I would say it depends if your experiment is exploratory or if you are planning on publishing those results shortly.

If you know you have good reproducibility with the machine and reagents you are using (low technuical replicate variability), and you are only testing a hypothesis or exploring some ideas, then you could reduce the technical replicates to 2 or even 1.

However, if preliminary analyses indicate your qPCR may give publication-grade results, don't skip steps, it's just not worth it. More technical replicates ultimately means more accurate and reproducible results, potentially increasing your statistical differences.

Personally, I always start with biological triplicates but no technical replicates. If the results are promising (which is rarely the case in fundamental research), I then go on with more replicates to have solid data.

Thanks Martin Chenal for answering!

But what I tried to say is that, given the space constraints of the plate, sometimes one is necessarily confronted with the necessity of limiting either biological or technical replicates. If your technical noise is low, the main problem can be biological noise. Suppose you have groups with 4 biological replicates, each one pipetted three times. If a biological outlier shows in one group, you should skip all three tech. replicates, leaving only 3 biological replicates in these group, then lowering the power of the downstream statistical test.

On the other hand, if you had 12 biological replicates, without technical replicates, the occurrence of the outlier could be easily resolved, skipping the unusual biological replica and leaving you with 11 amplifications that are in principle useful.

I think in many cases there would be no problem with going down from three to two tech replicates.

In some cases, it could lead to more, and no less, reliable results, due to the expansion of biological replicates (more representative sample analysed).

When you have many conditions/genotypes, one choice is to be made. By the way, searching in the literature, I have just found a recent publication ("The Ultimate qPCR Experiment: Producing Publication Quality, Reproducible Data the First Time" , Taylor et al., 2018) that deals with common problems of qPCR exp., and suggests "Avoid interplate variability by pipetting all samples for a given target on a single 96 or 384 well plate." The authors calculate that CV is expected to be around 10% when copy number in the sample is = 100, which is indeed a small quantity of target. So, except for low-quantity targets, there would be no great problem at all.

Therefore they argue that lower expressed genes have higher CV in their Cqs (tech reps), so they have to be more heavily technically replicated. So the convenience (or not) of doing several tech replicates perhaps depends on that. As you said, one has to have a prior knowledge of their system, i. e. the expression level of the genes to be tested, perhaps searching on existing databases when available.

Opinions?

I agree with most of your points. 2 tech replicates seems like a good compromise.

However, I strongly disagree with removing a "biological outlier" as you call it if it only has 1 tech replicate. There is no way you can tell it is an outlier, and removing it represents an important bias in my opinion.

For interplate variability, I never really thought about it because I work a certain way to avoid it. Whenever I have more than one plate, I make sure each plate contains all of the controls (reference genes), so that any given sample is normalized to the controls in the same plate.