Broad size range for Illumina RNA-seq library - secondary subsampling biases?

[JHess]

Hello!

we are currently having mRNA libraries of non-model organisms prepped and have been running into some issues resulting in very broad fragment size distributions.

Neither of the two library construction protocols we have tried (standard Illumina protocols and covaris shearing, afaik) show distinct peaks; The first one produced a range in fragment sizes from about 200 to 700 bps at nearly equal concentrations accross the range. The second resulted in a distribution that is slightly skewed towards larger fragment sizes but is still showing considerable concentrations of shorter fragments.

Seeing that we will use those libraries for de-novo transcriptome assembly, I feel that it would be a bad idea to run the libraries as they are due to the compromises in accuracy of insert size information.

One of the other options we have would be subsampling from these libraries by performing an additional size selection step. Theoretically this should not bias the libraries if shearing was random. I was wondering whether any of you had an opinion on this? What is it that could be causing this broad distribution and can I realistically expect shearing to be (mostly) random?

Also, has anyone tried assembling a library that had similarly broad distributions?

Thanks very much for your help!
Jacky

[pmiguel]

Hard to say without see an image. This could be the normal "bird-nesting/bubble product/daisy chain" hybe phenomenon one will often see with TruSeq libraries after the enrichment amplification. You could try denaturing aliquots from the library (95 oC for 2 minutes followed by "snap" cool on ice) prior to running them on a pico RNA chip. That should remove the inter-molecule hybridization that can produce the multiple/broad peaks and give you an additional a better idea of the true size range of your amplicons.

--
Phillip

[sehrrot]

I agree with pmiguel. that's very common in Truseq libraries. I don't know the reason but I've tried denaturing and it helps a bit. Also, I've considered if I put too much input for bioA but it's also observed in gel. So, last time I ran the gel and ut both 'larger peak' and 'normal peak', then sequenced. Both are basically same. I think it just conformationally bound to each other than increasing their size. I would say two peaks issue is fine but you would be bothered in measuring the mean size of your library, which is needed for your library QC and cluster density.

Apart from this issue, I wouldn't say that this issue is bird-nesting, I know this might be very arbitrary, but bird-nesting describes the situation that un-proper ligation makes two population within one sample - one is ligated (so shown in larger size) and the other is un- or partially-ligated (so shown in the size same as sheared DNA).

[pmiguel]

Really? I just got the term from the Epicentre blog post:

http://epicentral.blogspot.com/2010/03/tangled-fragments-are-for-birds-bird.html

Denaturation does not help for us, unless the sample is subsequently run on a denaturing (RNA) chip. If you run it on a DNA chip, we usually see more of the larger peak(s).

Here is a potentially useful overview of the topic:

http://seqanswers.com/forums/showpost.php?p=47928&postcount=2

--
Phillip