Our Smart-3SEQ LCM (PALM) FFPE results/learnings

[Adam Passman]

Hi Smart-3SEQers

I did my first Smart-3SEQ LCM FFPE run a bit more than 2 years ago now and had some really encouraging output. Our metric has been how many different genes we detect in each library; I know this has it's shortcomings, but we sequence the same type of tissue and more or less, the same amount of tissue for each library.

Unfortunately, for reasons I can't explain, my 2nd run had really poor gene counts. On and off for the next 2 years or so I tinkered accross an additional 6 sequencing runs, changing bead ratios and primer concentrations etc and finally, the last 2 runs have me returning to the same or better gene counts as the first run.

I thought I'd just share what made the world of difference in my case and that is changing the lysis temperature from 60C to 70C. For some reasons I can't remember, we decided that we weren't de-crosslinking our RNA sufficiently and saw that this could be achieved at 70C. We tinkered with doing 70C for 5 or 20min after an initial 60min at 60C and also 70C alone for 60 min (no 60C). All these massively increased library yeild, allowing me to use less PCR cycles. I've landed on just doing the 60min lysis all at 70C for simplicity.

Another thing we have found is that when we are working with small amounts of tissue, using 0.1X the recommended concentration of 1S primer (oligo dT) really helped shift our library size peak to the right. This concentration didn't have as large/any such effect when working with our regular amount of material (single crypts, or ~30-50 um^2 on a 10um thickness section:somewhere in the 100s of cells range), but didn't make things worse, so we have stuck with 0.1X.  I did check, using UV spectrophotometry, that I didn't accidentally make the stock up at 10X too concentrated by accident. 

Combined, these changes meant we could drop our bead ratio to be more stringent. When I was having issues, I started to increase my bead ratio all the way up to 1.1X just to keep enough material to sequence, and I was wondering how the recommendation was 0.7X and I think I posted about this on this forum. With the 70C lysis and 0.1X 1S primer concentration I've now done two sequencing runs, one at 0.85X beads and one at a 0.8X ratio. At 0.85X I got 5X+ the number of gene detections I was getting during my failure runs and at 0.8X beads I'm getting 10X+ gene counts. For us, sequencing 50 samples on a High Output Next-Seq, this means we're now averaging 5K+ gene detections with some libraries having 10K+. I should mention that we count a gene as "detected" if we find >4UMI, this is an arbitrary choice though and happy to hear what others think about it. I'm looking foward to dropping the beads down further in the next run!

Another thing we changed was using a UMI with 7N instead of 5N and in a test run we got more gene detections on average with our 7N libraries. This was a minor count difference though and by FAR the biggest difference came from our 70C lysis.

A few observations about samples: We've been getting decent libraries out of blocks that are over a decade old and also from de-waxed sections that have lived in the fridge/freezer for 1-2 years. We also tend to use the same blocks across multiple projects and having them sectioned in an RNA-friendly manner was unfeasable. We use fresh reagents for our de-wax and stain though. We also use methylgreen pyronin as our stain which gives a nice colour an contrast between epithelium and stroma. What I mean convey is that this protocol is very robust and we haven't had to be too concerned about those dreaded RNAses. 

Hope this helps someone.

All the best,

Adam

[Joe Foley]

Thanks for sharing these findings.

Our metric has been how many different genes we detect in each library; I know this has it's shortcomings, but we sequence the same type of tissue and more or less, the same amount of tissue for each library.

Better metrics of a library's information content would be estimated library size, which is intuitive but requires deduplication to calculate (and I'm not sure how much I trust the estimation when you haven't sequenced to saturation), or the entropy of read counts per genome position or per gene, which is easy to calculate but not as easy to interpret. Both of these are provided by the UMI-dedup program: https://github.com/jwfoley/umi-dedup

Combined, these changes meant we could drop our bead ratio to be more stringent.

How did these libraries look by electrophoresis and by % alignable reads? Generally we need the stringent bead ratio because otherwise we get a lot of byproducts with inserts too short to align, or no insert at all (adapter dimers).

Another thing we changed was using a UMI with 7N instead of 5N and in a test run we got more gene detections on average with our 7N libraries.

What software are you using to deduplicate your UMIs? When UMIs are short, we can't ignore the likelihood that two distinct molecules will receive the same UMI by chance, but most software assumes that never happens. So if your software makes that assumption, this difference could be due to that error, which would be worse with 5N. The "weighted_average2" algorithm in UMI-dedup appears empirically to correct that error, but even then it seems to merely remove the harm of bad deduplication without adding any noticeable benefit, so our practice is to use UMIs only for QC metrics and count all reads including duplicates for downstream analysis. You might try that for at least a sanity check. Generally the problem solved by deduplication doesn't seem to be very large to begin with, at least above the single-cell scale, so it doesn't take much of a distortion in the deduplication algorithm to cause more harm than good.

The big downside of having more N's in your adapter is you will get more adapter dimers by spurious annealing, so it would be interesting to see the electropherograms and alignability for this comparison as well. We just set it at 5N because that's the number of cycles used for cluster registration on the newer Illumina sequencers, whose two-color chemistry reads the subsequent GGG as no signal, otherwise I might have been tempted to make it even shorter.

We use fresh reagents for our de-wax and stain though. We also use methylgreen pyronin as our stain which gives a nice colour an contrast between epithelium and stroma.

I don't suppose you could share a photo from the microscope? I'm looking into a wider variety of stains lately myself.

What I mean convey is that this protocol is very robust and we haven't had to be too concerned about those dreaded RNAses.

Fortunately this protocol is more robust against RNase contamination than most, since there aren't many steps with naked RNA before cDNA synthesis. Just keep things clean while you're handling the tissue samples beforehand.


JWF

--
You received this message because you are subscribed to the Google Groups "Smart-3SEQ" group.
To unsubscribe from this group and stop receiving emails from it, send an email to smart-3seq+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/smart-3seq/50466290-a33f-4dca-b9f0-669927f55d70n%40googlegroups.com.

[Adam Passman]

Thanks for your reply.

With the changes we've made, our library peak has shifted to the right (tapestation) meaning I have more scope for bead stringency. The alignable reads have increased, because we do seem to get rid of more of those byproducts you mention.

Thanks for your comments about the UMI's, I'll have to look further into this for our libraries. I believe we've been using UMI-tools.

I'm having trouble attaching photos of our methylgreen pyronin staining. Perhaps I'll send to you directly. I used this stain because I found a paper once that tested RNA extraction using various stains and they concluded that MGP was best.

All the best,

Adam

[Joe Foley]

Thanks. What's the paper that compared the stains?

--
You received this message because you are subscribed to the Google Groups "Smart-3SEQ" group.
To unsubscribe from this group and stop receiving emails from it, send an email to smart-3seq+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/smart-3seq/c39ad9b6-b97f-4133-800a-5e96f8728d8fn%40googlegroups.com.