Flattened TSS enrichment x unique fragments from scATAC experiment

Question

[forestyang]

I am analyzing multiome snRNA/ATAC data from fresh frozen lung cancer samples that were collected 2019 onwards. The RNA side seems to have captured meaningful biological variation and thus looks good. I'm starting to look more into the ATAC side now. For some samples, the distribution of TSS enrichment x number of unique fragments per cell looks very good, like the one below (circular area of high density with >10 TSS enrichment and 10k-100k unique fragments):

However, other samples, like the two below, have a "flattened" distribution concentrated at a lower TSS enrichment, which is concerning:

Has anyone encountered this, knows what this "flattening" signifies or what could be causing it? Thanks!

Answer 1

[LChart]

My guess would be loss of chromatin structure and increase in background accessibility levels. An easy way to check this would be to do the following:

Load the cutsites.bigwig into IGV for "good" and bad samples; pick a 20MB window for normalization (effectively normalizing to the 99.5%ile); set windowing function to max, turn autoscale on and then off again.

Then zoom into regions around ubiquitous genes (EIF4e, etc). If you see a clear higher "background" signal in those suspicious samples (reads outside of "good sample" peaks; not piling up, etc) then the chromatin of those may have been a bit degraded.

The signature is "nonspecific" tagmentation, and lots of other things can do this (excess Tn5, excess DMF, too high tagmentation temperatures, high %dead), but if all the runs used identical protocols, then something may be off with those samples.

[forestyang]

Following the steps you suggested, I do see higher "background" signal in the suspicious samples (top, bottom) relative to the "good" one (middle) at constitutively expressed loci, so I suppose their chromatin is a bit degraded. Thank you for the great tip!

[jared.andrews07]

I expect those are just poor(ish) quality nuclei. If these are using the 10X multiome kit, the RNA tends to be a bit more robust than ATAC. So you can have nuclei that pass RNA QC that have basically useless ATAC. This was a bit surprising to us, but it has proven consistent. Generally, we dump those nuclei, though you could retain them for RNA-specific analyses if one wanted.

I expect removing the offending cells will clean up the background on your tracks.