Particle Partitioning. 3. Estimating Partition Coefficients

Feb 18, 2009 - Equilibrium gas/particle partitioning coefficients of terrestrial aerosols, Kip, are dependent on various intermolecular interactions t...
0 downloads 11 Views 1MB Size
Download PDF
Environ. Sci. Technol. 2009, 43, 1923–1929

Ambient Gas/Particle Partitioning. 3. Estimating Partition Coefficients of Apolar, Polar, and Ionizable Organic Compounds by Their Molecular Structure H A N S P E T E R H . A R P * ,† A N D K A I - U W E G O S S * ,§ Department of Environmental Engineering, Norwegian Geotechnical Institute (NGI), P.O. Box 3930 Ullevål Stadion, N-0806, Oslo, Norway, and UFZ, Helmholtz Center for Environmental Research - UFZ, Permoserstrasse 15, 04318 Leipzig, Germany

Received September 5, 2008. Revised manuscript received January 15, 2009. Accepted January 16, 2009.

Equilibrium gas/particle partitioning coefficients of terrestrial aerosols, Kip, are dependent on various intermolecular interactions that can be quantified by experimentally determined compoundspecific descriptors. For many compounds of environmental interest, such as emerging contaminants and atmospheric phototransformation products, these compound-specific descriptors are unknown or immeasurable. Often, only the molecular structure is known. Here we present the ability of two computer programs to predict equilibrium partitioning to terrestrial aerosols solely on the basis of molecular structure: COSMOtherm and SPARC. The greatest hurdle with designing such an approach is to identify suitable molecular surrogates to represent the dominating sorbing phases, which for ambient terrestrial aerosols are the water insoluble organic matter (WIOM) phase and the mixed-aqueous phase. For the WIOM phase, hypothetical urban secondary organic aerosol structural units from Kalberer et al. Science 2004, 303, 1659-1662 were investigated as input surrogates, and for the mixedaqueous phase mildly acidic water was used as a surrogate. Using a validation data set of more than 1400 experimentally determined Kip values for polar, apolar, and ionic compounds ranging over 9 orders of magnitude (including semivolatile compounds such as PCDD/Fs, pesticides, and PBDEs), SPARC and COSMOtherm were generally able to predict Kip values well within an order of magnitude over an ambient range of temperature and relative humidity. This is remarkable as these two models were not fitted or calibrated to any experimental data. As these models can be used for potentially any organic molecule, they are particularly recommended for environmental screening purposes and for use when experimental compound descriptor data are not available.

* Address correspondence to either author. Phone: ++47 2202 1988 (H.P.H.A.); ++ 49 341 235 1411 (K.-U.G.). E-mail: [email protected]; [email protected]. † NGI. § UFZ. 10.1021/es8025165 CCC: $40.75

