Predicting Partitioning and Diffusion Properties of ... - ACS Publications


Predicting Partitioning and Diffusion Properties of...

0 downloads 52 Views 2MB Size

Article pubs.acs.org/est

Predicting Partitioning and Diffusion Properties of Nonpolar Chemicals in Biotic Media and Passive Sampler Phases by GC × GC Deedar Nabi†,‡ and J. Samuel Arey*,†,§ †

School of Architecture, Civil, and Environmental Engineering, Ecole Polytechnique Fédérale de Lausanne (EPFL), 1015 Lausanne, Switzerland ‡ Bigelow Laboratory for Ocean Sciences, East Boothbay, Maine 04544, United States § Eawag, Swiss Federal Institute of Aquatic Science and Technology, 8600 Dübendorf, Switzerland S Supporting Information *

ABSTRACT: The chemical parameters needed to explain and predict bioavailability, biodynamics, and baseline toxicity are not readily available for most nonpolar chemicals detected in the environment. Here, we demonstrate that comprehensive two-dimensional gas chromatography (GC × GC) retention times can be used to predict 26 relevant properties for nonpolar chemicals, specifically: partition coefficients for diverse biotic media and passive sampler phases; aquatic baseline toxicity; and relevant diffusion coefficients. The considered biotic and passive sampler phases include membrane and storage lipids, serum and muscle proteins, carbohydrates, algae, mussels, polydimethylsiloxane, polyethylene, polyoxymethylene, polyacrylate, polyurethane, and semipermeable membrane devices. GC × GC-based chemical property predictions are validated with a compilation of 1038 experimental property data collected from the literature. As an example application, we overlay a map of baseline toxicity to fathead minnows onto the separated analyte signal of a polychlorinated alkanes (chlorinated paraffins) technical mixture that contains 7820 congeners. In a second application, GC × GC-estimated properties are used to parametrize multiphase partitioning models for mammalian tissues and organs. In a third example, we estimate chemical depuration kinetics for mussels. Finally, we illustrate an approach to screen the GC × GC chromatogram for nonpolar chemicals of potentially high concern, defined based on their GC × GCestimated biopartitioning properties, diffusion properties, and baseline toxicity.

1. INTRODUCTION Once released into the environment, contaminants typically become distributed unevenly among different biotic and abiotic phases, due to the interplay of partitioning, transport, and transformation processes.1 The resulting distributions of contaminant concentrations in organisms and environmental compartments lead to biotic chemical exposures, usually defined in terms of bioavailability, bioaccessibility, and/or chemical activity.2 Resulting toxic effects to organisms may arise from both dynamic3 and equilibrium1 exposure conditions. When complex contaminant mixtures arise in the environment, the associated bioavailabilities, biodynamics, and toxicities become difficult to assess, because many different compounds (potentially hundreds or more) are acting simultaneously.4,5 Even in cases where a complex mixture composition has been elucidated, it is typically infeasible to categorize the individual components according to their bioaccumulation potentials, bioavailabilities, and toxicities.6 This is because the partition coefficient, effective lethal concentration data, and kinetic parameters needed to describe these processes are unavailable for many compounds. Existing property estimation approaches do not resolve the above problem. For example, polyparameter linear free energy © XXXX American Chemical Society

relationships (pp-LFERs) based on the Abraham Solvation Model (ASM) are used widely to predict the partition coefficients of relevant biotic phases and passive sampling materials.7−20 Abraham solute descriptors are available for approximately 8000 chemicals,21,22 whereas >140 000 industrial chemicals used in consumer products are suspected to have aquatic toxicity.23 Abraham descriptors are not available for many nonpolar environmental contaminants such as polychlorinated n-alkanes (PCAs), polyhalogenated dibenzo-pdioxins (PHDDs), dibenzofurans (PHDFs),24,25 polybrominated diphenylethers (PBDEs), and emerging nonpolar contaminants.26 Ideally, approaches to assess contaminant biopartitioning, biodynamics, and toxicity should be targeted at compounds actually measured in environmental samples. Comprehensive two-dimensional gas chromatography (GC × GC) may provide the missing link between sample analysis and the estimation of exposure- and effect-related properties. GC × GC is a powerful Received: October 6, 2016 Revised: January 20, 2017 Accepted: January 27, 2017

A

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

github.com/jsarey/GCxGC-property-estimation), 33 which maps partitioning property values directly onto the GC × GC chromatogram.32 The model can be adapted to a wide range of possible GC × GC instrument programs that have been optimized for other needs (e.g., for sample separation), assuming a linear temperature ramp is employed. To adapt the model to a user-defined instrument program, the GC × GC retention times of ≥15 calibration analytes must be recorded and entered into the model.32 The model can then be applied to other analyzed chemicals. The location of each nonpolar chemical on the GC × GC chromatogram corresponds to a unique set of estimated partitioning property values for environmental and biotic media. For example, nonpolar contaminants analyzed by GC × GC can be screened for aquatic bioaccumulation potential and terrestrial biomagnification potential, based on visual inspection of partitioning property contours overlaid onto the GC × GC chromatogram.32 In the present paper, we illuminate a path whereby GC × GC can be used to estimate the physical properties needed to model bioavailability, biodynamics, and baseline toxicity (for invertebrates and small fish) for the nonpolar compounds analyzed. Specifically, we explore the following hypotheses: (i) Equilibrium partitioning properties of nonpolar contaminants in diverse biotic phases and passive sampler phases can be estimated by eqs 1, analogous to relationships developed in previous work.32 These partitioning properties are needed to model the organism-specific bioavailabilities and biodynamic behaviors of contaminants. (ii) Baseline toxicity (narcosis) can be estimated based on eqs 1, due to the inherent relationship between baseline toxicity and phase partitioning. Baseline toxicity is associated with nonspecific interactions between the contaminant and biomembrane,34 which are modeled in part i. (iii) Diffusion coefficients of nonpolar chemicals in biotic phases and passive sampler polymers can be estimated by eqs 1. The molecular diffusion coefficient is controlled primarily by molecular size and by solute−solvent interactions,35 which are properties that are proxied by GC × GC retention times for nonpolar solutes.31,32,36 Diffusion coefficients are used in models of chemical uptake and release for organisms and for passive samplers,3 because diffusion influences the rate at which contaminants can traverse biotic (e.g., lipid− water) or polymer−water phase boundaries. (iv) Dynamics of contaminant uptake and release by invertebrates, small fish, and passive samplers can be modeled based on chemical properties estimated by eqs 1. This assumes that uptake and release processes are controlled primarily by the partition coefficients and molecular diffusivities that describe interphase mass transfers in an organism or passive sampler of a given size and set of aero/hydrodynamic conditions.3,37,38 Finally, limitations of the approach are identified and discussed at the end of the article (section 3.7).

analytical technique that can separate hundreds or thousands of compounds existing as complex mixtures in sediments,27 passive samplers,28 and biological media.29,30 Further capitalizing on the separation capacity of GC × GC, recent work shows that the environmental partitioning properties of analyzed nonpolar contaminants can be predicted from simple transformations of the first- and second-dimension retention times that are determined during a typical GC × GC analysis, applicable to nonpolar chemicals having boiling point ≤402 °C.31,32 This is accomplished by use of the relationship: log Pxy , i = λ1u1, i + λ 2u 2, i + λ3

(1a)

where Pxy,i is the partition coefficient of chemical i between phases x and y, and where the solute parameters u1,i and u2,i are defined as u1, i = log L1, i

(1b)

u 2, i = log L 2, i − βorth log L1, i

(1c)

where log L1,i and log L2,i are base-10 log-transformed gasstationary phase partition coefficients (mol L−1 mol−1 L) at 120 °C for chemical i on stationary phases 1 and 2. The parameter βorth was assigned a constant value of 1.1353 for the set of stationary phases employed in that work, defined such that the vectors u1,i and u2,i were mutually orthogonal for a balanced training set of nonpolar chemicals.32 The coefficients λ1, λ2, and λ3 are specific to each partitioning system, xy, and these parameters were calibrated previously for 11 different partitioning properties, based on multiple linear regression with model training sets.32 To clarify the scope of applicability of eqs 1, we define “nonpolar chemicals” as those that undergo negligible or limited hydrogen bonding interactions with their environment.32 More strictly, one may consider the validated scope of eqs 1 as the chemical families employed for calibration and testing, which comprised acyclic aliphatic, fused and bridged cyclic aliphatic, aromatic, and double-bonded hydrocarbon chemicals that are unsubstituted or substituted with fluorine, chlorine, bromine, and/or iodine.32 To develop the theoretical basis for eqs 1, we previously demonstrated that only two independent dimensions of information are required to describe >99% of the variability observed in a set of ASM solute parameters of nonpolar chemicals.32 This effectively reduces a 6-parameter linear model to a 2-parameter linear model that can describe the partitioning of nonpolar solutes among diverse environmental phases. The two orthogonal components (u1,i and u2,i) are dominated by the intermolecular interactions that govern the solute−solvent interactions of nonpolar solutes: the solvent cavitation energy, London dispersion interaction, and Debye and Keesom interactions, with negligible contributions from hydrogenbonding interactions. An appropriately chosen GC × GC column set can capture these two orthogonal components effectively: we used a nonpolar stationary phase (dimethyl polysiloxane) for the first dimension column and a semipolar stationary phase (50% phenyl polysilphenylene-siloxane) for the second dimension column.32 Finally, relationships from chromatography theory were employed to estimate u1,i and u2,i by mathematical transformations of the first and second dimension retention times of nonpolar chemicals analyzed by GC × GC (eqs S1 and S2).32 In practice, the calibrated model described by eqs 1a−1c can be applied using a freely available computer program (https://

