The use of animal tests in risk assessment for immunotoxicology

The design and content of a screening battery using a 'tier' approach for detecting potential immunotoxic compounds in mice has been described (M. I. Luster et al., Fundamental and Applied Toxicology 1988, 10, 2-19). The database generated from these studies, which consists of over 50 selected compounds, has been analysed in an attempt to improve future testing strategies and provide information to aid in developing future quantitative risk assessment for immunotoxicity. In a recent study it was shown that as few as two or three immune parameters were needed to predict immunotoxicants in mice (M. I. Luster et al., Fundamental and Applied Toxicology 1992, 18, 200-210). The analyses described here focus on the use of this database to develop statistical models that examine the qualitative and quantitative relationship(s) between the immune function and host resistance tests. The conclusions derived from these analyses are as follows: (1) A good correlation exists between changes in the immune tests and altered host resistance, in that there were no instances where host resistance was altered without affecting one or more immune test(s). However, in some instances immune changes occurred without corresponding changes in host resistance. (2) No single immune test could be identified that was fully predictive for altered host resistance, although most assays were relatively good indicators (i.e. 70%). Several others, such as proliferative response to lipopolysaccharide and leucocyte counts, were found to be relatively poor indicators for host resistance changes. (3) The ability to resist infectious agent challenge is dependent on the degrees of immunosuppression and the quantity of infectious agent administered. (4) Logistic and standard regression modelling using one extensive chemical data set from the immunosuppressive agent, cyclophosphamide, indicated that most immune function-host resistance relationships followed linear rather than linear-quadratic (threshold-like) models. For most of the relationships this could not be confirmed using a large chemical data set and, thus, a more mechanistically based approach for modelling will need to be developed.