The carcinogenicity prediction and battery selection (CPBS) method: a Bayesian approach

Recently, a large number of relatively inexpensive in vitro short-term tests have been developed to help predict the carcinogenicity of chemicals. The carcinogenicity prediction and battery selection (CPBS) method utilizes the results of such short-term tests to screen for chemicals that are most likely to cause cancer. The method is an integrated approach for analyzing large, often sparsely filled, data bases containing short-term test results, which often have only marginal representation of known non-carcinogens. The CPBS method is developed for the purpose of (i) determining the reliability and predictive capability of individual and batteries of short-term tests, and (ii) developing a strategy for formulating and selecting optimally preferred batteries of short-term tests for screening chemicals for further testing. The term 'optimally preferred' connotes the best acceptable combination of tests in terms of trade-offs among the multiple attributes of each test and resulting battery (e.g., cost, sensitivity, specificity, etc). The CPBS method consists of 5 major tasks: (1) data consolidation, (2) parameter estimation, (3) predictivity calculation, (4) battery selection and (5) risk assessment. Although there is a great need for more research and improvement, the CPBS method at its present stage should add an important method to the maze of the thousands of new chemicals that are introduced into drugs, foods, consumer goods and to the environment every year. This method should also provide an enhanced identification procedure for classifying chemicals more accurately as suspected carcinogens or non-carcinogens.     

Invited contribution: An objective approach to the development of short-term tests predictive of carcinogenicity

The Carcinogenicity Prediction and Battery Selection procedure was developed to address two problems: (1) the identification of highly predictive, yet cost-effective, batteries of short-term tests and (2) the objective prediction of the potential carcinogenicity of chemicals based upon the results of short-term tests even when a mixture of positive and negative results is obtained. In the present report the usefulness of the Carcinogenicity Prediction and Battery Selection procedure is demonstrated using benzo[a]pyrene, benzoin and diethylstilbestrol as examples. In addition, its applicability in the analysis of all the possible outcomes of a battery is illustrated together with an analysis of the worth of additional testing.Abbreviations B[a]P benzo[a]pyrene - CASE Computer-Automated Structure Evaluation - CPBS Carcinogenicity Prediction and Battery Selection - DEHP diethylhexylphthalate - DES diethylstilbestrol - NTA nitrilotriacetate - TCDD 2,3,7,8-tetrachlorodibenzo-p-dioxin     

Cluster analysis in predicting the carcinogenicity of chemicals using short-term assays

Cluster analysis can be a useful tool for exploratory data analysis to uncover natural groupings in data, and initiate new ideas and hypotheses about such groupings. When applied to short-term assay results, it provides and improves estimates for the sensitivity and specificity of assays, provides indications of association between assays and, in turn, which assays can be substituted for one another in a battery, and allows a data base containing test results on chemicals of unknown carcinogenicity to be linked to a data base for which animal carcinogenicity data are available. Cluster analysis was applied to the Gene-Tox data base (which contains short-term test results on chemicals of both known and unknown carcinogenicity). The results on chemicals of known carcinogenicity were different from those obtained when the entire data base was analyzed. This suggests that the associations (and possibly the sensitivities and specificities) which are based on chemicals of known carcinogenicity may not be representative of the true measures. Cluster analysis applied to the total data base should be useful in improving these estimates. Many of the associations between the assays which were found through the use of cluster analysis could be 'validated' based on previous knowledge of the mechanistic basis of the various tests, but some of the associations were unsuspected. These associations may be a reflection of a non-ideal data base. As additional data becomes available and new clustering techniques for handling non-ideal data bases are developed, results from such analyses could play an increasing role in strengthening prediction schemes which utilize short-term tests results to screen chemicals for carcinogenicity, such as the carcinogenicity and battery selection (CPBS) method (Chankong et al., 1985).     

Predictive assay for rodent carcinogenicity using in vivo biochemical parameters: operational characteristics and complementarity.

