but small	supports
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	Metabolism and blood omics	Naviaux 2016's picture of a homogeneous 'hypometabolic' dauer-like state (80% of discrimin	supports
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	Metabolism and blood omics	Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across sever	supports
Naviaux 2016, PNAS data	45 vs 39	case-control, targeted LC-MS/MS (612 metabolites)	Metabolism and blood omics	Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-valida	supports
Nehme 2026, Brain Behav Immun	203	double-blind RCT	Infection and viruses	Temelimab, an antibody against HERV-W envelope protein, did not reduce fatigue versus plac	supports
Nelson 2019, Medicine (Baltimore)	64 studies	systematic review and meta-analysis	Nervous system and brain	Resting heart rate is higher and high-frequency heart-rate variability lower in ME/CFS. Th	supports
Newton 2016, Open Heart	CFS vs controls (cardiac MRI + red cell mass)	case-control	Blood flow, vessels and blood	Red cell volume and total blood volume are below predicted in a large minority to about ha	supports
Nguyen 2017, Asian Pac J Allergy Immunol data	18 patients vs 13 HC	flow cytometry of PBMC mast cell phenotypes	Immune system	Clinically relevant mast cell activation is reported in up to a quarter of ME/CFS patients	supports
Novak 2026, PLOS One	170 ME/CFS, 143 long COVID, 73 HC, 290 hEDS	retrospective autonomic + iCPET	Blood flow, vessels and blood	During upright invasive exercise testing most ME/CFS patients have abnormally low right-at	supports
Novak 2026, PLOS One	170 ME/CFS, 143 LC, 73 HC	retrospective TCD tilt	Blood flow, vessels and blood	Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, i	supports
Novak 2026, PLOS One	170 ME/CFS, 143 LC	retrospective	Blood flow, vessels and blood	POTS is present in roughly a fifth of ME/CFS patients on formal testing (prevalence varies	supports
Novak 2026, PLoS One	170 ME/CFS, 143 long COVID, 73 HC, 290 hEDS	retrospective autonomic-lab cohort, TCD + capnography	Nervous system and brain	Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 2	supports
Novak 2026, PLoS One	170 ME/CFS	retrospective autonomic-lab cohort	Nervous system and brain	POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%	supports
Novak 2026, PLoS One	170 ME/CFS, 143 LC	retrospective autonomic-lab cohort	Nervous system and brain	Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients fr	supports
Nunes 2022, Pharmaceuticals	25 ME/CFS vs 15 HC	case-control, fluorescence microscopy + TEG	Blood flow, vessels and blood	Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be	supports
Nunes 2022, Pharmaceuticals	25 vs 15	case-control	Blood flow, vessels and blood	Platelet activation findings in ME/CFS are inconsistent: one network reports hyperactivate	supports
O'Brien 2026, J Clin Med	50 long COVID	clinic cohort, NASA lean test + ETCO2	Nervous system and brain	Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of	supports
O'Neal & Hanson 2021, Front Med (critical review)	review	critical review	Infection and viruses	Enteroviral VP1 protein was found in stomach biopsies from 82% of 165 CFS patients vs 20%	contradicts
OMF metabolomics validation update (MEAction 2018)	not stated	unpublished replication cohort summary	Metabolism and blood omics	Naviaux 2016's picture of a homogeneous 'hypometabolic' dauer-like state (80% of discrimin	supports
Oaklander 2022, Neurol Neuroimmunol Neuroinflamm	17	case series	Nervous system and brain	Long COVID patients referred for neuropathy evaluation often have small-fibre neuropathy o	supports
Olimulder 2016, Neth Heart J	ME/CFS vs controls	cardiac MRI	Blood flow, vessels and blood	Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more tha	contradicts
Oppegaard 2026, Lancet Infect Dis (PANORAMIC Norway)	144	RCT (terminated early)	Infection and viruses	Treating acute COVID-19 with nirmatrelvir may reduce later long COVID: a small terminated	supports
Osiaevi 2024, Nephrol Dial Transplant	long COVID vs controls	case-control, sublingual video-microscopy	Blood flow, vessels and blood	Sublingual video-microscopy shows reduced perfused capillary density and glycocalyx change	supports
Paffrath 2024, J Clin Med	144 post-COVID women (78 ME/CFS)	repeat HGS	Muscle, exertion and energy	Repeated maximal handgrip testing shows lower strength, faster decline across repetitions	supports
Palomo 2026, J Med Virol	40 vs 16 (873 + 378 serial samples)	longitudinal case-control	Infection and viruses	Antibodies against herpesvirus dUTPases (EBV, HHV-6, VZV), proposed as a marker of lytic o	supports
Paul 1999, Eur J Neurol	10 vs 10	quadriceps fatiguing protocol with recovery to 24 h	Muscle, exertion and energy	Repeated maximal handgrip testing shows lower strength, faster decline across repetitions	supports
Peluso 2023, J Clin Invest	280	post-COVID cohort	Infection and viruses	Detectable cell-free EBV DNA in plasma has been reported more often in ME/CFS (24% vs 4% i	contradicts
Peluso 2023, J Clin Invest	280	post-COVID cohort, adjusted regression	Infection and viruses	Serologic evidence of recent EBV reactivation (EA-D IgG) is associated with long COVID fat	supports
Peluso 2024, Lancet Infect Dis	171 vs 250 pre-pandemic	cohort with true-negative comparator	Infection and viruses	SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVI	supports
Peluso 2024, Sci Transl Med	24 (PET) + 5 biopsies	imaging cohort + tissue ISH	Infection and viruses	SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a m	supports
Peterson 1990, Am J Med	30	RCT	Treatment trials and natural history	IVIG has no consistent controlled evidence: 1990s ME/CFS RCTs conflicted, and a 30-person	contradicts
Peterson/Brenu 2015, Mediators of Inflammation	18 patients	pilot CSF cytokine study	Immune system	Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and	supports
Petter 2022, J Transl Med	6 vs 6 women	23Na-MRI lower leg before and after plantar flexion	Muscle, exertion and energy	23Na-MRI in 6 women with ME/CFS and 6 controls found higher resting muscle sodium in all l	supports
Plank 2022, Neuroimage Clin	20 healthy	RCT crossover, typhoid vaccine	Nervous system and brain	MRS thermometry found raised temperature in several brain regions (insula, putamen, fronta	contradicts
Polo 2019, Fatigue: Biomed Health Behav	218	retrospective chart review	Treatment trials and natural history	The first sizeable placebo-controlled trial of low-dose naltrexone in post-COVID ME/CFS (6	contradicts
Quach 2025, Biomedicines	86	retrospective clinic series	Hormones	Lower cortisol in long COVID, reported in a 2023 Nature cohort, has not held up in later c	contradicts
Querec 2023 (MCAM), J Transl Med	174 ME/CFS vs 86 HC vs 10 ill controls	multi-site case-control, shipped-sample validated assay	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	contradicts
RECOVER-AUTONOMIC ivabradine, ACC 2026 (NIH release)	181	double-blind placebo-controlled RCT	Treatment trials and natural history	Ivabradine lowered standing heart rate but did not improve symptoms in 181 people with lon	supports
REGAIN 2025 (long COVID)	69	RCT	Treatment trials and natural history	Oxaloacetate 2 g/day reduced Chalder fatigue more than control (−25% vs −10%) in an 82-per	contradicts
Raijmakers 2019, J Psychosom Res (Qure follow-up) data	154	RCT follow-up (CBT vs doxycycline vs placebo)	Infection and viruses	Q fever fatigue syndrome (QFS) follows roughly 20% of acute Q fever and persists more than	supports
Raijmakers 2019, Journal of Infection (Q fever fatigue syndrome) data	20 QFS, 19 CFS, 19 seropositive controls, 25 HC	case-control ex vivo stimulation	Immune system	Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen af	supports
Raijmakers 2021, Neurol Neuroimmunol Neuroinflamm	9 CFS, 10 QFS, 9 HC	case-control PET	Nervous system and brain	TSPO-PET evidence of microglial activation in ME/CFS is contested: one Japanese study foun	contradicts
Raj 2005, Circulation	POTS vs controls	case-control	Blood flow, vessels and blood	The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not ra	supports
Raj 2005, Circulation	15 vs 14	case-control (POTS)	Hormones	Patients with POTS and with ME/CFS show normal or low renin and low aldosterone despite re	supports
Rajabi 2026, Front Med	18 studies pooled	systematic review and meta-analysis	Gut and microbiome	IBS is about seven times more common in ME/CFS (pooled OR 7.20; prevalence about 37%), and	supports
Ramadan 2025, Brain Behav Immun	418 vs 473	case-control, KIR allele sequencing	Genetics	Specific inhibitory KIR alleles (e.g. KIR3DL1020, KIR3DL2009) were more or less common i	supports
Rao 2009, Gut Pathog	39	RCT	Gut and microbiome	Probiotic trials in CFS are small; one 39-person RCT of Lactobacillus casei Shirota reduce	supports
Rekeland 2020 (CycloME), Frontiers in Medicine	40 (open-label)	open-label phase II cyclophosphamide	Immune system	Plasma-cell depletion with daratumumab produced marked, in some cases sustained, improveme	supports
Rekeland 2020, Front Med	40	open-label phase II	Treatment trials and natural history	In an open-label phase II study, 55% of 40 patients responded to six IV cyclophosphamide i	supports
Rekeland 2024, PLoS One	RituxME 151 + CycloME 40 (75.7% / 94.4% followed)	6-year follow-up of two trials	Treatment trials and natural history	B-cell depletion with rituximab did not improve fatigue or function versus placebo in a 15	supports
Rekeland 2024, PLoS One	36 of 40 followed	6-year follow-up	Treatment trials and natural history	In an open-label phase II study, 55% of 40 patients responded to six IV cyclophosphamide i	supports
Rekeland 2024, PLoS One	placebo arm ~54	6-year follow-up	Treatment trials and natural history	Full recovery from established adult ME/CFS is uncommon: a systematic review of 14 criteri	supports
Retraction notice 2025, Food Sci Nutr	-	retracted two-mediation MR study	Genetics	MR of DecodeME suggests higher mitochondrial DNA copy number is protective and higher Glyc	contradicts
Reuter & Evans 2011, J Intern Med	44 patients	case-control serum carnitine profile	Metabolism and blood omics	Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts,	supports
Roberts 2016, Lancet	2,147	retrospective cohort	Treatment trials and natural history	In 2,147 secondary-care CFS patients in London, all-cause and cancer mortality were not ra	supports
