How Do the Partitioning Properties of Polyhalogenated POPs Change


How Do the Partitioning Properties of Polyhalogenated POPs Change...

0 downloads 61 Views 1MB Size

Environ. Sci. Technol. 2008, 42, 5189–5195

How Do the Partitioning Properties of Polyhalogenated POPs Change When Chlorine Is Replaced with Bromine? T O M A S Z P U Z Y N , * ,† N O R I Y U K I S U Z U K I , † AND MACIEJ HARANCZYK‡ National Institute for Environmental Studies, Research Center for Environmental Risk, Exposure Assessment Research Section, 16-2 Onogawa, Tsukuba, Ibaraki, 305-8506 Japan, ´ and Faculty of Chemistry, University of Gdansk, Sobieskiego ´ 18, 80-952 Gdansk, Poland

Received January 23, 2008. Revised manuscript received April 11, 2008. Accepted April 11, 2008.

Information about the mobility and the partitioning properties of brominated persistent organic pollutants, the environmental levels of which are sometimes higher than those of the chlorinated analogues, is limited. We estimated n-octanol/ water (log KOW), n-octanol/air (log KOA), and air/water (log KAW) partition coefficients for 1436 chloro- and bromo-analogues of dibenzo-p-dioxins, dibenzofurans, biphenyls, naphthalenes, diphenyl ethers, and benzenes by employing quantitative structure-property relationship (QSPR) techniques. The searches for similar partitioning patterns were performed by means of twodimensional cluster analysis. Five classes of compounds were identified. Each of the class is characterized by similar partition coefficients and, in consequence, similar environmental properties. Finally the data was fitted into a simple multimedia model involving the partitioning map. In addition, we found that the changes in the partition coefficients upon the replacement of chlorine with bromine were constant: 0.11, 0.31, and -0.21 per bromine atom for log KOW, log KOA, and log KAW, respectively. On the basis of this observation, a method for rapid estimation of changes in the partition coefficient upon chlorine-bromine substitution was proposed.

Introduction The rapid increase in the production and use of brominated flame retardants (BFRs) such as polybrominated diphenyl ethers (PBDEs), polybrominated biphenyls (PBBs), tetrabromobisphenol A, and hexabromocyclododecane has resulted in increased levels of these chemicals in various environmental compartments (1), food (2), and human tissues (3). Moreover, the incineration of waste containing BFRs (e.g., discarded televisions and monitors, computers, paints, furnishings for car interiors) might be a significant source of bromo-substituted analogues of chlorodibenzop-dioxins (PCDDs) and chlorodibenzofurans (PCDFs); that is, polybrominated dibenzo-p-dioxins (PBDDs) and polybrominated dibenzofurans (PBDFs), respectively (4). In addition, PBDDs and PBDFs may also be present as impurities * Corresponding author: phone, +81-298-50-2888; fax, +81-29850-2920; e-mail, [email protected] † Exposure Assessment Research Section. ‡ University of Gdan ´ sk. 10.1021/es8002348 CCC: $40.75

Published on Web 06/10/2008

 2008 American Chemical Society

in the original BFR mixtures (5). Furthermore, PBDD/Fs might be biosynthesized naturally, and the levels of these compounds might be increased by global warming and eutrophication (6). If so, PBDD/Fs can be considered environmental stressors rather than pollutants. Whatever their sources, more attention should be paid to polybrominated compounds, including the dynamics of their environmental concentrations. Some of the polybrominated compounds, like their chlorinated analogues, may act in the way specific for very toxic 2,3,7,8-tetrachlorodibenzo-p-dioxin. They bind to the aryl hydrocarbon receptor (AhR) and, in consequence, induce changes in gene transcription leading to adverse changes in cellular processes and function (7). A comprehensive risk assessment requires information about environmental transport processes, and fate. Various types of multimedia models can be used to obtain this information (8). However, those models require detailed data on partition coefficients (n-octanol/water, log KOW; n-octanol/ air, log KOA; and air/water, log KAW). Unfortunately, because of the high costs of the necessary experiments, limited availability of congeners (pure standards), and analytical problems with separation of individual congeners, such data exist only for few brominated persistent organic pollutants (POPs) (9, 10). Because of structural similarities, the partition coefficients for PBDDs, PBDFs, PBBs, PBDEs, polybrominated naphthalenes (PBNs), and brominated benzenes (BBzs) should not differ substantially from those of their chlorinated analogues (7, 9). The environmental partitioning and transport mechanisms of Cl-POPs have been studied extensively over the last 20 years, and thus data for some of these compounds exist (10). Hence, one could expect that the comprehensive estimation of the key environmental properties of the brominated compounds is possible by the use of (i) the existing data for Cl-POPs, (ii) the limited data available for their brominated analogues, and (iii) application of the structural similarity assumption. In this study, we apply the presented points (i)-(iii) to demonstrate how changing a chlorine atom to a bromine atom affects the partitioning properties of a compound. In addition, three of the most important partition coefficients (log KOW, log KOA, and log KAW) are predicted for 1436 compounds, including all the PBDD, PBDF, PBB, PBDE, PBN, and BBz congeners and their chlorinated analogues. We used the same method to predict all the values so that the comparisons between the chlorinated and brominated analogues would be more reliable. Moreover, the partitioning space of the chloro- and bromo-derivatives (a threedimensional hypothetical space, where each partition coefficient defines one axis) is explored by searching for similar partitioning patterns and the ranking of the compounds according to their environmental mobility is proposed.

