Preprint Replicated blood-based biomarkers for Myalgic Encephalomyelitis not explicable by inactivity, 2024, Beentjes, Ponting et al

Good point. My hypothetical plot is simply based on the cohen's d effect size value from the raw data, the sample size they report for males and imagining that the mean is zero and SD is 1. The Z score values they report comes after accounting for age and an activity mediator so it would be a bit different.

My main guess for why it would be different is because I think to get their Z scores they have divided their estimates not by the standard deviation (which would indicate the variation in the data), but by the standard error of the mean (which represents the variation around your estimate of the true difference between hc and me).

Sorry it's a bit confusing to write out, but basically I think it doesn't so much represent the magnitude of the difference in the data, but instead represents the confidence that there is a real difference between the groups. When I divide the estimated difference by the SEM in the plot I made, I get a value around 5 - but of course we know the effect size cohen's d is only 0.8. I hope that makes sense.

I should say that while i've got some experience with generalised linear models I haven't any experience with the model they're using in this paper - so I'm just making an educated guess.

 

@chillier 's comparison the Hoel paper is with serum, the rest are plasma, this could have an effect.

Uniprot entries describing the concerned enzymes:
https://www.uniprot.org/uniprotkb/P06276/entry
https://www.uniprot.org/uniprotkb/P04180/entry

IMO it is kind of hard to draw conclusions from snippets of lipid related data alone especially from biofluids, where the lipids in circulation may be impacted by processes within cells / other tissues, rates of secretion or uptake, all in combination with reactions in circulation itself, such as those catalysed by Phosphatidylcholine-sterol acyltransferase (LCAT) (link two above). And then within that, we don't really know the exact roles of all of these lipids or all of the specific implications of them rising and falling (and implications for which tissues?)

An approach to look for a meaningful thread to tug on related to lipid data might be to try and associate it with related enzyme levels or activity. Then we can account for tissue/circulation specificity as well depending on the attributes of the enzyme (order of the following points doesn't matter):

eg:
a) Identify whether particular groups of lipids are affected in their levels, or in their composition (length, saturation, ether bond, lyso species, etc)
b) In parallel, check whether this is accompanies directionally-concordant dysregulation of enzymes known to drive whatever specific change you are seeing (or vice versa)

Random illustrative example: if one sees an accumulation of PS lipids and a reduction of PE lipids, and in parallel a downregulation of an enzyme that catalyses a key PS -> PE reaction, bob's your uncle and you have a clear hypothesis to test.

So from biofluid data I think a useful exercise would be to take an enzyme like LCAT that is active in reactions in circulation, that there are differences in here, and then look at the levels of the molecules that it's working on (input & output).

What is it you say, "analysis by-meme" (joking not critiquing)

Anyway now for some example interpretation, at the very least as a curiosity

LCAT catalyses PAF -> Lyso-PAF. The lyso form of PAF is inactivated. PAF ("Platelet-activating factor") is a very influential molecule. @chillier I can't find it in the paper, were LCAT levels elevated or reduced? The paper seems to report elevated platelet count. This could be caused by elevated levels of active PAF (PAF mobilises platelets). This could be driven by reduced PAF inactivation by LCAT. If LCAT levels are reduced there could be a testable hypothesis in here. The whole chain of events isn't completely evidenced in the dataset from what I can see (although I have only skimmed due to time constraints) but it's something.

Putting together a table of altered metabolites, putting together another table of altered gene products annotated with the metabolites involved in reactions that they catalyse, and then cross checking everything could be really useful. Having metabolite and protein data together is a goldmine. Hell you could even look for correlations between altered enzymes and metabolites and then check the significant outcomes for biological relevance.

I am doing something related in a paper draft atm but using the LCL collection so more removed from the body than here, but more linked to one kind of cell's metabolism. Please please please someone do this in biofluids, the datasets seem to be in place?

Is anyone aware of any publicly available combined metabolite/gene expression datasets for ME/CFS? I'm tempted to have a play around if so

 

Maybe the Deep Phenotyping study data if you have/can make a mapmecfs account. The names of all the datasets are viewable without an account: https://www.mapmecfs.org/dataset/?organization=nih-intramural

 

Didn't know this was available. Thanks!

 

Thanks that's interesting. They don't report on triglycerides or phosphatidylcholine in Germain et al 2020 but they do mention they are detectable in their lipid panel so it follows they would probably be insignificant.

I wanted to check anyway so went into the supplementary and calculated the p values for the phosphatidycholines and triglycerides (I log2 transformed the data like they do in the paper I believe). Here's a volcano plot for the phosphatidylcholines (each dot is a unique phosphatidylcholine):

Sure enough nothing significant, and also no particular fold change bias either way.

Triglycerides also aren't mentioned and also aren't significant (besides one but it won't survive multiple test correction).

However, across ~500 unique triglycerides there is a global bias towards increased levels. Only a small handful are lower and they contain big outliers. in a linear model using all triglycerides as a random effect to test whether there are global differences in triglyceride levels, it comes out as highly significant <2e-16. I tried summing them together too to compare total triglyceride levels, and while this trended up it was not significant.

