Predicting Modes of Toxic Action from Chemical Structure: Acute Toxicity in the Fathead Minnow (Pimephales Promelas) Steven P
Total Page:16
File Type:pdf, Size:1020Kb
Iowa State University From the SelectedWorks of Steven P. Bradbury May, 1997 Predicting Modes of Toxic Action from Chemical Structure: Acute Toxicity in the Fathead Minnow (Pimephales Promelas) Steven P. Bradbury Christine L. Russom Steven J. Broderius Dean E. Hammermeister Robert A. Drummond Available at: https://works.bepress.com/steven_bradbury/25/ Environmental Toxicology and Chemistry, Vol. 16, No. 5, pp. 948±967, 1997 Printed in the USA 0730-7268/97 $6.00 1 .00 PREDICTING MODES OF TOXIC ACTION FROM CHEMICAL STRUCTURE: ACUTE TOXICITY IN THE FATHEAD MINNOW (PIMEPHALES PROMELAS) CHRISTINE L. RUSSOM,* STEVEN P. B RADBURY,STEVEN J. BRODERIUS, DEAN E. HAMMERMEISTER and ROBERT A. DRUMMOND U.S. Environmental Protection Agency, National Health and Environmental Effects Research Laboratory, Mid-Continent Ecology Division, 6201 Congdon Boulevard, Duluth, Minnesota 55804 (Received 7 May 1996; Accepted 23 September 1996) AbstractÐIn the ®eld of aquatic toxicology, quantitative structure±activity relationships (QSARs) have developed as scienti®cally credible models for predicting the toxicity of chemicals when little or no empirical data are available. In recent years, there has been an evolution of QSAR development and application from that of a chemical-class perspective to one that is more consistent with assumptions regarding modes of toxic action. The objective of this research was to develop procedures that relate modes of acute toxic action in the fathead minnow (Pimephales promelas) to chemical structures and properties. An empirically derived database for diverse chemical structures of acute toxicity and corresponding modes of toxic action was developed through joint toxic action studies, the establishment of toxicodynamic pro®les, and behavioral and dose±response interpretation of 96-h LC50 tests. Using the results from these efforts, as well as principles in the toxicological literature, approximately 600 chemicals were classi®ed as narcotics (three distinct groups), oxidative phosphorylation uncouplers, respiratory inhibitors, electrophiles/proelec- trophiles, acetylcholinesterase inhibitors, or central nervous system seizure agents. Using this data set, a computer-based expert system has been established whereby chemical structures are associated with likely modes of toxic action and, when available, corresponding QSARs. KeywordsÐQuantitative structure±activity relationships Expert systems Toxic action mode Aquatic toxicology Pimephales promela INTRODUCTION years has challenged the notion that similarity in mode of toxic action is necessarily related to typical chemical classi®cation In the ®eld of aquatic toxicology, ®rst-generation quanti- schemes [4±12]. As a consequence, QSAR development and tative structure±activity relationships (QSARs) have devel- application have been evolving from a chemical class per- oped as scienti®cally credible tools for predicting the acute spective to one that is more consistent with assumptions re- toxicity of chemicals when little or no empirical data are avail- garding modes of toxic action [13,14]. The use of mode of able [1]. In part, the success in establishing these QSARs is action-based QSARs, therefore, requires an appreciation of dependent upon well-de®ned and quanti®able toxicity end- both toxic mechanisms and the critical structural characteris- points, such as the 96-h LC50 value for the fathead minnow tics and properties of a chemical that governs its action by a (Pimephales promelas). Although the accuracy of toxic po- speci®c mechanism. tency predictions from QSARs continues to improve, there Establishment of toxicologically credible techniques to as- remains signi®cant uncertainty in the appropriate selection of sess mode of toxic action from chemical structure requires QSARs for predicting adverse effects. The proper application toxicodynamic knowledge bases that are clearly de®ned with and continued acceptance of these predictive toxicology tech- regard to exposure regimes and biological models/endpoints, niques, therefore, require methods to systematically assign and based on compounds that adequately span a chemical prop- chemicals to appropriate QSARs or analogues. This critical erty space anticipated for future applications [13]. A typical process in the use of predictive techniques represents a major endpoint used in initial effect assessments for aquatic organ- area of uncertainty in ecological risk assessments for chemical isms is the 96-h LC50 value for the fathead minnow. Collab- stressors [2,3], where errors in QSAR selections can result in orative research undertaken through our laboratory has estab- 10- to 1,000-fold errors in toxic potency estimates. lished a