2684
J. Phys. Chem. A 2010, 114, 2684–2688
Correlation of the Photolysis Half-Lives of Polychlorinated Dibenzo-p-dioxins and Dibenzofurans with Molecular Structure Alan R. Katritzky,*,† Svetoslav H. Slavov,† Iva B. Stoyanova-Slavova,† and Mati Karelson‡ Center for Heterocyclic Compounds, Department of Chemistry, UniVersity of Florida, GainesVille, Florida 32611 and Department of Chemistry, Tallinn UniVersity of Technology, Ehitajate tee 5, Tallinn 19086, Estonia ReceiVed: NoVember 2, 2009; ReVised Manuscript ReceiVed: December 4, 2009
The photolysis half-lives of 70 polychlorinated dibenzo-p-dioxins and dibenzofurans are correlated with their molecular structures by a QSPR model (R2 ) 0.72) comprising three bond-energy-related descriptors. The photodegradation depends on the stability of the aromatic system and the C-O and C-C bond strengths. Model validation utilized leave-one-out (R2 ) 0.69), leave-many-out (R2 ) 0.72), and scrambling (R2 ) 0.19) procedures. Our results allow estimation of the photolysis half-lives of the remaining possible 140 PCDDs and PCDFs congeners. Introduction Polychlorinated dibenzo-p-dioxins and dibenzofurans (PCDD/Fs) are unwanted byproducts of a variety of industrial processes (including herbicide production, metallurgical processes, and bleaching processes in the pulp and paper industry) and incineration of carbonaceous materials containing chlorine, such as PVC plastic. PCDD/Fs are highly lipophilic which, combined with their chemical stability, result in a rapid accumulation in both organisms and organic phases of soils or sediments and hence behave as persistent organic pollutants.1 Their ubiquity and the strong toxicity of some congeners (such as 2,3,7,8substituted PCDD/Fs) identify PCDD/Fs as high-priority contaminants of potential risk to human health.2,3 Toxic responses induced in laboratory animals include thymic atrophy, reproductive changes, weight loss, impairment of immune response, and disruption of the endocrine system causing thyroid dysfunction, resulting in a decrease of T4 in plasma and consequent developmental problems, including ovarian growth impairment, etc.4,5 Photodegradation is an important and often dominant abiotic environmental process leading to degradation of PCDD/Fs,6–8 occurring in surface waters, soil, and the atmosphere.9,10 A study of the photodegradation of 2,3,7,8-TCDD in aqueous solution under sunlight irradiation11 found that photolysis in water had a half-life (t1/2) of about 4-5 days at 40°N latitude during the summer; it was concluded that C-O rather than C-Cl cleavage was a major route for PCDDs photodegradation in water. However, photolytic degradation of PCDDs in hexane has been reported to form lower chlorinated congeners through C-Cl bond cleavage.11–13 Several studies of the photodegradation of PCDD/Fs adsorbed on fly ash14,15 suggested that the photodegradation of PCDD/Fs bound to atmospheric particles was not a significant mechanism for elimination of these compounds from the environment. Radiation experiments at 254 nm16 demonstrate that photodegradation of PCDD/Fs on a plain surface may eliminate more than 89% of the contaminants. Photodegradation on vegetation significantly impairs the transfer of PCDD/ Fs from the atmosphere into the terrestrial food web.7,8,17 Photolysis of PCDD/Fs on plants probably takes place predominantly in the cuticle and especially the cuticular wax coating of the leaves.18 As a quantitative measure of the efficiency of the photochemical degradation and thus for assessment of the environmental risk of these * To whom correspondence should be addressed. E-mail: katritzky@ chem.ufl.edu. † University of Florida. ‡ Tallinn University of Technology.
