Exploring a genetic basis for the metabolic perturbations in ME/CFS using UK Biobank, 2025, Armstrong et al

[Nightsong]

Abstract:
Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a clinically heterogeneous disease lacking approved therapies. To assess genetic susceptibility towards a specific metabolic phenotype, we performed a genome-wide association study on plasma biomarker levels (mGWAS) in ME/CFS patients (n=875) and healthy controls (HCs) (n=36,033). We identified 112 significant SNP–biomarker associations in ME/CFS, compared with 4,114 in HCs.

Two SNPs specific to ME/CFS, mapping to HSD11B1 and SCGN, were associated to phospholipids in extra-large very low-density lipoproteins (VLDL) and total fatty acids respectively. Genetic effects of VLDL associations were among the least correlated between ME/CFS and HCs. Heterogeneity tests found differential effects for several lipid traits at ADAP1, NR1H3 and CD40, which are involved in immune regulation. ME/CFS mGWAS summary statistics were decomposed to uncover shared genetic-metabolic patterns, where enrichment analysis highlighted pathways in lipid metabolism, neurotransmitter transport, and inflammation. These findings provide a genetic and molecular rationale for patient heterogeneity and suggest a polygenic predisposition in which many small-effect variants may jointly perturb metabolic mechanisms.

Link | PDF (iScience, December 2025, open access)

 

[Hutan]

An interesting idea to investigate SNP and plasma molecule associations in ME/CFS and healthy control groups.

Our study population was selected from the UKB, which is a large biomedical database with half a million participants, aged 39-70 years, comprising genetic, lifestyle, and electronic health records. The initial cohort included participants with baseline NMR metabolic biomarkers
(n=274,353), where 875 ME/CFS cases (median age of 56 years, 76% females) and 36,033 HCs (median age of 54 years, 53% females) that passed genetic quality control (See methods) were used as our final groups (Table 1).

I'm just reading through this, but, I'm wondering why you would have a significantly different percentage of females in the two groups (76% versus 53%)? There were thousands of participants that could have been chosen as healthy controls, so there presumably could have been better matching if they had wanted to. I'll be interested to see how they adjust the data for that.

Of course, there are other potential differences too that might vary between the two groups - e.g. contraceptive use, activity levels, medicine use, supplement use, diet quality, sun exposure, but not have any impact on ME/CFS status. I'll be interested to see if they considered these differences.

ME/CFS cases were identified from self-reported clinical diagnosis, which may include some misdiagnosed individuals.

We have seen issues/questions with quality of the label of ME/CFS labels in the UK Biobank before. As the authors note, it's an issue to keep in mind.

Edited to add
21.6% of the ME/CFS group reported high levels of physical activity (versus 37.7% in the healthy controls).

Incidentally, there was a really big difference in hand grip strength between the two groups, but I guess that might because of the different percentages of females. Also lower testosterone.. as you might expect.

40.1% of the ME/CFS group were on blood pressure medication but only 0.2% were in the healthy controls. the healthy controls were ultra healthy - not even a case of hay fever. 11.3% of the ME/CFS group had high cholesterol, none of the healthy controls did.

In this study, we performed a quantitative GWAS on 135 NMR metabolic biomarkers including lipoprotein subclasses, lipids, fatty acid ratios and low molecular weight metabolites (Data S1) in both ME/CFS and HCs.

I'm not completely sure that 'biomarkers' is the right word here, although it's hard to think of a better one, given it isn't just molecules but also ratios and presumably groups of molecules. Features? Anyway, they looked at the data for 135 of them.

 

[Hutan]

Results

There were 87 biomarkers that were significantly different between ME/CFS and HCs at the Benjamini-Hochberg adjusted p<0.05

That's a lot of biomarkers that were different between the ME/CFS and HC groups. That's surprising and interesting in itself. I feel that we might have seen data on plasma features in the UK Biobank before?

Association tests were adjusted for confounding factors including age, sex, fasting time, body mass index (BMI), medication (cholesterol lowering and blood pressure), supplements (fish oil) and comorbid conditions (hypertension, asthma, depression, osteoarthritis, high cholesterol, irritable bowel syndrome, unclassifiable, hypothyroidism, hay fever, migraine and gastric reflux).

There's the answer to the confounders. Some adjustments were made. I'm still wondering why they didn't match the controls better. They could have created matched pairs with even things like co-morbidities and things like high cholesterol matched.

SNPs were grouped into independent regions based on linkage disequilibrium (LD)26, where the SNP with the lowest p-value was appointed to represent each region.18

So, they will use individual SNPs as labels, but it is actually regions that have been identified.

There was a total of 112 SNP region-biomarker associations in the ME/CFS cohort at p<1.61×10-9 (Data S3), using the standard genome-wide significant threshold of p<5×10-8, which was adjusted for the number of effective biomarker tests (Figure S1A).

