Genetic Risk Factors for ME/CFS Identified using Combinatorial Analysis, 2022, Das et al

published paper posted on this thread here


Preprint

Abstract
Background: Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a debilitating chronic disease that lacks known pathogenesis, distinctive diagnostic criteria, and effective treatment options. Understanding the genetic (and other) risk factors associated with the disease would begin to help alleviate some of these issues for patients.

Methods: We applied both GWAS and the PrecisionLife combinatorial analytics platform to analyze ME/CFS cohorts from UK Biobank, including the Pain Questionnaire cohort, in a case-control design with 1,000 cycles of fully random permutation. The results from this study were supported by a series of replication and cohort comparison experiments, including use of a disjoint Verbal Interview cohort also derived from UK Biobank, and results compared for reproducibility.

Results: Combinatorial analysis revealed 199 SNPs mapping to 14 genes, that were significantly associated with 91% of the cases in the ME/CFS population. These SNPs were found to stratify by shared cases into 15 clusters (communities) made up of 84 high-order combinations of between 3-5 SNPs. p-values for these communities range from 2.3 x 10-10 to 1.6 x 10-72. Many of the genes identified are linked to the key cellular mechanisms hypothesized to underpin ME/CFS, including vulnerabilities to stress and/or infection, mitochondrial dysfunction, sleep disturbance and autoimmune development. We noted similarities with genes associated with multiple sclerosis and long COVID, which share some symptoms and potentially a viral infection trigger with ME/CFS.

Conclusions: This study provides the first detailed genetic insights into the pathophysiological mechanisms underpinning ME/CFS and offers new approaches for better diagnosis and treatment of patients.

Competing Interest Statement
S.D., K.T., J.K, J.S, and S.G. are employees of PrecisionLife, Ltd. S.G. is a shareholder of PrecisionLife, Ltd.

Funding Statement
The project was funded entirely by PrecisionLife Ltd.

https://www.medrxiv.org/content/10.1101/2022.09.09.22279773v1

 

I am unclear what it would mean for SNPs to be significantly associated with 91% of a population. I would expect a statistical association with the population as a whole. Maybe 91% showed at least one relevant SNP allele but that may not mean much?

 

Out of curiosity, I looked at the SNPs and I have 8 out of 14. These are all very common SNPs. I didn't study statistics and so I'm not able to do a sophisticated statistical analysis to find out whether this is reliably above chance level or not.

Still being very curious, the best I could do is look at the european allele frequencies shown on dbSNP for these SNPs:
Summing them up for simplicity, the average reference allele frequency is 0.6951, and the average alternative allele frequency is 0.3048 (when a refSNP had more than one possible alteration in nucleotide, I went with the more common one. The less common one was extremely rare in all cases).

A person has two alleles so gets two dice rolls at having an alternate allele. This works out to a chance of about 0.519 of having this hypothetical "average SNP" with an allele frequency of 0.3048. Therefore the average person should have 14 * 0.519 = 7.27 of these SNPs and I'm barely above average in this respect.

That these SNPs are so common suggests they don't have much effect. But the reported association could be real. DecodeME can't come soon enough.

A potential problem is that the data from the UK Biobank is unreliable due to widespread under and overdiagnosis of ME/CFS.

 

"Many of the genes identified are linked to the key cellular mechanisms hypothesized to underpin ME/CFS, including vulnerabilities to --- autoimmune development. Well, since rituximab failed that statement ("--vulnerabilities to --- autoimmune development--") would be a bit surprising - i.e. if it were true.
Chris Ponting's GWAS study might provide clues re where to look and debunk some theories.

 

Remember that Rituximab eliminates CD-20 B-cells and not plasma cells and other cells which can be a driver to inflammation and pathogenic mechanism of ME. We have yet to find what pathway to disease is wrong. We are still unfortunately in the early days of finding out.
P.S. Willing to stand corrected if any of this is wrong.

 

We certainly are in the early days of finding out ---- my understanding is very limited but long lived plasma cells do produce antibodies and are presumably capable of producing autoantibodies/autoimmunity. I'm open to the idea that autoimmunity may explain some cases of ME/CFS - but there's no supporting evidence at this stage - genetic evidence of autoimmunity or identified autoantibodies.

The way to move this on is GWAS ---- lets hope that deliverers results in the near future.

 

Oh, which mechanisms are those? They list some symptoms, but not any cellular mechanisms. My guess is that they don't have a good grasp of ME, they're just trolling for newsworthy numbers. Another guess: they were hoping to find something amazingly different, such as 80% of PWME having an SNP that was only 1% in the general population. Since they didn't find that, they trolled the numbers to try to justify their funding (and get more).

 

Has anyone been able to find detailed descriptions of all the 15 "communities" of snp clusters they talk about? I've only been able to find brief descriptions for two of them. Or have I missed something?

Out of the 14 snps listed I'm heterozygous for 2 and homozygous for another 2, all from different "community" clusters and none of them amongst the 5 validated in the second cohort. No idea what that means if anything.

 

This is how they explain why their combinatorial approach is supposed to be better than GWAS for ME. Their description of the problem with just looking for single snps makes sense to me (not that I know what I'm talking about). But is their method for overcoming the problem up to the task or is it just a way of creating enough data points to be sure to find something or other?

They're planning on replicating their work on DecodeME data. @Andy @Simon M

 

Yes, informal discussions around that have already happened.