chemicals the photolysis half-life (t1/2) is often used. However, the availability of t1/2 data is rather scarce.9 Thus, QSPR models relating photolysis data to other physicochemical properties or molecular descriptors have been developed.19–23 Tysklind et al.19 reported a QSPR model for the photolysis half-lives of PCDDs in different organic solvents. Chen et al.21,22 proposed QSPR models of photolysis of PCDDs dissolved in water-acetonitrile solution and in cuticular wax exposed to sunlight. Niu et al.23 proposed a five-parameter PLS model for a set of 70 PCDD/Fs emphasizing the importance of the number of Cl atoms and the absolute hardness of these compounds. Recently, ab initio and DFT studies (most using second-order perturbative approach (CASSCF) or HF/DFT hybrid functionals like B3LYP) on the formation, chlorination, dechlorination, and destruction as well as vibrational frequencies and electronic spectra of dibenzop-dioxins and dibenzofurans were reported.24–28 Searching for new insight into the influence of molecular structure on the photolysis half-lives we treated this data set23 by the BMLR algorithm implemented in CODESSA-Pro, which provided us with a simple multilinear QSPR model for estimation of the t1/2 values of PCDDIFs, and revealed structural characteristics explaining much of the variance of the experimental half-life data. Data Set Recently, Niu et al.29 studied the photolysis of 28 PCDDs and 42 PCDFs adsorbed on spruce. The PCDD/F-contaminated spruce trees were exposed to full sunlight in Oberschleissheim, Munich, Germany (latitude 48.2°N) in July 2001 from 9:00 a.m. to 6:00 p.m. under clear sky conditions, and half-lives of the PCDD/Fs were measured. In the present manuscript we describe our reexamination of this data set (see Table 1). The probability density function (see Figure 1) shows that the distribution of the t1/2 values deviates significantly from normal. Thus, to improve the statistical distribution of the data a logarithmic function was applied (Log t1/2). However, in zeroth- or (pseudo) first-order reactions, t1/2 is inversely proportional to the rate constant, k, and thus to the change in the free energy ∆G‡ during the reaction
Log t1/2 ≈ -Log k ≈ ∆Gq Due to the above relationship and the specific nature of the descriptors used in the model of Table 2 (all are energy related) our model is simply an example of a linear free energy relationship (LFER).
10.1021/jp910470e 2010 American Chemical Society Published on Web 01/29/2010
Correlation of the Photolysis Half-Lives TABLE 1: Experimental and Predicted Photolysis Half-Lives
J. Phys. Chem. A, Vol. 114, No. 7, 2010 2685
2686
J. Phys. Chem. A, Vol. 114, No. 7, 2010
Katritzky et al.
TABLE 1: Continued
Computational Procedure Three-dimensional conversions and preoptimization were performed using the molecular mechanics (MM+) implemented in the HyperChem 7.5 package.30 Final geometry optimization of the molecules was carried out by using the semiempirical quantum-mechanical AM1 parametrization.31 The optimized
geometries were loaded into CODESSA PRO software,32 and more than 800 theoretical descriptors were calculated. The Best Multilinear Regression method (BMLR)33 encoded in CODESSA-PRO software was used to select significant descriptors for building multilinear QSAR models.18–23 Using subsets of noncollinear descriptors, the BMLR stepwise regres-
Correlation of the Photolysis Half-Lives
J. Phys. Chem. A, Vol. 114, No. 7, 2010 2687
Figure 1. Statistical distribution of the half-lives (t1/2 and Log t1/2).
TABLE 2: Best Three-Parameter BMLR Modela no.
X
0 1
–35.80 –0.7461
∆X
t test
name of descriptor
7.110 –5.035 intercept 0.06995 –10.67 total molecular 2-center resonance energy/no. of atoms 0.5552 0.1550 3.583 min resonance energy for bond C–O 0.06330 0.02132 2.969 min e–n attraction for bond C–C
2 3 a
N ) 70; R2 ) 0.720; R2cv ) 0.687; F ) 56.62; s2 ) 0.00333.
sion algorithm generates the best n-parameter regression equations (n g 2) based on the highest R2 and F values obtained in the process of calculations.33–35 During the BMLR process the descriptor scales are normalized and centered automatically, and the final result is given in natural scales. This procedure ensures the best representations of the property in a given descriptor space. To avoid statistically unreliable correlations arising from mutually intercorrelated variables, the BMLR rejects descriptors with intercorrelation coefficients larger than a certain threshold value. During the stepwise regression procedure, an important stage in developing successive QSPRs is when to stop adding independent variables to the model. Thus, the addition of descriptors to the regression equations was monitored during the BMLR. When no significant improvement of the statistical parameters s, F, and especially R2 was observed, the current model was considered optimum. Results and Discussion QSPR models with up to 5 descriptors were generated. However, a model with three descriptors was preferred on the grounds of simplicity (Occam’s razor), Figure 2. All of the descriptors in Table 2 are quantum-chemical in origin. According to the Student’s t test their significance decreases in the following order: total molecular two-center resonance energy/ no. of atoms > min resonance energy for bond C-O > minimum e-n attraction for bond C-C. The most significant descriptor, total molecular two-center resonance energy/no. of atoms, emphasizes the importance of the resonance energy (expressed by the resonance integral, β), that of the overlapping (expressed by the overlap integral, S), and the bond order for the photolysis process. This descriptor can be regarded as a measure of the stability of the aromatic system of PCDD/Fs, which depends on the number and orientation of the Cl substituents. Descriptors 2 and 3 of Table
Figure 2. Predicted vs observed Log t1/2 values.
