Issues with SOLiD run

[JueFish]

Hi all,

We finally got our SOLiD 5500 up and running and have run into a suite of problems we are currently trying to trouble-shoot. There are three specific issues which I will describe below, but I am wondering if anyone else has experienced anything similar and, if so, if they were able to fix/identify the problem. Keep in mind, all the protocols below will relate to RNA-seq protocols for which all quality controls were well within expectations outlined in SOLiD's protocol manuals. We ran 4 human libraries and 3 non-model fish libraries in this run and all subsequent analyses were done in sequence space (not colorspace).

Issue #1) Significant quality degradation across the read and a lot of primer sequence in final sequence.

Boxplots of quality values show rapid degradation of quality values across the read starting at 33 bp. We lose 15% of our reads due to quality trimming our data and discarding anything below a quality score of 10 and total read length of 30 bp.

Issue #2) We find LOTS of primer sequence in our final sequencing results.

After primer trimming, we lose another 45%-47% of the total reads, using the same length cut-off of 30bp minimum length. Of this, between 4%-10% of the total reads are identified as "primer only." Thus, we lose ~60-65% of our data to poor quality and primer trimming. None of the cDNA libraries have "primer" peaks (that one ~100bp) that exceed manual thresholds and most have very minor ones.

Issue #3) Lots of ribosomal sequences in final sequences.

Of the ~40% of the data that comes back as "usable", 15%-18% of the total sequence (~40% of what's left of the usable reads) has been identified as rRNA. This appears to be the case for both the human and fish samples. All samples were subject to Invitrogen RiboMinus protocols by different users using different kits and neither lab has experienced this kind of over-representation of rRNA in samples previously using similar methods, so while it could be a handling/technical issue, still seems odd. The final cDNA libraries for the human samples tend to indicate one large peak which might be indicative of rRNA contamination, but the fish samples look textbook and would not indicate the same relative amount of contamination even though that is what we find after sequencing.

A few other things to add: (1) we ran ERCC spike-ins (added before RiboMinus) and they look as expected with R-squared values around 0.95; (2) all libraries were cycled at 18 cycles during amplification as, from past experience, we have found it necessary to do so for visualization (using Bio-Rad Experion) and also to hit manual library size targets; (3) average read length after trimming was around 49bp with a standard deviation of 13.

Obviously, we weren't thrilled to only have about 20%-25% of our data usable at the end of everything. SOLiD has given us a few explanations/potential fixes for what they think might be the reasons behind these results (e.g. poor efficiency of RiboMinus kit and need to RiboMinus twice in the future - frustrating if that's the case; too much primer-dimer in final library - doesn't match particularly well wtih Experion or Bioanalyzer traces of library size distributions; primer-dimer disrupting sequencing and driving quality scores down - the lack of 75bp reads still disturbing). We think that it's also possible the Easy Bead prep did not work appropriately.

So we're curious, anyone else had one or more of these problems? Any thoughts or potential insight/wisdom to offer?

Thanks for your time,
Nate

[pmiguel]

We could never get RiboMinus to work at all. (Yields were insanely low.) But I think most of the loss was due to the ethanol precipitations/Invitrogen Concentration devices. We had the same issue with RiboZero (Epicentre) initially but they recommended we use Zymo columns instead of ethanol precipitations. After which it seemed to work well.

However your issue is that RiboMinus failed to sufficiently deplete your samples of rRNA. There you are at the mercy of the tissue you are working with. The SOLiD RNAseq protocol advised using 2 cycles of RiboMinus for rRNA depletion.

On the primer-dimer front. I have harped on this exhaustively in other threads -- but just because you don't see a primer dimer in your BioAnalyzer traces, does not mean it is gone. The primer dimer shares sequence homology with your adapters, so it can also bind to one of your good library molecules, rather than annealing back to another primer dimer strand. Since you are probably running your samples undenatured on the BioAnalyzer, you will not see these "hitchhikers".

--
Phillip

[NextGenSeq]

RiboZero has always worked well for us. We only had a problem once with a bacterial library which had an unusual rRNA variation.

[JueFish]

Thanks for the thoughts. We're definitely looking into the RiboZero kit as an alternative. What species have you guys used it on? Just human or other eukaryotes as well? We're wondering how well it will work for other non-model Eukaryotes. One weird thing for us was that RiboMinus yields were pretty much right where we'd expect them and most RiboMinus mRNA gel chips claimed little rRNA in sample. Also, Philip, you mentioned that tissue-type might have an effect on your ability to remove rRNA? Could you elaborate on that? Was there a specific type of sample you found to be problematic? We used a variety of types and got the same result. As for RiboMinus, SOLiD suggested we do another round of ribo-deplete in the future, Phillip, but, unfortunately, the newest manual I have only says to do that for poly-A selection. Wish they would have mentioned that one earlier...

As for primer-dimer issues, do you evaluate your libraries in a different fashion to identify/remove primer-dimer from your samples as opposed to the Bioanalyzer? Any special tricks to minimize those effects? Lower primer concentrations? Less cycles? Additional size selection? Gel cuts? I'll also poke around for those other threads you mentioned, Phillip. Do you find that primer-dimer binding to good library molecules to be common? If so, at what % of your library sequence?

Weird thing for us is that we see the same result in the end in all our libraries in terms of % sequence lost to variety of factors, but very different fragment profiles going into emPCR (e.g. human samples have a large peak at 258, which we are now being told is likely a 28S fragment, while fish samples look pretty much as manuals tell you they should).

As I mentioned, we're also worried it might be an issue with the EasyBead step failing? Anyone else had an experience where they felt that happened? What did your results look like?

Thanks again for your thoughts.

Cheers,
Nate

[pmiguel]

We have only used RiboZero on some sheep RNA. Seemed to work well.

I would think that anything that lowers the concentration of primer dimers in your un-sizeselected PCR reaction, would also lower the concentration of primer dimer strands associated with full length molecules.

You can denature your samples (95 oC 2 minutes, plus snap chill until loading) prior to running a pico chip on them. The results will be confusing, but what you are looking for is the amount of primer-dimer sized stuff as a molar fraction of your library peak. The denaturation "frees" the hitch-hiking primer dimer strands from association with the full length library molecules. But the conditions of the pico chip are not perfectly denaturing so you may see some re-naturation. Which will cause things to look a good deal more complex because, for the most part, single stranded molecules run slower on Agilent chips than double stranded molecules.

--
Phillip