What thermal profile are you using for RT-PCR. Have you standardized the primers annealing conditions with normal PCR.

We are using KAPA SYBR Fast qPCR Mastermix, thermal profile is according to manufacturer's instructions, and primers are from PrimerDesign, who after synthesis fully validated them in their laboratory. I think those should not be a problem, since all other primers (tried them around 30 in setup of this particular mastermix and primers) work perfectly. I genuinely think it is the low expression problem and not the setup.

You can do the normal PCR of your target gene with the higher concentration of template. Have you also did the dissociation curve analysis.

Just so you know: under most conditions, a single transcript will amplify to completion with a Cq value of around 35, so if you're seeing Cq values like this, chances are you're just looking at primer dimers, mispriming or stochastic noise.

This means even if you ARE getting a genuine signal, you can't necessarily trust the results, as stochastic partitioning will be a factor: a microlitre of cDNA with a transcript at 2 molecules per microlitre is quite likely to contain anything from zero to five or more transcripts: your errors will be huge.

If you are detecting actual transcripts (the fact your Cq value decreases if you add more template is encouraging), you can circumvent this partitioning effect somewhat by just performing more replicates (for templates in the 10-20 molecules.ul-1 range I usually use 5 replicate wells rather than 3).

Overall, easiest fixes: use more RNA to make your cDNA, use more cDNA per well. I realise you ARE using more cDNA per well, but still. Also, doublecheck to make sure your cDNA isn't inhibiting the PCR -cDNA synthesis buffers are often toxic to PCR, so using too much cDNA can be as bad as too little. I know: it's a nightmare. 

There ARE target-specific linear amplification methods you can use to increase signal, but every additional amplification step increases potential for error, so I wouldn't recommend it unless you're desperate.

Thank you for your comprehensive answer. I did the dissociation curve and also tried lower concentrations, 0,5- 50 ng/well and primer dimers are forming at lower concentrations.

Also, with concentrations 50 ng/well, I can't get acceptable efficiency. Reason is, that using higher concentration than 20 ng/well is problem by itself, since it is declarated in manual of mastermix, that one should not exceed 20 ng per reaction. 

Currently I'm  thinking of finding more robust mastermix, which wouldn't be so sensitive to higher concentrations of cDNA, but I don't know if that is the right way to approach this problem. I've also considered gene-specific enrichment methods, but I'm afraid of, as you mentioned, increasing the potential error. Any other suggestions?

Honestly, I dunno. I'm tempted to suggest that if the expression is this low, the transcript is likely to be of very little biological relevance, but of course it could be "high expression level in a very small number of cells". If the latter, could you enrich for that population?

It's tricky, really.

You might have some success isolating the polyA fraction from your RNA: this would let you use (a lot) more RNA for cDNA synthesis, but it's expensive and you'll inevitably lose some RNA in the process.

If that is true, it is hard to say, which population would have high expression of GOI, since samples are from liver tissue of a certain gene KO mice. I really don't think that is an realistic option. Thanks anyways for the suggestion.

Anyways, do you think that RT reaction, using oligo-d(T) primers would give much different (better) results in comparison to now used random hexamere primers?

Wouldn't hurt to try: it'd lower you ribosomal/ncRNA cDNA fraction to essentially zero, though be cautious: make sure your qPCR primers are to 3' sequence rather than 5' sequence, especially if it's a long transcript: RTase is not a massively processive enzyme, so has a tendency to fall off when reverse transcribing long mRNAs from the polyA tail.

This means you'll have a 3' bias in your sample: representative 3' sequences but progressively less representative 5' sequence as transcript length increases.

For reference, I work with dystrophin, a 16k transcript when mature, so I use random nonamers for cDNA or I'd never get any 5' sequence at all. :)

using gene specific primers in the RT step is meant to increase sensitivity... but then you can't use reference genes, so you'll have to do absolute quantification.

You could also try a different Polymerase. I'm not familiar with Kapa, so no idea if it is good or not.

35 cycles isn't optimal but reasonable for a lowly expressed gene. Can you detect the expression in your parental mouse strain?

You haven't mentioned what the Cqs on your -RT and H2O samples are like - as long as these are separated at least 6.6 Cts, you can use the data.

@Petra : can we not use a mix of specific primers (GOI and References) in the same RT reaction mix ?

Thomas, 

A very good question, that I do not know the answer for. I have not had to use the specific primers myself, so I don't know if this would be acceptable or if there are differences in cDNA amplification depending on the primers.