That is an interesting question. Unfortunately, it has not a direct and concise answer. Several factors are involved, i.e. (1) type of variable used, (2) variation of the variable determinations, and (3) power of the statistical model to be use. In general terms, a cardinal variable (numerical representation of the process to be studied) needs a rather small sample; nominal variable (categorical representation of the process to be studied), needs a large sampling. There are mathematical statistical models to calculate the minimum number of animals to a determined experiment. The problem is that it is necessary to postulate several assumptions that usually are not known. In our lab we use the method of repeating sampling of the variable in order to estimate the minimum value of variance. Should you be interested, I can give additional details later. Perhaps, in your case you should need about 8-10 animals if you are testing just a single dose.