Evaluating the Salting-Out Effect on the Organic ... - ACS Publications

Jul 11, 2016 - included in the training set, which may reflect differences in the salting-out effect on partitioning to organic carbon versus on solub...
0 downloads 11 Views 2MB Size
Article pubs.acs.org/jced

Evaluating the Salting-Out Effect on the Organic Carbon/Water Partition Ratios (KOC and KDOC) of Linear and Cyclic Volatile Methylsiloxanes: Measurements and Polyparameter Linear Free Energy Relationships Dimitri Panagopoulos,*,† Amelie Kierkegaard,† Annika Jahnke,‡ and Matthew MacLeod† †

Department of Environmental Science and Analytical Chemistry, ACES, Stockholm University, Svante Arrhenius väg 8, SE-114 18 Stockholm, Sweden ‡ Department of Cell Toxicology, Helmholtz Centre for Environmental Research, UFZ, Permoserstr. 15, DE-04318 Leipzig, Germany S Supporting Information *

ABSTRACT: Dissolved inorganic salts influence the partitioning of organic chemicals between water and sorbents. We present new measurements of the salting-out constants (Ks) for partition ratios between water and organic carbon (KOC) and between water and dissolved organic carbon (KDOC) of three cyclic volatile methylsiloxanes (cVMS), two linear volatile methylsiloxanes (lVMS), three polychlorinated biphenyls (PCBs), and α-hexachlorocyclohexane (α-HCH). Ks, KOC, and KDOC were derived from volatilization rates of the chemicals from mixtures of water and organic carbon with varying concentrations of sodium chloride in a purge-and-trap system. KOC and KDOC values at different salinities were determined by fitting their values to reproduce observed volatilization rates using a fugacity-based multimedia model and assuming first-order kinetics for volatilization. The Ks values of cVMS and lVMS ranged from 0.16−0.76. The log KOC of cVMS and lVMS in fresh water interpolated from our measurements ranged from 5.20 to 7.36 and the log KDOC values from 5.04 to 6.72. Polyparameter linear free energy relationships (PP-LFERs) trained with data sets without measurements for siloxanes failed to accurately describe the log KOC and log KDOC of cVMS and lVMS. Including our measurements for cVMS and lVMS substantially improved the fit. PP-LFERs trained with data for Ks from solubility measurements do not describe our new measurements well regardless of whether or not they are included in the training set, which may reflect differences in the salting-out effect on partitioning to organic carbon versus on solubility.



s

INTRODUCTION

K i,salt = K i10 K Cs

Inorganic salts dissolved in water influence the partitioning of organic chemicals.1,2 When inorganic salts are present, water molecules interact strongly with the ions and form hydration shells.1,2 Hydration shells make the water more ordered and more compressible and increase the free energy cost to form aqueous cavities that could accommodate organic chemicals.1,2 The “salting-out effect” can significantly influence the partitioning of chemicals in aquatic systems with dissolved salts, such as seawater.2−4 The relationship between salinity and the solubility of an organic chemical in water can be described as2

where Ki is the equilibrium partition ratio of an organic chemical between an aqueous phase and a sorbent and Ki,salt is the equilibrium partition ratio between an aqueous phase containing salt and the sorbent. Strictly speaking, the salting-out constants Ks in eq 1 and eq 2 are not identical. The two values will be similar only if the effect of dissolved salts on the activity of organic chemicals in water is much larger than their effect on the activity of organic chemicals in the pure phase or activity in the sorbent. Available measurements show that Ks depends on the organic chemical and the type of dissolved salt.2−4 Most Ks measurements are for solutions of sodium chloride (NaCl), and fewer are for ammonium sulfate ((NH4)2SO4).2−4 However, the data available in the literature, even for common salts like NaCl, are very limited.2−4

s

Sw,salt = Sw10 K Cs

(1)

where Sw is the solubility of an organic chemical in pure water, Sw,salt is the solubility in saltwater, Ks is the salting-out constant (L/mol) and Cs is the concentration of salt (mol/L). Similarly, the salting-out effect on partition ratios between water and a sorbent can be described as2 © XXXX American Chemical Society

(2)

Received: March 3, 2016 Accepted: June 14, 2016

A

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

(α-HCH), which is one of two chemicals, along with 1,4dichlorobenzene (1,4-DCB), that were used as reference chemicals to calibrate our model. We also attempted Ks measurements of 1,4-DCB but the volatilization rates for this substance were too fast and the differences in the volatilization rates at the different salinities were too small for reliable quantification. By interpolating our measurements to zero salinity we provide new measurements of KOC and KDOC for the siloxanes and PCBs. Previous studies have demonstrated that cVMS are outside the applicability domain of PP-LFERs unless they are included in the training set.18,22 In this study we evaluate the performance and applicability domain of PPLFERs with new measurements of Ks, KOC, and KDOC for PCBs, cVMS, lVMS and α-HCH combined with measurements for organic chemicals collected from the literature.

