Preprint Initial findings from the DecodeME genome-wide association study of myalgic encephalomyelitis/chronic fatigue syndrome, 2025, DecodeMe Collaboration

[EndME]

In ME/CFS we have the problem that we don't know what diagnostic criteria are best.

We can create different diagnostic criteria according to what we believe best describes the disease, but that's just an opinion-based process.

If we had an objective method for defining ME/CFS, then we could develop and refine diagnostic criteria until they could capture the disease with high accuracy.

I'm wondering if a GWAS can be used to provide some much needed objectivity and help validate or refine the diagnostic criteria, or discover subtypes.

Yes but how would that work? If you'd just include people with MS, or Ank Spond in your ME/CFS cohort you might artificially just get more significant findings but your diagnostic criteria to get there will mean that you're not actually doing anything meaningful anymore. Maybe a certain Fukuda cohort ends up having many hits due to MS etc.

DecodeME showed that you can just ask people whether they have ME/CFS (edit: in the sense as mentioned by Andi below) and actually get some genes from that. But it isn't clear whether that is applicable to other cohorts and to what extent (the failure to replicate suggests it won't work for badly selected cohorts but we don't know what the results will look like in clinical cohorts or say a "German DecodeME cohort"). I think a GWAS in a country including a clincal cohort and a DecodeME like cohort could be a useful next step to understand how the signals behave then.

I think it should be fairly simply to do a grading how symptoms correlated to significant gene presence, but I don't think that's necessarily meaningful for identifying diagnostic criteria.

The problem here would be the same as always: Against what value do you define high accuracy of disease? Every criteria is 100% accurate against itself and we don't know any diseases processes as of yet.

 

[Andy]

DecodeME showed that you can just ask people whether they have ME/CFS

But that isn't what we did. We asked our participants to confirm that they had a diagnosis from a medical professional AND we put them through a screening questionnaire to assess them against CCC and IOM criteria.

Given that we have confirmed that this method has produced a genetically reasonably well defined cohort I don't see much value in deviating from that method.

 

[Jonathan Edwards]

I am struggling with phone access to net but I think @Hoopoe 's idea is a good one but may not be realistic in practice.

 

[EndME]

But that isn't what we did. We asked our participants to confirm that they had a diagnosis from a medical professional AND we put them through a screening questionnaire to assess them against CCC and IOM criteria.

Given that we have confirmed that this method has produced a genetically reasonably well defined cohort I don't see much value in deviating from that method.

Yes, that is what I meant. What I meant with a "German DecodeME cohort" would be to just repeat this procedure (supposedly without telling them that you're doing this) and also have a clinical cohort as comparison (that also undergoes the questionnaire screening).

 

[Kitty]

I guess the caveat is that MECFS might be two similar looking diseases like DR4 encouraged arthritis and B27 encouraged arthritis. Ritux is no good for the latter.

But if they were treated as one disease because there was no practical way to distinguish them, I guess the drug would tell you a good bit about where not to look for an answer for the non-responders?

 

[Simon M]

Exactly @ME/CFS Science Blog. Any of a hundred genes can lead us to a treatment strategy that is not dependent on any of them. Ritux for RA works whether or not you have DR4, but DR4 pointed us to it.

Thanks, both. That have been my understanding, but I wanted to check it.

ADDED and I think this is an important message to convey to people who understandably wonder what the relevance of this big genetic study is to them when many won't have the relevant genetic differences, and heritability is so low.

I know my blog didn't address this

 

[Simon M]

The problem here would be the same as always: Against what value do you define high accuracy of disease? Every criteria is 100% accurate against itself and we don't know any diseases processes as of yet.

I would say the gold standard is diagnosis by expert clinicians. We know from a couple of studies that half of GP diagnoses are wrong – and almost all cases because that had been other, undiagnosed diseases (mostly biomedical, but also psychological) that explain the symptoms.

I suspect the specific criteria are less important, so long as we have Pem. (I'll be interesting to see, though, how.DecodeME cases would stack up against expert diagnosis as they, probably uniquely, have a decent definition of PEM, which does seem to be a very unusual symptom).

By expert clinicians, I mean those who are very familiar with the illness, and I will resource. The two papers that found about 50% of GP diagnosis were were led by Julian Newton and Peter White respectively. Plus, I think they had bigger teams for this to do a good job.

