In-Depth Analysis of the Plasma Proteome in ME/CFS Exposes Disrupted Ephrin-Eph and Immune System Signaling, 2021, Hanson et al

[ME/CFS Skeptic]

We do have a good candidate for a reliable biomarker (in over 10 studies): workrate at ventilatory threshold on the 2 Day CPET. The only problem is that it isn't practical for many patients to undergo.

I still have my doubts about this because most of these studies were really small. Not sure if these results will hold up in a proper study with a pre-specified protocol, large sample size, adequate control groups etc.

 

[Hoopoe]

There are lots of other clues, though. I guess when doctors were figuring out what causes Type 1 diabetes, they had to rely on these to diagnose it until someone realised that blood sugar might be important and worked out how to test it. The lack of a test didn't mean that it was impossible to diagnose it, and it's possible that other indications of health vs disease used at the time looked pretty normal.

One ancient method of diagnosing diabetes was to test whether ants were drawn to the urine of the patient.

In ME/CFS we're still not much further ahead than this it seems. Make patient exercise, if they get worse it's ME/CFS.

 

[Midnattsol]

Fukuda would be fine if they made sure it was Fukuda with PEM. Which I'm hoping Dr Levine would have done. But unfortunately not made clear in the paper.

Can anyone explain what the significance of protein ratios is and how they relate to "biological relevance"?

Protein ratios may have a biological relevance, they may not. If the proteins are connected in some way, like they are both important for a specific function, a change in ratio between them could imply that something is going on in that pathway. But we don't really know what the normal ratio between different proteins are, or how it would impact anything. They write that not all the ratios makes biological sense, this migh only be true at the present moment, there might be some biological thing linking them that we just don't know yet. Very frustrating

I really like their data sharing thing, although they lack some sample metadata in the data they've shared for metabolomics.

 

[FMMM1]

Yes but in ME/CFS there aren't even non-specific biomarkers.

A person can be incredibly sick yet his main test come back pretty much normal. I'm not sure but I suspect in most other illnesses it will be easier to find abnormalities, even with low specificity, if a person has gotten so ill.

I'd never even heard of this technique
"To gain insights into the molecular disruptions in ME/CFS, we utilized an aptamer-based technology that quantified 4790 unique human proteins, allowing us to obtain the largest proteomics dataset yet available for this disease, detecting highly abundant proteins as well as rare proteins over a nine-log dynamic range."

Firstly they've analysed more proteins - so more likely to find something. Not sure how low the concentrations for "Ephrin-Eph" were but it may have been lower than other (previously used) conventional methods. Also, I recall a caveat that the other protein studies relied on sample preparation which "lost" some of the proteins. So there may be important differences in this study.

In terms of something being readily observable in blood, Ron Davis pointed out that at least one scientist had a theory that ME was diabetes. In this paper you'll see that "glucose" is a key word here's the first reference to "glucose": "Ephrin-A5 is in the same family as Ephrin-A4 (Table 2) and has been shown to be involved in various biological processes including --- glucose-stimulated insulin secretion through pancreatic islet cell communications.". Check out the other references to glucose - they're all pretty interesting. So in "hindsight" there could potentially be clues in readily observable things in blood in ME.

As the authors suggest we'd need a much bigger study looking in more detail "Ephrin-Eph" e.g. lot more people with ME and people with diseases which have similar presentation - this study used healthy controls*.

People with ME are often dismissed, i.e. labelled as having a psychological rather than a biomedical disease, and thus not warranting biomedical research. This study indicates that the (biomedical) tools are getting better - so biomedical research is justified.

I'm not aware of a good biomedical test which detect Alzheimer's disease at an early (more treatable) stage - perhaps an example of a "worthy" disease which currently doesn't have a good biomarker/diagnostic test.

I'm assuming a study like this would be a prime candidate for funding under Horizon Europe --- perhaps it could include Lyme and Long covid as well as ME & healthy controls.

Anyone in Europe using this technique i.e. aptamer-based technology?

*"We identified nine proteins with individual classifiers greater than 0.85 between ME/CFS subjects and controls as well as nine protein ratios with classifiers above 0.92. As is practice, a diagnostic test for ME/CFS would not be used on subjects who are not complaining of fatigue or malaise; these protein differences must be tested against other fatiguing illness that might be confused with ME/CFS—such as depression, cancer, or chronic Lyme disease, to name a few. Such studies will be needed to determine how the sensitivity and specificity are affected by excluding individuals who do not exhibit fatigue or malaise."
In-Depth Analysis of the Plasma Proteome in ME/CFS Exposes Disrupted Ephrin-Eph and Immune System Signaling

 

[FMMM1]

It's still possible that, when we do unpick the mechanism, it will come to look quite quickly as if it should have been much more obvious from the clues.

