Prediction of octanol-air partition coefficients for polychlorinated biphenyls (PCBs) using 3D-QSAR models

Ecotoxicology and Environmental Safety 124 (2016) 202–212

Contents lists available at ScienceDirect

Ecotoxicology and Environmental Safety journal homepage: www.elsevier.com/locate/ecoenv

Prediction of octanol-air partition coefficients for polychlorinated biphenyls (PCBs) using 3D-QSAR models Ying Chen a,b, Xiaoyu Cai a,b, Long Jiang a,b, Yu Li a,b,n a b

Resource and Environment Institute of North China Electric Power University, Beijing 102206, China The State Key Laboratory of Regional Optimisation of Energy System, North China Electric Power University, Beijing 102206, China

art ic l e i nf o

a b s t r a c t

Article history: Received 25 June 2015 Received in revised form 6 October 2015 Accepted 20 October 2015

Based on the experimental data of octanol-air partition coefficients (KOA) for 19 polychlorinated biphenyl (PCB) congeners, two types of QSAR methods, comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA), are used to establish 3D-QSAR models using the structural parameters as independent variables and using log KOA values as the dependent variable with the Sybyl software to predict the KOA values of the remaining 190 PCB congeners. The whole data set (19 compounds) was divided into a training set (15 compounds) for model generation and a test set (4 compounds) for model validation. As a result, the cross-validation correlation coefficient (q2) obtained by the CoMFA and CoMSIA models (shuffled 12 times) was in the range of 0.825–0.969 (40.5), the correlation coefficient (r2) obtained was in the range of 0.957–1.000 ( 40.9), and the SEP (standard error of prediction) of test set was within the range of 0.070–0.617, indicating that the models were robust and predictive. Randomly selected from a set of models, CoMFA analysis revealed that the corresponding percentages of the variance explained by steric and electrostatic fields were 23.9% and 76.1%, respectively, while CoMSIA analysis by steric, electrostatic and hydrophobic fields were 0.6%, 92.6%, and 6.8%, respectively. The electrostatic field was determined as a primary factor governing the log KOA. The correlation analysis of the relationship between the number of Cl atoms and the average log KOA values of PCBs indicated that log KOA values gradually increased as the number of Cl atoms increased. Simultaneously, related studies on PCB detection in the Arctic and Antarctic areas revealed that higher log KOA values indicate a stronger PCB migration ability. From CoMFA and CoMSIA contour maps, log KOA decreased when substituents possessed electropositive groups at the 2-, 3-, 3′-, 5- and 6- positions, which could reduce the PCB migration ability. These results are expected to be beneficial in predicting log KOA values of PCB homologues and derivatives and in providing a theoretical foundation for further elucidation of the global migration behaviour of PCBs. & 2015 Elsevier Inc. All rights reserved.

Keywords: PCBs KOA 3D-QSAR CoMFA CoMSIA Migrate

1. Introduction Polychlorinated biphenyls (PCBs) are one of the most widespread and persistent groups of persistent organic pollutants (POPs) in the environment, including 209 congeners characterized by the number and position of the chlorine atoms on the biphenyl core. PCBs can harm human health and the ecological environment and are different from common organic pollutants due to their high stability, toxicity, environmental persistence, bioaccumulation, long-distance migration ability and other characteristics (Zhang et al., 2007; Alkhatib and Weigand, 2002). Therefore, studies on PCBs have received substantially more global attention. n Corresponding author at: Resource and Environment Institute of North China Electric Power University, Beijing 102206, China. Fax: þ 86 010 61773886. E-mail address: [email protected] (Y. Li).

http://dx.doi.org/10.1016/j.ecoenv.2015.10.024 0147-6513/& 2015 Elsevier Inc. All rights reserved.

Approximately 1.5 million tons of PCBs were produced globally in the form of complex mixtures serving as dielectric fluids in transformers and capacitors and as a plasticizer agent in paint and rubber sealants since the 1930s (Bidleman et al., 2010). In the 1970s, PCBs were banned globally because of their adverse effects on immunity, nerves and endocrine systems and, especially, because they are capable of being passed down to the next generations, resulting in further adverse effects (Zhang et al., 2011). The octanol-air partition coefficient (KOA), which is defined as the ratio of solute concentration in air versus octanol when the octanol-air system is at equilibrium, is a key physicochemical parameter for describing the partition of organic pollutants between the atmosphere and the environmental organic phase. Therefore, KOA has important significance for the evaluation of organic pollutants in the atmosphere (Liu et al., 2013), soil (Harner et al., 2001), plant (Platts and Abraham, 2000), and humans (Betterman et al., 2002), as well as in aerosol (Dachs and

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

Eisenreich, 2000; Wang et al., 2011) migration and allocation behaviours. Because of PCBs' long-range transport potential, they have been detected in remote areas that were devoid of human activities, such as the Arctic (Zhang et al., 2014), Antarctic (Marco et al., 2015), Tibetan Plateau (Zheng et al., 2012), Alps (Tato et al., 2011), and some high latitude areas, especially in the polar areas. Moreover, PCBs have been detected in a wide range of biological samples, including fish (Su et al., 2012), chicken egg yolks (Rawn et al., 2012) and in human samples, such as breast milk (Hassine et al., 2012), blood (Jotaki et al., 2011) and adipose tissue (Arrebola et al., 2010). Therefore, it is of critical importance to evaluate the properties and global mobility of PCBs among various compartments of the natural environment. QSAR model has a long history of development. For the method of 1D-QSAR, the affinity is correlated with global molecular properties of ligands, which is one value per property and ligand (pKa, log P, etc.) (Hopfinger, 1980). For the method of 2D-QSAR, the affinity is correlated with structural patterns (connectivity, 2D pharmacophore, etc.) without consideration of an explicit 3D representation of these properties (Hansch and Fljita, 1964; Fujita et al., 1964; Free and Wilson, 1964). And for the method of 3DQSAR, the affinity is correlated with the three-dimensional structure of the ligands (Crippen, 1979; Cramer et al., 1988; Cramer Iii et al., 1988; Klebe et al., 1994). Harner and Mackay (1995), as well as Harner and Bidleman (1996) have developed a generator column method to measure KOA values of 19 PCBs. Because the generator column method is time-consuming, Zhang et al. (1999) proposed a multicolumn method to estimate KOA for semi-volatile organic compounds. These experimental methods have many drawbacks, including the need for special equipment and samples, as well as large investments of money, time, and labour. Chen et al. (2002) utilized nine quantum chemical descriptors to construct the classical Hanschtype modeling and to predict the KOA values of PCBs. Comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) are two powerful methods in 3D-QSAR approach, which take the 3D-conformation property of compounds into consideration, can be helpful in exploring and visualizing useful structural information that influences the activity of compounds. Note that geometric and electronic descriptors depend on the 3D coordinates of the atoms. 3D-QSAR, which refers to the use of force field calculations to compute spatial properties of the threedimensional structure of compounds, provides valuable information on the forces and interactions of molecules (Cruciani, 2003; Langer and Bryant, 2008). 3D-QSAR models can aid the design of new beneficial compounds and may be useful in the screening of a large number of chemicals for migratory effects, as well as for gaining a deeper understanding of the migration mechanism (Salahinejad and Ghasemi, 2014). CoMFA and CoMSIA have been widely used to construct 3D models, which use 3D structure as descriptors. They overcome the limitations of the conventional 2D model in characterizing the relationship between property and structure and have a clearer physical meaning and more abundant information of the molecular field energy. The CoMFA method involves the generation of a common three-dimensional lattice around a set of molecules and calculation of the steric and electrostatic interaction energies at the lattice points (Cramer et al., 1988), while the CoMSIA method uses the similarity functions represented by Gaussian (Klebe et al., 1994). In this study, QSAR models were constructed with 3D descriptors according to the experimental values of log KOA for 19 PCBs congeners. Two types of QSAR methods, CoMFA and CoMSIA, are used to predict the log KOA values of the remaining 190 PCBs congeners and to investigate the relationship between the

203

structures of PCBs and their persistent migration ability. These results are expected to be beneficial in predicting the log KOA values of homologues and derivatives of PCBs and providing the theoretical basis for further elucidation of the global migration behaviour of PCBs.