TABLE 3: Results from Internal Validation of the Main QSPR Model training set
N
R2(Fit) R2cv(Fit) S2(Fit) test set
N
R2(Pred)
A+B B+C A+C average
47 46 47
0.731 0.708 0.731 0.723
23 24 23
0.710 0.744 0.690 0.712
0.681 0.656 0.681 0.673
0.0033 0.0034 0.0035 0.0034
C A B
2 are related, respectively, to the C-O and C-C bond strengths. The regression coefficients of both of these descriptors are as expected positive, a consequence of the direct dependence of the photolysis half-lives on the stability of the C-O and C-C bonds. The presence of descriptor 3, the “minimum resonance energy for bond C-O”, in the model supports the hypothesis of Dulin et al.11 that C-O rather than C-Cl cleavage is the major route for the photodegradation of PCDDs. For validation we used (a) the leave-many-out ABC procedure (shown in Table 3) and (b) a scrambling procedure in which the model was fitted to randomly reordered log(t1/2) values and then compared with the one obtained for the actual activities.36 Twenty randomizations, resulting in average R2 ) 0.194 (ranging from 0.174 to 0.0.207), were performed. The substantial difference between the actual R2 and the averaged R2 from the scrambling procedure indicates the stability of the model and the absence of chance correlations. Several authors23,37,38 encountered great difficulty in developing QSPR models able to predict the photolysis rate constants
2688
J. Phys. Chem. A, Vol. 114, No. 7, 2010
from molecular structure and suggested that t1/2 depends strongly on the matrix sorbing the compounds which often shields the effect of different molecular structure. For the currently used set of 70 PCDD/Fs, Niu et al.23 proposed a complex 17-descriptor PLS model with R2 ) 0.643. They also separated the PCDDs from PCDFs and for the reduced set of 42 PCDFs proposed a 5-descriptor model with R2 ) 0.740. However, both ELUMO - EHOMO and (ELUMO - EHOMO)2 descriptors were used simultaneously, which raises questions about the validity of the model. According to Table 4 of ref 23 all five descriptors are highly intercorrelated with R2 g 0.959. Using a similar technique, Lu et al.39 proposed a logarithmic t1/2 PLS model for a training set of 11 PAHs involving 10 independent variables. However, this model has virtually no statistical degrees of freedom and thus suffers from serious overfitting because of the use of 10 descriptors for only 11 data points. Except for the two above, we were unable to locate any QSPR models concerning the photolysis half-lives of PCDD/Fs, PAHs, or similar compounds with reliability sufficient for accurate predictions. Since only one-third of all possible 75 PCDD and 135 PCDF congeners have experimentally measured photolysis half-lives, we believe that our model will advance environmental studies of the pollution effects caused by PCDD/Fs. To summarize, the advantages of the currently proposed model are as follows: (1) its generality, it describes the photolysis half-lives of both PCDDs and PCDFs; (2) its simplicity, it involves only 3 molecular descriptors in a multilinear model; (3) its easy interpretability, all descriptors have clear physicochemical meaning directly related to the nature of photolysis; (4) its reliability, the validation of the model using three different test procedures all provided highly satisfactory statistical parameters. Conclusions A multilinear three-parameter QSPR model for the photolysis half-lives of 70 PCDD/Fs was developed. An interpretation of each quantum-chemical descriptor from the viewpoint of the physicochemical nature of the photolysis was carried out. It was found that the half-lives depend exclusively on the stability of the PCDD/Fs aromatic system and both the C-O and the C-C bonds strength. The model was tested extensively and compared to earlier reported models. The advantages of the currently proposed model lie in its generality, simplicity, easy interpretability, and reliability. Acknowledgment. The authors gratefully acknowledge support by the Kenan Trust, USA (to Prof. A. R. Katritzky) and the European Regional Development Fund through the Center of Excellence in Chemical Biology and by an Estonian target financing grant SF0140031As09 (to Prof. M. Karelson). References and Notes (1) Choudhary, G.; Keith, L. H.; Rappe, C. Chlorinated Dioxins and Dibenzofurans in the Total EnVironment; Butterworth: Boston, 1983; p 416.