The salting-out effect is potentially important for the environmental fate of hydrophobic organic chemicals with large molar volumes that are released in coastal areas. These chemicals are expected to have large Ks because of the high energy cost to form large cavities in water.3 Chemicals in effluents from wastewater treatment plants may be driven to volatilize or sorb to, e.g., organic carbon when entering seawater because of the change in salinity. Cyclic volatile methylsiloxanes (cVMS) and linear volatile methylsiloxanes (lVMS) are highly hydrophobic,5,6 have large molar volumes, and can be released to the environment via wastewater treatment plants.7−9 Among other uses, cVMS are carriers in personal care products and monomers in the production of silicone polymers.7−9 According to the report of Environment Canada on siloxanes and silicones,10 lVMS are primarily used as chemical intermediates in the production of other silicone polymers and are only used in personal care products on a small scale. In most analyses of personal care products, lVMS appeared in much smaller quantities than cVMS.11−13 However, in the study of Lu et al.,14 some personal care products from the Chinese market, especially skin lotions and makeup, had higher concentrations of lVMS than cVMS. The residence times of lVMS and cVMS that are released to surface water depend strongly on the extent of partitioning to organic carbon in water, suspended particles, and sediments.15−18 Therefore, the partition ratio of cVMS between organic carbon and water is a key parameter required to understand the environmental fate of cVMS. A classical challenge when trying to measure partition ratios of extremely hydrophobic chemicals, such as cVMS and lVMS, is the accurate determination of their low concentrations in water. In a previous study,18 we developed a method that allows for indirect measurements of the partition ratios between water and organic carbon (KOC) and between water and dissolved organic carbon (KDOC), circumventing the problem of measuring the concentrations in the dissolved phase in water. The method employs a purge-and-trap system, where the extent of chemical sorption to organic carbon is determined by the volatilization rates of the chemicals from the system under the assumption that volatilization follows first-order kinetics. Polyparameter linear free energy relationships (PP-LFERs) that we developed in our previous work indicate that the KOC and KDOC of cVMS are primarily controlled by their large molar volumes, which implies a high energy cost for cavity formation in water.18 Here we report measurements of KOC and KDOC of cVMS and lVMS at varying salinities and Ks values for these partition ratios as defined in eq 2. Our hypothesis is that the effect of salinity on sorption to organic carbon is stronger for cVMS and lVMS than for polychlorinated biphenyls (PCBs), which have similar KOC16−20 but smaller molar volumes.21 To test that hypothesis, we measured the KOC (lake sediment) and KDOC (Suwannee River fulvic acid) of three cVMS (octamethylcyclotetrasiloxane (D4), decamethylcyclopentasiloxane (D5), and dodecamethylcyclohexasiloxane (D6)), two lVMS (decamethyltetrasiloxane (L4) and dodecamethylpentasiloxane (L5)), and three PCBs (PCB 28, 52, and 180) at different salinities and compared our measurements to each other, to previously published measurements, and to predictions made by PP-LFERs. We also attempted measurements for tetradecamethylhexasiloxane (L6), but the volatilization rates in our purge-and-trap system for this substance were too low for reliable quantification. We also report new measurements of Ks of α-hexachlorocyclohexane



MATERIALS AND METHODS Materials. The chemicals used in this study were purchased from the following companies: D4, D5, 1,4-DCB, α-HCH, Aldrin standards, methanol, potassium hydroxide (KOH), and NaCl from Sigma-Aldrich Sweden AB (Stockholm, Sweden); D6 from Fluorochem (Derbyshire, UK); 13C4-D4, 13C5-D5, and 13 C6-D6 from Moravek Biochemicals Inc. (Brea, CA, USA); PCB 28, PCB 52, PCB 53, PCB 153, 13C12-PCB 28, 13C12-PCB 52, and 13C12-PCB 153 from Larodan (Solna, Sweden); Isolute ENV+ resin (hydroxylated polystyrene−divinylbenzene copolymer) from Biotage AB (Uppsala, Sweden); dichloromethane (DCM) (SupraSolv) and n-hexane (LiChrosolv) from Merck (Darmstadt, Germany); concentrated sulfuric acid (98%) from BDH AnalaR (Poole, England); and Suwannee River standard fulvic acid reference material (1R101F-1) from the International Humic Substances Society (St Paul, MN, USA). Water was filtered and deionized using a Milli-Q system (Merck Millipore, Solna, Sweden). Sediment with an organic carbon content of 6.4% is the same material used in previous studies in our group and was collected from Lake Ången, Sweden.18,24 Methods. The methods are described in our previous study18 with a few modifications. In brief, Ks values were measured for three cVMS (D4, D5, and D6), two lVMS (L4 and L5), three PCBs (PCB 28, PCB 52, and PCB153), and α-HCH. The Ks values were calculated from measurements of KOC and KDOC of the chemicals at five different NaCl concentrations (0, 0.5, 1, 1.5, and 2 mol/L). In the KOC experiments, cVMS, lVMS, PCBs, 1,4-DCB, and α-HCH were spiked into field sediment representing bulk organic carbon. The procedure for spiking the sediment is described in Text S1 in the Supporting Information. The amounts of the chemicals in the spiking solution, the final amounts in the sediment, and the spiking efficiencies are presented in Table S1. The spiked sediment was then immersed into 300 mL of continuously stirred water in the purge-and-trap system. In the KDOC experiments, 1.5 mg of Suwannee River fulvic acid was spiked with 10 μL of hexane containing cVMS, lVMS, PCBs, 1,4-DCB, and α-HCH. The amounts of the chemicals in the spiking solution, the amounts in the DOC, and the spiking efficiencies are presented in Table S2. The sample was allowed to dry for 30 min and then transferred into 300 mL of continuously stirred water in the purge-and-trap system. In both the KOC and KDOC experiments, the headspace of the system (∼300 mL) was purged with a stream of nitrogen (25 ± 0.5 mL/min, filtered through an ENV+ column), which carried chemicals that entered the headspace to a solid-phase extraction B