2. METHODS A Zoex GC × GC−time of flight−mass spectrometer (TOFMS) instrument was used to analyze the standards mixtures described below. The first column was a nonpolar dimethyl B

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

calibrate the coefficients (λ1, λ2, and λ3) of eqs 1 for 15 properties, as explained in section 3.1. 2.2. Data Selection and Conversion. Experimental data were compiled from available literature, including partitioning coefficients, diffusion coefficients, and uptake kinetic parameters in diverse passive sampler media and biotic media for nonpolar chemicals in the primary training set, contaminant set I, and set II. Aquatic toxicity data were also compiled. Experimental data for set I are provided in Tables S4−S30 in the Supporting Information, as explained below. Experimental data for set II are given in Tables S31−S41. For each GC × GC instrument program (A, B, C), we adapted the program to eqs 1 using eqs S1 and S2, as described in our previous work32 (Section S3 and Table S42 in the Supporting Information). Passive Sampler Partitioning Data. Equilibrium partition coefficient data were collected from several literature sources for selected passive sampler materials in the presence of air, water, or lipids; a complete list of data is provided in the Supporting Information (Tables S4−S14). These properties include: experimental PDMS−water and PDMS−air partition coefficients (KPDMS‑w and KPDMS‑a)7,10,15,39−41 (Tables S4 and S5); polyethylene (PE)−water and PE−air partition coefficients (KPE‑w and KPE‑a)9,39,42−45 (Tables S6 and S7); polyacrylate (PA)−water partition coefficient (K PA‑w) 11 (Table S8); polyoxymethylene−water partitioning (KPOM‑w)8 (Table S9); polyurethane foam−air partition coefficient (KPUF‑a)12 (Table S10); semipermeable membrane device (SPMD)−water partition coefficient (KSPMD‑w)37 at temperatures ranging from 17 to 30 °C (Table S11, S12); SPMD−air partition coefficient (KSPMD‑a)37 (Table S13); and lipid−PDMS partition coefficient (Klip‑PDMS),46 where Klip‑PDMS values were converted from grams per gram units to liters per liter using density values of 1170 g/L for SSP-M823 PDMS membrane46 and 912.7 g/L for olive oil47 (Table S14). To augment the Klip‑PDMS data set, reported activity coefficients of PAHs in lipid48 and PDMS49 were used to calculate Klip‑PDMS values of PAHs39 (Table S14). Experimental KPDMS‑w values for individual PCA congeners were measured in our laboratory (Section S2 and Table S15 in the Supporting Information). Biopartitioning Data. Experimental partitioning properties were collected from literature for whole organisms (bivalve mussel and algae) and also for biotic phases including storage lipids, membrane lipids, protein, and carbohydrate. These properties include: lipid−normalized algae−water partition coefficient (Kalgae‑w) reported for Chlorella sorokiniana50 (Table S16); bioaccumulation factor (BAF) values, defined as the ratio between concentrations in mussels and concentrations in water51 (Table S17); storage lipid−air (Ks.lip‑a) and storage lipid−water partition coefficient (Ks.lip‑w)17,52 (Table S18, S19); phospholipid−water partition coefficient (Kph‑w)53 (Table S20); blood protein (bovine serum albumin)−water partition coefficient (Kalbumin‑w)14,15 (Table S21); muscle protein−water partition coefficient (Kmuscle.protein‑w)54 (Table S22); carbohydrate−water partition coefficient (Kch‑w)55,56 (Table S23), and Setschenow (salting-out) coefficient (Ks)16,57 (Table S24). Baseline Toxicity Data. Narcosis data reported for fathead minnows (Pimephales promelas) were collected from literature18,58 (Table S25). The end point used is the molar 96 hLC50 (median lethal concentration), defined as the molar concentration of a contaminant required to produce lethal effect in 50% of the population during an exposure period of 96 h.

polysiloxane phase, Rxi-1MS (Restek, RT-13323), of 30 m length, 0.25 mm inner diameter, and 0.25 μm film thickness, and the second column was a semipolar 50% phenyl polysilphenylene-siloxane phase, BPX-50 (SGE-054740), with 1 m length, 0.1 mm inner diameter, and 0.1 μm film thickness. Three different GC × GC-TOF-MS instrument programs were used in this study, referred to as GC × GC instrument programs A, B, and C, with runtime durations of 383, 194, and 99 min, respectively. Three different GC × GC methods were applied in order to confirm that eqs 1 predictions can be made with differing instrument programs and were not idiosyncratic outcomes of specially optimized instrument parameters of a single GC × GC method. The primary oven (which contains the first column) initially was held isothermal at 37 °C for 5 min and then ramped to 320 °C at a rate of 0.75 °C/min (for program A), 1.5 °C/min (program B), or 3 °C/min (program C). The secondary oven, which contains the second column, initially was held at 67 °C for 5 min and then was increased to 350 °C using a ramp rate of 1.5 °C/min (for program A), 3 °C/min (program B), or 5 °C/min (program C). Section S1 in the Supporting Information provides detailed descriptions of the three instrument programs used in this study. 2.1. Chemicals. A standard mixture comprising a diverse set of 70 nonpolar organic chemicals was analyzed by GC × GCTOF-MS instrument programs A, B, and C. These chemicals were divided into two contaminant sets: I and II. Numerous reliable experimental property data are available for contaminant set I, which was used to evaluate predictions of 15 properties by eqs 1 after model calibration with the “primary model training set” (see below). Set I was also used to calibrate the eq 1 coefficients (λ1, λ2, and λ3) of 11 properties for which the primary training set could not be applied, as explained in section 3.2. Fewer property data were available for set II, which was used to further validate predictions of eqs 1 for the collective set of 26 properties after the model calibrations explained above. The overall methodology is summarized in Figure S1 in the Supporting Information. Set I comprised of 41 chemicals and included selected nalkanes, 1,1,2,3,4,4-hexachloro-1,3-butadiene, 4 linear alkyl benzenes (LABs) (n-propyl- to n-hexyl-), 13 polycyclic aromatic hydrocarbons (PAHs), five chlorinated benzenes (CBs), 11 polychlorinated biphenyls (PCBs) (PCB congeners 28, 52, 101, 105, 118, 128, 138, 153, 156, 170, and 180), and p,p′-DDE (Table S1). A list of chemical suppliers is given in section S1 in the Supporting Information. Set II comprised of 29 chemicals and included 13 polychlorinated n-alkane (PCA) individual standards, 2 toxaphenes (Parlar 11 and 32), α-hexachlorocyclohexane (αHCH), γ-HCH (lindane), β-HCH, δ-HCH, heptachlor, aldrin, trans-chlordane, cis-chlordane, mirex, 2,4,5,6-tetrachloro-mxylene, p,p′-DDT, 4-chlorodiphenyl ether, and 2 polybrominated diphenyl ethers (PBDEs 3 and 47) (Table S2). Polychlorinated n-Alkane Technical Mixture. A technical grade short chain (C10−C13) mixture of polychlorinated nalkanes (PCA) having an average degree of chlorination of 51.5% was analyzed with GC × GC instrument program A. Primary Model Training Set. Here, 79 nonpolar chemicals were used to formulate a primary model training set (Table S3), taken from our previous study.32 With the primary training set, we aim to include balanced representation of chemical types among several nonpolar chemical families that contain the elements C, H, F, Cl, Br, and I. This training set was used to C

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Table 1. Regression Coefficientsa of Equations 1 for 15 Properties Describing Partitioning, Diffusion, and Baseline Toxicity, Based on Fitting to the Primary Model Training Set (n = 79) property −1

