Eccentric medium spiny neuron (eMSN)

[ChronicallyOverIt]

Could be interesting. I'm focusing on other analyses right now, but if someone else want to take a look, I'm happy to provide suggestions.

I’ll try to spin up your repo today and get a grasp on it. No guarantees. This is how @ME/CFS Science Blog made these plots?

 

[ME/CFS Science Blog]

@tralfamadorian97 or @ME/CFS Science Blog would it be useful to do one of these for the OCD data vs ME/CFS?

Yeah, although I would be more interested in the results for the fibromyalgia GWAS if its data is publicly available.
The genetic architecture of fibromyalgia across 2.5 million individuals - PubMed

For the graph I posted, these analyses for eMSN were already done by Duncan et al. They provided the data in their supplementary material. I simply added the ME/CFS data from trafalmadorian79.

 

[Axel]

Tried to make an overview of the geographical distribution of eMSN.

An idea... the eMSNs in the amygdala may play a role in the sense of impending doom that many time precedes a PEM.

 

[Jesse]

I've combined the MAGMA analysis that Duncan et al. 2025 did for other diseases and the analysis @tralfamadorian97 did for ME/CFS using the 461 cell type cluster in the human brain analysis. I've coloured the signals for eMSN (clusters 222-234 and cluster 426) in red.

eMSN are involved in many of these conditions. For bipolar disorder, major depressive disorder, and schizophrenia, the peak signal seems to lie a bit further along the cell numbering (230-300, where the interneurons lie, than for ME/CFS. Also note how different Alzheimer and Multiple Sclerosis look.

View attachment 32509

Does this mean eMSNs are not as interesting as they would have been if it was unique to ME/CFS? I guess if anything it does at least strengthen the neurological angle.

 

[Ravn]

Neuro-Immune Cross-Talk in the Striatum: From Basal Ganglia Physiology to Circuit Dysfunction, 2021, Mancini et al

This one turned up when I was looking for papers connecting eMSNs with the immune system in some way (I’ve only skimmed a few sections based on word searches)

Turns out the review doesn’t mention eMSNs by name, only normal MSNs expressing “membrane heteromeric D1/D2 receptors”. I’m not clear if the latter are or include eMSNs

And the link to the immune system discussed is limited to the CNS in the context of neuroinflammation and neurodegeneration so may not be very relevant to us but just dropping the paper here in case it does contain some useful leads

 

[Nightsong]

In addition to the eMSN relevant papers already discussed, there is a 2025 study in Nature Comms (link) in mice that found GRPR-expressing nucleus accumbens medial shell neurons included both classical & eccentric SPNs; these neurons had high intrinsic excitability, received VTA / hippocampal / amygdala related inputs, and GRPR deletion increased motivation in a progressive-ratio task.

It also mentions that there are no mouse lines that selectively label eSPNs (which is an obvious way in which research on them could progress).

 

[ME/CFS Science Blog]

Does this mean eMSNs are not as interesting as they would have been if it was unique to ME/CFS? I guess if anything it does at least strengthen the neurological angle.

Yes probably but it could still be interesting. Made a comparison to immune cells, which is how I currently try to understand it.

Post in thread 'Eccentric medium spiny neuron (eMSN)'

Sunday at 7:09 PM

So perhaps one way to picture this is to make the analogy to immune disorders where the evidence can suggest a particular cell type (e.g. B-cells) are important. In most immune disorders multiple immune cell types (B, T, NK, neutrophils, dendritic, mast cells, etc.) play a role but there is often one that stands out a bit more. Multiple disorders may have links to B-cells but the cells can be doing other things so that the outcomes are very different, from immune deficiencies, autoimmune disorders to cancers.

For these neuronal cells clusters, perhaps one could make a similar reasoning...

• ME/CFS Science Blog

• 

 

[jnmaciuch]

When you have time/energy, would also be interested in your opinion on the brain imaging studies on fatigue during sickness behavior pointing to the left nucleus accumbens and putamen.

Finally had a chance to skim some of these, the Capuron 2007 one was familiar to me already. I think this is another situation where the general methodology looks okay to me but there might be some technical issue I’m just not able to catch from lack of familiarity.

