Preprint Dissecting the genetic complexity of myalgic encephalomyelitis/chronic fatigue syndrome via deep learning-powered genome analysis, 2025, Zhang+

Thanks, and to @forestglip for the explanations.

You could say that a GWAS is essentially scanning all of human biology to find clues about what processes have gone wrong. But with a relatively small study, such as DecodeME at 20k, it's likely you will miss a lot of clues.

 

Thanks for pointing that out. I’m vaguely aware of the connections to NFkB, though don’t know comprehensively how the changes identified in this paper would lead to changes in that signaling.

The one thing I’m not quite sold on is the fact that the proteasome is a large complex and the identified proteins all encode for the core 20S subunit (as opposed to the “immuno-subunits”).

I don’t know what the biological relevance would be of encoding more of a certain subunit but not others if those extra subunits don’t form functional complexes on their own. When I see proteasome pathways come up in transcriptomic studies, typically it’s nearly all the subunit genes that come up as differential.

If it was a situation of mutation in one subunit —> less efficient proteasome —> more proteasomes are produced to compensate, then I would understand. But with just two subunits, I can’t figure out what the story would be.

It’s possible that the proteomics assay just didn’t pick up on other subunits, there was too much variability to all come up as differential, or individual subunits might have functional roles that are not well characterized.

[Edit: I just realized it was less of those subunits in ME/CFS and not more, in which case it might end up creating a rate limit on proteasome assembly?]

Either way, papers like this are best served as launch points for hypothesis generation, so I don’t think that confusion is necessarily a point against the paper itself. I’d just caution against using these findings to interpret anything about the disease given how often I see proteasome-related findings come up in pretty much every context where there’s no other strong signal to focus on.

 

One thing we discovered when trying to work out the mechanism of RA was that in evolution gene products that are used for one thing can be repurposed to do completely different things in different tissues.

So the gene that codes for a joint lubricant is also used to code a protein involving in making blood platelets. VCAM-1 is used to control cell movement across joint surfaces and is also used in (later evolving) bone marrow to support B cell development. The gene for collagen II, which is the basis of cartilage, is also involved in neurodevelopment and the eye. Another B cell survival factor is expressed in brain.

The upshot of this seems to be that two things might apply. A gene might come up on a risk analysis that is expressed in both T cells and synapses and one of those may be a complete red herring. The other is that the link may actually be the clue to a specific 'Achilles heel' mechanism that involves both tissues. MS might be that - a weak spot in the signalling control that allows certain B cell to survive in brain.

 

That's interesting that Lansoprazole and ranitidine are both shown - both used to reduce stomach acid. Makes me think of Dr Chia and his link of Cocksakie virus in the stomach, duodenum, and illeum. He described that the virus was also linked to Gastritis. UK had a recorded Coxsakie B4 outbreak in Ayrshire Scotland that led to an ME/CFS outbreak in the early 1980's.

Maybe clutching at straws because I don't know how that would link to genetics.

 

Okay, having had a chance to read more in-depth, some assorted thoughts:

• I think my previous thoughts re: proteasome findings still apply. The fact that it's genetic mutations related to proteasomes assuages concerns that it's some downstream homestatic stress issue causing those findings. I just still have my concerns since it's such a large pathway that often comes up in the absence of other strong signals
• The AUC (metric of performance of a classification model) is not awful but not great--for reference, 0.5 means that the model is performing exactly as you would expect if it did nothing and was just assigning randomly. >0.9 would be considered a fantastic model. In practice, I haven't seen many polygenic models in a GWAS that score better than 0.75. The fact that it was consistent in the independent test cohort is also very encouraging.
  
• I think they have good justification for focusing on Loss of Function mutations and rare variants, though this limited search space should be kept in mind when interpreting the results. There might be a more common variant or a less deleterious mutation that is just as relevant to ME/CFS which simply wasn't included in the study.
  
• It might have been beyond the scope of the analysis, but theoretically if they had access to the fastQ files for the scRNA-seq analysis, they might be able to verify whether those individuals had any of the identified mutations as well.
  
• I remember reading the scRNA-seq paper that they got the data from and being highly skeptical about the accuracy of their immune subtype labeling on the basis of RNA-seq alone. It doesn't really change this paper's finding in that regard, it just means that what they're calling CD4+ cytotoxic T cells may not be exactly CD4+ cytotoxic T cells
  
• Similar to others, I don't see too much of interest coming from the overlap with other diseases.
  
• Overall, I think the top genes picked out by this paper might be interesting for the exact fact that they are quite diffuse across proteins and biological processes. I'd be more interested to see where the proteasomic proteins might overlap with those other top hits, rather than focusing on the proteasome itself as a potential disease mechanism.

 

Thanks @jnmaciuch .
I like the idea that CD4+ cytotoxic cells may not be exactly CD4+ cytotoxic cells.
I am getting the impression that although it is likely that there are some really useful hits amongst these data it may not be that easy to tell which ones they are amongst the 115?

My gut feeling is that when we know what DecodeME shows, using a different approach, things may become a lot clearer.

