RNA-seq from single cells?

[jwfoley]

We use a LNA-containing TSO and always have a better coverage at the 5´-end of transcripts than the Clontech kit which uses 3 riboG.

Not a fair comparison since your protocol has several other optimizations relative to the Clontech kit (or just its legacy version, I assume).

The unblocked LNA-containing TSO was still performing better, therefore we don´t block the TSO unless the "hedgehog" signal shows up.

Is it realistic to buy two versions of the oligo, and risk losing precious samples, just to squeeze out a slight performance gain? I would rather stick to the most robust version of the protocol.

[Simone78]

Originally Posted by jwfoley

Not a fair comparison since your protocol has several other optimizations relative to the Clontech kit (or just its legacy version, I assume).

You´re right. But if you check in the Suppl Info of our Nat Methods paper you will see different comparisons. We singled out the effect of the polymerase and the TSO and showed, among other things, that LNA-based TSO is better than rG3-based TSO, all the other factors IN THE SMART-SEQ2 PROTOCOL kept the same.

Originally Posted by jwfoley

Is it realistic to buy two versions of the oligo, and risk losing precious samples, just to squeeze out a slight performance gain? I would rather stick to the most robust version of the protocol.

These days we FACS-sort most of our cells. Every cell type requires few adjustment for optimal results (amount of PCR primers, number of cycles, etc). We always have to test few of them anyway once we start a new project. Only if the TSO creates concatamers we switch to the iso-TSO. The most robust version of the protocol is still the one with the unmodified version of the LNA-TSO (at least for us).

[jwfoley]

We singled out the effect of the polymerase and the TSO and showed, among other things, that LNA-based TSO is better than rG3-based TSO, all the other factors IN THE SMART-SEQ2 PROTOCOL kept the same.

If I'm reading this right (doi:10.1038/nmeth.2639, supplementary table 1, sheet B, rows 34 and 35) your LNA TSO yielded 7% shorter cDNAs than your pure RNA TSO, and 11% shorter than the SMARTer TSO, on average. This is consistent with my suspicion that the gain in yield is due in part to the TSO's mispriming in the middle of the cDNA by strand invasion.

These days we FACS-sort most of our cells. Every cell type requires few adjustment for optimal results (amount of PCR primers, number of cycles, etc). We always have to test few of them anyway once we start a new project. Only if the TSO creates concatamers we switch to the iso-TSO. The most robust version of the protocol is still the one with the unmodified version of the LNA-TSO (at least for us).

I guess we have different ideas of what robustness should be. To me, it means any inexperienced lab can perform the protocol as written and see it work well the first time. Reoptimizing every time, fallback versions, etc. are okay for large service centers or labs that do the same protocol a lot, but they discourage wide adoption of the method.

Also, you mention using iso nucleotides as your 5' blocker. Have you done the test using biotin instead? That seems more common in the literature, and is a lot cheaper (plus it's difficult to get LNAs and iso bases from the same company). In my experience it works just fine to eliminate the "hedgehog".

[Simone78]

Originally Posted by jwfoley

If I'm reading this right (doi:10.1038/nmeth.2639, supplementary table 1, sheet B, rows 34 and 35) your LNA TSO yielded 7% shorter cDNAs than your pure RNA TSO, and 11% shorter than the SMARTer TSO, on average. This is consistent with my suspicion that the gain in yield is due in part to the TSO's mispriming in the middle of the cDNA by strand invasion.

A possibility that can´t be excluded and a phenomenon that is happening for sure to some extent. I already said the protocol is not perfect...
What we wanted to show with that table as well as Suppl Figure 1, 3 and 5 (among others) is the synergic effect of betaine, MgCl2 and LNA-TSO. LNA-TSO was chosen because it improved the efficiency of the strand switch reaction, a limintations when using rG3. Undesired phenomenons are possible, though.

Originally Posted by jwfoley

I guess we have different ideas of what robustness should be. To me, it means any inexperienced lab can perform the protocol as written and see it work well the first time. Reoptimizing every time, fallback versions, etc. are okay for large service centers or labs that do the same protocol a lot, but they discourage wide adoption of the method.

Yes, I guess we have different ideas. If you would have tried the protocol yourself instead of just looking for mistakes or inaccuracies you would have realized how robust the protocol is. I didn´t talk about a complete makeover of the protocol each time you start with a new cell type. I talked about few adjustments which won´t take more than half a day. On the other hand, cells are different in terms of RNA content (to mention one factor) which requires a different number of cycles and amount of primers in order to avoid over/under-amplification, excess of primer dimers etc. I guess that people out there are able to do this without following a protocol (but we also mention it, if you would have read the Nat Protocols paper you would have noticed (step 14). To my knowledge, many groups have already switched to Smart-seq2, got good results and are happy with that. Verba volant scripta manent , to say it in Latin.

