I'm a little bit uncertain as to what you're doing. geNorm is more or less designed for the purpose of identifying suitable reference genes, not for generating a normalisation factor for you. You should be using geNorm on a subset of your samples, using a panel of candidate ref genes, then taking the best two or three suggestions.

You then use these two or three ref genes with your entire dataset and generate a normalisation factor from the geometric average of the linearised reference gene signals for each sample. Then use this normalisation factor to normalise your linearised expression data.

It is always better to use multiple reference genes whenever possible, though your suggestion that geNorm is giving you values of 0.95-1.05 suggests to me that either your cDNA levels are so well matched normalisation is more or less redundant, or that you're reading the wrong number.

Could we have a bit more detail?

Might be antiquated by now, but I find linreg to be a pretty good program for calculating primer efficiency and calculating gene expression. Check out their paper (Ruijter et al, 2009). You can also just google LinReg QRT-PCR. Never used GeNorm, so I am not sure if it will solve your problem...

I found this tool very efficient  "Relative Expression Software Tool - Multiple Condition Solver REST-MCS © - version 2 "  , Reference:  Pfaffl MW: A new mathematical model for relative quantification in real-time RT-PCR. Nucleic Acids Res 2001 May 1;29:e45.

I'm a little bit uncertain as to what you're doing. geNorm is more or less designed for the purpose of identifying suitable reference genes, not for generating a normalisation factor for you. You should be using geNorm on a subset of your samples, using a panel of candidate ref genes, then taking the best two or three suggestions.

You then use these two or three ref genes with your entire dataset and generate a normalisation factor from the geometric average of the linearised reference gene signals for each sample. Then use this normalisation factor to normalise your linearised expression data.

It is always better to use multiple reference genes whenever possible, though your suggestion that geNorm is giving you values of 0.95-1.05 suggests to me that either your cDNA levels are so well matched normalisation is more or less redundant, or that you're reading the wrong number.

Could we have a bit more detail?

I have come to realise from an answer by the GeNorm developer in a yahoo GeNorm group that the old GeNorm spreadsheet is only suitable for the standard curve method in terms of using the normalisation value. For ddCT method, the GeNorm is only good to deduce the best 2 or 3 reference genes, no more, and one should convert their value to a linear values first using dCT calculation before inserting their values into GeNorm with the ddCT method. 

They did have develop the new GeNorm incoporated in the qBase software that deduces the normalisation values also for the ddCT method, but there is a subscription fee attached with the software, so I guess I will just use GeNorm for selecting the most suitable ref genes for now and use the geometric means of the 2 or 3 ref genes for normalisation.

Hi John, might I ask when you say linearised reference gene signals, do you mean you have to convert the Ct into a linearised value (e.g. by the dCT calculation?)? I thought I could just the geometric means of the two reference genes and use that for normalising the Ct values of my GOIs?  

(e.g. Ref gene geometric mean Ct: 16.5 cycles

         GOI Ct: 24.5 cycles

dCT =  24.5 - 16.5 = 8, then 2^ddCT = 2^ (dCT of control-8))

Is this correct, or I have to linearise the ref gene level? Sorry I am not too familiarised with the ddCT method as I have been always using the standard curve method.

And do I need to linearse the Ct values before putting them into GeNorm?

I will be grateful if you can help and elaborate on this further. Thanks!

By linearise, I mean convert them to a linear scale from 0-1, where the highest expression for that gene, in that series of samples, corresponds to 1.

This can be done by raising reaction efficiency to the power of the difference between your sample Ct and the highest expression Ct.

Efficiency is here expressed as "extent of doubling each round", so for a 100% reaction (perfect doubling) it would be 2.0, and for a 95% efficient reaction, 1.95.

So if you had 100% efficiency, and Ct values for, say, expression of Csnk2a2:

Sample 1: 23.4

Sample 2: 24.1

Sample 3: 24.3

Sample 4: 23.6

For each sample you would take the highest expression (lowest Ct, so 23.4), and subtract your sample Ct from it, then raise 2 to the power of that.

So

Sample 1: 2.0(23.4-23.4) = 1

Sample 2: 2.0(23.4-24.1) = 0.62

Sample 3: 2.0(23.4-24.3) = 0.54

Sample 4: 2.0(23.4-23.6) = 0.87

These are your relative values, and reflect actual linear representations of expression levels (so while previously a Ct difference of 1 corresponded to a 2-fold difference, now a 2-fold difference corresponds to a 2-fold difference).

And yes, these linearised values are what you should plug into geNorm.

Do this for all your data, ignoring ddCt methods entirely (I just don't really like them, because I'm crap at keeping track of exponentials -I find linearising everything early makes it much easier to get a feel for the data, and spot outliers), then use your chosen reference genes to generate a normalisation factor series by taking the geometric mean (so for two ref genes, it's the square root of the product of the two relative values).

Then simply divide the relative value for your GOI by the normalisation factor.

This final data is linear, not logarithmic, and similar to ddCt methods, the numbers are all arbitrary units, but if you want to work out transcripts you'd need to use a standard curve.

Thanks for the detailed explanation John, really appreciated your effort!

Just a quick question, if the primer efficiency is not exactly (which is often the case these days, some of my primers have around 2.5-3) is it still good to linearise the value with 2^Ct?

Thanks!

If you have efficiencies greater than 2, I would suspect primer dimers or other mispriming events are contributing significantly to your signal. Are you running melt curves (assuming you're using sybr green). You really shouldn't get more than (at most) a doubling of product each cycle.