Replicate flasks are not necessary for in vitro chromosome-aberration assays in CHO cells

Some recommended protocols for in vitro chromosome-aberration assays call for two flasks per dose group. Use of replicate flasks allows for possible variation in percent aberrant cells (ABR) between flasks. We studied the magnitude of variation between replicate flasks of Chinese hamster ovary (CHO) cells using data from 211 assays from three laboratories, in order to assess the effect on assay sensitivity. Based on all 403 pairs of replicate "control" flasks, there was almost no excess variability between flasks. The standard deviation (SD) was only 4% larger than the value expected purely from sampling cells (P > 0.05). Data from all 366 pairs of replicate "treated" flasks showed that between-flask variation increased with the average percent aberrant cells (P < 0.001). The SD for 60 pairs of flasks with 3.0-7.5% ABR cells was 32% larger than the expected value. However, computer simulations based on these data showed use of replicate flasks has little effect on assay false-positive or true-positive rates. All assays with replicate treated flasks and at least three dose groups including control were re-analyzed as "single-flask" experiments. A "single-flask" experiment was defined by taking both control flasks but only one treated flask per dose. For each assay, all possible single-flask experiments were re-analyzed and the percent with positive results recorded. For most assays, conclusions were the same regardless of which treated flasks were selected, in spite of the fact that these single-flask experiments had only half as many cells scored per active dose group. For a very few assays with marginal results, the conclusion could change depending on which set of flasks was chosen, but these were such borderline results that a repeat assay was required in any case. Repeating the assay is a better way to resolve marginal results than examining replicate flasks. From our re-examination of the experimental data and from the computer simulation, we conclude that, while flask-to-flask variability exists, it has no practical effect on the test outcome, so that use of replicate flasks is not necessary for this assay.