Rodrigues 2019, Auto Immun Highlights	small (moderate + severe vs controls)	case-control qPCR	Infection and viruses	Human endogenous retrovirus (HERV) expression in blood cells differs between ME/CFS, fibro	supports
Roerink 2017, Annals of Internal Medicine / J Transl Med (cytokine arm)	50 randomised	RCT, anakinra vs placebo, 4 weeks	Immune system	Blocking IL-1 with anakinra for four weeks did not improve fatigue in CFS, and did not cha	supports
Roerink 2017, J Translational Medicine	50 vs 48	case-control, 92-marker PEA + ELISA, plus anakinra RCT samples	Immune system	The most-replicated single cytokine finding, elevated TGF-beta, is largely a blood-handlin	supports
Roerink 2017, PNAS (letter) and Roerink 2017, J Transl Med	50 vs 48	critique + independent case-control with 92-marker PEA	Immune system	In the largest single cytokine study, 17 cytokines rose in step with disease severity alth	contradicts
Roerink 2018, Netherlands Journal of Medicine	50 vs 48 and 90 vs 29	two case-control cohorts with pre-analytic comparison	Immune system	The most-replicated single cytokine finding, elevated TGF-beta, is largely a blood-handlin	supports
Rohrhofer 2025, Diagnostics	687 survey + 383 retrospective	survey plus retrospective clinical cohort	Immune system	Clinically relevant mast cell activation is reported in up to a quarter of ME/CFS patients	supports
Ross 2014, Am J Physiol (midodrine in POTS)	small	acute crossover	Treatment trials and natural history	Fludrocortisone did not beat placebo in a 100-person RCT in CFS with neurally mediated hyp	contradicts
Rowe 2001, JAMA	100	RCT	Hormones	Fludrocortisone, a synthetic salt-retaining steroid, did not improve CFS with neurally med	supports
Rowe 2001, JAMA	100	double-blind RCT	Treatment trials and natural history	Fludrocortisone did not beat placebo in a 100-person RCT in CFS with neurally mediated hyp	supports
Ruan 2025, J Transl Med	86 (70 assessed)	decentralised open-label	Treatment trials and natural history	Low-dose rapamycin (6 mg weekly) was followed by self-reported improvement in about three-	supports
Rucker, The Sick Times 2026 (conference report)	10	follow-up talk, Berlin 2026	Treatment trials and natural history	In an open-label pilot, 6 of 10 women with moderate-to-severe ME/CFS improved markedly aft	supports
Ruiz-Núñez 2018, Front Endocrinol	98 vs 99	case-control	Hormones	In one Dutch case-control study, CFS patients had normal TSH but lower total T3, T4 and fr	supports
Russell 2019, Psychoneuroendocrinology	55 HCV patients (18 persistent fatigue) + 54 CFS + 57 HV	prospective pre/during/post immune-trigger cohort	Immune system	In the interferon-alpha fatigue model, higher baseline IL-6 and IL-10 and a larger early c	supports
Ryback & Cowan 2025, Frontiers in Immunology	25 mild/mod + 36 severe vs 21 HC vs 28 MS	AIRR-seq case-control with disease control	Immune system	Reported skewing of the B-cell receptor repertoire (IGHV3-30/3-33 overuse) partially repli	supports
Ryback 2026, PLOS ONE data	67 ME vs 53 controls, all female	pre-registered, blinded replication of Fluge 2016; healthy myoblasts in 20% serum 6 days, Seahorse	Muscle, exertion and energy	The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pr	contradicts
Ryback 2026, PLOS ONE (via Solving ME/CFS re-analysis serum-check)	67 ME vs 53 HC	preregistered blinded serum-exposure replication in myoblasts	Immune system	Purified IgG from ME/CFS patients fragments mitochondria and shifts energetics in human en	contradicts
Rydland 2026, Biochem Biophys Rep data	49 vs 50	case-control, EV proteomics	Metabolism and blood omics	Plasma extracellular vesicle (EV) counts may be higher in ME/CFS, but no EV protein or mic	supports
Saha 2019, Clin Hemorheol Microcirc	small ME/CFS vs age-matched HC sample	case-control microfluidics	Blood flow, vessels and blood	Red blood cells from ME/CFS patients were stiffer in one microfluidic study, and real-time	supports
Salonen 2023, J Transl Med	11	RCT (donor vs autologous FMT)	Gut and microbiome	The only published randomised faecal microbiota transplant trial in CFS (n=11) found no be	supports
Samms & Ponting 2025, BMC Public Health	100,055 diagnosed	national hospital records	Treatment trials and natural history	ME/CFS affects roughly 0.2-1% of adults depending on criteria (US 2021-22: 1.3% self-repor	supports
Sandvik 2023, PLOS One	39 ME/CFS vs healthy controls	case-control within RCT	Blood flow, vessels and blood	Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia a	supports
Sardell 2026, J Transl Med	DecodeME cohorts vs UKB controls	combinatorial genetics	Genetics	Proprietary combinatorial SNP analyses (PrecisionLife) report thousands of multi-SNP signa	supports
Sasso 2026, Front Med	36 vs 42	case-control patch-clamp, two sites	Genetics	TRPM3 ion channel dysfunction in NK cells and TRPM3 SNPs are reported repeatedly by one re	supports
Sasso 2026, Frontiers in Medicine	36 vs 42	two-site case-control patch-clamp	Immune system	Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been report	supports
Sato 2021, Brain Behav Immun	case-control	BCR repertoire sequencing	Immune system	Reported skewing of the B-cell receptor repertoire (IGHV3-30/3-33 overuse) partially repli	supports
Sawano 2025, Lancet Infect Dis (PAX LC)	100	decentralised RCT	Infection and viruses	Three randomised placebo-controlled trials of 15-25 days of nirmatrelvir-ritonavir in esta	supports
Sawano 2025, Lancet Infect Dis (PAX LC)	100	phase 2 double-blind RCT	Treatment trials and natural history	Three randomised trials of nirmatrelvir-ritonavir (Paxlovid) for 15-25 days in established	supports
Scheibenbogen 2018, PLOS ONE	10 (open-label)	uncontrolled pilot	Immune system	Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against	contradicts
Scherbakov 2020, ESC Heart Fail	35 ME/CFS vs 20 HC	case-control, EndoPAT	Blood flow, vessels and blood	Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia a	supports
Scherbakov 2020, ESC Heart Fail	35 ME/CFS	case-control	Blood flow, vessels and blood	Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoac	contradicts
Schlick 2022, Int J Mol Sci	PCS vs controls	case-control, OCT-A	Blood flow, vessels and blood	Retinal imaging shows narrower arterioles, lower arteriolar-to-venular ratio, reduced flic	supports
Schreiner 2020, Immunohorizons	cell lines + 10 patient sera	in vitro proteomics / serum transfer	Infection and viruses	HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord t	supports
Schutzer 2011, PLoS One	~43 CFS vs controls (pooled)	CFS vs neurologic post-Lyme vs controls, CSF MS	Metabolism and blood omics	CSF proteomics in ME/CFS is limited to small cohorts and has not produced a replicated pro	supports
Schutzer 2023, Ann Med	15 vs 15 (ME/CFS +/- FM)	CSF proteomics, 2083 proteins	Nervous system and brain	CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct	contradicts
Schönberg 2024, J Cereb Blood Flow Metab	34 vs 34	task fMRI case-control	Nervous system and brain	After a single 30-minute exercise bout, women with ME/CFS performed worse on a demanding c	supports
Sepúlveda 2022, Front Med	92 vs 50	re-analysis of peptide array	Infection and viruses	Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (abse	supports
Seton 2023, medRxiv	5 vs 5 household	feasibility case-control	Gut and microbiome	Quadram Institute studies of severe ME/CFS with household-matched controls report altered	supports
Seton 2023, medRxiv (Quadram)	5 severe patients vs 5 household controls	feasibility case-control (IgG-Seq)	Gut and microbiome	Blood markers of bacterial translocation (LPS, LPS-binding protein, sCD14, IgA/IgM to ente	contradicts
Shah 2025, JAMA Netw Open (RECOVER)	12,276	prospective cohort	Hormones	ME/CFS and long COVID are more common in women (ME/CFS prevalence about 1.4% vs 0.9%; Deco	supports
Shan 2020, J Transl Med	63 studies	systematic review	Nervous system and brain	Conventional structural MRI (grey matter volume, routine clinical reads) shows no consiste	supports
Shan 2020, J Transl Med	9 studies	systematic review	Nervous system and brain	Brainstem differences (smaller volume, altered signal or connectivity) are among the more	supports
Sharpe 2015, Lancet Psychiatry	481 followed	long-term follow-up	Treatment trials and natural history	PACE (n=641) reported modest gains in self-rated fatigue and function with CBT and graded	supports
Shikova 2020, J Med Virol	58 vs 50	case-control serology + nested PCR	Infection and viruses	In large well-characterised cohorts, standard EBV and other herpesvirus seroprevalence and	contradicts
Shikova 2020, J Med Virol	58 vs 50	case-control nested PCR	Infection and viruses	Detectable cell-free EBV DNA in plasma has been reported more often in ME/CFS (24% vs 4% i	supports
Shikova 2020, J Med Virol	58 vs 50	PCR	Infection and viruses	HHV-6 microRNA (miR-aU14) was detected by in situ hybridisation in brain and spinal cord t	contradicts
Shungu 2012, NMR Biomed	15 CFS, 15 MDD, 13 HC	case-control MRSI + ASL + 31P	Nervous system and brain	Brain lactate measured by MR spectroscopy is raised in ME/CFS: ventricular CSF lactate in	supports
Shungu 2012, NMR Biomed	15 vs 13	ASL case-control	Nervous system and brain	Resting (supine) cerebral blood flow by ASL or SPECT is reported as lower in some ME/CFS s	supports
Shungu 2012, NMR Biomed	15 CFS, 15 MDD, 13 HC	1H MRS case-control	Metabolism and blood omics	Markers of oxidative stress (methionine sulfoxide, oxidised glutathione products, lower br	supports
Silverman 2011, Science (partial retraction)	-	retraction notice	Infection and viruses	The 2009 report of the mouse retrovirus XMRV in 67% of CFS patients was a laboratory conta	supports
Simmons 2011, Science	15 positive vs 15 negative (coded replicates)	blinded multi-laboratory	Infection and viruses	The 2009 report of the mouse retrovirus XMRV in 67% of CFS patients was a laboratory conta	supports
Singh 2026, Front Med	37 ME/CFS, 19 LC, 27 HC	case-control DTI/DKI	Nervous system and brain	Advanced diffusion MRI models report widespread white-matter differences in ME/CFS that co	supports
Skjevling 2024, BMJ Open (Comeback protocol)	80 planned	RCT protocol	Gut and microbiome	The only published randomised faecal microbiota transplant trial in CFS (n=11) found no be	supports
