Data selection and treatment of chemicals tested for genotoxicity and carcinogenicity.

A database containing qualitative and quantitative results of experimental studies in the fields of genotoxicity and carcinogenicity has been developed. By analyzing results of the studies performed by the U.S. National Toxicology Program, or by a similar program developed in Japan, or reported in the scientific literature, as well performed by private organizations, information has been collected relating to 3389 chemicals, identified by their CAS number. The studies considered for the database include three genotoxicity/mutagenicity short-term test (STTs), namely, two in vitro (Salmonella, gene mutation assay, and mammalian cells/human lymphocytes chromosome aberration assay) and one in vivo, the rodent bone marrow micronucleus assay. To investigate the possible predictive value of these STT assays for carcinogenicity, the results of animal long-term bioassays have also been collected. We have re-evaluated all the genotoxicity studies and the majority of those cases studied in different laboratories with contrasting results has been resolved; a small proportion of questionable cases is, however, still present in the database. In total, 2898 (85.5%) of the chemicals have been tested in the Salmonella assay; 1399 (41.3%) have been tested in the in vitro chromosome aberration assay; 319 (9.4%) have been tested in the in vivo rodent bone marrow cell micronucleus assay; 716 (21.2%) of the chemicals have been tested in the in vivo animal long-term bioassay. For 1118 chemicals tested in the Salmonella assay, 30,650 quantitative studies have been included in the database, thus allowing a possible classification of mutagenic chemicals according to their mutagenic potency.(ABSTRACT TRUNCATED AT 250 WORDS)


Introduction
The assessment ofgenotoxic effects ofchemicals may be performed by means of a series of assays based on a variety of biologic test systems. In vitro and in vivo assays are available to detect either one of three major types of irreversible damage in genetic material, namely, gene mutation, chromosomal aberration, and DNA damage and repair. The induction of such genotoxic effects is indicative also ofpotential carcinogenicity of at least one large class ofchemicals (genotoxic carcinogens); also short-term tests (STTs) are extensively performed for screening large numbers ofchemical substances to identify those chemicals that should have priority for long-term in vivo studies needed for carcinogenicity assessment. However, the choice of genetic end points or the choice ofbiologic systems, and/or the combinations ofend points and systems that might be more adequate for disishing carcinogens from noncarcinogens is still a matter ofdebate, and diffent tiered testing approaches ofSUT batteries are suggested for assessing potential carcinogenicity of chemical substances.
In the current practice, such assessment relies on qualitative xammination ofresponses obained in a number oftests, which are generally selected by experts on a case-by-case basis and are not necessarily the same tests. This strategy, as it stands, is not devoid ofsome degree ofsubjectivity and might lead to some differences in predictions of carcinogenicity of the same chemicals. The strategy would be more efficient if quantification of the dependence ofcarcinogenicity from STT genotoxicity data were available.
In the context ofevaluating chemical substances, the need for relying on predictive methodologies is generally agreed upon. At present such methodologies are used on a qualitative basis for predicting a number oftoxic end points. The rationale for these predictions basically stands on the experience that a) there are correlations between the structure ofthe molecules and their properties, including biologic activities and toxic effects and b) there are toxic effects that are correlated to other toxic effects. This strategy would be more efficient if quantification of these correlations could be made; in other words, ifmathematical models could be developed to define the dependence ofa given toxic effect from either the molecular structure (referred to as quantitative structure-activity relationship, QSAR) or other toxic effects (referred to as quantitative activity-activity relationship, QAAR). These two approaches are complementary to each other, especially in the area of genotoxicity. Whereas single genotoxic end points, or a combination of them, may be modeled by QSAR, carcinogenicity may tentatively be modeled by QAAR, under the assumption that carcinogenic activity may be predicted on the basis of responses obtained in suitable combination of mutagenicity data.
Computer modeling oftoxicity is a suggestive task, though not a simple one. It is not difficult to compile data tables, run them through some computer packages, and end up with a model, but criteria used in the compilation of data, the quality of the data, and the method ofdata analysis can hide weaknesses, sometimes difficult to perceive, that may render the predictive power ofthe model questionable.
To make toxicity models acceptable substitutes of toxicity testing many things are requested. These include proper use of the state of the art in a number ofbranches ofthe disciplines involved (chemistry, toxicology, and statistics) and adequate data. Toxicity of chemicals is related to complex phenomena so that fundamental models based on ab initio calculations are, at present, inadequate. Vice versa, statistical models based on the principle ofanalogy such as QSARs may work. A number ofconditions must, however, be met by the database, and the data analytic method must also be used. One ofthe most important conditions to take into account is the quality ofthe data used in the analysis.
The relevance ofthe present studies on the QSAR resides also on several other factors: a) the need for different government agencies such as the U.S. EPA to define the toxicologic studies to be carried out on the new chemical substances and to be notified under the Toxic Substances Control Act (TSCA) legislation; b) the need to predict with high probability the toxicologic hazard of those numerous substances present on the market for which there might be an emergency faced by public authorities; c) the need to know how many studies are requested for defining the hazard related to an unknown chemical substance; d) the need to identify the toxicological mechanism by which a great number of substances produce different types of biological adverse effects; and e) the need to provide the public with more realistic conclusions on the benefit-risk evaluation for a given chemical substance of large use.