database for this endpoint that contains values for Traditionally, the selection of structural analogues or approximately 600 chemicals [15±19] and which serves as a QSARs has been based on the assumption that compounds foundation for the development of QSARs. The chemical set from the same ``chemical class'' should behave in a toxico- chosen for study was based on an assessment of the U.S. logically similar manner. Although this working hypothesis industrial chemical inventory of discrete organic chemicals seems reasonable, the identi®cation of chemical classes is [20]. problematic, and research completed over the past several Using this chemical data set, we describe an investigation that relates modes of acute toxic action in the fathead minnow * To whom correspondence may be addressed. to chemical structures and properties. An empirically derived Electronic copies of the data sets used in this study are available upon request. Mention of trade names, models, or commercial prod- database of chemical structures and corresponding modes of ucts does not constitute endorsement or recommendation for use by toxic action was developed through joint toxic action studies, the U.S. Environmental Protection Agency. the establishment of toxicodynamic pro®les, and behavioral 948 Predicting modes of acute toxic action from chemical structure Environ. Toxicol. Chem. 16, 1997 949 and dose±response interpretation of 96-h LC50 tests. Using reference toxicants such as octanol, pentachlorophenol, phe- the results from these efforts, as well as principles in the tox- nol, and carbaryl, which results in a database of 753 tests. icological literature, the chemicals were classi®ed as either A detailed description of the biological and chemical test narcotics, oxidative phosphorylation uncouplers, respiratory protocols used for these exposures has been published [15,16]. inhibitors, electrophiles/proelectrophiles, acetylcholinesterase Brie¯y, all tests were conducted using Lake Superior water at (AChE) inhibitors, or central nervous system (CNS) seizure 25 6 18C. Aqueous toxicant concentrations were measured in agents. Using this data set, a computer-based expert system all tests with quality assurance criteria requiring 80% agree- has been established whereby chemical structures are associ- ment between duplicate samples and 90 to 110% spike recov- ated with likely modes of toxic action and, when there are ery. Flow-through exposures were conducted using cycling suf®cient data, corresponding QSARs. proportional [31], modi®ed Benoit [32], or electronic [16] di- luters. Tests conducted on the Benoit and electronic diluters MATERIALS AND METHODS did not have replicate tank exposures. Median lethal concen- trations (LC50s) were calculated using the Trimmed Spear- To develop an expert system to predict acute mode of toxic man±Karber Method, with 95% con®dence intervals being cal- action from chemical structure ®rst requires a knowledge base culated when possible [33]. from which rules can be derived. In this study, the knowledge Dose±response assessments. The change of LC50 values base was derived from analyses of the chemicals in the fathead over time (LC50 ratio) and the ratio of measured 96-h LC50 minnow acute toxicity database [15±19]. Based on empirical values to those predicted from a baseline narcosis (narcosis I) mode of action assessments, a knowledge base was established QSAR [21] were used as supportive data for assessing potential from which substructural fragments of chemicals were asso- modes of action. To characterize time until death, a ratio of ciated with modes of toxic action. In turn, these rules were the 24-h LC50 to the 96-h LC50 was calculated for each 96-h written in Fortran and linked to mode of action-speci®c fathead minnow exposure using Equation 1: QSARs, when available. Each chemical was classi®ed into one of eight modes of LC50 ratio 5 24 h LC50/96 h LC50 (1) action: base-line narcosis or narcosis I [21], polar narcosis or For exposures where an LC50 did not occur in the ®rst 24 h, narcosis II [12], ester narcosis or narcosis III [11,22], oxidative ratios of the 48- to 96- or 72- to 96-h values were used. In- phosphorylation uncoupling [23], respiratory inhibition [24], stances where an LC50 was not obtained until 96 h were noted. electrophile/proelectrophile reactivity [25,26], AChE inhibi- Ratios that were approximately 1.0 were considered indicative tion [24], or several mechanisms of CNS seizure responses of narcosis I, whereas ratios greater than 2 or cases where [27]. For the purpose of this paper, mode of toxic action should LC50s were not achieved in the ®rst 24 h were generally not necessarily be construed to impart the sense of distinct considered indicative of a different mode of action. It was molecular mechanisms. For example, CNS seizure agents and noted, however, that compounds