−1

log KPDMS‑w (mol L mol L) log KPDMS‑a (mol L−1 mol−1 L) log KPOM‑w (mol kg−1 mol−1 L) log KPA‑w (mol kg−1 mol−1 L) 37 °C log KPUF‑a (mol g−1 mol−1 mL) 15 °C log KSPMD‑w (mol L−1 mol−1 L) Ks Setschenow constant (mol−1 L) log Kph‑w (mol L−1 mol−1 L) log Ks.lipid‑w (mol L−1 mol−1 L) log Klipid‑a (mol kg−1 mol−1 L) 37 °C log Kblood.protein‑w (mol kg−1 mol−1 L) 37 °C log Kmuscle.protein‑w (mol kg−1 mol−1 L) 37 °C −log LC50 (mol L−1; fathead minnow) log Dwater (×10−5 cm2 s−1) log Dethanol (×10−5 cm2 s−1)

λ1

λ2

1.40 ± 0.04 1.58 ± 0.04 1.27 ± 0.03 1.31 ± 0.07 1.77 ± 0.03 1.07 ± 0.05 0.05 ± 0.01 1.34 ± 0.06 1.59 ± 0.05 1.73 ± 0.03 1.09 ± 0.07 1.23 ± 0.06 1.32 ± 0.07 −0.13 ± 0.01 −0.038 ± 0.005

−6.60 0.59 −3.01 −3.89 2.28 −3.46 −0.28 −4.30 −7.07 0.43 −3.64 −4.14 −4.16 0.46 0.10

± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

λ3 0.56 0.20 0.55 0.53 0.20 0.39 0.07 0.53 0.68 0.19 0.60 0.50 0.54 0.06 0.04

c

0.0 0.49 0.0 0.31 0.81 0.91 0.13 0.56 0.0 0.39 0.43 −0.69 1.42 0.23 0.24

± 0.08 ± ± ± ± ±

0.18 0.10 0.13 0.02 0.17

± ± ± ± ± ±

0.11 0.17 0.16 0.19 0.02 0.01

RMSEb

r2

0.48 0.16 0.37 0.37 0.12 0.30 0.03 0.34 0.50 0.17 0.39 0.33 0.37 0.04 0.02

0.91 0.995 0.94 0.94 0.998 0.94 0.79 0.95 0.92 0.995 0.91 0.95 0.94 0.94 0.78

Reported regression coefficient uncertainties correspond to a 95% probability interval of the fitted values, estimated using the bootstrap method106 with 1000 synthetic resamplings. bRMSE is the root mean squared error. cA reported value of 0.0 indicates that the fitted coefficient value was found to be statistically indistinguishable from zero; in these cases the corresponding variable was removed and the regression was repeated. a

Transport-Related Property Data. From the available literature, we compiled experimental data for: aqueous diffusion coefficient (Dw)35,59 (Table S26); diffusion coefficient in ethanol35 (Dethanol) (Table S27); diffusion coefficients in PDMS (DPDMS) and PE (DPE)60 (Table S28); rate coefficient (ke) values for PDMS−water exchange,61,62 which were converted into time to reach 95% of the equilibrium state (τ95) (Tables S29 and S46). The ke data are specific to the geometries and hydrodynamic conditions reported in the literature.61,62 Therefore, we recommend the recalibration of eqs 1 for cases where the geometries of passive samplers and hydrodynamic conditions differ from those assumed here (Section S9). Depuration rate coefficient (kdepuration) were also collected for PBDEs in blue mussels (Mytilus edulis)63 and kdepuration values for PAHs in unionid mussels (Elliptio complanata) and for PCBs in green-lipped mussels (Perna viridis)64 (Table S30).

property. Among the partitioning properties, regression statistics were excellent (r2 = 0.995−0.998) for those involving transfer from an organic material to air phase. Regressions statistics for partitioning properties involving an aqueous phase were slightly worse (r2 = 0.91−0.95). This is consistent with findings of our previous study.32 Exceptionally, the very good regression statistics for log KPDMS‑a (r2 = 0.995; RMSE = 0.16) can be attributed to the close correspondence between KPDMS‑a and L1,i. When fitted with the primary training set, eqs 1 gave log Dwater values with r2 = 0.94 and RMSE of 0.04 log unit (Table 1). Diffusion coefficient in ethanol, which mimics the diffusion in a mucus layer,35 can also be estimated (r2 = 0.78 and RMSE = 0.02 log unit). Both of the solute parameters of eqs 1 (u1,i and u2,i) contributed significantly in explaining the variability in all 15 partitioning and transport properties, as evidenced by the statistically nonzero values of the model coefficients, λ1 and λ2 (Table 1). This demonstrates that a two-parameter model is justified for these properties, enabling more flexible predictive capability than is allowed by a traditional single-parameter LFER (such as with the log octanol−water partition coefficient).32 For KPDMS‑w, eqs 1 is somewhat analogous to that proposed previously by Dewulf et al.,78 who estimating KPDMS‑w using a two-parameter linear regression with values of the Kováts retention index on a 100% dimethyl polysiloxane stationary phase and the experimental air−water partition coefficient (Ka‑w). To further test eqs 1 for the 15 properties listed in Table 1, these fitted model relationships were used to make predictions for contaminant sets I and II, which had been analyzed by GC × GC. For contaminant sets I and II, the L1,i and L2,i (and thus the u1,i and u2,i) values were determined by transformations of GC × GC retention times according to eqs S1 and S2. Property predictions for contaminant set I exhibited good agreement with experimental data (Table S44, Figure S2). For passive sampler partitioning properties, eqs 1 predictions exhibited RMSE values ranging from 0.16 to 0.45 log unit with respect to reported experimental values. RMSE values for biopartitioning properties ranged from 0.21 to 0.44 log unit, and the RMSE for predicted −log LC50 (baseline toxicity) for fathead minnows

3. RESULTS AND DISCUSSION 3.1. Calibration and Testing of Equations 1 for 15 Properties Describing Partitioning, Diffusivity, and Toxicity. Equations 1 were calibrated for 15 target properties (Table 1) using the 79 solutes of the primary model training set, as follows. Solute L1,i and L2,i values were estimated using the Abraham solvation model, based on model coefficients reported previously for the stationary phases SE-30 and OV17,65 and solute parameters (Table S3) collected from several sources.14,17,18,66−77 Target property values of the training set were assigned using available experimental data from the literature (listed in Methods); data gaps were filled with estimated values from established Abraham solvation models (Table S43).7,8,11−14,16,17,35,54 Multiple linear regression was used to optimize the coefficients λ1, λ2, and λ3, so as to produce the best (lowest sum-of-squared residuals) agreement with assigned property values of the primary training set. Regressions of eqs 1 with the primary training set produced good statistics for all 15 properties, with squared correlation coefficient (r2) values ranging from 0.78 for log Dethanol to 0.998 for log KPUF‑a (Table 1). Root-mean-squared error (RMSE) values ranged from 0.02 to 0.50 log unit, depending on the D

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Table 2. Regression Coefficientsa of Equations 1 for 11 Properties Describing Partitioning, Diffusion, and Exchange Kinetics, Based on Fitting to the Experimental Data from Contaminant Set I λ1

property log log log log log log log log log log log

−1

−1

KSPMD‑a (mol L mol L) KPE‑w (mol kg−1 mol−1 L) KPE‑a (mol kg−1 mol−1 L) Klip‑PDMS (mol L−1 mol−1 L) KCH‑w (mol L−1 mol−1 L) BAF (mol L−1 mol−1 L) Kalgae‑w (mol kg−1 mol−1 L) DPDMS (m2 s−1) DPE (m2 s−1) τ95 (PSD; d) kdepuration (d−1; mussels)