111 chemicals of known rodent carcinogenicity (49 carcinogens, 62 noncarcinogens), including many promoters of carcinogenesis, nongenotoxic carcinogens, hepatocarcinogens, and halogenated hydrocarbons, were selected for study. The chemicals were administered by gavage in two dose levels to female Sprague-Dawley rats. The effects of these 111 chemicals on 4 biochemical assays (hepatic DNA damage by alkaline elution (DD), hepatic ornithine decarboxylase activity (ODC), serum alanine aminotransferase activity (ALT), and hepatic cytochrome P-450 content (P450)) were determined. Composite parameters are defined as follows: CP = [ODC and P450), CT = [ALT and ODC), and TS = [DD or CP or CT]. The operational characteristics of TS for predicting rodent cancer were sensitivity 55%, specificity 87%, positive predictivity 77%, negative predictivity 71%, and concordance 73%. For these chemicals, the 73% concordance of this study was superior to the concordance obtained from published data from other laboratories on the Ames test (53%), structural alerts (SA) (46%), chromosome aberrations in Chinese hamster ovary cells (ABS) (48%), cell mutation in mouse lymphoma 15178Y cells (MOLY) (52%), and sister-chromatid exchange in Chinese hamster ovary cells (SCE) (60%). The 4 in vivo biochemical assays were complementary to each other. The composite parameter TS also shows complementarity to all 5 other predictors of rodent cancer examined in this paper. For example, the Ames test alone has a concordance of only 53%. In combination with TS, the concordance is increased to 62% (Ames or TS) or to 63% (Ames and TS). For the 67 chemicals with data available for SA, the concordance for predicting rodent carcinogenicity was 47% (for SA alone), 54% (for SA or TS), and 66% (for SA and TS). These biochemical assays will be useful: (1) to predict rodent carcinogenicity per se, (2) to 'confirm' the results of short-term mutagenicity tests by the high specificity mode of the biochemical assays (the specificity and positive predictivity are both 100%), and (3) to be a component of future complementary batteries of tests for predicting rodent carcinogenicity.     

Genomic Models of Short-Term Exposure Accurately Predict Long-Term Chemical Carcinogenicity and Identify Putative Mechanisms of Action

Background

Despite an overall decrease in incidence of and mortality from cancer, about 40% of Americans will be diagnosed with the disease in their lifetime, and around 20% will die of it. Current approaches to test carcinogenic chemicals adopt the 2-year rodent bioassay, which is costly and time-consuming. As a result, fewer than 2% of the chemicals on the market have actually been tested. However, evidence accumulated to date suggests that gene expression profiles from model organisms exposed to chemical compounds reflect underlying mechanisms of action, and that these toxicogenomic models could be used in the prediction of chemical carcinogenicity.

Results

In this study, we used a rat-based microarray dataset from the NTP DrugMatrix Database to test the ability of toxicogenomics to model carcinogenicity. We analyzed 1,221 gene-expression profiles obtained from rats treated with 127 well-characterized compounds, including genotoxic and non-genotoxic carcinogens. We built a classifier that predicts a chemical''s carcinogenic potential with an AUC of 0.78, and validated it on an independent dataset from the Japanese Toxicogenomics Project consisting of 2,065 profiles from 72 compounds. Finally, we identified differentially expressed genes associated with chemical carcinogenesis, and developed novel data-driven approaches for the molecular characterization of the response to chemical stressors.

Conclusion

Here, we validate a toxicogenomic approach to predict carcinogenicity and provide strong evidence that, with a larger set of compounds, we should be able to improve the sensitivity and specificity of the predictions. We found that the prediction of carcinogenicity is tissue-dependent and that the results also confirm and expand upon previous studies implicating DNA damage, the peroxisome proliferator-activated receptor, the aryl hydrocarbon receptor, and regenerative pathology in the response to carcinogen exposure.     