2. Materials and methods 2.1. Data set For 19 PCBs, log KOA values were determined directly byHarner and Mackay (1995) and Harner and Bidleman (1996) at 293 K using a generator column method. To facilitate the QSAR analysis, the logarithm of KOA (log KOA) was taken as the index of PCB migration ability. The whole data set (containing 19 compounds) was divided in the ratio of 4:1 into a training set (containing 15 compounds) for 3D-QSAR model generation and a test set (containing 4 compounds) for model validation. The selection of the test set was determined by its ability to appropriately represent the structural diversity of the whole data set and to cover the range of log KOA values (Li et al., 2012a, 2012b) 2.2. Molecular modelling and alignment of PCBs The 3D-QSAR and molecular alignment were performed using Sybyl-x 2.0 molecular modelling software package (Tripos Inc., St. Louis, MO). The 3D structure of each compound in the data set was constructed using the Sketch Molecule module in Sybyl. The geometry of these compounds was subsequently optimized using Tripos force field (Clark et al., 1989) with the Gasteiger–Hückel charges (Gasteiger and Marsili, 1980). Repeated minimizations were performed by the Powell method with a maximum iteration of 10,000 to reach an energy convergence gradient value of 0.005 kJ mol  1. The minimized structures were used as initial conformations for molecular alignment. To derive the best possible 3D-QSAR statistical model, ligand-based alignment was employed in this study. In this process, the 2, 2′, 3, 4, 4′, 5, 5′-heptachlorobiphenyl (PCB-180) molecule with the largest log KOA was used as the template to align the other compounds using the Align Database command in Sybyl. All of the molecules can have a very good alignment. 2.3. CoMFA and CoMSIA analysis The steric and electrostatic potential fields for CoMFA were calculated at each lattice intersection of a regularly spaced grid of 2.0 A. A sp3 carbon atom served as the probe atom to calculate steric field and electrostatic field energies with contributions truncated to 30 kcal mol  1, and the other parameters were set at default values. For CoMSIA analysis, five descriptor fields (steric, electrostatic, hydrophobic, hydrogen bond-donor and hydrogen bond-acceptor) were considered. CoMSIA descriptors were derived by the same lattice box as that used in CoMFA. Compared with the CoMFA method, the CoMSIA method uses the similarity functions represented by Gaussian and increases hydrophobic fields, hydrogen bond-donor and hydrogen bond-acceptor fields. The results of the CoMSIA method show a relatively small influence from compound matching rules and can more intuitively explain a compound's quantitative structure-activity relationship. The use of CoMSIA can overcome inherent defects of CoMFA but does not necessarily gain better results (Xu et al., 2002). Therefore, in this study, the use of two methods (CoMFA and CoMSIA) can verify and supplement each other to gain reliable predicting models.

204

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

2.4. Partial least square (PLS) analyses

Table 1 Statistical parameters of the CoMFA and CoMSIA models.

The 3D-QSAR of CoMFA and CoMSIA was directly yielded by partial least squares (PLS) analyses in which the log KOA served as the dependent variable, and the 3D structure of PCBs served as the independent variable. The leave-one-out (LOO) cross-validation procedure was performed to determine the optimum number of components (n) and the highest cross-validation correlation coefficient (q2) for the correlation models. Simultaneously, noncross-validated analysis was performed, and the conventional multiple correlation coefficients (r2), standard error of estimate (SEE) and the Fisher test (F) values were calculated (Li et al., 2014). The 3D colour-coded contour maps expressed the relationship between log KOA and each molecular field.

The internal predictive ability and robustness of the developed models were evaluated by leave-one-out (LOO) cross-validation procedure. The values of q2and r2 can represent the internal predictive ability and robustness of the models. To measure the bias of 2 the original calculations (the average r2 ( rboot ) and SEE (SEEboot)), a bootstrapping analysis for 100 runs was performed. However, Golbraikh and Tropsha (2002) showed that q2 alone is not a good parameter to estimate the prediction capability of QSAR models and external validation should be applied. The purpose of the external validation is to evaluate how well the model developed using the training set generalizes to an independent test set. The external validation method, as the most valuable validation method, was applied to assess predictive ability of the obtained model (Golbraikh and Tropsha, 2002). The overall predictive ability of the CoMFA and CoMSIA models were externally validated by predicting the activity of independent test set compounds (the compounds not included in the original training set) (Liao et al., 2009; Li et al., 2013). The predictive ability of the 2 models were expressed by the predicted r2 ( rpred > 0. 6), the standard error of prediction (SEP) of test set, the explained var2 iance in prediction ( Q ext > 0. 5), and external standard deviation error of prediction (SDEPext) (Golbraikh and Tropsha, 2002). 2 was calculated by using the following equation: The rpred

PRESS SD

(1)

where SD means the sum of squared deviations between the experimental activities of the compounds in the test set and the average activity of the compounds in the training set, PRESS means the sum of squared deviations between experimental and predicted activity values. 2 The external Q ext for the test set is determined with the following equation: 2 Q ext =1−

~ )2 ∑ (yi − y i ∑ (yi − y¯ )2

(2)

where yi and y~i are the observed and the calculated response values, respectively; and y¯ is the averaged value for the response variable of the training set. The external standard deviation error of prediction (SDEPext) is determined with the following equation:

SDEPext =

n ~ )2 ∑i =ext (y − y 1 i i

next

n

r2

q2

SEE

F

S

E

H

D

A

CoMFA CoMSIA

12 4

1.000 0.987

0.91 0.9

0.030 0.149

1584.523 190.387

0.239 0.006

0.761 0.926

– 0.068

– 0

– 0

r2: correlation coefficient; q2: the cross-validated value; SEE: standard error of estimated; F: Fischer test value; S: steric; E: electrostatic; H: hydrophobic; D: hydrogen bond donor; A: hydrogen bond acceptor.

3. Results and discussion 3.1. The prediction of log KOA CoMFA and CoMSIA models were used to forecast 209 types of PCBs, and the experimental and predicted log KOA values and residual values for PCBs are given in Table A1 (see Appendix A).

2.5. CoMFA and CoMSIA models evaluation and validation

2 rpred =1−

Model

3.2. CoMFA model analysis based on the predicted KOA value of PCBs The results of the CoMFA model are summarized in Table 1. CoMFA models are considered reliable and acceptable if q2 was greater than 0.50 and r2 is greater than 0.90 (Golbraikh and Tropsha, 2002). Using 15 components, this model yields the optimum number of components n of 12, a cross-validated q2 of 0.91 (40.5), a non-crossvalidated r2 of 1.000 ( 40.9), an SEE of 0.030, and an Fvalue of 1584.523, proving a good correlation between the experimental and predicted log KOA values. These statistical indexes are reasonably high, indicating that the CoMFA model has a strong predictive ability. The robustness and stability of the model are also verified by bootstrap validation method for 100 runs. The 2 average r2 ( rboot ) and SEE (SEEboot) of these 100 analyses of CoMFA 2 model are 0.999 and 0.047, respectively. The higher rboot and lower SEEboot are demonstrations of the robustness of the built model. To eliminate the possibility of a chance correlation and to further test the robustness of the CoMFA model, the Yvector (log KOA values) was shuffled 12 times. As a result, the q2 values obtained were in the range of  0.059 to 0.085, the r2 values obtained were in the range of  0.006 to 0.030, indicating that the results from the CoMFA model were not due to chance correlation. The ratio of r2 and q2 (shuffled 12 times) was in the range of 3.1–20.5% ( o25%), indicating that the models were not overfitted (Leach, 2001). The predicted values of 190 PCBs and the relative errors are listed in Table A1. In the CoMFA model prediction, the relative errors of the test set (4 compounds) were 0%, 0.79%, 0.18% and 0%, respectively, which are acceptable for small base values. It is noted that in this model, the electrostatic field is found to make a higher contribution (76.1%) to the log KOA values of PCBs than that of the steric field (23.9%). Fig. 1 depicts the correlations between the observed and predicted log KOA values for the training and the test sets. Further analysis of the log KOA predicted by the CoMFA model revealed a fine linear dependence (R2 of 0.999) among experimental values and predicted values, which can be observed from the scatter plot of experimental versus predicted values of log KOA (Fig. 1). External validation was also carried out to further assess the 2 reliabilities and the predictive ability of the built model. The rpred 2 of 0.979, Q ext of 0.975 and SDEPext of 0.123 were achieved, verifying the good external predictive ability of the CoMFA model.