0.75 1.54 1.45 0.26 0.52 1.27 1.49 −0.27 −0.51 1.45 −0.15

± ± ± ± ± ± ± ± ± ± ±

λ2 0.08 0.06 0.05 0.01 0.02 0.04 0.03 0.02 0.03 0.07 0.01

0.0 −4.62 1.31 0.80 4.11 −4.22 −4.47 0.0 0.0 −7.73 0.0

λ3

± ± ± ± ± ±

0.54 0.49 0.24 0.47 0.82 0.72

± 0.62

3.94 −0.70 1.00 0.0 0.0 0.0 0.0 −9.16 −10.58 −1.88 −0.22

± 0.33 ± 0.26 ± 0.22

± ± ± ±

0.07 0.11 0.32 0.06

RMSEb

r2

n

0.26 0.25 0.20 0.10 0.19 0.33 0.30 0.06 0.12 0.24 0.07

0.89 0.96 0.98 0.86 0.96 0.83 0.91 0.93 0.94 0.97 0.83

10 23 22 21 12 16 15 21 22 20 25

Reported regression coefficient uncertainties correspond to a 95% probability interval of the fitted values, estimated using the bootstrap method106 with 1000 synthetic resamplings. bRMSE is the root mean squared error. cA reported value of 0.0 indicates that the fitted coefficient value was found to be statistically indistinguishable from zero; in these cases the corresponding variable was removed and the regression was repeated. a

which are molecular properties typically used to explain variability in molecular diffusion coefficients in polymers.60 These fitted models (Table 2) should be used cautiously, because the experimental data sets used in the regressions are relatively small (n = 11−25; Tables S6, S7, S13, S16, S17, S23, S28−S30) and in some cases include few chemical families. Thus, one might tentatively consider the validated scope of these models as only the chemical families encompassed in the regressions and their structural analogues. However, when applied to chemicals from contaminant set II, these models exhibit good agreement with experimental values (Tables S31− S41), suggesting that the eq 1 coefficients are reasonably robust for these properties (Table 2). We evaluated additional statistical diagnostics to report further on the robustness of the regression models (Table S44). Overall, our results demonstrate that GC × GC currently can be used to estimate these properties for which Abraham solvation model relationships are presently unavailable. 3.3. Properties Describing Partitioning, Diffusion, and Toxicity Can be Mapped onto the GC × GC Chromatogram. We mapped 26 properties describing partitioning, diffusion, and toxicity onto the GC × GC chromatogram (Figures S1, S2), based on the relationships described in Table 1 (section 3.1) and Table 2 (section 3.2). In practice, this is achieved with a software program33 that first maps L1,i and L2,i values onto the GC × GC chromatogram using eqs S1 and S2 and then applies eqs 1 with the appropriate coefficients (Tables 1, 2). As an example, in Figure 1 we show the contours of baseline toxicity (−log LC50) for fathead minnows overlaid onto the GC × GC chromatogram of a short-chain PCA mixture believed to contain up to 7820 structurally distinct congeners.83 The results illustrate that each resolved PCA band aligns with a contour of −log LC50 and spans a width of about 0.5 log unit. Although not shown here, the ratio of pure (subcooled) liquid aqueous solubility (SwL) to LC50 value could also be used as a metric to assess the relative toxic potential of mixture components.84 This information can be obtained by GC × GC, since SwL can be estimated using eqs 1.32 To our knowledge, the predicted LC50 distribution of the separated PCA complex mixture shown in Figure 1 would not have been possible to obtain using other approaches. Baseline toxicity may be used to shed light on the mode of action that expresses the toxicity of a contaminant85,86 (Section

was 0.46 log unit. Finally, for diffusion coefficients in water and ethanol, log Dwater and log Dethanol values predicted by eqs 1 compared favorably with available experimental data, exhibiting RMSE values of 0.04 and 0.13 log units, respectively. The available experimental data for contaminant set II were more scarce (n = 101) and also had substantial variability among reported experimental values (Table S31−S41 and Section S5). For the 15 properties listed in Table 1, eqs 1 appeared to have prediction skill comparable to ASM-type models for both contaminant sets I and II, based on a comparison of RMSE values (Tables 3 and S44 and Section S4 in the Supporting Information). 3.2. Tentative Calibration of Equations 1 for 11 Additional Properties Describing Partitioning, Diffusion, and Uptake. Among the 26 partitioning and diffusion properties that we considered, there were 11 properties for which the primary training set could not be constructed, due to lack of available ASM coefficients. For these cases, we fitted eq 1 coefficients (λ1, λ2, and λ3) by multiple linear regression with available experimental property data for contaminant set I (Table 2). For this chemical set, L1,i and L2,i values were determined from GC × GC retention times by eqs S1 and S2. We obtained reasonable fit statistics for these 11 properties by regression of eqs 1 with contaminant set I (Table 2, Figure S2). The fitted models produced r2 values ranging from 0.83 to 0.98 and RMSE values ranging from 0.07 to 0.34 log unit when compared to experimental data, respectively. For partitioning into the commonly used semipermeable membrane device (SPMD) and polyethylene (PE) passive sampling materials, regressions of eqs 1 exhibited r2 values of 0.89 to 0.98 and RMSE values of 0.21 to 0.27 log units. For lipid-PDMS partition coefficient (log Klip‑PDMS), eqs 1 produced a fit having r2 = 0.86 and RMSE of 0.10 log unit. This property is relevant for ex vivo analysis, internal exposure estimates, in vitro toxicity dosing,79−81 and model biomagnification factors (BMF), and trophic magnification factors (TMF).20,46,82 For the BAF, algae−water partition coefficient, and carbohydrate−water partition coefficient, regressions of eqs 1 with experimental data exhibited r2 values of 0.85−0.97 and RMSE values of 0.19−0.34 log units. Diffusion coefficients in PDMS and PE were successfully explained by eqs 1 fits, having RMSEs of 0.06−0.12 log unit and r2 = 0.93−0.94. These regressions performed better than regressions with molar mass, molecular size, and total surface area (TSA) (eqs S3−S10, Section S6), E

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Table 3. Log-Transformed Partition Coefficient Values for Mammalian Fluids, Tissues, and Organs According to Experimental Reports, Predictions by Equations 1, and Predictions by the ASM chemical nonane decane benzene, propylbenzene, butylacenaphthene fluorene phenanthrene anthracene fluoranthene pyrene pyrene PCB 101 PCB 153 PCB 52

Figure 1. Predicted baseline toxicity −log LC50 (mol L−1; fathead minnows) contours overlaid onto a GC × GC chromatogram of polychlorinated n-alkane technical mixture (C10−C13 with 51.5% chlorine), based on eqs 1.

S7 in the Supporting Information). Based on a comparison of their observed (LC 50,experimental) and predicted baseline (LC50,baseline) values, most of the contaminants in sets I and II (for which LC50,experimental values were available) were classified as nonpolar narcotics (Table S25). Using the approach described in section 3.1, the eq 1 model of LC50 (shown here for fathead minnows) could also be parametrized for other organisms for which Abraham solvation model equations exist, including protozoa,87 bacteria,88 water fleas,88 and tadpole.89 By comparison, other toxicity models for complex mixtures such as the target lipid model (TLM)84 and PETROTOX6 rely on one-parameter LFERs with octanol−water partition coefficient to estimate lipid−water partition coefficient, organic carbon−water partition coefficient, and subcooled aqueous solubility for petroleum substances. These models could be improved by using eqs 1 to provide more accurate predictions of these properties. 3.4. Estimating in-Vivo and in-Vitro Partition Coefficients Using a Multiphase Partitioning Model. Contaminant exposure concentration at the target organ level is considered the most accurate exposure end point.90 However, the biopartitioning properties needed to assess this outcome usually are not available. We used a multiphase partitioning model to calculate in-vivo and in-vitro partition coefficients for several mammalian organs, tissues, and biological fluids (n = 30, Table 3), based on compositional information describing the biological system and the partition coefficients of the component phases involved91 (Section S8 in the Supporting Information). For example, liver−water partition coefficients (Kliver‑w) were estimated as20 Kliver‐w =

PCB 101 PCB 105 PCB 118 PCB 138 PCB 153 PCB 156 PCB 170 PCB 180 DDE decane pyrene PCB 101 PCB 153 pyrene PCB 101 PCB 153

species

experimental log K

eqs 1

ASM

milk−water milk−water milk−water

cow cow cow

4.73 5.34 2.44

4.95 5.34 3.15

4.30 4.88 2.60

milk−water milk−water milk−water milk−water milk−water milk−water milk−water liver−blood liver−blood liver−blood adipose− plasma adipose− plasma adipose− plasma adipose− plasma adipose− plasma adipose− plasma adipose− plasma adipose− plasma adipose− plasma adipose− plasma skin−blood lung−plasma skin−blood skin−blood muscle− plasma muscle− blood muscle− blood

cow cow cow cow cow cow cow rat rat rat human

2.89 2.65 3.08 3.50 3.53 3.90 3.98 0.37 0.78 1.08 1.90

3.46 2.73 3.08 3.09 3.17 3.77 3.65 0.94 0.94 0.81 2.30

3.15 3.00 3.41 3.46 3.61 4.30 4.06 1.05 0.93 0.97 2.36

human

1.70

2.41

2.42

human

2.11

2.51

2.43

human

2.25

2.45

2.43

human

2.04

2.43

2.47

human

2.60

2.36

2.35

human

2.49

2.47

2.48

human

2.56

2.48

2.51

human

2.43

2.53

2.51

human

2.23

2.42

2.46

rat rat rat rat rat

0.68 0.35 0.85 1.48 0.2

1.66 0.48 1.31 1.25 0.32

1.49 0.57 1.35 1.41 0.36

rat

0

0.27

0.25

rat

0.6

0.25

0.28

0.38

0.35

RMSE

organisms.20 Using GC × GC to estimate the partition coefficients, Ks.lip‑w, Kph‑w, Kalbumin‑w, and Kmuscle.protein‑w, we found good agreement between computed Kliver‑blood values and reported experimental steady state concentration ratios (Table 3). In a similar way, we evaluated in vitro and in vivo partitioning for several other mammalian organs, tissues, and fluids (Table 3). Values computed by eqs 1 agreed well with these experimental data, exhibiting an RMSE of 0.38 log unit (n = 30). Similar prediction skill was obtained using the Abraham solvation model (Table 3). However, using GC × GC, we can estimate the needed biopartitioning properties and resulting exposure concentrations at target organs for nonpolar contaminants for which Abraham solute descriptors are not available.