Published on Web 02/18/2009

 2009 American Chemical Society

Introduction The ability to accurately predict an environmentally relevant partitioning process solely on the basis of molecular structure would have countless advantages, including less need for challenging experimental measurements, the rapid screening of large chemical data sets (such as those outlined in appendices of legislation such as REACH (1)), and the inclusion of diverse transformation products in environmental fate models. All these advantages become particularly apparent when one considers ambient atmospheric gas/ particle partitioning. Continuously a wide variety of organic compounds are emitted into the atmosphere, where they are phototransformed into an even larger variety of daughter products. Some of these daughter products are more toxic than the parent compounds themselves (e.g., certain nitroPAHs are more toxic than PAHs (2)). Thus, there is a present need to be able to model compound-specific gas/particle partitioning constants, Kip, on the basis of molecular structure. Though there is a great deal of such models for chemical partition systems (3), relatively few models exist for environmental partition systems, such as terrestrial aerosols. Currently, the most simplistic and widely used method of predicting Kip values based on molecular structure is through single-parameter linear free energy relationships (SP-LFERs), which assume that Kip is log-log linearly correlated with either the subcooled liquid vapor pressure (p*iL) or the octanol-air partition coefficient (Kioa) (e.g., refs 4 and 5). Both of these parameters can be estimated via molecular structure using methods such as UNIFAC (6) or the SPARC online calculator (7, 8). However, SP-LFERs lack chemical robustness. It has long been established that when SP-LFERs are calibrated to a set of diverse polar and apolar chemicals weak statistical correlations result, because these models simplify or ignore the diversity of specific and nonspecific sorbent-sorbate interactions (9, 10). Thus, at best, SP-LFERs are only accurate for chemicals that exhibit similar sorbent-sorbate interactions and must be recalibrated for different compound classes. Earlier in this series, we demonstrated that to successfully model Kip values for apolar, polar, and ionizable organic compounds collectively for ambient terrestrial aerosols, only two sorption components need to be accounted for: the mixed-aqueous phase and the water-insoluble organic matter (WIOM) phase (11). Sorption to the mixed-aqueous phase can be modeled successfully by assuming it is similar to moderately acidic water (11). Sorption to the WIOM can be modeled successfully using a representative poly-parameter linear free energy relationship (PP-LFER) that uses Abraham descriptors to quantify the diverse specific and nonspecific interactions sorbates can undergo with WIOM (10). PP-LFERs based on Abraham descriptors generally boast a very high degree of accuracy while at the same time they are applicable to a diverse variety of polar and apolar compounds. However, the use of these PP-LFERs for compounds that have unknown Abraham descriptors is problematic. Methods to predict Abraham descriptors on the basis of molecular structure exist (e.g., ref 12), but the reliability of these methods is in question (13, 14). Thus, to accurately predict Kip values solely on the basis of molecular structure, different modeling approaches are needed than done heretofore. In this work, we present the ability of two computer programs to predict partitioning into the WIOM and mixedaqueous phase simultaneously solely on the basis of molecular structure: COSMOtherm (15, 16) and SPARC (7, 8). What makes these two programs suitable is that they allow VOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1923

the user to predict partitioning constants at infinite dilution for essentially any organic sorbate-sorbent pair. Instead of identifying sorbent LFER descriptors, users of these models must identify a surrogate molecule to represent the sorbing phase. Though the exact molecular structure of a heterogeneous natural organic phase may be impossible to use as input, such as aerosol WIOM whose structure has not really been identified yet, it may be possible to use a simpler organic structure provided it has similar properties. Our recent experience with humic acid (17) encouraged us to look for such a simple molecular model for aerosol WIOM.

Ki sorbent, air(m3/g) ) RT/(MWsorbent · γi sorbent · p*iL)

Materials and Methods Dual Phase Sorption Mechanism. Kip is defined as: Kip(m3/g) ) cip/ci air

(1)

where cip is the equilibrium concentration of compound “i” sorbed to the particle phase (molip/gp) and ci air is the equilibrium concentration of “i” in the air phase (moli air/m3air). In the present work, cip and therefore Kip are normalized to the dry aerosol mass, Mdry. Kip values normalized to Mdry for apolar, polar, and ionic compounds can be described using the following dual phase sorption equation (11): Kip ) fWIOM · KiWIOM + VwRH/(Diaw · Mdry)

(2)

where fWIOM is the mass fraction of WIOM in dry aerosol (gWIOM/Mdry) and KiWIOM is the sorption coefficient of WIOM (ciWIOM/ci air). As presented in part 2 of this series, KiWIOM appears to be similar for all ambient terrestrial aerosols, regardless of sampling location, though fWIOM can vary (10). VwRH in eq 2 is the volume of water in the aerosol sample (m3water) (which is dependent on the RH and the aerosol sample’s hygroscopicity), and Diaw is the dimensionless air-water distribution coefficient: Diaw ) Ria(Kiaw) · S for organic acids, and

(3)

Diaw ) (1 - Ria)(Kiaw) · S for organic bases

(4)