Though one thing that does concern me is that the focus is on those specific brain regions because other studies have implicated them. Which is generally a logical and valid thing to do especially if you don’t want to end up doing multiple testing corrections across hundreds of small brain loci. But I’ve become a bit disenchanted with some of the brain scan methodology seeing recent papers over the past few months—it makes me worry that there’s something weird about doing these kinds of scans that nearly always yields a statistically significant difference between groups (i.e. wherever you look, unless you’re looking at lots of places and several drop out in the post hoc correction).

Since people tend not to publish negative results I just don’t have a good sense of how often that does happen for these studies, whereas I am less worried about this in other types of experiments bc I know how often my peers end up frustrated. Might be a baseless concern but figured I’d voice it anyways.

 

[Chris Ponting]

@tralfamadorian97 is also doing interesting analyses using the DecodeME data, including the MAGMA analysis that links DNA results of ME/CFS patients to gene expression (scRNAseq) data from the Human Brain Atlas. So this is similar to what Paolo did, but with only the DecodeME data (not those from the veterans program).

MAGMA HBA - ME/CFS Biostatistics Home

trafalmadorian97.github.io

He also found the most significant results for eMSN. I think this is confirmation that the eMSN result isn't due to some possible fluke in the veterans data or an error in the meta-analysis that Paolo did. It's largely in DecodeME data.
View attachment 32448

The signal for eMSN in the veterans data seems rather weak (p = 10^-3 or 10^-2). So I think it mainly added statistical power, mostly from the controls.

I'm greatly impressed with this analysis - and others - being performed on DecodeME, and other, data. I'd like to add a small note of caution though. There is a p-value associated with each cell population. My caution is that this p-value doesn't necessarily indicate that one cell population is more relevant to ME/CFS than another. Even if the p-value of one cell population (cell type-A) is 10-fold 'weaker' than another (cell type-B), cell type-A could in truth be more biologically relevant to ME/CFS than cell-type B. How? Well, it could be that the experimental data for cell type-A has more technical noise than has cell type-B: for example, cell type-A doesn't survive the experimental conditions as well as cell type-B. Also, if the entire analysis (both DecodeME and the Human Brain Atlas) were to be redone, then these p-values (above) would change due to statistical fluctuations: in short, there is a level of imprecision in the data points (above), arising from statistical and technical issues that is not evident in these plots. In summary: a better p-value does not necessarily mean more biological relevance. On the other hand, that p-values are much stronger for multiple different neuron subpopulations than other cells (including glia), does suggest to me that neuron populations contribute most ME/CFS heritability. It's just that I'm not convinced we know yet which neuron subpopulation is most relevant.

 

[Hoopoe]

On the other hand, that p-values are much stronger for multiple different neuron subpopulations than other cells (including glia), does suggest to me that neuron populations contribute most ME/CFS heritability. It's just that I'm not convinced we know yet which neuron subpopulation is most relevant.

Is it possible that the imprecision leads to MAGMA being biased towards neuron-type cells in general?

 

[Chris Ponting]

Is it possible that the imprecision leads to MAGMA being biased towards neuron-type cells in general?

At the moment, I'd say it's unlikely that MAGMA is biased towards neuron populations. Do gain confidence, though, we need to use different data sets (and different data types) and gauge whether they're leading to the same conclusion. This is such an important question to answer, but only by taking due care at each step.

 

[Hoopoe]

There might also be a bias in how humans interpret and research gene expression. It was pointed out by another member here that immune cells also express a number of genes that are typically viewed as being associated with neurons. When a human sees these genes coming up in an analysis, they will think "neurons!", but maybe these genes are associated with the illness due their role in other cell types.

For example NCAM1 (Neural Cell Adhesion Molecule 1) is expressed by several types of immune cells.

That several neuron-related genes are coming up seems to speak against this possibility, but it's something to keep in mind.

 

[ME/CFS Science Blog]

it could be that the experimental data for cell type-A has more technical noise than has cell type-B: for example, cell type-A doesn't survive the experimental conditions as well as cell type-B.

Some counterpoints:

- The MAGMA analysis shows that the eMSN clusters have high BETA, BETA_STD and SE. So when estimating the slope between gene-associations with ME/CFS and gene expression in the cluster, they had high effects and high variation. So the p-value likely wasn't due to higher precision in this area.

- When it comes to the dissections and technical noise: eMSN are spread out over multiple brain regions and are rare in each one. So one would expect them to have more instead of less noise in preservation and sampling . The data also shows low p-values for multiple eMSN clusters from different dissections. That suggest the signal likely isn't driven by a technical artefact in sampling but the genes that eMSN represent.

