A statistical analysis of the third UKEMS collaborative trial

In the third UKEMS collaborative trial, nine laboratories optimized protocols for a variety of mammalian cell mutagenicity assays and used these protocols to test three reference mutagens. Data from the trial have been deposited with Mutagenesis under the scheme of the journal for a database of results of extensive mutagenicity testing programmes. A preliminary examination of the data using procedures advocated by the UKEMS guidelines on statistical evaluation of mutagenicity test data confirms that the assays are capable of giving clear, convincing data, in a form amenable to statistical analysis. The utility of the methods of analysis proposed by both the plate and fluctuation test statistical working groups was confirmed. In almost all cases variation between plates or trays from the same replicate treatment was little greater than Poisson or binomial, whereas variation between replicate treatments was substantially larger, confirming the recommendation of both groups for true independent replicates of each treatment. In six out of nine laboratories part of this variation between replicates was consistent for both viability and mutation, which might cause the guide-lines procedure for fluctuation test data to underestimate significance. In about half the cases examined, variation between experiments was significantly greater than variation between replicates within an experiment, which may create problems of interpretation. Supplementary tests, using the statistical package GLIM, suggested that although factors such as choice of expression time, level of S9, zero dose cloning efficiency or spontaneous mutation frequency might have an effect, they would not have been likely to interfere with detection of a mutagenic effect in the present data set.