Preprint Seven replicated genomic associations of myalgic encephalomyelitis/chronic fatigue syndrome: a biobank study, 2026, Slaughter, Ponting et al.
mariovitali
Senior Member (Voting Rights)
I would also like to see rs261902 (BICD1) being further investigated as @forestglip mentions,it appears to be a solid finding.
Last edited:
Chris Ponting
Senior Member (Voting Rights)
Thank you @ME/CFS Science Blog for your careful reading of the preprint. Some of this we can clear up: (i) human variants (including those on the array) are heavily skewed towards being rare & (ii) few SNPs are correlated because we didn’t use imputed variants, only the markers on the array.
Also, we know from our null & realistic simulations that TarGene shows good performance: https://academic.oup.com/biostatistics/article/26/1/kxaf030/8268272.
True, we still have work to do on why most variants are protective. We think we know why & will update. Thanks!
Thanks @Chris Ponting
I saw that in DecodeME, it seems to be skewed the opposite way, towards very few of the low p-value variants being rare. Is this related to the algorithm used, where maybe you've noticed TarGene returns a high number of rare variants when testing other datasets as well?
Chris Ponting
Senior Member (Voting Rights)
We have, indeed, tested for more common variants, such as an FTO variant, rs1421085 (allele frequency 42% in non-Finnish Europeans): Figure 5 in https://academic.oup.com/biostatistics/article/26/1/kxaf030/8268272?login=false.
As a side note, we show that other, linear model, approaches can overestimate the effect size.
Thanks, though mainly what I'm curious about is whether the high proportion of significant results in this study being rare is expected for TarGene. Even though rare variants are more common, the statistical power for these is lower, which I think is the reason they are less likely to be significant in typical GWAS.
Here is the distribution of MAF for the 176 significant variants given in the S1 sheet for this study:
To compare to the distribution of MAF in general, here are all 8,902,782 QCed variants in DecodeME:
I also filtered to only the 407,128 variants with INFO=1, and it looks similar (I'm assuming INFO=1 means directly genotyped as opposed to imputed):
Thus, the TarGene analysis seems to be skewing even more towards rare variants than would be expected based on the frequency of rare variants tested in a GWAS (though this is just a crude comparison with DecodeME. I can't be sure the UKB GWAS array didn't skew more towards rare variants.)
And for comparing to significant variants, here is the MAF distribution just for low p-value variants in DecodeME, using an arbitrary cut-off of -log10p > 5, both with and without filtering to INFO=1, showing the tendency for significant variants to have higher MAF in that study:
The skew towards rare, along with the unexpectedly high number of significant/replicated variants with a relatively small sample, is unexpected for me. It makes me wonder if maybe the algorithm is for some reason returning overly low p-values specifically for variants with low MAF.
Maybe a QQ plot split by MAF would be helpful to see whether there is inflation of p-values specifically for low MAF variants.
Yes, the protein encoded by the CLYBL gene is the enzyme that catalyzes the last step in the itaconate shunt: citramalyl-CoA → pyruvate + acetyl-CoA.
It's nice to have a highly significant, replicated GWAS hit that points to a key element of the itaconate shunt, but what I found most encouraging is that the shunt hypothesis predicts the same direction of the effect of CLYBL as found by Slaughter et al.
The itaconate shunt hypothesis predicts that more CLYBL will result in LESS ME risk because CLYBL is one of the pathways by which the two toxins (itaconate and itaconyl-CoA) are cleared from mitochondria. Using eQTL analysis, Slaughter et al. found that more CLYBL expression (in brain putamen) is associated with less ME risk.
V.R.T.
Senior Member (Voting Rights)
Hi Rob, good to see you on here again. It is interesting to see this gene pop up. As a layman I feel it's odd we only see it here and not in DecodeME if it's crucial to MECFS pathology though.
I was wondering if looking at recent developments like the DecodeME data, the recent long covid dopamine paper, the forthcoming CRH autopsy paper and so forth had prompted you to think of any other hypotheses besides your itaconate shunt idea?
