Causal relationship between immune cells and post-viral fatigue syndrome: a Mendelian randomization study, 2025, Wang et al

Wyva

Wyva

Senior Member (Voting Rights)
Zheyi Wang, Zetai Bai & Yize Sun

Abstract

Background
Accumulating evidence has hinted at a correlation between immune cells and post-viral fatigue syndrome (PVFS). However, it is still ambiguous whether these associations indicate a causal connection.

Objective
To elucidate the potential causal link between immune cells and PVFS, we performed a two-sample Mendelian randomization (MR) study.

Methods
We obtained summary data on PVFS cases (Ncase = 195) and controls (Ncontrol = 382,198) from the FinnGen consortium. Additionally, we retrieved comprehensive statistical information on 731 immune cell features. Our analysis encompassed both forward and reverse MR approaches. To ensure the reliability and validity of our findings, we conducted rigorous sensitivity analyses, addressing issues of robustness and heterogeneity.

Result
Our study presents compelling evidence of a probable causal link between immune cells and PVFS. Notably, we have pinpointed 28 distinct types of immune cell traits that potentially exhibit a causal association with PVFS. Among a pool of 7 31 immune cell traits, we identified 28 immune cell types that exhibited a potential causal association with PVFS. These included 9 B cells, 1 conventional dendritic cell (cDC), 1 maturation stage of T cell, 3 myeloid cells, 9 T, B, NK, and monocyte cells (TBNK), and 5 regulatory T cells (Treg).

Conclusion
Through genetic analyses, our study has unveiled profound causal connections between specific types of immune cells and PVFS, offering valuable guidance for forthcoming clinical investigations.

Open access: https://virologyj.biomedcentral.com/articles/10.1186/s12985-025-02809-4
 
Quite a lot of associations. Maybe someone can weave a story out of it.
12985_2025_2809_Fig3_HTML.png
Notably, the upregulation of CD24 on CD24+ CD27+ cells stands out with a significant OR of 1.795. This finding suggests that the expression of CD24 on this subset of B cells may play a pivotal role in the pathogenesis of PVFS.

The study has found that the expression of the glycoprotein CD24 is elevated in memory B cell subsets in the peripheral blood of patients with CFS, compared to age-matched healthy controls, both in terms of frequency and expression levels [21, 22].
Click to expand...

21. Armstrong CW, Mensah FFK, Leandro MJ, Reddy V, Gooley PR, Berkovitz S, Cambridge G. In vitro B cell experiments explore the role of CD24, CD38, and energy metabolism in ME/CFS. Front Immunol. 2023;14:1178882.
Article CAS PubMed Google Scholar S4ME

22. Mensah FFK, Armstrong CW, Reddy V, Bansal AS, Berkovitz S, Leandro MJ, Cambridge G. CD24 expression and B cell maturation shows a novel link with energy metabolism: potential implications for patients with myalgic encephalomyelitis/chronic fatigue syndrome. Front Immunol. 2018;9:2421.
Article PubMed PubMed Central Google Scholar S4ME
 
In this study, the p value generated by the IVW method serves as the primary indicator for evaluating the causal association between exposure and outcome. Other methods complement the evaluation of MR findings, and consistency in the direction of effect size across diverse methods suggests robust conclusions. To mitigate the concern of multiple testing of p-values stemming from the IVW method, we implemented the False Discovery Rate (FDR) correction. A p-value adjusted for FDR that falls below 0.05 signifies a definitive causal linkage between exposure and outcome. Conversely, a value surpassing 0.05 suggests a potential causal association.
Click to expand...
I'm looking at Supplementary Table 1 where there is a column called pval which matches the p values reported in the paper. They used almost all of the immune cell exposures that were under p<.05 (28/31). There's a p_fdr column where one row is .83 and the rest are over .98. So I'm not really sure if they did any kind of multiple test adjustment.

Furthermore, the evaluation of pleiotropy incorporates the MR-PRESSO test and the intercept of the MR-Egger regression. If both the p-value for the MR-PRESSO global test and the MR-Egger intercept exceed 0.05, this indicates the absence of pleiotropy [20].
Click to expand...
If I understand this right, p values less than .05 in either MR-Egger or MR-PRESSO indicate pleiotropy, and thus the exposure should be excluded. But looking at Supplementary Table 1, 7 of the 28 exposures they described in the paper have MR-Egger p values less than .05 (they don't seem to show data for MR-PRESSO).

Edit: Nevermind on the second point. Apparently MR-Egger can be used to detect both pleiotropy and causal effect, and they used it for both. I think the p values they report are for causal effect.
 
Last edited:
poetinsf said:
Could this be groundbreaking?
Click to expand...

I could be but I am really annoyed that the jargon of the presentation is impenetrable from my perspective. People who'd o genetics claim that they need to use this sort of jargon but the main justification seems to be to please other geneticists refereeing the paper. papers should be written for a wide academic audience, not of an in crowd.

At present I have no idea what they found. They don't mention any specific genes or how they come to infer a causal link. They use terms like 'trait' which mean something technical to them but God knows what.

Hopefully, some members can unpick this for us. Jo Cambridge has been going on about CD24 for ages now. It might be really important but I have not so far been able to see how one would built it in to a general model.