Originally Posted by jwfoley

EDIT: on further inspection, those bands are definitely the characteristic "hedgehog" signal from using unblocked oligos. Block them and it should be fine. I don't know why there are any protocols floating around out there that say blocking is optional; it's so cheap to fix.

Originally Posted by jwfoley

Also, you mention using iso nucleotides as your 5' blocker. Have you done the test using biotin instead? That seems more common in the literature, and is a lot cheaper (plus it's difficult to get LNAs and iso bases from the same company). In my experience it works just fine to eliminate the "hedgehog".

Originally Posted by jwfoley

Is it realistic to buy two versions of the oligo, and risk losing precious samples, just to squeeze out a slight performance gain? I would rather stick to the most robust version of the protocol.

Not really clear to me what you suggest...Do you want to block it or not? The "most robust version" according to what you say is the 3rG which is not blocked...but then you suggest biotin as blocking group...I´m a bit confused! Besides, biotin as well as 2-OMe group, phosphates, etc don´t make things better, at least in our hand...I would be really interested in seeing biotin working as a blocking group in this context, if you have results showing that...
Best,
Simone

[jwfoley]

If you would have tried the protocol yourself instead of just looking for mistakes or inaccuracies you would have realized how robust the protocol is.

I did try it, and I spent a couple of weeks testing various things before I ran it on the Bioanalyzer. Then I saw the hedgehog pattern, and realized I had wasted all that time optimizing my yield of concatamers, so I ordered biotin-blocked oligos, waited another week for them to arrive, and moved on.

Not really clear to me what you suggest...Do you want to block it or not? The "most robust version" according to what you say is the 3rG which is not blocked...but then you suggest biotin as blocking group...I´m a bit confused! Besides, biotin as well as 2-OMe group, phosphates, etc don´t make things better, at least in our hand...I would be really interested in seeing biotin working as a blocking group in this context, if you have results showing that...

Sorry if that was confusing. Yes, I always use a blocking group now, but I use biotin instead of unnatural bases because it is much cheaper. I was getting the "hedgehog" fairly consistently without the biotin, and I have never seen it with the biotin, though so far I've only run a few dozen samples of either type on the Bioanalyzer.

Here is the test that convinced me not to do any more preps with unblocked oligos:

Unfortunately we had a bad lot of Bioanalyzer reagents (this seems to be a common problem), so the markers ran too slowly in the first wells, but you can clearly see a hedgehog pattern in every sample where I used unblocked primers. Post-PCR, those peaks totally overload the Bioanalyzer. Even the water (no template) control doesn't have a hedgehog when I use biotinylated oligos - at least not before PCR.

[Simone78]

I think we are talking about 2 different things.
What you have in your Bioanalyzer when you use unblocked oligos looks to me like primer dimers (and not longer concatamers) which are a problem anyway and it´s interesting to see it works for you. We also have such a problem, but usually the final results are not affected. Yes, this "short stuff" gets tagmented by the tn5 and we waste reads sequencing it but it´s never too bad.
Since we also make our own Ampure beads I just changed the % of PEG in the bead buffer, because the % of PEG directly correlates with the beads cutoff. In this way we reduce even more the leftover dimers in the final library.
Anyway, what I mean when I talk about "concatamers" is a profile that looks more like the one I show here. Sample 5 and 6 have a "moderate" amount of concatamers, while samples 9 and 10 are much worse.

Btw, your post-PCR with biotinylated samples aren´t looking too good. You have a lot of short stuff between 60 and 90 sec that shouldn´t be there (premature termination of RT? degradation?). Have you sequenced any single cell using this protocol? I guess these libraries have a strong 3´-bias.

[jwfoley]

What you have in your Bioanalyzer when you use unblocked oligos looks to me like primer dimers (and not longer concatamers)

The smallest peak, which I also get from the blocked oligos, is primer dimer. Larger peaks must be concatamers (primer multimers), since I see them even when I do the TS-RT with no RNA template. You can call them whatever you want, but they contain no useful cDNA sequence and they get above the size range that I can safely size-select out (and at very high concentrations, apparently), so I strongly prefer the blocked oligos to keep it clean. Maybe you get the wider range of concatamer sizes because you're using the LNA TSO?

