I'm doing some qRTPCR data analysis on tumour samples and using macthed "normal" tissue as a control. Each normal and tumour was run in triplicate (technical). Now we want to use all the normal as calibrators or reference samples to see how much our gene of interest is up or downregulated versus the normal (pooled). How do you calculate the standard deviation of each normal DCt +/- SD? I can't average the SD, can't I?
Assuming I'm not misreading this, you're using the results of normal tissue qPCRs to calibrate the results of tumour tissue qPCRs?
That's not necessary a valid method, as it doesn't calibrate for RNA extraction efficiency, RNA quality, or cDNA synthesis efficiency, all of which can contribute to considerable variation.
I would first establish a panel of reference genes (three is ideal, two is usually sufficient, one is risky) suitable for normalising expression between normal and tumour tissue. Programs such as geNorm and Normfinder are very useful for this.
You may need to test a number of genes before you identify stable candidates (and don't just use the 'traditional' genes: beta actin and GAPDH frequently perform spectacularly badly for normalising at the RNA level).
Now this might be terribly patronising, so please forgive me if I'm stating the obvious, but here goes:
(forwarning: I dislike delta-deltaCT methods as they don't give you a very good handle on the actual numbers involved)
qPCR your tumour/normal samples with your reference genes and your test genes, in triplicate. Determine the average of each set of triplicates (discarding obvious outliers), then convert those CT values to relative quantities like so:
Efficiency of the reaction (equal to 1 + efficiency%*0.01, so 2.0 equals 100% efficiency) to the power of the lowest CT value observed for that gene, minus the sample CT value for that gene
So if a given gene gives CT values of 22, 24, 24 and 23, assuming efficiency 2.0, you would get
2^(22-22)
2^(22-24)
2^(22-24)
2^(22-23)
Or 1, 0.25, 0.25 and 0.5.
From those values, determine a normalisation factor from your reference genes by calculating the geometric average (so rather than (A+B)/2, it's the square root of A*B), and use THAT to normalise your expression levels from your test genes. You should now have expression levels of your genes of interest, expressed as arbitrary units, but which are wholly comparable between tumour and normal samples.
From this point, how you choose to display your data is more of a personal choice. If you want to display it as fold increases, then calculate the fold increase/decrease for each patient and take the average and SD of those numbers.
Overall, doing it this way allows you to easily identify possible sources of error as you go along, i.e. if you generate your normalisation factor and find you have
0.34
0.67
1.00
0.03
Thanks so, so much for your answer. Unfortunately, it doesn't apply to our case. We have a set of tumours with usually methylated gene (promoter) whose expression is low. We have a set of matched normal samples with high expression level. So for calculations we want to know what is the expression of each tumour when compared with out normal dataset, rather than one by one (tumour vs match).
So...you have a large number of normal samples which you've used to generate a single value which you consider 'normal', and you want to compare your individual tumours to this one 'normal' value? Because you imply that one by one matching isn't what you want to do.
How are you calibrating for RNA extraction efficiency, RNA integrity and cDNA synthesis efficiency?
In other words, how are you normalising the data you use to generate your 'normal' value?
- Badges
- Science method