GP's unfamiliar with the illness and often dismiss it. Even specialist clinics are frequently held up by lack of resources, or by impeded by beliefs.

I believe that DecodeME is the best large cohort we have available, and the best possible job that could be done.

But it would still miss cases where there are undiagnosed alternative diagnosis. That's why I think it's use of a good pen question is so important.

 

[mariovitali]

I am sharing the initial results from the analysis I have been performing using 5 different reasoning engines to suggest causal hypotheses given the GWAS results and the candidate genes. @Hutan it was difficult to extract text because of the table.

I will not provide the details on how it was done here, I am providing this information in hope that helps experts to formulate their hypotheses.

Tagging @Chris Ponting @DMissa @Jonathan Edwards

EDIT : The hypotheses generated are not discussed here but on a later post below.

First the areas of agreement in the 5 hypotheses generated by the reasoning engines :








And here are the areas of disagreement.

 

[Jonathan Edwards]

Hi @mariovitali ,

To me those responses are just a rehash of all the trendy stuff constantly recycled by people thinking inside a box with no real understanding of global biodynamics. There is no actual explanatory model, just buzzwords.

I think asking @jnmaciuch or @Snow Leopard and a few others here, some with little or no technical training in the field but a good dose of common sense, is likely to provide a more interesting answer.

 

[mariovitali]

Hi @mariovitali ,

To me those responses are just a rehash of all the trendy stuff constantly recycled by people thinking inside a box with no real understanding of global biodynamics. There is no actual explanatory model, just buzzwords.

I think asking @jnmaciuch or @Snow Leopard and a few others here, some with little or no technical training in the field but a good dose of common sense, is likely to provide a more interesting answer.

Thank you for your reply. Actually the responses are not included in my message (=the hypotheses) . One hypothesis text for example states the following :

ME/CFS arises when a genetically primed bottleneck in endosomal/ER trafficking and ER‑phagy (RABGAP1L, ARFGEF2, CCPG1) meets oxidative–lipid stress on mitochondria and endothelium (FBXL4, PRDX6, PEBP1, HFE), tipping immune checkpoints toward non‑resolving activation (BTN2A2, TNFSF4/RC3H1, TRIM38) and culminates in microvascular hypoperfusion (SERPINC1, PTGIS) and dysautonomia (CA10, PEBP1), with epigenetic locking (SUDS3) that sustains the state after diverse triggers (viral or toxicants such as organophosphates).

How the GWAS hits define the core failure points
- Vesicle and receptor trafficking bottleneck
- RABGAP1L and ARFGEF2 (BIG2): control Rab/ARF-dependent endosomal and Golgi trafficking. A deficit here impairs antigen processing, cytokine receptor recycling, and GPCR trafficking (including adrenergic/muscarinic receptors), prolonging or distorting signaling and antigen persistence.
- CCPG1: an ER‑phagy receptor that clears stressed ER. Lower ER‑phagy capacity makes secretory/immune and barrier cells intolerant to acute ER stress (infection, toxins), amplifying the unfolded protein response and DAMP signaling.
- Energetics and oxidative resilience
- FBXL4: maintains mitochondrial DNA and biogenesis; even partial impairment lowers ATP reserve and raises ROS under stress.
- Immune co-regulation and mucosal priming
- BTN2A2: butyrophilin family co-regulatory molecule shaping T cell responses (including γδ T cells); variants can skew immune tone toward either overactivation or poor resolution.
- OLFM4 (Tier 2 but a GWAS hit): neutrophil/intestinal stem cell protein modulating innate responses; links mucosal barrier events to systemic inflammation.
- Neuro–autonomic vulnerability
- CA10: brain/synaptic carbonic anhydrase–like protein (noncatalytic) that modulates synaptic architecture and excitability; variants can destabilize central autonomic networks and sensory processing.
- Transcriptional set-point
- SUDS3: SIN3–HDAC scaffold; shifts gene-expression programs (metabolic and immune), facilitating a chronic “stuck” state.

