Genetics: Chromosome 20: ARFGEF2, CSE1L, STAU1

[forestglip]

I downloaded the GWAS summary stats for that cingulate gyrus metric to see if there's evidence of colocalization.

Visually, at first it looks promising:

The most significant variants in both traits are in the far left portion of the locus.

Zooming in on that most significant little portion, it looks messier:

I ran coloc here, like I did for a trait at the CA10 locus. In this case, the posterior probability of a shared causal variant is 37.6%. So it seems more likely that the traits might have causal variants near each other, but where it's not the same variant.

 

[forestglip]

Looking at phenotypes associated with ARFGEF2 in the AstraZeneca rare variant study results, two traits reach their suggestive significance threshold, one each from binary and continuous traits:

Phenotype	Ancestry	Collapsing model	P value	Odds ratio	Effect size	Type
41202#M7986# Other specified soft tissue disorders Lower leg	European	ptvraredmg	3.62e-7	2.11		Binary
1072#426#CD22#P20273#OID20637#v1# Inflammation	European	flexnonsynmtr	6.48e-7		0.25	Continuous

The second is expression of the protein CD22. From conclusion of a paper about CD22 (Clark 2018):

CD22 plays a key role in affecting B cell responses to Ags [antigens] and innate immune signals, and CD22-CD22L interactions are essential for maintaining self-tolerance. Despite the evidence implicating CD22 in murine lupus, human genetic studies do not support CD22 as a major disease susceptibility locus in SLE. However, it is likely that defects in CD22 combined with other genetic factors have additive or synergistic effects on disease susceptibility. The ability of CD22 to regulate both BCR and TLRs represents an attractive therapeutic strategy for manipulating B cell responses in autoimmunity.

I also found it interesting that fibromyalgia was the third most significant trait:

Phenotype	Ancestry	Collapsing model	P value	Odds ratio
41202#M797# Fibromyalgia	European	flexnonsynmtr	1.12e-5	6.95

When looking at the page for fibromyalgia, ARFGEF2 is the third most significant gene for that trait.

 

[mariovitali]

I found an association of ARFGEF2 (BIG2) with Ehlers Danlos via Periventricular Heterotopia which is mentioned in @Hutan post (1st post of this thread) :

https://www.s4me.info/threads/genetics-chromosome-20-arfgef2-cse1l-stau1.45515/#post-630791

Title : Periventricular Heterotopia: New Insights into Ehlers-Danlos Syndrome
Study : https://pmc.ncbi.nlm.nih.gov/articles/PMC1288408/

A much more rare autosomal recessive form of PH with microcephaly results from mutations in the ARFGEF2 gene which encodes the brefeldin-inhibited guanine exchange factor 2 (BIG2) protein.4 Filamin A interacts with the actin cytoskeleton and thereby regulates various aspects of cell shape, motility and function.5 BIG2 regulates vesicle trafficking from the trans-Golgi apparatus to the cell surface in various tissues of the body. Potentially, the secreted or surface proteins transported through BIG2-dependent vesicles may interact with filamin A near the cell surface and disruption of these as proteins could give rise to PH by impairing neuronal migration.

 

[mariovitali]

ARFGEF2 discussed in the following study mentioning pediatric POTS

https://pmc.ncbi.nlm.nih.gov/articles/PMC12137463/

 

[Jonathan Edwards]

I found an association of ARFGEF2 (BIG2) with Ehlers Danlos via Periventricular Heterotopia which is mentioned in @Hutan post (1st post of this thread) :

I think this is intriguing but potentially confusing. The paper is about some people with a rare genetic disorder, periventricular heterotopia, who had also been told they had 'EDS'. I think the problem here is that 'EDS' can be used to describe anyone with hypermobility. It may be that ARFGEF2 variants encode hypermobility of various degrees. Some rare variants may encode significant hypermobility and skin elasticity. However, it seems likely that these are extremely rare.

In this sense 'EDS' is not one disease entity and so claims of association may be misleading.

I t would be very interesting if some genetic factors influencing hypermobility did, for some obscure indirect reason, also endow risk for ME/CFS but in fact the little reliable epidemiological data we have does not suggest a significant association.

 

[Hutan]

As far as I understood from that paper, there are two causes of Periventricular Heterotopia. One is caused by mutations in the FLNA gene. another is caused by mutations in the ARFGEF2 gene. The paper is mostly about the FLNA gene sort, and it is that sort that seems to be associated with EDS type symptoms such as heart issues and joint mobility.

