A SWATH-MS analysis of ME/CFS peripheral blood mononuclear cell proteomes reveals mitochondrial dysfunction: Sweetman,Vallings,Tate et al Aug 2020

So this version is still undergoing review in the Journal of Translational Review. Does anyone know if/how it's possible to make community comments?


(Potentially, there looks to be quite a good, transparent review process. I don't know if that is standard practice for journals or not.)

 

@Snow Leopard's good comments notwithstanding, I think this is an interesting study done by the Dunedin team who operate with very little funding, most of it coming from donations from people with ME/CFS and their families.

It will be good when authors of biology papers about ME/CFS don't feel the need to include advocacy in the text.

I think we are probably at the point where continuing to talk about controversy about whether ME/CFS is seen as a legitimate condition in a proteomic analysis paper can cause more harm than help.

They describe ME/CFS with the word that makes me groan every time I see it - "complex". This leaves the door open for people to assume that all sorts of factors might be involved in causing it. We've discussed the word elsewhere. Yes, ME/CFS is poorly understood, but that's not the same as a complex disease.

Anyway, to the biology. It's difficult to assess how real the differences that were found are without seeing some scatter plots and the sample is small. There's a Principle Component Analysis. PCA is new to me.



Principle Component 1 ( along the x axis) is reported as explaining 38% of the whole sample variance, but the controls and ME/CFS people aren't really separated out along the x axis. Principle Component 2 is reported as explaining 16% of the variance - and 5 or 7 of the ME/CFS people do look different to the controls on the y axis. I'm a bit doubtful about their conclusion that the red (ME/CFS) dots above the dotted grey line really represent a well-defined separate group.

Bearing in mind Snow Leopard's comments about the number of proteins that were assessed (and so the risk of random findings) I like the lists of proteins that seem to be differently expressed between the controls and ME/CFS people (Table 2 and 3). It's encouraging that these seem to support the findings of Fisher/Missailidis about problems with the electron transport system/ATP production in mitochondria.

I'm not sure what you mean WTM. Fisher's team did publish this:
An Isolated Complex V Inefficiency and Dysregulated Mitochondrial Function in Immortalized Lymphocytes from ME/CFS Patients Missailidis et al. 2019


The Dunedin team aren't over-claiming. I liked their conclusion:

Hopefully there will indeed be more validation done with larger samples.

 

Does the NIH ME/CFS study include muscle biopsies?

I reckon there are probably lots of biopsies that could be gathered up from gastroscopies and colonoscopies done in people with ME/CFS.

 

Welcome to the forum @DMissa. It's great to have you here; thanks for taking the time to comment.

 

The "goal" of PCA is not to get separation on the x or y axis, you just want separation. And they do get separation here, even if the patients are not in a nice cluster (Although to me P1 and P7 could have been left with the controls). However, if they only have very few samples/participants and a lot of proteins PCA can find separation by chance, the results are more valid when there are many samples and fewer variables (proteins in this case).

For the rest, agree with @Snow Leopard . Hopefully there will be more omics studies, and it sounds like that is the case

Edit: I wonder if something like this could be used to understand PEM:

https://www.youtube.com/watch?v=R1WUuSZimpo

(they have measured fatty acid concentrations in the blood of patients with a metabolic disease, and by using computer modelling they found that an enzyme that metabolized these fatty acids was substrate inhibited causing problems with energy. This disease is genetic in origin if I remember correctly, I haven't watched the video in ages, but there are adults with the same mutation that grow up without showing symptoms. So there must be some pathway that can compensate for the faulty gene).

 

But PC1 doesn't look to be adding anything to the separation of the ME/CFS sample from the controls. Arguably, only the 5 out of 11 of the ME/CFS dots are a separate cluster - and they are separated out on the basis of PC2. And PC2 only accounts for 16% of the sample variance. That's not a very strong finding.

For example, P1, P11, P10, P7, P4 and P9 are a lot closer to the controls on these measures than to other ME/CFS people.

 

PCA is an unsupervised method so does not try to separate between samples and controls, that's why getting separation at all is considered good/useful. The variance is frustrating to me as I've gotten as many different explanations for how much I should care about it as I've had teachers. I've been told low variance PC's sometimes can give better explanation of anomalies, but I'm not up to explaining the argument atm. For data exploration looking at more components until you find the best separation might provide a starting point for further analysis but it could also mean you're just looking at noise (or fishing for a better result than what you actually have, but then I guess it would be easier to use a supervised method like OPLS that actually force separation between groups no matter how small it is to begin with).

Agreed that P4 and P9 are close to a group of controls, but there really are very few participants, and probably many other things that could be used to categorize the participants into different groups besides disease state. Sex, age, physical activity level, BMI, diet..

Going with their dotted line:
We don't have the loadings (again something I've either been told not to care about or to find interesting), but loadings show which variables/proteins pull the samples in any which direction. So here I'd be interested in the proteins that would show up in the upper left and lower right corners because those are the ones that seem important for separating the pwME and controls. If some proteins makes sense biologically I might not want to throw it away as "noise" just yet. Even if the variance is low for the PC overall. They might explain what I'm interested in.

With so many proteins, if only a few are different between disease state and healthy state (and I assume only a few of all proteins are actually involved in the disease) they wouldn't explain a lot of variance because they're drowning in all the other things that can explain variance better, by affecting more proteins to a larger degree, like some of the examples I already mentioned might.

Just for making an example: Maybe the y axis can be explained by physical activity level. Some patients are less disabled than others and would be closer to the controls.

Edit: Wikipedia had a note about low variance PC's being able to show possibly important variables, citing this: https://www.jstor.org/stable/2348005?origin=crossref&seq=2#metadata_info_tab_contents which may be where I got it from.

Edit2: Clarified a bit