QSARs in Predictive Environmental Toxicology Polar Narcosis Series: QSARs in Predictive Environmental Toxicology Series Editors: Joop L.M. Hermens, Ph.D. and Henk J.M. Verhaar, Ph.D., Research Institute of Toxicology,P.O. Box 80176, NL-3508 TD Utrecht, The Netherlands Articles published in this column: J.L.M. HE~MENS, H.J.M. VERHAAR:Preface - From Mechanistic Studies to Applications in Risk Assessment ESPR 3/2, pp. 96-98 (1996) E.M.J. VERBRUGGEN, W.M.G.M. VAN LOON, J.L.M. HERMENS: Hydrophobicity of Complex Organic Mixtures ESPR 3/3, pp. 163-168 (1996) D. GAMBERGER, D. HORVAT1C, S. SEKUSAK, A. SABLIJC: Applications of Experts' Judgement to Derive Structure-Biodegradation Relationships ESPR 3/4, pp. 224-228 (1996) E, RORIJE, L. ERIKSSON, H. VERBOOM, H.J.M. VERHAAR, J.L.M HERMENS, W.J.G.M. PEiJNENBURG: Predicting Reductive Transformation Rates of Halogenated Aliphatic Compounds Using Different QSAR Approaches ESPR 4/1, pp. 47-54 (1997) Polar Narcosis" Designing a Suitable Training Set for QSAR Studies Efiaut Urrestarazu Ramos, Wouter H.J. Vaes, Henk J.M. Verhaar, Joop L.M. Hermens Research Institute of Toxicology (RITOX), Utrecht University, I:O. Box. 80.176, NL-3508 TD Utrecht, The Netherlands Corresponding author: Efiaut Urrestarazu Ramos; e-maih enaut@ritox.dgk.ruu.nl Abstract Substituted phenols, anilines, pyridines and mononitrobenzenes can be classified as polar narcotics. These chemicals differ from non-po- lar narcotic compounds not only in their toxic potency (normalized by log Kow), but also in their Fish Acute Toxicity Syndrome pro- files, together suggesting a different mode of action. For 97 polar narcotics, which are not ionized under physiological conditions, 1 t physico-chemica[ and quantum chemical descriptors were calculated. Using principal component analysis, 91% of the total variance in this descriptor space could be explained by three principal components which were subsequently used as factors m a statistical design. Eleven compounds were selected based on a two- level full factorial design including three compounds near the center of the chemical domain (a 23+3 design). QSARs were developed for both the design set and the whole set of 63 polar narcotics for which guppy and/or fathead minnow data were available in the literature. Both QSARs, based on partial least squares regression (3 latent variables), resulted in good models (R2=0.96 and Q2=0.82; RZ=0.86 and Q2=0.83 respectively) and provided similar pseudo-regression coefficients. In addition, the model based on the design chemicals was able to predict the toxic- ity of the 63 compounds (R2=0.85). Models show that acute fish toxicity is determined by hydropho- bicity, HOMO-LUMO energy gap and hydrogen-bo~d acceptor ca- pacity. Keywords: Fish acute LC50; PCA; PLS; polar narcosis; QSAR; statistical design 1 Introduction Several major factors such as i) the increase in the number of chemicals produced and/or released (environmental fac- tor), ii) the expensive and time-consuming toxicity tests needed to assess these chemicals (economical factor), and iii) the number of test animals required for these tests (eth- ical factor) make predictive environmental toxicology a necessary approach. These three factors require that our knowledge about the toxic hazard of chemicals must be updated in a fast, cheap and ethical way; moreover, besides being accurate, we preferably must be able to make predic- tions for chemicals that have not been investigated or might not even exist yet. In the past years, Quantitative Structure-Activity Relation- ship (QSAR) techniques have been shown to constitute use- ful tools for this issue [1, 2]. However, since QSARs are mainly statistical models, modeling local variability of one aspect (toxicity, pharmacological efficiency, etc.) with local variability of another aspect (molecular or physico-chemi- cal descriptors) within a 'homologous' group [3], and es- pecially since their use is very relevant in policy making, care must be taken in their development so that their ap- plicability is well known. To this effect, an appropriate strategy has been described by ERIKSSONET AL. [4]. This strategy consists of six princi- pal steps: o') ESPR - Environ. Sci. & Pollut. Res. 4 (2) 83-90 (1997) o.) 9 ,~r n,~hlisher~ D-~C;g99 l.andsherm Germany