• If my efficiencies are > 80% and within 5% of one another I routinely use the Livak method providing that I discover that any gene expression differences are > 2 x SEM; that is significant a P=0.05
• This is not the prescribed method but my rational is providing discovered differences are as stated by the Livak method then any departure from the theoretical analysis based on say pfaffl is unlikely to change the data trend

To my opinion it is essential, at least to study relatively small expression changes (less than 10-fold or so).

Suboptimum efficiency - if it was really determined carfully and correctly! - for me is a sign that the PCR is not stable and, therefore, results are not reliable - no matter if they are somehow mathematically considered or "corrected".* It is very implausible that a suboptimum efficiency is kept constant during all cycles, what is in implicit assumption of the quantification. It is much more plausible that the efficiency often is only "appearantly" low because the amplification is severely compromised during the very first 1-3 cycles, whereas the following cycles (amplification of PCR product, less influence of target molecules) the efficiency is actually close to the optimum. But even if the efficiency is low throughout the PCR, I strongly doubt that it is constant. Only when it is optimal at the end, it must have been optimal throuout the PCR and thus must iit have been constant. Only then you can trust that the assumptions used to caclulate the initial amounts (or relative differences) are met and that the results are correct (accurate, unbiased - not neccesarily precise, that's a different thing).

An additional problem is introduced when the efficiency is determined badly or very unprecisely. Unprecise measurements don't allow any reliable statements about the efficiency anyway. The problem here is that the precision of the efficiency estimates are rarely given. If one states an efficiency of, say, 80%, what is the 95% confidence interval? Is this intervall small enough to consider the estimate useful? And frequently I see that the efficiency is determined on amplification curves that haven not been well corrected for background drifts (of the signal). This leads to biased estimates for the efficiency. This sometimes is evident when the slope of the log-linear parts of the amplification curves indicate a considerably larger efficiency than calculated from the ct-values of the dilution series. This discrepancy often seems to remain unrecognized by the researcher.

---

* If there is enough material it is simple to crosscheck if the correction works: the quantifications should be carried out in 2 different dilutions (10- to 100-fold); when the two results are similar, the correction works. This is applicable only when the ct value of the 100-fold dilution is still above the lower limit of quantification, for sure.

The more perfect the efficiency of the reactions, the more straight-forward the math. But, as Jochen responsibly mentions, math is not a cure for poor reactions. Folk-remedy has it that 100% efficient qPCReaction data can be handled by the Livak method, while less than perfect - even very imperfect ones can be dealt with by the Pfaffl method. Rutledge-Stewart 2008 showed that the efficiency is not constant from cycle to cycle (as Jochen pointed out) - so, what we are really seeing in the tube is a human-made phenomenon - pulling off the illusion of a 100% continuously efficient process when all things are 'going well' ... but realize it is still an illusion.  With suboptimal reactions - that illusion is not only shattered, but it may be way further off from idealty than we think it is.  Now go and try to have a good night's sleep after that ;], e.g., if a person uses a horse-shoe to perform a very delicate brain surgery, and the surgery surprisingly "worked" - kick the horse! 

Keeping things as constant as possible and using the same amount of template per each perfectly-rendered reaction with primers and/or primers-probe that pull off the illusion of a 100% efficient 2n process is the best place to start.

As well, doing an up-front analysis of all targets over a 20-point serial 1:2 dilution of a representative sample or sample mixture will show you what you're working with (dynamic range-wise) for each target.  Working within the most log-linear/high efficiency portion of that range, and being certain your sample unknowns enter the qPCR within a valid part of that range is an absolute requirement.  Ah... just kick the horse!  ;]

Dear all,

Thank you all for your valuable suggestions and insight.

My PCR efficiencies were determined by using supercoiled plasmids that were diluted 10 fold serially from 107 copies/ul to 10 copies/ul as templates. PCR with each plasmid dilution was run in triplicate.

I used the Ct mean obtained and plotted the an efficiency curve (Ct mean vs lg. no of copies), and used the gradient to find the PCR efficiency.

I found that at high copy numbers (107 copies and 106 copies), the Ct mean difference was much smaller than 3.3, which is expected for an efficiency of 2. 

Perhaps it was poor pipetting that led to poor dilutions and hence suboptimal PCR efficiency; or it could have been that the supercoiled form of the plasmids inhibits PCR at high copy numbers (which I believe to be the case).

I will redilute my plasmids and re-assess the PCR efficiency with my primers. 

It would be good to relax the plasmids by nicking or even by cutting (to get ssDNA template).

The attached publication ties in nicely here...