 

Precision Life Ltd https://find-and-update.company-information.service.gov.uk/company/08687703/filing-history


I'm unclear what the practical purpose of this study is, and the repeated reference to Precision Life's "platform" gives the overall sense that this is something of a promo.

The study population seems to be 735 people who 10+ years ago became UK biobank volunteers and a) filled out a pain questionnaire and b) told the UK biobank interviewer they had an ME or a CFS diagnosis (hardly any use of ME/CFS back then). A feature of the UK biobank is that recruitment was limited to those aged 40 - 69 between 2006 and 2010. Participation was voluntary and there was an inevitable bias towards those with good mobility and high social participation.

Assuming that ME/CFS is broadly present in the 500,000 UK biobank cohort as it is in the population as a whole (125,000 > 250,000 per 67mn) then UK biobank should have recorded a representative sample of 933 >1866 PwME. Would that self selected sample of (in 2006 - 2010) mobile or mobility supported and socially participatory PwME be a good sample for a genetic study ? I'm somewhat doubtful that whatever statistical somersaults are performed that anything meaningful can be extracted.

COI - I'm a UK biobank volunteer, I signed up early on, at a time when I was more well than I had been for years and before I developed notable co-morbidities. I think the UK biobank project is a very important piece of epidemiology however I'm very doubtful as to its ability offer detailed insights into illness that have low presence (<1%) in the whole UK population.

 

My Nebula Genomics report said that I have more risk factor variants for MS than almost everyone else in their database. I doubt they have a large database but still.

 

is this statistically sound? multiple comparisons or so?

what is combinatorial analysis? is there an explosion and what is done to reduce it? apologies for asking a q that cold be looked up.

 

I'm aiming to get a blog out later in the week and I'm talking to 1 of the researchers this morning. It's onw of the most complex papers I've ever read, so it might take me a while.

Briefly, this approach, by a techbio company, uses a novel analytical method on existing genetic data (from the UK Biobank) to find a signal where others have found none.

Two things are different about PrecisionLife's approach:
1. They use a combinatorial approach. They are not looking at individual DNA differences but combinations of between three and five SNPs. That makes for a more specific signal. They found 84 of these 3-5 SNP combinations, called disease signatures.
2. They look for individuals who share the same SNP combination. In essence, they are hunting for subgroups. This makes a lot of sense: most researchers think ME is several different illnesses or multiple overlapping illnesses. You will get a stronger signal by focusing on a subgroup.

Note that they combine the disease signatures into 15 subgroups that they call "communities", made up of overlapping disease signatures. These provide the key p values, which range from 2 x 10^-10 to 5x10^-57.

Analysing the 199 SNPs that make up the 84 SNP combinations, they identify 25 "critical SNP's" by statistical means. It is these 25 SNP is that then map to the 14 genes.

I am is a bit sceptical about the gene findings, because I have doubts about the quality of diagnosis in the UK biobank. However, I think the approach is very interesting, and it will be very interesting to see what they come up with when they look at data from DecodeME.

Note this is a black box method — its patented, because they aim to make money by selling their analysis to pharmaceuticals, to help them home in on those promising candidates for drug development.

The paper is also a preprint and so hasn't been peer reviewed.

Also note the replication they managed was very limited.

Cohorts
UK Biobank pain questionnaire , which has a question askig if a doctor has ever diagnosed them with ME/CFS (2382 patients, 4764 controls.
The other cohort asked a similar question in a verbal interview, but only asked about CFS. There were 735 shared between cohorts and these were excluded from the verbal interview cohort, creating a "disjoint" cohort with 1273 patients and 4137 controls that's all I can manage from their, might not be back till later in the week..

ADDED
Is it real?
The big question. They were first out of the block with Covid n June 2020, using just 700 cases in the UK biobank. They found 68 genes, and say 70% of these have'validation' - or biological plausibility - through other approaches e.g. protein-protein interactions. Some of the 70% were replicated in genetic studies, waiting for a number on that.

 

Might this depend on how well defined the cohort is? If we're dealing with several related yet discrete diseases, or if some subset is misdiagnosed (which is entirely plausible given the weak diagnostic process), you might see an association lower than 100%.

 

I see they are presenting at the ME Genetics Research Summit this week

 

PrecisionLife would argue that this could be due to subgroups. The largest subgroup they identify covers 31% of patients. Let's say this is an autoimmune subgroup: if only 30% of those treated with rituximab had an autoimmune disease (and assuming rituximab was the right autoimmune drug), it would be hard to get a positive result. This is a general problem with subgroups, and a big way that precision life thinks its approach can make a difference, by sub grouping patients with different groups responding to different drugs..

 

Hi, I'm looking for some help with my blog.

Assuming everyone on this thread is aware that in this study Precision Life uses combinations of SNP's (between three and five), which it calls disease signatures. Then it searches for all the patients with the same disease signature (actually, it expands this to create subgroups of patients who share overlapping disease signatures). Just a reminder, they found 84 disease signatures, which aggregate into 15 "communities" or subgroups.

I tried to find a visual way to explain how this works and why I think it's clever. Please let me know what you think - I'd rather know if this is a dud idea.

The first image shows patients from three different subgroups. OK, I coloured them slightly different hues, but the basic idea is you can't tell which is which.


How does precision life find the subgroups? It takes a disease signature and searches through the patients for other people people who have the same signature (or similar signature from the same subgroup). For me, it's as if they're using the disease signature to guide a pen, searching through all the patients, joining the dots between the ones that match.

Then you can clearly see the two subgroups (the grey circles are people in neither subgroup).


Thanks