You have a lot of short stuff between 60 and 90 sec that shouldn´t be there (premature termination of RT? degradation?). Have you sequenced any single cell using this protocol? I guess these libraries have a strong 3´-bias.

These particular samples contained fragmented RNA, so that's expected.

No, I haven't brought it down to single-cell amounts of input yet. But if I do, I may not have the choice to test one batch of expensive oligos to see if it works, and then switch to a different batch of expensive oligos if it doesn't, on the same sample. So I'm only using blocked oligos for every sample.

[Simone78]

Originally Posted by jwfoley

No, I haven't brought it down to single-cell amounts of input yet. But if I do, I may not have the choice to test one batch of expensive oligos to see if it works, and then switch to a different batch of expensive oligos if it doesn't, on the same sample. So I'm only using blocked oligos for every sample.

the cost/sample of the "batch of expensive oligos" is negligible...you don´t have to order buckets of it...
it would be interesting to see the gene body coverage of your TSO compares with the kit and with a "batch of expensive oligo".
Since we made the protocol so cheap I think I´ll stick to a batch of expensive oligos for now, at least until you prove the rest of the world wrong...good luck with the tests!

[jwfoley]

the cost/sample of the "batch of expensive oligos" is negligible...you don´t have to order buckets of it...

Oh, maybe that's why we see this differently. With my supplier, the smallest possible order of the TSO is about 50 micrograms, for about USD 500. Obviously that's enough for hundreds or thousands of libraries, but if you only want to make a dozen, it's a pretty big commitment, especially when the rest of the reagents are off-the-shelf and/or can be bought in much smaller effective amounts. So doubling the amount of TSO you must purchase, just to try a risky way of getting a slight yield increase that may require falling back to the foolproof way anyhow, is rather unattractive if you only plan to do one experiment rather than start a service center. Otherwise you might just buy the kit.

at least until you prove the rest of the world wrong

I don't know what the rest of the world does, but the STRT-seq folks aren't even outside your institution: "Critical: This oligo must be 5′ biotinylated to prevent template switching at its 5′ end (which would generate useless background reads)." (doi:10.1038/nprot.2012.022) At any rate, your own protocol's note that unblocked oligos sometimes result in concatamers is more than enough to persuade me not to waste any time on them.

[Simone78]

Originally Posted by jwfoley

I don't know what the rest of the world does, but the STRT-seq folks aren't even outside your institution: "Critical: This oligo must be 5′ biotinylated to prevent template switching at its 5′ end (which would generate useless background reads)." (doi:10.1038/nprot.2012.022) At any rate, your own protocol's note that unblocked oligos sometimes result in concatamers is more than enough to persuade me not to waste any time on them.

I suggest you reading both papers from the Linnarsson group once again (the Genome Research 2011 and the Nature Protocols 2012). Neither of them has a biotinylated TSO, don´t know where you read it...
In the Nat Prot paper the biotinylation refers to the oligodT primer (STRT-V3-T30VN oligo), while in the Genome Research paper they biotinylated both the STRT-V3-T30VN and the STRT-PCR primer (to capture cDNA later on). It obviously works for them but it was less successful for us (because of the different TSO? I don´t know). That´s why I was trying to block the TSO instead.

[wacguy]

Hi Simone,

I've been using your Smart-seq2 protocol to construct over 20 libraries in the last couple of months w/ great results (Arabidopsois root tissue; around few dozens to few hundreds cells, starting w/ ~30pg-3ng total RNA). This week it stopped working. The problem is somewhere in one of the reactions b4 Tagmentation. There is almost no trace for the ~2000bps peak in the BioAnalyzer after pre-amplification reaction. I know this is one of the weird things we, biologist experience and there is usually no straight forward answer but maybe you have any suggestion or noticed that one of the materials is highly sensitive (e.g., freeze and thaw of the KAPA enzyme).

Besides that, less critical issues, what is the % of mapped reads you get and do you get a lot of rRNA reads (I do). Why did you include the VN nts at the end of the poly(dT) primers and did you try primers that do not have these nts.

Another question is why not using lower Ampure beads ratio after the KAPA reaction same as after the XT reaction to get rid of all oligos that are <200bps.

BTW, I've tried to add more of the poly(dT) primers and it seems that less rRNA is present in the final reads, but as I said, currently, nothing works.

