Pseudoreplication and comparing RNA-seq samples from the same animal

Question

[achamess]

Hi all,

This is as much an experimental design question as a bioinformatic one, but it's something many of doing RNA-seq analysis deal with, so I thought I'd ask here.

I'm looking to compare enrichment of genes in a GFP-labelled set of neurons in mouse. A colleague recommended that as the comparison, we use total cells (GFP+ and GFP-) from the same animals (3 animals). I said we should collect total cells from separate animals (an additional 3 animals). My reasoning is that taking cells from the same animals would constitute pseudoreplication, since the biological units are not really different. It's the same animal. So in my final DE analysis, I couldn't say that n=3 for GFP+ and n=3 for total (GFP+ and GFP-), since I sampled only from the same animals. But if I sample from different animals, then it really is n=3.

What do you think?

Answer 1

[Devon Ryan]

I'm not sure why you're including GFP+ cells in your comparison group, GFP+ vs. GFP- seems like a more reasonable comparison.

Regarding harvesting the comparison group from the same mice, that's not as problematic as you think, it's a paired design. This is more powerful (statistically) and generally preferred when variance is high. This is the same reason one uses litter-mate controls rather than controls from different litters when using mice (as an aside, make sure your mice are from different litters).

I'd interested in how you're getting the RNA out of the GFP+ neurons. It's not like you're going to FACS sort neurons without losing dendritic trees (and we know how important local translation is in neurons, so that'd be a problem). Are you going the "suck them up through an electrode" route?

Answer 2

[achamess]

I'm not sure why you're including GFP+ cells in your comparison group, GFP+ vs. GFP- seems like a more reasonable comparison.

The GFP cells make a considerable (~30%) of all the cells in the tissue. There are other types of cells too (non-neuronal). I was concerned that if I compared to GFP-, I would enrich for genes that are generally neuronal vs non-neuronal, and not cells that are specifically enriched in the GFP+ neurons.

Regarding harvesting the comparison group from the same mice, that's not as problematic as you think, it's a paired design. This is more powerful (statistically) and generally preferred when variance is high. This is the same reason one uses litter-mate controls rather than controls from different litters when using mice (as an aside, make sure your mice are from different litters).

OK. Good to know.

I'd interested in how you're getting the RNA out of the GFP+ neurons. It's not like you're going to FACS sort neurons without losing dendritic trees (and we know how important local translation is in neurons, so that'd be a problem). Are you going the "suck them up through an electrode" route?

Neuronal nuclei actually. Nuclei are good surrogates for whole cells and much easier to release compared to intact neurons. https://www.nature.com/articles/s41598-017-04426-w

[GenoMax]

Please use ADD COMMENT/ADD REPLY when responding to existing posts to keep threads logically organized.

This entire section should have gone under @Devon's answer.

[Devon Ryan]

Sounds like you need a generic neuronal nuclear marker (truth be told, I'm still hesitant about reading too much into nuclear RNA, especially in neurons). I've never looked for one, but I wonder if Stefan Bonn's group has one since they have a LOT of experience sequencing from neuronal nuclei.