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

1,4-DCB and α-HCH were calculated using the PP-LFER of Nguyen et al.,25 and the KDOC values were calculated using the PP-LFER of Neale et al.26 for Suwannee River fulvic acid KDOC. The calibration of the MTCs was done by adjusting the MTCs in the model so that the modeled volatilization rates matched the measured volatilization rates for 1,4-DCB and α-HCH. PP-LFERs. For Ks, we constructed new PP-LFERs as suggested by Goss23 using our Ks measurements and a Ks data set derived from solubility measurements made by Endo et al.3 We compared our measurements to the estimates given by the PP-LFER before and after including our data in the training set. The performance of all of the PP-LFERs was assessed quantitatively by calculating the root-mean-square error (RMSE) of the relationships. For KOC, we constructed new PP-LFERs using the average log KOC measurements for each chemical from this study and our previous study18 and the data set of measured KOC for organic chemicals compiled by Nguyen et al.25 For KDOC, we constructed new PP-LFERs for Suwannee River fulvic acid KDOC using our new measurements and the data set compiled by Neale et al.26 The solute descriptors used in the PP-LFER calculations are presented in Table S3. The robustness of the PP-LFERs was evaluated by internal cross-validation using the leave-one-out approach, following the example of Zhao et al.27 Instead of leaving out individual measurements, we divided the data sets into groups of compounds with similar structures, and PP-LFERs were created that left out one of the groups each. In each case, the PP-LFERs were used to predict the Ks, KOC, or KDOC values for the compounds that were in the group that was left out of the training set. The cross-validation root-mean-square error (CV RMSE) for the entire data set was then calculated as

(SPE) column containing 25 mg of ENV+. After approximately 2, 4, 21, 29, and 48 h in the KOC experiments and after approximately 2, 4, 8, 22, 30, 45, 72, and 97 in the KDOC experiments, the ENV+ columns were exchanged, and the trapped chemicals were immediately eluted with 1 mL of DCM. The ENV+ columns were tested for breakthrough, and no significant losses were observed. At the end of the experiment, the bulk water including the sediment or fulvic acid was analyzed to assess the mass balance of the system. Extraction blanks, which included water, extraction solvents, and ENV+, were run along with every experiment. 1,4-DCB and α-HCH were used as benchmark chemicals to calibrate the mass transfer coefficients at the air− water interface in the mass balance model of the purge-and-trap system. The PCBs were used as reference chemicals to verify the performance of the method because of their wellcharacterized partitioning properties and Ks values. Partition ratios were determined by fitting the measured volatilization rates of chemicals with a fugacity-based multimedia model assuming that volatilization follows first-order kinetics, as described below.



MODELING Determining KOC and KDOC. The modeling is described in detail in our previous study.18 In brief, we used a first-order kinetic model to describe the volatilization of cVMS, lVMS, PCBs, α-HCH, and 1,4-DCB from bulk water in the purge-andtrap system. The model (eq 3) accounts for the formation of a nonavailable fraction and degradation of the chemicals in the system: I(t ) = [100 − b(t )]e−kt +b(t )

(3)

where I(t) is the percentage of the chemical remaining in bulk water at time t, k is the volatilization rate constant of the chemical, and b(t) is the percentage of the chemical at time t that is not available for volatilization because of the formation of a nonavailable fraction and/or degradation. We used a first-order kinetic model (eq 4) to describe the formation of the nonavailable fraction, and we assumed that 100% of the chemical was available at time zero:

b(t ) = B(1 − e−kbt )

n

CV RMSE =

∑i = 1 (yexp − ypred )2 n

(5)

where yexp is the experimentally measured parameter value, ypred is the value of the parameter predicted by the PP-LFER, and n is the total number of compounds in each data set. We applied internal cross-validation analysis to predict Ks, KOC, and KDOC using data sets that either excluded or included the siloxanes. If the data for the siloxanes are outliers that are not consistent with the other data, then the CV RMSEs predicted by PP-LFERs developed from data sets that include siloxanes will be significantly higher than the corresponding CV RMSEs from data sets where they are excluded. Conversely, if the data for the siloxanes are consistent with the other data, then PP-LFERs developed from data sets that include siloxanes will have CV RMSEs that are similar to those for PP-LFERs developed from data sets excluding siloxanes.

(4)

where B is the maximum percentage of the nonavailable fraction and kb is the formation rate constant. The model was fit to the measured values of I(t) at various time points using leastsquares minimization by optimizing values of k, B, and kb in eq 3 and eq 4 with the added constraint of B ≤ 20 as described by Panagopoulos et al.18 For the least-squares minimization, we used the Solver function in Microsoft Excel. We used a fugacity-based multimedia model parametrized to describe the experimental system and to calculate the KOC and KDOC values from the experimentally determined k value. The fugacity model assumes equilibrium partitioning between organic carbon and water and nonequilibrium conditions between air and water. The rate of volatilization of chemicals calculated in the model depends on KOC or KDOC and the mass transfer coefficients on the air and water sides of the interface (MTCa and MTCw, respectively). The measured volatilization rates of 1,4-DCB and α-HCH at 0 mol/L NaCl and their KOC and KDOC were used to calibrate the MTCs, and then the calibrated model was used to fit observed volatilization rates for other chemicals by adjusting KOC or KDOC. The KOC values of



RESULTS Mass Balance Control. The average total recoveries from ENV+ traps and total extraction of bulk water at the end of the KOC experiments (n = 5) ranged from 71.4% to 101% for cVMS and lVMS and from 72.2% to 112% for PCBs and the benchmark chemicals. The average total recoveries in the KDOC experiments (n = 5) ranged from 67.9% to 84.6% for cVMS and lVMS and from 66.6% to 106% for PCBs and the benchmark chemicals (Table S4). The recoveries were calculated on the basis of the amount of chemical collected summing up the purge-and-trap and water/organic carbon samples relative to C

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

Figure 1. Percentages of spiked cyclic (top panels) and linear (bottom panels) volatile methylsiloxanes remaining in the purge-and-trap system as a function of time in five experiments at different salinities to measure KOC. Symbols are experimental data for cVMS and lVMS at 0, 0.5, 1, 1.5, and 2 mol/L NaCl. The lines are the model fits using eq 3 and eq 4. I(t) is the amount of the chemical that has not volatilized and been trapped by the ENV+ sorbent at time t, and I0 is the amount spiked into the system. The model fitting parameters for eq 3 and eq 4 can be found in Table S5. The different scales of the y axes in the figures should be noted; they were adjusted so that the fits of the model to the experimental data became clearly visible.

