A quantitative structure-toxicity relationships model for the dermal sensitization guinea pig maximization assay.

We have developed quantitative structure-toxicity relationship (QSTR) models for assessing dermal sensitization using guinea pig maximization test (GPMT) results. The models are derived from 315 carefully evaluated chemicals. There are two models, one for aromatics (excluding one-benzene-ring compounds), and the other for aliphatics and one-benzene-ring compounds. For sensitizers, the models can resolve whether they are weak/moderate or severe sensitizers. The statistical methodology, based on linear discriminant analysis, incorporates an optimum prediction space (OPS) algorithm. This algorithm ensures that the QSTR model will be used only to make predictions on query structures which fall within its domain. Calculation of the similarities between a query structure and the database compounds from which the applicable model was developed are used to validate each skin sensitization assessment. The cross-validated specificity of the equations ranges between 81 and 91%, and the sensitivity between 85 and 95%. For an independent test set, specificity is 79%, and sensitivity 82%.

Assessment of n-octanol/water partition coefficient: when is the assessment reliable?

A model, VLOGP, has been developed for assessment of n-octanol/water partition coefficient, log P, of chemicals from their structures. Unlike group contribution methods, VLOGP is based on linear free energy relationship (LFER) approach and employs information-rich electrotopological structure quantifiers derived solely from molecular topology. VLOGP, a robust and cross-validated model derived from accurately measured experimental log P values of 6675 diverse chemicals, has a coefficient of determination, R2, of 0.986 and a standard error of estimate of 0.20. When applied to the training set, the largest deviation observed between experimental and calculated log P was 0.42. VLOGP is different from other log P predictors in that its application domain, called Optimum Prediction Space (OPS), has been quantitatively defined, i.e., structures to which the model should not be applied for predicting log P can be identified. A computer-assisted implementation of this model within HDi's toxicity assessment software package, TOPKAT 3.0, automatically checks whether the submitted structure is inside the OPS or not. VLOGP was applied to a set of 113 chemicals not included in the training set. It was observed that for the structures inside the OPS the average deviation between experimental and model-calculated log P values is 0.27, whereas the corresponding deviation for structures outside the OPS is 1.35. This demonstrates the necessity of identifying the structures to which a model is not applicable before accepting a model-based predicted log P value. For a set of 47 nucleosides, the performance of VLOGP was compared with that of four published log P predictors; a standard deviation of 0.33 was obtained with VLOGP, whereas the standard deviation from other log P predictors ranged between 0.46 and 1.20.

Prediction of Salmonella mutagenicity.

The ability of a number of prediction systems was examined to determine how well they could predict Salmonella mutagenicity. The prediction systems included two computer-based systems (CASE and TOPKAT), the measurement of a physiochemical parameter (ke) and the use of structural alerts by an expert chemist. The computer-based systems operators and the chemist were supplied with the structures of 100 chemicals that had been tested for mutagenicity in the Salmonella test; the actual chemicals were needed for the physiochemical measurement. None of the participants was provided with the chemical names or Salmonella test results prior to submitting their predictions. The three systems that predicted the mutagenicity from the structure of the chemicals produced equivalent results (71-76% concordance with the Salmonella results); the physiochemical system produced a lower (60-61%) concordance.

Assessment of effect levels of chemicals from quantitative structure-activity relationship (QSAR) models. I. Chronic lowest-observed-adverse-effect level (LOAEL).