Animal carcinogenicity studies: implications for the REACH system

The 2001 European Commission proposal for the Registration, Evaluation and Authorisation of Chemicals (REACH) aims to improve public and environmental health by assessing the toxicity of, and restricting exposure to, potentially toxic chemicals. The greatest benefits are expected to accrue from decreased cancer incidences. Hence the accurate identification of chemical carcinogens must be a top priority for the REACH system. Due to a paucity of human clinical data, the identification of potential human carcinogens has conventionally relied on animal tests. However, our survey of the US Environmental Protection Agency's (EPAs) toxic chemicals database revealed that, for a majority of the chemicals of greatest public health concern (93/160, i.e. 58.1%), the EPA found animal carcinogenicity data to be inadequate to support classifications of probable human carcinogen or non-carcinogen. A wide variety of species were used, with rodents predominating; a wide variety of routes of administration were used; and a particularly wide variety of organ systems were affected. These factors raise serious biological obstacles that render accurate extrapolation to humans profoundly difficult. Furthermore, significantly different International Agency for Research on Cancer assessments of identical chemicals, indicate that the true human predictivity of animal carcinogenicity data is even poorer than is indicated by the EPA figures alone. Consequently, we propose the replacement of animal carcinogenicity bioassays with a tiered combination of non-animal assays, which can be expected to yield a weight-of-evidence characterisation of carcinogenic risk with superior human predictivity. Additional advantages include substantial savings of financial, human and animal resources, and potentially greater insights into mechanisms of carcinogenicity.     

Computer-aided rodent carcinogenicity prediction

The potential of the computer program PASS (Prediction Activity Spectra for Substances) to predict rodent carcinogenicity for chemical compounds was studied. PASS predicts carcinogenicity of chemical compounds on the basis of their structural formula and of structure-activity relationship analysis of known carcinogens and non-carcinogens. The data on structures and experimental results of 2-year carcinogenicity assays for 412 chemicals from the NTP (National Toxicological Program) and 1190 chemicals from the CPDB (Carcinogenic Potency Database) were used in our study. The predictions take into consideration information about species and sex of animals. For evaluation of the predictive accuracy we used two procedures: leave-one-out cross-validation (LOO CV) and leave-20%-out cross-validation. In the last case we randomly divided the studied data set 20 times into two subsets. The data from the first subset, containing 80% of the compounds, were added to the PASS training set (which includes about 46,000 compounds with about 1500 biological activity types collected during the last 20 years to predict biological activity spectra), the second subset with 20% of the compounds was used as an evaluation set. The mean accuracy of prediction calculated by LOO CV is about 73% for NTP compounds in the 'equivocal' category of carcinogenic activity and 80% for NTP compounds in the 'evidence' category of carcinogenicity. The mean accuracy of prediction for the CPDB database is 89.9% calculated by LOO CV and 63.4% calculated by leave-20%-out cross-validation. Influence of incorporation of species and sex data on the accuracy of carcinogenicity prediction was also investigated. It was shown that the accuracy was increased only for data on male animals.     

Concomitant observations of UDS in the liver and micronuclei in the bone marrow of rats exposed to cyclophosphamide or 2-acetylaminofluorene

Oral dosing of between 5–30 mg/kg of cyclophosphamide (CP) to Alderley Park rats induced micronuclei in the bone marrow between 12 and 36 h after dosing, but failed to induce unscheduled DNA synthesis (UDS) in the liver at similar dose levels and treatment periods. Dose levels of > 30 mg/kg were toxic to the liver. In contrast, 2-acetylaminofluorene (2AAF) induced UDS in the rat liver between 4–36 h after dosing, but gave only a weak response in the bone marrow assay at dose levels between 0.5 and 2 g/kg. Selected observations were made for each chemical using both tissues of the same test animal.

It is concluded that an assessment of the genotoxicity in vivo of chemicals defined as genotoxic in vitro will contribute to an assessment of their possible mammalian carcinogenicity, and that these should involve assays conducted using both the bone marrow and the liver of rodents. Due to its relative ease of commission, the bone marrow micronucleus assay will usually be conducted first; in the case of negative results it is recommended that a liver genotoxicity assay should also be conducted. The case for employing in vivo short-term genotoxicity tests to predict the possible organotropic carcinogenicity or germ cell mutagenicity of a new in vitro genotoxin is discussed.     

