induced mutation by the test compound. In the CHO Cell/BrdU-VL assay, each 
cell screened will either be auxotrophic for one or more of the nine nutrilites 
omitted for FI 2D (operational definition of mutant), or it will not be 
(operational definition of wild-type). The actual criteria employed to classify 
mammalian cell variants as true mutants are somewhat complex, and have been 
reviewed recently (16, 40, 46). For the purpose of this paper, the operational 
definitions given above shall be used. 
In the standard procedure, a sample of 10 1 cells from each test or control 
population is distributed among fifty 60 mm dishes, and subjected to the 
selective process. Surviving clones are then sampled and tested for the presence 
of auxotrophs. The total number of auxotrophs expected per 10^ viable cells is 
then estimated from the data by equation [2], where (y) is the estimated 
number of auxotrophs, (x) is the number of auxotrophs observed, (n) is the 
number of replica experiments, (A) is the total number of cells surviving 
selection, (B) is the total number of colonies picked and tested, (C) is the 
initial number of 60 mm dishes, (D) is the final number of dishes (some may be 
lost to contamination during the course of the experiment), and (E) is the 
absolute plating efficiency (defined as the ratio of macroscopic colonies 
produced to cells inoculated) as measured in low density control dishes. 
Because mutants are randomly distributed among wild-types in mixed 
populations, the probability that any given survivor will be a mutant should be 
constant over all survivors. Moreover, as only a small number of mutants is 
generally found in any given population, the distribution of mutants in such 
populations should be Poisson. Accordingly, mutagenesis data from sets of 
replica experiments were tested for goodness of fit to a Poisson model, and 
found to be consistent with this type of distribution (33). 
For two independent Poisson variables (X, Y), a new statistic (V) has been 
proposed by Best (9) for testing the difference between two Poisson 
expectations (e.g., the estimated mean number of mutants in experimental (X) 
versus control (Y) populations). This statistic, given by equation [3], is similar 
in performance to the more familiar square root of the Poisson Index of 
Dispersion (20), except in the tails of the distribution where (V) is superior. 
Although (V) is a function of the Poisson variables (X, Y), (V) itself shows an 
approximately normal distribution. This statistic may be particularly applicable 
to mutagenesis data where the difference in variance observed between 
experimental and control populations is large. This is the situation at the 
present time with the CHO Cell/BrdU-VL system where mutants are rarely 
observed in control populations. All mutagenesis data considered below were 
scaled via equation [2] and compared to an historical control (Y) in 
accordance with equation [3] and appropriate confidence limits. The model 
given by equation [3] and appropriate confidence limits. The model given by 
86 