where Kiaw is the dimensionless air-water partition coefficient for pure water (ci air/ci water), S is a unitless empirical factor to account for deviation from the sorption behavior in pure water (it can assumed to be 1 for terrestrial aerosols (11)), and Ria is the fraction of the compound in the protonated form (necessary only for ionizing compounds) and is dependent on the aerosol’s pH and compound’s acid dissociation constant, pKa. Kip Validation Set. Using a recently developed technique involving inverse gas chromatography (IGC) (18), we determined more than 1300 Kip values covering diverse ranges of RH (50-90%), temperature (15-55 °C), aerosol samples (covering the four seasons, desert dust, rural, coastal, urban, and suburban particles), and compound classes (polar, apolar, ionizable) (10, 11, 18). All these data are used here as part of the validation set along with more than 100 Kip values for semivolatile organic compounds (SVOCs) from elsewhere in the literature, including data for polyaromatic hydrocarbons (PAHs) (19-21), polychlorinated biphenyls (PCBs) (22, 23), polychlorodibenzodioxins and -furans (PCDD/ Fs) (19, 24, 25), n-alkanes (26), organochlorine pesticides (27), and polybrominated diphenyl ethers (PBDEs) (28). Note that the Kip values for SVOCs were measured using so-called “sample-and-extract” methods, which are prone to more substantial sampling artifacts than the IGC method (such as fluctuating ambient conditions, removal of nonexchangeable compounds, filter sorption artifacts, etc.) (18). COSMOtherm. COSMOtherm (v. 2.1, COSMOlogic) (15, 16) is a commercial software that performs density functional quantum chemical continuum solvation calculations with 1924

9

statistical thermodynamics to determine activity coefficients and vapor pressures. Input files are generated on the basis of the molecular structure of the sorbent and sorbate using software packages such as Turbomole (used here, v. 5.0) or Gaussian. COSMOtherm was used to calculate the p*iL values and the activity coefficients of the considered chemicals in various sorbents, γi sorbent, such as water and various molecular surrogates for the WIOM phase (see below). The COSMOtherm output was then converted to Ki WIOM or Kiaw on the basis of the general definition of partitioning between air and a condensed phase (29):

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 6, 2009

(5)

where Ki sorbent, air is the equilibrium partitioning constant between the sorbent and the air phase, R is the ideal gas constant (m3 · Pa /mol · K), T is the temperature (K), and MWcondensed is the molecular weight of the sorbent (g/mol). SPARC. SPARC is a free, online Web application (http:// ibmlc2.chem.uga.edu/sparc/) that explicitly calculates sorbate-sorbent interactions by using various empirical molecular descriptors that are derived from molecular structure (7, 8). SPARC requires the input of molecular structure in the form of a SMILES string. As with COSMOtherm, SPARC calculates both p*iL and γi sorbent separately, but SPARC can be set up to give the output directly as a K value. SPARC is periodically updated online. During the preparation of this article, two versions were available: v. 4.1 (accessed March 7 and 8, 2008) and v. 4.2 (accessed March 26 and 27, 2008). Here, outputs from both versions are presented. Input Structures in COSMOtherm and SPARC. To calculate γi sorbent for the mixed-aqueous phase, γi mixed-aqueous, the choice of the input structure in COSMOtherm and SPARC was clear: H2O. The input structure required to estimate γi sorbent for WIOM (i.e., γi WIOM), was not so immediately straightforward. To address this, we sought a molecular structure that was based on the molecular composition of ambient WIOM itself, rather than that of an arbitrary molecule that happens to give occasional good correlations, such as n-octanol, which is commonly used as a surrogate for environmental phases. Organic matter in ambient urban aerosols consists largely of secondary organic aerosol (SOA) polymers (30). These polymers are not very hygroscopic and are therefore likely candidates for being major components of the ambient WIOM phase. Typical precursors for SOA polymers in urban atmospheres are small volatile aromatic compounds, such as 1,3,5-trimethylbenzene (TMB), the photooxidation products of which readily undergo acetal polymerization with methyglyoxal as the main polymer unit (30). Many oxidation products can be incorporated into the resulting oligomers. Three possible structural units for the polymerization of TMB photooxidation products are given by Kalberer et al. (30), which are shown here in Figure 1 and are referred to as WIOM A, WIOM B, and WIOM C (from the pure methyglyoxal polymer).

