Statistical evaluation of inter- and intra-laboratory variations of the Ames test, as related to the genetic stability of Salmonella tester strains

A statistical analysis was performed on the data resulting from an international collaborative study of the Ames test according to a standardized experimental protocol, which involved the comparative testing of 4NQO (4 doses), in 3 separate experiments for each of the 38 participating laboratories, by using a common reference (R) culture and in-house laboratory (L) cultures of 5 strains of S. typhimurium. Despite some toxicity phenomena recorded at the highest dose of 4NQO, the majority of the dose-response curves in individual laboratories were linear on a bi-log scale and their mean values fitted a linear regression framework. Scattering of data around mean values of laboratories was Gaussian-like even at the highest dose of 4NQO, toxic effects being expressed as a dose-related increase of variance. A weighted least-square analysis could therefore take into account toxic effects without resorting to a sophisticated non-linear model incompatible with log transformation. Various analytical approaches--e.g. the weighted estimates of linear regression parameters, a multifactor (laboratory, experiment, dose, culture of each strain) analysis of variance with all the possible interactions, the assessment of correlations in individual laboratories and of coefficients of variation for induced and spontaneous mutability--could detect some statistically significant differences between L and R cultures. However, at a critical evaluation on an individual basis, only few of these differences, without any peculiar involvement of given strains, were convincing in view of the existence of real phenomena of genetic drift. Therefore, on the whole, the genetic drift of Salmonella tester strains appears to lend a negligible contribution to the considerable inter- and intra-laboratory variability detected in this study. With a background variability between replications averaging 26%, a dose-related variability was evident both between experiments (28-54%) and between laboratories (44-127%).