Thread by @samuel_marsh

As always @LGMartelotto asks the best discussion kickoff questions!! 😀 so here’s a thread (I promise it’s worth full read 😉) on my thoughts and some considerations/caveats for use of transcriptional/translational inhibitor cocktail we describe: biorxiv.org/content/10.110…
1/n

https://twitter.com/lgmartelotto/status/1422319039637069827

Broadly the tl:dr is:
•scRNA-seq w/enzymes > 4°C = Yes
•scRNA-seq mechanical dissociation > 4°C = Yes
•scRNA-seq mechanical dissociation < 4°C = Not necessary but not harmful either*
•snRNA-seq = Not necessary**
2/n

We find in brain and through reanalysis of public datasets in other organs (🤞published soon) that nearly any dissociation that includes steps above 4°C induces signature. Most of these protocols were optimized long before single cell and were simply continued with scRNA-seq
3/n

This includes RT mechanical protocols including, spin steps, and likely those with prolonged FACS steps. We show in our preprint that short FACS (8-10 min sorts) does not induce this signature in whole cells. Though we haven’t tested explicitly we also show that…
4/n

…issues during a sort that result in increased time at RT or increased number of cold>RT>cold transitions can also induce signature. Inhibitors don’t explicitly help here because we’ve washed them out at this point.
5/n

•scRNA-seq mechanical dissociation < 4°C = Not necessary but not harmful either*

*At least in our testing in brain we didn’t see any negative gene expression changes when comparing cold Dounce (4°) with or without inhibitors.
6/n

•snRNA-seq = Not necessary**

**If tissue frozen as quickly as possible (model systems). We don’t see signature in nuclei (even w/FANS) from mice as long as mice are perfused & tissue is dissected & snap-frozen quickly.
Human tissue is different story and we’ll come to that.
7/n

In humans (again assuming tissue is frozen as quickly as possible) presence of signature most likely represents acute pre-/post-mortem variables (see our preprint) and/or specific disease conditions (i.e. cancer, etc):
pubmed.ncbi.nlm.nih.gov/32910905/
8/n

Now for some final considerations.
1. We use transcrip./translat. cocktail. I think this is probably particularly important/advantageous in two contexts a) measurement of intracellular protein (e.g. inCITE-seq) where you don’t want protein translation occurring ex vivo.
9/n

b) RNA velocity. Now this is complicated and my thoughts here don’t reflect any actual experiments but more concepts on what is being measured. RNA velocity is predicated on measuring the dynamics of RNA abundance/splicing at time point of interest. Any ex vivo…
10/n

…degradation of transcripts will bias these assumptions. So transcr. inhib blocks new mRNA and while mRNA decay is complex one aspect is translation mediated decay which would theoretically be inhibited here.
11/n

2. Are you comparing blood and tissue resident immune cells? Well then you need to take extra care! @rmassonix, @hoheyn, and co have already detailed time dependent sampling effects in PBMCs:
genomebiology.biomedcentral.com/articles/10.11…

But what about when you compare PBMCs to tissue?
12/n

Well if you are digesting your tissue with enzymes then those cells are going to be inherently subjected to very different conditions than the cells from blood. And what do immune cells especially well do when subjected to altered microenv.? They RESPOND!
13/n

More that I can hopefully share fully soon (in revised manuscript; though happy to share/talk privately so shoot me DM/email) but in short this can significantly confound identification of tissue-resident vs. circulating signatures.
14/n

3. Orthogonal validation is your friend (in model systems*). If you see something in scRNA-seq but can’t see in smFISH or analogous method then you need to question that. This also goes for ambient RNA contamination issues (but that’s another thread; coming soon 😉).
15/n

As mentioned though while still valuable it’s less applicable for validating/ruling out artificial signature in PM human tissue. snRNA-seq prep itself isn’t inducing changes, they are largely baked in as result of pre-/post-mortem factors. Meaning you still see them via…
16/n

…smFISH but that doesn’t mean they don’t reflect pre-/post-mortem factors rather than the condition you are studying. And certain conditions may be be colinear with these factors. For instance so you want to study aging. You get brains from people of different ages.
17/n

Well people who die in their 20s, 50s, 80s are dying of very different things, in different ways. @lopeskp, @towfiqueRaj, & co showed cause of death was 2nd highest known variable contributing to variance in overall gene expression in bulk microglia
biorxiv.org/content/10.110…
18/n

This is why 4) publishing as much meta data as is possible for PM studies is critical and why it’s unfortunate that so many current studies lack detailed meta data.
19/n

Some final thoughts before Twitter let’s me know I’ve been rambling on too long with thread creation limit 😂.
-These artificial signatures apply in vitro too. How do you lift adherent/semi-adh. cells? Enzymes plus warm temps. So it’s not surprising…
20/n

…similar sig can be found from cel culture RNA-seq if you’re not careful. Not universal but just putting cells on ice can work pretty well!
21/n

-Biological replicates are key!! Even when working with protocol that shouldn’t induce signature. We show this in our preprint (below). In this case we knew what made Sample 4 different but often it’s small difference in processing that can cause shift and if…
22/n

…samples were pooled it wouldn’t be possible to know where the issue was this condition or a technical artifact.
23/n

Finally think critically about protocols before beginning experiment. It’s not exactly surprising that enzymes+heat results in cleavage of extracellular proteins and significant gene expression changes. But that hasn’t stopped the field going back to microarrrays
24/n

Even just flow cytometry or cyTOF can be significantly biased due to extracellular protein cleavage. What protocol you use depends on tissue, experimental question, and downstream assay. Think about what you need and how to get there. Science will always have caveats…
25/n

…but the key is minimizing those caveats to gain the most usable data from experiment.
I think our protocol using transcriptional/translational inhibitors does this and does so in way that requires minimal optimization to apply to range of tissues/enzymes/protocols.
26/n

But I’m always open to someone showing the field another improvement to our protocol, finding a flaw in our protocol, or coming up with a completely new and better way to do things!!
27/n

Now I’ll leave it at that, after failing to finish before Twitter made me publish first part of 🧵 😂.If anyone has Q’s about our preprint, protocol, or related issue again feel free to shoot me email or DM (and don’t be afraid to ping me again if I don’t get back to you).
28/28