Results and Discussion To test the accuracy of KiWIOM predictions based on WIOM A, B, and C, only Kip data for compounds that partition mainly in the WIOM phase are considered. Thus, potentially ionizable compounds and small polar compounds (ethanol, 1-propanol, 2-propanol, n-propanoic acid, and 1,4-dioxane) are excluded as these favor the mixed-aqueous phase. Further, only IGC-measured Kip values at exactly 15 °C and 50% RH are included in the validation data set to eliminate interferences caused by fluctuating ambient conditions, and because the aqueous phase is comparatively negligible at this RH (ref 11). For this chemical data set, eq 2 simplifies to Kip ) fWIOM · KiWIOM. For the initial comparison, a typical fWIOM value of 0.1 is assumed, as

FIGURE 1. Hypothetical structural units present in SOA polymers from the photooxidation of 1,3,5-trimethylbenzene as presented in Kalberer et al. (30). The SMILES strings for these compounds are WIOM A [CC(dO)C1OC2OC(C)(OC2(C)O1)C(O)dO], WIOM B [c(c(cc1C(OC2(OC3C(dO)C)C)OC2O3)C)c(c1)C], and WIOM C [CC(dO)C1OC2OC(C)(OC2(C)O1)C3(C)OC(O)C(C)(O)O3].

FIGURE 2. Comparison of measured and predicted log Kip values (15 °C, 50% RH) for more than 70 diverse apolar and polar compounds. Predictions are done using COSMOtherm (a-c) and SPARC v. 4.2 (d-f) using an fWIOM value of 0.1 and either WIOM A, WIOM B, or WIOM C as input structures of the sorbing phase. The measured values are averages of nine diverse aerosol samples. No calibration or fitting was done. measured fWIOM values range between 5 and 17% (based on refs 19 and 31 and assuming that approximately half the total OM is WIOM, after refs 32 and 33). Average experimental Kip values (15 °C, 50% RH) from our nine aerosol samples for more than 70 diverse polar and apolar compounds are plotted against predicted COSMOtherm and SPARC v. 4.2 values (15 °C, fWIOM ) 0.1) in Figure 2. As is evident from Figure 2, the COSMOtherm and SPARC v. 4.2 output for the WIOM B structure exhibited the best agreement with experimental log Kip values (Figure 2b,e). Similar results can be found for SPARC v. 4.1, as shown in the Supporting Information. Variations of WIOM B were also tested as input structures, such as WIOM B with a keto group removed (making it less polar) and a combination of WIOM A and B (making it more polar). However, all attempted variations resulted in no significant improvement (data not shown). The agreement between WIOM B and the average experimental data is encouraging, especially when one considers that these data were neither calibrated nor fitted. They are pure, estimated values. The clustering of data near the 1:1 line for the WIOM B predictions in Figure 2 indicates that the fWIOM value of 0.1 was an appropriate assumption; otherwise, the data would

be clustered systematically away from the 1:1 line. In Part 1 of the Supporting Information, it is presented that fWIOM values that are fitted, by regressing estimated KiWIOM values against experimental Kip values for specific aerosols (instead of average values), deviate from 0.1 by at most a factor of 3, depending on the aerosol sample and estimation software. The average of all fitted fWIOM values was 0.12 ( 0.07. Thus, we conclude that an fWIOM value of 0.1 along with aerosol surrogate WIOM B is an appropriate assumption for the PM 10 fraction of terrestrial aerosols (as used in this study) for environmental screening purposes and larger scale fate models, though a more appropriate fWIOM value may be needed in special cases (for more details, refer to Part 1 of the Supporting Information), such as specific samples and smaller particle size fractions. For the remainder of this article, all predictions are done using WIOM B as an aerosol surrogate. Simulation of Temperature Dependence of WIOM. To simulate the temperature dependence of WIOM sorption, both COSMOtherm and SPARC were used to predict the enthalpy of partitioning into the WIOM phase, ∆HiWIOM. This was performed using predicted Ki WIOM values at three different VOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1925

