You should have at least six samples per group to check the distribution of your data (Shapiro-Wilk's test). After you find the distribution, you can select the correct statistical methods for your case.

A lot depends on the type of data you are collecting. Is it go no-go? (the rat died or did not dive) that will require a larger amount of observations than if it is categorical. Continuous data (what is the weight gain of the mouse) can get by small sets of data, however it depends on the standard deviation of the resulting data.

Statistically, you will need at least 5 in each group to be able to generate any useful values. That is a bare minimum, and may not be sufficient for future acceptance. It also decreases robustness as any validly rejected data from a set would necessarily make the entire set invalid.

RESEARCH is not just what we do individually as scientists! It STARTS with understanding what has already been done. Previous experiments using rat populations for any reason can be read to find out what the "normal" range of sample sizes is, and how they were justified and used. You can also use them as references in your own work as to why you selected a certain number.

Previously, I used 5 rats per group, but recently I have increased to 6 rats per group, this is due to standard error in statistics and also the possibility of outliers in data. It all depends but I'll always advise at least 5 rats per group