3.3. CoMSIA model analysis based on the predicted KOA value of PCBs

(3)

where yi and y~i are the observed and the calculated response values, respectively; and next is the is the number of test set.

For CoMSIA analysis, five descriptor fields (steric, electrostatic, hydrophobic, hydrogen bond-donor and hydrogen bond-acceptor) were considered. This CoMSIA model gives the optimum number

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212 11.5

CoMFA Model

11.0

205

CoMSIA Model

11.0

10.5 10.5

10.0

Predicted log K O A

Predi ct ed log KOA

10.0

9.5 9.0 8.5

Training set

8.0

Test set

9.5 9.0 8.5

Training set 8.0

Test set

7.5 7.5

7.0 7.0

6.5 6.5

7.0

7.5

8.0

8.5

9.0

9.5

10.0

10.5

11.0

Observied logK

6.5

7.0

7.5

8.0

8.5

9.0

Observied logK

OA

9.5

10.0

10.5

11.0

OA

Fig. 1. The plot of observed vs predicted log KOA values using CoMFA and CoMSIA models.

of components n of 12, a cross-validated q2 of 0.9 ( 40.5), a noncrossvalidated r2 of 0.987 ( 40.9), a standard error of estimation (SEE) of 0.149 and an Fvalue of 190.387, proving a good correlation between the experimental and predicted log KOA values. These results demonstrated the robustness and the statistical confidence 2 ) and SEE (SEEboot) of of the CoMSIA model. The average r2 ( rboot these 100 analyses of CoMSIA model were 0.997 and 0.072, re2 and lower SEEboot were demonstrations spectively. The higher rboot of the robustness of the built model. To eliminate the possibility of chance correlation and to further test the robustness of the CoMSIA model, the Yvector (log KOA values) was shuffled 12 times. As a result, the q2 values obtained were in the range of 0.049 to 0.045, and the r2 values obtained were in the range of 0.000– 0.008, indicating that the results from the CoMSIA model were not due to chance correlation. The ratio of r2 and q2 (shuffled 12 times) was in the range of 4.3–12.8% ( o25%), indicating that the models were not overfitted (Leach, 2001). The predicted values of 190 PCBs and the relative errors are listed in Table A1. In the CoMSIA model prediction, the relative errors of the test set (4 compounds) were 0.34%, 1.85%, 0.21% and 3.57%, respectively, which are acceptable for small base values. CoMSIA analysis revealed that the corresponding percentages of the variance explained by steric, electrostatic and hydrophobic fields were 0.6%, 92.6%, and 6.8%, respectively. However, hydrogen bond donor and acceptor fields contribute minimally to the log KOA value of PCBs when steric, electrostatic and hydrophobic fields were introduced. The results reveal that electrostatic interaction was the major contribution to the log KOA of PCBs. An external test set including 4 compounds was used to further evaluate the reliability and applicability of the CoMSIA model. The correlations between the experimental and predicted log KOA values are depicted in Fig. 1. Further analysis of the log KOA predicted by the CoMSIA model revealed a fine linear dependence (R2 of 0.985) among experimental values and predicted values, which can be observed from the scatter plot of experimental versus predicted values of log KOA (Fig. 1). External validation was also carried out to further assess the 2 reliabilities and the predictive ability of the built model. The rpred 2 of 0.927, Q ext of 0.914 and SDEPext of 0.229 were achieved, verifying the good external predictive ability of the CoMSIA model.

The correlation analysis of the relationship between the log KOA predicted by the CoMFA and CoMSIA models and the log KOA values of PCBs in the literatures (Zhang et al., 1999,; Chen et al., 2002) revealed the fine linear dependence (R2 of 0.9018, 0.9223, 0.9391 and 0.95127, respectively), which indirectly indicated the prediction of 190 PCBs was validated. 3.4. CoMFA and CoMSIA contour analysis To visualize the field effects on the target compounds in 3D space, the contour diagrams of the final CoMFA and CoMSIA models are shown in Fig. 2. The diagrams could be helpful in identifying the important regions where variations in the steric, electrostatic, hydrophobic, hydrogen bond donor and acceptor fields around the compound explain differences in the log KOA value of PCBs. The contour maps of all of the fields were constructed using the field type “stdev*-coeff” (the standard deviation and the coefficient) with default values of 80% favoured and 20% disfavoured contributions. To aid in visualization, the most active compound (PCB-180) was selected as a reference compound and was overlaid on the map. The CoMFA steric contour map is shown in Fig. 2(a), where the sterically favourable regions are represented in green and the unfavourable regions in yellow. As seen from this figure, a large green contour located at the 3-position of the A ring indicated that the sterically bulkier substituent is favoured at that position, as illustrated by the fact that the log KOA value of PCB-180 was stronger than that of compound PCB-153. A medium-sized region of green contour covered the 4- and 5-positions of the A ring, indicating that these positions were preferred for larger substituents to increase log KOA. The higher log KOA of PCB-180 compared with PCB-146 and PCB-138 was an example of such a case. In addition, the small green coloured contours were mapped near the 3′- and 6′-positions of the B ring, indicating that a bulkier group at this position may increase log KOA, as illustrated by the fact that the log KOA value of PCB-194 was stronger than that for compound PCB-180. The appearance of green contour and the absence of yellow contour demonstrated that the chlorine atom at the benzene ring was important for log KOA. The electrostatic contour map of CoMFA is displayed in Fig. 2 (b). The electrostatic field is denoted by blue and red coloured

206

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

Fig. 2. Contour maps of the CoMFA and CoMSIA models, CoMFA model: (a) steric fields; (b) electrostatic fields; CoMSIA model: (c) steric fields; (d) electrostatic fields; (e) hydrophobic fields. Table 2 Isomer ratios of Cl in the common PCB mixtures (George et al., 1996). The Cl isomer

The common PCB mixtures

Chlorobiphenyl Dichlorobiphenyl Trichlorobiphenyl Tetrachlorobiphenyl Pentachlorobiphenyl Hexachlorobiphenyl Heptachlorobiphenyl Octachlorobiphenyl Nonachlorobiphenyl Decachlorobiphenyl

Aroclor1016 (%)

Aroclor1242 (%)

Aroclor1248 (%)

Aroclor1254 (%)

Aroclor1260 (%)

0.7 17.5 54.7 26.6 0.5 0 0 0 0 0

0.8 15.0 44.9 32.6 6.4 0.3 0 0 0 0

0 0.4 22.0 56.6 18.6 2.0 0.6 0 0 0

0 0.2 1.3 16.4 53.0 26.8 2.7 0 0 0

0 0.1 0.2 0.5 8.6 43.4 38.5 8.3 0.7 0

13

13

CoMFA Model R² = 0.9987

12

The average of LogKoa

The average of LogKoa

12 11 10 9 8 7 6

CoMSIA Model R² = 0.9813

11 10 9 8 7

0

1

2

3

4

5

6

7

8

9 10 11

6

0

1

The number of Cl atom Fig. 3. The plot of Cl atoms vs log KOA value.

2

3

4

5

6

7

8

The number of Cl atom

9 10 11

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

207

Table A1 Predicted KOA values of PCBs (293 K) through CoMFA and CoMSIA models. No.

Compounds

Obs.

CoMFA Pred.