- eMSN come up a lot in other neurological and psychiatric diseases but in schizophrenia or depression it's more the interneurons that stand out most, rather than eMSN. Both are related so they often pop up together, but the different pattern might also suggests the eMSN result for DecodeME isn't driven by an artefact.

 

[ME/CFS Science Blog]

if the entire analysis (both DecodeME and the Human Brain Atlas) were to be redone, then these p-values (above) would change due to statistical fluctuations: in short, there is a level of imprecision in the data points (above), arising from statistical and technical issues that is not evident in these plots.

I can see there might be variation in the brain cell samples as Siletti et al. only used 4 donors. So the clustering and gene expression per cluster might be slightly different if this was done again. But the same reasoning likely applies to GTEx tissue expression or eQTL-based mapping that the preprint already used or any other external dataset used to help interpret the GWAS data.

The eMSN signal also stands out at different levels of clustering (31, 461 or 2082 cell types) and from different dissections and brain regions. And thousands of genes were used to generate the cluster. So not sure if this signal would change much with different donors or sampling. The results were also largely the same using the mouse cell type data from DropViz.

 

[forestglip]

One point is that in my wide-ranging FUMA cell type analysis on brain cell types (1573 cell types), the most significant were not eMSNs, but instead were types like intratelencephalic neurons of the motor cortex or just "neurons". The most significant eMSN had a p-value of 0.005, while the most significant overall had p=3.47E-07.

Of course, I tested a lot more cell types than Paolo, so the very low p-value could partly be due to randomness (though it passed multiple test correction) and so might not be directly comparable to other study p-values. But this is just to say that the "most significant" can change depending on which cell types are tested.

FUMA doesn't appear to have a way to share cell type results in the browser, but I'll attach the results zip that they provide.

The data also shows low p-values for multiple eMSN clusters from different dissections. That suggest the signal likely isn't driven by a technical artefact in sampling but the genes that eMSN represent.

Good point about it likely not being a technical sampling artifact.

 

[ME/CFS Science Blog]

The most significant eMSN had a p-value of 0.005, while the most significant overall had p=3.47E-07.

Thanks. Do you know which eMSN cell type this was and from which database? Because both the 461 and 2082 clustering from the Siletti atlas resulted in eMSN cell types with p < 10^−6, if I recall correctly.

 

[forestglip]

Thanks. Do you know which eMSN cell type this was and from which database? Because both the 461 and 2082 clustering from the Siletti atlas resulted in eMSN cell types with p < 10^−6, if I recall correctly.

This is what is given for that cell type:

22_Siletti_CerebralCortex.SPL.A5-A7_Human_2022_level2 - Eccentric_medium_spiny_neuron

The FUMA documentation says:

Brief summary: In Siletti et al. 2023, postmortem tissues were isolated from 3 donors. Neurons were enriched from ~100 locations across the forebrain (cerebral cortex, hippocampus, cerebral nuclei, hypothalamus, and thalamus), midbrain, and hindbrain (pons, medulla, and cerebellum). In total, there were ~3M cells that formed 31 superclusters, 461 clusters, and 3313 subclusters. From the data downloaded from cellxgene, we processed 105 datasets that we categorized to different specific regions of the brain. For each of the 105 datasets, we created two files for 2 level of cell type annotations. For more information on how this data was processed, please check: https://github.com/tanyaphung/scrnaseq_viewer/blob/main/notes/14_Silletti_adult_human_brain_notes.md

* I think the provided link has moved to: https://github.com/tanyaphung/scrnaseq_viewer/blob/main/notes/14_Silletti_2023_notes.md

So I'm not clear on the details, but it seems like the data is from the same study, but FUMA somehow processes it so that the cell types are for specific brain regions, while I think the clusters are not necessarily local to any specific region.

Edit: Or maybe it's the same data, just renamed? I'm not really sure.

Edit: I think it may be that these are based on the same data, but I just didn't test the specific eMSN dataset that tralfamadorian tested. I list the 86 datasets I tested in this post, and this does not include every single Siletti dataset. I'm rerunning it now just on all Siletti datasets. I think there are 209 of those. [Edit: 208 datasets. I was including an unrelated Siletti dataset.]

 

[ME/CFS Science Blog]

My guess is that it only tested one of the less notable eMSN such as those in the cerebral cortex but not those in the striatum and amygdala, which is where most eMSNs are and where most of the significant results came from.