I am doing time course study for gene expression in cell culture. I am doing 0, 1, 2, 4, 8, 16, 24 hours. Each time point I have treated and untreated (control). I should normalised my individual treated time point samples to control at that time point also and relative to the 0hr treatment sample?
Hi Shin,
I would take each time point individually (1,2,4,8hrs etc) lysing the cells and freezing. The day after your experiment extract all RNA at the same time and normalise to a housekeeper gene (or two if you can - normalise to the geometric mean) such as BActin, Gapdh, Bmn etc. Then for each timepoint, normalise to the untreated control so that you have a value for fold-up/down regulation and average over your technical replicates (at least 3 for each time point).
You can then plot this as a scatter plot to see how regulation of your gene of interest changes over time. Normalising to time0 at this point is useful for when you run biological replicates (i.e. when you repeat the experiment).
John
If you are interested in the difference in the time courses between the treated and the controls, then an appropriate model would be (using the model formula notation)
dCt ~ TIME x TREATMENT
where dCt are the delta-Ct values from qPCR, TIME is the variable coding the time points (if you have a good and relatively simple functional model, like for instance a linear or logistic change, then you can use TIME as a metric variable with appropriate transformations; otherwise I would suggest to code TIME as a categorical variable), and TREATMENT is the categorical variable specifying the treatmen ("control" or "treated"). The variables TIME and TREATMENT are crossed, so the model estimates main effects and the TIME:TREATMENT interaction, the latter being the interesting thing, telling you how much different the time-course is between the two treatments.
If your experiment is designed in a way that the measures within a time-course are not independent* then a mixed model accounting for the dependency-structure would be appropriate, like
dCt ~ TIME x TREATMENT + (1 | ID)
where ID is the factor specifying the ID of correlated measurements, and (1 | ID) is the random factor allowing the model to use (randomly) varying intercepts for each series sharing the same ID. If it is reasonable to assume that the time-course might be correlated with the ID, the random factor could be specified as (TIME | ID).
* You might get a correlation of measurements with the "day" of the experiment when each time course is measured individually (on individual "days". In this case ID would be the "day". You can substitute "day" by "stock preparation" or similar things. You may get a correlation to the "culture" or "well" if the measurements during the time course are taken (repeatedly) from the same well. In this case ID would be the "well". Depending on your design you can even have several such variables, what would lead to several (possibly also nested and/or crossed) random factors in your model.
I strongly suggest reading
Gelman & Hill: Data Analysis using Regression and Multilevel/Hierarchical Models. Cambridge, 2009.
Hi Shin,
I would take each time point individually (1,2,4,8hrs etc) lysing the cells and freezing. The day after your experiment extract all RNA at the same time and normalise to a housekeeper gene (or two if you can - normalise to the geometric mean) such as BActin, Gapdh, Bmn etc. Then for each timepoint, normalise to the untreated control so that you have a value for fold-up/down regulation and average over your technical replicates (at least 3 for each time point).
You can then plot this as a scatter plot to see how regulation of your gene of interest changes over time. Normalising to time0 at this point is useful for when you run biological replicates (i.e. when you repeat the experiment).
John
Obviously, this depends on your particular research question. Ideally, you would need non-treated control for each time point. Usually this is possible without multiplying your samples simply by using a countdown treatment scheme; i.e., you add the inducer at -24 hours, -16 hours, ... and so on and harvest all samples simultaneously (including the untreated control) at time "zero"
Dear Shin,
I think that each sample should be normalized against an invariant endogenous control in the assay (a good choice is 18S rRNA , b-actin and GAPDH are known to be less than perfect since their expression can be modified under some conditions), and this normalization should be prefferably done with samples from the same timepoint. This is done so as to keep all experimental variables unchanged except the one you are studying, that is the effect of the treatment in the course of time on gene expression. Given that normalization is carried out with an invariant endogenous control, there should be no difference in the expression irrespective of time point.
The same logic follows on other comparisons: you should compare the expression of your gene of interest with controls from the same timepoint. Still, i would not expect any difference in the control samples due to time alone - if there is a difference, then it could be difficult to pin down the change to time,treatment and their interplay (you would be studying concomitantly two variables that both change)
Dear All,
Thanks for your advice and suggestion. Another question I would like to ask. If I normalise my treated sample to control at each time point, does the SD change? If yes, may I know how to calculate it? The SD will change as well if the treated samples at each time points are relative 0 hour treated sample, right?
I think you should normalize each value first by the control gene at each given time point, calculating the SD as you would with the raw data. Then you should express it in relation to the time 0, which should not change the SD, just do the same calculation with the data. It is advisable to do three biological replicates per time point/treatment.
- Badges
- Science topic