The raw data looks the same - albeit higher. When the 'negative' curve is really bad the initial background reading is also really high - this then drops to a low number when the curve levels out (all values are positive). When the baseline is set, the plot becomes negative, but doesn't change shape.
I had a theory that there may be something damaging the probes and releasing free fluorophore prior to the PCR. This would give a high background, and with each round of PCR the 'free' dye gets bleached, decreasing the signal. When all the free dye is quenched the levels flatten out (as in the negative amps on the graph shown). Any dye on intact probes is protected from photobleaching by the quencher, so the positive signals still work. Or that might be nonsense (if it was true, wouldn't the plateau actually decrease as well, since the free dye is bleached, or could it be that this would happen if you continued the PCR for long enough after the probe is exhausted?). Either way, I've ordered a new synthesis of the worst probe to see if it's any better.
This problem is worst with FAM. Yakima Yellow and Rox are affected, but only slightly. Cy5 works properly.
toxlab - I am using real time in genotyping rather than expression analysis. At higher gDNA concentrations I start to get inhibition. I will go back and check the theoretical primer/probe interactions to see if there is any pattern connecting the worst to the negative curves.
Thanks for all your input so far