Predictive QSAR models for estimating ... - Science Direct

3 downloads 0 Views 296KB Size Report
program IRA) and discriminant analysis (ADAPT, 1989, program TILSQ) .... Leisinger, T., A.M. Cook, R. H/itter and J. N/isch (Eds), 1981. Microbial Degradation of.
The Science of the Total Environment, 109/110 (1991) 253-259 Elsevier Science Publishers B.V., Amsterdam

253

Predictive QSAR models for estimating biodegradation of aromatic compounds P. Degner, M. Nendza and W. Klein Fraunhofer-Institut j~r Umweltchemie und Okotoxikologie, D-5948 Schmallenberg-Grafschaft, FRG

ABSTRACT The development of valid structure biodegradation relationships (SBRs) is restricted by the lack of reproducible published data and by the considered endpoint of degradation data. A classification scheme is required for comparative evaluation of degradation data obtained by different test methods. SBRs based on substructure indicators are available for application to most compounds, but the reliability is still uncertain. SBRs based on physico-chemical parameters are only available for a few classes of compounds based on specific test methods. A combination of several SBRs covering the various transformation pathways provides a promising tool for predicting biodegradability. Two models describing biodegradation are introduced.

INTRODUCTION Persistence constitutes an important factor in evaluating the environmental hazard of chemicals. Several biotic and abiotic processes determine the residence time. Biodegradation has been recognized as a major pathway contributing to the transformation of xenobiotics. M o d e and rate of biodegradation depend on the chemical structure o f the substance as well as on the experimental conditions (Kawasaki, 1980). Biodegradability does not represent a well-defined endpoint with respect to mechanistic considerations. Various methods have been employed to discriminate between persistent and readily biodegradable compounds. Physico-chemical properties and substructure-related parameters have been used to evaluate microbial transformation by structure biodegradation relationships (SBRs). DATA ANALYTICAL METHODS Biodegradation data F o r harmonization o f test procedures, guidelines for standardized testing of chemicals have been published by the O E C D . 0048-9697/91/$03.50 © 1991 Elsevier Science Publishers B.V. All rights reserved

254

P. DEGNER ET AL.

Biodegradation data (O-consumption rates) for 112 disubstituted aromatic compounds (Kitano, 1987) estimated using the MITI degradation test procedure (OECD 301 C, 1984) were used for evaluating SBRs. Oxygen consumption >70% within 28 days characterizes readily degradable compounds; oxygen consumption < 70% characterizes compounds that are not readily degradable.

Chemical descriptors According to the proposed degradation pathways, different types of descriptors were examined. Physico-chemical and geometrical parameters were calculated using ADAPT software (ADAPT, 1989). Electronic descriptors were estimated with fully geometry-optimized molecules using MNDO procedures (modified neglect of diatomic overlap) in the AMPAC package (Dewar and Thiel, 1977). Countings of certain substructures were obtained from substructure keys.

Mathematical procedures For evaluating SBR models, multiple regression analysis (ADAPT, 1989, program IRA) and discriminant analysis (ADAPT, 1989, program TILSQ) were employed to examine parameters associated with biodegradability of aromatic compounds. SBR BASED ON STRUCTURAL FEATURES (MODEL 1)

Enhanced biodegradability of chemicals has been demonstrated in the presence of functional groups such as carboxyl-, hydroxyl-, and methyl-, and decreased degradability has been demonstrated in the presence of nitro-, amino-, cyano- and halogen- substituents (Kawasaki, 1980; Kobayashi, 1981; Paris et al., 1982; Pitter, 1985). Substructure counts for nine functional groups (Table 1) are used for evaluating a regression model discriminating readily degradable and not readily degradable compounds (ADAPT, 1989, program IRA). Applying regression model 1, a biodegradability factor (B) > 0 classifies a readily degradable chemical. SBR BASED ON ELECTRONIC PROPERTIES (MODEL 2)

