strand specific RNAseq not strand specific

[druncie]

We are using a slightly modified version of the following protocol to generate strand-specific RNAseq libraries for Arabidopsis thaliana:

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0026426

It is a dUTP method with a UDG digestion of the 2nd strand before PCR amplification.

We ran a test lane of 24 multiplexed libraries, and everything looked OK (~90% read pairs mapped with tophat2 using fr-firststrand). But RSeQC's infer_experiment.py function finds only ~ 60% of reads are oriented the right way. Here's an example output.

This is PairEnd Data
Fraction of reads failed to determine: 0.0073
Fraction of reads explained by "1++,1--,2+-,2-+": 0.3922
Fraction of reads explained by "1+-,1-+,2++,2--": 0.6004

The statistics are similar for the other libraries.

Is there a likely place where our protocol may have gone wrong and lost the strand specificity?

Our protocol has a couple modifications:
-we are using Superscript III instead of Superscript II
-we are using the inline barcodes from this protocol: Kumar et al 2012

Thanks.

[Michael.Ante]

Hi Druncie,

you migth re-run RSeQC's infer_experiment with a more stringent setting:

• Limit the reference gene model to genes w/o known antisense/overlapping elements (e.g. start with a couple of your favourite genes)
• Increase the sample size
• Increase minimum mapping quality

The genomic strandedness is obfuscated by a lot of ambiguity and the dynamic range within the transcript abundances.
If you need the correct measure for strand-specificity, I suggest to use the ERCC spike-in transcripts for further experiments. They have known length, known concentrations, and have a defined strand-specificity.

[druncie]

Hi Druncie,

you migth re-run RSeQC's infer_experiment with a more stringent setting:

• Limit the reference gene model to genes w/o known antisense/overlapping elements (e.g. start with a couple of your favourite genes)
• Increase the sample size
• Increase minimum mapping quality

The genomic strandedness is obfuscated by a lot of ambiguity and the dynamic range within the transcript abundances.
If you need the correct measure for strand-specificity, I suggest to use the ERCC spike-in transcripts for further experiments. They have known length, known concentrations, and have a defined strand-specificity.

Hi Michael.Ante,

Thank you for these suggestions. I actually noticed this first by looking at a few of my favorite genes with IGV. I've gone ahead and re-run RSeQC gene-by-gene and the results are pretty consistent for each gene. In the attached plot, I show the percentage of forward (1+-,1-+,2++,2--) read pairs for each transcript against the total number of reads on that transcript. As #reads goes up, the percentage of forward direction reads converges to ~60%. So, it doesn't seem to be a problem with specific genes with lots of antisense transcription. All of this is done only with uniquely-mapped reads.

Next time, we'll see about adding some ERCC spike-ins. But any suggestions for where in the protocol something may have gone wrong?

Thanks

[druncie]

Here are a couple suggestions from off this forum:

-our protocol does not uses Actinomycin D during 1st strand synthesis with Superscript III, possibly allowing self-priming and 2nd strand synthesis during the RT reaction.
-our protocol uses UDG, rather than the USER mix (with DNA glycosylase-lyase Endonuclease VIII to break the backbone of the UTP containing strand)

Does anyone have any experience that would suggest that either of these would be the cause of the strand-specificity failure? Is adding ActD going to significantly reduce the efficiency of the 1st strand synthesis as suggested in this forum?