Materials and Methods QSPR Modeling. In the first stage of the study, we used the QSPR approach to calculate log KOW and log KOA values for all 1436 compounds. The method is based on the paradigm that the variance in the physicochemical properties of chemical compounds is determined by the variance in their molecular structures. Thus, if experimental data are available only for some chemicals in a group, one can predict the missing data from molecular descriptors calculated for the whole group and a suitable mathematical model (11). The usual QSPR procedure involves (i) collection of experimental data; (ii) calculation of molecular descriptors for all the studied compounds; (iii) splitting of the compounds VOL. 42, NO. 14, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

5189

for which experimental data exist into two sets, a training set and a validation set; (iv) calibration of the model with the training set; (v) external validation with the validation set; and (vi) if the model passes the validation criteria, prediction of values for new compounds (11, 12). The Organization for Economic Co-operation and Development has recommended that, for regulatory purposes, a correctly validated model should be associated with appropriate statistical measures of goodness-of-fit, robustness, and predictivity; information on the domain of applicability; and a mechanistic interpretation, if possible. Moreover, the predicted endpoint should be well defined, and the modeling algorithm must be unambiguous, so that anyone can easily repeat the modeling, if necessary (12). Many authors highlight the importance of the experimental data quality and their possible influence on the QSPR modeling results (11, 13). In many cases, the reported data on the physicochemical properties, as water solubility and partition coefficients can differ even up to 2 orders of magnitude. It is relatively easy to find and eliminate the errors resulting from multilevel referencing, incorrect citation, or copying of the data (“literature errors”). But the evaluation of uncertainties related to different analytical methods and the procedures applied by different laboratories are usually very difficult (13). In those cases, when more than one value of the partition coefficients had been available for a compound, we have selected the final one according to the recommendation given by the Handbook of Physical-Chemical Properties and Environmental Fate for Organic Compounds (10). In ref 10 the authors had comprehensively validated the existing data and recommended single, the most reliable, coefficient values for each compound. According to the information given by the authors of the handbook (10), the evaluation of the data had been guided by (i) the acknowledgment of previous supporting or conflicting values; (ii) the method of determination: experimental direct (i.e, generator column, shakeflask), experimental indirect (i.e., these based on HPLC or GC retention times), calculation (i.e., QSPR); (iii) the perception of the objectives of the authors, not necessarily as an indication of competence, but often as an indication of the need of the authors to obtain accurate values; (iv) the reported values for structurally similar, or homologues compounds. It should be noted that only the experimentally determined (not calculated) values have been taken into account. Similarly, only the partition coefficients determined directly at 25 °C have been taken. For the rest of cases, we were searching for the original papers presenting the data. The log KOW values for polychlorinated diphenyl ethers (PCDEs) were taken from ref 14; for PBDEs from (15); for PBBs (16); and for polychlorinated naphthalenes (PCNs) (17). The experimental log KOA values were taken from the following studies ref 18 for PCDDs and PCDFs; refs 19, 20 for PBDEs; refs 20, 21 for polychlorinated biphenyls (PCBs); refs 10, 22, 23 for PCNs; and refs 10, 20 for chlorinated benzenes (CBzs). These data were evaluated by the same criteria as those proposed by the handbook (10). Of course, the probability of making a “literature error” when data are copied directly from the original paper is lower, than if the data are derived from a compilation such as the handbook (13). However, we assumed that even if some values had been cited incorrectly in the handbook and the “literature error” was high, it would be possible to identify that when the applicability domain of the model was tested by the Williams plot (a plot of the standardized residuals vs leverage values). The leverage value describes how a given compound differs from the others at the molecular level (for details please refer to the Supporting Information). Great residual value and low leverage might indicate the “literature error” for the compound. If both the residual and the leverage 5190

