Hi there

I don't quite get the formula you're using. Given two samples (a and b), shouldn't it be something like "(1+Efficiency)^[-(Cta - Ctb)]"? This would give you the relative expression (a/b) of a given gene (target or reference). Next, to normalise, simply divide the results for the target gene by the results for the reference gene.

Another thing you can do is select the most stable reference gene(s) using NormFinder or GeNorm before normalising. I can help you out if you want. Good luck =)

Hey Gerson,

Thank you for your help and the suggestion of a different formula. I tried these one on my data and the result was slightly better. I still have the problem though that I get a different result for each reference gene on some samples.

To make things more interesting its not always the same sample and not always the same reference gene that gives these conflicting results.

I ask myself now - should I calculate the mean of my data for all reference genes or should I follow the general trend - when two of three suggest the gene is upregulated then it is? I wonder if this is really a problem of my choosen reference genes.

Regarding the programs you suggest - I will get into Normfinder and see what it will say.

Hello!

Hm, if you're seeing this much variation depending on the reference gene chosen, there could be a problem with their stability. I also use multiple reference genes in my experiments, but I use genorm or normfinder to select the best one(s). Then, based on those results, if you were to choose more than one to normalise data, use the geometric mean of the relative expression values of your chosen references.

Rather stupid question, but are you applying equal amounts of cDNA per reaction? Are your primer efficiencies ok (e.g. 85-100%)? And I reckon you're not using 'classic' reference genes (like GAPDH, TBP, SDHA), right? Maybe it's a matter of stability indeed.

Good luck!

Hey there,

Yes for all of your mentioned points. I will see what Normfinder suggests. How many putative reference genes have you tested before you settled with those you use currently?

Thank you for your help so far!

No problem!

We usually test 5 different reference genes (GAPDH, SDHA, HMBS, HPRT1, TBP), but I guess the more, the merrier. :) But of course the selected reference genes will change depending on which samples and what experimental conditions we are testing.

Cheers

Thank you, Mikael, for your input!

Using NormFinder I stumbled over some problems, definition-wise. The NormFinder Add-In requires me to organize my data using "2 to 8 samples per group".

My experiment is quite complex. The plants I investigate form four treatment groups. Each group is harvested at three time points (=1 sample is taken) so thats makes 12 samples; three samples per treatment from three time points. I harvest tissue of whole plants each time so there is no possibility to take several samples per treatment at each harvesting. (Since I work with Arabidopsis sample biomass is very low).

Question: Is it convenient to see the three timepoint-samples of each treatment as a group of samples and compare those in Normfinder or should each sample from each treatment be treated seperately ( and "samples" then refer to the set of PCRs I performed on this sample)?

So far I used the latter approach and gave Normfinder the PCRs of one timpoint as the set of samples per group.

I usually run three PCRs with each set of samples and a set of genes of interest, using one of three reference genes in each cycle. So I have three repetitions of each reference gene for each set of timepoint-sample.

Anyone not confused now? (:

I am thankful for any input on that matter!

If I got that right, you have a huge experiment that requires three qPCR plates, right?

The endogenous controls should be ran for all samples from all groups - within a defined experimental set. If you intend to compare data from treatment 1 against treatment 2, I think you should consider all samples (from all treatments/subgroups) as a whole group, because you're interested in finding the reference gene that varies the less amongst all your treatments (so you can compare treatments).

If the treatments are independent (treatment 1 won't be directly compared to treatment 2), maybe you could, indeed, consider each treatment as a separate group. This way you'll find the best reference gene for each treatment. That's my guess, I don't know if you agree. :)

As I understand it, you may benefit from using all 3 ref. genes in your normalisation process - and the equation (as I understand it to be) is attached here. But Gerson (and your statements) bring up some key points regarding treatments influencing ref. gene expression differentially.

This issue (since using 3 ref. genes is a "requirement") is indeed the crux of qPCR/RT-qPCR. Also, Gerson mentioned being careful in trying to load the exact same amounts of sample per reaction: that way the ref. genes don't have to 'work so hard' to accomplish what their purpose is (correcting for sample loading).

There is another way to normalise involving use of all of the targets examined in a study - but I don't grasp the full ramifications of that approach just yet.

Thanks for the attachment, Jack. Is there a source for this eqation?

I don't know what to think about the implemantations of this - basically averageing your ref. genes to include them in the formula. I was thinking of something like this myself but then I wonder what the direct biological implication of that is.

I decided against it because I think it will produce results that are not accurate.

Example: Treatment one (solution 1) results in upregulation of gene X, a trend shown with all ref. genes. Treament two (a mixture of solution1+2) results in stronger upregulation, confered by all three ref. genes. Treatment three (solution 2) results in upregulation (ref.gen 1), downregulation (ref.gen 2) or no regulation at all (ref.gen3).

You might think now that obviously the third treatment is the one that gives the ref.genes a hard time. This might even be the case due to physiological changes, similar to cancer symptoms. However, conflicting results like that also occur with treatment two and one. And I chose my ref.genes based on their stability in a microarray that was done with solution 2 to avoid the problems I face now...

You see my dilemma ? (:

I still wonder how other researches have dealt with the variation in their results when using several ref. genes. I can't believe nobody else has had this problem.

The equation I offered is the one I myself unraveled after many hours in Excel based on Vandesompele, Pfaffl, Bustin and others. It is likely a proprietary algorithm tucked inside of commercial software. But it does not undo the reality that you voice so well. Sometimes perfect math and perfect reality never coincide.

It is a labour of love that goes on - and, eventually, you must actually report exactly how you arrived at the conclusions you ultimately do. AKA: in my opinion, there is no good answer to your question. Correct error propagation makes the available software reasonable to invest in - but it is still cryptic to the buyer what the black box is doing.

Reading (and re-reading) Vandesompele's (et. al) stalwart 2002 publication regarding internal control (reference genes) is always a good thing to revisit when this matter arrives.

MIQE warns us that what we think is possible (with qPCR/RT-qPCR) in all situations, may not actually be. I am a member of the MIQE alliance.

Hah! The black box indeed! I spent already too many hours on thinking about that ...

Thank you very much for your valuable insight, Jack! I am relieved that my doubts are not too far from what other - especially more experienced - scientists think. It helps a lot to put things into perspective.

Danke.

C. Savio Chan, that is a powerful link you provided~! Thank you.

With respect to the black box discussion... Biogazelle's qbase+ software is entirely based on peer-reviewed and open access algorithms and formulas (Hellemans et al., Genome Biology, 2007; Vandesompele et al., Genome Biology, 2002).

To answer the question of conflicting results when normalizing with gene A, B or C. First, make sure all 3 are stably expressed (using one of the many published algorithms, e.g. geNorm). Second, using the geometric mean of the relative quantities of all 3 reference genes will do a much better job for normalization (remove better the experimentally induced variation). Third, also consider the magnitude and statistical significance of the fold-change; the conflicting results may be marginal differences.

Article Hellemans J, Mortier GR, De Paepe A, Speleman F, Vandesompel...

Thank you for clearing this up Jo~ I wasn't sure if the algorithms were proprietary or not.