The patient cohort is again small, which is a bit worrying.
 
But come to think of it, if the difference in CD24 expression on cells in people with ME/CFS is due to a genetically programmed regulatory shift then it could slot in to an adaptive immune model. If you are a CD24 overexpresser under certain conditions then maybe you are destined to trip into an acquired loop.
 
Jonathan Edwards said:
What is a 'phenotype' in this context?
Click to expand...

Paper said:
The GWAS data on immune cells were derived from a comprehensive study of 3,757 European-ancestry individuals, which analyzed genetic variants associated with 731 immune cell traits, including absolute counts (AC), median fluorescence intensities (MFI), and morphological parameters across seven major immune cell types: T cells, B cells, dendritic cells (DCs), monocytes, myeloid cells, natural killer (NK) cells, and regulatory T cells (Tregs) [13, 14].
Click to expand...

From the GWAS they reference:
We profiled by flow cytometry 539 immune traits, including 118 absolute cell counts, 389 MFIs of surface antigens and 32 morphological parameters. In addition, we considered 192 relative counts (ratios between cell levels) for a total of 731 cell traits assessed in a general population cohort of 3,757 Sardinians (Fig. 1, Extended Data Figs. 1–6, Supplementary Table 1 and Supplementary Information).
Click to expand...

So for example, for the trait "CD24 on CD24+CD27+", the original GWAS detected the SNPs associated with expression of CD24 on these cells. In theory this thread's study then checks if any of the immune cell-associated SNPs from that other study are associated with having PVFS, from which they infer that these immune cell traits (cell counts, protein expression on cells, etc) cause PVFS.

Mendelian randomization also uses some statistical tools (such as MR-Egger) to exclude SNPs borrowed from the other study which may be directly causing PVFS themselves, though I'm not sure how these tools do this.
 
OK, that makes sense. So this is a way of pre-selecting a relatively small number of allelic variants that you think might be of interest and then doing a linkage study on fewer patients but with a more manageable P value correction than a comprehensive SNP search?

If this is valid then it would seem to be a very nice confirmation of Jo's hunch that CD24 is of major importance to whatever is going wrong.
 
Jonathan Edwards said:
OK, that makes sense. So this is a way of pre-selecting a relatively small number of allelic variants that you think might be of interest and then doing a linkage study on fewer patients but with a more manageable P value correction than a comprehensive SNP search?
Click to expand...
I think that basically covers it. With the added component, which is vital to MR, of excluding SNPs that can directly cause the disease itself [edit: through a pathway other than through the] intermediate trait, so that one can infer causality of the intermediate trait (immune cell trait in this case) on the disease.

Jonathan Edwards said:
If this is valid then it would seem to be a very nice confirmation of Jo's hunch that CD24 is of major importance to whatever is going wrong.
Click to expand...
CD24 is the most significant trait, so there might be something there, but I'd be cautious as it seems they didn't do any multiple test correction and reported anything below an unadjusted p-value of .05 as significant. Even if none of these traits were actually associated with PVFS, around 37 of them would be expected to be below .05 by chance. I emailed the author to double check I'm understanding it correctly, though.
 
Last edited:
forestglip said:
I'm looking at Supplementary Table 1 where there is a column called pval which matches the p values reported in the paper. They used almost all of the immune cell exposures that were under p<.05 (28/31). There's a p_fdr column where one row is .83 and the rest are over .98. So I'm not really sure if they did any kind of multiple test adjustment.
Click to expand...

In the methods they state "To mitigate the concern of multiple testing of p-values stemming from the IVW method, we implemented the False Discovery Rate (FDR) correction. A p-value adjusted for FDR that falls below 0.05 signifies a definitive causal linkage between exposure and outcome." But then they don't really clarify if they p values they list in the paper are the FDR adjusted ones or not. You need to go into the supplementary material (like you did) to figure that out, and it appears they didn't use the FDR adjusted ones in the paper. Papers should be presented a lot more clearly and transparently than this.
 
The authors confirmed that all traits reported in the paper were based on unadjusted p-values. While FDR values were calculated, the only place they were reported was in the supplementary table. They weren't used for identifying traits of interest, and the authors acknowledged that the wording in the methods section could have been clearer.

So in my opinion, there's a good chance that most or all of these 28 traits are false positives, as there were no more significant findings than would be expected by pure chance.

Maybe if there's good reason to consider CD24 or something else interesting before this study, seeing it have the lowest unadjusted p-value might somewhat support the finding. But keep in mind that out of the 731 traits they tested, 11 of them were about expression of the CD24 protein, only one of which was found to be significant in the paper:
CD24 on CD24+ CD27+
CD24 on IgD+ CD24+
CD24 on IgD+ CD38-
CD24 on IgD+ CD38- unsw mem
CD24 on IgD+ CD38br
CD24 on IgD- CD38-
CD24 on IgD- CD38dim
CD24 on memory B cell
CD24 on unsw mem
CD24 on sw mem
CD24 on transitional
Click to expand...
So having many such traits increases the chance that a trait related to CD24 could have the lowest p-value by chance. Maybe it's still surprising enough to consider it a bit of a replication of CD24, I'm not sure.

Edit: Originally put info here, but moved to new post so people would see.
 
Last edited:
You must log in or register to reply here.
Back
Top