ARTICLE pubs.acs.org/est
Predicting Sorption of Pesticides and Other Multifunctional Organic Chemicals to Soil Organic Carbon Guido Bronner*,†,‡ and Kai-Uwe Goss*,‡ †
Institute for Biogeochemistry and Pollutant Dynamics (IBP), Swiss Federal Institute of Technology (ETH) Zurich, Universit€atstrasse 16, CH-8092 Zurich, Switzerland ‡ Helmholtz Centre for Environmental Research UFZ, Permoserstrasse 15, 04318 Leipzig, Germany
bS Supporting Information ABSTRACT: Chemicals of current environmental concern are often multifunctional and more polar and more complex than classical pollutants such as polychlorinated biphenyls (PCB) or polycyclic aromatic hydrocarbons (PAH). Traditional models for predicting the partitioning in the environment such as group contribution methods or correlations with octanol-water partitioning cannot be expected to work well for such complex chemicals. In contrast, poly parameter Linear Free Energy Relationships (pp-LFERs) have been proven to describe partitioning of polar and nonpolar chemicals in all kinds of sorbing systems. Here, a pp-LFER model for soil-water partitioning was calibrated with data for 79 polar and nonpolar compounds that cover a very wide range of the relevant intermolecular interactions. The data set used for the model calibration in this work is more diverse and covers a wider range of the chemical space than other pp-LFERs published so far. Subsequently, the experimental data for about 50 pesticides and pharmaceuticals -not involved in the model calibration- were used as independent validation of this new calibrated model. The model performs well with a standard error of 0.25 log units for fitting the calibration data and with a root-mean-square error of 0.4 log units for the pesticides and pharmaceuticals. The validation with the independent data set for pesticides and pharmaceuticals also shows that the pp-LFER model reported here performs better compared to earlier published pp-LFER models and to the traditional log Kow correlation.
’ INTRODUCTION Pesticides are often multifunctional, highly polar, and biologically active chemicals and thus complex. Because of their complex structure the prediction of their chemical properties is a challenge. Every year thousands of tons of these chemicals are introduced in the environment which makes the assessment of their environmental partitioning, fate, and toxicity mandatory. To this end, among other properties, soil organic matter (SOM)/water sorption coefficients, typically normalized to organic carbon content and denoted as Koc, is a central issue in risk assessment and appropriate predictive tools are urgently needed. The simplest, oldest, and still most often used method to predict Koc is using established correlations between log Koc and log Kow (Kow octanol/water partitioning coefficient). It has been demonstrated1 that such single-parameter Linear Free Energy Relationships (sp-LFERs) for SOM/water and other partitioning processes are principally limited in their applicability to sorbates with similar sorptive interactions, in particular nonpolar sorbates (also see Figure SI-1 right in the SI). For complex organic sorbates poor correlation between log Koc and log Kow can be expected as it has been demonstrated in theory1 and as can be seen using data provided within this paper (see Figure SI-1 left). It has been shown that the principal shortcomings of sp-LFERs can be overcome by poly parameter Linear Free Energy Relationships (pp-LFERs).1 pp-LFERs use 5 sorbate descriptors that explicitly account for all relevant types of intermolecular interactions. The original pp-LFER equation for partitioning of a chemical i between water, w, and another condensed phase, cp, was r 2010 American Chemical Society
established by Abraham and co-workers2 and has the following form log K cp=w ¼ ecp=w Ei þ scp=w Si þ acp=w Ai þ bcp=w Bi þ vcp=w Vi þ ccp=w ð1Þ The capital letters are sorbate descriptors for the various types of interactions: Ei, the excess molar refractivity and Vi, the McGowan volume (cm3/mol)/100 represent nonspecific interactions (van der Waals interactions and cavity formation).2 The remaining three descriptors stand for the H-donor property, Ai, the H-acceptor property, Bi, and Si refers to the dipolarity/polarizability of the sorbate.3 The E-descriptor has the disadvantage that for solid compounds it cannot be calculated from experimental refractive indices but has to be estimated with empirical methods from molecular structure which involves considerable uncertainty.4,5 This problem can be circumvented by using eq 2, a modified version of eq 1, in which the Ei-descriptor is replaced by Li, the logarithm of the hexadecane/air partition constant of i at 25 °C in units of [Lair/Lhexadecane]6 log Ks=w ¼ ls=w Li þ ss=w Si þ as=w Ai þ bs=w Bi þ vs=w Vi þ cs=w ð2Þ Li can be measured for liquid and many solid organic chemicals,4 except for very large and very polar molecules, for which estimates of Li will be needed. Received: July 29, 2010 Accepted: December 14, 2010 Revised: December 11, 2010 Published: December 31, 2010 1313
dx.doi.org/10.1021/es102553y | Environ. Sci. Technol. 2011, 45, 1313–1319
