I've always worked on a principle that 35 cycles is sufficient to reach reaction saturation even with only a single starting template molecule.

If you're seeing Cq values higher than this, I'd would suggest either "primer dimer/mispriming" or low reaction efficiency: 35 cycles is sufficient at an efficiency of 100% (i.e. doubling each cycle) but at 75% efficiency you might end up looking at 40+ cycles.

Assuming it IS genuine signal, a 75% efficiency is not good (you can make a rough estimate of your efficiency from the amplification curve), and even accounting for that, you're looking at very low transcript numbers, which means you're in the realms of stochastic noise.

5000 molecules distributed between three replicate wells will give you something very close to 1666 molecules in each, whereas 5 molecules distributed between three replicate wells clearly cannot give you 1.66 molecules per well, you'll get 1,2,2 or 1,1,3 or 2,0,3 and so on. Thus to get meaningful numbers from low transcript numbers you generally have to do more than 3 replicates (in my experience, 5 or 6 replicate wells seems sufficient for transcripts in the 5-20 range).

You could also take a purified PCR product of your HPV qPCR reaction (run it on a gel to check it's the right thing, gel extract, clean up, quantify via spec) and add it to your reactions (using the same buffers and conditions as your sample qPCRs -ideally even using actual negative control sample) at serial dilutions down to 1-10 molecules per well (or even lower): this would show you what sort of signal you'd have under those conditions. if you still get values in the 40s, then your values are probably real (but need to be repeatedly measured -see above), whereas if with your dilution series you get values that increase linearly until they reach around 35 or so and then jump to either "40" or "nothing", then you're looking at mispriming or primer dimers.

Hi There

I agree with both previous commentators but would add that maybe you should perform a 1:2 serial dilution of your starting cDNA and run a standard curve protocol. You might finds based on that using double or 4 x your current cDNA load will bring you down to 30 cycles for your Ct. I should add that a standard curve will also tell you something about the efficiency of your primers: You want them really to be >85% and if they are not then you might want to consider your primer design .I should add that at higher cycle numbers triplicates associate dwith the std curve tend to vary much more with repect to individual Cts; only select a a cDNA amount that yields concordant triplicate Ct values (R squared > 0.98) and generally that tends to be below 30 cycles (but not always)

If you need any more help drop me a line

Agree with other commentators, and in most of the case, Ct value more than 30 is considered as primer dimer or unspecific amplification. Maybe quantify your RNA first?In addition, a melting curve is suggested if you are not using probes.

Thanks a lot Kaushal, John, Laurence and Fang!   

This is entirely dependent upon the assay: hardware platform, chemistry, cDNA input concentrations, reaction volume and Cq threshold setting; SYBRGreen based assays typically give lower Cqs than probe based (e.g. TaqMan). It is possible to derive a usable cut-off from an absolute quantification method - if you can create (or buy) an accurate control of your gene of interest (such as a plasmid quantitated by digital PCR), then by running a standard curve (typically 10^6 - 10^1 molecules per reaction volume) you can calculate an intercept, which is the hypothetical Cq for 1 molecule. Guidelines for molecular monitoring in the BCR-ABL1 setting (CML) recommend a cut-off for positivity of the Intercept Cq+1; although the quantitative range will be defined by your lowest plasmid dilution point (typically 10). (Foroni L, et al. (2011). Guidelines for the measurement of BCR-ABL1 transcripts in chronic myeloid leukaemia. Br J Haematol. 153:179–190). If using a SYBRGreen based relative qPCR assay, then something like 4 Cqs lower than the lowest Cq of your NTC would be a reasonable rule of thumb.

All previous answers make good points ans suggestions. To answer the initital question - how to decide where to draw the threshold for your assay - it would be good to increase your cDNA input to get a smaller Ct value. Kaushal already stated the easiest way to do this: Put more (or higher concentrated) RNA in your cDNA preparation, and/or put more cDNA into your rtPCR reaction. From then on (once you have shifted the Ct to 38 or lower) I agree with John and Laurence: try serial dilutions and monitor the Ct values as they describe to judge if your signal is "real".

I assume you are wondering if it is an analytical positive (clinically relevant is a completely different story). Others have made many valid points. However, a Ct of 40 is completely reasonable for a positive sample. It depends on lots of factors that have been mentioned. There is no magic number cutoff. Is it reproducible? What does a standard curve look like? I would also recommend a melt curve (which can be done with the right kind of probes). If you get a melt at the right temp, then it is real. If you can't do a melt, and see it with no template controls, it is probably primer dimer.