Figure 2. Measured log KOC of cVMS and lVMS as a function of NaCl concentration. The lines are the best-fit trend lines to the experimental data (dots), and the shaded areas represent the 95% confidence intervals of the trend lines. The slope of the linear regression equation is the Ks value, and the intercept is the log KOC measurement at 0 M salinity.

Measurements. Volatilization Curves. The volatilization losses for cVMS and lVMS over time in the KOC experiments are presented in Figure 1 along with model fits, and those in the KDOC experiments are presented in Figure S1. The volatilization losses for PCBs, 1,4-DCB, and α-HCH are presented in Figures S2 and S3. Volatilization rates and correlation coefficients (R2) derived from model fits for all of the chemicals are shown in detail in Tables S5 and S6. Out of 110 observations, the nonavailable fraction, B, at the end of the experiments was found to be between 0% and 10% for 54 cases and between

the measured amount of chemical in the spiked sediment and in the Suwannee River fulvic acid. In most cases the concentrations in the extraction blanks were lower than 15% of the concentration of the chemicals in the samples, and only in five blanks were the concentrations of the chemicals between 15% and 20%. No blank correction was made. Generally, lVMS showed lower background levels than cVMS and the PCBs showed lower background levels than lVMS. D

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

Figure 3. Measured log KDOC of cVMS and lVMS as a function of NaCl concentration. The lines are the best-fit trend lines to the experimental data (dots), and the shaded areas represent the 95% confidence intervals of the trend lines. The slope of the linear regression equation is the Ks value, and the intercept is the log KDOC measurement at 0 M salinity.

Table 1. Ks Measurements for All of the Chemicals in This Study; The Numbers in Parentheses Are the Standard Errors of the Measurements

10% and 20% for 56 cases (Tables S5 and S6). For the benchmark chemicals, B was in all cases between 0% and 5%. Out of 110 curve fits to experimental data, 82 had R2 > 0.9 and 104 had R2 > 0.7. Only six had R2 < 0.7, and the minimum value was 0.37 for PCB 153 measured at 1.5 M salt concentration (Tables S5 and S6). The values of MTCa and MTCw were derived from volatilization of 1,4-DCB and αHCH in the KOC and KDOC experiments at 0 M salt concentration and were found to be 2.06 and 0.056 m/h for the KOC experiment and 5.68 and 0.038 m/h for the KDOC experiment. The difference in the MTCa values in the two experiments could be explained by a difference in the air flow, and the difference in the MTCw values can be explained by a difference in stirring of the water. As we observed in our previous study, the MTC values in one series of purge-and-trap systems are nearly identical, but they do vary between experiments because the air flow and the stirring cannot be fully replicated. However, since the benchmarking chemicals likewise are subject to this variation, it does not affect the calculations of Ks, KOC, and KDOC. We excluded measurements from the Ks, KOC, and KDOC calculations when no measurable concentrations were observed for two or more consecutive time points in the volatilization curves. The data that were excluded were the 2 mol/L data for D6 and L5 and the 0.5, 1, 1.5, and 2 mol/L data for 1,4-DCB. The volatilization rate of L6 was too low to be reliably quantified in all of the experiments. Ks Measurements. The Ks values are the slopes of the regression lines shown in Figures 2, 3, S4, and S5. All of the Ks values with their uncertainties are presented in Table 1. The Ks values of cVMS and lVMS (Ks KOC: 0.18−0.42; Ks KDOC: 0.43− 0.76) were not higher than those of the PCBs or that of αHCH (Ks KOC: 0.20−0.45; Ks KDOC: 0.38−1.33). log KOC and log KDOC. The log KOC and log KDOC values for all of the chemicals at 0 M NaCl derived from the intercepts of

Ks KOC D4 D5 D6 L4 L5 PCB 28 PCB 52 PCB 153 α-HCH

0.42 0.34 0.16 0.25 0.18 0.20 0.29 0.42 0.45

(0.08) (0.06) (0.10) (0.07) (0.05) (0.11) (0.08) (0.06) (0.11)

Ks KDOC 0.47 0.59 0.47 0.76 0.43 0.38 0.69 1.05 1.33

(0.12) (0.16) (0.10) (0.03) (0.25) (0.03) (0.30) (0.19) (0.43)

the linear regressions are presented in Table 2. The measurements for all of the chemicals at all salt concentrations are presented in detail in Table S7. Generally, the KOC and KDOC measurements of cVMS and lVMS increased with increasing size. PP-LFER Predictions Based on Literature Data. The Ks PP-LFER trained with the data from Endo et al.3 describes our measurements for PCBs (RMSE = 0.08) and α-HCH (RMSE = 0.16) more accurately than those for cVMS (RMSE = 0.56) and lVMS (RMSE = 0.42). However, even for PCBs and α-HCH the RMSEs are larger than that of the training set (RMSE = 0.03) (Figure 4A). The PP-LFER trained with data from Nguyen et al.25 that included our measurements for PCBs but not those for siloxanes in the training set describes our measurements of log KOC for PCBs more accurately (RMSE = 0.33) than those for cVMS (RMSE = 0.88) and lVMS (RMSE = 1.41) (Figure 5A). Similarly, the PP-LFER trained with data from Neale et al.26 for Suwannee River fulvic acid DOC and our measurements for PCBs describes our measurements of log KDOC for PCBs more E

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

Table 2. log KOC and log KDOC Measurements for All of the Chemicals at 0 M NaCl Concentration Determined in This Study and Measurements from the Literature; The Numbers in Parentheses Are the Standard Errors of the Measurements log KOC this study (Lake Ången sediment) D4 D5 D6 L4 L5 PCB 28 PCB 52 PCB 153

5.20 6.47 7.29 6.24 7.36 4.77 4.66 5.19

(0.10) (0.08) (0.10) (0.09) (0.05) (0.13) (0.10) (0.07)

Panagopoulos et al.18 (Lake Ången sediment) 5.06 6.12 7.07 − − 5.36 5.44 5.52

(0.04) (0.04) (0.05)

(0.18) (0.05) (0.11)

log KDOC other lit. values 4.2217 5.1717 − 5.1617 − 3.27−5.8019,20 3.48−6.1519,20 4.42−7.6019,20

this study (Suwannee River fulvic acid) 5.04 5.91 6.68 5.90 6.72 3.70 3.44 4.09

(0.15) (0.20) (0.09) (0.04) (0.23) (0.04) (0.37) (0.23)

Panagopoulos et al.18 (Aldrich humic acid) 5.05 6.13 6.79 − − 5.36 5.54 6.04

(0.04) (0.05) (0.02)

(0.24) (0.14) (0.04)

Figure 4. Literature PP-LFER-derived Ks (A) vs revised PP-LFER predictions of Ks (B) with our measurements included in the training set of the PP-LFERs. The original training set was compiled by Endo et al.3 Solid lines show 1:1 agreement, and the gray dotted lines show the deviation from the 1:1 line equal to 2 times the RMSE of the training set.

descriptors of the PP-LFERs that included only cVMS or only lVMS. The solvent descriptors of the PP-LFERs that included both cVMS and lVMS in the training set were within the uncertainty ranges of the PP-LFERs that included only cVMS or only lVMS in the training set (Table S8). Cross-Validation of PP-LFERs. The cross-validation of the Ks PP-LFERs indicated that cVMS and lVMS are outliers in the Ks data set. When cVMS and lVMS were included in the data set, the CV RMSE for non-siloxane groups of chemicals increased from 0.05 to 0.15, and it is apparent that predictions of the PP-LFERs were biased for most of the chemical groups (Figure S6). In the case of the KOC PP-LFERs, when cVMS and lVMS were included in the KOC data set, the CV RMSE of the nonsiloxane chemical groups increased only marginally, from 0.45 to 0.47 (Figure S7). The KOC data for siloxanes are not wellpredicted by PP-LFERs that do not include siloxanes in the training set, but the data for the siloxanes are not outliers since models trained with data for siloxanes and other chemicals perform well (Figure 5). Finally, in the case of the KDOC PP-LFERs, when cVMS and lVMS were included in the KDOC data set, the CV RMSE of the non-siloxane chemical groups again increased only marginally, from 0.14 to 0.24 (Figure S8). Similar to the case for KOC, the KDOC data for siloxanes are not well-predicted by PP-LFERs that do not include siloxanes in the training set, but the data for

accurately (RMSE = 0.28) than those for cVMS (RMSE = 1.40) and lVMS (RMSE = 2.02) (Figure 6A). No log KOC or log KDOC data at 0 M were calculated for α-HCH since it was used as a benchmark chemical and its log KOC and log KDOC values at 0 M were already calculated using the PP-LFERs constructed with the data sets of Nguyen et al.25 and Neale et al.26 PP-LFER Predictions Based on Literature Data and Our Measurements. Adding our Ks measurements to the PPLFERs improved the fits for cVMS (RMSE = 0.38) and lVMS (RMSE = 0.11) but doubled the RMSE of the PP-LFER derived from only the original data set3 (RMSE = 0.06). Even though the RMSEs for cVMS and lVMS were improved, they remained high compared with those for the PCBs and the chemicals in the original data set (Figure 4B). Adding the log KOC and log KDOC measurements for cVMS or lVMS to the training sets of the log KOC and log KDOC PPLFERs improved the fits for both cVMS and lVMS (Figures 5 and 6). PP-LFERs trained with both cVMS and lVMS (Figures 5D and 6D) were better than PP-LFERs trained with only cVMS (Figures 5B and 6B) or only lVMS (Figures 5C and 6C) at describing the log KOC and log KDOC of cVMS and lVMS. This observation is also reflected in the individual descriptors of the log KOC and log KDOC PP-LFERs (Figure 7): the values of the solvent descriptors of the PP-LFERs that included both lVMS and cVMS are between the values of the solvent F

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

Figure 5. Literature PP-LFER-derived log KOC (A) vs revised PP-LFER predictions of log KOC (B−D) with our measurements included in the training set of the PP-LFERs. Literature-based PP-LFER for log KOC refers to the PP-LFER constructed using the data set compiled by Nguyen et al.25 Revised PP-LFERs refer to the PP-LFERs constructed using that data set along with our measurements. Solid lines show 1:1 agreement, and dashed lines show ±1 log unit deviations from the 1:1 lines.

increasing number of oxygen atoms in the molecules and thus also with increasing V. However, the V/B ratio is not constant for all cVMS and lVMS. The V/B ratio decreases with increasing molecular size, indicating that B becomes more influential as cVMS and lVMS increase in size. Thus, it is plausible that the effect of increasing V on Ks of siloxanes is offset by the effect of increasing B. Previous studies have interpreted the salting-out effect on chemicals with reference to the molar volume. In their review of the effect of salts on the solubility of organic chemicals, Xie at al.28 presented a moderate correlation between molar volume and Ks (R2 = 0.49). In their study on salting-out effects for a diverse set of chemicals, Endo et al.3 also reported a moderate correlation between Ks and McGowan molar volume (R2 = 0.49). Our results suggest that the degree of salting-out of siloxanes is not solely related to the molar volume (Figure S9). Other studies have also demonstrated that molar volume does not solely determine Ks.3,4,29 For example, polycyclic aromatic hydrocarbons (PAHs) have large molar volumes, and one might expect them to exhibit high Ks. However, previous studies have presented evidence that there is no correlation between the Ks and molar volume or hydrophobicity of PAHs, which, like siloxanes, are substances that can act as H-bond acceptors in specific interactions with water.2,29 Most Ks measurements reported in the literature evaluate the effect of salinity on the solubility of organic chemicals, whereas we measured the effect on partitioning between organic carbon

the siloxanes are not strong outliers in models that include siloxanes with other chemicals in the training set (Figure 6).



DISCUSSION Mass Balance. Total recoveries ranged from 66.6% to 112%. The maximum recovery differed from 100% by 12%, while the minimum differed from 100% by almost 33%. This difference indicates that some losses occurred during the experiments. Possible loss mechanisms are formation of a nonavailable fraction that resisted solvent extraction, hydrolysis, and leakage when changing cartridges. The potential loss mechanisms are discussed in detail in our previous study.18 Ks Measurements. Our results do not support our original hypothesis that the volatile methylsiloxanes would have higher Ks values than PCBs. The Ks values of cVMS and lVMS (Ks KOC: 0.18−0.42; Ks KDOC: 0.43−0.76) were in similar ranges as those of the PCBs and α-HCH (Ks KOC: 0.20−0.45; Ks KDOC: 0.38−1.33). A possible explanation is the ability of cVMS and lVMS to engage in specific H-bond-acceptor interactions with water that counteract the effect of molar volume. We found strong negative correlations between the KOC Ks values of the five siloxanes and the molecular descriptors V (molar volume), B (H-bond acceptor), and L (log hexadecane/air partition ratio) (Figure S9). Endo et al.3 suggested that compounds with high capability for specific and nonspecific interactions deviate from the general rule that Ks increases with V. cVMS and lVMS have higher B compared with PCBs, and B increases with G

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

Figure 6. Literature PP-LFER-derived log KDOC (A) vs revised PP-LFER predictions of log KDOC (B−D) with our measurements included in the training set of the PP-LFERs. Literature-based PP-LFER for log KDOC refers to the PP-LFER constructed using the data set for Suwannee River DOC compiled by Neale et al.26 Revised PP-LFERs refer to the PP-LFERs constructed using that data set along with our measurements. Solid lines show 1:1 agreement, and dashed lines show ±1 log unit deviations from the 1:1 lines.

and water. Turner30 noted that salting-out constants derived from solubility experiments are conceptually different from salting-out constants derived from partitioning experiments. In solubility experiments, a chemical molecule is salted out from water to other molecules of the same chemical until it forms a second phase that is immiscible with water, in other words, until it reaches its solubility limit. In that sense, the extent of the salting-out of a chemical depends on the molecular interactions of a chemical with water molecules and with molecules of the same chemical. In partitioning experiments, the extent of the salting-out of a chemical depends on the chemical’s interactions with water molecules and with organic carbon molecules. These differences in molecular interactions may lead to different values of Ks. Differences between the Ks measurements derived from the KOC experiments and the Ks measurements derived from the KDOC experiments were also observed. Although the exact mechanism is not fully understood, these differences are yet another indication that Ks does not only depend on the saltingout of a chemical from water but also on the sorbent to which it is salted-out. Ks, log KOC, and log KDOC Measurements and PP-LFER Predictions Based on Literature Data. Our measured Ks values for PCBs and α-HCH showed a smaller RMSE than those for cVMS and lVMS but still higher than that for the original data set. All of the data from the Endo et al.3 data set are from solubility experiments, and the deviation between the

predictions of the PP-LFERs and our measurements may be another indication that Ks measurements from solubility experiments are conceptually different from Ks measurements from partitioning experiments. The discrepancy becomes even more pronounced in the case of cVMS and lVMS. This observation is again consistent with different molecular interactions playing significant roles in solubility experiments versus partitioning experiments. The KOC of PCBs is primarily controlled by their capability for nonspecific interactions with organic carbon molecules, such as London dispersion forces and dipole−induced dipole and dipole−dipole interactions. The KOC of cVMS and lVMS, however, is primarily controlled by their large molar volumes and less by their capability for nonspecific interactions with organic carbon molecules. Therefore, one would expect that the difference between Ks from solubility experiments and Ks from partitioning experiments is smaller for PCBs than for cVMS and lVMS. Our measured log KOC values for PCBs are at the low end of the range of previously reported measurements19,20 but still within ±1 log unit deviation from the PP-LFER predictions. The log KOC values for PCBs measured in this study are all lower than those measured in our previous study using the same lake sediment as a source of organic carbon, but the measurements in the two studies agree within the associated uncertainty ranges for two of the three PCBs (Table 2).18 The only difference in the experimental setup is the amount of sediment used in the two studies. In our previous study we used H

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

Figure 7. Changes in solvent descriptors for the log KOC and log KDOC PP-LFERs when cVMS, lVMS, or both cVMS and lVMS are added to the training sets. The training sets without VMS were compiled by Nguyen et al.25 and Neale et al.26

dissolved organic substances. However, in the case of cVMS, no large differences were observed. A plausible explanation of why we did not see the same effect for cVMS as we saw for PCBs is that the KOC and KDOC of PCBs are mainly controlled by nonspecific interactions while the KOC and KDOC of cVMS are mainly controlled by their large molar volume and the high energy cost for cavity formation in water.18 Differences in the chemical composition of organic carbon could possibly result in differences in the log KOC and KDOC of PCBs but would not influence the partitioning of cVMS to the same extent. Similarly to the observations for KOC, the literature-based log KDOC PPLFER calibrated with a data set that did not include siloxanes26 failed to accurately predict our log KDOC measurements for cVMS and lVMS (Figure 6A). Ks, log KOC, and log KDOC Measurements and PP-LFER Predictions Based on Literature Data and Our Measurements. The Ks PP-LFER that included cVMS and lVMS in the training set can describe the Ks of cVMS and lVMS more accurately than the PP-LFER that was trained without cVMS and lVMS. However, even in this case, the RMSEs of cVMS and lVMS remain high relative to the RMSEs of the training set. Also, the addition of cVMS and lVMS to the training set increased the RMSE of the training set by a factor of 2. This failure of the PP-LFER to accurately describe the Ks of cVMS and lVMS even after they are included in the training set is yet another indication that Ks measurements from solubility experiments are conceptually different from Ks measurements from partitioning experiments. The log KOC PP-LFER that included cVMS in the training set of organic chemicals from Nguyen et al.25 describes the log KOC of PCBs, cVMS, and lVMS within ±1 log unit deviation from the 1:1 line (Figure 5B). Similarly, the log KOC PP-LFER that included lVMS in the training set describes PCBs, cVMS, and lVMS within ±1 log unit (Figure 5C). For predicting KOC

6 mg of sediment in 300 mL of water, while in this study we only used 2.5 mg of sediment in 300 mL of water. We chose a smaller amount of sediment in this study to speed up the elimination of chemicals from the system so that elimination curves could be derived for highly hydrophobic cVMS and lVMS and at high salinities over the course of the experiment. Our measured log KOC values for cVMS are in good agreement with previous measurements17,18 but not in good agreement with the predictions of the PP-LFER constructed from training sets of literature data that do not include siloxanes. This poor fit was also observed in previous studies and attributed to siloxanes being outside the domain of applicability of PP-LFERs that do not include siloxanes in the training set.18,22 The largest difference observed between the results of this study and our previous study18 was for D5 (0.35 log units). The log KOC of D4 was 0.14 log units higher in this study, and that of D6 was 0.11 log units higher. Experimental data for log KOC values for lVMS were found only for L4. Our measured log KOC of L4 is 0.84 log units higher than the one measured by Kozerski et al.17 This difference could be attributed to the different sources of organic carbon. In our study we used freshwater sediment from Lake Ången, Sweden, whereas Kozerski et al.17 used soil. Differences in the molecular composition of the organic carbon can influence the KOC.31 No experimental data on Suwannee River fulvic acid KDOC could be found for any of the cVMS or lVMS. Our measured log KDOC values for PCBs were in good agreement with the predictions of the PP-LFER created from the data set compiled by Neale et al.26 Our measurements of the Suwanee River log KDOC values for PCBs were lower than our previous measurements for Aldrich humic acid (Table 2). It has been previously reported26 that Aldrich humic acid has a higher absorptive capacity than naturally occurring humic acids and therefore log KDOC measurements for this humic acid usually are higher than log KDOC measurements for naturally occurring I

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

(4) Wang, C.; Lei, Y. D.; Endo, S.; Wania, F. Measuring and modeling the salting-out effect in ammonium sulfate solutions. Environ. Sci. Technol. 2014, 48, 13238−13245. (5) Xu, S.; Kropscott, B. Method for simultaneous determination of partition coefficients for cyclic volatile methylsiloxanes and dimethylsilanediol. Anal. Chem. 2012, 84, 1948−1955. (6) Xu, S.; Kropscott, B. Evaluation of the three-phase equilibrium method for measuring temperature dependence of internally consistent partition coefficient (KOW, KOA, and KAW) for volatile methylsiloxanes and trimethylsilanol. Environ. Toxicol. Chem. 2014, 33, 2702−2710. (7) Brooke, D. N.; Crookes, M. J.; Gray, D.; Robertson, S. Risk Assessment Report: Octamethylcyclotetrasiloxane; Environment Agency of Great Britain: Rotherham, U.K., 2009. (8) Brooke, D. N.; Crookes, M. J.; Gray, D.; Robertson, S. Risk Assessment Report: Decamethylcyclopentasiloxane; Environment Agency of Great Britain: Rotherham, U.K., 2009. (9) Brooke, D. N.; Crookes, M. J.; Gray, D.; Robertson, S. Risk Assessment Report: Dodecamethylcyclohexasiloxane; Environment Agency of Great Britain: Rotherham, U.K., 2009. (10) Environment Canada. Screening Assessment for the Challenge: Siloxanes and Silicones, di-Me, Hydrogen-Terminated, Chemical Abstracts Service Registry Number 70900-21-9; Environment Canada: Ottawa, ON, 2011. (11) Horii, Y.; Kannan, K. Survey of organosilicone compounds, including cyclic and linear siloxanes, in personal-care and household products. Arch. Environ. Contam. Toxicol. 2008, 55, 701−710. (12) Wang, R.; Moody, R. P.; Koniecki, D.; Zhu, J. Low molecular weight cyclic volatile methylsiloxanes in cosmetic products sold in Canada: Implication for dermal exposure. Environ. Int. 2009, 35, 900. (13) Dudzina, T.; von Goetz, N.; Bogdal, C.; Biesterbos, J. W. H.; Hungerbühler, K. Concentrations of cyclic volatile methylsiloxanes in European cosmetics and personal care products: Prerequisite for human and environmental exposure assessment. Environ. Int. 2014, 62, 86. (14) Lu, Y.; Yuan, T.; Wang, W.; Kannan, K. Concentrations and assessment of exposure to siloxanes and synthetic musks in personal care products from China. Environ. Pollut. 2011, 159, 3522−3528. (15) Whelan, M. J.; Sanders, D.; van Egmond, R. Effect of Aldrich humic acid on water-atmosphere transfer of decamethylcyclopentasiloxane. Chemosphere 2009, 74, 1111−1116. (16) Whelan, M. J.; van Egmond, R.; Gore, D.; Sanders, D. Dynamic multi-phase partitioning of decamethylcyclopentasiloxane (D5) in river water. Water Res. 2010, 44, 3679−3686. (17) Kozerski, G. E.; Xu, S.; Miller, J.; Durham, J. Determination of soil-water partition coefficients of volatile methylsiloxanes. Environ. Toxicol. Chem. 2014, 33, 1937−1945. (18) Panagopoulos, D.; Jahnke, A.; Kierkegaard, A.; MacLeod, M. Organic carbon/water and dissolved organic carbon/water partitioning of cyclic volatile methylsiloxanes: measurements and polyparameter linear free energy relationships. Environ. Sci. Technol. 2015, 49, 12161− 12168. (19) Seth, R.; Mackay, D.; Muncke, J. Estimating the organic carbon partition coefficient and its variability for hydrophobic chemicals. Environ. Sci. Technol. 1999, 33, 2390−2394. (20) Mackay, D.; Shiu, W. Y.; Ma, K. C. Illustrated Handbook of Physical-Chemical Properties and Environmental Fate for Organic Chemicals, 2nd ed.; Lewis Publishers: Chelsea, MI, 1992. (21) Endo, S.; Watanabe, N.; Ulrich, N.; Bronner, G.; Goss, K.-U. UFZ-LSER Database, version 2.1 [Internet]; Helmholtz Centre for Environmental Research-UFZ: Leipzig, Germany, 2015; available at http://www.ufz.de/index.php?en=31698&contentonly=1&lserd_ data[mvc]=Public/start (accessed Feb 19, 2016). (22) Endo, S.; Goss, K.-U. Predicting partition coefficients of polyfluorinated and organosilicon compounds using polyparameter linear free energy relationships (PP-LFERs). Environ. Sci. Technol. 2014, 48, 2776−2784.

of other VMS, we recommend the log KOC PP-LFER trained with both cVMS and lVMS (Figure 5D). PP-LFERs for log KDOC developed from the data set of Neale et al.26 with and without the cVMS and lVMS included in the training set followed a similar pattern as those for KOC described above (Figure 6). In this case, the PP-LFER estimates for L4 are consistently lower than our measurements. This error in describing the log KDOC measurement of L4 could reflect a bias in our measurement for L4 or limitations associated with the Neale et al.26 data set, which does not include substances with log KDOC as high as those of PCBs, cVMS, and lVMS. However, the chemical applicability domain of a PP-LFER does not solely depend on the range of log KOC or KDOC but also on the range and variability of the molecular descriptors of the chemicals that were included in the training set of the PP-LFER. The addition of both cVMS and lVMS to the training set of the log KDOC PP-LFER improved the fit for both cVMS and lVMS, and we recommend that the PP-LFER illustrated in Figure 6D be used for estimating log KDOC of other siloxanes. From these results we can conclude that there is some evidence to suggest that Ks measurements derived from solubility experiments are conceptually different from Ks measurements from partitioning experiments, especially for cVMS and lVMS. Including our measurements in the training set of the PP-LFER did not result in accurate descriptions of the Ks of cVMS and lVMS. On the other hand, including siloxanes in the training sets of log KOC and log KDOC PPLFERs is necessary in order for the PP-LFERs to describe the log KOC and log KDOC of siloxanes within ±1 log unit deviation. PP-LFERs that are trained with data for cVMS perform well in predicting the partitioning of lVMS, and vice versa. The measurements and PP-LFERs developed here provide a basis for modeling the fate of siloxanes released from wastewater treatment plants to freshwater or marine systems.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jced.6b00196. Description of the experimental conditions and procedures and supporting figures and tables (PDF)



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Funding

This study was funded by the Swedish Research Council FORMAS (2011-484). Notes

The authors declare no competing financial interest.



REFERENCES

(1) Millero, F. J. Chemical Oceanography, 2nd ed.; CRC Press: Boca Raton, FL, 1996. (2) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry; John Wiley & Sons: Hoboken, NJ, 2003. (3) Endo, S.; Pfennigsdorff, A.; Goss, K.-U. Salting-out effect in aqueous NaCl solutions: trends with size and polarity of solute molecules. Environ. Sci. Technol. 2012, 46, 1496−1503. J

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX

Journal of Chemical & Engineering Data

Article

(23) Goss, K.-U. Predicting the equilibrium partitioning of organic compounds using just one linear solvation energy relationship (LSER). Fluid Phase Equilib. 2005, 233, 19−22. (24) Jahnke, A.; Mayer, P.; McLachlan, M. S.; Wickström, H.; Gilbert, D.; MacLeod, M. Silicone passive equilibrium samplers as ’chemometers’ in eels and sediments of a Swedish lake. Environ. Sci.: Processes Impacts 2014, 16, 464−472. (25) Nguyen, T. H.; Goss, K.-U.; Ball, W. P. Polyparameter linear free energy relationships for estimating the equilibrium partition of organic compounds between water and the natural organic matter in soils and sediments. Environ. Sci. Technol. 2005, 39, 913−924. (26) Neale, P. A.; Escher, B. I.; Goss, K.-U.; Endo, S. Evaluating dissolved organic carbon-water partitioning using Polyparameter linear free energy relationships: Implications for the fate of disinfection byproducts. Water Res. 2012, 46, 3637−3645. (27) Zhao, Q.; Yang, K.; Li, W.; Xing, B. S. Concentration-dependent polyparameter linear free energy relationships to predict organic compound sorption on carbon nanotubes. Sci. Rep. 2014, 4, 3888. (28) Xie, W. H.; Shiu, W. Y.; Mackay, D. A review of the effect of salts on the solubility of organic compounds in seawater. Mar. Environ. Res. 1997, 44, 429−444. (29) Jonker, M. T. O.; Muijs, B. Using solid phase micro extraction to determine salting-out (Setschenow) constants for hydrophobic organic chemicals. Chemosphere 2010, 80, 223−227. (30) Turner, A. Salting-out of chemicals in estuaries: implications for contaminant partitioning and modeling. Sci. Total Environ. 2003, 314− 316, 599−612. (31) Bronner, G.; Goss, K.-U. Sorption of organic chemicals to soil organic matter: Influence of soil variability and pH dependence. Environ. Sci. Technol. 2011, 45, 1307−1312.

K

DOI: 10.1021/acs.jced.6b00196 J. Chem. Eng. Data XXXX, XXX, XXX−XXX