How Tier 1 genes reinforce these modules
- Immune activation and off-switches
- TNFSF4 (OX40L) and RC3H1 (Roquin-1): coherent pair—RC3H1 normally destabilizes OX40/ICOS mRNAs; variants combined with higher OX40L tilt toward sustained T cell help, autoimmunity, and poor resolution.
- TRIM38, ZNFX1: tune innate RNA-sensing and type I IFN—risk of a chronic, smoldering antiviral program even after acute infection clears.
- BTN3A3 (with BTN2A2): further modulates γδ T cell circuitry.
- Trafficking, degradation, and stress responses
- KLHL20, CSE1L, STAU1, DDX27: ubiquitin/trafficking, RNA handling, ribosome biogenesis—together lower the ceiling for rapid recovery from cellular stress.
- Redox–lipid injury and ferroptosis susceptibility
- PRDX6 (peroxidase/iPLA2), PEBP1 (RKIP; also scaffolds 15‑LOX to peroxidize phosphatidylethanolamine), HFE (iron handling): converge on lipid peroxidation control and ferroptosis threshold.
- B4GALT5: glycosphingolipid biosynthesis that tunes membrane microdomains in platelets/endothelium and immune cells.
- Microvascular tone and thromboregulation
- SERPINC1 (antithrombin) and PTGIS (prostacyclin synthase): prothrombotic/vasoconstrictive tilt → microclots and impaired perfusion.

Tier 2 proximity-only genes (lower confidence, but consistent with the model)
- FBXL4 (also a top GWAS hit, so still high importance), OLFM4, CCPG1: while they lacked eQTL colocalization in this analysis, their known functions fit the core axes above (mitochondria, innate/mucosal immunity, ER‑phagy).

 

[Evergreen]

Regarding pain, being discussed by @voner, @Simon M and others above, here's an excerpt of table 4 from Pendergrast et al. 2016. Scroll down for pain.



For what it's worth, I had little or no new pain for the first 5 years, with the exception of when I did GET (joint pain bad enough to see a specialist, vanished when I stopped GET). The only real pain I had when I was mild and mild-moderate was if I really and truly overdid it to a daft degree, and it receded within a few days. And sometimes feeling in my quads as if I had been exercising heavily the day before (more like muscle fatigue), despite not having done anything that should have made quads tired. A lot of pain now.

 

[Kitty]

And sometimes feeling in my quads as if I had been exercising heavily the day before (more like muscle fatigue), despite not having done anything that should have made quads tired.

That describes part of my pain very well, I've had it throughout the illness except for a couple of remissions. But when I relax the quads they burn too, and that's the worst of it. I've never had pain in any other muscle group.

I was diagnosed in the late 90s (after years of illness), and I remember reading something produced by the MEA. It said quads and intercostals were the most common pain sites.

 

[Evergreen]

That describes part of my pain very well, I've had it throughout the illness except for a couple of remissions. But when I relax the quads they burn too, and that's the worst of it. I've never had pain in any other muscle group.

I was diagnosed in the late 90s (after years of illness), and I remember reading something produced by the MEA. It said quads and intercostals were the most common pain sites.

Yeah, it's interesting that the aftermath you've described in other posts seems to be the opposite of the aftermath for me. I absolutely have to lie down and rest them and that eases the fatigue/burn (somewhat). But it seems like the central issue is the same.

 

[chillier]

A couple more thoughts and open questions

- cell type most clearly implicated is the neuron

- Which type of neurons are affected and where? Peripheral ones or Central ones, acetycholinergic ones or glutamatergic ones? Or any type anywhere. GWI might suggest acetylcholinergic, and the HOMER-SHANK-DLGAP complex that came up in Zhang et al's WGS study might suggest glutamatergic, but I get the feeling from decode that it's probably pan neuronal.

- Does a pathogen, probably a virus, ever actually infect a neuron directly? We have non immunological genes affecting neurons, immunological genes that don't necessarily affect neurons, probably explicitly immune related genes that specifically affect neurons too - but i suppose that doesn't have to mean a neuron is dealing with a pathogen directly. I know S4ME is not very keen on the viral reservoir idea aka a brain infection but why is that?

- What effect do the variants we see have on the neuron? I think the genes might suggest a dysfunction in the ability to form, maintain or regulate synapses in some way, but how exactly? Would it be excess synapse formation leading to excess firing or an inability to form them leading to a loss of connection. Or would the affect result in a change in scaling, synapses being more prone to scaling either up or down but not the other way. Perhaps the situation is more complex and leads to an inflexibility to change in either direction, or poorly constructed synapses lead to neurotransmitter spillover and more sketchy unstable firing patterns in some form. This is surely the kind of thing we could have a guess at with the evidence we have, and would otherwise be testable in cell culture.