With the multitude of new chemicals being synthesized and the paucity of long-term test data on chemicals that could be introduced into the environment, innovative approaches must be developed to determine the health and environmental effects of chemicals. Research was conducted to employ quantitative structure-activity relationship (QSAR) techniques to study the feasibility of developing models to estimate the noncarcinogenic toxicity of chemicals that are not addressed in the literature by relevant studies. A database of lowest-observed-adverse effect level (LOAEL) was assembled by extracting toxicity information from 104 U.S. EPA documents, 124 National Cancer Institute/National Toxicology Program (NCI/NTP) reports, and 6 current reports from the literature. A regression model, based on 234 chemicals of diverse structures and chemical classes including both alicyclic and aromatic compounds, was developed to assess the chronic oral LOAELs in rats. The model was incorporated into an automated computer package. Initial testing of this model indicates it has application to a wide range of chemicals. For about 55% of the compounds in the data set, the estimated LOAELs are within a factor of 2 of the observed LOAELs. For over 93%, they are within a factor of 5. Because of the paucity or absence of long-term toxicity data, the public health and risk assessment community could utilize such QSAR models to determine initial estimates of toxicity for the ever-increasing numbers of chemicals that lack complete pertinent data. However, this and other such models should be used only by expert toxicologists who must objectively look at the estimates thus generated in light of the overall weight of evidence of the available toxicologic information of the subject chemical(s).

Assessment of developmental toxicity potential of chemicals by quantitative structure-toxicity relationship models.

Statistically significant quantitative structure-toxicity relationship (QSTR) models have been developed for assessing developmental toxicity potential (DTP) of chemicals. Three submodels, one each for aliphatic, heteroaromatic and carboaromatic compounds, have been cross-validated to ascertain their robustness. The specificities of the models range from 86% to 97%, and their sensitivities between 86% and 89%. For convenient computer-assisted application, the models are installed in a toxicity assessment software package, TOPKAT, which has been recently enhanced with algorithms to identify whether or not a query structure is inside the optimum prediction space (OPS) of a QSTR model. Different functionalities of the TOPKAT program have been explained by assessing the DTP of a number of compounds not used in the model training sets. The DTP of 18 existing drugs was assessed using these models; the DT assay results were available for 5 of these. Three of these 5 molecules were identified to be inside the OPS and their TOPKAT assessment matched their experimental assignment.

Carcinogenicity of azathioprine: an S-AR investigation.

Besides its use in the treatment of a variety of presumed autoimmune diseases, azathioprine is given as an immunosuppressant to patients who have had renal transplants. Though epidemiological studies have provided "sufficient" evidence of its carcinogenicity in humans, the carcinogenicity tests in rats and mice are considered to be inconclusive because of limitations in the design and results of these tests (IARC, 1981, 1987). Rosenkranz and Klopman (1991) used the CASE program to identify the structural features responsible for its carcinogenicity. They concluded that this genotoxic chemical was a carcinogen due to the presence of the molecular fragment C"-S-C=. The finding was based on the presence of this biophore fragment in five other compounds, namely: 2-amino-5-nitrothiazole, 2-mercaptobenzothiazole, fenthione, 4,4'-thiodianiline and nithiazide. Recently, Ashby (1992) has expressed concern over the validity of their findings. With the aim of contributing to this debate on the mechanism of carcinogenicity of azathioprine, we have analyzed the structural basis of carcinogenicity of azathioprine and the five support compounds using the carcinogenicity predictor of our toxicity prediction program, TOPKAT. The results, more in line with Ashby's concerns, indicate that no molecular fragment involving the S atom is associated with the carcinogenic properties of these molecules. According to the TOPKAT program the carcinogenicity, if any, of azathioprine is due to the NO2 electrophile because its other major structural features are found to be either associated with non-carcinogenicity or do not discriminate carcinogens from non-carcinogens.

Estimation of maximum tolerated dose for long-term bioassays from acute lethal dose and structure by QSAR.

A quantitative structure-activity relationship (QSAR) model has been developed to estimate maximum tolerated doses (MTD) from structural features of chemicals and the corresponding oral acute lethal doses (LD50) as determined in male rats. The model is based on a set of 269 diverse chemicals which have been tested under the National Cancer Institute/National Toxicology Program (NCI/NTP) protocols. The rat oral LD50 value was the strongest predictor. Additionally, 22 structural descriptors comprising nine substructural MOLSTAC(c) keys, three molecular connectivity indices, and sigma charges on 10 molecular fragments were identified as endpoint predictors. The model explains 76% of the variance and is significant (F = 35.7) at p less than 0.0001 with a standard error of the estimate of 0.40 in the log (1/mol) units used in Hansch-type equations. Cross-validation showed that the difference between the average deleted residual square (0.179) and the model residual square (0.160) was not significant (t = 0.98).