Biodegradation rates depend on the interaction between microbial enzymes and substrate and are influenced by the substituents on the aromatic ring. Determining the electron density on the aromatic ring, substituents enhance

PREDICTIVE QSAR MODELS FOR ESTIMATING BIODEGRADATION OF AROMATIC COMPOUNDS

255

TABLE 1 Regression model 1 discriminating readily biodegradable and not readily biodegradable aromatic compounds using substructure coefficients given by Eqn (1). Applying regression model 1, a biodegradability factor (B) > 0 classifies a readily degradable compound; B < 0 classifies a not readily degradable c o m p o u n d Eqn ( 1 ) B

=

a * Fi + b , F 2

B

=

+ ...

n * F, + intercept

biodegradability factor

F~, F: . . . . .

number of substructure occurrences

a,b . . . . .

coefficients

Model 1 Substructure

Coefficient

OH-aryl CH3-aryl NH2-aryl CI-aryl Br-aryl NO2-aryl SO3-aryl COOH-aryl NC-aryl Intercept

+ + + +

0.08 0.06 0.26 0.50 0.57 0.46 0.61 0.10 0.38 0.29

or reduce the probability of enzymatic attack resulting in ring cleavage. Volume and radius of the substituents determine the steric fit to the interaction site. Model 2 was derived based on electronic and steric parameters. Five descriptors {(a) volume/molecular weight ratio (V/MW), (b) difference in energy between the highest occupied orbital and the lowest unoccupied orbital (HOLU), (c) heat of formation (HEAT), (d) ionization potential (IP), (e) centre of maximal superdelocalizability towards nucleophilic reactants [DN(max), Schfiiirmann, 1990]} were calculated using ADAPT software (ADAPT, 1989) and MNDO methods (Dewar and Thiel, 1977). Separation of readily degradable and not readily degradable aromatic compounds was achieved by discriminant analysis (ADAPT, 1989, program TILSQ). Biodegradability of the aromatic compounds is satisfactorily predicted using the discriminant equation model 2 (Table 2). Validation of models 1 and 2 with 112 aromatic compounds results in approximately 90% correct classification for both models (Figs 1 and 2). Correct classification for most aromatic compounds applying model 2 indicates a similar biodegradation pathway. Misclassification of aromatics

256

i,. D E G N E R ET AL.

TABLE 2 Discriminant model 2 discriminating readily biodegradable and not readily biodegradable aromatic compounds using the weight vector given by Eqn (2). Applyingmodel 2, a biodegradability factor (B) > 0 classifiesa readily degradable compound; B < 0 classifiesa not readily degradable compound Eqn(2) B = a*F l + b,F2 + . . . + n*Fn + intercept B /71, F2. . . . .

a,b. . . . . Model 2

= biodegradability factor normalized descriptor value factors of the weight vector

Weight vector: Factors

Parameter

Mean

Standard deviation

+ 0.64 - 0.30 - 0.32 + 0.60 - 0.19 - 0.05

V/MW HOLU HEAT IP DN(max) Intercept

3.74 9.00 - 11.62 9.73 - 0.53

0.48 0.40 55.52 0.79 0.02

containing sulphate groups by both models is an indication o f a deviating mode o f transformation: desulfonation is likely prior to aromatic ring cleavage for some compounds (Leisinger et al., 1981). Misclassification applying the substructural model is mainly a problem o f an insufficient number of substructure indicators available for classification in a small data base. Various compounds are characterized by the same substructure keys. F r o m a statistical point o f view, the substructure m i n i m u m occurrence o f five times in a data set is required before using as discrimination criteria. Opposite to physico-chemical descriptors (minimum: five chemicals per descriptor), the number o f substructure keys is not limited for a data set. Deriving substructure-based models from an extensive data base could result in a more reproducible classification. DISCUSSION Development o f predictive biodegradation models requires an extensive data set based on reliable biodegradation data. Neglect o f compounds representative for a particular pathway results in limited predictive power. When validating and applying correlations, the protocol used for the generation of the biological data has to be considered, as the biodegradability of chemicals depends on the experimental conditions. Biodegradability

257

PREDICTIVEQSAR MODELSFOR ESTIMATING BIODEGRADATIONOF AROMATICCOMPOUNDS

Model I pred.

reedy degradable

A

not ready

A a

z~

degradable

A A

A~A AX

A

A

A A

not ready degradable

reedy degradable

exp.