Slaghekke 2025, ME/CFS Research Conference poster	~11 per group	electron microscopy	Blood flow, vessels and blood	Muscle biopsies show fewer capillaries per fibre in ME/CFS and less capillary supply per f	supports
Slaughter 2026, medRxiv	1,268 discovery; 319 + 371 replication	GWAS with two replications	Genetics	A 2026 UK Biobank GWAS with stricter, multi-source case definitions (1,268 cases) reports	supports
Smith 2011, Neuropsychobiology	40 vs 40	convergent SNP + expression	Genetics	Before DecodeME, small GWAS and candidate-gene studies (GRIK2, NPAS2, TPPP, TRA locus and	supports
Snell 2013, Phys Ther	51 CFS vs 10 controls, women	2-day CPET	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
Solving ME/CFS re-analysis (3-cohort species meta) data	NIH, Raijmakers, Columbia	random-effects meta-analysis	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	contradicts
Solving ME/CFS re-analysis (EBV load MR) data	DecodeME; MVP	MR with Kamitaki 2026 instruments	Genetics	Three DecodeME loci lie near infection-response genes (OLFM4, RABGAP1L, BTN2A2/TRIM38), an	supports
Solving ME/CFS re-analysis (FinnGen/FinRegistry Risteys) data	about 1,622 and 348 cases	registry co-occurrence (O/E)	Hormones	Diagnosed adrenal insufficiency and hypopituitarism are 7 to 13 times more common than exp	supports
Solving ME/CFS re-analysis (FinnGen/FinRegistry) data	~1,600 G93.3 cases	registry co-occurrence	Nervous system and brain	Uncontrolled clinic series report high rates of craniocervical 'obstruction', low-lying to	contradicts
Solving ME/CFS re-analysis (MVP-only and FinnGen replication) data	3,891 MVP cases; 283 FinnGen	replication of lead SNPs	Genetics	DecodeME (15,579 PEM-defined cases vs 259,909 controls) identified 8 genome-wide significa	contradicts
Solving ME/CFS re-analysis (Mendelian randomisation) data	DecodeME 15,579; MVP	two-sample MR	Gut and microbiome	Plasma indole-3-propionate (IPA), a gut-bacteria product of tryptophan, is lower in ME/CFS	contradicts
Solving ME/CFS re-analysis (PEM) data	60 vs 45, 404 samples	within-person change, mixed models, permutation	Metabolism and blood omics	Germain 2022 reported growing metabolic divergence after a 2-day CPET; within-person re-an	contradicts
Solving ME/CFS re-analysis (TCR, PRJNA940635)	~38 per group per cell type	re-analysis of public TCR repertoire	Infection and viruses	Charité studies report a deficient EBV-specific memory B- and T-cell response in CFS (abse	supports
Solving ME/CFS re-analysis (Walitt 2024 CSF SomaScan, GEO GSE251790)	~17 vs 21	re-analysis	Nervous system and brain	CSF cytokine and proteome studies report differences (e.g. raised eotaxin/CCL11, distinct	contradicts
Solving ME/CFS re-analysis (datasets / deep2) data	4 plasma cohorts	meta-analysis	Metabolism and blood omics	Indole-3-propionate (IPA), a gut-bacterial tryptophan product, is lower in ME/CFS plasma i	supports
Solving ME/CFS re-analysis (datasets 02, deep2 microbiome_function) data	4 plasma cohorts; 231 paired stool-plasma	meta-analysis + cross-validated prediction	Gut and microbiome	Plasma indole-3-propionate (IPA), a gut-bacteria product of tryptophan, is lower in ME/CFS	supports
Solving ME/CFS re-analysis (datasets 03_steroids_crosscheck) data	4 cohorts	re-analysis with covariate adjustment	Hormones	Lower plasma androgenic steroids in ME/CFS (Cornell metabolomics) is mostly an age and sex	supports
Solving ME/CFS re-analysis (datasets) data	two Cornell cohorts, 575 shared compounds	cross-cohort concordance	Metabolism and blood omics	Blood metabolome differences between ME/CFS and controls are real at group level but small	contradicts
Solving ME/CFS re-analysis (datasets) data	4 cohorts	meta-analysis	Metabolism and blood omics	Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts,	contradicts
Solving ME/CFS re-analysis (datasets) data	4 cohorts	meta-analysis	Metabolism and blood omics	Ketone bodies are associated with ME/CFS in UK Biobank NMR data, but 3-hydroxybutyrate sho	contradicts
Solving ME/CFS re-analysis (datasets) data	two Cornell cohorts, 575 shared compounds	cross-cohort concordance	Metabolism and blood omics	Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-valida	contradicts
Solving ME/CFS re-analysis (deep2 HLA) data	DecodeME + MVP	SNP-level MHC re-analysis	Genetics	Reported HLA associations are inconsistent: Lande 2020's HLA-C07:04 and DQB103:03 risk a	contradicts
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME 15,579 cases	Mendelian randomisation	Hormones	ME/CFS genetic risk is enriched in brain tissues and the pituitary, but genetically predic	supports
Solving ME/CFS re-analysis (deep2 genetics_mr) data	15,579 + MVP	two-sample MR	Gut and microbiome	ME/CFS genetic risk overlaps with IBS and is enriched in enteric neurons, but genetically	supports
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME 15,579; MVP 3,891; FinnGen 283	LDSC rg and sign concordance	Genetics	The genome-wide DecodeME signal replicates in independent cohorts: genetic correlation 0.6	supports
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME + MVP	bidirectional MR	Genetics	ME/CFS genetics overlaps most with chronic pain, fibromyalgia, IBS and migraine, and subst	supports
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME 15,579; MVP 3,891	cis-pQTL MR with replication	Genetics	Proteome-wide cis-MR finds genetically higher plasma EPHB1 (a pain-sensitisation receptor)	supports
Solving ME/CFS re-analysis (deep2 genetics_mr) data	DecodeME + MVP	MR with replication	Genetics	MR of DecodeME suggests higher mitochondrial DNA copy number is protective and higher Glyc	supports
Solving ME/CFS re-analysis (deep2 metabolome-wide meta-analysis) data	10 cohorts, ~600 cases / ~430 controls	harmonised meta-analysis of public per-person data (Metabolon, Fiehn, LC-MS/MS, NMR)	Metabolism and blood omics	Blood metabolome differences between ME/CFS and controls are real at group level but small	supports
Solving ME/CFS re-analysis (deep2 metabolome-wide meta-analysis) data	10 cohorts, ~600 cases / ~430 controls	harmonised meta-analysis of public per-person data (Metabolon, Fiehn, LC-MS/MS, NMR)	Metabolism and blood omics	Naviaux 2016's picture of a homogeneous 'hypometabolic' dauer-like state (80% of discrimin	contradicts
Solving ME/CFS re-analysis (deep2 metabolome-wide meta-analysis) data	10 cohorts, ~600 cases / ~430 controls	harmonised meta-analysis of public per-person data (Metabolon, Fiehn, LC-MS/MS, NMR)	Metabolism and blood omics	The IDO metabolic-trap model predicts tryptophan-kynurenine pathway disruption, but blood	contradicts
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	up to 10 cohorts	meta-analysis	Metabolism and blood omics	Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across sever	supports
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	5 cohorts (BioMapAI, Germain 2018/2020/2022, Naviaux)	harmonised meta-analysis with confounder models	Metabolism and blood omics	Four aromatic 2-hydroxy acids (phenyllactate, 4-hydroxyphenyllactate, indolelactate, imida	supports
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	multiple cohorts	meta-analysis	Metabolism and blood omics	Free carnitine and long-chain acylcarnitines are slightly lower in ME/CFS across cohorts,	supports
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	multiple cohorts	pre-specified ratio meta-analysis	Metabolism and blood omics	Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, T	contradicts
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	4 Metabolon + 4 other cohorts	meta-analysis	Metabolism and blood omics	Collagen breakdown products (hydroxyproline, galactosyl-hydroxylysine and related) are hig	supports
Solving ME/CFS re-analysis (deep2 metabolome-wide) data	7 cohorts	meta-analysis	Metabolism and blood omics	Markers of oxidative stress (methionine sulfoxide, oxidised glutathione products, lower br	supports
Solving ME/CFS re-analysis (deep2 microbiome_function) data	231 paired; 5 plasma cohorts	cross-validated prediction, hypothesis-free scans	Gut and microbiome	The replicated deficit of aromatic lactates (indolelactate, phenyllactate, 4-hydroxyphenyl	supports
Solving ME/CFS re-analysis (deep2 proteome convergence) data	4 omic layers	pathway-score correlation with permutation null	Metabolism and blood omics	Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-valida	contradicts
Solving ME/CFS re-analysis (deep2 proteome) data	~150 vs ~105 per-person + UKB	random-effects meta-analysis of protein sets	Metabolism and blood omics	Muscle-derived intracellular proteins (myoglobin, troponins, CA3 and similar) are lower in	supports
Solving ME/CFS re-analysis (deep2 proteome) data	Hoel 50/24 + Germain + Walitt + UKB	artefact score and leak-adjusted meta-analysis	Metabolism and blood omics	Most of the ~845 FDR-significant serum proteins in Hoel 2026 appear driven by a sample-han	contradicts
Solving ME/CFS re-analysis (deep2 proteome) data	4 blood cohorts	set and single-protein meta-analysis	Metabolism and blood omics	Hepatocyte enzymes (ADH4, GSTA1, AGXT, BCHE) are modestly higher, and vascular mural/matri	supports
Solving ME/CFS re-analysis (deep2 proteome) data	Germain 79/51 per-person	within-person change	Metabolism and blood omics	Germain 2025 reported patient-specific proteomic changes during recovery after a 2-day CPE	contradicts
Solving ME/CFS re-analysis (deep2 proteome) data	Rydland + Giloteaux EV sets	re-analysis	Metabolism and blood omics	Plasma extracellular vesicle (EV) counts may be higher in ME/CFS, but no EV protein or mic	contradicts
Solving ME/CFS re-analysis (deep2 proteome) data	Walitt CSF + Bragee	re-analysis	Metabolism and blood omics	CSF proteomics in ME/CFS is limited to small cohorts and has not produced a replicated pro	contradicts
Solving ME/CFS re-analysis (deep2 proteome) data	3 SomaScan cohorts + UKB	re-analysis	Hormones	Patients with POTS and with ME/CFS show normal or low renin and low aldosterone despite re	contradicts
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	Hoel, Germain, Walitt SomaScan + UKB Olink	secondary analysis	Blood flow, vessels and blood	The hormones that should rise to restore blood volume (renin, aldosterone, EPO) are not ra	supports