For HPV specifically, a Ct of 40 is right at the clinically validated cutoff for the Roche HPV Test. Cts higher than 40 can still be analytically positive, but are not reported because they are not clinically relevant. One difference is that this test is for HPV DNA, but shouldn't make a big difference. For RNA viruses at my institution, it is not uncommon to see Cts >40 for positive samples. A recent assay I developed (one-step RT-PCR) has a Ct of ~40 for the lowest member of the standard curve (RNA transcript), which equates to about 100 copies/reaction. The LOD (6 of 6 replicates) is approximately 12 copies/reaction, which is an average Ct of ~41.5. The variability at this level is high, as you'd expect.

I am using TaqMan for my real time PCR. So there won't be any primer dimer giving the reading. Whatever things Kaushal has mentioned (in 1st answer) were already taken into consideration for the assay.

I think I should load the products of  varied Cts on a gel to compare and see if they give a band of the expected size at high Ct and repeat the real time assay for such samples to see if they are consistently showing the same result.

Thanks Gareth for providing references for calculation. Thanks Nicholas and Michael for your valuable inputs!    

I know my answer will not be helpful in your Taqman format. This is a suggestion to look over HIV test by COBAS and Abbott which were approved by FDA. They have used internal control to prevent false positive. For the same issue, the following group (including me) developed their assay not only detection but also quantification. As we use internal standards mixture (known copy numbers) which have a target and ACTB, reference. We do not need to worry about loading in PCR. Also, as they put this mixture with unknown sample, it was possible two time of PCR (pre-amplification) because the ratio between internal standard and unknown samples does not change. We saw Cq as 26 about 6 copies of viral load after two time of PCR. You can low Cq controlling dilution of 1st PCR products to discriminate background signal. In our case, ACTB expression was too high (about 10 Cq). That's reason we diluted 1st amp sample as 1/1000.

I know developing a new assay will be challenge. However, it depends how often you need this assay, how much it is a long term project, and how many genes you may need to change (it's hard to put additional assays after making one mixture but you can compare any data if you use a same mixture over the year). FYI, we used BHQplus probes and ABI7500. Using same method we published for FFPE sample. Also, an group in our site developed an assay for sorted B cell. 

"Accurate Detection and Quantification of the Fish Viral Hemorrhagic Septicemia virus (VHSv) with a Two-Color Fluorometric Real-Time PCR Assay"

"A Multiplex Two-Color Real-Time PCR Method for Quality-Controlled Molecular Diagnostic Testing of FFPE Samples"

That's more easily said than done, though: with qPCR (especially with low copy numbers) you sometimes have "no Cq value", and sometimes you get "a Cq value" even with a negative control.

So if you run five negative controls and get 45, nothing, nothing, nothing and 42, then run five test samples and get 37, 42, nothing, 40 and nothing, deciding how best to generate mean Cq values for statistics is...challenging.

Generating a standard curve is much more informative, as it tells you the cut-off between "something" and "nothing/noise", and a standard curve diluted into a negative control sample will show you this distinction under exactly the same conditions as your test samples.

If, under these conditions, your tenfold dilution series goes something like 25.1, 28.4, 31.7, 35, 45, no signal, 40, 35, no signal,  then you've got pretty good evidence that a Cq of 35 represents the lowest detectable signal, and that signals in the 40s are simply noise. If your curve instead continues to proceed linearly down to the 40s, signals in that range are likely to be valid.

OK, nice idea, but given that the Cq (Ct) metric is an exponential rather than a arithmetic scale, is SD an appropriate statistic to use? I suppose you could use the delta-Cq calculation to re-linearise, but then how many negative-control samples would you need to run on each plate to have reasonable precision around the mean? Do you use this method and if so, is it validated for clinical use?

And yes, you're completely right - the standard curve does not indicate positivity of an individual sample - the Intercept+1 is however an internationally accepted threshold for determining, given the limits of the technology, whether it is reasonable to consider that sample to be positive or undetectable. It is a bit of a fudge, but given the context of the clinical diagnostic work that this derives from (molecular monitoring in haematological malignancies) it is appropriate since it takes into account the inter-run changes in sensitivity, specificity and quantitative range of each run. Whether this is appropriate for the HPV work under discussion is another matter...