1 2 3a 4 5 6 7 8 9 10 11 12 13 14 15a 16 17 18 19 20 21 22 23 24 25 26 27 28 29a 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49b 50 51 52 53a 54 55 56 57 58 59 60 61a 62 63 64 65 66a 67 68 69 70 71 72 73

2-Chlorobiphenyl 3-Chlorobiphenyl 4-Chlorobiphenyl 2,2′-Dichlorobiphenyl 2,3-Dichlorobiphenyl 2,3′-Dichlorobiphenyl 2,4-Dichlorobiphenyl 2,4′-Dichlorobiphenyl 2,5-Dichlorobiphenyl 2,6-Dichlorobiphenyl 3,3′-Dichlorobiphenyl 3,4-Dichlorobiphenyl 3,4′-Dichlorobiphenyl 3,5-Dichlorobiphenyl 4,4′-Dichlorobiphenyl 2,2′,3-Trichlorobiphenyl 2,2′,4-Trichlorobiphenyl 2,2′,5-Trichlorobiphenyl 2,2′,6-Trichlorobiphenyl 2,3,3′-Trichlorobiphenyl 2,3,4-Trichlorobiphenyl 2,3,4′-Trichlorobiphenyl 2,3,5-Trichlorobiphenyl 2,3,6-Trichlorobiphenyl 2,3′,4-Trichlorobiphenyl 2,3′,5-Trichlorobiphenyl 2,3′,6-Trichlorobiphenyl 2,4,4′-Trichlorobiphenyl 2,4,5-Trichlorobiphenyl 2,4,6-Trichlorobiphenyl 2,4′,5-Trichlorobiphenyl 2,4′,6-Trichlorobiphenyl 2,3′,4′-Trichlorobiphenyl 2,3′,5′-Trichlorobiphenyl 3,3′,4-Trichlorobiphenyl 3,3′,5-Trichlorobiphenyl 3,4,4′-Trichlorobiphenyl 3,4,5-Trichlorobiphenyl 3,4′,5-Trichlorobiphenyl 2,2′,3,3′-Tetrachlorobiphenyl 2,2′,3,4-Tetrachlorobiphenyl 2,2′,3,4′-Tetrachlorobiphenyl 2,2′,3,5-Tetrachlorobiphenyl 2,2′,3,5′-Tetrachlorobiphenyl 2,2′,3,6-Tetrachlorobiphenyl 2,2′,3,6′-Tetrachlorobiphenyl 2,2′,4,4′-Tetrachlorobiphenyl 2,2′,4,5-Tetrachlorobiphenyl 2,2′,4,5′-Tetrachlorobiphenyl 2,2′,4,6-Tetrachlorobiphenyl 2,2′,4,6′-Tetrachlorobiphenyl 2,2′,5,5′-Tetrachlorobiphenyl 2,2′,5,6′-Tetrachlorobiphenyl 2,2′,6,6′-Tetrachlorobiphenyl 2,3,3′,4-Tetrachlorobiphenyl 2,3,3′,4′-Tetrachlorobiphenyl 2,3,3′,5-Tetrachlorobiphenyl 2,3,3′,5′-Tetrachlorobiphenyl 2,3,3′,6-Tetrachlorobiphenyl 2,3,4,4′-Tetrachlorobiphenyl 2,3,4,5-Tetrachlorobiphenyl 2,3,4,6-Tetrachlorobiphenyl 2,3,4′,5-Tetrachlorobiphenyl 2,3,4′,6-Tetrachlorobiphenyl 2,3,5,6-Tetrachlorobiphenyl 2,3′,4,4′-Tetrachlorobiphenyl 2,3′,4,5-Tetrachlorobiphenyl 2,3′,4,5′-Tetrachlorobiphenyl 2,3′,4,6-Tetrachlorobiphenyl 2,3′,4′,5-Tetrachlorobiphenyl 2,3′,4′,6-Tetrachlorobiphenyl 2,3′,5,5′-Tetrachlorobiphenyl 2,3′,5′,6-Tetrachlorobiphenyl

7.01

7.88

8.03

8.57

8.24

8.90

9.22

6.299 7.010 7.010 7.406 7.435 7.305 7.193 7.195 7.094 6.914 8.186 8.033 7.947 7.755 7.874 8.225 8.127 8.109 7.595 8.446 8.234 8.310 8.146 7.837 8.212 8.106 8.273 8.076 8.036 7.732 7.977 8.178 8.342 7.840 8.986 8.742 8.957 8.530 8.463 9.165 8.913 8.991 8.882 8.657 8.588 8.346 8.890 8.880 8.570 8.284 8.439 8.538 8.240 7.916 9.258 9.460 9.161 8.967 8.851 9.098 9.009 8.625 9.006 8.692 8.568 9.235 9.060 8.743 8.942 9.129 8.894 8.636 8.415

CoMSIA Relative error (%)

0

 0.07

0.075

0

0

1.22

0.16

Pred. 6.645 7.176 7.176 7.003 7.693 7.615 7.378 7.492 7.493 7.000 8.231 8.019 7.911 8.117 7.795 8.045 7.737 7.847 7.311 8.664 8.421 8.541 8.542 8.044 8.349 8.464 7.933 8.225 8.223 7.734 8.340 7.616 8.457 8.112 8.961 9.081 8.754 8.515 8.410 9.013 8.773 8.892 8.889 8.538 8.398 8.348 8.584 8.577 8.599 8.158 8.050 8.341 8.153 7.672 9.393 9.506 9.513 9.161 9.016 9.269 9.266 8.772 9.389 8.891 8.886 9.191 9.194 8.846 8.781 9.305 8.662 8.961 8.468

Relative error (%)

2.37

 1.08

2.4

0.34

 1.06

4.11

 0.31

208

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

Table A1 (continued ) No.

Compounds

Obs.

CoMFA Pred.

74 75 76 77a 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95b 96a 97 98 99 100 101a 102 103 104 105a 106 107 108 109 110 111 112 113 114 115 116 117 118a 119 120 121 122 123 124 125 126a 127 128 129 130 131 132 133 134 135 136 137 138b 139 140 141 142 143 144 145 146 147

2,4,4′,5-Tetrachlorobiphenyl 2,4,4′,6-Tetrachlorobiphenyl 2,3′,4′,5′-Tetrachlorobiphenyl 3,3′,4,4′-Tetrachlorobiphenyl 3,3′,4,5-Tetrachlorobiphenyl 3,3′,4,5′-Tetrachlorobiphenyl 3,3′,5,5′-Tetrachlorobiphenyl 3,4,4′,5-Tetrachlorobiphenyl 2,2′,3,3′,4-Pentachlorobiphenyl 2,2′,3,3′,5-Pentachlorobiphenyl 2,2′,3,3′,6-Pentachlorobiphenyl 2,2′,3,4,4′-Pentachlorobiphenyl 2,2′,3,4,5-Pentachlorobiphenyl 2,2′,3,4,5′-Pentachlorobiphenyl 2,2′,3,4,6-Pentachlorobiphenyl 2,2′,3,4,6′-Pentachlorobiphenyl 2,2′,3,4′,5-Pentachlorobiphenyl 2,2′,3,4′,6-Pentachlorobiphenyl 2,2′,3,5,5′-Pentachlorobiphenyl 2,2′,3,5,6-Pentachlorobiphenyl 2,2′,3,5,6′-Pentachlorobiphenyl 2,2′,3,5′,6-Pentachlorobiphenyl 2,2′,3,6,6′-Pentachlorobiphenyl 2,2′,3,4′,5′-Pentachlorobiphenyl 2,2′,3,4′,6′-Pentachlorobiphenyl 2,2′,4,4′,5-Pentachlorobiphenyl 2,2′,4,4′,6-Pentachlorobiphenyl 2,2′,4,5,5′-Pentachlorobiphenyl 2,2′,4,5,6′-Pentachlorobiphenyl 2,2′,4,5′,6-Pentachlorobiphenyl 2,2′,4,6,6′-Pentachlorobiphenyl 2,3,3′,4,4′-Pentachlorobiphenyl 2,3,3′,4,5-Pentachlorobiphenyl 2,3,3′,4′,5-Pentachlorobiphenyl 2,3,3′,4,5′-Pentachlorobiphenyl 2,3,3′,4,6-Pentachlorobiphenyl 2,3,3′,4′,6-Pentachlorobiphenyl 2,3,3′,5,5′-Pentachlorobiphenyl 2,3,3′,5,6-Pentachlorobiphenyl 2,3,3′,5′,6-Pentachlorobiphenyl 2,3,4,4′,5-Pentachlorobiphenyl 2,3,4,4′,6-Pentachlorobiphenyl 2,3,4,5,6-Pentachlorobiphenyl 2,3,4′,5,6-Pentachlorobiphenyl 2,3′,4,4′,5-Pentachlorobiphenyl 2,3′,4,4′,6-Pentachlorobiphenyl 2,3′,4,5,5′-Pentachlorobiphenyl 2,3′,4,5′,6-Pentachlorobiphenyl 2,3,3′,4′,5′-Pentachlorobiphenyl 2,3′,4,4′,5′-Pentachlorobiphenyl 2,3,4′,5,5′-Pentachlorobiphenyl 2,3′,4′,5′,6-Pentachlorobiphenyl 3,3′,4,4′,5-Pentachlorobiphenyl 3,3′,4,5,5′-Pentachlorobiphenyl 2,2′,3,3′,4,4′-Hexachlorobiphenyl 2,2′,3,3′,4,5-Hexachlorobiphenyl 2,2′,3,3′,4,5′-Hexachlorobiphenyl 2,2′,3,3′,4,6-Hexachlorobiphenyl 2,2′,3,3′,4,6′-Hexachlorobiphenyl 2,2′,3,3′,5,5′-Hexachlorobiphenyl 2,2′,3,3′,5,6-Hexachlorobiphenyl 2,2′,3,3′,5,6′-Hexachlorobiphenyl 2,2′,3,3′,6,6′-Hexachlorobiphenyl 2,2′,3,4,4′,5-Hexachlorobiphenyl 2,2′,3,4,4′,5′-Hexachlorobiphenyl 2,2′,3,4,4′,6-Hexachlorobiphenyl 2,2′,3,4,4′,6′-Hexachlorobiphenyl 2,2′,3,4,5,5′-Hexachlorobiphenyl 2,2′,3,4,5,6-Hexachlorobiphenyl 2,2′,3,4,5,6′-Hexachlorobiphenyl 2,2′,3,4,5′,6-Hexachlorobiphenyl 2,2′,3,4,6,6′-Hexachlorobiphenyl 2,2′,3,4′,5,5′-Hexachlorobiphenyl 2,2′,3,4′,5,6-Hexachlorobiphenyl

