Unexpected RNASeq result in shRNA experiment

Question

[noushin.farnoud]

Dear All,

I have RNASeq data from an shRNA experiment in mouse, but I see more than 2.7 fold over-expression in the shRNA target gene (i.e. setd1a). The shRNA assay had been verified by qRT-PCR with TaqMan probes, where it showed that the expression value was reduced almost by half in our target sample.

I am confident that samples have not been swapped during the computational analysis.

Is there any experimental reason that can explain this observation? Have you seen anything like this before, and if so what further analysis did you do to verify its source?

I greatly appreciate your feedback.

Noushin

[Martombo]

how many counts does the gene have in either condition? how does the distribution of the reads look like? if everything looks fine I would bet on a sample swap (sorry for being skeptical)... you say you're sure this didn't happen during the computational analysis, but it could have possibly happened before, couldn't it?

[noushin.farnoud]

I will give you specific stats later, but the number of samples in the test and control groups are different (3 for control group and 4 for setd1-a group), which makes it very hard to believe a sample swap is possible, don't you think so?

samples                                 control-1   control-2   control-3   setd1a-1   setd1a-2   setd1a-3   setd1a-4
setd1a coverage before normalization    24          39          47          317        659        167        341
setd1a coverage after normalization     26.09       33.1        51.7        395.7      570.6      166.04     276.01

[Martombo]

Isn't it down regulated?

... I mean ...

you're comparing condition "setd1a" condition to condition "control". condition "setd1a" is where the shRNA is present. in that condition the counts of the gene setd1a are fewer compared to the control.

I see no problem here. or am I completely wrong about something?

Edit: the labels were swapped before

[noushin.farnoud]

I corrected the table labels now. Yes, I expect to see down regulation of setd1a compared to control group, but in fact I see the opposite effect. Thanks for your input!

Answer 1

[Michael]

Some ideas here, but mainly speculation:

• you sequenced fragments of the shRNA
• more likely: RNAseq is not that good in detecting post-transcriptional regulation, RNA-interference leads to cleavage of mRNA which is then degraded more quickly, still you'd depend on the degradation to see any down-regulation. The remaining fragments still can be there longer than we thought and are aligned and counted.
• PCR based methods require a transcript that is complete between the primers, so if the mRNA is cleaved in the correct location, the cleaved transcript cannot be detected by qPCR.
• a regulatory feedback loop could increase transcription due to lack of product, leading to increased numbers of (then cleaved) transcripts, again detected mostly by RNA-seq but not by qPCR.

[Matteo Schiavinato]

The remaining fragments still can be there longer than we thought and are aligned and counted.

I know that 2.4 years have passed by since this comment, but: do you have a literature reference to cite for this sentence? I desperately need one and can't find it. I refuse to cite a biostars post in a paper :D

Answer 2

[Gary]

How about the expression profile of setd1b? If Cufflinks was used to quantify the gene expression level, reads expressed from other paralogous genes, e.g. setd1b, could increase the estimated expression level of setd1a. For duplicated genes, HTSeq could be a better choice to do quantification based on my experience, because HTSeq could remove reads mapped onto both setd1a and setd1b. I hope this could be helpful.

[noushin.farnoud]

Thanks Gary for your input. In fact I had used HTSeq to generate the read counts. But that is an interesting point, maybe I should go back to the data and investigate setd1b coverage as well, just in case.