Help decide on appropriate test for experiment design

Hello - new to the forums here!

I work in cell biology and a colleague and I are designing an experiment to determine the appropriate concentration of a toxin used as a selection agent in cell line development. The toxin is methotrexate (MTX) and the cells used are Chinese Hamster Ovary (CHO) cells that are typically transfected with a recombinant plasmid containing a gene for MTX resistance.

We want to determine the optimum concentration for MTX in CHO cells that will allow selection against unsuccessfully transfected cells but allow for optimum growth of successfully transfected cells.

The design is to expose CHO cells to varying concentrations of MTX ranging from 0 to 100nM MTX (independent variable) and measuring growth of colonies in our media over time (which may be a single measuring point of perhaps 3 days, but could also be measured in multiple days of, perhaps, 3, 5, 7, 10 days post seeding into media).

We want to show that that whatever concentration appears to be ideal is also significantly different from other concentrations, that the varying concentrations are significantly different from each other, etc.

Would a one-way ANOVA show this?

If multiple days are used though, would this require repeated measures one-way ANOVA, or does this fall into one-way MANOVA?

I'll likely be using Graphpad Prism for the analysis.

Thanks in advance!
Actually now that I think about more, there are two independent variables: time and MTX concentration. The dependent variable is number of colonies.

What I'm expecting to see is data that shows that over time, different MTX concentrations will result in different number of colonies. I would think that any statistics used to describe the effects of drug dosage over time would work?
Hi, would it make sense to do a multiple linear regression? assuming the number of colonies eventually decline at certain levels of both of the independent variables, you could use Hessian matrix to determine the values of time and concentration at which this happens.
That does make sense and I'm leaning towards that direction. The number of colonies (dependent variable) is a number that is not known to absolute - there is some error (it's number of cells/ml - and will range to millions of cells with a standard deviation of perhaps 0.1 million cells/ml). I believe I can assume the dependent variable is normally distributed, but the cell counter doesn't give me access to raw data, only the average that it computes and the standard deviation.


TS Contributor
I wonder if you need the time in this experiment? Obviously you will have more colonies at time t+Dt then at t, so an effect will be given. Do you expect a time-toxin interaction? E.g. that colonies treated with different amounts of the toxin will somehow remember their dosis and behave differently over longer time stretches (unless the toxin causes mutations I cannot see this happening, but i am no biologist). It could simplify your experiment to just measure at a given time for different doses,

That's a good point, Rogojel. The effect of the toxin is not affected by time, I just measure cell growth (increase in number of cells) roughly every 24 hours over several days after a single administration of the toxin at time=0. It would definitely simplify things to limit this to one independent variable (toxin concentration) and one dependent (number of cells) after a pre-determined number of days.