Environmental Science & Technology In order to turn the general forms of eqs 1 or 2 into a specific model for sorption to SOM a calibration is required from which the values for e, l, s, a, b, v, and c are derived. This is done by regressing eq 1 or 2 to a diverse set of experimentally determined sorption coefficients for sorbates with known sorbate descriptors. The applicability domain of the resulting model is directly determined by the diversity of the calibration data set. Some pp-LFER models have already been published for sorption in soil organic matter. Nguyen et al. and Poole and Poole have calibrated pp-LFER equations for sorption to SOM based on experimental literature data.7 In more recent work, Endo et al. presented a pp-LFER model for sorption at low concentrations from water into a reference peat (Pahokee Peat). Their model is based on consistent experimental data and might be representative for solid soil organic matter.8 These calibrations may however be insufficient for predicting Koc of highly polar, multifunctional sorbates because either a) the available literature data for calibration were biased to simple or nonpolar chemicals, or b) the calibration was not validated with complex sorbates, or c) the sorption data used for calibration did not refer to one soil which could lead to wrong fitting parameters if the sorption data are not well distributed over the different soils, considering the possible variability of SOM sorption properties, or d) for solid compounds the estimated descriptor of the van der Waals interaction (E value, see eq 1) can lead to predictions errors of about one log units. The quality of the sorption data for the calibration of the existing pp-LFERs have already been discussed in Nguyen et al.7 and Endo et al.8 and will not be repeated here. Instead it has to be noted that although all pp-LFERs for sorption in SOM have the same overall statistical fit quality, the single parameter in the different pp-LFERs deviate significantly and substantially, and the consequences for prediction of sorption of sorbates with extreme properties and thus high descriptor values such as multifunctional sorbates are unknown. Table SI-7 and Table SI-8 summarize the more noteworthy pp-LFER equations for Koc and the descriptor ranges of the sorbates that have been used for calibration. These ranges are far too small in comparison with the values of compound descriptors that pesticides show according to ref 5. Further, just a few large polycyclic aromatic hydrocarbons contribute already half of the range of covered E values (up to the value of 4) in former calibrations giving too much weight to this compound class for this descriptor. Niederer et al. have provided experimental data and calibrated pp-LFER equations for the sorption of up to 90 diverse chemicals from air to 10 hydrated humic and fulvic acids.9 It remained unclear though whether these data can be converted from air to water by simply using Henry’s law constants because a significant dependence of sorption coefficients of polar sorbates from the humidity content in the humic acid was found. Furthermore, these data do not contain any information on sorption to humin, the base-unextractable fraction of soil organic matter that contributes a large fraction to the total solid organic matter. Sorption nonlinearity is an effect that has to be taken into account when extrapolations over concentration ranges of several orders of magnitude have to be done. This was systematically investigated by Endo et al.10 and is not the focus of this work which is limited to a narrow concentration range. We thus consider the work of Endo et al. as complementary to the work presented here. The aim of this work was to establish a predictive model for the sorption of polar and nonpolar neutral organic chemicals from
ARTICLE
water to nonsoluble soil organic matter that overcomes the weaknesses of Kow-models and existing pp-LFERs. Our work puts an emphasis on the diversity of the calibration data by including multifunctional molecules which should allow for a highly robust model. Finally with a data set of 47 complex sorbates consisting of pesticides and several pharmaceuticals and hormones the applicability of the model to highly complex chemicals is assessed and its performance is compared to other existing models.
’ MATERIALS AND METHODS Experimental Work. We have used column (dynamic) sorption experiments and batch (static) sorption experiments for measuring sorption coefficients. The batch experiments followed a standard procedure details of which are described in the SI. Pahokee Peat (PP), from the International Humic Substances Society (IHSS), was used as sorbent (see Table SI-4 Pahokee Peat properties). Column Experiments. The column method was chosen because it offers several advantages: a) a large sorption data set can be obtained with only a small amount of sorbent, b) the conditions in the column (e.g., salt, pH) can be maintained or easily changed as desired through the eluent, and c) the fast equilibration (see below) limits the time of the sorption experiment from several minutes to a few hours thus excluding significant degradation in the aqueous solution and covalent binding to SOM for most investigated sorbates. For the column sorption experiments the Pahokee Peat (PP) was ‘diluted’ with silicon carbide SiC (trade name: “SiC dunkel F1200” diameter 3 ( 0.5 μm, density 3.16 g/cm3 obtained from “ESK-SiC GmbH, Germany) an inert, low sorbing material in a ratio PP:SiC of 3:100 and 12:100 (w/w). About 0.1 g of these mixtures was then packed as a stationary phase into chromatographic columns (0.3 cm i.d., 1.0 cm length). Milli-Q water with 5 mM CaCl2 served as mobile phase. The column was connected to an HPLC-pump and flushed with the mobile phase (5 mM CaCl2 aqueous solution) for 1 day at 0.05 mL/min to allow stabilization of the packing material (preflushing). A restrictor generating a pressure of 100 bar was connected to the outlet of the column to ensure that air was completely dissolved and thus removed from the column. The mobile phase was collected for TOC analysis. The loss of dissolved organic carbon was found to be 3.2% (w/w) with respect to the total mass of the initial oc-content. This loss of DOC is not considered as an artifact because this work is focused on solid organic matter and not on dissolved organic matter. After equilibration the column was connected to an HPLCSystem. For peak detection either a UV/vis detector (Jasco 870UV, Jasco, Japan and Lambda 1010, Bischoff Analysentechnik GmbH, Germany) or a fluorescence detector (RF-1002, Shimadzu Corporation, Japan) were used. All experiments were conducted at 25 ( 3 °C. Sorption constants, Ki peat/water concentration of compound i in the peat ð3Þ Ki peat=water concentration of compound i in water