FIGURE 3. COSMOtherm and SPARC (v. 4.2) log Kip predictions compared with average literature values. Note that the COSMOtherm for PBDEs (35) and PCDD/Fs (36) are based on experimental p*iL values from the literature combined with calculated γiWIOM values. temperatures (5, 15, and 25 °C) and inputting these data into the van’t Hoff relationship (eq 6). log KiWIOM ) (∆HiWIOM + RTa)/(2.303 · R · T) + constant (6) where Ta is the average temperature (here 15 °C). All calculated enthalpies using COSMOtherm and SPARC v. 4.2 are shown in Part 2 of the Supporting Information (unfortunately, SPARC v. 4.1 went offline before the enthalpy calculations could be completed). Previously, we measured the enthalpy of gas/particle partitioning, ∆Hip, for two aerosol samples, referred to as the Duebendorf Fall sample (51 compounds) and Roost sample (50 compounds) (10). Of the 44 compounds that were measured for both samples, the ∆Hip values were on average within 16% of each other, that is, ∆Hip Duebendorf Fall ) (1.02 ( 0.16) · ∆Hip Roost. A comparison of measured ∆Hip values for both aerosol samples with estimated ∆HiWIOM values using COSMOtherm and SPARC is presented in eqs 7 and 8: ∆Hip ) (0.91 ( 0.17) · ∆HiWIOM (COSMOtherm, n ) 101) (7) ∆Hip ) (1.13 ( 0.34) · ∆HiWIOM (SPARC v. 4.2, n ) 101) (8) Thus, from eq 7, experimental ∆Hip values are on average 0.91 of the COSMOtherm-predicted ∆HiWIOM value, though they can deviate by more than 17% (which is similar to the standard deviation for the enthalpies for the two aerosols samples considered). This correlation is similar to that previously reported between measured ∆Hip and experimental enthalpies of vaporization, ∆Hi vap (10). Thus, use of COSMOtherm to predict the temperature dependence of Kip values is considered suitable over narrow, ambient temperature ranges. SPARC v. 4.2 predictions are not as good as those of COSMOtherm, especially for apolar compounds (Part 2 of the Supporting Information). SVOCs. The available ambient gas/particle partitioning data for SVOCs from the literature are limited to apolar and weakly polar compounds, which have a much higher affinity for organic phases relative to the aqueous phase. Thus, the COSMOtherm and SPARC predictions performed here assume the SVOCs sorb exclusively to WIOM. It should be noted beforehand that COSMOtherm is known to have difficulty predicting p*iL of certain highly halogenated SVOCs (34). Correspondingly, we found that COSMOtherm predicted p*iL values deviated by more than an order of magnitude from reported p*iL values for PBDEs (35) and highly chlorinated PCDD/Fs (36). As a result, for PBDEs and PCDD/Fs only, we replaced COSMOtherm-predicted p*iL values with literature 1926

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 6, 2009

values (35, 36) and used COSMOtherm for predicting γiWIOM only. In Figure 3, log Kip predictions for SVOCs using COSMOtherm and SPARC v. 4.2 (fWIOM ) 0.1, T ) 15 °C) are compared with average literature Kip values from various studies. Note that these average literature values combine data covering various aerosol samples, ambient temperatures (from below 0 to 40 °C), and sampling methods and are only averaged here for the sake of being concise. In Part 3 of the Supporting Information, plots similar to Figure 3 can be found for each of the individual compound classes using nonaverage, original Kip values from the literature, as well as a plot showing all SVOCs and VOCs. As is evident from Figure 3, SPARC and the (partially p*iL adjusted) COSMOtherm predictions at 15 °C agree with average literature Kip values for most compounds within an order of magnitude and almost always within 2 orders of magnitude. Regarding compounds that are generated in combustion processes, namely PAHs and PCDD/Fs, average Kip are generally well estimated or underestimated. Note that certain Kip measurements of PAHs are associated with experimental errors, especially the extraction of PAHs from soot cores, which causes a scatter in reported Kip values of more than 2 orders of magnitude (Part 3 of the Supporting Information and ref 10). The estimations here correlated with the minimum of reported Kip for PAHs. Noncombustion SVOCs, such as PBDEs, pesticides, and PCBs, are generally well predicted or overpredicted by COSMOtherm and SPARC. However, this overprediction is only more than 1 order of magnitude for COSMOtherm predictions of hexachlorohexanes (which were not p*iL-corrected) and SPARC predictions of PBDEs (which is partially explained by temperature; see below). Ignoring sampling artifacts, the predictions in Figure 3 are biased because of the choice of two assumed input parameters: fWIOM ) 0.1 and T ) 15 °C. For SVOCs in particular, the input temperature can have a substantial influence on the output, much more so than fWIOM. Predicted ∆HiWIOM values for many SVOCs are near 100 kJ/mol; thus, a change in input temperature by 15 °C can influence predictions by more than an order of magnitude. Thus, the Kip of PBDEs being generally overestimated by our models is partly influenced by the measurements being taken at temperatures 12-17 °C higher than the temperature assumed in our estimations (28). Simulation of Dual Phase Sorption. To validate if COSMOtherm and SPARC can successfully predict sorption into both the WIOM and aqueous phases simultaneously, Kip data for moist aerosols at 70 and 90% RH, along with compounds that could readily sorb into the aqueous phase

FIGURE 4. Comparison of experimental log Kip values (15 °C) of polar, apolar, small polar, and ionizable compounds for the Berlin Spring aerosol sample at various RH values with COSMOtherm predictions (15 °C, fWIOM ) 0.1), assuming the mixed-aqueous phase is negligible and partitioning occurs into the WIOM phase only.

