Comparative sensitivity of survival-adjusted chi-square and normal statistics for the mutagenesis fluctuation assay.

Three statistics for analysis of microtitre plate mutagenesis fluctuation tests were studied by simulation, and in enzyme-activated assays of dimethylnitrosamine and diethylnitramine. A survival-adjusted chi 2 statistic ('Gsq') was compared with Katz's normally distributed statistic ('Phi'), and with the survival-independent statistic ('Zsq') of Gilbert. When toxicity was either very low or high, the Phi statistic either could not be evaluated over the whole range of possible background mutant frequencies, or sometimes it indicated unusually high levels of statistical significance, even when the other tests were negative. The survival-adjusted Gsq closely followed the Zsq statistic throughout the experimentally useful range of toxicities and mutant background values, with some improvement in sensitivity. Within the range 80 +/- 10% survival approximately, Katz's statistic 'Phi' was the most sensitive. The choice of statistical test could affect the estimate of the minimal effective mutagenic concentration by a factor of 10-100. For screening unknowns, both types of test (Phi and Gsq (or Zsq] may help in detecting suspect pro-mutagens and in designing a confirmatory assay. Bacterial population statistics are needed to assess the value of statistically positive results.