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 14, 2008

were low and the number of training compounds was high (as it was in our study), the error should not influence the results of the QSPR-based prediction significantly. The structures of the 1436 studied compounds were expressed in a format of internal coordinates by means of the ConGENER program (24). This program combinatorially generates families of congeners and facilitates their characterization by means of quantum-chemistry software packages. The geometry of each of the molecular structures was optimized at the of the PM6 semiempirical level using MOPAC 2007 package (25). For the optimized geometries, eight molecular descriptors were calculated: the standard heat of formation (∆H), the dipole moment (µ), the solvent-accessible surface area (SAS), the energy of the highest occupied molecular orbital (HOMO), the energy of the lowest molecular orbital (LUMO), the most positive and negative partial charges on atoms (q+ and q-) and the mean molecular polarizability (R). In our previous study (26), it was demonstrated that by employing the molecular descriptors from the PM6 method, one can obtain QSPR models of quality similar to that of the models based on density functional theory descriptors for hundreds of chemicals in a relatively short time. However, our previous study dealt only with Cl-POPs. The developers of the PM6 method (25) indicate that the Hamiltonian was also reparameterized for the bromine atom, so our current results are expected to be of good quality. To avoid the danger of chance correlation described by Topliss and Edwards (27), we applied only those descriptors which had been found to be useful and interpretative in previous QSPR studies for POPs (26). The compounds, for which experimental data was available, were divided into training and validation sets. The compounds were ranked according to their endpoints (the experimentally determined values), and beginning from the third compound, every fifth compound was labeled as a validation compound and removed from the training set; the first and second compounds were arbitrarily included in the training set and the validation set, respectively. This commonly used method produces two sets that accurately represent the data. Later, the data was used to calibrate and validate the models. We selected multiple linear regression (MLR), which is one of the most commonly used and transparent modeling techniques for QSPR (11). The goodness-of-fit was expressed in terms of the squared correlation coefficient (R2) and the root-mean-square error of calibration (RMSEC); the robustness of the model, by the cross-validation coefficient (Q2CV) and the root-mean-square of cross-validation (RMSECV); and the predictivity, by the external validation coefficient (Q2Ext) and the root-mean-square error of prediction for the external validation set (RMSEP) (12). The optimal combination of the descriptors in the models has been found by a backwise regression on the initially reduced set of the eight descriptors. The initial reduction included (i) calculation of the correlation coefficients between the individual descriptors and (ii) selection of one representative descriptor from each group of collinear descriptors. The quality of the models was judged based on the RMSECV value. The model characterized by the lowest RMSECV value in the case of each partition coefficient has been selected as the most optimal one. The following four the most predictive descriptors were finally selected for the models: R, HOMO, SAS, and µ. The optimized models were thus externally validated, and we used them to calculate log KOW and log KOA values for all the compounds. The third partition coefficient (log KAW) was then calculated for each compound using the following equation: log KAW ) log KOW - KOA

(1)

FIGURE 1. Two-dimensional cluster analysis of the partitioning space. Five natural clusters (classes) in the data were identified. Colors represent the autoscaled values of the partition coefficients: green color, the mean value of a given partition coefficient (zero value on the scale); yellow and red colors, the values higher than the mean value of a given partition coefficient (higher up to 2.5 standard deviations); light and dark blue colors, the values lower than the mean values of a given partition coefficient (lower up to 2.5 standard deviations). When moving up the figure from Class I to Class V, log KOA and log KOW systematically increase, and log KAW decreases. VOL. 42, NO. 14, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

5191

TABLE 1. Relationship between the Number of Halogen Atoms and Membership in the Similarity Classes for the Studied Groups of Compounds group

PCDDs PBDDs PCDFs PBDFs PCDEs PBDEs PCBs PBBs PCNs PBNs CBzs BBzs

Class I

Class II

Class III

Class IV

0 0 0–4 0–3

0–1 0–1 0–2 0–1 0–1 0–1 0–2 0–1 1–4 1–3 5–6 4–5

2–5 2–3 3–6 2–4 2–6 2–4 3–7 2–4 5–7 4–5

6–8 4–5 7–8 5–6 7-10 5–6 8-10 5–7 8 6–7

Class V 6 –8 7–8

Results and Discussion

7–10

QSPR Models. We developed predictive single QSPR models of log KOW and log KOA for the applicability domain covering all the studied compounds (including both chloro- and bromo-analogues). The first model, which explained 92% of the variance in log KOW, was based on two descriptors, µ and SAS (eq 3):

8–10 8

6

It is worth noting that this equation assumes the negligence of the mutual solubility of water and octanol. Exploration of the Partitioning Space. The data obtained from the QSPR modeling was used to explore the similarities in the partitioning patterns of Cl- and Br-POPs, as well as to rank the compounds according to their environmental mobility. First, we calculated mean values of the partition coefficients for each individual homologue (mono-, di-, tri-, etc.) group of PCDDs, PCDFs, PCBs, PCDEs, PCNs, CBzs, PBDDs, PBDFs, PBBs, PBDEs, PBNs, and BBzs. Then, the values (characterizing the homologue groups of the studied compounds) were autoscaled to make the influence of each variable (partition coefficient) equal. The autoscaling was made according to eq 2: ˜ xij )

xij - xj σj

(2)