FIGURE 5. Comparison of experimental log Kip values (15 °C) for various polar, apolar, small polar, and ionizable compounds at various RH values for the Berlin Spring aerosol sample using COSMOtherm (a-c, fWIOM ) 0.1) and SPARC v. 4.1 (d-f, fWIOM ) 0.1) and the dual phase sorption equation (eq 2). (i.e., small polar compounds and ionizable compounds) were included in the validation data set. To illustrate the importance of partitioning to the aqueous phase, in Figure 4 experimental values for the Berlin Spring aerosol sample (from ref 11) at increasing RH are compared with COSMOtherm-predicted Kip assuming partitioning into the WIOM phase only. As is evident from Figure 4, as RH increases the Kip value of several ionizable, small polar, and other polar compounds also increases with respect to the estimated value based on WIOM exclusive partitioning. This is attributable to additional partitioning occurring into the mixed-aqueous phase. At 90% RH (Figure 4c), predictions based on WIOM exclusive sorption underestimate several experimental Kip values by more than 1 order of magnitude. To account for additional partitioning into the mixedaqueous phase, COSMOtherm- and SPARC-predicted Kiaw

and KiWIOM values were placed in the dual phase sorption equation (eq 2), which also accounts for the pH of the aqueous phase and the pKa of the sorbate. The results for the Berlin Spring aerosol sample are displayed in Figure 5. Similar plots for the Duebendorf-Sahara and Roost sample using COSMOtherm and SPARC v. 4.1 and v. 4.2 are presented in Part 4 of the Supporting Information, using fitted and assumed fWIOM values. As evident from Figure 5 and similar plots in Part 4 of the Supporting Information, COSMOtherm and SPARC predictions based on the dual phase sorption equation were almost always within 1 order of magnitude from the measured values for all compounds. In fact, these nonfitted models performed comparably to estimations based on fitted PP-LFERs and literature Kiaw values (as presented in ref 11). Thus, the accuracy is remarkable for a nonfitted model. Note that the VOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1927

general tendency for the estimated values to be slightly larger than the measured value is largely accountable by the assumed fWIOM values in Figure 5 of 0.1 being approximately twice the fitted value for the Berlin Spring aerosol sample of 0.044. Use and Limitations of COSMOtherm and SPARC Models for Environmental Screening. Both COSMOtherm and SPARC are capable of modeling Kip values of terrestrial aerosols with an accuracy that is sufficient for environmental screening purposes. This applies to all compounds be they apolar, polar, ionizable, or semivolatile. There are, however, uses and limitations for these models that users must take into account. Appropriateness of Molecular Surrogate. The ability to predict γiWIOM accurately depends on the accuracy of the estimation model used and the choice of molecular surrogate. It cannot be identified here whether WIOM B performed well as a molecular surrogate because it is a typical structure in terrestrial WIOM, or simply because it is representative of the various specific and nonspecific absorption interactions that WIOM can undergo with organic compounds (WIOM likely consists of a large heterogeneous mix of compounds and polymers). Though WIOM B is proposed as a structural unit specifically for urban SOA, on the basis of the results shown in part 2 of this series, it is not surprising that the same molecular surrogate worked for both urban and rural aerosols, as the WIOM in terrestrial aerosols was found to exhibit similar sorption behavior regardless of location and season (10). Correspondingly, it has been hypothesized that urban SOA polymers are structurally similar to rural SOA polymers, even though rural SOA precursors are different (30). Whether WIOM B is appropriate for modeling the organic phase in nonterrestrial aerosols, such as marine, combustion, or indoor aerosols, is unknown. Fluctuating fWIOM, Temperature, RH, and pH. Because of natural variability, ambient conditions (RH and temperature) and aerosol properties (fWIOM, pH, hygroscopicity) will vary. Thus, for environmental fate models, a sensitivity analysis for all these parameters is recommended. Chemical Applicability Domain. Though Kip values for most compounds were predicted with a good degree of accuracy, some compound classes were unaccountably and substantially under- or overestimated by either COSMOtherm or SPARC. Compounds that were systematically underestimated include 2,6-dichlorophenol (SPARC v. 4.2) and 2,6dimethylaniline (COSMOtherm). Compounds that were overestimated are several highly halogenated SVOCs (COSMOtherm) and fluorotelomer alcohols (FTOHs) (SPARC v. 4.1). In the case of SPARC, the outlying compound classes were identified to be dependent on the software version. FTOHs were overestimated by several orders of magnitude using SPARC v. 4.1 (Supporting Information) but were estimated accurately using SPARC v. 4.2. Similarly, 2,6chlorophenol was estimated accurately with SPARC v. 4.1 but often was underestimated by more than an order of magnitude with SPARC v. 4.2. Though the chemical applicability domain of COSMOtherm and SPARC is definitely the largest and most robust available for modeling Kip values, the existence of outlying compound classes indicates that this domain is not universal. Fortunately, the chemical applicability domains are different between COSMOtherm and SPARC, and even among differing versions of SPARC. Thus, we recommend predicting Kip values using more than one software package to check if the values are similar. SP-LFERs Are Obsolete. Despite the above-mentioned limitations, the chemical application domain of these models is clearly more robust than those based on simple SP-LFERs mentioned in the Introduction, mainly because the γi sorbent term in eq 5 is dealt with explicitly. It is not assumed to be 1928