Genotoxicity testing of extracts of a Swedish moist oral snuff

The present study was designed to investigate the potential genotoxicity of aqueous and methylene chloride extracts of Swedish moist oral snuff. The test systems were selected to provide optimal data for the prediction of carcinogenicity in rodents and included assays for the induction of mutation in bacteria, sister-chromatid exchanges (SCE) in human lymphocytes, of chromosome aberrations and gene mutations in V79 Chinese hamster cells and of micronuclei in mouse bone marrow cells. In addition, the methylene chloride extract was tested for the induction of sex-linked recessive lethal mutations in Drosophila melanogaster. The aqueous extract of 'Snus' induced SCE in human lymphocytes and chromosome aberrations in V79 cells, the latter effect being observed both with and without metabolic activation. No induction of point mutations was detected with the Ames test or in V79 cells and the micronucleus test in mice was negative. It was demonstrated that the induction of chromosome aberrations without metabolic activation may be due to a high salt concentration, indicating that the clastogenic agent(s) in this extract required metabolic activation. The methylene chloride extract showed genotoxicity in the Ames test, the SCE test and the chromosome aberration test, whereas no induction of gene mutations in V79 cells was observed. Once again, the results suggested that metabolism is required for genotoxicity. The methylene chloride extract did not cause induction of micronuclei in mice or of sex-linked recessive lethal mutations in Drosophila melanogaster. These combined data on genotoxicity were analyzed using various models for the prediction of carcinogenicity. In a sequential testing model, the probabilities that the aqueous and methylene chloride extracts of 'Snus' are carcinogenic due to a genotoxic mechanism were both predicted to be low. Using carcinogenicity prediction by battery selection (CPBS), the probabilities of the methylene chloride and aqueous extracts being correctly identified as non-carcinogens are 71 and 77%, respectively. Up to date, the CPBS approach has been validated primarily for individual compounds, so some caution should at present be exercised in interpreting the results using this method. Based on these results, the carcinogenic potential of Swedish 'Snus' should be considered to be low, a conclusion in agreement with the low incidence of oral cancer in Sweden compared to other countries.     

Knowledge-based battery design of short-term tests based on dose information

A construction of batteries of short-term tests (STTs) is described which is based on a classification of 73 chemicals in regard to their carcinogenicity. The 73 chemicals were studied within the U.S. National Toxicology Program (Ashby and Tennant, 1988). The batteries are validated using the classification of 35 additional chemicals. They are defined by logically structured combinations of rules. The single rules are defined by the z-scores of the logarithmic values of the limiting doses obtained from the 4 in vitro STTs used in the study by Ashby and Tennant. The limiting dose is defined as the lowest effective dose or the highest ineffective dose (Waters et al., 1987). The batteries are constructed by minimizing the number of disagreements with the classification by Ashby and Tennant. Compared with the results obtained from single STTs, 2 batteries of 3 STTs have higher concordances with the carcinogenicity data, namely 70% for the NTP data and 74-77% for the independent test data. In addition, a theoretical result shows that the proposed battery design, for a large enough learning set of chemicals, leads to results which are replicated with high probability on a large enough validation set. Based on the first results obtained with a limited number of chemicals it is concluded that the knowledge-based battery design is worth further development.     

The ability of plant genotoxicity assays to predict carcinogenicity

A number of assays have been developed which use higher plants for measuring mutagenic or cytogenetic effects of chemicals, as an indication of carcinogenicity. Plant assays require less extensive equipment, materials and personnel than most other genotoxicity tests, which is a potential advantage, particularly in less developed parts of the world. We have analyzed data on 9 plant genotoxicity assays evaluated by the Gene-Tox program of the U.S. Environmental Protection Agency, using methodologies we have recently developed to assess the capability of assays to predict carcinogenicity and carcinogenic potency. All 9 of the plant assays appear to have high sensitivity (few false negatives). Specificity (rate of true negatives) was more difficult to evaluate because of limited testing on non-carcinogens; however, available data indicate that only the Arabidopsis mutagenicity (ArM) test appears to have high specificity. Based upon their high sensitivity, plant genotoxicity tests are most appropriate for a risk-averse testing program, because although many false positives will be generated, the relatively few negative results will be quite reliable.     