Katritzky et al. (2) Mocarelli, P.; Brambilla, P.; Gerthoux, P. M.; Patterson, D.G. J.; Needham, L. L. Lancet 1996, 348. (3) Boening, D. W. Ecotoxicol. EnViron. Safety 1998, 39, 155. (4) Dı´az-Ferrero, J.; Rodrı´guez-Larena, M. C.; Comellas, L.; Jime´nez, B. Trends Anal. Chem. 1997, 16, 563. (5) Bosveld, A. T. C.; Van den Berg, M. Fresenius’ J. Anal. Chem. 1994, 348, 106. (6) Crosby, D. G.; Wong, A. S.; Plimmer, J. R.; Woolson, E. A. Science 1971, 173, 748. (7) Crosby, D. G.; Wong, A. S. Science 1977, 195, 1337. (8) McCrady, J. K.; Maggard, S. P. EnViron. Sci. Technol. 1993, 27, 343. (9) Larson, R. A.; Weber, E. J. Reaction Mechanisms in EnVironmental Organic Chemistry; Lewis Publishers, CRC Press, Inc.: Ann Arbor, 1994; p 359. (10) Mill, T. Chemosphere 1999, 38, 1379. (11) Dulin, D.; Drossman, H.; Mill, T. EnViron. Sci. Technol. 1988, 20, 72. (12) Dobbs, A. J.; Grant, C. Nature 1979, 278, 163. (13) Dung, M. H.; O’Keefe, P. W. EnViron. Sci. Technol. 1994, 28, 549. (14) Tysklind, M.; Rappe, C. Chemosphere 1991, 23, 1365. (15) Koester, C. J.; Hites, R. A. EnViron. Sci. Technol. 1992, 26, 502. (16) Borwitzky, H.; Schramm, K.-W. Chemosphere 1991, 22, 485. (17) Welsch-Pausch, K.; McLachlan, M. S. Organohalogen Compd. 1995, 24, 509. (18) Schuler, F.; Schmid, P.; Schlatter, C. H. Chemosphere 1998, 36, 21. (19) Tysklind, M.; Lundgren, K.; Rappe, C.; Eriksson, L.; Jonsson, J.; Sjostrom, M.; Ahlborg, U. G. EnVrion. Sci. Technol. 1992, 26, 1023. (20) Chen, J. W.; Peijnenburg, W. J. G. M.; Quan, X.; Chen, S.; Martens, D.; Schramm, K.-W.; Kettrup, A. EnViron. Pollut. 2001, 114, 137. (21) Chen, J. W.; Quan, X.; Peijnenburg, W. J. G. M.; Yang, F. L. Chemosphere 2001, 43, 235. (22) Chen, J. W.; Quan, X.; Yang, F. L.; Peijnenburg, W. J. G. M. Sci. Total EnViron. 2001, 269, 163. (23) Niu, J.; Huang, L.; Chen, J.; Yu, G.; Schramm, K.-W. Chemosphere 2005, 58, 917. (24) Altarawneh, M.; Dlugogorski, B. Z.; Kennedy, E. M.; Mackie, J. C. Prog. Energy Combust. Sci. 2009, 35, 245. (25) Altarawneh, M.; Carrizo, D.; Ziolkowski, A.; Kennedy, E. M.; Dlugogorski, B. Z.; Mackie, J. C. Chemosphere 2009, 74, 1435. (26) Sun, X. M.; Sun, T. L.; Zhang, Q. Z.; Wang, W. Sci. Total EnViron. 2008, 402, 123. (27) Ljubic, I.; Sabljic, A. J. Phys. Chem. A 2006, 110, 4524. (28) Arulmozhiraja, S.; Morita, M. J. Phys. Chem. A 2004, 108, 3499. (29) Niu, J.; Chen, J.; Henkelmann, B.; Quan, X.; Yang, F.; Kettrup, A.; Schramm, K.-W. Chemosphere 2003, 50, 1217. (30) Hyperchem, v. 7.5, 2003; Hypercube Inc.: Gainesville, FL; www.hyper. com. (31) Dewar, M. J. S.; Zoebisch, E. G.; Healy, E. F.; Stewart, J. J. P. J. Am. Chem. Soc. 1985, 107, 3902. (32) CODESSA PRO Software; University of Florida: Gainesville, FL, 2002; www.codessa-pro.com. (33) Karelson, M. Molecular Descriptors in QSAR/QSPR; WileyInterscience: New York, 2000. (34) Todeschini, R., Consonni, V. Molecular Descriptors for Chemoinformatics, 2nd ed.; Wiley-VCH: Weinheim, 2009. (35) Topological Indices and Related Descriptors in QSAR and QSPR; Devillers, J., Balaban, A. T., Eds.; Gordon and Breach: The Netherlands, 1999. (36) Eriksson, L.; Jaworska, J.; Worth, A. P.; Cronin, M. T. D.; McDowell, R. M.; Gramatica, P. EnViron. Health Perspect. 2003, 111, 1361. (37) Yokley, R. A.; Garrison, A. A.; Wehry, E. L.; Mamantov, G. EnViron. Sci. Technol. 1986, 20, 86. (38) Behymer, T. D.; Hites, R. A. EnViron. Sci. Technol. 1988, 22, 1311. (39) Lu, G.-N.; Zhi Dang, Z.; Tao, X.-Q.; Yang, C.; Yi, X.-Y. Sci. Total EnViron. 2007, 373, 289.
JP910470E