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 6, 2009

similar to γi octanol (as when using SP-LFERs based on Kioa) nor to be similar for all compounds (as often assumed when using SP-LFERs based on p*iL). Unlike SP-LFERs, the models presented here are shown to perform well even without calibration and without the need for the experimental determination of compound data (such as p*iL or Kioa). Thus, because of the explicit nature of the presented COSMOtherm and SPARC modeling approaches, the robustness of their chemical application domains, the lack of need for any experimental data, and the ease of use and the general accuracy of these models, the presented modeling approaches show much greater promise than that conceivable with SP-LFERs.

Acknowledgments The Research Council of Norway (NFR) contributed financially through Project 178141 in the “HavKyst” (Coast and Sea) program.

Supporting Information Available Discussion on fWIOM variability, model output, and additional plots. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) Scheringer, M.; Boschen, S.; Hungerbuhler, K. Will we know more or less about chemical risks under REACH? Chimia 2006, 60 (10), 699–706. (2) Zielinska, B. Atmospheric transformation of diesel emissions. Exp. Toxicol. Pathol. 2005, 57, 31–42. (3) Lei, Z. G.; Chen, B. H.; Li, C. Y.; Liu, H. Predictive molecular thermodynamic models for liquid solvents, solid salts, polymers, and ionic liquids. Chem. Rev. 2008, 108 (4), 1419–1455. (4) Pankow, J. F. An absorption model of gas/particle partitioning of organic compounds in the atmosphere. Atmos. Environ. 1994, 28 (2), 185–188. (5) Finizio, A.; Mackay, D.; Bidleman, T.; Harner, T. Octanol-air partition coefficient as a predictor of partitioning of semi-volatile organic chemicals to aerosols. Atmos. Environ. 1997, 31 (15), 2289–2296. (6) Asher, W. E.; Pankow, J. F.; Erdakos, G. B.; Seinfeld, J. H. Estimating the vapor pressures of multi-functional oxygencontaining organic compounds using group contribution methods. Atmos. Environ. 2002, 36 (9), 1483–1498. (7) Hilal, S. H.; Karickhoff, S. W.; Carreira, L. A. Prediction of the vapor pressure boiling point, heat of vaporization and diffusion coefficient of organic compounds. QSAR Comb. Sci. 2003, 22 (6), 565–574. (8) Hilal, S. H.; Karickhoff, S. W.; Carreira, L. A. Prediction of the solubility, activity coefficient and liquid/liquid partition coefficient of organic compounds. QSAR Comb. Sci. 2004, 23 (9), 709–720. (9) Goss, K.-U.; Schwarzenbach, R. P. Linear free energy relationships used to evaluate equilibrium partitioning of organic compounds. Environ. Sci. Technol. 2001, 35 (1), 1–9. (10) Arp, H. P. H.; Schwarzenbach, R. P.; Goss, K.-U. Ambient gas/ particle partitioning. 2: The influence of particle source and temperature on sorption to dry terrestrial aerosols. Environ. Sci. Technol. 2008, 42 (16), 5951–5957. (11) Arp, H. P. H.; Schwarzenbach, R. P.; Goss, K.-U. Ambient gas/ particle partitioning. 1. Sorption mechanisms of apolar, polar, and ionizable organic compounds. Environ. Sci. Technol. 2008, 42 (15), 5541–5547. (12) Platts, J. A.; Abraham, M. H.; Butina, D.; Hersey, A. Estimation of molecular linear free energy relationship descriptors by a group contribution approach. 2. Prediction of partition coefficients. J. Chem. Inf. Comput. Sci. 2000, 40 (1), 71–80. (13) Schuurmann, G.; Ebert, R. U.; Kuehne, R. Prediction of physicochemical properties of organic compounds from 2D molecular structure - Fragment methods vs. LFER models. Chimia 2006, 60 (10), 691–698. (14) Schuurmann, G.; Ebert, R. U.; Kuhne, R. Prediction of the sorption of organic compounds into soil organic matter from molecular structure. Environ. Sci. Technol. 2006, 40 (22), 7005– 7011. (15) Eckert, F.; Klamt, A. Fast solvent screening via quantum chemistry: COSMO-RS approach. AIChE J. 2002, 48, 369.