C liver Cw

= 0.73 + 0.019K s.lip‐w + 0.046K ph‐w + 0.0019 K albumin‐w + 0.17K muscle.protein‐w

system

(2)

where Cliver is contaminant concentration in liver tissue, and Cw is the freely dissolved concentration of contaminant in the aqueous phase. The numerical coefficients in eq 2 represent the volume fractions of water, storage lipids, phospholipids, blood proteins, and muscle proteins reported for a set of mammalian F

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 2. Example screening approach for identifying chemicals of potentially high environmental concern with the GC × GC chromatogram retention space. Top panels show GC × GC chromatograms of (a) the short-chain polychlorinated alkanes technical mixture and (b) a mixture of injected standards. In panels c−f, the contours of (c) log BAF, (d) −log LC50 (mol L−1; fathead minnow), (e) log kdepuration (d−1; mussels), and (f) log Dethanol (×10−5 cm2 s−1) are overlaid onto GC × GC chromatogram. Black and green triangles indicate analytes in contaminant set I and set II, respectively. The shaded yellow region indicates the area where analytes have predicted properties values indicate high environmental concern, according to the example screening approach.

3.5. Modeling Dynamics of Contaminant Uptake into Passive Sampling Devices. Dynamic aspects of contaminant uptake and release in passive sampling devices (PSD) can be modeled using the diffusion coefficients and partition coefficients that are estimated by eqs 1. Exchange dynamics are controlled by the size, geometry, and composition of the PSD, as well as the hydro/aerodynamic conditions and properties of the contaminant.3,92,93 Under well-defined assumptions (Section S9), the exchange processes in PSDs can be modeled using diffusion-based models with the following form:3,37,38,93,94 ke =

APSD VPSDαKPSD‐w

sampling device, water, and polymer, respectively. The parameter α describes the chemical’s extent of association to dissolved organic matter (DOM). The rate constant, ke, can also be converted into a more conceptually intuitive parameter, for example, time to reach 95% equilibrium: τ 95 =

+

δp DpKPSD‐w

(4)

We inputted the GC × GC-estimated partition coefficients (KPDMS‑w) and diffusivities (DPDMS and Dwater) values in the mass transfer model (eq 3) to simulate ke values of PAHs for the geometry and hydrodynamic conditions given in the work of Mayer et al.61 The simulated ke values compared favorably (RMSE = 0.64 log unit) with reported experimental ke values.61 By comparison, the input of experimental DPDMS, experimental KPDMS‑w values, and Hayduk−Laudie-estimated Dwater values95 into eq 3 produced ke estimates that exhibited RMSE = 0.73 log unit with respect to experimental ke data. Separately, we also

1 δw Dw

−ln(0.05) ke

(3)

where ke is the exchange rate constant, and where A and V denote area and volume of the sampler. The term δ is the diffusion path length, D is diffusion coefficient, and K is partition coefficient. Subscripts PSD, w, and p indicate passive G

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

processes, performs satisfactory for the prediction of dynamic accumulation of hydrophobic nonpolar chemicals in algae, protozoa, small fish, and invertebrates such as bivalve, crustacean, insect, and worm.3,98 Additionally, the baseline toxicity model that is used (as an example application) in the present work does not account for the kinetically limited uptake of very hydrophobic chemicals, due to the short duration of the uptake test (e.g., 96 h) that is usually employed.101 Finally, the partitioning properties discussed in the present work assume the subcooled liquid state (e.g., for aqueous solubility) and this does not take into account the melting point toxicity cutoff that is relevant for pure solid chemicals.102 3.8. Screening for Chemicals of Potentially High Concern by GC × GC. Using the approach developed in this paper, researchers can make deductions relevant to environmental risk by projecting combinations of environmental partitioning properties onto the GC × GC chromatogram. For example, by overlaying the property estimates for BAF, −log LC50 (fathead minnow), depuration rate constant (mussels), and diffusion coefficient in ethanol onto the GC × GC chromatogram, we can identify chemicals of potentially high concern (Figure 2). The analytes that elute within yellowhighlighted areas are highly bioaccumulative (BAF > 5000) and are toxic (LC50 < 0.01 mg/L) to a test organism (fathead minnow). These chemicals also have significant diffusivities (Dethanol > 1.16 × 10−5 cm2 s−1) and short exchange half-lives in mussels (kdepuration > 0.09 d−1), which determine how quickly these organisms can reach lethal body burdens under dynamic conditions (e.g., accidental chemical spills). Based on these or other selected criteria, chemicals could be categorized as chemicals of potentially high concern within an analyzed chemical set or environmental sample.103 The above approach could also be adapted to recently updated biomagnification factor (BMF),104 the trophic magnification factor (TMF),104 and associated depuration rate constants.105

conducted direct regression of eqs 1 with reported experimental log τ95 values for contaminant set I, finding an r2 of 0.97 and an RMSE of 0.24 log units (Table 2). The mass transfer resistance (Section S9) of contaminants in polymer phases is an important criterion for selecting the appropriate PSDs, and this property can be estimated using eqs 1. For example, the parametrization of eq 3 revealed that the mass transport resistances for PAHs in aqueous boundary layers were 2 orders of magnitude higher than those in the polymer film (Table S46). Taken together, these results illustrate that eqs 1 can be used to provide useful insights into the mass transfer processes of passive samplers and phase interfaces in biota. 3.6. Modeling Dynamics of Contaminant Depuration for Whole Organisms. Chemical property estimates from eqs 1 can also be used to predict contaminant depuration dynamics for invertebrates and small fishes used by regulatory agencies for risk assessment. Depuration dynamics in these organisms are controlled by organism composition, age, growth rate, size, and geometry, as well as ambient hydro/aerodynamic conditions and properties of the contaminant.3,92 Using wellknown assumptions, chemical exchange can be modeled using a diffusion-based model:3,96 kdepuration =

Ao Mo(1 − γ ) + γK m‐w

1 δw Dw

+

δm DmK m‐w

(5)

where the subscript m indicates that the property is assigned to a biomembrane and o indicates the organism. The terms Ao and Mo refer to the surface area and the mass of the organism, respectively. The parameter γ represents lipid content of the organism. Experimental kdepuration data reported for bivalve mussels were employed here as a way to evaluate the use of eqs 1. The assumptions considered here92,96 were the following: (1) chemical exchange across the gill surface dominates exchange from fecal and urinary discharge; (2) biotransformation of chemicals is negligible; (3) chemical exchange across the gills follows a simple passive diffusion model; and (4) diffusion boundary layers are controlled by the anatomical configuration of the gills and not by the hydrodynamic conditions, resulting in constant aqueous and membrane layers. Therefore, for a given weight and age, depuration of contaminants is assumed to be a function of water-membrane partitioning and of diffusion coefficients in water and membrane (eq 5). Thus, it was expected that these chemical properties would explain the variance in the depuration rate constant data for the mussels. Equations 1 produced good fits for the depuration rate constant data, exhibiting an RMSE value of 0.07 log units and r2 of 0.83 (Table 2), despite the uncertainties97 associated with the measurements of depuration kinetics in mussels. Taken together, these results support the conclusion that diffusion and biopartitioning properties predicted by eqs 1 can be used to model the dynamics of contaminant exchange between environment and organism or PSD, within the limitations discussed below. 3.7. Limitations of the Approach. Equations 1 is capable of modeling the parameters that describe diffusion and equilibrium partitioning properties for nonpolar compounds. Thus, it does not account for chemical biotransformation and chemical ionization. It also does not account for chemical uptake that is limited by dietary exposure, breathing rate, blood flow rate, and growth dilution; these may be the limiting processes for most vertebrates.99,100 The diffusion-based kinetic model (eq 5), which does not take into account above limiting



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.6b05071. Text sections: GC × GC-TOF-MS instrument programs; measurement of PDMS−water partition coefficient; how to adapt eqs 1 to an individual GC × GC instrument program; performance of eqs 1 for contaminant set II; data quality of hydrophobic chemicals considered in the evaluation of eqs 1; correlations of Dwater, DPE, and DPDMS with molar mass, molecular size, and total surface area; discrimination of excess toxicity using GC × GC; estimating in vivo and in vitro partition coefficients using a multiphase partitioning model; derivation of a diffusion-based kinetic model; and the experimental database used to validate eqs 1 (PDF)