I need to look at the data but I think I formed the impression that the group of patients selected demonstrated an issue re Disrupted Ephrin-Eph and Immune System Signaling
If that's correct then perhaps many people with ME have a similar underlying problem --- too early to say though.

 

[Snow Leopard]

I'd never even heard of this technique
"To gain insights into the molecular disruptions in ME/CFS, we utilized an aptamer-based technology that quantified 4790 unique human proteins, allowing us to obtain the largest proteomics dataset yet available for this disease, detecting highly abundant proteins as well as rare proteins over a nine-log dynamic range."

Aptamers are really cool, I had to study them for my undergraduate degree over 10 years ago.

The key is to have an off-the-shelf kit with validated sensitivity/specificity. (which is not perfect BTW, this is primarily a discovery tool)
The set of aptamers has only been commercially available (with over 4000 targets) since 2017.
https://en.wikipedia.org/wiki/SomaLogic

 

[Adrian]

I haven't read the paper yet but it looks like this group are well aware that the small sample in this study means that this is just a start. I hope they get more funding for this.

With the potential for possibly two or more illnesses being included under the ME/CFS label, it would be interesting to look at the scatter plots of some of the proteins.

I would quite like to see a cluster analysis putting all the interesting proteins for each patient in a vector. It would be interesting if there are clear clusters between the two groups but also if the ME group forms different clusters,

 

[wigglethemouse]

According to the paper this was funded by Cornell and NIH. The NIH grant was a 5 year funded grant 2013-2017 totalling $2,956,287

Cornell is also allowing on private donors to fund their work. This study was funded partly by Cornell which I assume is private donors.

@Squashedhedgi
We are grateful to a number of private donors who are supporting our work on the molecular basis of ME/CFS--our Center maintains a donation page here: https://neuroimmune.cornell.edu/donate.

 

[FMMM1]

Aptamers are really cool, I had to study them for my undergraduate degree over 10 years ago.

The key is to have an off-the-shelf kit with validated sensitivity/specificity. (which is not perfect BTW, this is primarily a discovery tool)
The set of aptamers has only been commercially available (with over 4000 targets) since 2017.
https://en.wikipedia.org/wiki/SomaLogic

Thank you Snow Leopard.

Vaguely recall a guy in a corner of the lab (30+ years ago!) working on biotin binding ---- not sure that's at all relevant!

I expect that once you've identified compounds of interest then you can determine them using other methods. E.g. if you Google "Ephrin-Eph + mass spectrometry" then you'll see a bunch of stuff (detection methods). That may indicate off the shelf [mass spectrometry] methods to give more accurate measurements of e.g. Ephrin-Eph(?).

Noticed Cornell have applied for a patent - maybe someone could find it.
[6. Patents
A provisional patent application (U.S. Serial No. 63132722) concerning the data has been filed by Cornell University.]

 

[FMMM1]

I would quite like to see a cluster analysis putting all the interesting proteins for each patient in a vector. It would be interesting if there are clear clusters between the two groups but also if the ME group forms different clusters,

I recall this paper which compared Lyme and ME [https://pubmed.ncbi.nlm.nih.gov/21383843/] It's a protein study of Lyme and CFS - data is presented in the way you suggest.

So yea, an update which used this technique Aptamers would seem to be in order --- would it need to be expanded to include Long covid ---? Mind you if you included Lyme and Long covid (as well as ME) then it might be easier to get funding!

 

[Adrian]

I recall this paper which compared Lyme and ME [https://pubmed.ncbi.nlm.nih.gov/21383843/] It's a protein study of Lyme and CFS - data is presented in the way you suggest.

So yea, an update which used this technique Aptamers would seem to be in order --- would it need to be expanded to include Long covid ---? Mind you if you included Lyme and Long covid (as well as ME) then it might be easier to get funding!

Yes the point is it brings a multivarient view onto the data an allows for combined effects. I've always like the hierarchical cluster analysis technique they use - but these days I tend to use TSNE which does a projection from a high dimentional space to a lower one whilst trying to preserve distance. It all depends if we see each protein acting (or change caused) independently or as a set.

 

[Snow Leopard]

Noticed Cornell have applied for a patent - maybe someone could find it.
[6. Patents
A provisional patent application (U.S. Serial No. 63132722) concerning the data has been filed by Cornell University.]

Patents are horrible to read so I'll pass on that!

I am kind of glad they have patented this as a patent by the wrong people could hold up progress...

As for biotin, it's commonly used in bionanotechnology with Avidin to stick things together.