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	3 SomaScan cohorts + UKB Olink + LC UK	secondary meta-analysis	Blood flow, vessels and blood	Pericytes and vascular smooth-muscle cells have not been examined directly in ME/CFS tissu	supports
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	UKB	competitive gene-set test	Blood flow, vessels and blood	Pericytes and vascular smooth-muscle cells have not been examined directly in ME/CFS tissu	contradicts
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	~150 ME/CFS SomaScan + UKB Olink	secondary analysis	Blood flow, vessels and blood	Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoac	contradicts
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	SomaScan + Olink cohorts	secondary analysis	Blood flow, vessels and blood	VEGF-A has been reported lower, not higher, in ME/CFS plasma, and proteomic re-analysis fi	contradicts
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	SomaScan + Olink cohorts	secondary analysis	Blood flow, vessels and blood	Fibrinaloid microclots (thioflavin-T-positive amyloid fibrin deposits) are reported to be	contradicts
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	SomaScan + Olink cohorts	secondary analysis	Blood flow, vessels and blood	Routine coagulation markers (D-dimer, fibrinogen, PT/APTT) are generally normal in ME/CFS;	contradicts
Solving ME/CFS re-analysis (deep2 proteome), 2026 data	4 cohorts	secondary meta-analysis	Blood flow, vessels and blood	In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetica	contradicts
Solving ME/CFS re-analysis (deep2 single-cell/methylation) data	4 cohorts, 177 vs 107	composition-adjusted EWAS, clocks	Metabolism and blood omics	Reported genome-wide blood DNA methylation differences have not replicated across cohorts,	contradicts
Solving ME/CFS re-analysis (deep2 single-cell/methylation, proteome) data	503,739 cells (58 people); 4 bulk RNA cohorts	person-level pseudobulk DE, cross-cohort meta	Metabolism and blood omics	Blood transcriptome and single-cell RNA studies report monocyte and platelet abnormalities	contradicts
Solving ME/CFS re-analysis (deep2/epi_literature literature tally) data	19 studies	literature contradiction count	Immune system	A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficie	contradicts
Solving ME/CFS re-analysis (deep2/epi_literature, FinnGen/FinRegistry) data	~1,622 FinRegistry / 348 FinnGen postviral fatigue cases	registry enrichment	Immune system	A sizeable minority of ME/CFS patients have IgG subclass or mannose-binding-lectin deficie	supports
Solving ME/CFS re-analysis (deep2/genetics_mr) data	DecodeME + Million Veteran Program	Mendelian randomisation and replication	Immune system	Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type,	supports
Solving ME/CFS re-analysis (deep2/proteome) data	4 blood RNA cohorts	meta-analysis of public data	Immune system	The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-	supports
Solving ME/CFS re-analysis (deep2/proteome) data	~150 cases/105 controls across 4 proteomic cohorts + UK Biobank	harmonised meta-analysis	Immune system	Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF	contradicts
Solving ME/CFS re-analysis (deep2/proteome) data	4 blood protein cohorts + UK Biobank	harmonised proteomic meta-analysis	Immune system	Clinically relevant mast cell activation is reported in up to a quarter of ME/CFS patients	contradicts
Solving ME/CFS re-analysis (deep2/proteome) data	4 blood protein cohorts	meta-analysis	Immune system	Newly reported elevations of serum IL-11 and MMP-9 in female ME/CFS patients are unreplica	contradicts
Solving ME/CFS re-analysis (deep2/proteome; Germain 2-day CPET) data	SomaScan, 2-day CPET	re-analysis	Immune system	Whole-blood immune responses to microbial antigens are exaggerated in ME/CFS and worsen af	contradicts
Solving ME/CFS re-analysis (deep2/proteome; NIH Walitt CSF) data	NIH CSF and serum	re-analysis	Immune system	Cerebrospinal fluid cytokine profiles differ from controls with raised eotaxin (CCL11) and	contradicts
Solving ME/CFS re-analysis (deep2/proteome; long COVID Olink sets) data	UK Explore 51 vs 51; IRIS 36 vs 9; Bansal 12 vs 44	re-analysis of long COVID proteomics	Immune system	Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to	contradicts
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	503,739 PBMCs, 58 people; 5 cohorts	re-analysis of public scRNA, methylation, repertoires	Immune system	Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been report	contradicts
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	304 T-cell repertoire samples	re-analysis, UK ME/CFS Biobank	Immune system	T-cell receptor repertoires do not distinguish ME/CFS from controls, arguing against a dom	supports
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	58-person scRNA; 123-person BCR repertoire	re-analysis	Immune system	The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-	contradicts
Solving ME/CFS re-analysis (deep2/singlecell_methyl)	123 people, UK ME/CFS Biobank	re-analysis of B-cell repertoires	Immune system	Reported skewing of the B-cell receptor repertoire (IGHV3-30/3-33 overuse) partially repli	contradicts
Solving ME/CFS re-analysis (deep2/singlecell_methyl, GSE214283)	30 vs 28, 503,739 PBMCs	re-analysis with CellTypist + marker gating	Immune system	CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exha	contradicts
Solving ME/CFS re-analysis (deep2/singlecell_methyl, GSE214283)	30 vs 28, 503,739 PBMCs	independent re-analysis	Immune system	Single-cell transcriptomics showed classical-monocyte dysregulation at baseline and platel	contradicts
Solving ME/CFS re-analysis (deep2/singlecell_methyl; downloads)	5 cohorts; 58-people scRNA; 5 blood RNA cohorts	re-analysis	Immune system	Increased mucosal-associated invariant T (MAIT) cell proportions and a shift in CD8+ memor	contradicts
Solving ME/CFS re-analysis (downloads)	5 blood RNA cohorts + cfRNA + SomaScan	re-analysis	Immune system	Plasma cell-free RNA carries an immune-weighted ME/CFS signature that classifies cases wit	contradicts
Solving ME/CFS re-analysis (downloads, DecodeME MAGMA) data	15,579 cases vs 259,909 controls	GWAS enrichment	Immune system	Reduced TRPM3 ion-channel current and calcium influx in patients' NK cells has been report	contradicts
Solving ME/CFS re-analysis (downloads; DecodeME) data	15,579 cases vs 259,909 controls	GWAS	Immune system	Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes	contradicts
Solving ME/CFS re-analysis (epi_literature tally) data	24 study groups	vote count of published studies	Hormones	On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and mor	supports
Solving ME/CFS re-analysis (epi_literature; FinnGen R12 / FinRegistry) data	1,622 (FinRegistry), 348 (FinnGen) G93.3 cases	registry co-occurrence (O/E, Fisher, BH-FDR)	Infection and viruses	In Finnish registry data, ME/CFS (G93.3) is enriched 3-6-fold for prior viral infections (	supports
Solving ME/CFS re-analysis (genetics_mr), using Kamitaki 2026 instruments, DecodeME and MVP data	DecodeME ~15k cases; MVP	two-sample Mendelian randomisation	Infection and viruses	Genetic instruments for higher EBV DNA load in blood (Kamitaki 2026) are associated with M	supports
Solving ME/CFS re-analysis (locus tissue permutation)	8 loci	permutation	Genetics	Genome-wide ME/CFS heritability is enriched in brain tissues and neurons (including glutam	contradicts
Solving ME/CFS re-analysis (proteome) data	Germain 79/51 per-person	within-person change	Muscle, exertion and energy	Plasma proteomics of 79 ME/CFS and 53 sedentary controls across a 2-day CPET found the lar	contradicts
Solving ME/CFS re-analysis (proteome) data	Hoel 50/24, Germain 78/51, Walitt 21/21, plus UK Biobank	meta-analysis of public proteomics	Muscle, exertion and energy	Muscle-derived proteins (myoglobin, troponins, CA3 and similar) are lower in ME/CFS blood	supports
Solving ME/CFS re-analysis (proteomics) data	~150 patients across cohorts	re-analysis	Nervous system and brain	Markers of neuronal or glial injury (NfL, GFAP, tau, UCH-L1) are not raised in ME/CFS bloo	supports
Solving ME/CFS re-analysis (resting metabolomics, several cohorts)	Germain 2022 60 vs 45 plus 5 cohorts for aromatic lactates	re-analysis of public metabolomics	Muscle, exertion and energy	Fluge 2016 proposed impaired pyruvate dehydrogenase (PDH) function from low TCA-fuelling a	contradicts
Solving ME/CFS re-analysis of BioMapAI data	143 patients	within-patient association	Gut and microbiome	Links between specific gut bacteria and symptom severity are inconsistent: F. prausnitzii	contradicts
Solving ME/CFS re-analysis of BioMapAI (Xiong 2025) data	143 vs 88	species association, adjusted	Gut and microbiome	Ruminococcus gnavus, Flavonifractor plautii, Ruthenibacterium lactatiformans and Enteroclo	supports
Solving ME/CFS re-analysis of BioMapAI plasma metabolome data	about 236	case-control re-analysis	Hormones	On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and mor	contradicts
Solving ME/CFS re-analysis of Germain 2022 data	60 vs 45 (404 samples)	within-person change scores, mixed models, permutation	Muscle, exertion and energy	Some patients produce more lactate for a given workload, and this worsens on a repeat test	contradicts
Solving ME/CFS re-analysis of Germain 2022 data	60 vs 45 (404 samples)	within-person change scores, mixed models, permutation	Muscle, exertion and energy	Plasma metabolomics across a 2-day CPET shows clear resting differences in ME/CFS, but onc	contradicts
Solving ME/CFS re-analysis of Ryback 2026 supplementary data	67 vs 53	participant-level recomputation	Muscle, exertion and energy	The claim that ME/CFS serum changes healthy myoblast mitochondrial respiration failed a pr	contradicts
Solving ME/CFS re-analysis of Walitt 2024 NIH stool	15 vs 17	pre-specified signature test	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	supports
Solving ME/CFS re-analysis, negative-control serologies	same	MR specificity check	Infection and viruses	Genetic instruments for higher EBV DNA load in blood (Kamitaki 2026) are associated with M	contradicts
Song 2025, Sci Rep	UK Biobank CFS GWAS	two-sample MR	Genetics	ME/CFS genetics overlaps most with chronic pain, fibromyalgia, IBS and migraine, and subst	contradicts
