Multiple melt curves for primer in a sample that does not express gene - (Feb/25/2014 )

I have validated a set of primers by dilution curve analysis using cDNA which expresses the gene. 

I recently used these primers to look at the over-expression of this gene in a cell line that does not endogenously express this gene (FPKM = 0 in RNAseq).  It worked well for all samples that were induced with my over-expression vector, but as a negative control, I included the non-induced cells as well.  Here I get a CT of 32 but the melt curves look very bizarre with multiple peaks.  I was expecting to obtain a CT of high 30s/40/undetermined.  Also, when using two different amounts of cDNA for this sample (10 fold difference), I do not see the cycle number change in this sample where as it does for the induced samples.  Wondering if I am nonspecifically amplifying something in this sample? Do you think this is real amplification?  Is it a bad primer? Thanks

Could be nonspecific amplification, but since you said that Ct didn't change with different DNA concentration, I'd go for primer dimers ?

Ya, I just rechecked the data.  There is a 1 cycle difference (from CT 32 to CT 31) when using 10 times less cDNA in the uninfected line.  

Where as in the lines expressing this gene, there is a 4 cycle difference when using 10 times less cDNA.  

I'd run the samples on an agarose gel after amplification in order to check the size of the amplicon and check if they could be primer dimers (approx. 50 bp size).

So it looks like there is some non-specific amplification only in the control, non-expressing samples.  Why would I not see this in all my samples?  It even seems variable across technical replicates....is this possible?  

The CT also increases when the primer concentration is halved, so I think there is also primer dimer going on as well. 

In that case, could you just give it a try with different primers ?