Deep phenotyping of post-infectious myalgic encephalomyelitis/chronic fatigue syndrome, 2024, Walitt et al

Yep, if they were looking for adulation, they could have had it in spades. Will be interesting to see if, what, when and how subsequent papers will be published.

 

I downloaded the raw data from mapmecfs to dig around in. First little thing I did was get the correlation of every molecule from the metabolomics dataset with the group. Used point biserial correlation and Benjamini–Hochberg correction.

Here are the top 40 out of 445 total molecules:



And the top five molecules:

 

I thought that was weird too, but the first 102 molecules are all lower in ME/CFS. Only 27/445 molecules are higher. Which seems fishy so I may have done something wrong, but I'll manually check some of these, probably tomorrow.

 

Yes, I think these are correct. I just averaged the metabolites for each group directly in the raw data file and the mean levels in ME/CFS are lower than HC in the top five negative correlations and higher in the top five positive correlations. (ME/CFS is 1, HC is 0.)

Highest negative correlations:

Highest positive correlations

Maybe we're just generally metabolite deficient and that will be the answer.

 

Also appears to match with the paper. I don't know if this heatmap is ordered by most correlated at the top, but the top four are the same as the top four in my list, although in a slightly different order. And it seems to also show that most of the metabolites they chose to illustrate are higher in HV.

 

This is the CSF data if I recall right? I believe from when I looked at this earlier in the year that the ones in that heatmap are the top results from a simple mann whitney u test between groups. They also analysed it using a more complicated (but probably more appropriate) glm model.

If it's not a pain, it would be cool to see strip plots like you've done for those metabolites for pyrroline-5-carboxylate and maybe also hypoxanthine. They've both come up in a few places before and I think I remember seeing them here too.

Also 1-steroyl-GPC (18:0) - from third to the bottom of that heatmap - is a phosphatidylcholine. Interesting that one of their top results here is again a phosphatidylcholine and is also low ( I don't think they do a broader lipidomics panel do they EDIT: They do).

 

Yes, this is CSF.

Sure here are those plots. Also box plots, with the white dot being mean and the yellow line being median.

Molecule | Correlation (point-biserial) | uncorrected p value | adjusted p (BH)
S-1-pyrroline-5-carboxylate | -0.3542989681 | 0.02907775564 | 0.08665556102


hypoxanthine | -0.2336116431 | 0.1580701679 | 0.247194837


1-stearoyl-GPC (18:0) | -0.4503725604 | 0.004548911271 | 0.0427324321

Yes, I'm hoping to look at that as well. You think Mann-Whitney would be more appropriate?

 

That's great thank you Mann whitney is a safe bet since it doesn't make any assumptions about distribution and simply asks the question of whether one group is higher or lower than another, so their p values are probably fine. Biserial correlation I think is ok too but assumes that the data is normally distributed, and sometimes metabolomics data needs a log transform first to better meet that assumption. A (generalized) linear model can also be nice because it can handle confounders like sex and age.

 

Ok, much appreciated, thanks!

 

Ok, I did Mann-Whitney on both the metabolomics and the lipidomics datasets, and the BH adjusted p-values are based on both datasets combined.

Metabolomics order changed a bit:


And here's lipidomics top 40:


Top 5:

 

From study:

 

This looks really interesting - no individual TAG passes adjusted p<0.05, but (like with Germain 2020) they all seem to be biased higher in ME. I'd quite like to repeat the analysis I did on that data (ie volcano plot and testing global TAG differences):

@forestglip how would you feel about sharing the lipidomics data with me? Is it in a convenient csv type format?

 

It is, but it's against the terms of mapmecfs to share raw data:

I did email to ask if sharing charts on this forum made from the data was allowed and they said yes to that. Can you make an account?

 

That's fair enough! I'll make an account in that case

 

Took about four days for them to approve my account, FYI.

 

What could explain why almost all metabolites tested were lower in ME/CFS? Most of these aren't statistically significant on their own, but as @chillier was saying about triglycerides in the lipidomics study, there seems to be a general trend here too.

The comparisons are based on medians. The lists are in alphabetical order, this just says if the medians of one group of metabolite are at all higher than the other, and says nothing about significance.

Metabolites that were higher (n=36)

Metabolites that were lower (n=409)

 

I was curious about energy expenditure after looking at my own food tracking over the past six months, which shows I ate an average of 2,950 calories per day, yet have gained at most, maybe five pounds, if any. There's a chance the recorded intake is a bit higher than reality, as I was eating beef from the farmer's market often, and I probably guessed too high for the fat content. But at most it'd be about 150 calories too high per day.

So I'm wondering if my body is burning tons of energy for some reason, even though I'm mostly sedentary. Average 2,800 steps per day over the past month. I think a typical energy expenditure for someone my size and age and activity is around 2,250, according to a calculator on Mayo Clinic's website. 2850 for "very active" which is quite far from my activity level.

Looking at the data from the study, it doesn't look like that's a universal thing, if my own recorded calorie count is even accurate for me. The average intake seems to be similar for both groups. There are some outliers both high and low, but their body weights are also pretty high and low, respectively, compared to most of the other participants.

Definition of the metric: "total energy expenditure in kcals/day from the full whole room indirect calorimeter measurement on the Baseline chamber night". So that was before the CPET.

Looks like it's maybe trending up in the three days after the CPET:


Largest difference seems to be during sleep on the third day after CPET:


The study says no differences found, although a p-value of 0.07 for that last sleep measurement.

S12:


Fig S9A-D:

I might try to see what the data looks like if the energy expenditure is normalized to body weight.