where x˜i,j is the transformed mean value of the jth partition coefficient for the ith homologue group; xi,j is the original mean value of the jth partition coefficient for the ith homologue group; xj j is the mean value of the jth partition coefficient calculated across all homologue groups of all studied compounds; and σj is the standard deviation of the jth partition coefficient calculated across all homologue groups of all studied compounds. We employed two-dimensional cluster analysis to search for similar partitioning patterns in the transformed data. This method permits the simultaneous comparison of samples (compounds) and variables (the partition coefficients). The Euclidean distance was used as a quantitative measure of similarity, and Ward’s method of clustering was selected to enhance differences in the variances between the clusters (28). On the basis of these clusters, we divided the studied compounds into mobility classes. The information available for chlorinated members of each class was used to characterize the class. We assumed that the rest of the compounds in the class have similar environmental fate (i.e., brominated compounds), since they have similar values of the partition coefficients. This assumption is correct, because the partition coefficients and, additionally, degradation half-lives in air, water and soil are commonly used by most of the screening multimedia models as sufficient inputs for studies of the environmental fate of chemicals. If persistence of the analyzed compounds and the emission scenario are the same, the only factor determining their fate in the modeled ecosystem is partitioning between the individual compartments (air, water, soil) (8, 29). Since the availability of the half-life values for chloro- and bromo-POPs was insufficient, we used “the worst-case scenario”, assuming that the compounds were 5192

9

perfectly persistent (nondegradable in the environment). Moreover, a period of 10 years of continuous emission to the atmosphere was assumed according to the previous work by Wania (29). Finally, we made a simple comparison between the values of the partition coefficients for pairs of chlorinated and brominated homologues (e.g., 1,2,3-triCDD versus 1,2,3triBDD), and we used these to derive quantitative relationships between the values.

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 14, 2008

log KOW ) -0.3587(0.1521) - 0.1220(0.0238)µ + 0.0247(0.0006)SAS (3) The statistics for the model are as follows: number of training compounds, n ) 178; number of compounds in the validation set, nval ) 59; F ) 1013; p < 0.001; R2 ) 0.920; Q2CV ) 0.918; Q2Ext ) 0.924; RMSEC ) 0.315; RMSECV ) 0.320; RMSEP ) 0.302. The mechanistic interpretation of the model is intuitive: chemicals with lower polarity (measured by the dipole moment) and larger surface area (SAS) are more soluble in octanol than in water, so their log KOW values are higher. The predictive ability of the second model (eq 4) was also satisfactory: log KOA ) 7.3108(1.4538) + 0.7408(0.1554)εHOMO + 0.2862(0.0056)R (4) although the statistics were slightly worse n ) 77; nval ) 26; F ) 1287; p < 0.001; R2 ) 0.972; Q2CV ) 0.970; Q2Ext ) 0.961; RMSEC ) 0.320; RMSECV ) 0.333; RMSEP ) 0.376. The interpretation of this model is also strightforward. Mean polarizability (R) is a measure of the tendency of a charge distribution to be distorted from its normal shape by an external electric field. Staikova et al. (30) demonstrated that halogenated POPs with higher polarizabilities are more soluble in octanol and have higher log KOA values than those with lower polarizabilities. This is in agreement with our results; the tendency of forming induced dipoles in octanol solution was an important factor determining the value of log KOA. In addition, the mean polarizability is correlated to the number of substituents: the polarizability increases as the number of halogen atoms increases. Also we found that HOMO is a useful descriptor, because it depends on both the substitution pattern and the type of halogen. The substitution pattern is important when comparing the log KOA values between the congeners having the same number of substituents. According to our observation, laterally (2, 3, 7, 8 positions) substituted congeners of dibenzo-p-dioxins and furans are characterized by higher log KOA values in comparison to their homologues. The same for R-substituted (1, 4, 5, 8 positions) naphthalenes. Polyhalogenated biphenyls and diphenyl ethers with more than two substituted ortho positions have lower values of the partition coefficient than their non-ortho homologues. Moreover, these congeners of PCDEs/PBDEs, which have none or only one halogen atom in ortho position and at least one substituent in para position were characterized by the highest values of log KOA within the individual homologue group. Our conclusions are in agreement with the results obtained by other authors (31–33). It should be also noted that both descriptors (R and HOMO) are able to distinguish between the presence of a Br and Cl atom in the molecule, thus they are slightly correlated.

However, since the common part of the information is very low (r2 ) 0.23), it does not influence the model. We investigated the applicability domains of the two QSPR models by means of three methods. The simplest method involved analysis of the descriptor ranges only (34), the second employed the Williams plots (12), and the third, which was developed by us, was based on a surface plot of the residual values (differences between the observed and predicted values of the partition coefficients), where x and y axis represented the descriptors used in the models (for details, see the Supporting Information). Investigation of the applicability domain confirmed that chloro- and bromoanalogues can be modeled as a single group. We found no restrictions for the first model (log KOW); all the compounds were situated within its applicability domain. The second model (log KOA) was calibrated with fewer compounds than the first model. Unsurprisingly, the predictions for the highly brominated compounds (eg., for nona- and decabromobiphenyls and diphenyl ethers) were found to be less precise than for the other species because the log KOA values were extrapolated rather than interpolated by the model (for a more detailed discussion, see the Supporting Information). However, we decided to consider those values anyway. A spreadsheet containing (i) the collected experimental data, (ii) the method for splitting the data into the training and validation sets, and (iii) the predicted values of the partition coefficients for all 1436 compounds is available in the Supporting Information.

where nClfBr is the number of chlorine atoms replaced by bromine atoms. The partition coefficients derived in this way represent only a first approximation of the actual value, but it should nevertheless be useful when the experimental data are not available. Moreover, this simple increment-based method has the advantage that it does not require molecular descriptors (i.e., R, HOMO, SAS, and µ), which have to be obtained by employing specialized quantum-mechanics or molecular modeling software.