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

J. Samuel Arey: 0000-0002-3189-1585 Notes

The authors declare no competing financial interest. H

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology



(18) Hoover, K. R.; Acree, W. E.; Abraham, M. H. Chemical toxicity correlations for several fish species based on the Abraham solvation parameter model. Chem. Res. Toxicol. 2005, 18 (9), 1497−1505. (19) Poole, C. F.; Ariyasena, T. C.; Lenca, N. Estimation of the environmental properties of compounds from chromatographic measurements and the solvation parameter model. J. Chromatogr. 2013, 1317, 85−104. (20) Endo, S.; Brown, T. N.; Goss, K.-U. General model for estimating partition coefficients to organisms and their tissues using the biological compositions and polyparameter linear free energy relationships. Environ. Sci. Technol. 2013, 6630. (21) ACD/ADME suite, version 5.0. Advanced Chemistry Development, Inc.: Toronto, ON, Canada, 2010. (22) Endo, S.; Watanabe, N.; Ulrich, N.; Bronner, G.; Goss, K.-U. Ufz-lser database, v 2.1; Helmholtz Centre for Environmental Research: Leipzig, Germany, 2015. https://www.ufz.de/index. php?en=31698&contentonly=1&m=0&slerd_data[mvc]=Public/start. (23) 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 (8), 2018−2027. (24) van den Berg, M.; Denison, M. S.; Birnbaum, L. S.; DeVito, M. J.; Fiedler, H.; Falandysz, J.; Rose, M.; Schrenk, D.; Safe, S.; Tohyama, C.; Tritscher, A.; Tysklind, M.; Peterson, R. E. Polybrominated dibenzo-p-dioxins, dibenzofurans, and biphenyls: Inclusion in the toxicity equivalency factor concept for dioxin-like compounds. Toxicol. Sci. 2013, 133 (2), 197−208. (25) Weber, L. W. D.; Greim, H. The toxicity of brominated and mixed-halogenated dibenzo-p-dioxins and dibenzofurans: An overview. J. Toxicol. Environ. Health 1997, 50 (3), 195−216. (26) Howard, P. H.; Muir, D. C. G. Identifying new persistent and bioaccumulative organics among chemicals in commerce. Environ. Sci. Technol. 2010, 44 (7), 2277−2285. (27) Frysinger, G. S.; Gaines, R. B.; Xu, L.; Reddy, C. M. Resolving the unresolved complex mixture in petroleum-contaminated sediments. Environ. Sci. Technol. 2003, 37 (8), 1653−1662. (28) Manzano, C. A. The identification and quantitation of complex polycyclic aromatic hydrocarbon mixtures in environmental samples using comprehensive two-dimensional gas chromatography with timeof-flight mass spectrometry. PhD thesis, Oregon State University, 2012. (29) Booth, A. M.; Sutton, P. A.; Lewis, C. A.; Lewis, A. C.; Scarlett, A.; Chau, W.; Widdows, J.; Rowland, S. J. Unresolved complex mixtures of aromatic hydrocarbons: Thousands of overlooked persistent, bioaccumulative, and toxic contaminants in mussels. Environ. Sci. Technol. 2007, 41 (2), 457−464. (30) Mondello, L.; Tranchida, P. Q.; Dugo, P.; Dugo, G. Comprehensive two-dimensional gas chromatography-mass spectrometry: A review. Mass Spectrom. Rev. 2008, 27 (2), 101−124. (31) Arey, J. S.; Nelson, R. K.; Xu, L.; Reddy, C. M. Using comprehensive two-dimensional gas chromatography retention indices to estimate environmental partitioning properties for a complete set of diesel fuel hydrocarbons. Anal. Chem. 2005, 77 (22), 7172−7182. (32) Nabi, D.; Gros, J.; Dimitriou-Christidis, P.; Arey, J. S. Mapping environmental partitioning properties of nonpolar complex mixtures by use of GC × GC. Environ. Sci. Technol. 2014, 48 (12), 6814−6826. (33) Arey, J. S.; Nabi, D. GC × GC property estimation algorithm. https://github.com/jsarey/GCxGC-property-estimation. (34) Wezel, A. P. v.; Opperhuizen, A. Narcosis due to environmental pollutants in aquatic organisms: Residue-based toxicity, mechanisms, and membrane burdens. Crit. Rev. Toxicol. 1995, 25 (3), 255−279. (35) Hills, E. E.; Abraham, M. H.; Hersey, A.; Bevan, C. D. Diffusion coefficients in ethanol and in water at 298 K: Linear free energy relationships. Fluid Phase Equilib. 2011, 303 (1), 45−55. (36) Arey, J. S.; Nelson, R. K.; Reddy, C. M. Disentangling oil weathering using GC × GC. 1. Chromatogram analysis. Environ. Sci. Technol. 2007, 41 (16), 5738−5746.

ACKNOWLEDGMENTS The authors gratefully acknowledge support for the GC × GCTOF-MS from Swiss National Science Foundation R’Equip Grant 206021-128753/1. Three anonymous reviewers are thanked for valuable comments, as well as Drs. Martin Scheringer, Peter Schmid, Felippe de Alencastro, and Philip Mayer. We thank Mehmet Coelhan for kindly supplying individual PCA standards.



REFERENCES

(1) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry, 2nd ed.; Wiley: Hoboken, NJ, 2003. (2) Reichenberg, F.; Mayer, P. Two complementary sides of bioavailability: Accessibility and chemical activity of organic contaminants in sediments and soils. Environ. Toxicol. Chem. 2006, 25 (5), 1239−1245. (3) Bayen, S.; ter Laak, T. L.; Buffle, J.; Hermens, J. L. M. Dynamic exposure of organisms and passive samplers to hydrophobic chemicals. Environ. Sci. Technol. 2009, 43 (7), 2206−2215. (4) Donnelly, K. C.; Lingenfelter, R.; Cizmas, L.; Falahatpisheh, M. H.; Qian, Y.; Tang, Y.; Garcia, S.; Ramos, K.; Tiffany-Castiglioni, E.; Mumtaz, M. M. Toxicity assessment of complex mixtures remains a goal. Environ. Toxicol. Pharmacol. 2004, 18 (2), 135−141. (5) Altenburger, R.; Backhaus, T.; Boedeker, W.; Faust, M.; Scholze, M. Simplifying complexity: Mixture toxicity assessment in the last 20 years. Environ. Toxicol. Chem. 2013, 32 (8), 1685−1687. (6) Redman, A. D.; Parkerton, T. F.; McGrath, J. A.; Di Toro, D. M. Petrotox: An aquatic toxicity model for petroleum substances. Environ. Toxicol. Chem. 2012, 31 (11), 2498−2506. (7) Sprunger, L.; Proctor, A.; Acree, W. E., Jr; Abraham, M. H. Characterization of the sorption of gaseous and organic solutes onto polydimethyl siloxane solid-phase microextraction surfaces using the Abraham model. J. Chromatogr. 2007, 1175 (2), 162−173. (8) Endo, S.; Hale, S. E.; Goss, K.-U.; Arp, H. P. H. Equilibrium partition coefficients of diverse polar and nonpolar organic compounds to polyoxymethylene (POM) passive sampling devices. Environ. Sci. Technol. 2011, 45 (23), 10124−10132. (9) Lohmann, R. Critical review of low-density polyethylene’s partitioning and diffusion coefficients for trace organic contaminants and implications for its use as a passive sampler. Environ. Sci. Technol. 2012, 46 (2), 606−618. (10) DiFilippo, E. L.; Eganhouse, R. P. Assessment of PDMS-water partition coefficients: Implications for passive environmental sampling of hydrophobic organic compounds. Environ. Sci. Technol. 2010, 44 (18), 6917−6925. (11) Endo, S.; Droge, S. T. J.; Goss, K.-U. Polyparameter linear free energy models for polyacrylate fiber-water partition coefficients to evaluate the efficiency of solid-phase microextraction. Anal. Chem. 2011, 83 (4), 1394−1400. (12) Kamprad, I.; Goss, K.-U. Systematic investigation of the sorption properties of polyurethane foams for organic vapors. Anal. Chem. 2007, 79 (11), 4222−4227. (13) Abraham, M. H.; Ibrahim, A. Gas to olive oil partition coefficients: A linear free energy analysis. J. Chem. Inf. Model. 2006, 46 (4), 1735−1741. (14) Endo, S.; Goss, K.-U. Serum albumin binding of structurally diverse neutral organic compounds: Data and models. Chem. Res. Toxicol. 2011, 24 (12), 2293−2301. (15) Endo, S.; Mewburn, B.; Escher, B. I. Liposome and protein− water partitioning of polybrominated diphenyl ethers (PBDEs). Chemosphere 2013, 90 (2), 505−511. (16) 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 (3), 1496−1503. (17) Geisler, A.; Endo, S.; Goss, K.-U. Partitioning of organic chemicals to storage lipids: Elucidating the dependence on fatty acid composition and temperature. Environ. Sci. Technol. 2012, 46 (17), 9519−9524. I

