Missing cardiomyocyte detection in DNA methylation data of cardiac ventricles

[SteffanChristiansen]

Hi,

Thank you for providing this interesting resource.

While reviewing cell type outputs, I have noticed an odd absence of cardiomyocytes based on EPICv2 array data from cardiac samples. I would expect the fraction of cardiomyocytes to be estimated in the range of 25-60% among the samples. Yet, in approximately half of the samples, they constitute less than 2% while the max value was less than 20%.

I used the provided DNAm reference from hearts (https://www.biosino.org/episcore/) with weights based on 207 genes. Based on the the output from wRPC, there are 142 genes left in wRPC_output$ref. Using a weight cutoff of 0.4, there are 35 genes in the CM category while the remaining four categories hold between 55-73 genes. Have you observed similar patterns or any explanation for this odd observation?

Thank you for your help.
Steffan

[aet21]

Dear Steffan- thanks for your query. The relative proportion of cardiomyocytes (CM) and smooth muscle cells (SMC) is know to vary very substantially depending on where the samples derive from. If from the aorta, the CM fraction should indeed be very low, as SMC is the dominant fraction and I note that in our publication we validated the Heart DNAm reference against aortic dissection and BAV samples, where the CM fraction was indeed very low. I presume therefore that your samples are from the myocardium, as you expect much higher CM fractions. From what you state, it would appear that our Heart DNAm panel is underestimating the true CM fraction by around 20-30%, and my instinct would be the estimated SMC fraction will correspondingly be overestimated by that amount, although you don't state that. Because both are muscle cell-types, it is possible that there is some confusion between the two if the marker genes that we identified from scRNA-Seq data are not distinguished via promoter DNAm levels. This is certainly possible, or alternatively, the lower number of CMs in the scRNA-Seq assay may have compromised the detecting of CM-specific markers. Depending on your needs there are a number of possible solutions:

1. Use a hiearchical approach where you treat SMC+CM as ONE muscle cell-type. Build a separate DNAm panel to discriminate SMC from CM and use that panel to estimate the subfractions. For this, you will need to sorted DNAm data of SMCs and CMs which by now may be available.
2. Build a new mRNA expression panel using a newer higher quality scRNA-Seq atlas and then rebuild the DNAm panel with EpiSCORE.
3. If you don't care so much about muscle subtype, you can just add the fractions of SMC+CM and proceed with the 4 cell-types: muscle, endothelial ,fibroblast and macrophage, which is better than nothing I guess.
4. Build a new DNAm panel for cardiac samples using existing sorted DNAm data (i.e not using the scRNA-Seq + EpiSCORE imputation approach)
   One final note: although absolute cell-type fractions may be off by 20%, it is likely that the relative fractions are meaningful, and so if are going to use the fractions as covariates in linear models say, then it does not matter greatly if you have offsets, because the regression statistics will be invariant to scaling factors......what matters is the relative fractions.
   Hope this helps,
   Andrew Teschendorff

[SteffanChristiansen]

Dear Andrew,
I much appreciate your detailed and fast response. My samples are derived from the myocardium. As you suggest, the SMC are strongly overrepresented in all samples. For this specific purpose the relative fractions are meaningful, and therefore I'll proceed in that direction.

Thanks a lot!

Best,
Steffan