 

[Adrian]

Noticed Cornell have applied for a patent - maybe someone could find it.
[6. Patents
A provisional patent application (U.S. Serial No. 63132722) concerning the data has been filed by Cornell University.]

If they have only just filed then it won't be published for a while (I can't remember the current rules but it used to be a year in the UK system and I think the US was similar). Initially just the title is published. Granting can take multiple years.

 

[FMMM1]

Yes the point is it brings a multivarient view onto the data an allows for combined effects. I've always like the hierarchical cluster analysis technique they use - but these days I tend to use TSNE which does a projection from a high dimentional space to a lower one whilst trying to preserve distance. It all depends if we see each protein acting (or change caused) independently or as a set.

Thanks. I'm laughing here - you obviously understand this much better than I do - which is great.

 

[FMMM1]

Yes but in ME/CFS there aren't even non-specific biomarkers.

A person can be incredibly sick yet his main test come back pretty much normal. I'm not sure but I suspect in most other illnesses it will be easier to find abnormalities, even with low specificity, if a person has gotten so ill.

It's occurred to me that this technique [aptamer-based proteomics] is similar to genome-wide association study (GWAS). I.e. you don't know what the cause [of ME] is and you measure everything - in the hope you find something different between people with ME & healthy controls. OK the authors highlight that you then need to look at similar diseases (Lyme etc.).

The MRC panel @Jonathan Edwards sat on identified GWAS as a worthwhile study area* and that led to Chris Ponting's GWAS study being funded. Not sure if the MRC are still using that approach i.e. convene a panel to consider worthwhile research areas - like aptamer-based proteomics!


*https://www.s4me.info/threads/georg...dian-article-21-1-21.18562/page-4#post-319043

 

[Midnattsol]

Yes the point is it brings a multivarient view onto the data an allows for combined effects. I've always like the hierarchical cluster analysis technique they use - but these days I tend to use TSNE which does a projection from a high dimentional space to a lower one whilst trying to preserve distance. It all depends if we see each protein acting (or change caused) independently or as a set.

Why TSNE and not PCA?

 

[Adrian]

Why TSNE and not PCA?

It depends on what PCA gives. TSNE allows nice 2d plots which are easy to interpret where as with PCA it depends on how much of the varience is captured with the first 2 principle components. I have seen people doing PCA as a dimension reduction and then TSNE (as it is quite costly).

The difference as I see it is that PCA rotates the axis such that they are ordered with the first axis capturing the most varient and the 2nd the next most (given a linear transformation) etc. Where as TSNE tries to keep similarities between data and hence may deal better with non-linear relationships.

But as with any clustering there is a certain amout of choice and see what separates data and then ask why.

With some data I've also done things like try to discard non-interesting features (say that aren't IID) such that they don't affect the clustering - this really helps with things like anomaly detection.

 

[Midnattsol]

It depends on what PCA gives. TSNE allows nice 2d plots which are easy to interpret where as with PCA it depends on how much of the varience is captured with the first 2 principle components. I have seen people doing PCA as a dimension reduction and then TSNE (as it is quite costly).

The difference as I see it is that PCA rotates the axis such that they are ordered with the first axis capturing the most varient and the 2nd the next most (given a linear transformation) etc. Where as TSNE tries to keep similarities between data and hence may deal better with non-linear relationships.

But as with any clustering there is a certain amout of choice and see what separates data and then ask why.

With some data I've also done things like try to discard non-interesting features (say that aren't IID) such that they don't affect the clustering - this really helps with things like anomaly detection.

I find PCA score plots easier to interpret than TSNE plots, but yes non-linear relationships might work better with TSNE (I've never used TSNE, just read about it and decided on PCA as it was less costly and worked well with my data). Is it possible to do a supervised version of TSNE?

 

[Adrian]

I find PCA score plots easier to interpret than TSNE plots, but yes non-linear relationships might work better with TSNE (I've never used TSNE, just read about it and decided on PCA as it was less costly and worked well with my data). Is it possible to do a supervised version of TSNE?

I've found the oposite but I suspect it is very data dependant. I'm not aware of a supervised version of TSNE but if you have labels you can build a classifier - if you want to then have an understanding of what features are important then using tree based classifiers can help as they are interpretable (something like random forests or XGBoost can give good performance) - or there are explanation techniques for neural nets (but that is an area I've not read enough about - its on my list of things to do but has been for the last few years!).

 

[Hoopoe]

If proteins related to the cytoskeleton, cell-cell communication, extracellular matrix, etc. keep showing maybe it's time to physically look at cells of patients and the spaces inbetween them. My understanding is that different kinds of ephrins in different tissues which would be a clue as to there the problem could be. Is this possible to do?