relative qPCR analysis when gene is not expressed at baseline - (Feb/01/2012 )
qPCR is used routinely in our lab, however I'm only just starting to seriously use it and although I've worked out most of the basics from the internet, there's a few specifics with analysing my experiments that I can't seem to work out (I've given up asking my lab to explain).
The experiments involve taking lung samples over a timecourse in mice after treatment. Some experiments have just one treatment (virus), others have two (allergen and/or virus), giving either two groups (control, virus) or four for the more complex (control, virus, allergen, allergen and virus). There are usually 4 mice per group.
I take one lung lobe and store it in RNAlater, then process it using QIAgen kits. Finally I do the RT reaction on this RNA. I Nanodrop the RNA to check concentration and that it doesn't vary too much, but I'm not currently varying what I put in the RT reaction (it's not varying too much).
What I'm wondering is this: To analyse the data I either do absolute quantification off a standard curve (which is what I'm doing for the viral load in the lung), or relative quantification via either ddCt (if the assays are equally efficient) or the standard curve method (if they're not). For relative quantification we use 18S (we'll skip over whether this is best), and to get fold change I need to use a calibrator sample, which in these experiments the control group for each time point would be ideal. However I'm looking at cytokine mRNA expression, and it's undetectable in my controls, which is not unexpected. This leads to division by zero getting ddCt, and a headache. I've tried purer cell populations with no luck in getting a baseline expression.
So, how do I analyse this data? What I think might work is:
A - normalisation of RNA into RT reaction across whole experiment and read the Ct values against a standard curve (which I have), allowing comparison along the whole timecourse and between groups (though encouraging comparison between experiments and other groups which aren't neccesarily comparable if different amounts of RNA go in) - essentially copies of gene per RNA conc.
B - report just normalised dCt, i.e. Ct(target)-Ct(18S), which I'm guessing is OK to compare along the timecourse but I'm not sure and the numbers are a bit difficult to understand at a glance as they're arbitrary
C - use a calibrator from a high-expressing mouse to get fold changes, but this seems a bit irrelevent as their treatment would be something different
I'm not sure if any of the above make sense. Either way I attempt to purify and run the whole experiment in one go, but this isn't always possible.
My second query is how would you use a calibrator sample anyway, as the 4 baseline mice aren't paired with treatment mice? Would you just average their dCt and subtract from each mouse dCt in the treatment groups so you still got statistics?
Any suggestions gratefully received! Apologies for the long post.
I personally think it would be best to just report absolute quantities of all mRNAs, since it seems you can calculate them. For the control cytokine levels you can indicate that they are undetectable, and then the notion of fold change will be irrelevant.
Sorry I forgot to check back! Thanks for your reply
Absolute numbers does seem like the best option, I'm just worried about what that number means still (as in is it concentration in the lung? What to I work it back to? This is what I do with the flow cytometry results though) and also what happens if I get a particularly low yield of RNA so I can't put enough in the reaction? Normalisation seems to be something that people are pretty keen on, so I'm wondering if normalisation to RNA concentration would be acceptable.
Because I can't detect levels in control mice, I think this is the best way, just want to make sure others have done this!
Normalise to RNA concentration, then to house keeping gene and then to a control sample, because its relative mRNA levels. But relative to what. If you have a treated vs untreated, then that's what you compare. Or if you have a WT animal vs transgenic.
Does this make sense?
I get that, but I'm trying to figure out what you do when there's no expression of the gene in your control samples i.e. nothing comes up after 40 cycles of qPCR.
So I can:
Step 1 - normalise all my reactions to total RNA,
Step 2 - normalise my animals expressing gene to 18S by subtracting 18S Ct from the gene Ct (giving delta Ct) but what about animals without any gene expresion, as with my controls? What Ct value to you assign them? I've seen people get round this by calling their unexpressed samples 40Ct, but it's a bit arbitrary; what if you ran it to 45 cycles and still nothing? 100 cycles and still nothing?
Basically, without gene expression in controls it's not possible to show relative/fold change expression.
I'm trying to find out what people do in this situation as none of the literature covers this. The only thing that makes sense is absolute quantification (read off a curve, report mRNA concentration), but then would you still need to state some sort of qualifier or normalisation, "copies per" what? How do you make this data comparable?
I think the normalisation of RNA is an important first step though, at least you're comparing like for like (I think?)
Relative quatification compares two samples. If one is always zero in all timepoints, it's quite pointless to compare to it, although this could be done by measuring the detection limit of your assay and assigning the negative samples Ct = detection limit+1. You obviously cannot say "there is 3 fold higher expression in my samples" not with relative or absolute quantification.. But you know there is change from the control animals as any expression is always more than no expression.
So you have to either increase the senstitivity of detection (real-time PCR in theory should be able to detect one copy in your sample) or compare to something else. You can choose a model that express your gene. Or you can compare everything to first timepoint. Or you can do absolute quantfication of both your gene and 18S and then show it as 'copies of your gene/1000 copies of 18S' or other suitable amount. That way you normalize your absolute quantification results to housekeeping gene.
But anyway, use the same amount of RNA in your RT reaction, efficiency of RT varies with different concentrations.
Thanks for your reply and going through the options. I think the best thing to do is to run everything with a standard, including 18S, and then that allows analysis by any method. Then when it comes to the genes not expressed in controls, I can show absolute quantification, or if required/looks better/18S variable I can normalise to a quantity of 18S that gives readable number. I think running the 18S in any case is a good check on the amount of cDNA in the mix.
This means making standard curves for everything which I haven't yet managed for a couple of genes, but I think that's just my technical ability!
So in summary; normalise by taking same lung lobe, adding the same amount of RNA to the RT reaction, adding the same amount of cDNA to the qPCR, running 18S to confirm and normalise to "gene copies per 1000 copies 18S"
Thanks for your help everyone, and if anyone knows of some publication examples of similar methods that would be useful, but I'll go have a look myself now anyway.
I found this paper: http://www.gene-quantification.de/integromics-qpcr-statistics-white-paper.pdf
Hope it helps!
Cool, that looks good, will have a read. Thanks!