Fig. 1. Comparison of experiment (exp.) and predicted (pred.) biodegradability. Application of regression model l to the degradation data of 112 aromatic compounds results in approximately 90% correct classification (zx). The 13 compounds misclassified are denoted by ,.

measured according to OECD Guidelines represents an accepted endpoint useful for modeling and predicting ready degradability. From a statistical point of view, transformation kinetics data would best allow classification, but there is a lack of reproducible data. Different testing procedures may result in various non-uniform classifications for the same compounds, hence impeding any modeling effort. The application range of SBRs for predictive purposes varies, depending on whether substructure indicators or physico-chemical parameters are used for the modeling. Models based on physico-chemical descriptors are only valid for strictly homogeneous series with uniform degradation pathways. Prediction of biodegradation for diverse data sets requires a combination of numerous models based on physico-chemical parameters. The derivation of physico-chemical descriptor-based models with reproducible data is required. Application of substructure indicators is more general, as many different

258

P. DEGNER ET AL.

Model 2 pred. &

ready

deoredeble

A A ,

not ready

degradable

A ZEx

/x

not reedy degradable

ready degradable

exo.

Fig. 2. Comparison of experimental (exp.) and predicted (.pred.) biodegradability. Application of discriminant model 2 to the degradation data of 112 aromatic compounds results in approximately 90% correct classification (A). The 13 compounds misclassifiedare denoted by *.

pathways may be considered simultaneously. The prediction of various compounds becomes feasible, but a large number of degradation data has to be available. ACKNOWLEDGEMENT

This study was supported by the Federal Environmental Ministry and Federal Environmental Agency (BMU/UBA). REFERENCES ADAPT, 1989. ADAPT Computational Chemistry Software Rev. 3.0. Molecular Design Ltd, San Leandro, CA, USA. Dewar, M.J.S. and W. Thiel, 1977. Ground states of molecules. 38. The MNDO method. Approximations and parameters. J. Am. Chem. Soc., 99: 4899-4907. Kawasaki, M., 1980. Experience with the test scheme under the chemical control law of Japan:

PREDICTIVE QSAR MODELS FOR ESTIMATING BIODEGRADATION OF AROMATIC COMPOUNDS

259

an approach to structure-activity correlations. Ecotoxicol. Environ. Saf., 4: 444--454. Kitano, M., 1987. Chemicals Inspection & Testing Institute, Tokyo, Japan, personal communication. Kobayashi, K., 1981. Safety examination of existing chemicals - - Selection, testing, evaluation and regulation in Japan. In: Proc. Workshop Control of Existing Chemicals Under the Patronage of the Organisation for Economic Co-operation and Development, June 10-12, Reichstagsgeb/iude, Berlin. OECD, Paris, pp. 141-163. Leisinger, T., A.M. Cook, R. H/itter and J. N/isch (Eds), 1981. Microbial Degradation of Xenobiotics and recalcitrant compounds, FEMS Symp. 12. Academic Press, London. OECD 301C, 1984. Ready biodegradability: modified MITI test (1). Off. J. Eur. Communities L 251, 27: 199-211. Paris, D.F., N.L. Wolfe and W.C. Steen, 1982. Structure-activity relationships in microbial transformation of phenols. Appl. Environ. Microbiol., 44: 153-158. Pitter, P., 1985. Correlation of microbial degradation rates with the chemical structure. Acta Hydrochim. Hydrobiol., 13: 453-460. Sch/iiirmann, G., 1990. Quantitative structure-property relationships for the polarizability, solvatochromic parameters and lipophilicity. Quant. Struct.-Act. Relat., 9: 326-333.