9.96

9.06 8.77

9.31

10.27

10.08

10.61

10.09

8.905 8.584 8.853 9.976 9.618 9.734 9.245 9.450 9.865 9.829 9.299 9.667 9.636 9.343 9.249 8.959 9.633 9.333 9.309 9.215 8.944 8.988 8.770 9.398 9.021 9.627 8.911 9.305 10.432 8.913 8.583 10.261 10.036 10.162 9.778 9.652 9.843 9.676 9.682 9.345 9.858 9.469 9.420 9.408 10.069 9.553 9.586 9.517 9.958 9.743 9.634 9.397 10.599 10.215 10.754 10.596 10.333 10.110 9.266 10.294 9.751 9.904 9.606 10.373 10.072 9.981 9.625 10.060 9.942 9.623 9.647 9.326 10.033 9.945

CoMSIA Relative error (%)

0.16

 0.79 0

 0.05

 0.09

 0.11

 0.10

 0.18

Pred. 9.070 8.581 8.951 9.804 9.807 9.924 9.580 9.249 9.741 9.857 9.315 9.620 9.613 9.267 9.127 9.077 9.736 9.245 9.383 9.236 9.192 8.892 8.702 9.383 9.196 9.424 8.892 9.071 10.229 9.001 8.519 10.234 10.238 10.355 9.890 9.745 9.858 10.010 9.393 9.512 10.114 9.619 9.611 9.733 10.036 9.510 9.691 9.628 10.000 9.684 9.799 9.306 10.650 10.420 10.582 10.581 10.235 10.156 9.300 10.351 9.806 10.158 9.668 10.461 10.111 9.974 9.924 10.107 9.961 9.916 9.621 9.431 10.227 10.083

Relative error (%)

 1.57

 1.85  0.77

 2.57

 0.35

 0.44

0.21

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

209

Table A1 (continued ) No.

Compounds

Obs.

CoMFA Pred.

148 149 150 151 152 153a 154 155a 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171b 172 173 174 175 176 177 178 179 180a 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 a b

2,2′,3,4′,5,6′-Hexachlorobiphenyl 2,2′,3,4′,5′,6-Hexachlorobiphenyl 2,2′,3,4′,6,6′-Hexachlorobiphenyl 2,2′,3,5,5′,6-Hexachlorobiphenyl 2,2′,3,5,6,6′-Hexachlorobiphenyl 2,2′,4,4′,5,5′-Hexachlorobiphenyl 2,2′,4,4′,5,6′-Hexachlorobiphenyl 2,2′,4,4′,6,6′-Hexachlorobiphenyl 2,3,3′,4,4′,5-Hexachlorobiphenyl 2,3,3′,4,4′,5′-Hexachlorobiphenyl 2,3,3′,4,4′,6-Hexachlorobiphenyl 2,3,3′,4,5,5′-Hexachlorobiphenyl 2,3,3′,4,5,6-Hexachlorobiphenyl 2,3,3′,4,5′,6-Hexachlorobiphenyl 2,3,3′,4′,5,5′-Hexachlorobiphenyl 2,3,3′,4′,5,6-Hexachlorobiphenyl 2,3,3′,4′,5′,6-Hexachlorobiphenyl 2,3,3′,5,5′,6-Hexachlorobiphenyl 2,3,4,4′,5,6-Hexachlorobiphenyl 2,3′,4,4′,5,5′-Hexachlorobiphenyl 2,3′,4,4′,5′,6-Hexachlorobiphenyl 3,3′,4,4′,5,5′-Hexachlorobiphenyl 2,2′,3,3′,4,4′,5-Heptachlorobiphenyl 2,2′,3,3′,4,4′,6-Hepatchlorobiphenyl 2,2′,3,3′,4,5,5′-Heptachlorobiphenyl 2,2′,3,3′,4,5,6-Heptachlorobiphenyl 2,2′,3,3′,4,5,6′-Heptachlorobiphenyl 2,2′,3,3′,4,5′,6-Heptachlorobiphenyl 2,2′,3,3′,4,6,6′-Heptachlorobiphenyl 2,2′,3,3′,4,5′,6′-Heptachlorobiphenyl 2,2′,3,3′,5,5′,6-Heptachlorobiphenyl 2,2′,3,3′,5,6,6′-Heptachlorobiphenyl 2,2′,3,4,4′,5,5′-Heptachlorobiphenyl 2,2′,3,4,4′,5,6-Heptachlorobiphenyl 2,2′,3,4,4′,5,6′-Heptachlorobiphenyl 2,2′,3,4,4′,5′,6-Heptachlorobiphenyl 2,2′,3,4,4′,6,6′-Heptachlorobiphenyl 2,2′,3,3′,5,5′,6-Heptachlorobiphenyl 2,2′,3,4,5,6,6′-Heptachlorobiphenyl 2,2′,3,4′,5,5′,6-Heptachlorobiphenyl 2,2′,3,4′,5,6,6′-Heptachlorobiphenyl 2,3,3′,4,4′,5,5′-Heptachlorobiphenyl 2,3,3′,4,4′,5,6-Heptachlorobiphenyl 2,3,3′,4,4′,5′,6-Heptachlorobiphenyl 2,3,3′,4,5,5′,6-Heptachlorobiphenyl 2,3,3′,4′,5,5′,6-Heptachlorobiphenyl 2,2′,3,3′,4,4′,5,5′-Octachlorobiphenyl 2,2′,3,3′,4,4′,5,6-Octachlorobiphenyl 2,2′,3,3′,4,4′,5,6′-Octachlorobiphenyl 2,2′,3,3′,4,4′,6,6′-Octachlorobiphenyl 2,2′,3,3′,4,5,5′,6-Octachlorobiphenyl 2,2′,3,3′,4,5,5’,6’-Octachlorobiphenyl 2,2′,3,3′,4,5,6,6′-Octachlorobiphenyl 2,2′,3,3′,4,5′,6,6′-Octachlorobiphenyl 2,2′,3,3′,5,5′,6,6′-Octachlorobiphenyl 2,2′,3,4,4′,5,5′,6-Octachlorobiphenyl 2,2′,3,4,4′,5,6,6′-Octachlorobiphenyl 2,3,3′,4,4′,5,5′,6-Octachlorobiphenyl 2,2′,3,3′,4,4′,5,5′,6-Nonachlorobiphenyl 2,2′,3,3′,4,4′,5,6,6′-Nonachlorobiphenyl 2,2′,3,3′,4,5,5′,6,6′-Nonachlorobiphenl Decachlorobiphenyl

