# Dual Luciferase assay data analysis issues - (Aug/28/2012 )

Hi All,

I want to investigate the functional properties of a promoter fragment of interest. I cloned the promoter into a pGL3 plasmid and co transfesfected this promoter plasmid with a renilla vector in HEk293T cells to assess transfection efficiency. After following the protocol for the Dual-Glo luciferase assay system and assaying luminescence, I normalized the data to empty vector control as recommended by Promega.

Just looking at the data, it seems that there is a significant difference in my experimental promoter fragment compared to empty vector controls. The luciferase normalized ratios are on the order of 1.7 compared to empty vector at 0.7. and an even greater difference when calculating relative response ratios. However doing a 2-tailed student's T test shows that it is just under statistical significance. I recently read a paper (Zhuang, H et al 2009) in Nature Protocols that shows a slightly different method of analysis. When I do this analysis, I get statistical significance.

Please help!! I'm not sure which one is best or better represents my data since I get two different results. Does anyone know a good way to analyze/normalize the luciferase values compared to emtpy controls?

For the correct answer to this question you need to ask yourself two things: ~~1)~~~~do I feel lucky?, 2) well, do ya, Punk?~~ 1) should the data that you are generating form a bell-shaped (normal distribution) curve, if you took enough samples? 2) Do you have more than 30 samples?

If you answered no to either of these, then a Student's t-test is not the correct test to be performing, as it is only for normally distributed data for large sample numbers. The equivalent, slightly less powerful, but more statistically valid tests are known as non-parametric tests, they are for low sample numbers and/or non-normally distributed data. The tests you probably want to perform are either the Mann-Whitney test (for unpaired groups) or the Wilcoxon test (paired groups). Kruskal-Wallis is the equivalent to ANOVA if you need it.

Totally agree with Bob's comment about small samples for student's t-test. To add to this: if you see hundreds of articles on pubmed with student's t-test with 3 samples, they are all wrong! Take the above advice and perform the right statistical tests

Andreea

Thanks so much for your quick replies.

Bob, you are correct, I have a small sample size and cannot assume a normal distribution curve. I will try the calculations with different tests and see how the data looks.You are right as well Andreea; I have seen numerous papers on pubmed using a student's t-test with few samples, so that is where some of the confusion occured.

I actually subcloned the promoter plasmid into eighths so I essentially am testing the activity of 7 smaller promoter fragments, the full promoter fragment, an empty vector control, and mock transfected cells in triplicate. So essentially, my sample size is 10. One thing that I have notcied is that, most papers don't give statistical significance and rather just mention a fold change or relative luciferase activity. So would these tests hold true in this scenario as well or should I just stick with the "norm" and comment on a fold change?

Fold changes are fine too for this sort of data, so long as the fold change is consistent, even if the actual values aren't (for whatever reason). For example if in one set of data the control value is 100 and the treatment value is 200, and in another test, the control is 50 and the treatment is 100, the fold change is still the same. Obviously, you don't want the actual values to be too different between experiments, as this would indicate that there is something wrong with the procedure.

If you can generate a standard deviation on your data and plot it, basically if the standard deviations overlap, then there is no statistical significance.

HawkeyeGrad on Wed Aug 29 15:56:13 2012 said:

I actually subcloned the promoter plasmid into eighths so I essentially am testing the activity of 7 smaller promoter fragments, the full promoter fragment, an empty vector control, and mock transfected cells in triplicate. So essentially, my sample size is 10. One thing that I have notcied is that, most papers don't give statistical significance and rather just mention a fold change or relative luciferase activity. So would these tests hold true in this scenario as well or should I just stick with the "norm" and comment on a fold change?

Actually, you have 10 separate groups of samples, each of which I would expect to have a normal distribution of the measured luciferase values, due to the nature of the way these sorts of experiments are measured. You need to be using an ANOVA (Kruskal-Wallis) style approach to analyse your data, or you run the risk of hitting a type one statistical error. You can do post-hoc testing (e.g. Tukey test) to find out which group(s) are statistically different amongst the groups.