Sorensen 2009, Mol Med	CFS vs controls, exercise	case-control, exercise challenge	Blood flow, vessels and blood	In ME/CFS, complement findings are heterogeneous: an exercise-induced C4a rise, a genetica	supports
Spronk 2023, Epidemiol Infect	368	cross-sectional, ≥10 years post-infection	Infection and viruses	Q fever fatigue syndrome (QFS) follows roughly 20% of acute Q fever and persists more than	supports
Stein 2024, Lancet Regional Health Europe	20 (open-label)	prospective pre-post cohort	Immune system	Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against	contradicts
Stein 2025, Lancet Reg Health Eur	20	prospective open-label cohort (repeat IA, beta2-AR antibody positive)	Treatment trials and natural history	Removing antibodies by immunoadsorption was no better than a sham procedure in two control	supports
Stein, conference case series 2026	15	uncontrolled, autoantibody-positive	Treatment trials and natural history	Removing antibodies by immunoadsorption was no better than a sham procedure in two control	contradicts
Stortz 2026 (IAMPOCO), Lancet Reg Health Eur	40	sham-controlled crossover	Immune system	IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve	contradicts
Stortz 2026 (IAMPOCO), Lancet Regional Health Europe	40	randomised patient-blinded sham-controlled crossover	Immune system	Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against	supports
Stortz/Kraus et al. 2026, IAMPOCO (Mainz repository; Lancet Reg Health Eur correspondence)	40	randomised patient-blinded sham-controlled crossover	Treatment trials and natural history	Removing antibodies by immunoadsorption was no better than a sham procedure in two control	supports
Strawbridge 2019, Neuroscience & Biobehavioral Reviews	42 studies	systematic review and meta-analysis	Immune system	Meta-analysis of 42 studies finds small but significant elevations of TNF, IL-2, IL-4, TGF	supports
Strayer 2012, PLoS One	234	phase III double-blind RCT, 12 sites	Treatment trials and natural history	Rintatolimod (Ampligen) gave a modest placebo-adjusted gain in treadmill time in a 234-per	supports
Strayer 2020, PLoS One (post hoc)	ITT population	post hoc subgroup	Treatment trials and natural history	Rintatolimod (Ampligen) gave a modest placebo-adjusted gain in treadmill time in a 234-per	contradicts
Streeten & Bell 2000, Am J Med Sci	15 CFS	case series, Cr-51 red cell mass	Blood flow, vessels and blood	Red cell volume and total blood volume are below predicted in a large minority to about ha	supports
Stussman 2024, BMJ Neurol Open	10 ME/CFS vs 9 controls	mixed-methods (VAS + qualitative interviews) over 72 h after CPET	Muscle, exertion and energy	PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took	supports
Stussman 2025, Front Neurol	34 long COVID exercise cohort, 9 ME/CFS, 9 controls; 244 questionnaire cohort	CPET with serial VAS and interviews over 72 h	Muscle, exertion and energy	PEM is usually delayed by hours to a day, lasts days to weeks, and after a 2-day CPET took	contradicts
Su 2022, Cell	309	prospective multi-omic cohort (COVID)	Infection and viruses	Detectable cell-free EBV DNA in plasma has been reported more often in ME/CFS (24% vs 4% i	supports
Swank 2022, medRxiv (later published in Clin Infect Dis)	63	case-control plasma antigen	Infection and viruses	SARS-CoV-2 spike, S1 or nucleocapsid antigen is detectable in plasma more often after COVI	supports
São Paulo group 2025, ATS abstract	32	RCT 6-10 weeks, long COVID	Treatment trials and natural history	A single 60 mg pyridostigmine dose raised peak oxygen uptake relative to placebo on a repe	contradicts
Sørland 2021, Front Med	35 ME/CFS vs 66 HC	case-control within open-label trial	Blood flow, vessels and blood	Large-artery flow-mediated dilation and microvascular post-occlusive reactive hyperaemia a	supports
Taub 2021, J Am Coll Cardiol	22	double-blind crossover RCT	Treatment trials and natural history	Ivabradine lowered standing heart rate but did not improve symptoms in 181 people with lon	supports
Teuschl 2025, Clin Auton Res	n/a	expert commentary	Nervous system and brain	On the 10-minute NASA lean test, a subgroup of recent-onset patients shows a rising heart	contradicts
Theorell 2017, Frontiers in Immunology	48 patients, two cohorts, matched HC	case-control, deep cytotoxic-lymphocyte phenotyping	Immune system	Natural-killer-cell cytotoxicity is reduced in ME/CFS in meta-analysis of the older litera	contradicts
Tokumasu 2021, BMJ Case Reports	1	case report	Hormones	Diagnosed adrenal insufficiency and hypopituitarism are 7 to 13 times more common than exp	supports
Tomas 2017, PLOS ONE	CFS fresh 25/frozen 38 vs control fresh 3/frozen 12	Seahorse PBMC	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	supports
Tomas 2019, PeerJ	myotubes n=6; PBMC 13 CFS vs 6 controls	permeabilised-cell extracellular flux	Muscle, exertion and energy	Muscle cells grown from ME/CFS patients failed to activate AMPK and raise glucose uptake a	contradicts
Tomas 2019, PeerJ	13 CFS vs 6 controls (PBMC)	permeabilised cells	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	contradicts
Trivedi 2018, PLoS One data	13 vs 12 (+33/31 validation)	EPIC array + pyrosequencing	Metabolism and blood omics	Reported genome-wide blood DNA methylation differences have not replicated across cohorts,	supports
Tuomaala 2026, J Neurol	14 LC, 11 HC, 13 MS	case-control PET	Nervous system and brain	In long COVID, one [11C]PBR28 PET study found widespread raised TSPO signal that correlate	contradicts
USF 2013 report (secondary source: press/HHV-6 Foundation)	small clinic cohort	cross-sectional	Infection and viruses	Inherited chromosomally integrated HHV-6 (iciHHV-6, ~1% of people) has been reported at ro	supports
Unger 2024 (INSPIRE), via Solving ME/CFS re-analysis	~4,500	prospective cohort, test-positive vs test-negative	Infection and viruses	In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after S	contradicts
Vahratian 2023, NCHS Data Brief 488	NHIS 2021-22	national survey	Treatment trials and natural history	ME/CFS affects roughly 0.2-1% of adults depending on criteria (US 2021-22: 1.3% self-repor	supports
VanElzakker 2024, Brain Behav Immun	12 vs 43	case-control PET	Nervous system and brain	In long COVID, one [11C]PBR28 PET study found widespread raised TSPO signal that correlate	supports
VanNess 2003, Med Sci Sports Exerc	189 CFS	clinical series, single CPET	Muscle, exertion and energy	On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled a	supports
Venturini 2019, Oxid Med Cell Longev	small	uncontrolled pilot	Gut and microbiome	Probiotic trials in CFS are small; one 39-person RCT of Lactobacillus casei Shirota reduce	supports
Vermeulen 2010, J Transl Med	15 vs 15 women	PBMC OXPHOS plus 2-day CPET	Muscle, exertion and energy	Mitochondrial function in patients' blood cells has been reported as lower (PBMC maximal r	contradicts
Vermeulen 2014, J Transl Med	178 CFS women, 172 idiopathic fatigue, 11 healthy women (plus men)	CPET with non-invasive cardiac output (Nexfin)	Muscle, exertion and energy	Invasive exercise testing shows two peripheral limits in ME/CFS: low cardiac filling (prel	supports
Vernino 2024, Clin Auton Res	30	RCT	Treatment trials and natural history	Placebo and natural-course improvement in ME/CFS and long-COVID trials is large (35% respo	supports
Vernino 2024, Clin Auton Res (iSTAND)	16 vs 14	RCT vs albumin	Treatment trials and natural history	IVIG has no consistent controlled evidence: 1990s ME/CFS RCTs conflicted, and a 30-person	supports
Vernon 2025, Front Neurol	82	secondary analysis	Treatment trials and natural history	Oxaloacetate 2 g/day reduced Chalder fatigue more than control (−25% vs −10%) in an 82-per	supports
Vernon 2025, J Gen Intern Med	11,785 infected vs 1,439 uninfected	prospective observational cohort	Infection and viruses	In RECOVER-Adult, new ME/CFS (IOM criteria) developed at 2.66 per 100 person-years after S	supports
Vernon 2025, J Gen Intern Med	11,785 infected vs 1,439	prospective cohort	Treatment trials and natural history	About 4.5% of SARS-CoV-2-infected adults in RECOVER met IOM ME/CFS criteria (vs 0.6% uninf	supports
Vink & Vink-Niese 2026, J Clin Med	review	narrative review of severe ME	Nervous system and brain	Self-reported light and sound sensitivity is common in ME/CFS. One ERP study found normal	supports
Vollmer-Conna 1997, Am J Med	99	RCT, adolescents/adults	Treatment trials and natural history	IVIG has no consistent controlled evidence: 1990s ME/CFS RCTs conflicted, and a 30-person	contradicts
Vollmer-Conna 2004, Psychological Medicine	21 + 48 acute infections	prospective symptom-cytokine correlation	Immune system	Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes	supports
Vollmer-Conna 2008, Clinical Infectious Diseases	Dubbo cohort	genotype-phenotype analysis	Immune system	Acute-infection symptom severity, and the pro-inflammatory cytokine response and genotypes	supports
Vroegindeweij 2024, Psychoneuroendocrinology	12 ME/CFS, 20 Q-fever fatigue, 8 post-COVID, 19 JIA vs 57 controls	case-control (LC-MS/MS hair cortisol)	Hormones	On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and mor	supports
Vu 2024, Cell Rep Med data	ME/CFS vs controls, before and after exercise	single-cell RNA-seq of PBMC	Muscle, exertion and energy	After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), a	contradicts
Vu 2024, Cell Rep Med data	~30 vs 28 (PBMC, pre/post CPET)	scRNA-seq	Metabolism and blood omics	Blood transcriptome and single-cell RNA studies report monocyte and platelet abnormalities	supports