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology (37) Huckins, J. N.; Petty, J. D.; Booij, K. Monitors of organic chemicals in the environment: Semipermeable membrane devices; Springer: New York, 2006. (38) Tuduri, L.; Millet, M.; Briand, O.; Montury, M. Passive air sampling of semi-volatile organic compounds. TrAC, Trends Anal. Chem. 2012, 31 (0), 38−49. (39) Smedes, F.; Geertsma, R. W.; Zande, T. v. d.; Booij, K. Polymerwater partition coefficients of hydrophobic compounds for passive sampling: Application of cosolvent models for validation. Environ. Sci. Technol. 2009, 43 (18), 7047−7054. (40) Mayer, P.; Vaes, W. H. J.; Hermens, J. L. M. Absorption of hydrophobic compounds into the poly(dimethylsiloxane) coating of solid-phase microextraction fibers: High partition coefficients and fluorescence microscopy images. Anal. Chem. 2000, 72 (3), 459−464. (41) Nabi, D. Estimating environmental partitioning, transport, and uptake properties for nonpolar chemicals using GC × GC. PhD Dissertation, EPFL, Lausanne, Switzerland, 2014. (42) Morgan, E. J.; Lohmann, R. Detecting air−water and surface− deep water gradients of PCBs using polyethylene passive samplers. Environ. Sci. Technol. 2008, 42 (19), 7248−7253. (43) Kennedy, K. E.; Hawker, D. W.; Müller, J. F.; Bartkow, M. E.; Truss, R. W. A field comparison of ethylene vinyl acetate and lowdensity polyethylene thin films for equilibrium phase passive air sampling of polycyclic aromatic hydrocarbons. Atmos. Environ. 2007, 41 (27), 5778−5787. (44) Bartkow, M. E.; Hawker, D. W.; Kennedy, K. E.; Müller, J. F. Characterizing uptake kinetics of PAHs from the air using polyethylene-based passive air samplers of multiple surface area-to-volume ratios. Environ. Sci. Technol. 2004, 38 (9), 2701−2706. (45) Thompson, J. M.; Hsieh, C.-H.; Luthy, R. G. Modeling uptake of hydrophobic organic contaminants into polyethylene passive samplers. Environ. Sci. Technol. 2015, 49 (4), 2270−2277. (46) Jahnke, A.; McLachlan, M. S.; Mayer, P. Equilibrium sampling: Partitioning of organochlorine compounds from lipids into polydimethylsiloxane. Chemosphere 2008, 73 (10), 1575−1581. (47) Werner, M.; Baars, A.; Eder, C.; Delgado, A. Thermal conductivity and density of plant oils under high pressure. J. Chem. Eng. Data 2008, 53 (7), 1444−1452. (48) Mayer, P.; Toräng, L.; Glæsner, N.; Jönsson, J. Å. Silicone membrane equilibrator: Measuring chemical activity of nonpolar chemicals with poly(dimethylsiloxane) microtubes immersed directly in tissue and lipids. Anal. Chem. 2009, 81 (4), 1536−1542. (49) Reichenberg, F.; Smedes, F.; Jö nsson, J.-Å.; Mayer, P. Determining the chemical activity of hydrophobic organic compounds in soil using polymer coated vials. Chem. Cent. J. 2008, 2 (1), 8. (50) Hung, W.-N.; Chiou, C. T.; Lin, T.-F. Lipid−water partition coefficients and correlations with uptakes by algae of organic compounds. J. Hazard. Mater. 2014, 279 (0), 197−202. (51) Smedes, F. Chapter 19 monitoring of chlorinated biphenyls and polycyclic aromatic hydrocarbons by passive sampling in concert with deployed mussels. In Comprehensive analytical chemistry; Greenwood, G. M., Vrana, B., Eds.; Elsevier: 2007; Vol. 48, pp 407−448. (52) Abraham, M. H.; Andonian-Haftvan, J.; Whiting, G. S.; Leo, A.; Taft, R. S. Hydrogen bonding. Part 34. The factors that influence the solubility of gases and vapours in water at 298 K, and a new method for its determination. J. Chem. Soc., Perkin Trans. 2 1994, 0 (8), 1777− 1791. (53) Tcaciuc, A. P.; Nelson, R. K.; Reddy, C. M.; Gschwend, P. M. Estimating phospholipid membrane-water partition coefficients using comprehensive two-dimensional gas chromatography. Environ. Sci. Technol. 2012, 46 (6), 3449−3456. (54) Endo, S.; Bauerfeind, J.; Goss, K.-U. Partitioning of neutral organic compounds to structural proteins. Environ. Sci. Technol. 2012, 46 (22), 12697−12703. (55) Goss, K.-U. Predicting equilibrium sorption of neutral organic chemicals into various polymeric sorbents with COSMO-RS. Anal. Chem. 2011, 83 (13), 5304−5308.

(56) Hung, H.-W.; Lin, T.-F.; Chiou, C. T. Partition coefficients of organic contaminants with carbohydrates. Environ. Sci. Technol. 2010, 44 (14), 5430−5436. (57) Ni, N.; Yalkowsky, S. H. Prediction of Setschenow constants. Int. J. Pharm. 2003, 254 (2), 167−172. (58) Lyakurwa, F.; Yang, X.; Li, X.; Qiao, X.; Chen, J. Development and validation of theoretical linear solvation energy relationship models for toxicity prediction to fathead minnow (pimephales promelas). Chemosphere 2014, 96, 188−194. (59) Gustafson, K. E.; Dickhut, R. M. Molecular diffusivity of polycyclic aromatic hydrocarbons in aqueous solution. J. Chem. Eng. Data 1994, 39 (2), 281−285. (60) Rusina, T. P.; Smedes, F.; Klanova, J. Diffusion coefficients of polychlorinated biphenyls and polycyclic aromatic hydrocarbons in polydimethylsiloxane and low-density polyethylene polymers. J. Appl. Polym. Sci. 2010, 116 (3), 1803−1810. (61) Mayer, P.; Fernqvist, M. M.; Christensen, P. S.; Karlson, U.; Trapp, S. Enhanced diffusion of polycyclic aromatic hydrocarbons in artificial and natural aqueous solutions. Environ. Sci. Technol. 2007, 41 (17), 6148−6155. (62) Rusina, T. P.; Smedes, F.; Koblizkova, M.; Klanova, J. Calibration of silicone rubber passive samplers: Experimental and modeled relations between sampling rate and compound properties. Environ. Sci. Technol. 2010, 44 (1), 362−367. (63) Gustafsson, K.; Björk, M.; Burreau, S.; Gilek, M. Bioaccumulation kinetics of brominated flame retardants (polybrominated diphenyl ethers) in blue mussels (mytilus edulis). Environ. Toxicol. Chem. 1999, 18 (6), 1218−1224. (64) Booij, K.; Smedes, F.; Van Weerlee, E. M.; Honkoop, P. J. C. Environmental monitoring of hydrophobic organic contaminants: The case of mussels versus semipermeable membrane devices. Environ. Sci. Technol. 2006, 40 (12), 3893−3900. (65) Abraham, M. H.; Poole, C. F.; Poole, S. K. Classification of stationary phases and other materials by gas chromatography. J. Chromatogr. 1999, 842 (1−2), 79−114. (66) Hale, S. E.; Martin, T. J.; Goss, K.-U.; Arp, H. P. H.; Werner, D. Partitioning of organochlorine pesticides from water to polyethylene passive samplers. Environ. Pollut. 2010, 158 (7), 2511−2517. (67) Abraham, M. H.; Al-Hussaini, A. J. M. Solvation descriptors for the polychloronaphthalenes: Estimation of some physicochemical properties. J. Environ. Monit. 2001, 3 (4), 377−381. (68) Abraham, M. H.; Andonian-Haftvan, J.; Whiting, G. S.; Leo, A.; Taft, R. S. Hydrogen bonding. Part 34. The factors that influence the solubility of gases and vapours in water at 298 K, and a new method for its determination. J. Chem. Soc., Perkin Trans. 2 1994, 0 (8), 1777− 1791. (69) Abraham, M. H.; Ibrahim, A. Gas to olive oil partition coefficients: A linear free energy analysis. J. Chem. Inf. Model. 2006, 46 (4), 1735−1741. (70) Abraham, M. H.; Nasezadeh, A.; Acree, W. E. Correlation and prediction of partition coefficients from the gas phase and from water to alkan-1-ols. Ind. Eng. Chem. Res. 2008, 47 (11), 3990−3995. (71) ACD/ADME. ACD/ADME suite, version 5.0; Advanced Chemistry Development, Inc., Toronto, ON, Canada, 2010. (72) Endo, S.; Schmidt, T. C. Prediction of partitioning between complex organic mixtures and water: Application of polyparameter linear free energy relationships. Environ. Sci. Technol. 2006, 40 (2), 536−545. (73) Goss, K.-U. Prediction of the temperature dependency of Henry’s law constant using poly-parameter linear free energy relationships. Chemosphere 2006, 64 (8), 1369−1374. (74) 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 by-products. Water Res. 2012, 46 (11), 3637−3645. (75) van Noort, P. C. M. Solvation thermodynamics and the physical−chemical meaning of the constant in Abraham solvation equations. Chemosphere 2012, 87 (2), 125−131. J