Similarities in the Partitioning Space

Environmental Implications

By using two-dimensional cluster analysis to explore the partitioning space, we identified five natural clusters (classes) in the data (Figure 1). The classes consist of compounds with similar values for all three partition coefficients. The partition coefficients indicate that the mobility of the classes should decrease in the order Class I > Class II > Class III > Class > IV > Class V. Comparison of the coefficients for homologous Cl- and Br-POPs indicates that the brominated compounds generally have higher log KOA and log KOW values and lower log KAW values than the chlorinated compounds. This result agrees with results reported in literature (7, 9). However, a more detailed and quantitative comparison between the groups provides interesting information. Table 1 shows how the partitioning properties vary with the number of the halogen atoms for each group. For the lower-halogenated congeners, there is no significant difference between the chloro- and bromo-derivatives. For instance, both monochloro- and monobromodibenzo-p-dioxin groups fall into Class II. But the differences in the partitioning properties between chloroand bromo-homologues increase as the number of substituents increases; for example, octachlorodibenzo-p-dioxin belongs to Class IV, whereas octabromodibenzo-p-dioxin belongs to Class V. In fact, this “class-shift effect” is observed for the most highly halogenated compounds (hepta- and octa-halogenated) in all the pairs of compound types: PCDDs/ PBDDs, PCDFs/PBDFs, PCDEs/PBDEs, PCBs/PBBs, PCNs/ PBNs, and CBzs/BBzs. The “class-shift effect” probably results from relatively large difference in the atomic weight between chlorine and bromine atoms, i.e. the mass of Br atom is more than twice the mass of Cl atom. The difference in the partitioning properties becomes more evident as more chlorine atoms are replaced by heavier bromine counterparts.

For better interpretation of the environmental transport processes and the fate of the investigated POPs (the brominated compounds in particular), it was necessary to fit our results into a simple multimedia model. However, a direct calculation by means of multimedia models was impossible for all (especially brominated) compounds, because persistence data (which are necessary for such calculations) were available for only some of the Cl-POPs. Therefore, we decided to employ a partitioning-map analysis based on the map originally developed by Wania (29). Wania simulated the phase distribution of hypothetical chemicals using the GloboPOP model (35). He assumed (i) the compounds were perfectly persistent (nondegradable in the environment), (ii) they were emitted to the atmosphere, and (iii) the period of continuous emission was 10 years. Using the predicted environmental distribution and the plot of log KOA vs log KAW (the partitioning map), he divided the compounds into three main categories: “fliers” (log KAW > 0; log KOA < 6.5), “multiple hoppers” (-4 < log KAW < 0; 6 < log KOA < 10), and “single hoppers” (log KAW < 0; log KOA > 10). The categories correspond to various modes of environmental transport, including “global distillation” and the “grasshopper effect,” which are discussed in the literature (36). When our data was analyzed using this method (Figure 2), we found that compounds from Classes I and II and the majority of compounds in Class III can be qualified as “multiple hoppers,” which means that they are semivolatile substances that relatively readily exchange between the atmosphere and the surface. They are able to revolatilize and redeposit according to the changing temperature. Thus, their environmental transport occurs through a series of deposition–volatilization cycles. Consequently, multiple hoppers have the highest target-oriented long-range transport characteristics, such as the arctic contamination potential. Compounds in Classes IV and V are “single hoppers”, which means that they are considerably less mobile than the rest of the compounds and that after emission to the atmosphere they are quickly deposited on the land or water surfaces (29, 36). Note that even if our QSPR results for the highly brominated compounds are not highly precise (as indicated

Empirical Relationships between the Partition Coefficients of Chloro- And Bromo-Derivatives Inspired by the observations discussed above, we compared the partition coefficients of individual chloro- and bromoanalogues (e.g., 1,2,3,4-tetraCDD and 1,2,3,4-tetraBDD) and

pointed out that the corresponding values were strongly correlated. The correlation coefficients were R ) 0.99, R ) 0.99, and R ) 0.95 for log KOW, log KOA and log KAW, respectively. Moreover, the differences between the chlorinated and brominated analogues per one chlorine atom replaced with bromine atom were almost constant and characteristic for each type of partition coefficients. Therefore, the mean values of these differences could be calculated and used as the increments enabling to estimate the partition coefficients for a given brominated compound from the value for the corresponding chlorinated analogue, according to the eqs 5–7: log KOW [Br] ) log KOW [Cl] + 0.11 · nClfBr