Vu 2024, Cell Reports Medicine data	case-control, paired pre/post exercise	scRNA-seq of PBMCs	Immune system	Single-cell transcriptomics showed classical-monocyte dysregulation at baseline and platel	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	16-20 per group	grip MVC, time to failure, repetitive grip	Muscle, exertion and energy	Repeated maximal handgrip testing shows lower strength, faster decline across repetitions	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	deep phenotyping incl. 40-min tilt	Nervous system and brain	Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 2	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control tilt	Nervous system and brain	POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	14 vs 19 (Holter)	case-control	Nervous system and brain	Resting heart rate is higher and high-frequency heart-rate variability lower in ME/CFS. Th	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control, thigh + leg biopsies	Nervous system and brain	Skin biopsies show reduced small-nerve-fibre density in about 31-53% of ME/CFS patients fr	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control 3T MRI incl. DTI	Nervous system and brain	Conventional structural MRI (grey matter volume, routine clinical reads) shows no consiste	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	15-direction DTI	Nervous system and brain	Advanced diffusion MRI models report widespread white-matter differences in ME/CFS that co	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control task fMRI	Nervous system and brain	During a sustained grip task, ME/CFS patients fatigue earlier while their brain motor-netw	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	~17 vs 21	case-control CSF	Nervous system and brain	In the NIH cohort, CSF levels of the catechol metabolites DOPA, DOPAC and DHPG were lower	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	~17 vs 21	CSF metabolomics	Nervous system and brain	Targeted CSF metabolomics found raised serine and phospholipids and reduced 5-methyltetrah	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	~17 vs 21	case-control Simoa	Nervous system and brain	Markers of neuronal or glial injury (NfL, GFAP, tau, UCH-L1) are not raised in ME/CFS bloo	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	15 vs 16 (EEfRT)	case-control	Nervous system and brain	The NIH claim that altered 'effort preference' defines post-infectious ME/CFS rests on the	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	case-control, 15 tests	Nervous system and brain	Meta-analysis shows moderate-to-large deficits in information-processing speed, attention,	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17	clinical PSG	Nervous system and brain	Objective sleep in adult ME/CFS shows modestly longer sleep latency, more wake after sleep	contradicts
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 PI-ME/CFS vs 21 HV	deep phenotyping incl. PBMC RNA, flow, metabolomics	Metabolism and blood omics	Blood transcriptome and single-cell RNA studies report monocyte and platelet abnormalities	supports
Walitt 2024, Nat Commun nih-walitt-2024 mwb-st003178	17 vs 21	deep phenotyping, multiple compartments	Metabolism and blood omics	Even the deepest blood omics studies sample mainly at rest, in mild-to-moderate patients,	supports
Walitt 2024, Nat Commun (NIH) nih-walitt-2024 mwb-st003178	about 17 PI-ME/CFS vs 21 controls	deep phenotyping	Muscle, exertion and energy	On a single maximal exercise test, ME/CFS patients have lower peak oxygen uptake (pooled a	supports
Walitt 2024, Nat Commun (authors' own analysis) nih-walitt-2024 mwb-st003178	15 vs 17	case-control, per-species tests	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	contradicts
Walitt 2024, Nature Communications nih-walitt-2024 mwb-st003178	17 PI-ME/CFS vs 21 HV (immune measures; 16 vs 20 for switched memory)	deep phenotyping case-control	Immune system	The NIH deep-phenotyping study found more naive and fewer switched-memory B cells in post-	supports
Walitt 2025 reply, Nat Commun	n/a	reply	Nervous system and brain	The NIH claim that altered 'effort preference' defines post-infectious ME/CFS rests on the	supports
Wallraven 2026, BMC Med	102 PCS (62 ME/CFS), 102 recovered, 96 never-infected	prospective case-control, retinal vessel analysis	Blood flow, vessels and blood	Retinal imaging shows narrower arterioles, lower arteriolar-to-venular ratio, reduced flic	supports
Wallraven 2026, BMC Med	102 PCS (62 ME/CFS) vs 198 controls	case-control	Blood flow, vessels and blood	Proteomic panels do not show a consistent rise in classic endothelial-activation or vasoac	supports
Wang 2023, PNAS	index patient; transgenic mice; small ME/CFS biopsy cohort	mechanistic case study plus cohort biopsies	Muscle, exertion and energy	WASF3 protein is raised in ME/CFS muscle, and in cells and mice it disrupts respiratory su	supports
Wassif 1994, Ann Clin Biochem	10 CFS vs 10 alcohol misuse	ischaemic lactate/ammonia test and biopsy histomorphometry	Muscle, exertion and energy	Older ME/CFS muscle biopsy studies mostly found normal or nonspecific histology, without t	contradicts
Watt 2012, J Med Virol	61	retrospective uncontrolled chart review	Infection and viruses	In a 30-patient RCT of 6 months of valganciclovir in CFS patients selected for high HHV-6	supports
White 2011, Lancet (PACE)	641	4-arm RCT	Treatment trials and natural history	PACE (n=641) reported modest gains in self-rated fatigue and function with CBT and graded	supports
White 2011, Lancet (PACE)	641	RCT	Treatment trials and natural history	NICE's 2021 guideline (NG206) told clinicians not to offer graded exercise therapy, and pa	contradicts
White 2011, Lancet (PACE)	160 APT vs 160 SMC	RCT	Treatment trials and natural history	Pacing (staying within one's energy limits) is the top-rated approach in patient surveys a	contradicts
White 2012, Psychosom Med	22 CFS, 20 MS, 23 controls	pre/post moderate exercise qPCR	Muscle, exertion and energy	After moderate exercise, leukocyte mRNA for metabolite-detecting (ASIC3, P2X4/5, TRPV1), a	supports
Wielscher 2026, Research Square preprint	DecodeME 15,579	genetic correlation + MR	Gut and microbiome	ME/CFS genetic risk overlaps with IBS and is enriched in enteric neurons, but genetically	supports
Wielscher 2026, Research Square preprint	DecodeME 15,579	rg, pleiotropy, MR	Genetics	MR of DecodeME suggests higher mitochondrial DNA copy number is protective and higher Glyc	supports
Williams 2002, Eur J Clin Invest	about 30	randomised crossover (melatonin, phototherapy)	Hormones	Melatonin may help the subgroup with delayed melatonin onset, and melatonin plus zinc mode	contradicts
Wilshire 2018, BMC Psychol	641 (re-analysis)	protocol-based re-analysis	Treatment trials and natural history	PACE (n=641) reported modest gains in self-rated fatigue and function with CBT and graded	contradicts
Wittke/Scheibenbogen, CFS_CARE, presented 2026	97	uncontrolled rehab study	Treatment trials and natural history	NICE's 2021 guideline (NG206) told clinicians not to offer graded exercise therapy, and pa	supports
Wong 1992, Chest	22 vs 21	31P-MRS during exercise to exhaustion	Muscle, exertion and energy	Some patients produce more lactate for a given workload, and this worsens on a repeat test	supports
Wong 1992, Chest	22 vs 21	31P-MRS gastrocnemius	Muscle, exertion and energy	31P-MRS studies from Newcastle found exaggerated muscle acidosis and up to ~4-fold slower	contradicts
Woo 2026, Molecular Psychiatry	46 studies; 1,388 vs 1,349	systematic review and meta-analysis	Hormones	On average, people with ME/CFS have mildly lower free cortisol (salivary awakening and mor	supports
Woo 2026, Molecular Psychiatry	12 challenge studies within 46	meta-analysis	Hormones	Challenge tests show a blunted cortisol response to ACTH and exaggerated cortisol suppress	supports
Wossidlo 2026, Infection	18	prospective exploratory, 5 sessions	Immune system	Removing autoantibodies by immunoadsorption helps in open-label cohorts but fails against	supports
Wossidlo 2026, Infection	18	prospective exploratory, uncontrolled	Treatment trials and natural history	Removing antibodies by immunoadsorption was no better than a sham procedure in two control	contradicts
Wu 2025, eClinicalMedicine (long COVID)	58	24-week RCT nicotinamide riboside	Metabolism and blood omics	NAD+ depletion is a proposed mechanism but has not been directly established in ME/CFS coh	contradicts
Xie 2023, JAMA Intern Med	35,717 vs 246,076	observational VA cohort	Infection and viruses	Treating acute COVID-19 with nirmatrelvir may reduce later long COVID: a small terminated	supports
Xiong 2023, Cell Host Microbe data	short-term 75, long-term 79 vs 79 controls	case-control, stool metagenomics + plasma metabolomics	Metabolism and blood omics	Lower plasmalogens, some phosphatidylcholines and sphingomyelins are reported across sever	supports
Xiong 2023, Cell Host Microbe data	short-term 75, long-term 79 vs 79 controls	case-control, stool metagenomics + plasma metabolomics	Metabolism and blood omics	Indole-3-propionate (IPA), a gut-bacterial tryptophan product, is lower in ME/CFS plasma i	supports
Xiong 2023, Cell Host Microbe data	154 vs 79	case-control, metagenomics + plasma metabolomics	Gut and microbiome	Stool from ME/CFS patients has fewer butyrate-producing bacteria (notably Faecalibacterium	supports
Xiong 2023, Cell Host Microbe data	75 short, 79 long vs 79	cross-sectional two-cohort comparison	Gut and microbiome	Gut dysbiosis is strongest in the first years of illness and largely resolves in patients	supports
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	Metabolism and blood omics	Blood metabolome differences between ME/CFS and controls are real at group level but small	contradicts
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	Metabolism and blood omics	Indole-3-propionate (IPA), a gut-bacterial tryptophan product, is lower in ME/CFS plasma i	supports
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	Metabolism and blood omics	Multi-omic machine-learning classifiers (BioMapAI and others) perform well in cross-valida	supports
Xiong 2025 (BioMapAI), Nat Med data	249 participants, 4-year longitudinal	deep neural net on stool metagenome, plasma metabolome, immune profiling, labs, symptoms	Metabolism and blood omics	Even the deepest blood omics studies sample mainly at rest, in mild-to-moderate patients,	supports
Xiong 2025 (BioMapAI), Nat Med data	249	longitudinal multi-omics	Gut and microbiome	Plasma indole-3-propionate (IPA), a gut-bacteria product of tryptophan, is lower in ME/CFS	supports
Xiong 2025, Nat Med data	249	longitudinal multi-omics, deep learning	Gut and microbiome	Microbiome-based machine-learning classifiers distinguish ME/CFS from controls in held-out	supports
Xu 2019, Analyst	small (PBMCs from patients and controls)	single-cell Raman microspectroscopy	Metabolism and blood omics	Raman spectroscopy of PBMCs or plasma separates ME/CFS from controls in single-centre stud	supports