Mutations in two genes have been identified as the genetic cause of PH. The more common X-linked dominant form of PH is due to mutations in the filamin A (FLNA) gene.1,2 At least 80% of all familial cases of PH are due to a FLNA mutation with some 20% of sporadic cases attributable to this gene.3

A much more rare autosomal recessive form of PH with microcephaly results from mutations in the ARFGEF2 gene which encodes the brefeldin-inhibited guanine exchange factor 2 (BIG2) protein.4

Filamin A interacts with the actin cytoskeleton and thereby regulates various aspects of cell shape, motility and function.5 BIG2 regulates vesicle trafficking from the trans-Golgi apparatus to the cell surface in various tissues of the body. Potentially, the secreted or surface proteins transported through BIG2-dependent vesicles may interact with filamin A near the cell surface and disruption of these as proteins could give rise to PH by impairing neuronal migration.

Thus, the finding of PH and EDS due to FLNA mutations would suggest a novel intracellular cause of EDS.

In the autosomal recessive form of PH due to ARFGEF2mutations, magnetic resonance images also demonstrate bilateral near-continuous nodular PH but have distinguishing features of microcephaly, mild ventriculomegaly and white matter changes.4

The sort of PH caused by ARFGEF2 mutations doesn't seem to be associated with EDS symptoms. At least, the paper doesn't seem to be claiming that.

 

[Jonathan Edwards]

The paper is mostly about the FLNA gene sort, and it is that sort that seems to be associated with EDS type symptoms such as heart issues and joint mobility.

I couldn't get that far! But again, it suggests that this may be a very indirect link if any.

 

[Hutan]

ARFGEF2 discussed in the following study mentioning pediatric POTS

https://pmc.ncbi.nlm.nih.gov/articles/PMC12137463/

We have a thread for that study here:
The genetic landscape of pediatric Postural Orthostatic Tachycardia Syndrome, 2025, Qu et al

 

[ME/CFS Science Blog]

I was playing around with jnmaciuch's DecodeME dataset in the UCSC Genome Browser and was intrigued by how so many of the top DecodeME hits all land on this "ENSG00000294533" guy,

Yeah, I also think this is where the action is.



I the graph above, I finemapped this locus using SUSIE and LD data from UK biobank, similar to what Trafalmadorian did here.

In summary, this uses the correlation among SNPs in the big UK biobank cohort to estimate where the signal is coming from. You can think of it as testing: what if SNP1 was a real signal, how would the other SNPs z-value then look like based on their correlations with SNP1? Tools like SUSIE en FINEMA are able do this search using smart computational tricks. They estimate how many causal signals there are and give a probability for each SNP being the causal one in that set.

One issue is that a mismatch between the LD panel and DecodeME could mess up the results. That's why I added a filter that excluded SNPs whose z-value didn't match the reference panel (kriging_rss' z standardized difference > 3). Using this filter, I got 1 credible set of 60 SNPs. These were selected so that the probability of one of them being the causal one exceeds 95%. But most of the probability is concentrated in 12 SNPs that make up around 85% of the probability and these are all inside ENSG00000294533.

But the big question is, which of these 12 SNPs is driving the effect and what is ENSG00000294533 doing?

 

[jnmaciuch]

But the big question is, which of these 12 SNPs is driving the effect and what is ENSG00000294533 doing?

I should append to one of my earlier posts in this thread: an anti-sense transcript doesn't only affect nearby genes by getting transcribed and then binding somewhere else according to complementarity. It could also help regulate transcription by just getting transcribed itself--keeping the chromatin open and transcriptional machinery in place for other important genes nearby. This might be a helpful overview (full disclosure I have only skimmed):

Using both strands: The fundamental nature of antisense transcription - PMC

Non-coding transcription across the antisense strands of genes is an abundant, pervasive process in eukaryotes from yeast to humans, however its biological function remains elusive. Here, we provide commentary on a recent study of ours, which ...

pmc.ncbi.nlm.nih.gov

Since several of the hits are close to some "CA" regulatory regions ("chromatin accessible" in epigenetic screens), I'm guessing that the presence of an anti-sense gene is just an extra indication that it is particularly biologically important for the door to be kept open around this regulatory region. In which case a mutation in this region might have two potential effects: cutting off anti-sense gene transcription, or just directly affecting some regulatory region on the other strand. I also see at least one CTCF binding site in there, which indicates that this might be a region which loops around and regulates other genes beyond the one closest to it.

The SNPs most likely to affect binding of transcriptional machinery are going to be the longer insertions and deletions, since most TFs have some wiggle room for single nucelotides in their binding footprint. There's one SNP that immediately stands out to me: chr20:48914264 TTGC>T. The region is marked as highly chromatin accessible with high TF occupancy, and the mutation would remove a large chunk of the binding motif for MAFF/MAFB (looking at the JASPAR TF track). Might be wrong about it being the one relevant to ME/CFS, but insofar as any of these SNPs can lead to testable hypotheses, that's very promising.

 

[ME/CFS Science Blog]

There's one SNP that immediately stands out to me: chr20:48914264 TTGC>T.