can be derived from peak retention law Ki peat=water ¼ ðVret i - Vret i background Þ=mpeat
ð4Þ
where Vreti is the retention volume of i, Vret i background is the elution volume of i in a column filled with pure SiC, and mpeat is 1314
dx.doi.org/10.1021/es102553y |Environ. Sci. Technol. 2011, 45, 1313–1319
Environmental Science & Technology the amount of peat present in the column. The volumes Vret i and Vret i background are determined from the volumetric flow rate of the mobile phase and the retention time of the respective peaks. The retention time was obtained from their first statistical moment.11 Sorption to the chromatographic system (capillaries, frits, column wall) and SiC should be small compared to sorption to the peat in order to obtain a Vret i value being much larger than Vret i background. Experimental Koc were not converted to a standard sorbed concentration because many sorbates have an S-value higher than 1.0 and the estimation of Freundlich coefficients according to Endo et al. might be unreliable;10 instead of that the sorbed concentration range in the experiments was kept as narrow as possible. The major challenge in the column sorption experiments lies in meeting the (local sorption) equilibrium requirement. The rate limiting step in the sorption process is diffusion of the sorbate inside the sorbent matrix. In order to facilitate sorption equilibrium in our experiments the peat was micronized as follows: the peat was suspended in deionized water in portions of 20 gpeat/Lwater and stirred with Ultra-Turrax (IKA-Werke GmbH & CO. KG, 79219 Staufen, Germany) for 10 min beginning at 11000 and ending at 24000 rpm. This suspension was then introduced into a high pressure homogenization device (NS1001L Panda 2k, NIRO SOAVI S.p.A., Italy). Here the suspension was passed through a small opening under high pressure 10 times, each time using a smaller opening, resulting in a higher conveying pressure. The observation of the suspension with an optical microscope revealed no particles larger than 20 μm after the treatment. Eventually the suspension was freeze-dried to obtain a powder and then mixed with appropriate amounts of SiC in a mortar. Model Calibration. Calibration sorbates were carefully selected so that wide ranges of descriptor values are covered, while cross correlation of the descriptors within the calibration data set is minimized. Highly complex sorbates (i.e the pesticides and drugs) were consciously excluded from the calibration data set for two reasons. First the sorbate descriptors of the pesticides and pharmaceuticals have a higher uncertainty than those of more simple sorbates and for the calibration data the reliability of the descriptors has to be high. Second we eventually aimed to show that the sorption properties of even complex strucutures can be reduced to the 5 sorbate descriptors of the pp-LFER model. A list of 79 sorbates used as training set for the calibration of the pp-LFER equations can be found in Table SI-1 together with the respective descriptors. This set is more diverse than in previous calibrations in the literature for sorption to soil organic matter and covers a much wider range of interaction descriptor values: E: -0.24 to 2.06; S: 0 to 1.95; A: 0 to 0.99; B: 0 to 1.1; V: 0.64 to 2.56; L: 1.95 to 11.1. Using eq 2 the cross correlation between any one of the descriptors and the remaining ones is small (R2 < 0.76) with the exception of V and L (R2 = 0.86 and 0.91, respectively; see Table SI-3 for a cross correlation analysis). The correlation between V and L is given and thus will not cause biased predictions unless the model is extrapolated to perfluorinated compounds.12 Including these compounds could further enlarge the model applicability, though due to expected artifacts by sorption at the glass and water surface accurate determination of sorption coefficients may be a challenge. Hence, the application domain of the model equations presented in this work excludes perfluorinated compounds. Note that including chemicals with S-values higher than 2.0 into the calibration data was omitted because it would have resulted in a considerable cross-correlation between the S-value and the other, size-related descriptors V and L.
ARTICLE
Application to Pesticides and Pharmaceuticals. In a second step the experimental Koc of 56 pesticides and pharmaceuticals (evaluation data set) were compared with the predictions of the pp-LFER model established in this work to assess its applicability to complex, environmentally relevant chemicals. For the evaluation data set and for some sorbates of the calibration data set hexadecane/air partitioning coefficients (L-values) were not available in the literature and were thus experimentally determined. For the evaluation data set the sorbate descriptors (S, A, B) were refined based on the measured L values, eq 2 and the data of Tuelp et al.5 The refined descriptors were used in this study and listed in Table SI-2. The details will be published elsewhere.
’ RESULTS AND DISCUSSION Background Sorption in Column Experiments. For the calibration sorbates the background sorption on silicon carbide and the column material was measured. The retention volume as well as the percent contribution to the total retention on the 3% peat column is reported in Table SI-6. The background sorption was found to be