Using Short-Term Tests to Predict Carcinogenic Activity in the Long-Term Bioassay

A method for classifying chemicals with respect to carcinogenic potential based on short-term test results is presented. The method utilizes the logistic regression model to translate results from short-term toxicity assays into predictions of the likelihood that a chemical will be carcinogenic if tested in a long-term bioassay. The proposed method differs from previous approaches in two ways. First, statistical confidence limits on probabilities of cancer rather than central estimates of those probabilities are used for classification. Second, the method does not classify all chemicals in a data base with respect to carcinogenic potential. Instead, it identifies chemicals with highest and lowest likelihood of testing positive for carcinogenicity in the bioassay. A subset of chemicals with intermediate likelihood of being positive remains unclassified, and will require further testing, perhaps in a long-term bioassay. Two data bases of binary short-term and long-term test results from the literature are used to illustrate and evaluate the proposed procedure. A cross-validation analysis of one of the data sets suggests that, for a sufficiently rich data base of chemicals, the development of a robust predictive system to replace the bioassay for some unknown chemicals is a realistic goal.     

Assembly and preliminary analysis of a genotoxicity data base for predicting carcinogens

With a view to developing methodologies for predicting the carcinogenicity of chemicals on the basis of the results of short-term assays and selecting highly predictive batteries of short-term tests, a data base was assembled. The present is a compilation of data extracted from the reports of Gene-Tox working groups, Salmonella mutagenicity data obtained from the U.S. National Toxicology Program and the Environmental Mutagen Information Center and results from BHK21 transformation assays.     

Genetic toxicity in relation to receptor-mediated carcinogenesis

There is growing agreement on the types and number of assays required to assess the ability of a chemical to mutate or to affect in a heritable manner the expressional integrity of DNA. This usually involves measurement of the ability of a chemical to induce chromosomal aberrations or gene mutations in cultured cells, coupled to confirmation of genetic toxicity in rodents. The results of such assays, coupled to assessment of the chemical structure of the agent for sites of actual or potential electrophilicity, provide a major and primary input to estimation of whether a rodent carcinogen is operating by a genotoxic or a non-genotoxic mechanism. The extent and sites of carcinogenesis also contribute to this decision. In cases where the mechanism of action of a carcinogen is to be studied in detail, additional assessments of genetic toxicity can be made in the species/gender/tissue subject to carcinogenesis. Suitable assays include measurements of DNA adducts (e.g., ³²P post-labelling), assessment of DNA damage using, for example, the single-cell gel electrophoresis (Comet) assay, or the determination of transgenic mutation frequencies in appropriate rodent model systems. The genetic toxicity of o-anisidine, methyl clophenipate, etoposide and taxol are discussed to illustrate these concepts.

The present need is for high quality genetic toxicity data to be derived and integrated with other relevant toxicological data on a new carcinogen in order to provide an informed estimate its most likely mechanisms of carcinogenic action.     

Beryllium: genotoxicity and carcinogenicity

Beryllium (Be) has physical-chemical properties, including low density and high tensile strength, which make it useful in the manufacture of products ranging from space shuttles to golf clubs. Despite its utility, a number of standard setting agencies have determined that beryllium is a carcinogen. Only a limited number of studies, however, have addressed the underlying mechanisms of the carcinogenicity and mutagenicity of beryllium. Importantly, mutation and chromosomal aberration assays have yielded somewhat contradictory results for beryllium compounds and whereas bacterial tests were largely negative, mammalian test systems showed evidence of beryllium-induced mutations, chromosomal aberrations, and cell transformation. Although inter-laboratory differences may play a role in the variability observed in genotoxicity assays, it is more likely that the different chemical forms of beryllium have a significant effect on mutagenicity and carcinogenicity. Because workers are predominantly exposed to airborne particles which are generated during the machining of beryllium metal, ceramics, or alloys, testing of the mechanisms of the mutagenic and carcinogenic activity of beryllium should be performed with relevant chemical forms of beryllium.     