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

(94) Greenwood, R.; Mills, G. A.; Vrana, B. Passive sampling techniques in environmental monitoring; Elsevier Science Limited: 2007; Vol. 48. (95) Hayduk, W.; Laudie, H. Prediction of diffusion coefficients for nonelectrolytes in dilute aqueous solutions. AIChE J. 1974, 20 (3), 611−615. (96) Sijm, D. T. H. M.; van der Linde, A. Size-dependent bioconcentration kinetics of hydrophobic organic chemicals in fish based on diffusive mass transfer and allometric relationships. Environ. Sci. Technol. 1995, 29 (11), 2769−2777. (97) Richardson, B. J.; Tse, E. S. C.; De Luca-Abbott, S. B.; Martin, M.; Lam, P. K. S. Uptake and depuration of PAHs and chlorinated pesticides by semi-permeable membrane devices (SPMDs) and greenlipped mussels (perna viridis). Mar. Pollut. Bull. 2005, 51 (8−12), 975−993. (98) Xia, X.; Li, H.; Yang, Z.; Zhang, X.; Wang, H. How does predation affect the bioaccumulation of hydrophobic organic compounds in aquatic organisms? Environ. Sci. Technol. 2015, 49 (8), 4911−4920. (99) McLeod, A. M.; Paterson, G.; Drouillard, K. G.; Haffner, G. D. Ecological implications of steady state and nonsteady state bioaccumulation models. Environ. Sci. Technol. 2016, 50 (20), 11103−11111. (100) Mackay, D.; Arnot, J. A.; Gobas, F. A. P. C.; Powell, D. E. Mathematical relationships between metrics of chemical bioaccumulation in fish. Environ. Toxicol. Chem. 2013, 32 (7), 1459−1466. (101) Kwon, J.-H.; Lee, S.-Y.; Kang, H.-J.; Mayer, P.; Escher, B. I. Including bioconcentration kinetics for the prioritization and interpretation of regulatory aquatic toxicity tests of highly hydrophobic chemicals. Environ. Sci. Technol. 2016, 50 (21), 12004−12011. (102) Mayer, P.; Reichenberg, F. Can highly hydrophobic organic substances cause aquatic baseline toxicity and can they contribute to mixture toxicity? Environ. Toxicol. Chem. 2006, 25 (10), 2639−2644. (103) Regulation for registration, evaluation, authorisation and restriction of chemicals (REACH); European Commission, 2007. (104) Gobas, F. A. P. C.; de Wolf, W.; Burkhard, L. P.; Verbruggen, E.; Plotzke, K. Revisiting bioaccumulation criteria for POPs and PBT assessments. Integr. Environ. Assess. Manage. 2009, 5 (4), 624−637. (105) Goss, K.-U.; Brown, T. N.; Endo, S. Elimination half-life as a metric for the bioaccumulation potential of chemicals in aquatic and terrestrial food chains. Environ. Toxicol. Chem. 2013, 32 (7), 1663− 1671. (106) Efron, B.; Tibshirani, R. Bootstrap measures for standard errors, confidence intervals, and other measures of statistical accuracy. Stat. Sci. 1986, 1, 54−77.

(76) van Noort, P. C. M.; Haftka, J. J. H.; Parsons, J. R. Updated Abraham solvation parameters for polychlorinated biphenyls. Environ. Sci. Technol. 2010, 44 (18), 7037−7042. (77) West, C.; Lesellier, E. Characterisation of stationary phases in subcritical fluid chromatography with the solvation parameter model: III. Polar stationary phases. J. Chromatogr. 2006, 1110 (1−2), 200− 213. (78) Dewulf, J.; Van Langenhove, H.; Evaraert, M. Solid-phase microextraction of volatile organic compounds estimation of the sorption equilibrium from the Kováts index, effect of salinity and humic acids and the study of the kinetics by the development of an “agitated/static layer” model. J. Chromatogr. 1997, 761 (1−2), 205− 217. (79) Jahnke, A.; Mayer, P. Do complex matrices modify the sorptive properties of polydimethylsiloxane (PDMS) for non-polar organic chemicals? J. Chromatogr. 2010, 1217 (29), 4765−4770. (80) Zhang, X.; Oakes, K. D.; Hoque, M. E.; Luong, D.; Metcalfe, C. D.; Pawliszyn, J.; Servos, M. R. Pre-equilibrium solid-phase microextraction of free analyte in complex samples: Correction for mass transfer variation from protein binding and matrix tortuosity. Anal. Chem. 2011, 83 (9), 3365−3370. (81) Jahnke, A.; Mayer, P.; Broman, D.; McLachlan, M. S. Possibilities and limitations of equilibrium sampling using polydimethylsiloxane in fish tissue. Chemosphere 2009, 77 (6), 764−770. (82) Jahnke, A.; Mayer, P.; Adolfsson-Erici, M.; McLachlan, M. S. Equilibrium sampling of environmental pollutants in fish: Comparison with lipid-normalized concentrations and homogenization effects on chemical activity. Environ. Toxicol. Chem. 2011, 30 (7), 1515−1521. (83) Tomy, G. T.; Fisk, A. T.; Westmore, J. B.; Muir, D. C. G., Environmental chemistry and toxicology of polychlorinated n-alkanes. In Reviews of environmental contamination and toxicology; Ware, G., Ed.; Springer: New York, 1998; Vol. 158, pp 53−128. (84) Di Toro, D. M.; McGrath, J. A.; Stubblefield, W. A. Predicting the toxicity of neat and weathered crude oil: Toxic potential and the toxicity of saturated mixtures. Environ. Toxicol. Chem. 2007, 26 (1), 24−36. (85) Verhaar, H. J. M.; van Leeuwen, C. J.; Hermens, J. L. M. Classifying environmental pollutants. Chemosphere 1992, 25 (4), 471− 491. (86) Verhaar, H. J. M.; Urrestarazu Ramos, E.; Hermens, J. L. M. Classifying environmental pollutants. 2: Separation of class 1 (baseline toxicity) and class 2 (‘polar narcosis’) type compounds based on chemical descriptors. J. Chemom. 1996, 10 (2), 149−162. (87) Bowen, K. R.; Flanagan, K. B.; Acree, W. E., Jr; Abraham, M. H. Correlating toxicities of organic compounds to select protozoa using the Abraham model. Sci. Total Environ. 2006, 369 (1−3), 109−118. (88) Hoover, K. R.; Flanagan, K. B.; Acree, J. W. E.; Abraham, M. H. Chemical toxicity correlations for several protozoas, bacteria, and water fleas based on the Abraham solvation parameter model. J. Environ. Eng. Sci. 2007, 6 (2), 165−174. (89) Bowen, K. R.; Flanagan, K. B.; Acree, W. E., Jr; Abraham, M. H.; Rafols, C. Correlation of the toxicity of organic compounds to tadpoles using the Abraham model. Sci. Total Environ. 2006, 371 (1−3), 99− 109. (90) Escher, B. I.; Hermens, J. L. M. Peer reviewed: Internal exposure: Linking bioavailability to effects. Environ. Sci. Technol. 2004, 38 (23), 455A−462A. (91) Schmitt, W. General approach for the calculation of tissue to plasma partition coefficients. Toxicol. In Vitro 2008, 22 (2), 457−467. (92) Barber, M. C. A review and comparison of models for predicting dynamic chemical bioconcentration in fish. Environ. Toxicol. Chem. 2003, 22 (9), 1963−1992. (93) ter Laak, T. L.; Busser, F. J. M.; Hermens, J. L. M. Poly(dimethylsiloxane) as passive sampler material for hydrophobic chemicals: Effect of chemical properties and sampler characteristics on partitioning and equilibration times. Anal. Chem. 2008, 80 (10), 3859−3866. K

DOI: 10.1021/acs.est.6b05071 Environ. Sci. Technol. XXXX, XXX, XXX−XXX