(5)

log KOW [Br] ) log KOA[Cl] + 0.32 · nClfBr

(6)

log KAW [Br] ) log KAW [Cl] - 0.21 · nClfBr

(7)

VOL. 42, NO. 14, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

5193

homologues were found in fish by many authors in high concentrations (39). Further studies in this area should be devoted to determination of the environmental persistence (i.e., degradation half-lives) of the individual Br-POPs and to carrying out more complex multimedia modeling that also includes detailed information on the persistence. Although our results seem to indicate that Br-POPs are less mobile than their chlorinated analogues, long-range atmospheric transport and bioaccumulation probably still play important roles in the environmental fate of the former. If environmental Br-POP levels continue to increase, these compounds may become a serious environmental problem over the next decades.

Acknowledgments

FIGURE 2. Partitioning map for the studied compounds fitted into categories of environmental mobility introduced by Wania (29). by the applicability domain analysis), this fact should not strongly influence the mobility ranking and the environmental interpretation of that ranking. Nona- and decabromobiphenyls and diphenyl ethers should be considered as single hoppers because their log KOA values are >10. None of the studied compounds was classified as a flier. Wania (29) introduced also a fourth category, “swimmers,” for compounds with relatively low log KOW values (e7) and log KAW values less than -2. These compounds, because of their relatively high water solubility, can be easily transported in the hydrosphere. Wania pointed out that some compounds can be both swimmers and multiple hoppers and thus be transported partially in the atmosphere and partially in the oceans. Our data seem to indicate that compounds in Classes I, II, and III can be categorized as swimmers. MacLeod and Mackay (37) developed a similar categorization scheme and concluded that such approaches can successfully describe the mechanisms of the environmental transport on both global and regional scales. However, once again we must emphasize that these classification schemes involve many simplifications. In our case, the most important simplification is the neglect of all the naturally occurring degradation processes; that is, we assumed the worst-case scenario with regard to degradation in the environment. Because the energy of the BrsC bond is lower than that of the ClsC bond, Br-POPs should be less persistent than their chlorinated analogues (9). Therefore, because of greater degradability in the atmosphere, water, and soil results in decreased mobility (29), some of the Br-POPs labeled as multiple hoppers might in fact be less mobile than that classification would suggest. From an ecotoxicological viewpoint, our results suggest that organisms can accumulate some of the brominated compounds from water (Class III) and from sediments and soil (Class III and IV). The compounds could also be biomagnified in trophic pyramids. These conclusions are supported by the predicted log KOW values between 5 and 8 (38). However, the class-shift effect and the mentioned difference in persistence, should determinate also a difference in bioaccumulation and biomagnification between the chloro- and bromo-analogues, which should be observed especially for the higher-substituted congeners. This hypothesis is supported by the results of a study published by Haglund et al. (6), who evaluated levels of mono- through heptaBDDs/Fs in perch (Perca fluviatilis) from Swedish Baltic waters and detected only up to tetrabrominated congeners in the fish body. In the case of polychlorinated dibenzo-pdioxins and dibenzofuranes, the higher (even octaCDDs/Fs) 5194

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 14, 2008

T.P. thanks the Japan Society for the Promotion of Science (JSPS) for a Postdoctoral Fellowship for Foreign Researchers and the National Institute for Environmental Studies for hosting him as a JSPS fellow. Part of this study is supported by the Environment Technology Development Fund, Japan. M.H. holds the award for the Foundation for the Development of University of Gdan ´ sk (FRUG). We thank Takeo Sakurai, Yoshitaka Imaizumi and Jun Kobayashi for their helpful suggestions. We also thank to prof. Guibin Jiang, the Associate Editor of ES&T, and anonymous reviewers for their detailed consideration of our work and valuable comments that have led to improvement of the article.