Prediction of probability of carcinogenicity for a set of ongoing NTP bioassays.

Forty-four compounds currently undergoing carcinogenesis bioassay by the National Toxicology Program were submitted to the TOPKAT program for prediction of their potential carcinogenicity. Sixteen compounds could not be handled by TOPKAT. Of the 28 for which predictions were made, 26 (93%) had a confidence level in the estimate of at least moderate. Seventeen were predicted to be carcinogens and 11 non-carcinogens. These results will be compared with the assay results as the assays are completed.

Salmonella mutagenicity and rodent carcinogenicity: quantitative structure-activity relationships.

Based on a compilation of 222 reports of rodent nominal lifetime carcinogenicity bioassays by the NCI/NTP on the one hand, and corresponding Salmonella mutagenicity bioassays (Ames tests) on the other, Ashby and Tennant (1988) have divided the carcinogens and non-carcinogens into genotoxic (Ames test positive) and non-genotoxic (Ames test negative) groups and discussed structural characteristics common to each of these groups. The Ames test alone was deemed to be adequate for the identification of genotoxicity because other short-term bioassays, and even combinations, or batteries, appeared to offer no significant advantages. From the results of this study it is possible to achieve (1) a division of the carcinogens into the same genotoxic and non-genotoxic groups, and (2) a division of the non-genotoxic compounds into the same carcinogenic and non-carcinogenic groups, solely on the basis of structure-activity relationships, with a classification accuracy of approx. 95%. (1) An equation comprising 8 sigma molecular charge descriptors, 2 molecular connectivity indices (MCIs), 2 kappa molecular shape descriptors and one MOLSTAC substructure descriptor achieved discrimination between genotoxic and non-genotoxic carcinogens with an accuracy of 94.5%. (2) Another equation comprising 8 sigma molecular charge descriptors, 3 MCIs, one kappa shape descriptor and 12 substructural descriptors achieved discrimination between non-genotoxic carcinogens and non-genotoxic non-carcinogens with an accuracy of 95.2%. These SAR models are suitable for the distinction between (1) genotoxic and non-genotoxic carcinogens and (2) carcinogenic and non-carcinogenic non-genotoxins, both in the absence of animal bioassay data.

A QSAR model for the estimation of carcinogenicity: example application to an azo-dye.

Since carcinogenicity bioassays are time-consuming, costly, and use animal resources, structure-activity relationship equations which model toxicological end-points have been developed to make available alternative methods which approximate the results that could be obtained from bioassays but which are less expensive and time-consuming and use fewer, if any, animals. These equations are based on sets of bioassay results and explain the end-point under consideration in terms of substructural and other parameters which describe the chemical entities. The resulting equations--or models--can then be used to estimate--or predict--the end-point for new structures. The estimation is followed by validation procedures.

An overview of structure-activity relationships as an alternative to testing in animals for carcinogenicity, mutagenicity, dermal and eye irritation, and acute oral toxicity.