Thanks and have a nice weekend,
Guy

[Simone78]

Originally Posted by wacguy

Hi Simone,

I've been using your Smart-seq2 protocol to construct over 20 libraries in the last couple of months w/ great results (Arabidopsois root tissue; around few dozens to few hundreds cells, starting w/ ~30pg-3ng total RNA). This week it stopped working. The problem is somewhere in one of the reactions b4 Tagmentation. There is almost no trace for the ~2000bps peak in the BioAnalyzer after pre-amplification reaction. I know this is one of the weird things we, biologist experience and there is usually no straight forward answer but maybe you have any suggestion or noticed that one of the materials is highly sensitive (e.g., freeze and thaw of the KAPA enzyme).

Hi Guy,
Difficult to say why it´s not working. KAPA Pol is not that sensitive. In fact, on their website they say you could even leave it at RT over the weekend and it´ll still work (but don´t do it!). The other reagents are not sensitive either. Maybe your RNAse inhibitor has gone bad or you started using a new batch of primers that are not good (problems in the synthesis)? My only suggestion is to replace ALL the reagents and start over, which I know is a bit annoying.

Originally Posted by wacguy

Besides that, less critical issues, what is the % of mapped reads you get and do you get a lot of rRNA reads (I do). Why did you include the VN nts at the end of the poly(dT) primers and did you try primers that do not have these nts.

We usually get 60% uniquely mapping reads, 20% multi-mapping and 20% unmappable. We keep only the uniquely mapping of course and, of those, we have 60% reads mapping to exons, 20% intronic and 20% intergenic. rRNA is never an issue, always <5%, usually around 1-2%.
We include the "VN" to be sure that the oligo dT will anneal at the beginning of the poly-A tail, thus avoiding an unnecessarily long stretch of A that might cause problems to the reverse transcriptase and, later, to the KAPA Pol.

Originally Posted by wacguy

Another question is why not using lower Ampure beads ratio after the KAPA reaction same as after the XT reaction to get rid of all oligos that are <200bps.

Initially that was our intention, but we noticed that we were losing also some of the long fragments along with the short. We are now making our own beads and we change the buffer composition according to the size of fragments we want to recover but it´s never perfect anyway. A solution would be to block the primers, of course. I did limited trials on that but I think that is the only thing that might solve the problem.
Best,
Simone

[wacguy]

I appreciate your reply, changing EVERYTHING, and checking the PCR machine was and is my next plan.

Best,
Guy

[konglongjidan]

for TSO, is HPLC purification necessary for this protocol? and could you recommend any reliable vendors in US for locked nucleic acid. We usually order oligos from IDT, but they don't provide service for LNA. Thanks!

[jwfoley]

Exiqon is the only LNA vendor in North America... though I believe they actually get their orders produced by IDT anyway.

[wacguy]

Hi konglongjidan,

I used the same companies mentioned in the SMART2-seq article (Biomers & Exiqon). I work at Duke so I guess you would be able to do the same. They were all HPLC purified, I wouldn't change that.

Good luck,
Guy

[linampli]

thank you guys for an exciting and very informative exchange of ideas and protocols. smartseq2 with LNA seems to work better than 3rG as mentioned in the paper, at least in my hands. just for information, i get my LNA TSO synthesized by eurogentec (in europe) at 40nmol scale, the yield was 95 micrograms and it cost me 30 euros. so it is not expensive at all.

[eab]

Simone,
Really appreciate all your advice on this forum.
How much have you played with the concentration of betaine during the RT, or investigated other duplex destabilizers?
Many thanks for any feedback.
Eli

[eab]

Also, if one were to leave out the betaine, you would predict a lower yield of library, and bias in the data due to selective loss of reads from transcripts with strong secondary structures?

[Simone78]

I tried 0.5, 1, 1.5 and 2M betaine (all final conc). While the cDNA yield after preampl was higher for 1M when compared to 0.5M, no further improvements were observed when using more than 1M. Additionally, the highest conc of betaine solution you can buy (or make) is 5M, which means that already 1/5 of your reaction volume is already taken when using 1M. Adding >1M becomes impractical.
As we show in the Nat. Method paper, betaine alone is not sufficient but somehow it works only when combined with 6 mM or higher MgCl2 (9-12 mM to maximize yield). I tried betaine in PCR as well (or only in the PCR but no RT) but results were not great and I dropped it. We don´t know if leaving out betaine will cause of lower detection of genes with strong secondary structure, but we do know that we have a worse coverage at the 5´-end of the genes, which might be an indirect proof that we do lose highly structured genes.
As alternatives to betaine I tried trehalose (the second best), sorbitol (bad), DMSO (bad) and formamide (bad). For experiments with betaine, trehalose and sorbitol on tot RNA please see Suppl Table 1 in NM paper.
Best,
Simone