10.04 9.16

10.51

10.75

9.607 9.707 9.416 9.647 9.349 10.027 9.591 9.161 11.025 10.758 10.633 10.551 10.473 10.144 10.653 10.315 10.316 10.077 10.250 10.571 10.210 11.078 11.470 10.723 11.059 10.537 10.595 10.701 10.164 9.984 10.353 9.993 10.773 10.659 10.274 10.353 9.961 10.353 9.950 11.998 9.939 11.517 11.094 11.104 10.940 11.033 11.905 11.148 11.380 10.873 11.124 11.101 10.274 9.710 10.304 11.025 10.529 11.885 11.801 10.613 11.124 11.805

CoMSIA Relative error (%)

 0.13 0.01

2.03

0.21

Pred. 10.039 9.737 9.549 9.621 9.538 9.916 9.731 9.252 11.079 10.728 10.587 10.735 10.232 10.241 10.848 10.122 10.350 10.355 10.458 10.529 10.040 11.146 11.422 10.885 11.076 10.644 10.883 10.999 10.397 10.213 10.649 10.159 10.952 10.808 10.764 10.465 10.278 10.649 10.263 12.235 10.385 11.573 10.961 11.079 11.080 11.193 11.914 11.373 11.724 11.238 11.487 11.448 10.753 10.744 10.921 11.300 11.110 11.918 12.212 11.538 11.719 12.557

Relative error (%)

 1.24 1.00

3.57

1.88

Training set. Test set.

contours, where the blue regions represent the electropositive groups near these regions favourable to log KOA, and the red regions indicate that the electronegative groups close to these regions may increase log KOA. It has been recognized that the 3′- and 6-positions are encompassed by a medium-sized region of blue contours, suggesting that these positions were preferred for larger

substituents to increase log KOA. The large red contour around the 3-position of A ring depicts its favour for the electropositive groups. The fact that the log KOA value of PCB-180 was stronger than compound PCB-153 verifies this observation. A small red coloured contour was mapped near the 4′-position of the B ring and this is the same case for the comparison between PCB-180 and

210

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

PCB-141. To aid in visualization, the most active compound (PCB-180) was selected as a reference compound and was overlaid on the map of the CoMSIA model. The CoMSIA steric contour map is displayed in Fig. 2(c) in which the sterical favourable regions are represented in green and the unfavourable regions in yellow. It has been recognized that the 3- and 3′-positions are encompassed by larger green contours, suggesting that bulkier groups are favoured at this position. A medium-sized region of green contour covered the 4′- and 5-positions of the benzene ring, indicating that these positions were preferred for larger substituents to increase log KOA. The fact that the log KOA value of PCB-180 was stronger than that for compound PCB-141 and PCB-138 is a good example of this phenomenon. The electrostatic contour map obtained from the CoMSIA model is depicted in Fig. 2(d). The blue area suggests that a stronger electronegativity of the substituted group indicates a smaller log KOA of PCBs. It can be observed that strong electronegative groups generally result in a lower log KOA. A mediumsized region of blue contour covering the 4- and 5-position of the carbon atom revealed that the electropositive groups are favourable for enhancing log KOA. A large red contour covering the 3-position indicates that an electronegative group at this position may decrease log KOA. The higher log KOA value of PCB-180 compared with PCB-153 is an example of such a case. The hydrophobic contour map of the CoMSIA model is presented in Fig. 2(e) in which the yellow regions were hydrophobic favourable for the log KOA, while the white region was hydrophilic favourable. A lager white region was observed close to the 3-, 4and 5-positions of the A ring, indicating that a hydrophobic group at these positions was good for increasing log KOA. Simultaneously, a larger white region was located at the 3′-, 4′- and 5′-position of the B ring. The appearance of a white contour and the absence of yellow contour demonstrated that the hydrophilic group at the benzene ring was important for the log KOA value of PCBs. 3.5. Analysis of differences and similarities between the CoMFA and CoMSIA models From the statistical parameters, the CoMFA and CoMSIA models were considered reliable and acceptable because q2 was greater than 0.50, and r2 was greater than 0.90. The statistical indexes of the CoMFA and CoMSIA models are reasonably high, indicating that the two models have reached the internal inspection standard of models. External validation was also carried out to further evaluate the reliabilities and the predictive ability of the constructed model. Further analysis of the log KOA predicted by the CoMFA and CoMSIA models revealed a fine linear dependence among experimental values and predicted values, which can reach the external inspection standard of models (Development OFEC, 2007; Worth et al., 2004). The validations suggest that the models exhibit optimum stability and good predictive power. Thus, the two models can be used to predict the log KOA of the same types of compounds. From the contribution rates of the descriptor fields, the two models mutually verify and prove that electronic effects primarily influence the log KOA. According to the 3D isogram of the CoMFA and CoMSIA models, the information for the modification of PCBs was generally consistently obtained. Although the models show satisfactory fitting ability and acceptable predictive ability, the CoMFA model only contains two descriptor fields (steric and electrostatic fields), presenting certain limitations in terms of the analysis of the effect of the descriptor field and the modification of the information on the compounds. The CoMSIA model contains five descriptor fields, which will provide a more comprehensive understanding of the effect on the physical and chemical properties of PCBs. It is noted that in this

model, the five descriptor fields play an important role in renovating PCBs and designing new types of compound molecules. For example, an analysis of the hydrophobic field could reveal that a larger white region was observed close to the 3-, 4- and 5-positions of the A ring, which indicated that a hydrophobic group would increase log KOA. 3.6. The migration ability analysis of PCBs A homologue is a subset of PCB congeners defined by the degree of chlorination, i.e., monochlorobiphenyls through decachlorobiphenyls. PCBs within a homologue group have the same molecular weight and number of chlorines attached to the biphenyl ring, but the pattern of chlorination varies among the individual isomers within the homologue group (Tala and Michael, 2003). Although PCBs include 209 congeners, only 130 types of congeners were found in the commercial mixtures of PCBs. Commercial products of PCBs are composed of chlorides on the diphenyl moiety. According to the degree of chlorination, the commercial mixtures of PCBs around the world have different names. Monsanto is the largest producer of the commercial mixtures of PCBs in the United States; therefore, its PCBs product names are used globally. According to the degree of chlorination, the PCB mixtures of Monsanto are called Aroclor. Seven types of the commercial mixtures of PCBs are most commonly used, whose names include the addition of four numbers after Aroclor, including Aroclor 1016, Aroclor 1221, Aroclor 1232, Aroclor 1242, Aroclor 1248, Aroclor 1254 and Aroclor 1260. Each Aroclor is composed of many independent congeners. Except for Aroclor 1016, the first two digits of each Aroclor are 12, representing the number of carbon atoms in the molecules, and then the double digit represents the isomer ratios of Cl in the commercial mixtures of PCBs (Xie, 2013). The isomer ratios of Cl in the common commercial PCB mixtures are shown in Table 2. As shown in Table 2, the Cl isomers from trichlorodiphenyl to heptachlorobiphenyl make up most of the commercial mixtures of PCBs, and chlorobiphenyl and dichlorobiphenyl occupy a certain proportion. However, the Cl range from octachlorobiphenyl to decachlorobiphenyl is not commonly used in the commercial products of PCBs. Cummins (Tala, 2003) estimated that sixty-five percent of the commercial mixtures of PCBs existed in capacitors and transformers and were in landfills and stored on land, while thirty-one percent entered the environment during use and disposal and have been shown to be ubiquitous contaminants; the remainder were degraded and burned. Once released into the environment, PCBs first polluted the atmosphere and then spread primarily via atmospheric movement, which led to global pollution. Twenty-nine PCB congeners were detected in the Arctic by Zhang et al. (2014), including (numbered by the IUPAC system) 18, 28, 44, 52, 66, 77, 81, 87, 101, 105, 110, 114, 118, 123, 126, 128, 138, 153, 155, 169, 170, 180, 187, 189, 194, 195, 200, 205, 206, which the range is from trichlorodiphenyl to nonachlorobiphenyl. The following PCB congeners were detected in Arctic seabird species by Marit et al. (2011): 18, 28, 31,33, 37, 47, 52, 66, 74, 99, 101, 105, 118, 123, 141, 149, 153, 138, 167,128, 156, 157, 170, 180, 183, 187, 189, 194, 206, 209. The range of these congeners is from trichlorodiphenyl to decachlorobiphenyl. Twenty-three PCB congeners were detected in Antarctic air by Vecchiato et al. (2015) and Piazza et al. (2013), including 18, 28, 31, 44, 52, 70, 74, 92, 99, 105, 107, 110, 118, 146, 149, 151, 153, 158, 174, 180, 183, 187, 206, and this range is also from trichlorodiphenyl to nonachlorobiphenyl. These studies indicate the strong migration ability of PCBs between trichlorobiphenyl and decachlorobiphenyl, which can undertake global migration. However, the migration ability of chlorobiphenyl and dichlorobiphenyl is low due to their small values of log KOA;