Supporting Information Available A detailed presentation of the QSPR models, including discussion on their applicability domains and a Microsoft Excel spreadsheet that contains (i) the collected experimental data, (ii) the method for splitting the data into training and validation sets, and (iii) the predicted values of the partition coefficients for the 1436 compounds. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) Moon, H. B.; Kannan, K.; Choi, M.; Choi, H. G. Polybrominated diphenyl ethers (PBDEs) in marine sediments from industrialized bays of Korea. Mar. Pollut. Bull. 2007, 54, 1402–1412. (2) Covaci, A.; Voorspoels, S.; Vetter, W.; Gelbin, A.; Jorens, P. G.; Blust, R.; Neels, H. Anthropogenic and naturally occurring organobrominated compounds in fish oil dietary supplements. Environ. Sci. Technol. 2007, 41, 5237–5244. (3) Choi, J. W.; Fujimaki, S.; Kitamura, K.; Hashimoto, S.; Ito, H.; Suzuki, N.; Sakai, S.; Morita, M. Polybrominated dibenzo-pdioxins, dibenzofurans, and diphenyl ethers in Japanese human adipose tissue. Environ. Sci. Technol. 2003, 37, 817–821. (4) Wang, L. C.; Chang-Chien, G. P. Characterizing the emissions of polybrominated dibenzo-p-dioxins and dibenzofurans from municipal and industrial waste incinerators. Environ. Sci. Technol. 2007, 41, 1159–1165. (5) Hanari, N.; Kannan, K.; Miyake, Y.; Okazawa, T.; Kodavanti, P. R. S.; Aldous, K. M.; Yamashita, N. Occurrence of polybrominated biphenyls, polybrominated dibenzo-p-dioxins, and polybrominated dibenzofurans as impurities in commercial polybrominated diphenyl ether mixtures. Environ. Sci. Technol. 2006, 40, 4400–4405. (6) Haglund, P.; Malmvarn, A.; Bergek, S.; Bignert, A.; Kautsky, L.; Nakano, T.; Wiberg, K.; Asplund, L. Brominated dibenzo-pdioxins: A new class of marine toxins. Environ. Sci. Technol. 2007, 41, 3069–3074. (7) Birnbaum, L. S.; Staskal, D. F.; Diliberto, J. J. Health effects of polybrominated dibenzo-p-dioxins (PBDDs) and dibenzofurans (PBDFs). Environ. Int. 2003, 29, 855–860. (8) Fenner, K.; Scheringer, M.; MacLeod, M.; Matthies, M.; McKone, T.; Stroebe, M.; Beyer, A.; Bonnell, M.; Le Gall, A. C.; Klasmeier, J.; Mackay, D.; Van De Meent, D.; Pennington, D.; Scharenberg, B.; Suzuki, N.; Wania, F. Comparing estimates of persistence and long-range transport potential among multimedia models. Environ. Sci. Technol. 2005, 39, 1932–1942. (9) Polybrominated Dibenzo-P-Dioxins and Furans; World Health Organization: Geneva, 1998; http://inchem.org/documents/ ehc/ehc/ehc205.htm (Accessed: Jan, 2008).

(10) Mackay, D.; Shiu, W. Y.; Ma, K.-C.; Lee, S. C. Handbook of Physical-Chemical Properties and Environmental Fate for Organic Chemicals, 2nd ed.; Taylor & Francis: Boca Raton, London, New York, 2007. (11) Cronin, M. T. D.; Livingstone, D. J. Predicting chemical toxicity and fate; CRC Press: Boca Raton, FL, 2004. (12) Gramatica, P. Principles of QSAR models validation: internal and external. QSAR Comb. Sci. 2007, 26, 694–670. (13) Pontolillo, J.; Eganhouse, R. P. The Search for Reliable Aqueous Solubility (SW) and Octanol-Water Partition Coefficient (KOW) Data for Hydrophobic Organic Compounds: DDT and DDE As a Case Study, Water-Resources Investigations Report 01-4201; U.S. Geological Survey: Reston, VA, 2001. (14) Kurz, J.; Ballschmiter, K. Vapour pressures, aqueous solubilities, Henry’s law constants, partition coefficients between gas/water (Kgw), n-octanol/water (Kow) and gas/n-octanol (Kgo) of 106 polychlorinated diphenyl ethers (PCDE). Chemosphere 1999, 38, 573–586. (15) Hayward, S. J.; Lei, Y. D.; Wania, F. Comparative evaluation of three high-performance liquid chromatography-based KOW estimation methods for highly hydrophobic organic compounds: polybrominated diphenyl ethers and hexabromocyclododecane. Environ. Toxicol. Chem. 2006, 25, 2018–2027. (16) Doucette, W. J.; Andren, A. W. Correlation of octanol water partition-coefficients and total molecular-surface area for highly hydrophobic aromatic-compounds. Environ. Sci. Technol. 1987, 21, 821–824. (17) Opperhuizen, A.; van der Velde, E. W.; Gobas, F. A. P. C.; Liem, D. A. K.; van der Steen, J. M. D. Relationships between bioconcentration in fish and steric factors of hydrophobic chemicals. Chemosphere 1985, 14, 1871–1896. (18) Harner, T.; Green, N. J. L.; Jones, K. C. Measurements of octanolair partition coefficients for PCDD/Fs: A tool in assessing airsoil equilibrium status. Environ. Sci. Technol. 2000, 34, 3109– 3114. (19) Harner, T.; Shoeib, M. Measurements of octanol-air partition coefficients (KOA) for polybrominated diphenyl ethers (PBDEs): Predicting partitioning in the environment. J. Chem. Eng. Data 2002, 47, 228–232. (20) Wania, F.; Lei, Y. D.; Harner, T. Estimating octanol-air partition coefficients of nonpolar semivolatile organic compounds from gas chromatographic retention times. Anal. Chem. 2002, 74, 3476–3483. (21) Kömp, P.; McLachlan, M. S. Octanol/air partitioning of polychlorinated biphenyls. Environ. Toxicol. Chem. 1997, 16, 2433– 2437. (22) Su, Y. S.; Lei, Y. D.; Daly, G. L.; Wania, F. Determination of octanol-air partition coefficient (KOA) values for chlorobenzenes and polychlorinated naphthalenes from gas chromatographic retention times. J. Chem. Eng. Data 2002, 47, 449–455. (23) Harner, T.; Bidleman, T. F. Measurement of octanol-air partition coefficients for polycyclic aromatic hydrocarbons and polychlorinated naphthalenes. J. Chem. Eng. Data 1998, 43, 40–46. (24) Haranczyk, M.; Puzyn, T.; Sadowski, P. ConGENER - A tool for modeling of the congeneric sets of environmental pollutants. QSAR Comb. Sci. 2008, DOI: 10.1002/qsar.200710149.