Yamano 2016, Sci Rep	133 samples (93 train / 40 validation)	case-control, CE-MS plasma	Metabolism and blood omics	Several studies infer impaired carbohydrate oxidation (pyruvate dehydrogenase restraint, T	supports
Yamazaki 2026, Physiol Rep	49 ME/CFS	cross-sectional, blood volume + 10-min lean test	Blood flow, vessels and blood	Red cell volume and total blood volume are below predicted in a large minority to about ha	supports
Yamazaki 2026, Physiol Rep	49 ME/CFS	cross-sectional	Blood flow, vessels and blood	Red cell volume and total blood volume are below predicted in a large minority to about ha	contradicts
Yamazaki 2026, Physiol Rep	49 ME/CFS	lean test with capnography	Blood flow, vessels and blood	Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exerci	supports
Yamazaki 2026, Physiol Rep	49	cross-sectional, lean test with capnography + Daxor blood volume	Nervous system and brain	Many patients over-breathe when upright and lower their end-tidal CO2. This drives part of	supports
Yamazaki 2026, Physiol Rep	49	cross-sectional	Nervous system and brain	On the 10-minute NASA lean test, a subgroup of recent-onset patients shows a rising heart	supports
Yin 2024, Nature Immunology	LC vs non-LC, 8 months post-infection	multi-omic + serology case-control	Immune system	Long COVID immune profiling finds exhausted T cells, myeloid shifts, raised antibodies to	supports
Yin 2024, Nature Immunology (long COVID)	deep immune profiling, 8 months post-infection	case-control omics + serology	Immune system	CD8+ effector-memory T cells in ME/CFS show transcriptional and chromatin features of exha	supports
Yu 2025, Sci Rep	PI and GO subgroups vs sedentary HC	case-control diffusion MRI	Nervous system and brain	Advanced diffusion MRI models report widespread white-matter differences in ME/CFS that co	supports
Yu 2026, Hum Brain Mapp	67 vs 67	case-control diffusion MRI (NII model)	Nervous system and brain	Advanced diffusion MRI models report widespread white-matter differences in ME/CFS that co	supports
Zhang 2025 (HEAL2), medRxiv	multi-site whole genomes	deep-learning rare-variant analysis	Immune system	Common-variant genetics shows no enrichment of ME/CFS risk in any blood immune cell type,	contradicts
Zhang 2025, Vasc Health Risk Manag	65 vs 65 children (POTS)	case-control	Hormones	Patients with POTS and with ME/CFS show normal or low renin and low aldosterone despite re	contradicts
Zhang 2025, medRxiv (HEAL2)	464 vs 611	WGS + deep learning	Genetics	Whole-genome sequencing of 464 cases and 611 controls with a deep-learning rare-variant mo	supports
Zilberman-Itskovich 2022, Sci Rep	73	double-blind sham RCT, 40 sessions at 2.0 ATA	Treatment trials and natural history	Forty daily sessions of hyperbaric oxygen at 2.0 ATA improved cognition and symptoms versu	supports
Zollner 2022, Gastroenterology	46	endoscopy cohort (IBD)	Infection and viruses	SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a m	supports
Zuo 2024, Lancet Infect Dis	225 patients, 317 samples	cross-sectional tissue cohort	Infection and viruses	SARS-CoV-2 RNA and protein persist in gut and other tissues for months to two years in a m	supports
de Sa / Iwasaki 2026, Cell	LC cohort + mouse transfer	autoantibody mapping + passive transfer	Immune system	IgG from long COVID patients transfers pain, fatigue-like behaviour and small-fibre nerve	supports
de Vega 2014, PLoS One	PBMC case-control	450K array	Metabolism and blood omics	Reported genome-wide blood DNA methylation differences have not replicated across cohorts,	supports
van Campen 2020, Clin Neurophysiol Pract	429 ME/CFS vs 44 HC	case-control, tilt with extracranial Doppler	Blood flow, vessels and blood	Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, i	supports
van Campen 2020, Clin Neurophysiol Pract	429 vs 44	case-control, tilt with extracranial Doppler	Nervous system and brain	Blood flow to the brain falls much more than normal during head-up tilt in ME/CFS (about 2	supports
van Campen 2020, Clin Neurophysiol Pract	429	tilt cohort	Nervous system and brain	POTS occurs in a minority of ME/CFS patients, and published rates vary widely (about 6-43%	supports
van Campen 2020, Healthcare	82 women (31 mild, 31 moderate, 20 severe)	2-day CPET by severity	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
van Campen 2021, Clin Neurophysiol Pract	ME/CFS patients	tilt with post-tilt recovery	Blood flow, vessels and blood	Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, i	supports
van Campen 2021, Front Neurol (compression)	16 ME/CFS	within-subject tilt	Blood flow, vessels and blood	Raising blood pressure with phenylephrine during tilt blocked hyperventilation, restored c	supports
van Campen 2021, Healthcare	50 ME/CFS vs 51 idiopathic chronic fatigue (women); 26 vs 25 (men)	2-day CPET, disease-control comparison	Muscle, exertion and energy	ME/CFS patients perform worse on a second maximal exercise test 24 h after the first, most	supports
van Campen 2021, Healthcare	50 vs 51 women	2-day CPET, idiopathic chronic fatigue control	Muscle, exertion and energy	Several designs (aerobic-capacity-matched controls, idiopathic-fatigue and MS comparison g	supports
van Campen 2023, Physiol Rep	female ME/CFS vs HC	tilt with PETCO2 and CBF	Blood flow, vessels and blood	Many ME/CFS patients over-breathe when upright (orthostatic hypocapnia), and during exerci	contradicts
van Campen 2024, J Clin Med	233 ME/CFS with POTS + normal-response patients vs HC	retrospective tilt with echocardiographic SVI	Blood flow, vessels and blood	Left-ventricular volumes and stroke volume are smaller in ME/CFS at rest and fall more tha	supports
van Campen 2024, J Clin Med	233 ME/CFS-POTS	retrospective tilt	Blood flow, vessels and blood	POTS is present in roughly a fifth of ME/CFS patients on formal testing (prevalence varies	supports
van Campen 2025, J Clin Med	ME/CFS with POTS	retrospective tilt CO + CBF	Blood flow, vessels and blood	Cerebral blood flow falls about 26% during head-up tilt in ME/CFS versus 7% in controls, i	supports
van Heukelom 2006, Eur J Neurol	29	open-label, DLMO-selected	Hormones	Melatonin may help the subgroup with delayed melatonin onset, and melatonin plus zinc mode	supports

Glossary

16S rRNA sequencing	A cheaper method that identifies bacteria by one marker gene, usually only to genus level.
2-day CPET	Two CPETs 24 hours apart, used to see whether performance drops on the second day.
23Na-MRI	An MRI scan tuned to sodium, used to measure salt content in tissue.
31P-MRS	A magnetic resonance scan that measures phosphorus-containing energy molecules (PCr, ATP) and acidity inside working muscle.
Acylcarnitines	Fats bound to carnitine for transport into mitochondria; their profile hints at how fat is burned.
Antigenemia	Viral protein detectable in the blood.
ATA / bar	Units of pressure. 1 ATA is sea-level air pressure; HBOT trials use 2.0-2.4.
AUC	How well a test separates two groups: 0.5 is a coin flip, 1.0 is perfect. Values from the same data used to build a model are usually too optimistic.
Autoantibody	An antibody that targets the body's own tissue instead of a germ.
Basement membrane	A thin protein layer around each capillary; thickening can slow oxygen passing to tissue.
Beighton score	A 9-point bedside test of joint flexibility; 4 or more (adults often 5) suggests generalised hypermobility.
Butyrate	A short-chain fatty acid made by some gut bacteria from fibre. It fuels the cells lining the colon and dampens inflammation.
Cell danger response (CDR)	Naviaux's idea that cells under threat switch into a protective, low-energy mode, which in ME/CFS fails to switch off.
Cell-free RNA (cfRNA)	RNA fragments floating free in blood plasma, released when cells die or turn over. They give a readout of what many tissues and cells are doing.
Chronotropic incompetence	The heart rate fails to rise as much as expected during exercise.
Colocalisation	A test of whether two genetic signals in the same region share the same causal variant.
Complement	About 30 blood proteins in the innate immune system that tag and punch holes in targets; terminal complex (sC5b-9) marks full activation.
Copeptin	A stable fragment released with vasopressin. It is easier to measure accurately than vasopressin itself.
Cortisol awakening response	The normal rise in cortisol in the 30–45 minutes after waking, measured in saliva.
CPET	Cardiopulmonary exercise test: cycling or treadmill to exhaustion while breathing gases and heart rate are measured.
Craniocervical instability (CCI)	Proposed abnormal movement at the skull-neck junction that compresses the brainstem; mainly discussed in hypermobile patients.
Cross-validation vs external validation	Testing a model on parts of the same data set vs on a completely separate group of people. Only the second shows real-world performance.
CSF	Cerebrospinal fluid, which bathes the brain and spinal cord; sampled by lumbar puncture.
Cytokine	A small signalling protein immune cells use to talk to each other; also what makes you feel ill during an infection.
Cytotoxicity assay	A lab test that mixes immune cells with target cells and measures how many targets they kill. It needs fresh blood, which is why results vary between labs.
Dexamethasone suppression test	A dose of a strong steroid to see how much the body's own cortisol is switched off (feedback sensitivity).
Diffusion MRI / DTI	MRI that tracks water movement along nerve fibres to infer white-matter structure.
DLMO	Dim-light melatonin onset, the evening time melatonin starts rising. It is a marker of body-clock timing.
DNA methylation / epigenetic clock	Chemical tags on DNA that change gene activity; clocks estimate biological age from them. Results depend heavily on the mix of cell types.
dUTPase	An enzyme some herpesviruses make when partly reactivated. It can itself stir up inflammation, and antibodies to it are used as a sign of reactivation.
Dysbiosis	An imbalance in the gut microbial community compared with healthy people.
EA-D, VCA, EBNA	Parts of Epstein-Barr virus that antibodies are measured against. EA-D antibodies suggest recent reactivation, VCA IgM suggests a new infection, and EBNA IgG shows an old past infection.
EEfRT	Effort-Expenditure for Rewards Task: repeated choices between an easy and a hard button-pressing task for money; used by NIH to define 'effort preference'.