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

therefore, they were not detected in the Arctic and Antarctic. This is consistent with the finding by Cleverly (2005) that the degree of chlorination directly affects the PCB behaviour in the environment. As shown in Table 2, the productions and usages of chlorobiphenyl and dichlorobiphenyl were more than those of PCBs between octachlorobiphenyl and decachlorobiphenyl, and many were released into the environment, although they were not detected in the Arctic and Antarctic area. This result indicates that stronger KOA values of PCBs imply a stronger migration ability of PCBs. Thus, the PCB migration ability gradually increased with the increasing number of Cl atoms. PCBs between chlorobiphenyl and decachlorobiphenyl served as independent variables, and the average log KOA values of PCBs served as the dependent variable. The relationship between the number of Cl atoms and the average log KOA values of PCBs is depicted in Fig. 3. As can be observed from Fig. 3, the log KOA values gradually increased with the increasing number of Cl atoms, and the higher the log KOA values are, the stronger the PCB migration ability, which was consistent with the previous discussions.

4. Conclusions QSAR models were constructed with 3D descriptors according to the experimental values of log KOA for 19 PCB congeners. Two types of QSAR methods, CoMFA and CoMSIA, are used to predict the remaining 190 PCB congeners and to investigate the relationship between the structures of PCBs and their persistent migration properties. The main research conclusions are as follows: (1) CoMFA and CoMSIA models show satisfactory fitting ability and acceptable predictive ability of the 190 PCBs. The two models mutually verify and prove that the electrostatic descriptors play a more significant role than steric descriptors and hydrophobic descriptors. However, hydrogen bond donor and acceptor fields contribute minimally to the log KOA value of PCBs when steric, electrostatic and hydrophobic fields were introduced. (2) In an electrostatic field, the introduction of electropositive groups in some positions of PCBs can significantly reduce the log KOA values and migration ability. (3) The correlation analysis of the relationship between the number of Cl atoms and the average log KOA values of PCBs revealed that the log KOA values gradually increased as the number of Cl atoms increases. Simultaneously, related studies on the detection of PCBs in the Arctic and Antarctic prove that higher log KOA values indicate stronger PCB migration ability.

Acknowledgements The research was supported by the Fundamental Research Funds for the Central Universities in 2013 (JB2013146) and the Key Projects in the National Science & Technology Pillar Program in the Eleventh Five-Year Plan Period (2008BAC43B01).

Appendix A See Table A1. References Alkhatib, E., Weigand, C., 2002. Parameters affecting partitioning of 6 PCBs congeners in natural sediments. Environ. Monit. Assess. 78, 1–17.

211

Arrebola, J.P., Fernandez, M.F., Porta, M., Rosell, J., dela Ossa, R.M., Olea, N., MartinOlmedo, P., 2010. Multivariate models to predict human adipose tissue PCB concentrations in Southern Spain. Environ. Int. 36, 705–713. Betterman, S., Zhang, L., Wang, S., Franzblau, A., 2002. Partition coefficients for the trihalomethanes among blood, urine, water, milk and air. Sci. Total Environ. 284, 237–247. Bidleman, F.T., Helm, A.P., Braune, M.B., Gabrielsen, W.G., 2010. Polychlorinated naphthalenes in polar environments – a review. Sci. Total Environ. 408, 2919–2935. Chen, J.W., Xue, X.Y., Schramm, K.W., Quan, X., Yang, F.L., 2002. Antonius Kettrup. Quantitative structure–property relationships for octanol-air partition coefficients of polychlorinated biphenyls. Chemosphere 48, 535–544. Clark, M., Cramer, R.D., Vanopdenbosch, N., 1989. Validation of the general purpose tripos 5.2 force field. J. Comput. Chem. 10, 982–1012. Cramer Iii, R.D., Bunce, J.D., Patterson, D.E., Frank, I.E., 1988. Crossvalidation, Bootstrapping, and Partial Least Squares Compared with Multiple Regression in Conventional QSAR Studies. Quant. Struct. Act. Relat. 7, 18–25. Cramer, R.D., David, E.P., Jeffrey, D.B., 1988. Comparative Molecular Field Analysis (CoMFA). 1. Effect of Shape on binding of steroids to carrier proteins. J. Am. Chem. Soc. 110, 5959–5967. Crippen, G.M., 1979. Distance geometry approach to rationalizing binding data. J. MedChem. 22, 988–997. Cruciani, G., 2003. 25–Three-dimensional quantitative structure–property relationships. In: Camille, G.W. (Ed.), The Practice of Medicinal Chemistry, Second Edition Academic Press, London, pp. 405–416. Dachs, J., Eisenreich, S.J., 2000. Adsorption onto aerosol soot carbon dominates gasparticle partitioning of polycyclic aromatic hydrocarbons. Environ. Sci. Technol. 34, 3690–3697. Development OFEC, 2007. Guidance document on the validation of (quantitative) structure-activity relationship [(Q)SAR] models[R]. OECD Environment Health and Safety Publications, Series on Testing and Assessment, No. 69, ENV / JM / MONO, vol. 2, OECD Paris, France. Free, S.M., Wilson, J.W., 1964. A mathematical contribution to structure-activity studies. J. Med. Chem. 7, 395–399. Fujita, T., Iwasa, J., Hansch, C., 1964. A new substituent constant, π, derived from partition coefficients. J. Am. Chem. Soc. 86, 5175–5180. Gasteiger, J., Marsili, M., 1980. Iterative partial equalization of orbital electronegativity–a rapid access to atomic charges. Tetrahedron 36, 3219–3228. George, M.F., Jack, W.C., Soren, S.B., 1996. J. High. Resol. Chromatogr. 19, 657–668. Golbraikh, A., Tropsha, A., 2002. Beware of q2!. J. Mol. Graph. Modell. 20, 269–276. Hansch, C., Fljita, T., 1964. A Method for the Correlation of Biological Activity and Chemical Structure. J. Am. Chem. Soc. 86, 1616–1626. Harner, T., Bidleman, T.F., 1996. Measurement of octanol–air partition coefficients for polychlorinated biphenyls. J. Chem. Eng. 41, 895–899. Harner, T., Bidleman, T.F., Jantunen, L.M.M., Mackay, D., 2001. Soil/air exchange model of persistent pesticides in the United States cotton belt. Environ. Toxicol. Chem. 20, 2001, 1612–1621. Harner, T., Mackay, D., 1995. Measurement of octanol-air partition coefficient for chlorobenzenes, PCBs and DDT. Environ. Sci. Technol. 29, 1599–1606. Hassine, S.B., Ameur, W.B., Gandoura, N., Driss, M.R., 2012. Determination of chlorinated pesticides, polychlorinated biphenyls, and polybrominated diphenyl ethers in human milk from Bizerte (Tunisia) in 2010. Chemosphere 89, 369–377. Hopfinger, A.J., 1980. A QSAR investigation of dihydrofolate reductase inhibition by Baker triazines based upon molecular shape analysis. J. Am. Chem. Soc. 102, 7196–7206. Jotaki, T., Fukata, H., Mori, C., 2011. Confirmation of polychlorinated biphenyl (PCB) distribution in the blood and verification of simple quantitative method for PCBs based on specific congeners. Chemosphere 82, 107–113. Klebe, G., Abraham, U., Mietzner, T., 1994. Molecular Similarity Indices in a Comparative Analysis (CoMSIA) of drug molecules to correlate and predict their biological activity. J. Med. Chem. 37, 4130–4146. Langer, T., Bryant, S.D., 2008. Chapter 29 – 3D quantitative structure–property relationships. In: Georges, W. Camille (Ed.), The Practice of Medicinal Chemistry, Third Edition Academic Press, New York, pp. 587–604. Leach, A.R., 2001. Molecular Modelling: Principles and Applications. Person Education Limited, Harlow, England. Liao, S.Y., Qian, L., Miao, T.F., Lu, H.L., Zheng, K.C., 2009. CoMFA and docking studies of 2-phenylindole derivatives with anticancer activity. Eur. J. Med. Chem. 44, 2822–2827. Li, M., Wei, D.B., Zhao, H.M., Du, Y.G., 2014. Genotoxicity of quinolones: Substituents contribution and transformation products QSAR evaluation using 2D and 3D models. Chemosphere 95, 220–226. Li, X.L., Ye, L., Shi, W., Liu, H.L., Liu, C.S., Qian, X.P., Zhu, Y.L., Yu, H.X., 2013. In silico study on hydroxylated polychlorinated biphenyls as androgen receptor antagonists. Ecotoxicol. Environ. Saf. 92, 258–264. Liu, H.G., Shi, J.Q., Liu, H., Wang, Z.Y., 2013. Improved 3D QSPR analysis of the predictive octanoleair partition coefficients of hydroxylated and methoxylated polybrominated diphenyl ethers. Atmos. Environ. 77, 840–845. Li, X.L., Ye, L., Wang, X.X., Wang, X.Z., Liu, H.L., Qian, X.P., Zhu, Y.L., Yu, H.X., 2012a. Molecular docking, molecular dynamics simulation, and structure-based 3D QSAR studies on estrogenic activity of hydroxylated polychlorinated biphenyls. Sci. Total Environ. 441, 230–238. Li, X.L., Ye, L., Wang, X.X., Wang, X.Z., Liu, H.L., Zhu, Y.L., Yu, H.X., 2012b. Combined 3D QSAR, molecular docking and molecular dynamics study on thyroid hormone activity of hydroxylated polybrominated diphenyl ethers to thyroid

