Profiles of chemically-induced tumors in rodents: quantitative relationships

Abstract

The rodent carcinogenicity bioassay has been used for several decades for evaluating hundreds of chemicals, with the two aims of better understanding the etiologies of cancer, and of assessing the hazard posed by environmental and industrial chemicals. This has generated an enormous wealth of data and information on the phenomenon of chemical carcinogenicity. However, this information cannot be appreciated easily, since too many details may obscure the general trends present in the data; on the contrary, the use of computerized data analysis techniques suitable for the exploration of large databases makes its investigation much more fruitful, and its results more reliable. For this work, we collected a database of 536 rodent carcinogens, and we investigated the profiles of tumors (target organs) induced in the four experimental systems which are usually employed (rat and mouse, male and female). The analysis was performed with an Artificial Neural Network called Kohonen Self-Organizing Map, which is a computer-intensive method aimed at making the relevant information emerge automatically from the data itself. The analysis generated a global view, as well as a quantitative measure of the associations among the individual tumor types, and among the tumor profiles induced by the chemicals. In the complex interplay between the organ and species specificity of tumor induction, the species specificity generally overcame organ specificity, except for a few tumors (namely Lymphatic System, Brain, Forestomach, Stomach and Thyroid Gland). Moreover, the species specificity was remarkably stronger than the trans-species sex specificity. For three chemical classes (Aromatic Amines, Electrophilic/Alkylating Agents, Nitroarenes) most represented in the database, we investigated the hypothesis that a single mechanism of interaction with DNA would produce one, or a few very similar tumor profiles. Our analysis pointed out that no obvious association exists between chemical/mode of action class, and tumor profile. On the contrary, none of these classes induces a single tumor or pattern of tumors, but rather it appears that each class produces tumors at a wide range of sites. This suggests that an important determinant of the differences in tumor profile are the events that surround the ultimate mechanism of interaction with DNA.