 

For sure, there's evidence that in general serum contains higher levels of many amino acids and differing levels of pyruvate:lactate than plasma - possibly due to platelet activation in serum. I don't know about lipids.

This would be great, and of course it's difficult to get a mechanistic understanding from just metabolomics. I'd love it if something like this was possible. I wonder though whether a lot of these things could be fallout from another upstream problem (eg Liver issues). The effect sizes are small and the differences in lipid levels seem like they might be quite broad spectrum within their classes (PC, TAG, CE etc).

There are other proteins in the data like Apolipoproteins which sit in lipoproteins and therefore may give clues as to what kind of density lipoproteins (carrying what kinds of lipids) are involved - but they're not enzymes so probably not great for anything mechanistic. In this study Apolipoprotein A(1) is low according to both blood biochemistry and NMR data, and Apolipoprotein D is low in men only in the proteomics data. These are both associated with HDL, and accordingly HDL cholesterol as well as HDL concentration in general is down in the data.

Interestingly ApoD is also associated with LCAT and they are co-expressed together supposedly.

Interesting! LCAT levels are increased, but it's one of the proteins that is only significant after multiple testing correction in the combined analysis. It is significant in both male and female cohorts analyzed separately but it does not survive multiple test correction in either. The Cohen's d for men in the raw data is 0.55 in male and 0.33 in female cohorts.

 

Probably not needed for most people here, but here's a lay summary I wrote for te MEA website.
Research: Myalgic Encephalomyelitis is clear to see in the blood

https://twitter.com/user/status/1831301007764246889

 

I need it! Thank you

 

I am interested in the SOd3 finding, but can’t quite follow the information in the article. Was it raised or lowered? And what might this mean?

“Repeating this NDE analysis using the UKB 874 mediator on levels of 2,923 proteins, measured using antibody-based assays, yielded only a single protein, extracellular superoxide dismutase or SOD3, whose abundance was significantly altered (FDR < 0.05) between cases and controls in both females and males.”

 

If I remember rightly it appears in a scatterplot figure at bottom left suggesting low in both?

 

Also, I wonder whether there were any common findings between this study and this (much smaller) one on potential biomarkers in POTS
https://www.ncbi.nlm.nih.gov/pmc/articles/pmid/36414707/

I’m afraid I’m too tired to check. On heavy medications for pinched nerve.

Thanks everyone and the researchers for excellent discussion.

 

New Scientist piece on this:
https://www.newscientist.com/articl...-could-be-quickly-diagnosed-via-a-blood-test/

 

That's pretty unhelpful coverage I think.
Ponting's study does not claim to provide any sort of diagnostic test. Moreover, diagnostic tests are not what we need. The comment shows a lack of understanding of the basic process of clinical decision-making.

The point of such studies is to give us at least some clues as to what is actually going on in at least some people with ME/CFS.


The discussion arising from this has been moved to Biomarkers for ME/CFS - Discussion thread on the next steps for testing biomarkers and why we need them

 

To come back to the paper itself, I think what it shows that Is valuable is a kind of biological footprint of the illness – even if you can only see it in a large sample.

Although the pre-print says it shows it’s not a psychological illness, I don’t think that will convince doubters without showing data that shows, for example, anxiety and depression have a different biomarker profile (depression at least will have some biomarkers).

But evidence that shows what causes at least a subset of ME is, sadly, I think the only thing that will start to shift minds away from Psychosocial explanations that prevail.

Maybe a diagnostic test would also come along at that point, but the bigger gain is showing biological causation. Unfortunately we’re not there yet.

Perhaps Decode ME will find something (genetic studies are good at finding causal signals).

ADDED: I guess I’m just agreeing with @Jonathan Edwards.

 

I agree that if specific findings can be replicated in a population with tighter diagnostic ascertainment they may be a great importance. And if they can be linked to DecodeME findings in terms of genes all the better.

My worry about the sample size is that the bigger your sample the more likely you are to get significant differences due to systematic confounders related to the way cohorts have been gathered. As has been pointed out by others, only a tiny proportion of 'ME' subjects had a CRP above 1, so we can be reasonably sure that whatever is driving ME/CFS symptoms it doesn't normally involve any chronic inflammation.

 

Is it actually sensible to move thee posts since the discussion is central to analysis of this paper? I appreciate that it may seem tidier to have topics all in one place but if we are going to have dialogue with scientists on specific projects then it becomes a bit pointless if discussion of key points gets removed. The more general parts of the discussion have been covered several times before so the thread for biomarkers is likely just to become a repetition.

 

Except that this discussion is central to the significance of the paper. Individuals may disagree about the implications, but that is what make a discussion valuable. My concern is that by not appreciating the complexity of the 'diagnostic' concept the authors may have significantly misinterpreted their data. That would be a huge pity because they data are very valuable. If much of the paper focuses on mathematical analysis that is misconceived and the genuinely important facts are not brought out key information may be buried in confusion.

But at the cost of losing important discussion in relation to this particular study, because most people will not realise that moved threads may be crucial to the discussion. No topic exists in isolation., It is always relevant in different ways to many different problems. Dumping the same arguments repeatedly in one thread, dislocated from the situations they relate to seems to me unhelpful.