The use of structure-activity relationships (SAR) has proven practical for the development of equations which can be used to estimate the above-listed endpoints for a large variety of chemicals. The SAR models predict these endpoints correctly in 85 to 97% of the cases and often surpass in their predictive ability the results obtainable from the equivalent biological assays. These SAR models are being used at several levels: drug, or more generally, chemical discovery; prioritization for testing; regulatory affairs; investigation of detoxification mechanisms; and risk estimation. In the new compound (discovery) use, potential toxic effects of a set of related compounds are investigated before synthesis to select those chemicals with the lesser probabilities of producing toxic effects for further investigation, at considerable savings in research expenditure since fewer compounds need to be synthesized, and the avoidance of blind alleys. Prioritization for testing is used in numerous instances, such as selecting those chemicals in an environment which are most likely to have toxic effects for priority attention. SAR models are used by regulatory agencies to determine the possible toxic effects of chemicals for which data insufficient to render decisions have been submitted, and to gain insight into possible toxicity problems. SAR models are also used to investigate possible metabolites, and toxicity mechanisms due to the ability of making computer-based structural modifications and observing the effects on the modelled toxic endpoints. Risk analysis is a natural outgrowth of several of the above applications, and is particularly useful for SAR models of carcinogenicity. SAR models as alternatives to animal bioassays should be used in the context of other information for the chemicals of concern. Just as bioassays and in vitro methods have their limitations, so do SAR models. These include the sometimes limited data base on which to base an SAR model, the temptation to extrapolate beyond the confines of the model, and the noise inherent in the bioassays on which the models are based. Within these constraints SAR models have a considerable potential in reducing the number of animals used in toxicity testing.

Multiparameter pilot study of adolescent idiopathic scoliosis.

Seventy-two adolescent females with idiopathic scoliosis were studied to determine the possible relationships between parameters previously studied by others, as well as several new parameters. Dichotic listening tests showed that patients with less right-ear advantage for language were more likely to have progressive curves than patients with greater right-ear advantage, implying a difference in hemispheral dominance between the two groups. This finding, if confirmed with larger populations, could lead to a prognostic test for the determination of therapeutic modality. Platelet aggregation was found to be lower than in a group of controls; however, similar findings were also obtained from a group of patients with chronic orthopaedic conditions. Mitral valve prolapse, the electroencephalogram and the mother's age at patient's birth were all found to be nonspecific for the patients as compared to control groups. The level of autonomic system activity in the patients was found to be higher than that observed in controls.

Teratogenesis: a statistical structure-activity model.

This structure-activity model of teratogenicity was developed to provide the ability to rank untested compounds by their probability of teratogenicity. The model is based on 430 compounds collected from various sources in the literature and scored from zero to one as to evidence of teratogenicity. A discriminant equation then separates those compounds in the extremes of this distribution. The false positive classification rate based on the compounds in the equation is approximately 8% and the false negative rate approximately 10%. Approximately 22% of the compounds are not classifiable as either teratogens or nonteratogens with this equation.

Mutagenicity (Ames): a structure-activity model.

A statistical structure-activity model of the Salmonella typhimurium (Ames) test has been devised based on 472 chemicals for which this endpoint has been measured. The model uses substructural fragments as the independent parameters to explain the difference in mutagenicity of the different chemicals. The model is able to classify 86% of the chemicals into their correct categories; the false-positive rate is 4.7%, and the false-negative rate 5.3%. Approximately 10% of the chemicals cannot be classified by the existing equation. This structure-activity model can be used as a preliminary screen prior to other testing as well as for setting priorities for more detailed investigations.

Carcinogenesis: a predictive structure-activity model.

A statistical structure-activity equation has been developed for the estimation of carcinogenic potential for chemicals that have not been subjected to carcinogenesis assays. This discriminant equation is based on substructural fragments and molecular weight obtained for 343 compounds. The data base used for the design of the equation was obtained from the monographs of the International Agency for Research on Cancer. The accuracy of classification for the carcinogens in the model is between 87 and 91%, and for noncarcinogens between 78 and 80% in the presence of between 5.5 and 10.2% of the compounds not being classifiable. The false negative rate ranges between 4 and 5%; the false positive rate is near 11%.

A test of the hypothesis that time to tumor in humans fits the "one-hit" model.

Data from the Connecticut Tumor Registry for 1935-1968 for 14 tumor sites were subjected to formal statistical tests of the hypothesis that the time to tumor has an exponential distribution (and hence is a result of one random "hit"). All of the resulting 270 test statistics rejected the hypothesis.