212

Y. Chen et al. / Ecotoxicology and Environmental Safety 124 (2016) 202–212

receptors β. Toxicol. Appl. Pharm. 265, 300–307. Marco, V., Stefano, Z., Elena, A., Carlo, B., Andrea, G., Rossano, P., 2015. Polychlorinated biphenyls (PCBs) and polybrominated diphenyl ethers (PBDEs) in Antarctic ice-free areas: Influence of local sources on lakes and soils. Microchem. J. 120, 26–33. Marit, N.H., Lionel, C., Lisa, B.H., Raffaella, B., Geir, W.G., Francesco, R., 2011. Hepatic antioxidant responses related to levels of PCBs and metals in chicks of three Arctic seabird species. Comp. Biochem. Physiol. Part C 154, 28–35. Piazza, R., Gambaro, A., Argiriadis, E., Vecchiato, M., Zambon, S., Cescon, P., Barbante, C., 2013. Development of a method for simultaneous analysis of PCDDs, PCDFs, PCBs, PBDEs, PCNs and PAHs in Antarctic air. Anal. Bioanal. Chem. 405, 917–932. Platts, J.A., Abraham, M.H., 2000. Partition of volatile organic compounds from air and from water into plant cuticular matrix: an LFER analysis. Environ. Sci. Technol. 34, 318–323. Rawn, D.F.K., Sadler, A.R., Quade, S.C., Sun, W.F., Kosarac, I., Hayward, S., Ryan, J.J., 2012. The impact of production type and region on polychlorinated biphenyl (PCB), polychlorinated dibenzo-p-dioxin and dibenzofuran (PCDD/F) concentrations in Canadian chicken egg yolks. Chemosphere 89, 929–935. Salahinejad, M., Ghasemi, J.B., 2014. 3D QSAR studies on the toxicity of substituted benzenes to Tetrahymena pyriformis: CoMFA, CoMSIA and VolSurf approaches. Ecotoxicol. Environ. Saf. 105, 128–134. Su, G.Y., Liu, X.H., Gao, Z.S., Xian, Q.M., Feng, J.F., Zhang, X.W., Giesy, J.P., Wei, S., Liu, H.L., Yu, H.X., 2012. Dietary intake of polybrominated diphenyl ethers (PBDEs) and polychlorinated biphenyls (PCBs) from fish and meat by residents of Nanjing, China. Environ. Int. 42, 138–143. Tala, H., Michael, J.D.V., 2003. Non—Dioxin—like PCBs: Effects and Consideration in Ecological Risk Assessment. US EPA, USA. Tato, L., Tremolada, P., Ballabio, C., Guazzoni, N., Parolini, M., Caccianiga, M., Binelli, A., 2011. Seasonal and spatial variability of polychlorinated biphenyls (PCBs) in vegetation and cow milk from a high altitude pasture in the Italian Alps.

Environ. Pollut. 159, 2656–2664. Wang, Z., Wang, T., Gao, R., Xue, L.K., Guo, J., Zhou, Y., Nie, W., Wang, X.F., Xu, P.J., Gao, J., Zhou, X.H., Wang, W.X., Zhang, Q.Z., 2011. Source and variation of carbonaceous aerosols at Mount Tai, North China: Results from a semicontinuous instrument. Atmos. Environ. 45, 1655–1667. Worth, A.P., Van Leeuwen, C.J., Hartung, T., 2004. The prospects for using QSAR in a changing political environment—high expectations and a key role for the European commission's joint research centre. SAR QSAR Environ. Res. 15, 331–343. Xie, T.M., 2013. The Analysis of Polychlorinated Biphenyls (PCBs) in Environmental Samples. China Environmental Science Press, China. Xu, M., Zhang, A.Q., Han, S.K., Wang, L.S., 2002. Studies of 3D-quantitative structure–activity relationships on a set of nitroaromatic compounds: CoMFA, advanced CoMFA and CoMSIA. Chemosphere 48, 707–715. Zhang, P., Ge, L.K., Gao, H., Yao, T., Fang, X.D., Zhou, C.H., Na, G.S., 2014. Distribution and transfer pattern of Polychlorinated Biphenyls (PCBs) among the selected environmental media of Ny-Alesund, the Arctic: As a case study. Mar. Pollut. Bull. 89, 267–275. Zhang, P., Ge, L., Zhou, C., Yao, Z., 2011. Evaluating the performance of accelerated solvent extraction microwave assisted extraction and ultrasonic extraction for determining PCBs, HCHs and DDTs in sediment. Chin. J. Ocean. Limnol. 29, 1103–1112. Zhang, P., Song, J.M., Liu, Z.G., Zheng, G.X., Zhang, N.X., He, Z.P., 2007. PCBs and its coupling with eco-environments in Southern Yellow Sea surface sediments. Mar. Pollut. Bull. 54, 1105–1115. Zhang, X.M., Schramm, K.W., Henkelman, B., Klimm, G., Kaune, A., Kettrup, A., Lu, P. Z., 1999. A method to estimate the octanol–air partition coefficient of semivolatile organic compounds. Anal. Chem. 71, 3834–3838. Zheng, X.Y., Liu, X.D., Jiang, G.B., Wang, Y.W., Zhang, Q.H., Cai, Y.Q., Cong, Z.Y., 2012. Distribution of PCBs and PBDEs in soils along the altitudinal gradients of Balang Mountain, the east edge of the Tibetan Plateau. Environ. Pollut. 161, 101–106.