What should I do if death cell counts and types biologically important in my single cell data?

In single-cell analysis, I know that the first steps typically involve filtering out dead cells based on mitochondrial genes, ribosomal genes, low count, and low feature numbers. However, due to the nature of my wet lab experiment, the number of dead cells is important for me. In this case, what should I do? I know that some cells will die due to the experimental conditions. But how can I determine whether the dead cells in my single-cell data are truly due to the experiment or caused by technical issues during the analysis?

scrna cell-death single-cell

• 2,465 views • link

updated 1.7 years ago by i.sudbery

• written 1.9 years ago by carolofharvest •

2 answers

If you are interested in the amount of dying or stressed cells, you could add the percent MT counts per cell instead of removing them. After clustering and dimensionality reduction, make a UMAP plot, color by %MT, and see if the %MT explains any clusters.

Let x be your Seurat object:

x$mito <- PercentageFeatureSet(x,pattern="^MT-"); # change MT to mt for mouse
[...] # all intermediate WF steps until UMAP
FeaturePlot(x, feature="mito", label=TRUE)
FeaturePlot(x, feature="nCount_RNA", label=TRUE) # colour by rna counts as well

Also, look at the barcode rank plot in your cellranger output.

• 0 views • link

written 1.9 years ago by Michael Dondrup

A low-quality cluster is forming in the UMAP plot. However, how can I determine whether the cells in this cluster are truly showing a high mitochondrial ratio due to cell lysis in the droplets, or if they are cells in the early stages of apoptosis? If they are apoptotic cells—but I don't see apoptotic markers based on their transcriptions—I would like to keep them. But if they are due to technical issues and broken cells, I would like to exclude them.

If I keep this cluster despite its low quality, wouldn't it create noise during cell annotation and DEG analysis, making it harder to assign other cells correctly?

• 0 views • link

written 1.9 years ago by carolofharvest •

I don’t have a great answer to your question since I’ve always excluded cells that I think are dead/bad/low-quality/etc. However, since you’re doing a wet lab experiment, can’t you sequence dying cells and see what they would look like?

• 0 views • link

written 1.9 years ago by dsull

If you run FeaturePlot after UMAP as given in the code above, you will get the UMAP plot colored by mitochondrial percentage. You can also generate such plots using the gene names of some apoptotic marker genes. Nobody forces you to follow a "standard" to remove high MT cells if it isn't appropriate for your experiment. Indeed, in the sc workshop I attended, it was recommended to always leave them in and instead color them in the dimensionality reduction plot. The problem is rather with the experimental question. What could you possibly expect to learn from a dead cell? Therefore you shouldn't focus on them but the living ones that still have a coordinated stress response.

Btw, the reason why MT% is used as a filter is that in a dead cell, most other RNA is washed out through the ruptured plasma membrane and only the MT and plastid transcripts may remain because they are held inside the compartment.

• 0 views • link

written 1.9 years ago by Michael Dondrup

Great question! When dead cells are important to your analysis, filtering them out based on mitochondrial and ribosomal genes may remove valuable data. To differentiate between dead cells caused by your experiment and those from technical issues, try these strategies:

Compare Mitochondrial and Ribosomal Content: Check if mitochondrial and ribosomal gene percentages are higher in specific conditions compared to controls. Consistent patterns in experimental samples suggest biological effects, not technical artifacts.

Use Viability Stains: Stains like 7-AAD or propidium iodide during cell sorting can help identify dead cells, linking them to their profiles in your single-cell data.

Check Gene Expression Patterns: Look for stress or apoptosis markers to identify dead cells linked to experimental conditions.

Optimize rRNA Depletion: Enhancing data quality helps distinguish between biological and technical sources of dead cells. Zymo Research’s PureRec Duplex-Specific Nuclease (DSN) effectively reduces ribosomal RNA (rRNA) and globin transcripts, which are often elevated in damaged cells. This improves signal-to-noise ratios and provides clearer gene expression data.

These steps should help you accurately assess dead cells in your single-cell analysis. Good luck with your research!

• 0 views • link

written 1.7 years ago by jnechacov •

Dear @jnechacov thank you for joining our community.

We note that all of the posts you have added in the 12 hours since you joined mention a particular product. Note that while there is no community ban on recommending products which you are associated with, associations must be excplicitly disclosed for posts not to be marked as spam.

Secondly, the use of generative AI in composing posts is again, not outside the rules of the community, but posts using generative AI must be specifically useful to the question or risk being deleted as off-topic. Your answer has a structure common to answers from many generative AI tools, such as chatGPT. Your answer here is superficially relevant, so I'll leave it up for now. But be aware that obviously GPT generated answers are not likely to engender confidence in your product recommendations.

• 0 views • link

written 1.7 years ago by i.sudbery