The QSAR studies are particularly interesting in the field of mutagenesis and carcinogenesis due to their partial overlapping and the irreversible nature of their biological implications. For a long time attempts have been made to employ the mutagenicity short-term studies to predict the carcinogenic potential activity of the chemical substances, not only for biological considerations, but also for economic reasons, ifone considers the cost of the long-term studies.
According to the International Agency for Research on Cancer of Lyon (IARC-WHO), in 1990 there were 732 chemical sub-stances that had been tested for carcinogenicity through adequate studies on animals: 85 % ofthem have proven to be carcinogenic. In contrast, the literature reports that almost 10,000 chemical substances have been tested for mutagenicity. For this reason it is quite simple to understand why mutagenicity studies (most of them using in vitro methodologies) have been and are being carried out and why the results are used to develop mathematical models to predict the potential carcinogenic activity ofmutagenic molecules. For molecules ofnew chemicals such studies attempt to assess toxicological potential.
On the basis of independent systems that make use of the chemical structure of substances, the presence of structural genotoxic alert fragments, or computerized evaluation systems, such as CASE, COMPACT, etc., several researchers have developed a prediction hypothesis on the possible different results that might be produced by the long-term carcinogenicity studies currently carried out on 44 chemical substances by the National Toxicology Program (1-4).
Our objective is to develop a predictive model for genotoxicity. In the present paper we have assumed as a basis ofthis objective the collection and the evaluation ofat least two series of testing procedures recognized as indicative ofthe genotoxicity property of a chemical. These two methodologies are represented by a) the Salmonella typhimurium reverse mutation assay for analyzing the ability to induce molecular (gene) mutations in the genetic material and b) the chromosome aberration-mammalian cell growth in vitro assay for analyzing the ability to induce structural (chromosomal) mutations in the genetic material that is organized in chromosomes in the cell nucleus. These two methodologies, on the basis oftheir experimental procedures, represent the most suitable technical approach to maxiiize the cell exposure to a chemical solution, which is a basic condition for a chemical to enter into a cell structure and to react with the DNA (genetic) material, if it is a genotoxic agent.
Ashby (5)  combination of the Salmonella mutation assay and one for the assessment ofchromosome aberrations in vitro. The indication that a chemical is positive in these two in vitro assays clearly defines what is today known as an "in vitro genotoxin." Because it has be shown that not all in vitro genotoxins are carcinogenic to mammals, it has been recommended that all newly discovered in vitro genotoxins should be assessed in vio using very few additional tests (Fig. 1). The experimental data have shown clearly the weight of evidence resulting from the application of two very simple in vitro genotoxicity assays in the evaluation of the mutagenic potential of the chemicals. Our conclusion is that these types of assays could well represent the basis for the correct classification ofa "genotoxin" and that these data should be used for discriminating a mathematical model for predicting genotoxicity of chemical substances. In addition to collecting available data, in this preliminary analysis we have also made an attempt to define thepossible correlations existing between in vitro STT results obtained with the two tests mentioned aboveand in vivo results by applying theprevious hypothesis (5). For the in vivo test, we have chosen the rodent bone marrow micronucleus assay because considerable data existthatmaketheevaluationofthesecorrelationspossible. Moreover, the collection ofresults derived from long-termcarcinogenic tests performed on chemicals has allowed us to make an attempt to investigatethepredictive value ofthe STTs forcarcinogenicity.

Materials
A database for a QSAR study includes a number of chemical substances and, for each ofthem, a number ofnumerical descriptors ofthe molecular structure (x-data) and a number ofmeasured biological responses (y-data). Geneticists are awareofthe factthat a single genetic end point is insufficient to evidentiate the genotoxic profile ofachemicalbecause a variety ofgenetictoxic effects and impairments ofgenetic material processes may lead to an irreversible change in the genetic structure ofan organism.
The literature provides material for our present analysis that could not always be used for developing a model. The data reported in the literature, when these three assays have been used (this applies also to other genotoxicity assays) are extremely variable for a number ofreasons: a) the data are presented only graphically; b) different protocols have been used; c) a maximum dose for the analysis has not always been applied, especially in the negative results; d) a replication ofthe experiment is not present in many studies; e) positive and negative controls rarely have been reported; andJ) criteria for defining a positive series ofresults are different in different laboratories. For these reasons we have proceeded to a particular selection ofthe data to be used in the present study.
