You are here:
SAR STUDY OF NASAL TOXICITY: LESSONS FOR MODELING SMALL TOXICITY DATASETS
Citation:
Richard, A. AND M Greenberg. SAR STUDY OF NASAL TOXICITY: LESSONS FOR MODELING SMALL TOXICITY DATASETS. Presented at Quantitative Structure Activity Relationships Gordon Research Conference, Tilton, New Hampshire, July 25-30, 2000.
Description:
Most toxicity data, particularly from whole animal bioassays, are generated without the needs or capabilities of structure-activity relationship (SAR) modeling in mind. Some toxicity endpoints have been of sufficient regulatory concern to warrant large scale testing efforts (e.g., mutagenicity and rodent carcinogenicity). More common, however, are the small, structurally diverse data sets relative to a wide range of toxicity endpoints, e.g. that may be generated in the course of whole animal bioassays. These toxicity endpoints pose particular challenges to SAR modelers, not only due to limited data for model development, but also due to limited capability for model testing and validation. In such cases, we must determine if a data set is suitable for modeling and what steps can be taken in model development and evaluation to raise the level of confidence in such models.
This study describes application of SAR/QSAR modeling to a small, structurally diverse set of Clean Air Act (CAA) gaseous pollutants, for which concentration-response data are available for rat nasal lesions induced by chronic inhalation exposure. Such lesions serve as sentinels of potentially more serious systemic effects or noncarcinogenic effects in the lower respiratory tract. In addition, histopathological evidence of nasal damage, available for a limited number of CAA chemicals, constitutes critical effects in the derivation of EPA's Inhalation Reference
Concentrations (RfCs). No prior SAR models were available for nasal toxicity and little is known of the mechanism(s) for toxicity. The original modeled data set consisted of 11 substituted aromatic compounds; 6 were nasal toxicants, with LOAEL values spanning 5 log units, and 5 were inactive. The goal was to screen a small number of potentially relevant descriptors and develop models for: ( I) classifying an untested chemical as capable of inducing nasal lesions; and (2) estimating the relative potency of such chemicals. A single parameter, molecular weight accounted for 98% of the variance in potencies for the 6 actives. In addition, a discriminant analysis based on two electronic parameters (Ehomo and dipole) correctly classified, with high probability (>90%), 10/ 11 of the chemicals according to activity ( +1- ). These models predicted both the correct activity classification ( +)and approximate potency of toluene diisocyonate in a limited validation exercise. The implications are that electronic properties may dictate the potential of a chemical to be a nasal toxin, whereas size-related factors may be important for modulating potency among the actives. The second phase of this work is extending modeling efforts to 26 aliphatics for which nasal toxicity data is available.
This abstract does not necessarily reflect EPA policy.