(25) Stewart, J. J. P. MOPAC2007, ver. 7.065W; Stewart Computational Chemistry 2007; http://OpenMOPAC.net (Accessed: Jan, 2008). (26) Puzyn, T.; Suzuki, N.; Haranczyk, M.; Rak, J. Calculation of quantum-mechanical descriptors for QSPR at the DFT level: Is it necessary? J. Chem. Inf. Model. 2008, DOI: 10/1021/ci800021p. (27) Topliss, J. G.; Edwards, R. P. Chance factors in studies of quantitative structure-activity relationships. J. Med. Chem. 1979, 22, 1238–1244. (28) Ward, J. H. Hierarchical Grouping to Optimize an Objective Function. J. Am. Stat. Assoc. 1963, 58, 236–244. (29) Wania, F. Potential of degradable organic chemicals for absolute and relative enrichment in the arctic. Environ. Sci. Technol. 2006, 40, 569–577. (30) Staikova, M.; Wania, F.; Donaldson, D. J. Molecular polarizability as a single-parameter predictor of vapour pressures and octanolair partitioning coefficients of non-polar compounds: a priori approach and results. Atmos. Environ. 2004, 38, 213–225. (31) Öberg, T. Prediction of vapour pressures for halogenated diphenyl ether congeners from molecular descriptors. Environ. Sci. Pollut. Res. 2002, 9, 405–411. (32) Chen, J.; Xue, X.; Schramm, K.-W.; Quan, X.; Yang, F.; Kettrup, A. Quantitative structure-property relationships for octanol-air partition coefficients of polychlorinated biphenyls. Chemosphere 2002, 48, 535–544. (33) Zeng, X. L.; Wang, Z. Y.; Ge, Z. G.; Liu, H. X. Quantitative structureproperty relationships for predicting subcooled liquid vapor pressure (PL) of 209 polychlorinated diphenyl ethers (PCDEs) by DFT and the position of Cl substitution (PCS) methods. Atmos. Environ. 2007, 41, 3590–3603. (34) Netzeva, T. I.; Worth, A.; Aldenberg, T.; Benigni, R.; Cronin, M. T.; Gramatica, P.; Jaworska, J. S.; Kahn, S.; Klopman, G.; Marchant, C. A.; Myatt, G.; Nikolova-Jeliazkova, N.; Patlewicz, G. Y.; Perkins, R.; Roberts, D.; Schultz, T.; Stanton, D. W.; van de Sandt, J. J.; Tong, W.; Veith, G.; Yang, C. Current status of methods for defining the applicability domain of (quantitative) structure-activity relationships. The report and recommendations of ECVAM Workshop 52. ATLA;Altern. Lab. Anim. 2005, 33, 155–173. (35) Wania, F.; Daly, G. L. Estimating the contribution of degradation in air and deposition to the deep sea to the global loss of PCBs. Atmos. Environ. 2002, 36, 5581–5593. (36) Gouin, T.; Mackay, D.; Jones, K. C.; Harner, T.; Meijer, S. N. Evidence for the ”grasshopper” effect and fractionation during long-range atmospheric transport of organic contaminants. Environ. Pollut. 2004, 128, 139–148. (37) MacLeod, M.; Mackay, D. Modeling transport and deposition of contaminants to ecosystems of concern: a case study for the Laurentian Great Lakes. Environ. Pollut. 2004, 128, 241–250. (38) Czub, G.; McLachlan, M. S. Bioaccumulation potential of persistent organic chemicals in humans. Environ. Sci. Technol. 2004, 38, 2406–2412. (39) Sakurai, T.; Kim, J. G.; Suzuki, N.; Matsuo, T.; Li, D. Q.; Yao, Y. A.; Masunaga, S.; Nakanishi, J. Polychlorinated dibenzo-p-dioxins and dibenzofurans in sediment, soil, fish, shellfish and crab samples from Tokyo Bay area, Japan. Chemosphere 2000, 40, 627–640.

ES8002348

VOL. 42, NO. 14, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

5195