And also if you forsee a point where you can definitely prove or disprove the itaconate shunt in the near future?
The genes that matter for ME should be seen as the UNION of those found by all the GWASs on ME.
Earlier this year, I asked a group of 15 senior ME researchers if they had formulated any new hypotheses for ME based on the DecodeME GWAS hits. They all shook their heads, "no."
Last week, I visited colleagues at the University of Luxembourg who were "there at the beginning" of the itaconate immunometabolism idea when Karsten Hiller's lab was still at LCSB. Carole Linster is an expert on CLYBL whose work would complement the groups at Stanford and the University of Utah who are working on ACOD1. She was visibly excited by the Slaughter et al. preprint.
What we need, and what I am working on with these and other collaborators, is 13C-tracer kinetic evidence of a significant itaconate shunt flux (measured by mass spectrometry) in cells that display an ME phenotype. Amanda Ooi's Biolog experiments in Ron Davis's lab at Stanford give us the phenotype of decreased NADH production in ME PBMCs. We are planning 13C-glucose and/or 13C-glutamine experiments with our colleagues at Utah and possibly in Luxembourg. These experiments require a lot of cells from patients and controls, and the tracers are expensive, but it's the best way I know to demonstrate increased shunt flux in ME-affected cells.
As you know, I always emphasize that ME is probably cell autonomous. This means every sample of ME cells (from any tissue) may contain a large predominance of healthy cells. I think this possibility is the most difficult hurdle faced by ME research; it may dilute every informative signal in every patient sample. I'm hoping the tracer experiments can overcome this hurdle and rule the itaconate shunt hypothesis in or out.
Jonathan Edwards
Senior Member (Voting Rights)
Indeed, I think members here have come up with at least a dozen new theories. And James is already well into a project based on CA10.
V.R.T.
Senior Member (Voting Rights)
Thank you for your detailed reply about the itaconate project. I hope you can develop the tracers swiftly and answer this question once and for all. Do you have any kind of timeline?
I echo @Utsikt's frustration vis a vis the lack of DecodeME theorising. It is arguably the strongest lead we currently have so to discover 15 of the top MECFS researchers aren't grappling with what it might mean is disappointing.
I often say that half of what I know about ME I learned from patients. It will not surprise me if a smart, motivated patient or group of patients solves this disease. I know there is a group of GWAS-knowledgeable patients following the DecodeME hits into neurobiology. It would be fabulous if they discover the key.
It's also possible, of course, that those 15 scientists each have a hot new hypothesis based on DecodeME and simply want to test it before they tell me about it.
Last edited:
The timeline is always the same: ASAP.
Concentric_Spiral
Senior Member (Voting Rights)
Hoopoe
Senior Member (Voting Rights)
@Chris Ponting I'm curious about the genetics behind the two age peaks in ME/CFS. Would it be feasible and make sense to perform two analyses of DecodeME data, one for onset before the age of about 22 and one with a later onset?
Last edited:
Hi Rob - as an ME/CFS patient, I’ve been v interested in your Itaconate Shunt Hypothesis for a while. Best theory out there IMO. It’s what brought me to this forum.
Myself, Jonathan Edwards and others here, have been working recently on the possibility of a larger loop within which your itaconate shunt explanation of PEM and brain fog takes place — it depends on your theory. This is, I hasten to add, lay patient-led theorizing, with me as penholder, a non-scientist — but with Jonathan and other scientists here nudging and pruning along the way. A latest draft of a parsimonious summary of the theory’s explanation for ME/CFS onset and prolongation is here (again, health warning: drafted by me, a non scientist, so it’s teleological at best - but it’s short):
I’d be really glad if you were to take a look.
Chris Ponting
Senior Member (Voting Rights)
Thanks. We (DecodeME) should have asked the direct question: "At what age did you first have ME/CFS symptoms?" We didn't, rather we know their age of first symptoms only in age-ranges, and many people indicated >10 years, which doesn't help. These ">10 years" people were not used for Audrey Ryback et al.'s analysis. We could do the same (i.e., remove the ">10 years" people) but this would greatly reduce the comparison's power. So my answer is "perhaps"... Thanks.