 

I feel semi-confident that they are mostly cytotoxic from the CCL5 and GZMB levels. In my own experience doing scRNA-seq on T cells, that expression pattern is usually quite strong. Though from what I recall of their clustering, they didn’t get a clear divide between their CD4s and CD8s. So might be a mix of both in there.

I think someone with more expertise might be able to give a better answer. From my perspective, an association is an association especially if it’s validated in another cohort. There might be some false positives among there, though they’ve already [edit: pared] down their subset quite a bit.

So theoretically, they might all be biologically relevant somehow, it’s just a matter of whether that relevance can be sorted out meaningfully.

If it’s relevant but via 12 steps of indirect mechanisms, it’s probably not going to be useful for the purposes of zeroing in on a pathological mechanism. In that regard, you’re exactly right, since we have no way of knowing the difference until we start investigating each candidate in earnest.

 

Regarding CD4 cells Mark Davis gave a talk at the NIH Fatigue Conference/Webinar Sep 2021 that Jamie Seltzer transcribed on Twitter. He presented results for an autoimmune CD4 cell subtype that was elevated in males and females in ME/CFS (small patient sample size).


I cant find the talk to see how they defined autoimmune CD4 cells. Maybe someone here can find it.

What I remember from about that time was that they needed lots of tubes of blood to obtain sufficient quantities of a rare T cell type to quantify and analyse (I donated blood and asked why so many tubes were needed). Sadly the researcher performing the work left Stanford the following year so I don't know if they ever followed up on this or if it was a dead end.

 

I have yet to be convinced of a bona fide autoimmune T cell other than maybe with AIR mutations.

 

I’m guessing it wouldn’t make a great deal of sense of people to start pursuing 115 different candidates on the basis of a small study and a potentially over fitted model.
But comparing these results with other studies might be very useful eg the DecodeME GWAS and the precision life work on DecodeME

As would, as @Sasha suggests, applying this new approach to DecodeME data, which will hopefully be available later this year.

That would be a big shift in the field. Until very recently, we haven’t been in a position where researchers could compare study results looking at very much the same issue (genetics here), and certainly not to have access to a very large data set that could be used and analysed to check other findings.

 

Talk by Mark Davis on the rare CD4 T cell type is here starting at 4hrs 48mins
https://videocast.nih.gov/watch=42563

He talks about classical autoimmune diseases - Celiac disease and mice with EAE autoimmune disease where they saw bursts of CD4 and gamma delta T cells. The gamma delta T cells expressed IL17 and apprently not much is known about them, and they saw about 1 in 1000 CD4 cells were this particular rare type. Then moves onto Lupus and MS patients, and finally ME/CFS. So he covers how they discovered these cells in autoimmune diseases and then they showed up in some ME/CFS patients.

 

Yes im definitely looking forward to DecodeME results. I think it will end up being a process where investigators try to come up with a story of a biological process that would incorporate some subset of the genes [edit: e.g. a bi-directional regulatory relationship where mutations at multiple points might result in something getting thrown out of whack] and then that hypothesis would be testable directly. I don’t think investigation of any of the genes individually would yield much use since there’s no evidence of ME being a Mendelian disorder.

 

Agreed, it’s not a Mendelian disorder.

But each individual hit could point to a biological process that plays a part in the illness. It could be one of several parts. Or it could reflect some of the likely-several different sub groups.

Ideally, you would find several genes that played a part in the same pathway, or the same broad biological process.

But I think they could be relatively simple interpretations of the data. Hopefully, we’ll see the results before too long, with data available to researchers not long after that.

 

Well, the T cells in coeliac aren't autoimmune if they are reacting to gliadin. It's the B cells reacting to tissue transglutaminase that are autoimmune! And mouse EAE isn't a spontaneous human autoimmune disease so probably has nothing to do with MS! And so on.

When people talk of 'they discovered these cells' in immunology it is worth adding a spoonful of salt.

 

I agree. And if those who know the methodology are happy that there are real associations here I think it is worth underlining how important this all is.

When the first draft of the Beentjes paper suggested that there was proof that ME/CFS was not 'psychological' (to paraphrase) I thought that might be a little premature and ill-judged. But as Chris P has himself said, what is happening now is not just about any one study. This Zhang study may be the first to hit the public domain where we can truly say there must be a biological causation because there is an identifiable genetic component. In the long run that is not what matters and I personally think that GWAS data will be easier to get my head around, but if we have reliable data that point to genetic links relevant to T cells and synapses then I think we have got a purchase on what we are looking for. And maybe NAD is in there as well from what they say.

With RA our best shot at a mechanism had 55 steps, implicating maybe 5 functional domains. We had accumulated, from various sources, genetic evidence for 3. Fitting things together enough to go ahead with treatment design did not take very long. ME/CFS may be a level more heterogeneous but I think our claim in 2016 that it was a solvable problem will turn out to be valid.

 

You've said elsewhere that we won't be able to say that ME/CFS is a disease, rather than an illness, until we can demonstrate pathology. If researchers reliably show that there's an identifiable genetic component, can we call it a disease?

If so, does this Zhang study allow us to do that, or do we need more replication?