From the analysis of literature data, we have selected sets of studies for their intrinsic homogeneity. These sets of data are: 1. The National Toxicology Program (NTP) developed by the U.S. Department of Health and Human Services as a cooperative effort to strengthen and coordinate esearch and testing on toxic chemicals, established in 1978 Reports contining the results ofall studies conducted by the NTP have been published (6)(7)(8)(9)(10)(11)(12). The quantitative results ofthe NTPdeveloped ST1h have been reported in the literature (13)(14)(15)(16)(17)(18)(19)(20)(21)(22). In its Cellular and Genetic Toxicology Program, the NTP is involved in development, improvement, and validation of STTs for mutagens and carcinogens, using STTs to detect and characterize chemicals that may pose carcinogenic or genetic risks to humans. The NTP is focused on developing and validating in vitro and in vio systems for determining the genotoxicity ofchemicals. In the Annual Plan ofthe fiscal year 1988 (11), the NTP reports that "testing with Salmonella strains has been completed on a total of 1566 samples and 1190 unique chemicals since the initiation of the testing program." The NTP has developed a database of STT results, created by using chemicals tested for carcinogenicity by the National Cancer Institute and the NTP. This database allows for the evaluation ofthe Salmonella assay and several other STTs with respect to their ability to predict carcinogenesis or other short-term assay results. Although no assay can detect all carcinogens, in this database a positive result in Salnoella was a better predictor ofcarcinogenicity in rodents than a positive result in other assays. The Salmonella assay differed markedly in its response to chemicals ofdifferent classes.
Chemicals that are carcinogenic only in rats or mice were tested to determine to what extent their mutagenicity depends on the mouse or rat S-9 activation system. There does not appear to be any correlation between the species specificity of the carcinogens and the rodent liver S-9 requirements for mutagenicity. This finding has implications for the use of mutagenicity results for predicting carcinogenicity. A similar testing pgram, although not at the same level ofquantitative development, is conducted under NTP on the evaluation ofthe cytogenetic damage induced in mammalian cells in vitro.

The Institute for Future Technology, Japan, Cooperative
Program on Long-Term Assays for Carcinogenicity (23).
This progrm includes data on Salmonella, as well on mammalian cells for chromosome aberration. The Salmonella data have been originated in several laboratories; the in vitro chromosome aberration results have been originated in only one laboratory and they have been reported in publications by Ishidate in 1983 (24) and in 1988 (25).
3. Other Salmonella results or chromosomal aberration results have been collected from selected papers in the scientific literature (26)(27)(28)(29)(30)(31)(32)(33). 4. The data on the micronucleus assay have been collected from either a paper by Ishidate et al. (26), from Hayashi et al. (32), or from private studies. From the available results, we have collected the data for the present analysis presented in Table 1. Several attempts were made to classify the "questionable" mutagens or nonmutagens by reanalyzing the experimental data to draw our conclusion. In several cases our conclusion was definitive for the classification ofa chemical based on a more critical review ofthe available data. The overall results are reported in Figure 2; for each type of assay the numbers of chemicals classified positive, negative, or questionable are reported. Among the chemicals tested in the two in vtro STT (Salmonella and chromosome aberrations) the database includes 1053 chemicals that have been tested in both assays; for these chemcials, the results ofthe different combinations are reported in Figure 3. Moreover, 270 chemicals have been tested in the two in vitro STTs, as well in the in vivo rodent bone marrow micronucleus test: that data are reported in Figure  4. One or both the in vitro Slms with negative results indicate a chemical with a low probability (25-15%) ofproducing a positive result in the in vio micronucleus test. This probability rises to about 50% ifthe results from both the in vitro STh are positive (Fig. 5).
When comparing the results observed in one single STT or in a combination ofdirrerent STTs with those observed in the carcinogenicity study, their accuracy varies according to a specific combination: the data of such a calculation are reported in Table  2. The accuracy values observed when the results from Salmonella or chromosome aberrations are considered more than 60% as observed by other authors (18). The inclusion of the in vivo micronucleus assay in a combination with the two STTs allows the accuracy value to raise to more than 80%. The best performance of the STTS for predicting the carcinogenicity is that one observed when the two in vitro STTS (Salmonella and chromosomal aberrations) are combined with the in vivo micronucleus assay; in this case a 92.5 % value ofthe accuracy has been calculated (Fig. 6). In Thble 3 the positive and the negative predictivity ofSTTs for carcinogenicity has been calculated and reported.
The inclusion of an in vivo STT in a battery of genotoxicity SiTs, such as the in vio rodent bone marrow cells micronucleus assay, represents an improvement in the present strategy of using STTs to prescreen the potential carcinogenic compounds. Specific data and results on this category ofchemical susbstances will be reported elsewhere (N. Loprieno, in preparation    In the present analysis we found 11 chemicals that were classified negative in Salmonella and chromosome aberration assays but positive in the in vivo micronucleus assay (Fig. 4). These chemicals are reported in Table 4. Five chemicals out of the 11 have been classified also as animal carcinogens. Our opinion is that with these chemicals we are dealing with a specific class of compounds, such as trichloroethylene, vincristine, toluene, and chlorobenzene for which the mechanism of genotoxicity could not be fully applied. We are at present collecting more information on this class of chemicals.
The databases developed as reported in this paper represents an adequate resource for developing quantitative structureactivity relationships.