Are mitochondrial genes to exclude in scRNA-seq such as ribosomal genes?

Question

In this answer, it is stated that ribosomal genes should be excluded prior to normalization in scRNA-seq as contaminants.

Do mitochondrial genes have to be excluded as well? I plotted the top 50 expressed genes for a specific dataset and they tend to appear often (for example MT-ATP6). My assumption is that, given that they work for mitochondrial function and may be highly expressed, they can dilute the signal of gene differential across cell types but expressed at lower levels. Is this biologically sound?

Additionally, in this course, mitochondrial genes are used to filter cells when they contribute above a certain ratio of total RNA of single cells. However, I could not find an explanation for this filtering. What is the underlying process this procedure is used to account for?

Answer 1

According to Ilicic et al. (2016), on upregulation of mtRNA in broken cells:

There is an extensive literature on the relationship between mtDNA, mitochondrially localized proteins, and cell death [34, 35]. However, upregulation of RNA levels of mtDNA in broken cells suggests losses in cytoplasmic content. In a situation where cell membrane is broken, cytoplasmic RNA will be lost, but RNAs enclosed in the mitochondria will be retained [...]

According to this, I think that mtRNA levels relative to endogenous RNA can be used as a control for low quality (broken) cells, in a similar way to ERCC spike-ins.

• Ilicic T, Kim JK, Kolodziejczyk AA, Bagger FO, McCarthy DJ, Marioni JC, Teichmann SA. Classification of low quality cells from single-cell RNA-seq data. Genome biology. 2016 Feb 17;17(1):29.

Answer 2

Mitochondrial genes do not always have to be removed, though you may need to do so for a variety of technical reasons.

Mitochondrial reads are innately different from rRNA due to rRNA typically being excluded during normal library preparation (so any of it getting carried over is essentially a contaminant). Some scRNAseq programs (e.g., RaceID) can use sampling at various stages, which can break if you have a lot of your signal going to mitochondrial transcripts. Then it's not that they should be excluded because they're from mitochondria, but rather that the results are heavily affected by the presence of VERY highly expressed genes and your results are suspect if you don't remove these RNAs. When your methods aren't affected by that (e.g., you use more robust scaling methods) or you don't have super high expression of mtRNA then that's not explicitly needed.

Further, sometimes you very much want to keep mtRNAs, since you might be interested in metabolic changes. Since I work in an institute partially dedicated to immunology, we often look at things like metabolic changes involved in immune activation and removing mtRNAs would hinder that. But if you're a priori not interested in metabolism then excluding genes you won't care about isn't unreasonable.