 

[Nightsong]

Re pain in ME/CFS: muscle & joint pain was highly prevalent in the 1999 Jason community sample (link), and in the 2011 Nacul primary-care sample (link) - Nacul et al reported "joint (76% to 91%) and muscle pain (74% to 95%)". There was also a 2010 paper based on interviews with 50 pwME, including 10 severely affected, that attempted to characterise the types of pain they experienced.

There was also this from IOM:

Recent preliminary data from the Centers for Disease Control and Prevention’s (CDC’s) Multi-Site Clinical Study of CFS indicate that 80 percent of patients enrolled had experienced pain in the past week (Unger, 2013). Muscle aches and pains were the most common pain complaint (reported by 72 to 79 percent of patients), followed by joint pain (reported by 58 to 60 percent of patients) and headaches (reported by 48 to 56 percent of patients). Less common pain complaints included tender lymph nodes (37 to 39 percent), abdominal pain (32 percent), sore throat (25 to 28 percent), eye pain (23 percent), and chest pain (15 percent).1,2

Historical surveys also showed significant proportions of pwME experiencing pain - AfME's 2001 membership survey reported that "77% experienced severe pain because of the illness", for instance.

 

[Simon M]

cell type most clearly implicated is the neuron

How sure of this can we be at this stage? The two specific bits of evidence I'm aware of, the most relevant CA 10 for pain, which is directly affects neuron. However, there's also a microglial gene. Microglial are glia not neurons. Then there are the 13 tissues where candidate ME/CFS genes are overexpressed – all brain regions. But I don't think we know if they are glial cells or neurones. Or have I missed something?.

 

[Simon M]

Luckily gene X points to the problem: the dam is breaking!

Good analogy

One variant of the gene creates a slightly stronger protein than the other variant to support the dam

I'm pretty sure genes identified by GWAS make identical proteins, and variants instead alter gene regulation, usually not by very much. That's one reason why effect sizes are small.

So it may be that the dam is stronger because a bit more of this protein is produced. But the point is the same, this could still point to a good target for an effective drug

ADDED
I recently saw a pre-print or paper (sorry, can't find link) that found 92% of gene variants identified by GWAS were non-coding.

 

[chillier]

I'm sure of this can we be at this stage? The two specific bits of evidence that's the most relevant CA 10 for pain, which is directly affects neuron. However, there's also a microglial gene. Microglial are glia not neurons. Then there are the 13 tissues where candidate ME/CFS genes are over expressed – all brain regions. But I don't think we know if they are glial cells or neurones. Or have I missed something?.

That's fair. I'm using a lot of the single cell type annotations from proteinatlas.org as a bit of a guide, and then some of the descriptions of the genes. ARFGEF2 for example supposedly has a role in GABA receptor recycling from endosome to membrane, which would suggest a neuronal function to me. Others have evidence that suggest a role in neurite development (like KHLH20) or neuropeptide (PEBP), but it's true these roles don't need to be carried by the neurons themselves, but could be by supporting glia to affect the function of the neuron.

I'm also factoring in a bit the SHANK complex found in zhang et al WGS, which is post synaptic density on glutamatergic neurons. Could be a good exercise to examine which cell types each of the decode genes most likely point to.

 

[Mij]

I thought pain was really common. It doesn’t have to be extreme pain, and some of it might also be more aches, stiffness and soreness.

I've never had pain related to M.E in 34 years. I have experienced deep bone aching/soreness during the early PVFS stage and only during immune events like viral reactivations. I never described it as pain and it doesn't require medications.

I've never had inflammation markers based on my test results, so the term M.E doesn't fit my experience.

 

[Hoopoe]

I know S4ME is not very keen on the viral reservoir idea aka a brain infection but why is that?

The thinking is that in the terminal stage of the illness, the immune suppression of the virus would fail. Since symptoms and signs of brain infection don't seem to be a common phenomenon in terminal cases, brain infection doesn't seem convincing. There are going to be patients who also have some persistent infection or difficulty controlling infection but it seems to be small minority so there's no clear link.

I know there's some evidence that viruses can produce ongoing responses even while fully suppressed. Van Elzakker was interested in this idea.