Hoopoe
Senior Member (Voting Rights)
Is it difficult to contact the participants again to collect age of onset data?
hotblack
Senior Member (Voting Rights)
I wanted to give some time to read and digest this. I’ve been through a fair bit of the paper and this thread but probably need a few more listens to digest it more, there’s a lot there! But here’s my initisl notes, thoughts, questions. Not to anybody in particular but maybe they’ll help discussion.
- First off this looks like the sort of good rigorous analysis we’ve come to expect from the group in Edinburgh. Thank you as ever.
- The preprint server does a great job of formatting the paper in a way screen readers can read it, so I was able to listen, nice!
- I don’t understand it all so have some more reading and learning to do!
- That there was replication seems significant.
- Even if it wasn’t possible with DecodeME, what this tells us could be really interesting and important. The problem is working out what this study is telling us, it feels a bit like 42 being the answer to the ultimate question but we need to understand what the question is.
- It seems to reinforces the neurological signal. But what do the loci tell us about the condition? What do the differences tell us about the cohorts?
- Can all the different genetic findings in different cohorts all be pointers to something still hidden? More x’s on the treasure map leading us to something still buried?
- I’m interested in the methodology here. Can we use these approaches and techniques on different datasets or subsets, on DecodeME data?
- We’re comparing different groups with different criteria and different methods at the moment. It is perhaps unsurprising we get different results. And maybe more surprising we got replication with separate cohorts.
- This brings us back to looking at groups by severity imho. This is s something we talked about before on DecodeME, can or should we subgroup by severity? I think the numbers were said to be too small to look at severe only but does this approach change that? Or if not what about excluding severe?
- Study/patient selection criteria seem more important than ever. Does this study give us any other insights into how to best to define groups to best understand the condition in other studies?
- Findings like the B12 one are interesting given how that’s a topic which has cropped up and been an area of interest for many over the years.
- The difference between a poorly performing and completely broken B12 gene/pathway reminds me of some of the sickle cell treatments where through CRISPR treatments a different functioning pathway can be activated. Maybe there’s a similar possibility for some patients here?
- First off this looks like the sort of good rigorous analysis we’ve come to expect from the group in Edinburgh. Thank you as ever.
- The preprint server does a great job of formatting the paper in a way screen readers can read it, so I was able to listen, nice!
- I don’t understand it all so have some more reading and learning to do!
- That there was replication seems significant.
- Even if it wasn’t possible with DecodeME, what this tells us could be really interesting and important. The problem is working out what this study is telling us, it feels a bit like 42 being the answer to the ultimate question but we need to understand what the question is.
- It seems to reinforces the neurological signal. But what do the loci tell us about the condition? What do the differences tell us about the cohorts?
- Can all the different genetic findings in different cohorts all be pointers to something still hidden? More x’s on the treasure map leading us to something still buried?
- I’m interested in the methodology here. Can we use these approaches and techniques on different datasets or subsets, on DecodeME data?
- We’re comparing different groups with different criteria and different methods at the moment. It is perhaps unsurprising we get different results. And maybe more surprising we got replication with separate cohorts.
- This brings us back to looking at groups by severity imho. This is s something we talked about before on DecodeME, can or should we subgroup by severity? I think the numbers were said to be too small to look at severe only but does this approach change that? Or if not what about excluding severe?
- Study/patient selection criteria seem more important than ever. Does this study give us any other insights into how to best to define groups to best understand the condition in other studies?
- Findings like the B12 one are interesting given how that’s a topic which has cropped up and been an area of interest for many over the years.
- The difference between a poorly performing and completely broken B12 gene/pathway reminds me of some of the sickle cell treatments where through CRISPR treatments a different functioning pathway can be activated. Maybe there’s a similar possibility for some patients here?
Last edited: