Environ. Sci. Technol. 2009, 43, 7047–7054
Polymer-Water Partition Coefficients of Hydrophobic Compounds for Passive Sampling: Application of Cosolvent Models for Validation F O P P E S M E D E S , * ,†,§ RINZE W. GEERTSMA,† TON VAN DER ZANDE,† AND KEES BOOIJ‡ Ministry of Transport, Public Works and Water Management, National Institute for Coastal and Marine Management/RIKZ, P.O. Box 207, 9750 AE Haren, The Netherlands, and Royal Netherlands Institute for Sea Research, P.O. Box 59, 1790 AB Texel, The Netherlands
Received April 4, 2009. Revised manuscript received July 7, 2009. Accepted July 17, 2009.
Polymer-waterpartitioncoefficients(Kpw)ofhexachlorobenzene, 41 polychlorinated biphenyls (PCBs), and 26 polyaromatic hydrocarbons (PAHs) were determined for low-density polyethylene (LDPE) and five different silicone rubbers. Partition coefficients were determined in ultra pure water and in a range of methanol-water mixtures. Different cosolvent models for the effect of methanol concentration on the polymermixture partition coefficient (Kpm) were used to validate the Kpw in pure water. Linear regression of logKpm against the mole fraction (x) methanol over range 0 < x < 0.3 yielded the best results. The obtained logKpws were best described by a correlation with molecular weight, for PCBs in combination with the fraction of chlorine atoms in the meta and para positions (standard deviations of ∼0.08 log units). Correlations with logKow were less good (standard deviations of ∼0.21 log units), partly as a result of uncertainties in the logKow estimates that were used. Similar Kpws were found for different batches of silicone rubber from the same supplier. Differences in logKpws for silicone rubbers obtained from different suppliers ranged from 0.16-0.58.
Introduction In the aqueous environment, the freely dissolved concentration (Cw) of hydrophobic contaminants is an important parameter in environmental risk assessment (1). This freely dissolved fraction cannot be separated from the fraction bound to dissolved organic carbon (DOC) by filtration, because the bound fraction partly passes the filter and the freely dissolved fraction partly adsorbs to the filter (2, 3). Passive sampling with triolein-filled semipermeable membrane devices (SPMDs) has proven to be a valuable method for measuring Cw (4). Next to SPMDs, a number of single* Corresponding author phone: + 31 (0)8886 62190; e-mail:
[email protected]. † Ministry of Transport, Public Works and Water Management. ‡ Royal Netherlands Institute for Sea Research. § Present address: Deltares, P.O. Box 85467, 3508 AL Utrecht, The Netherlands. 10.1021/es9009376 CCC: $40.75
Published on Web 08/11/2009
2009 American Chemical Society
phase polymeric samplers have been introduced as passive samplers for the water phase: LDPE (5), polyoxymethylene (POM) (6), and silicone rubbers (mainly polydimethylsiloxane: PDMS) (7). If the polymer sampler has reached equilibrium with the water it is deployed in, the freely dissolved concentrations can be calculated using the polymer-water partition coefficients (Kpw). In kinetic sampling, the aqueous phase concentrations are calculated from sampling rates that are obtained from the release of performance reference compounds (PRCs) (4, 8). In both cases, the quality of the results strongly depends on the accuracy of the Kpws used. Kpw is often determined by equilibration of the material with a water phase containing the solutes of interest, followed by the analysis of both phases. However, the concentrations in the aqueous phase often are very low and difficult to measure accurately. Furthermore, sorption of solutes may occur to the wall of the container and to particulate or dissolved organic matter present in the water. Depending on the analytical procedure, this may result in an overestimation of the freely dissolved concentration. The addition of cosolvents (e.g., methanol, acetone) stabilizes the solutions and increases solubility in the aqueous phase and consequently lowers the polymer-mixture partition coefficient (Kpm). The resulting concentrations are further above the analytical detection limit and therefore can be measured more accurately. In addition, the effect of dissolved organic matter on the apparent partition coefficient vanishes with increasing cosolvent concentration. Therefore, extrapolation of Kpms to pure water may yield more accurate and/or more precise Kpw than direct measurements in cosolventfree water. The cosolvent method has been applied for the measurement of sediment-water partition coefficients, where the risk of overestimation of aqueous phase concentrations is evident due to the presence of dissolved organic matter and particulates (9, 10). Recently, this method was also reported for the measurements of Kpws for silicone rubber (11). The selection of the extrapolation method is of crucial importance but is often made without justification. In this paper, different models to predict Kpw from measured Kpms in methanol-water mixtures were evaluated. Predicted values were compared with the experimental values in ultra pure water, and the difference between the two were used to select the best Kpw estimates. Finally, the obtained Kpw values were correlated with logKow, molecular weight, and, for PCBs, chlorine substitution. A number of compounds were included that can be used as performance reference compounds (PRCs) in passive sampling. Theory. Cosolvent models have been developed for modeling solubility as a function of the volume fraction or mole fraction of cosolvent (12-14). Since water-solids partition coefficients are strongly related to the solubility, cosolvent models also have been applied to the estimation of sediment-water partition coefficients (Kd), mainly to eliminate the effect of sorption to DOM (9, 10, 15), by plotting logKd versus the volume fraction (10, 15) or the mass fraction (9) of cosolvent. The cosolvent concentration range over which this plot appears to be linear is identified, and logKd is estimated from the intercept obtained by linear regression. An implicit assumption is that any curvature in the low cosolvent range is caused by sorption to dissolved organic matter or by poor experimental control over highly hydrophobic compounds in aqueous solutions. This assumption need not be true, because a pronounced curvature is sometimes observed for solutes with low hydrophobicity in some cosolvent-water mixtures (16-18). Solvation models VOL. 43, NO. 18, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7047
TABLE 1. Numbers, Properties, and Suppliers of Investigated Materials Nr
material
thickness (mm)
density (g cm-3)
supplier
1 2 3 4 5 6 7 8
Silastic A SR batch 0 LDPE SR-RED SR-TF AlteSil Translucent (batch 11425b) AlteSil Translucent (batch 11756c) AlteSil Translucent (batch 11963b)
0.4 0.5 0.07 0.5 0.5 0.5 0.5 0.5
1.15 1.2 0.9 1.2 1.15 1.2 1.2 1.2
Dow Corning Vizo, Zeewolde, The Netherlands Brentwood Plastics, Brentwood, MO, USA J-flex Industrial rubber productsa J-flex Industrial rubber productsa Altecweb UKb Altecweb UKb Altecweb UKb
a
See www.j-flex.com.
b
See www.altecweb.com.
FIGURE 1. Log(Kpm/Vm) (mol kg-1) of Altesil polymer 678 as a function of the volume fraction methanol (left) and the mole fraction (right) for acenaphthene (ACE), chrysene (CHR), and PCB153. Lines represent the model fits for the LL5 model (dashed, left graph), KC2 model (drawn, right graph) and the MF5 model (dotted, right graph). Model fits were based on the filled symbols for LL5 and MF5 and on all data for KC2. may help to understand the relationship between logKpm and cosolvent concentration, and aid in selecting the appropriate extrapolation method. Because models for solute-solvent interactions in binary solvents have been developed for modeling solubility data, a translation step is needed to apply these models to measured partition coefficients. The activities ai,p and ai,m of solute i in the polymer and in the mixture are given by (19)
conveniently expressed on a volume basis (Cm ) xi,m/Vm, where Vm is the molar volume of the mixture). Thus, the polymer-mixture partition coefficient Kpm, given in volume/ mass units, equals
ai,p ) γi,pφi,p
(1)
and the ratio of Kpm to the partition coefficients in pure water (Kpw) is given as
ai,m ) γi,mxi,m
(2)
where φi,p is the volume fraction of the solute in the polymer, xi,m is its mole fraction in the mixture, and γi,p, γi,m are the solute’s activity coefficients (Raoult’s law convention) in the polymer and in the mixture, respectively. An expression for the equilibrium partition coefficient K′pm follows from the condition ai,p ) ai,m
′ ) Kpm
γi,m φi,p ) xi,m γi,p
(3)
Concentrations in the polymer are more conveniently expressed on a mass basis (Cp ) φi,p/[ViF], where Vi is the molar volume of the solute and F is the density of the polymer). Similarly, concentrations in the mixture are more 7048
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 18, 2009
Kpm )
Cp γi,mVm ) Cm γi,pViF
γi,mVm Kpm ) Kpw γi,wVw
(4)
(5)
where Vw is the molar volume of pure water. Here we assume that the effect of the cosolvent on the activity coefficient in the polymer phase can be neglected. This seems to be a fair assumption in view of the small swelling observed by Rusina et al. (20) for methanol (∼2.4-5.5% for various silicone rubbers and 0.1% for LDPE), but direct evidence is absent. Since activity coefficients at infinite dilution are equal to the inverse of the solubilities of the subcooled liquids, eq 5 can be written as log
Kpw Sw Kpm ) log + log Vm Vw Sm
(6)
where Sw and Sm are the mole fraction solubilities in water and in the mixture, respectively. A large number of models exist that describe solubilities as a function of cosolvent concentration in binary mixtures (13, 14, 21, 22). The log-linear model (LL) model predicts that the logarithm of the solubility is linearly proportional to the volume fraction of the cosolvent (12). Combining the LL model equation for log(Sw/Sm) with eq 6 gives log
Kpw Kpm ) log - σf Vm Vw
(7)
where f is the (nominal) volume fraction of cosolvent, and σ is a measure of the free energy difference between water and the cosolvent at the solute-solvent interface. For nonpolar solutes, σ is a linear function of logKow (23). In the Khossravi-Connors (KC) model (16), the free enthalpy change of dissolution is expanded into contributions that arise from crystal interactions, cavity formation, and solvation of the solute by the solvent components. The free enthalpy change for cavity formation is proportional to the solute’s surface area (A), the surface tension (Γ) of the solvent, and a dimensionless curvature effect factor (g) of the solute-solvent interface. Solvation is modeled as the formation of complexes between the solute and n solvent molecules, that are characterized by equilibrium constants Ki (i ) 1,..., n). Inserting the KC expression for Sw/Sm in eq 6 for n ) 2 (KC2) gives Kpm Kpw ) log + Vm Vw 2gAΓ′ gAΓ′ - ln K1 K1x(1 - x) + - ln(K1K2) K1K2x2 kT kT × (1 - x)2 + K1(1 - x)x + K1K2x2 log e (8)
log
[(
)
(
)
]
where x is the mole fraction of cosolvent, and Γ′ ) (Γc Γw)/2, (Γc, Γw are the surface tensions of cosolvent and water, respectively). Similarly, with n ) 1, the one-step KC model (KC1) yields Kpm Kpw log ) log + Vm Vw
[
- ln K )K x ( gAΓ′ kT 1
(1 - x) + K1x
1
]
× log e
(9)
Application of the KC2 model to solubility data of nonpolar and moderately polar solutes in various cosolvent-water mixtures yielded residual errors of about 2.5% (16). Solubilities of PCB3, PCB30, and PCB155 in alkanol-water mixtures were
best described by the KC models (residual errors ∼0.14 log units) compared with four other models (residual errors ∼0.15-0.34 log units) (13). Jouyban and co-workers compared a large number of models for the prediction of the solubilities of pharmaceuticals in mixed aqueous solutions (14, 21). They concluded that the Jouyban-Acree (JA) model yielded the smallest mean percentage deviation (MPD) of all tested models and that the largest MPD was observed for the LL model. With the JA model, logSm is fitted as a power series of the cosolvent volume fraction. Inserting the JA expression for Sw/Sm in eq 6 gives log
Kpm Kpw ) (1 - f )log + Vm Vw f log
Kpc - f (1 - f ) Vc
n
∑ A (2f - 1)
i
(10)
i
i)0
where Kpc is the partition coefficient between the polymer and the pure cosolvent, Vc is the molar volume of the cosolvent, and n is chosen to be as low as possible while still giving a good fit (typically, n ) 2). Data were modeled with n ) 1 (JA1) and n ) 2 (JA2). The JA models are equivalent to (n + 2) order polynomials in f. The above models were used to estimate Kpw by extrapolation from measured Kpm. The extrapolated Kpw were then used to validate the experimental Kpw. The choice for the KC2 model was based on the observation by Li and Andren (13) that solubility data of PCBs in alkanol-water mixtures were best described by this model. In addition, the log-linear model was selected because of its simplicity and popularity, and the JA models were chosen because these models were recently developed.
Materials and Methods Materials. Silicone rubbers from five different suppliers, including three batches from one supplier, and LDPE (Table 1) were investigated. Silastic sheets (nr 1) were prepared by spreading the rubber paste on 0.12 mm polyethylene between two spacers of 0.25 mm. The rubber was then covered with a second layer of polyethylene and pressed with a metal roller to form a layer in accordance with the spacers. This was covered with a wet tissue and a glass plate and allowed to cure for a week. All polymers were cut into approximately 1 g pieces with different shapes in order to distinguish them from each other. Prior to use, the sheets were Soxhlet extracted for 100 h with ethylacetate to remove any nonpolymerized material. Glass bottles of sizes ranging from 0.05 to 5 L were equipped
FIGURE 2. Average bias (difference between modeled and experimental logKpw) of the applied models using different data sets. Error bars represent the standard deviation of the bias. VOL. 43, NO. 18, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7049
FIGURE 3. LogKpw (L kg-1) versus molecular weight (MW) and logKow for polymer 678 (upper panel) and LDPE (lower panel). The number of Cl atoms is indicated by different symbols. Filled symbols: nonortho and mono-ortho PCBs; circled symbols: tetra-ortho PCBs. LogKows were adopted from Hawker and Connell (35). with screw caps lined with aluminum foil. Bottles of 10 L had glass stoppers. Equilibration Experiments. All polymers were equilibrated together with a single liquid phase ranging from 0 to 100% v/v methanol in steps of about 10%. A 50 mL bottle was used for incubations in pure methanol, and with increasing water content larger bottles were selected up to 10 L for 10% methanol and pure water. For methanol concentrations lower than 50% the sheets were, after addition of the spike (0.5 mL), pre-equilibrated in 80 mL 50% methanol for 6 days in 250 mL bottles. After that period 120 mL water was added and the bottles were shaken another 6 days to promote sorption of low hydrophobic compounds. Then the liquid phase was discarded and the sheets were transferred to the water or the methanol-water mixtures (10-40%) in appropriate bottle size. In the equilibrations in 0, 10, 20, and 30% methanol, the sheets were clamped together with a stainless steel spring in order to promote diffusive transport between them. The incubations at 50 and 80% methanol were performed in duplicate for the polymers Nr. 1, 6, 7, and 7050
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 18, 2009
FIGURE 4. LogKpw (L kg-1) of PAHs versus MW (b) and logKow (O) for polymer 678 and LDPE. LogKow were adopted from Mackay et al. (37). 8. All bottles were shaken upright for two months using orbital shakers (Gerhardt, RO 500, Germany) with amplitude of 30 mm at 100 rpm in a constant temperature (20 °C) room. Analysis. After equilibration, solvent and sheets were separated and sheets transferred to 50 mL glass vials equipped with screw caps lined with aluminum foil. Internal standard CB143 (400 ng) and 30 mL methanol-acetonitrile (1:1) were added and the vials were shaken overnight at 150 rpm. The extraction was repeated and the combined extracts were Kuderna Danish evaporated to around 1 mL. To transfer the analytes to isooctane, 10 mL of hexane and one mL of isooctane were added, followed by Kuderna Danish evaporation to about 2 mL. The extract was brought to 1 mL with a nitrogen stream. Extracts were analyzed by an Agilent HP 6890-HP 5972 GCMS in selected ion mode using electron impact ionization. The methanol-water phases with f g 0.3 were transferred to separating funnels (2 L), and water was added in the separating funnel to reduce the methanol content to below 25%. After internal standard CB143 (100-400 ng) was added, the mixture was extracted twice with 100 mL pentane for three minutes. The aqueous phase was extracted a second time using a second separating funnel (2 L). Volumes lar-
wall or closure. The larger volumes from the incubations in 0 and 10% methanol were extracted for 24 h using a large volume batch extractor as described by Hermans et al. (3). Here the aqueous phase was transferred to a clean bottle so as not to include analytes adsorbed to the wall. Internal standard (CB143) was added to the sample. All pentane extracts were Kuderna Danish evaporated, transferred to isooctane and analyzed as described above. Repeated extraction confirmed complete recovery. Mass balance calculations yielded values of 97 ( 7%. The bottle with 20% methanol showed breaks after the first week of shaking and sheets and solution were transferred to a new bottle.
TABLE 2. Applied Cosolvent Models Relating Log(Kpm/Vm) with Methanol Content abbreviation
model type
independent parameter
applied range
LL LL5 LL3 MF5 KC2 JA1
log-linear log-linear log-linear log-linear solvation 3th order polynomial 4th order polynomial
volume fraction (f) volume fraction (f) volume fraction (f) mole fraction (x) mole fraction (x)
0-1 0-0.5 0-0.3 0-0.3 0-1
volume fraction (f)
0-1
volume fraction (f)
0-1
JA2
Results and Discussion From the duplicates at 50 and 80% methanol an average standard error of 0.02 log units in experimental logKpms was calculated and to obtain equally structured data sets for all polymers the results of these duplicate measurements were
ger than the funnel allowed were extracted in portions passing both separating funnels. Equilibration bottles were not rinsed to avoid inclusion of analytes that might be sorbed on the
TABLE 3. Regression Parameters for PCBs of LogKpw with Compound Properties polymer type model
1
2
3 (LDPE)
4
5
678
Avg sa
log Kpw ) alog Kow + b
I
a b R2 s
0.95 0.12 0.90 0.22
0.94 0.00 0.89 0.23
1.18 -1.26 0.95 0.18
0.96 0.17 0.92 0.19
0.95 -0.02 0.89 0.22
0.97 0.14 0.92 0.19
0.05 0.30
0.92 0.11 0.62 0.97 0.13
0.94 0.24 0.49 0.97 0.12
0.03 0.18 0.06
0.0129 1.79 0.97 0.11
0.0128 2.09 0.95 0.15
0.0004 0.14
0.0124 0.52 1.93 0.98 0.09
0.0003 0.07 0.10
0.0119 0.64 1.99 0.20 0.99 0.08
0.0003 0.06 0.08 0.05
log Kpw ) alog Kow + b + b1b
II a b b1b R2 s
0.93 0.24 0.58 0.97 0.13
0.91 0.13 0.62 0.96 0.13
1.15 -1.16 0.52 0.99 0.09
0.93 0.27 0.50 0.97 0.12 log Kpw ) aMW + b
III a b R2 s
0.0129 1.97 0.97 0.12
0.0128 1.80 0.97 0.11
0.0153 1.23 0.94 0.20
0.0127 2.09 0.96 0.14
log Kpw ) aMW + a1MPF + b
IV a A1 b R2 s
0.0126 0.35 1.86 0.98 0.09
0.0126 0.29 1.71 0.98 0.10
0.0146 0.77 0.99 0.98 0.11
0.0123 0.50 1.94 0.98 0.09
0.0127 0.29 1.70 0.98 0.09
log Kpw ) aMW + a1MPF + b + b1b
V a a1 b b 1b R2 s a
0.0121 0.48 1.94 0.22 0.99 0.07
0.0120 0.44 1.80 0.25 0.99 0.07
0.0141 0.90 1.06 0.21 0.99 0.09
Average of the standard deviations of individual polymers.
0.0118 0.62 2.01 0.20 0.99 0.08 b
0.0121 0.43 1.79 0.25 0.99 0.07
Intercept b1 is only added for tetra-ortho substituted PCBs.
VOL. 43, NO. 18, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7051
averaged. For further interpretation the results for the three batches of Altesil rubber were averaged and referred to as polymer 678. Tested Cosolvent Models. Plots of log(Kpm/Vm) versus f are slightly S curved with more or less straight parts in the range 0 < f < 0.3 and 0.3 < f < 0.8 (Figure 1A). Because of the shift in slope at f ≈ 0.3, the log-linear model was also applied to the range 0 e f < 0.3 (LL3 model), and 0 e f < 0.5 (LL5 model). The similar curvature for acenaphthene, chrysene, and PCB153 at low f values indicates that this curvature is not caused by sorption to DOC. The JA1 and JA2 model closely followed the measured data (SI S1f and g). The same log(Kpm/Vm) data are plotted versus x in Figure 1B. The solid lines represent the KC2 model showing an excellent fit. Since it was noted that log(Kpm/Vm) shows a good linear relationship with x in the range from 0-0.3 (0 e f < 0.5), this model was also included (MF5 model, dotted line in Figure 1B). All tested models are listed in Table 2. Performance of Cosolvent Models. The JA and KC models give an excellent fit if all data are included. However, a good cosolvent model should yield consistent log(Kpw/Vw) estimates when the data for the water-rich incubations are excluded. To this end, we define the bias as the difference between the experimental log(Kpw/Vw) and the extrapolated value, and evaluated this bias after excluding (1) the data in pure water and (2) both the data in pure water and 10% methanol. The results of this analysis show that the LL and LL5 models are quite sensitive to the exclusion of data from the water-rich incubations (Figure 2, and Supporting Information (SI) S2). This is supported by the observations by Banerjee and Yalkowsky (24) that the log-linear model underestimates solubilities at cosolvent volume fractions
