CURRENT STATUS OF MODELS FOR TESTING ANTIBIOTIC RESIDUES

Models for testing the effect of antibiotic residues on the human gut flora should be (i) feasible (quality of data, statistics, price, dura tion), (ii) sensitive (baseline resistance, background variability, re sponse to low doses), and (iii) relevant (composition and ecology of t he flora, predictive value). An evaluation of existing models led to t he following conclusions: 1 - Flora. Relevant models must include anae robic flora that antagonize colonization by minor bacterial population s. In vitro tests on pure strains, batch or continuous flow cultures, do not take this barrier effect into account. 2 - Drug treatment. Trea tment should be long enough to allow the detection of indirect selecti on due to a weakening of the barrier effect (i.e., greater than 1 5 da ys). For dose response analysis, trials should include at least 3 grou ps: a control group, an experimental group given the antibiotic at res idue levels, and a positive control group given a high dose of the sam e drug. 3 - Target strain(s). The target bacteria should be identified , resident in the gut, accurately countable, drug susceptible but with drug resistant variants, and related to a pathogen. Some gut aerobes meet these requirements: sensitive to the ecosystem balance, easy to c ount accurately, opportunistic pathogens, and R-plasmid carriers. Tota l anaerobes are not suitable. An anaerobe species may serve as a targe t strain if it is particularly susceptible to, or selected by, the tes t antibiotic. Thus, each in vivo protocol should be based on in vitro data. 4 - Measure of resistance. The determination of the minimum inhi bitory concentration (MIC) of selected clones from the model flora is time consuming, does not detect subdominant resistance (less than 1%); and a MIC shift is difficult to test statistically. In contrast, dire ct counting of bacteria on drug supplemented media allows for the rapi d measure of minor resistant populations. 5 - Statistics. Most publish ed designs do not include a power calculation, nor yield enough data p oints for accurate statistical analysis. Sufficient statistical power is critical for studies using conventional humans and animals, because the data are highly variable and not normally distributed. 6 - Human trials. The lowest level of antibiotic tested in human volunteers (2 m g oxytetracycline/day for 7 days in 6 subjects) increased the proporti on of resistant fecal enterobacteria (p=0.05). However, the large day- to-day and inter individual variation in the human flora limit the sig nificance of this change. Confirmation of this finding would require a large number of volunteers. Moreover, our data showed that, when comp ared to bacterial contamination from food containing resistant enterob acteria, the relative contribution of antibiotic residues for selectin g resistance is low. 7 - Human flora associated (HFA) mice. HFA mice a re in vivo models appropriate for testing the effect of any drug on th e human gut flora. HFA mice retain a high number of anaerobes from the human gut inoculum which make a strong barrier effect. Interindividua l and day-to-day variations of bacterial populations are lower in HFA mice than in humans. The model is, hence, highly sensitive to antibiot ics. The predictive value of data from this model to humans is confirm ed by the similar effects of therapeutic drugs in HFA mice and in volu nteers, and by the correlation between the selective effect of antimic robial growth promoters in piglets and in gnotobiotic mice.