Unexpected differential expression

[tboothby]

I was wondering if anyone had any input on what follows below:

I have a pretty straightforward experimental setup for de novo transcriptome assembly and differential expression analysis:

2 conditions: Stressed and unstressed
3 replicates for each condition
RNA extracted first by trizol/chloroform, followed by PolyA selection -> Illumina mRNA lib prep kit v2 (replicates individually indexed)


I get back ~half a billion reads. Pool everything and assembly a de novo transcriptome. I estimate abundance with RSEM for each of my 6 replicate samples and then run differential expression with edgeR.

What I find puzzles me (I am looking for genes that are upregulated in response to stress). Between my 2 conditions I get ~700 differentially expressed genes. I look at those upregulated in my stressed condition and BLAST results are ~99% unknown. I look at those genes down regulated in the stress condition and some of them all look like things that I would expect to be UP not down regulated.

Do you think I have completely messed up my experiment? Besides actual down regulation of a genes transcription I wonder what could explain these results.

If a transcript is translated a lot will that led to faster turn over?

If anyone has any ideas about some biological explanation, I love to hear it.

Cheers,
T

[bruce01]

No biological explanation I am afraid, but did you check back against raw counts? I had the same thing happen before, but found that the script I wrote put conditions back to front in the results, if you get me. So in edgeR: conds<-c("cond1","cond2"), but the results were + for up-regulation of cond2, where I expected cond1... I now always double check with raw reads to see if it makes sense. Just take one highly DE gene, compare mean count of the replicates between conditions.

[tboothby]

No biological explanation I am afraid, but did you check back against raw counts? I had the same thing happen before, but found that the script I wrote put conditions back to front in the results, if you get me. So in edgeR: conds<-c("cond1","cond2"), but the results were + for up-regulation of cond2, where I expected cond1... I now always double check with raw reads to see if it makes sense. Just take one highly DE gene, compare mean count of the replicates between conditions.

Thanks for the reply. Yes I have been looking at the raw reads and they tell the same story. In my stressed condition the raw reads are very low for things I would expect to be high, relative to unstressed.

[bruce01]

Out of interest what sort of genes are these that you expect from stress response? Any chance that the stress is causing a stop in transcription giving the effect? So a pathway is being diverted by this?

[chadn737]

Maybe you mixed up your samples.

Do you have any evidence, say from independent qPCR or microarrays that would tell you that these genes should be higher expressed. I hate the term "up regulated".

[tboothby]

Maybe you mixed up your samples.

Do you have any evidence, say from independent qPCR or microarrays that would tell you that these genes should be higher expressed. I hate the term "up regulated".

No, not yet. I am planning on trying to confirm some of my RNAseq data with qPCR.

Out of interest what sort of genes are these that you expect from stress response? Any chance that the stress is causing a stop in transcription giving the effect? So a pathway is being diverted by this?

Desiccation stress... the organism dried out and becomes ametabolic, after rehydration it recommences metabolism. I thought that initial drying out would trigger a transcriptional stress response and would be detectable in the dried samples. There are a lot of assumptions there, maybe the transcripts are degraded... or transcription stops before complete drying all together.

[rskr]

What is your stress response? If it something like draught, then the samples might not be expressing much of anything, which could lead to the neutral vector problem, where the proportions aren't independent.