Strange MT% in 10X scRNA-seq data analysis

Question

[maria2019]

I have 10X scRNA-seq data for a hESC and after the alignment using cellrenger, I used seurat V3 to cluster cells. Since this is a normal cell line, I expected not to see any clusters of cells. However, the mt% in seurat is VERY strange

and has 2 clusters which finally results in getting 2 separate clusters in the final scRNA-seq analysis (I only keeo mt%< 6). Is such mt% normal? removing 2-6 mt% will reduce the reads from 500,000 to 15,000 and does not seem right! removing 0-2% mt and keeping 2-6 %mt also does not look right to me.

Any suggestions why there is such mt%? Is it only experiment problem? if not, how should I treat it so that i will not lose a lot of reads.

[ATpoint]

Please read and follow How to add images to a Biostars post. I did the changes in the toplevel question for you this time.

[leahurosen]

Hi Maria, were you ever able to figure out what caused this? We're seeing a very similar thing with 10X hESC data of our own: Around 50% of our cells have very low, or even 0 percent.mt. The distribution is clearly bimodal. The cells look happy otherwise (similar nUMI and nFeatures), and importantly the low mito cells aren't a distinct celltype, but rather we see a low and high mito cluster for each of our celltypes (this is confirmed with marker genes). Any hints you have would be most appreciated, because we have no idea what this could be.

[Friederike]

do they have the same cell cycle status? GSEA of metabolic pathways for the marker genes of the low-mito vs. high-mito group of the same celltype would be interesting

[leahurosen]

Based on Seurat cell cycle inference, they do: all span G1, G2M, and S. GSEA would indeed be interesting. Thanks for the idea!

[Friederike]

also, I assume these are cell lines? I've seen very clear bimodal populations for mito-content in spatial transcriptomics data from a tissue section where everything immediately fell into place once we looked at the image and saw very clearly that the two mito-content-populations corresponded to histologically/phenotypically distinct cell types

[tao2013]

It seems such low mt% clusters could be found in many datasets.

Answer 1

[Arup Ghosh]

Can you plot the MT percent graph with y.max =20 ? There can be several reasons for high mitochondrial content like

1) An issue during the library preparation.

2) A higher number of dead and dying cells.

3) MT content will vary among different cell types.

According to most of the sources up to 5-10% mitochondrial content is fine to continue with. Just filter out the cells with more than 10% or 5% depending on the number of cells.

Ref: Biostars.

Ref: 10xportal

Ref: Seurat Example

[maria2019]

Thanks for your comment and the provided links. Below is the link to the Ymax=20 https://hmaryam0.wixsite.com/seurat

The problem that I have is that my mt% is not consistent and it is separated to 2 parts. There are also low% MTs with high read counts. I do not have high mt% in particular, but I have very strange mt% and I am not sure if that is the result of the experiment or the nature of the human embryonic stem cell it self. Does it make sense to remove cells that have below 1.75% mt and only keep those with 1.75-6% mt%?

[GenoMax]

Not sure if this is applicable in your case but something to consider: Mitochondrial Gene percentage threshold in single cell RNA-Seq

[Arup Ghosh]

Maybe there are two metabolically distinct subpopulations is present in the dataset. Also, the UMAP shows two distinct population of cells. I think there will not be any major issue if you consider the cells having <2% MT content, maybe these cells are differentiated and became metabolically less active. It will be best to check top 10/20 features from the different clusters to identify the cell types.