Quantitative structure-activity relationships for predicting skin and respiratory sensitization.

Quantitative structure-activity relationships (QSARs) for predicting skin and respiratory sensitization are reviewed. Overall, progress has been hampered by the sparseness of good quality experimental data, a fact that makes it difficult, at this time, to recommend one or two QSARs for predicting skin and respiratory sensitization. Creation of appropriate data sets for uninvestigated classes of chemicals by experimentation should facilitate the development of more robust QSARs for predicting skin and respiratory sensitization. Such QSARs will be valuable in the evaluation of identifiable toxic hazards where dose responses are relevant, as is the case for skin and respiratory sensitization.

Comparison of models for the estimation of biological partition coefficients.

Several models have been published for calculating blood-air, tissue-air, or tissue-blood partition coefficients of volatile organic chemicals in human or rat tissues, from functions of their octanol-water partition coefficients or solubilities in vegetable oil and water. In this work, the relative accuracy, strengths, and limitations of the various models are examined. Comparison of predicted human tissue-air and tissue-blood partition coefficients with experimental values has been made for 12 chemicals, covering a wide range of lipophilicity (acetone, isopropanol, diethylether, methylene dichloride, benzene, toluene, trichloroethylene, trichloroethane, n-pentane, cyclohexane, n-hexane, and n-heptane). Seven published models for human tissue-air and 10 models for tissue-blood partition coefficients have been compared. Fewer models are available for predicting rat tissue-air and rat tissue-blood partition coefficients, but a similar comparison has been made. The ratio of predicted to experimental partition coefficients and their mean, R(mean), and the mean magnitude of the difference between predicted and experimental values of log(10) P, E, were used to assess the accuracy of each model. For the test set the most accurate for human blood-air partition coefficients were the empirical equations of Meulenberg and Vijverberg (R(mean) = 1.1 +/- 0.46, E = 0.156) and the empirical solvation equation of Abraham and Weathersby (1994) (R(mean) = 0.93 +/- 0.38, E = 0.166). For rat blood, predictions are much less accurate due to difficulties in modeling the effects of protein binding, which are much larger. Overall, for rat blood-air partition coefficients the equation of Meulenberg and Vijverberg (1999) (R(mean) = 0.74 +/- 0.50, E = 0.236) was the most accurate. The tissue-composition-based equations of Poulin and Krishnan, using solubilities in vegetable oil, performed well for human liver-air partition coefficients (R(mean) = 1.21 +/- 0.28, E = 0.079) for log(octanol-water partition coefficients) > 0.7 and for fat-air partition coefficients, but overestimated solubilities in human kidney and brain tissues (e.g., for kidney tissue, R = 1.88 +/- 0.58, E = 0.255). The equations of Meulenberg and Vijverberg (2000a), Abraham and Weathersby (1994), and Paterson and Mackay (1989) also performed moderately well for human tissue-air partition coefficients. For rat muscle-air, liver-air, and fat-air partition coefficients the model of Poulin and Krishnan (1996a) gave the most accurate predictions. For tissue-blood partition coefficients, generally good agreement with experimental values is obtained by the empirical model of Balaz and Lukacova (1999) (e.g., for human kidney, R(mean) = 1.15 +/- 0.38, E = 0.085) and, if solubility in fat is known, by the equations of Fiserova-Bergerova and Diaz (1986) (e.g., for human muscle, R(mean) = 1.10 +/- 0.39, E = 0.107). The equations of DeJongh et al. (1997) gave the most accurate predictions for rat muscle-blood, liver-blood and fat-blood partition coefficients (e.g., for rat muscle R(mean) = 1.03 +/- 0.39, E = 0.149), but predictions were less accurate than for human tissue-blood partition coefficients, attributable to difficulties in modeling the effect of protein binding. The choice of equation for use in physiologically based pharmacokinetic (PBPK) models depends on the species, tissue, and chemical lipophilicity.