The structural basis of the mutagenicity of chemicals in Salmonella typhimurium: the Gene-Tox data base

The CASE structure-activity methodology has been applied to a Gene-Tox derived Salmonella mutagenicity data base consisting of 808 chemicals. Based upon qualitative structural features, CASE identified 29 activating and 3 inactivating structural determinants which correctly predicted the probability of carcinogenicity of 93.7% of the known mutagens and non-mutagens in the data base (sensitivity = 0.998, and specificity = 0.704). Additionally, based upon a qualitative structure-activity analysis, CASE's performance was even better, leading to a sensitivity of 0.981 and a specificity of 1.000. Using the structural determinants identified in this data base, CASE gave excellent predictions of the mutagenicity of chemicals not included in the data base. The identified biophores and biophobes can also be used to investigate the structural basis of the mutagenicity of various chemical classes.     

Cytogenetic effects in the mouse of 17 chemical mutagens and carcinogens evaluated by the micronucleus test.

2 dialkylnitrosamines, 4 oxazaphosphorines, 6 aryldialkyltriazenes, urethane, N-hydroxyurethane, 4-nitroquinoline-1-oxide, procarbazine (natulan) and the inorganic carcinogen potassium chromate were studied for cytogenetic activity in the micronucleus test on mouse bone marrow. Except diethylnitrosamine, all chemicals were active. The results are compared with those known from studies in other mammalian and sub-mammalian test systems. The results of the micro nucleus test correlate well with results from other mutagenicity tests and with the carcinogenicity of the chemicals. The lack of an effect on N-nitrosodiethylamine (DENA) is discussed with regard to the short life-time of the ultimate mutagen.     

Animal carcinogenicity studies: 1. Poor human predictivity

The regulation of human exposure to potentially carcinogenic chemicals constitutes society's most important use of animal carcinogenicity data. Environmental contaminants of greatest concern within the USA are listed in the Environmental Protection Agency's (EPA's) Integrated Risk Information System (IRIS) chemicals database. However, of the 160 IRIS chemicals lacking even limited human exposure data but possessing animal data that had received a human carcinogenicity assessment by 1 January 2004, we found that in most cases (58.1%; 93/160), the EPA considered animal carcinogenicity data inadequate to support a classification of probable human carcinogen or non-carcinogen. For the 128 chemicals with human or animal data also assessed by the World Health Organisation's International Agency for Research on Cancer (IARC), human carcinogenicity classifications were compatible with EPA classifications only for those 17 having at least limited human data (p = 0.5896). For those 111 primarily reliant on animal data, the EPA was much more likely than the IARC to assign carcinogenicity classifications indicative of greater human risk (p < 0.0001). The IARC is a leading international authority on carcinogenicity assessments, and its significantly different human carcinogenicity classifications of identical chemicals indicate that: 1) in the absence of significant human data, the EPA is over-reliant on animal carcinogenicity data; 2) as a result, the EPA tends to over-predict carcinogenic risk; and 3) the true predictivity for human carcinogenicity of animal data is even poorer than is indicated by EPA figures alone. The EPA policy of erroneously assuming that tumours in animals are indicative of human carcinogenicity is implicated as a primary cause of these errors.     

