VIII SIMPOSIO INTERNAZIONALE ICPBR

HAZARDS OF PESTICIDES TO BEES

Bologna, 4-6 Settembre 2002

Metodologie dei test e studi sugli effetti dei pesticidi sulle api

Modeling the acute toxicity of pesticides to Apis mellifera

James Devillers (1) , Minh-Hà Pham-Delègue (2), Axel Decourtye (2), Hélène Budzinski (3), Sophie Cluzeau (4) and Gilbert Maurin (4)

(1) CTIS, 3 Chemin de la Gravière, 69140 Rillieux La Pape, France. E-mail: j.devillers@ctis.fr
(2) Laboratoire de Neurobiologie Comparée des Invertébrés, INRA, BP 23, 91440 Bures-sur-Yvette, France
(3) LPTC, UMR 5472 CNRS, Université de Bordeaux I, 351 Cours de la Libération, 33405 Talence CEDEX, France
(4) ACTA, Association de Coordination Technique Agricole, 149 rue de Bercy, 75595 PARIS CEDEX 12, France

Investigations into the development and use of Quantitative Structure-Activity Relationship (QSAR) models [1] to rapidly predict the ecotoxicity of xenobiotics from their molecular structure and/or physicochemical properties have increased dramatically over the past decades in order to save time and money in the design of safer chemicals. These QSAR models are now integrated in most of the methodological frameworks designed for estimating the environmental hazard and risk of organic chemicals. Surprisingly, while numerous QSAR models are available for estimating the toxicity of chemicals to various Arthropods, the number of models derived on the honey bee is very scarce. Indeed, only Vighi et al. [2], more than ten years ago, proposed a structure-toxicity model for estimating the acute toxicity of organophosphorus pesticides to Apis mellifera. Their regression equation was derived from only 15 pesticides and included six molecular descriptors.

Consequently, the aim of our study was to propose a more powerful model allowing to simulate the acute toxicity of all the families of pesticides to the honey bee. The model was derived from a training set of 89 chemicals and tested on an external testing set of 11 pesticides. Pesticides were described by means of autocorrelation descriptors [3] encoding lipophilicity, molar refractivity and the H-bonding acceptor ability of the pesticides. A three-layer feedforward neural network trained by the back-propagation algorithm4 was used as statistical engine for deriving the QSAR model. The root mean square residual values for the training and testing sets were 0.430 and 0.386, respectively. The practical interest of this original model was deeply discussed.

Acknowledgement: This study was granted by the French Ministry of the Environment (MATE-01133- Evaluation et réduction des risques liés à l’utilisation des pesticides).

References:

[1] KARCHER, W. & DEVILLERS, J. (1990). Practical Applications of Quantitative Structure-Activity Relationships (QSAR) in Environmental Chemistry and Toxicology. Kluwer Academic Publishers, Dordrecht, The Netherlands, p. 475.
[2] VIGHI, M., GARLANDA, M.M. and CALAMARI, D. (1991). QSARs for toxicity of organophosphorus pesticides to Daphnia and honeybees. In, QSAR in Environmental Toxicology-IV (J.L.M. Hermens and A. Opperhuizen, Eds.). Elsevier, Amsterdam, pp. 605-622.
[3] DEVILLERS, J. & BALABAN, A.T. (1999). Topological Indices and Related Descriptors in QSAR and QSPR. Gordon and Breach Science Publishers, The Netherlands, p. 811.
DEVILLERS, J. (1996). Neural Networks in QSAR and Drug Design. Academic Press, London, p. 284.