 

[Hutan]

We identified 112 significant genetic associations in ME/CFS, across 8 biomarker groups including high-density lipoprotein (HDL) (n=37), fatty acids (n=19), low-density lipoprotein (LDL) (n=19), very low-density lipoprotein (VLDL) (n=17), other lipids (n=9), IDL (n=6), apolipoproteins (n=4) and amino acids (n=1) (Data S3). There was a total of 27 significant SNPs that mapped to 15 genes (Figure 1A, F and G). Three of those SNPs had not been previously reported in the NHGRI-EBI GWAS catalogue27 or were significantly associated in HCs.

The way I'm interpreting that is there were 112 genetic regions where variation significantly affected levels of biomarkers in the ME/CFS group. So, having a particular SNP would tend to push the level of a biomarker one way or another, compared to not having the particular SNP. This variation is going on within the ME/CFS group, it's nothing to do with the healthy controls. The results are shown in Fig 1A.



This analysis not like DecodeME, where they were looking for different prevalences of SNPs between ME/CFS and healthy control groups. It is finding the associations between genetic variation and variation in the levels of molecules in the plasma.

Figure 1B is the healthy controls. The associations are much stronger because the sample size for the control group is so much bigger (check out the -p values on the y axis scale). You can see that some of the relationships are the same in the ME/CFS and healthy control Manhattan charts. For example at gene 11 I think that means that there are SNPs that have some predictive value for biomarkers. And, that's just what we would expect. Genes make proteins, some of which affect the amount of molecules in the plasma, so, some genetic variation will affect levels of the biomarkers.

There seems to be some suggestion that there are some relationships between genes and biomarker levels that are unique to the ME/CFS group (presumably the ones labelled with the RS numbers in Figure 1A. And, I suppose that might be interesting. But, a number of the differences seem to be related to cholesterol. Although it is said that adjustments were made for confounders, I think the complete removal of anyone with high cholesterol from the healthy controls and the 11.3% of the ME/CFS group with high cholesterol makes that comparison very tricky, It will be interesting to see what the authors say about that.

As far as I can see from a quick google, around 70% of people in the UK in a similar age group have high cholesterol, and women tend to have higher cholesterol than men So, these healthy controls, all with their perfect cholesterol are not normal, and even most of the ME/CFS group are looking pretty healthy on that front.

Happy to be corrected on any of this, especially by @MelbME.

There's more, but that's me done for the day on this.

 

[Hutan]

But my summary so far if you can't be bothered to read my thoughts as I plodded through the paper (and fair enough too) is that this was a nice idea.

But the healthy controls are ultra healthy. So, I'm worried that what has been found has more to do with the differences between a cohort with some people with the common comorbidities of 50 year olds and 75% women and a cohort that is extremely healthy and only around 50% women than any differences attributable to ME/CFS.

Not to mention the issues with the self-reported ME/CFS in the UK Biobank, especially when a substantial proportion of the ME/CFS group report being in the highest physical activity category.

There may well be something interesting here and it's worth spending some time looking for it, but I'm worried that it might be buried under the confounding effects of a poorly matched control group and a poorly identified disease group.

 

[Amw66]

The way I'm interpreting that is there were 112 genetic regions where variation significantly affected levels of biomarkers in the ME/CFS group. So, having a particular SNP would tend to push the level of a biomarker one way or another, compared to not having the particular SNP. This variation is going on within the ME/CFS group, it's nothing to do with the healthy controls. The results are shown in Fig 1A.


View attachment 29524
This analysis not like DecodeME, where they were looking for different prevalences of SNPs between ME/CFS and healthy control groups. It is finding the associations between genetic variation and variation in the levels of molecules in the plasma.

Figure 1B is the healthy controls. The associations are much stronger because the sample size for the control group is so much bigger (check out the -p values on the y axis scale). You can see that some of the relationships are the same in the ME/CFS and healthy control Manhattan charts. For example at gene 11 I think that means that there are SNPs that have some predictive value for biomarkers. And, that's just what we would expect. Genes make proteins, some of which affect the amount of molecules in the plasma, so, some genetic variation will affect levels of the biomarkers.

There seems to be some suggestion that there are some relationships between genes and biomarker levels that are unique to the ME/CFS group (presumably the ones labelled with the RS numbers in Figure 1A. And, I suppose that might be interesting. But, a number of the differences seem to be related to cholesterol. Although it is said that adjustments were made for confounders, I think the complete removal of anyone with high cholesterol from the healthy controls and the 11.3% of the ME/CFS group with high cholesterol makes that comparison very tricky, It will be interesting to see what the authors say about that.

As far as I can see from a quick google, around 70% of people in the UK in a similar age group have high cholesterol, and women tend to have higher cholesterol than men So, these healthy controls, all with their perfect cholesterol are not normal, and even most of the ME/CFS group are looking pretty healthy on that front.

Happy to be corrected on any of this, especially by @MelbME.

There's more, but that's me done for the day on this.

So we have SNPs but do we know their expression..whether they are on or off makes a huge difference.

 

[Hutan]

Yes, there is the issue of epigenetics as well. This study measured the levels of molecules in the blood, so, the outcome of the genes and epigenetics versus the genetic variation.

There are some violin plots in Figure 1 that I haven't looked at yet that I think illustrate the impact of a SNP on biomarker levels (with no, 1 or 2 copies of a SNP).