Smart-3SEQ: a cheap, fast protocol that is sensitive enough for FFPE single cells

[jwfoley]

Our lab has just posted a preprint about our new RNA-seq protocol, Smart-3SEQ. It combines the previous 3SEQ method for 3' digital gene expression and Smart-seq/Smart-seq2 methods for working with very low amounts of input. The resulting protocol is actually simpler and faster than any previous one (~3 hours) and it uses very cheap off-the-shelf reagents (~$5). Combined with relatively inexpensive sequencing conditions (we use fewer than 10M 1x76 nt reads per human sample) it can be very useful for large-scale studies or just for saving money.

On the small scale, Smart-3SEQ is sensitive and accurate even down to single-cell inputs. Like 3SEQ but unlike Smart-seq/Smart-seq2, it is robust to fragmented RNA, such as from FFPE tissue. In the manuscript we demonstrate this by using Smart-3SEQ to profile single cells extracted from clinical archival tissue, which was not previously possible. Thus Smart-3SEQ enables a range of new studies in molecular pathology and other preclinical research.

The protocol itself is included with the preprint as Supplemental File 2. It includes special instructions for working with the Arcturus XT laser-capture microdissection system on either fresh or FFPE tissue. I've also attached two spreadsheets for ordering the oligonucleotides from IDT: one that you can copy and paste into the "Bulk Input" tool (the 2S primer must be ordered as Custom RNA and the rest as Custom DNA) and one that you can upload to order the barcoded PCR primers in a 96-well plate (you still need the 1S and 2S from the other file). Note that it may be possible to save more money by ordering the PCR primers with IDT's "Ultramer" synthesis, and then you don't need HPLC purification, plus you might only need one phosphorothioate instead of two - but we haven't tested that yet. The current PCR primers are for 48-plex single-indexing but we're now testing 96-plex unique dual indexes as well.

Another way to save money on the protocol is to make your own SPRI bead mix.

I'm happy to answer any questions here about the manuscript, the protocol, LCM, RNA-seq, etc.

[luc]

Thanks for the information Joseph,

I would have two questions:
Why should the Smart oligos be 5' biotinylated?
How much of a balanced library do you need to spike in to sequence through the template-switching Cs without running into read quality problems?

Thanks in advance!

[jwfoley]

Both great questions.

Why should the Smart oligos be 5' biotinylated?

Well, first of all, it has nothing to do with streptavidin, which always confuses people, and that's why our diagram (fig. S1 in Supplemental File 1) just calls it a blocking group. Biotin is actually just a very cheap modification to stick on custom oligos. When it's at the 5' end of the primers in template-switching reverse transcription, it prevents the formation of concatamers: if you didn't have it, then at the ends of your ds-cDNA with both adapters on it, you might just extend additional C-tails, and then additional G-overhang primers could come in and anneal again, until by the end your molecule is a series of duplicated adapters at both ends. Biotin seems to be sufficient to stop this, presumably just by steric hindrance. Another solution people have used in the literature is unnatural deoxynucleotides (iso-dC and iso-dG) at the 5' end, which prevent the extension of the unwanted 3' C-tail by maintaining a 3' underhang as there are no complementary dNTPs to pair with them. However, those are a lot more expensive.

How much of a balanced library do you need to spike in to sequence through the template-switching Cs without running into read quality problems?

Actually we have some degenerate bases (N) between the sequencing adapter and the G-overhang, so the resulting reads all begin with NNNNNGGG before the unique cDNA sequence (fig. S3A). This guarantees sequence diversity in the cycles used for cluster registration on Illumina sequencers, so we only use a 1% PhiX spike-in, and that's just in case the sequencer fails and we need to troubleshoot. The NNNNN also serves as a short UMI.

There can still be some funny business at the beginning of the read, because for three cycles after cluster registration you'll only have G. On the two-color Illumina sequencers in particular, this is read as no signal. So the per-cycle quality graphs look weird. I'm not sure what consequences this has for the 9th cycle, but another thing that can happen is you get more than 3 G's at the beginning, because the C-tailing activity of MMLV RTase doesn't guarantee a particular length. This again screws up the quality metrics, but in practice it doesn't do much harm to the final data because modern read aligners (we used STAR) "soft-clip" non-matching bases at the ends of the read.

[luc]

Thanks Joseph!
Have you perhaps tried extending the UMI a bit?

[jwfoley]

Have you perhaps tried extending the UMI a bit?

No, I'm concerned that having a lot of random bases would invite mispriming and other molecular tomfoolery, plus it's a waste of sequencing cycles if you're using short reads, as we do. Oh, and template switching is sensitive to the length of the oligo. Better instead to use a deduplication algorithm that's robust against UMI collisions, but that's a subject for another paper...