Yeah I think that's quite possible, it has the second largest probability (PIP) in SUSIE.

Below are the 12 SNPs that carry 85% of the probability. The two most likely candidates are those at the top because (1) they have the largest probability and (2) they are the only ones that are insertions and deletions which likely has a bigger effect in introns that a mere DNA letter change. They are both are quite common mutations with a minor allele frequence around 30%.


The top hit
The leading SNP is the one with the smallest p-value in DecodeME. Here an A is added to position 48914387 on chromosome 20. It’s pretty close to the splice region, the shift from intron to exon.



But my understand is that adding an A here would have little effect. Because this RNA gene is read from the other side we have to read from right to left and switch the letters. So, it basically becomes an extra T in a region that already has multiple after each other as a signal that the intron is about to end. So the signal stays the same, isn't disrupted and probably won’t have a large effect.

Plus strand <-
Reference: ...CCATCTCTACTTAAAAAAAAAAATACA...
Variant: ...CCATCTCTACTTAAAAAAAAAAAATACA...

lncRNA direction ->
Reference: ...TTGTATTTTTTTTTAAGTAGAGATGG...
Variant: ...TTGTATTTTTTTTTTAAGTAGAGATGG...

20:48914264:TTGC:T
The second SNP might be more interesting. It’s a deletion the TTGC pattern becomes just a T. That TGC is part of a pattern (‘TGCTGAC’) that transcription factors MAFF and MAFB recognize and like to bind as shown below in the graph (under JASPAR database). The deletion would probably disrupt this binding.



But it is not clear what effect this would have and through which gene(s). One curious thing is that this SNP came up earlier when I tried to run them all through AlphaGenome, the Google AI model that tries to predict what mutations do. You can test their effect on multiple ‘tracks’ or cell lines. 20:48914264:TTGC:T had the biggest predictive effect, and it was linked to chromatin accessibility and active histone marks in myeloma cell lines (so cancer of plasma cells), namely:

DNase-seq - RPMI8226
DNase-seq - MM.1S
ATAC-seq - NCI-H929
H3K27ac, H3K4me1, H3K4me2 - MM.1S, KMS-11, NCI-H929

Haven’t been able to make sense out of this yet (could be a hallucination) but want to mention it anyway, in case anyone else is able to check this more thoroughly. Given the Daratumumab trial, it might be an interesting finding, but the tools are all quite new to me. Will need to digg deeper to understand them better first.

 

[jnmaciuch]

The second SNP might be more interesting. It’s a deletion the TTGC pattern becomes just a T. That TGC is part of a pattern (‘TGCTGAC’) that transcription factors MAFF and MAFB recognize and like to bind as shown below in the graph (under JASPAR database). The deletion would probably disrupt this binding.

yup that's. what I was saying. thanks for including the screenshot

20:48914264:TTGC:T had the biggest predictive effect, and it was linked to chromatin accessibility and active histone marks in myeloma cell lines (so cancer of plasma cells), namely:

Yes that tracks with the CA-TF regulatory marks I mentioned, and the fact that there is an antisense gene at all. Fits with my intuition that the mere transcription of the gene matters more than any secondary function of the lncRNA. Does AlphaGenome give more specific information on the direction of the signal? That would tell me if the mutation is associated with opening or closing chromatin at that region. My money is on closing, since it looks like that region is held open normally (hence the CA notation) and if you lose the binding motif you probably lose the TF machinery that's functioning as a doorstop of sorts.

 

[V.R.T.]

I have been thinking about this hit and wondering what on earth could be going on here lately after being reminded it was the biggest hit in DecodeME. Glad to see some people are on the case!

 

[jnmaciuch]

If we are right about the MAFF/MAFB binding motif being important, it may also give a clue as to cell type relevance because MAFF definitely isn't expressed in all cell types.

Here's the overall protein expression:

When you click to single cell resolution, organ-specific expression is either lymphocytes (for the immune-related tissues) or endothelial cells. The brain, for example:



This mirrors the expression of ARFGEF2, where the brain signal is also primary driven by endothelial cells.


Even if ARFGEF2 ultimately isn't the functionally relevant gene for this SNP, the overlap between its expression and MAFF's expression is a strong hint that this specific binding site for MAFF near ARFGEF2 might be most relevant in endothelial cells.

If the epigenetic predictions @ME/CFS Science Blog got from AlphaGenome can be trusted, it might also point to lymphocytes, which also have high MAFF expression (and some level of ARFGEF2 expression, indicating that the specific MAFF locus is active in those cells).

So two starting places. The first step would be double checking whether this variant is associated with opening or closing that chromatin region. That might be achievable with existing datasets, I can try to check that when I have time if someone else doesn't get to it first. Second step would be figuring out the functional effect of opening/closing that chromatin region in either endothelial cells or lymphocytes.