Effect size (SD, d)	How far apart the group averages are, in units of normal person-to-person spread. 0.2 is small, 0.5 moderate, 0.8 large.
Enterovirus VP1	A coat protein of enteroviruses (the family of polio, coxsackie and echoviruses) used to stain tissue for evidence of infection.
Extracellular vesicles (EVs)	Small membrane packages released by cells into blood, carrying proteins and RNA.
Fibrinaloid microclot	A clot-like particle of misfolded fibrin seen under a microscope with an amyloid dye; its significance is disputed.
Flow-mediated dilation (FMD)	Ultrasound measure of how much an arm artery widens after a cuff is released; a test of the vessel lining.
FMT	Faecal microbiota transplantation: transferring stool from a healthy donor to a patient.
Genetic correlation (rg)	How much the genetic risk for two traits overlaps (from −1 to 1).
GPCR autoantibody (beta2, M3, M4)	Antibodies said to target receptors that nerves use to control blood vessels, heart rate and glands. Commercial tests exist; better methods have not confirmed differences between patients and controls.
GWAS	Genome-wide association study: compares millions of DNA letters between cases and controls to find regions linked to a disease.
Handgrip fatigability	The fall in grip strength across repeated maximal squeezes, and how well it recovers after rest.
Heart-rate variability (HRV)	Beat-to-beat variation in heart rhythm; higher high-frequency HRV reflects stronger vagal (calming) nerve activity.
HERV	Human endogenous retroviruses: remnants of ancient viral infections that make up about 8% of our DNA and can sometimes be switched back on.
HLA	Immune genes that present fragments of microbes to T cells; often linked to autoimmune disease.
HPA axis	The hypothalamus-pituitary-adrenal chain that controls cortisol release.
Hypocapnia	Low carbon dioxide in the blood, usually from over-breathing; it narrows brain arteries.
Hypocapnia / POSH	Low carbon dioxide from over-breathing. POSH means CO2 drops when upright; low CO2 narrows brain blood vessels.
IBS	Irritable bowel syndrome, a disorder of gut-brain interaction causing pain and altered bowel habit.
iciHHV-6	Inherited chromosomally integrated HHV-6: about 1 in 100 people carry HHV-6's genome in every cell, inherited from a parent. It can make blood HHV-6 tests look falsely 'high'.
iCPET	Invasive cardiopulmonary exercise test: a bike test with catheters in the heart's arteries to measure filling pressures and oxygen use.
IDO metabolic trap	A model in which the tryptophan-processing enzyme IDO1 gets stuck in a low-activity state because of substrate inhibition and inactive IDO2 variants.
IGF-1	A blood marker of growth hormone activity.
IgG subclass deficiency	Being low in one of the four flavours of the main blood antibody, which can make some infections more frequent.
Immunoadsorption / apheresis	Filtering antibodies out of the blood through a column, like a short-term dialysis for antibodies.
Indole-3-propionate (IPA)	A protective tryptophan breakdown product made only by certain gut bacteria.
Invasive CPET (iCPET)	An exercise test with catheters in the heart and an artery, measuring filling pressures and oxygen use directly.
Locus (loci)	A region of the genome. A GWAS 'hit' is a locus, not necessarily a known gene.
Low T3 syndrome	Low active thyroid hormone with normal TSH, often seen in chronic illness rather than thyroid disease.
LPS / sCD14 / LBP	LPS is a component of bacterial walls. sCD14 and LBP are proteins the body makes in response to it, used as markers of bacteria leaking into blood.
Lytic vs latent	Herpesviruses hide quietly in cells for life (latent). When they 'wake up' and make new virus or viral proteins, that is lytic activity, or reactivation.
MAIT cell	An unconventional T cell that patrols mucosal surfaces such as the gut and responds to bacterial metabolites.
Mannose-binding lectin (MBL)	A blood protein of the innate immune system that tags microbes for destruction; low levels are a common inherited variation.
Mast cell activation	Over-release of histamine and other mediators by tissue immune cells, causing flushing, itching, gut symptoms and sensitivity to foods, smells and drugs.
Mendelian randomisation (MR)	Using genetic variants as a natural experiment to test whether a trait (like EBV load) plausibly causes a disease.
Metabolomics	Measuring hundreds to thousands of small molecules (sugars, fats, amino acids, drug traces) in a sample at once.
Microbiome	The community of bacteria, viruses and fungi living in the gut.
MR spectroscopy (MRS)	An MRI method that measures chemicals such as lactate, choline and glutathione in brain tissue, and can estimate tissue temperature.
NASA lean test	10 minutes leaning with shoulders on a wall while pulse and blood pressure are taken each minute. Cheap, but not validated for formal diagnosis.
Natural killer (NK) cell	A white blood cell that kills virus-infected and abnormal cells without needing to be trained first.
NfL / GFAP	Proteins released when neurons (NfL) or astrocytes (GFAP) are damaged; raised in MS, stroke and dementia.
OCT-A	Optical coherence tomography angiography: a non-invasive scan of the tiny vessels in the retina.
Odds ratio per allele	How much one copy of a variant changes the odds of disease. 1.08 means 8% higher odds, which is a very small effect.
Open-label	Everyone, patients and doctors, knows who got the drug. Prone to placebo and expectation effects.
Orthostatic intolerance (OI)	Symptoms (dizziness, fog, fatigue, nausea) that come on when upright and ease when lying down.
Oxford criteria	A 1991 definition of CFS needing only six months of disabling fatigue; it does not require post-exertional malaise and captures broader fatigue states.
OXPHOS	Oxidative phosphorylation: the mitochondrial process that uses oxygen to make most of the cell's ATP.
Oxygen extraction	How much oxygen tissues remove from the blood that flows through them.
Passive transfer	Injecting antibodies from a patient into an animal to see whether the symptoms travel with the antibodies. It is the strongest available test of whether antibodies cause something.
PBMC	Peripheral blood mononuclear cells: lymphocytes and monocytes isolated from a blood sample.
PDH (pyruvate dehydrogenase)	The enzyme that lets the product of sugar breakdown enter mitochondria; if blocked, cells make more lactate.
Peak VO2	The most oxygen the body can take up and use per minute at maximal effort; the standard measure of aerobic capacity.
PEM (post-exertional malaise)	A worsening of symptoms after physical, mental or emotional effort, often delayed by hours to a day and lasting days or longer.
Pericyte	A cell wrapped around capillaries that controls their width and helps keep their walls intact.
Placebo response	Improvement in people given a dummy treatment, from expectation, natural fluctuation and enrolling when at a low point.
Plasmalogens	Membrane fats made partly in peroxisomes; thought to protect against oxidation.
Polygenic	Risk spread across thousands of variants, each with a tiny effect.
PORH	Post-occlusive reactive hyperaemia: the skin blood-flow surge after a cuff is released, a test of small vessels.
Post-infective fatigue syndrome (PIFS)	A lasting illness of fatigue, pain, poor concentration and unrefreshing sleep that begins after a documented infection.
POTS	Postural orthostatic tachycardia syndrome: heart rate rises by 30+ bpm (40+ in teenagers) within 10 minutes of standing, without a big blood-pressure drop, plus symptoms.
Pre-analytic artefact	A difference created by how blood was collected, spun or stored rather than by the illness. TGF-beta in ME/CFS turned out to be one.
Preload	How much blood fills the heart before each beat. Low preload means each beat pumps less.
Proteomics (SomaScan, Olink)	Measuring thousands of proteins at once; SomaScan uses DNA-like aptamers, Olink uses antibody pairs. Both can be distorted by how a sample is handled.
Red cell volume	The total amount of red blood cells in the body, measured with tracer methods, not by a standard blood count.
Renin and aldosterone	Kidney and adrenal hormones that make the body hold salt and water when blood volume is low.
SDH (succinate dehydrogenase)	A mitochondrial enzyme whose staining is used as a marker of mitochondrial content in muscle sections.
Seahorse assay	A plate-based instrument measuring how fast cells use oxygen and make acid, used to estimate mitochondrial function.
SF-36 physical function	A 0-100 questionnaire score of physical ability; healthy adults score around 90-100, many ME/CFS patients 20-50.
Sham-controlled	The comparison group gets a fake version of a procedure (for example blood passed through a filter that removes nothing) so no one knows who got the real thing.
Sham-controlled trial	A trial where the comparison group gets a convincing fake procedure, which matters enormously here because 35% improved on placebo in the rituximab trial.
Shotgun metagenomics	Sequencing all DNA in stool to identify species and their genes, more detailed than 16S sequencing.
Small-fibre neuropathy (SFN)	Loss of the thinnest nerves in skin and organs, which carry pain/temperature and control sweating and blood vessels; diagnosed by counting fibres in a 3 mm skin biopsy (IENFD).
SMR	Standardised mortality ratio: observed deaths divided by the number expected for people of the same age and sex.
SNP heritability	The share of variation in risk explained by common DNA variants measured on chips.
Switched-memory B cell	A B cell that has already met an antigen and changed the class of antibody it makes. Fewer of them, with more untrained (naive) cells, can suggest an unusual pattern of immune stimulation.
Synacthen (ACTH stimulation) test	An injection of synthetic ACTH, with cortisol measured afterwards, to check whether the adrenal glands work.
T-cell exhaustion	A worn-out state that killer T cells enter after long periods of fighting an infection or tumour; they stay present but respond poorly.
TRPM3	A calcium channel in cell membranes. One group reports it works poorly in patients' NK cells.
TSPO-PET	A PET scan using a tracer that binds a protein raised in activated microglia and astrocytes, used as a marker of brain inflammation.
Type I / type II fibres	Type I fibres are slow, endurance, mitochondria-rich; type II are faster and more reliant on glycolysis.
Vasopressin (ADH)	The pituitary hormone that makes the kidneys keep water.
Ventilatory (anaerobic) threshold	The point during exercise when breathing rises faster than oxygen use because muscles lean more on lactate-producing fuel use.
Virome / phages	The viruses in the gut, most of which infect bacteria (bacteriophages).
WASF3	A protein that, when raised, disrupts the assembly of mitochondrial energy-producing complexes.
XMRV	A mouse-derived retrovirus once reported in CFS. The finding was later shown to be lab contamination and retracted.

About this atlas

Built by the Solving ME/CFS project from published papers, preprints and our own re-analyses of public datasets. Grades are editorial judgements about replication, not formal GRADE ratings. Corrections and missing studies are welcome. The underlying data is in atlas-data.json.