Comparative investigation of multiple organs of mice and rats in the comet assay

Mice and/or rats are usually used to detect chemical carcinogenicity and it has been known that there are species differences in carcinogenicity. To know whether there are species difference in genotoxicity, we conducted comparative investigation of multiple organs of mice and rats in the comet assay. Since the sensitivity to xenobiotics is different for different species, we queried species difference in the genotoxic sensitivity at one equitoxic level but not at one equidose. Therefore, groups of four mice or rats were treated once intraperitoneally or orally with a chemical at highest dose without death and distinct toxic manifestation. When the death was not observed at 2000 mg/kg of a chemical, 2000 mg/kg was used for the comet study. The stomach, colon, liver, kidney, bladder, lung, brain, and bone marrow were sampled 3, 8, and 24h after treatment. Among chemicals tested, benzyl acetate, chlorodibromomethane and p-chloro-o-toluidine are carcinogenic to mice but not rats, and aniline, azobenzene, o-phenylphenol Na, and D-limonene are carcinogenic to rats but not mice. Although the two species differed in genotoxicity target organs and migration values, the judgement of a positive or negative response was the same for all chemicals studied except for 2,4-dimethoxyaniline, 2,5-diaminotoluene, and p,p'-DDT when chemicals with positive responses in at least one organ are judged to be comet assay-positive. 2,4-Dimethoxyaniline and 2,5-diaminotoluene that are Ames test-positive non-carcinogens in both species were positive in one organ (urinary bladder for 2,4-dimethoxyaniline and stomach for 2,5-diaminotoluene) in rats, but negative in all mouse organs. p,p'-DDT, which is an Ames test-negative but in vitro cytogenetic test-positive hepatic carcinogen in mice and rats, was positive in multiple rat organs, but not in any mouse organ. These results suggest that species differences in genotoxicity at one equitoxic level are not consistent with species difference in carcinogenicity and that the use of both species is appropriate to indicate a carcinogenic potential in the comet assay with multiple organs, when chemicals being positive in at least one organ are judged to be comet assay-positive.     

Classification according to chemical structure, mutagenicity to Salmonella and level of carcinogenicity of a further 42 chemicals tested for carcinogenicity by the U.S. National Toxicology Program

This paper is an extension and update of an earlier review published in this journal (Ashby and Tennant, 1988). A summary of the rodent carcinogenicity bioassay data on a further 42 chemicals tested by the U.S. National Toxicology Program (NTP) is presented. An evaluation of each chemical for structural alerts to DNA-reactivity is also provided, together with a summary of its mutagenicity to Salmonella. The 42 chemicals were numbered and evaluated as an extension of the earlier analysis of 222 NTP chemicals. The activity patterns and conclusions derived from the earlier study remain unchanged for the larger group of 264 chemicals. Based on the extended database of 264 NTP chemicals, the sensitivity of the Salmonella assay for rodent carcinogens is 58% and the specificity for the non-carcinogens is 73%. A total of 32 chemicals were defined as equivocal for carcinogenicity and, of these, 11 (34%) are mutagenic to Salmonella. An evaluation is made of instances where predictions of carcinogenicity, based on structural alerts, disagree with the Salmonella mutagenicity result (12% of the database). The majority of the disagreements are for structural alerts on non-mutagens, and that places these alerts as a sensitive primary screen with a specificity lower than that of the Salmonella assay. That analysis indicates some need for assays complementary to the Salmonella test when screening for potential genotoxic carcinogens. It also reveals that the correlation between structural alerts and mutagenicity to Salmonella is probably greater than 90%. Chemicals predicted to show Michael-type alkylating activity (i.e., CH2 = CHX; where X = an electron-withdrawing group, e.g. acrylamide) have been confirmed as a structural alert, and the halomethanes (624 are possible) have been classified as structurally-alerting. To this end an extended carcinogen-alert model structure is presented. Among the 138 NTP carcinogens now reviewed, 45 (33%) are non-mutagenic to Salmonella and possess a chemical structure that does not alert to DNA-reactivity. These carcinogens therefore either illustrate the need for complementary genetic screening tests to the Salmonella assay, or they represent the group of non-genotoxic carcinogens referred to most specifically by Weisburger and Williams (1981); the latter concept is favoured.