(16) Eckert, F.; Klamt, A. COSMOtherm, C2.1; COSMOlogic: Leverkusen, Germany, 2005. (17) Niederer, C.; Goss, K.-U. Quantum chemical modeling of humic acid/air equilibrium partitioning of organic vapors. Environ. Sci. Technol. 2007, 41 (10), 3646–3652. (18) Arp, H. P. H.; Schwarzenbach, R. P.; Goss, K.-U. Determination of ambient gas-particle partitioning constants of non-polar and polar organic compounds using inverse gas chromatography. Atmos. Environ. 2008, 42 (2), 303–312. (19) Mader, B. T.; Pankow, J. F. Study of the effects of particle-phase carbon on the gas/particle partitioning of semivolatile organic compounds in the atmosphere using controlled field experiments. Environ. Sci. Technol. 2002, 36 (23), 5218–5228. (20) Ngabe, B.; Poissant, L. Polycyclic aromatic hydrocarbons in the air in the St. Lawrence Basin (Quebec). Environ. Sci. Technol. 2003, 37 (10), 2094–2099. (21) Ligocki, M. P.; Pankow, J. F. Measurements of the gas particle distributions of atmospheric organic compounds. Environ. Sci. Technol. 1989, 23 (1), 75–83. (22) Falconer, R. L.; Bidleman, T. F.; Cotham, W. E. Preferential sorption of non-ortho-polychlorinated and mono-ortho-polychlorinated biphenyls to urban aerosols. Environ. Sci. Technol. 1995, 29 (6), 1666–1673. (23) Kaupp, H.; McLachlan, M. S. Gas/particle partitioning of PCDD/ Fs, PCBs, PCNs and PAHs. Chemosphere 1999, 38 (14), 3411– 3421. (24) Correa, O.; Rifai, H.; Raun, L.; Suarez, M.; Koenig, L. Concentrations and vapor-particle partitioning of polychlorinated dibenzop-dioxins and dibenzofurans in ambient air of Houston, TX. Atmos. Environ. 2004, 38 (39), 6687–6699. (25) Kadowaki, S.; Naitoh, H. Gas-particle partitioning of PCDD/Fs in Nagoya urban air, Japan. Chemosphere 2005, 59 (10), 1439– 1453. (26) Liang, C. K.; Pankow, J. F.; Odum, J. R.; Seinfeld, J. H. Gas/ particle partitioning of semivolatile organic compounds to model inorganic, organic, and ambient smog aerosols. Environ. Sci. Technol. 1997, 31 (11), 3086–3092. (27) Cotham, W. E.; Bidleman, T. F. Laboratory investigations of the partitioning of organochlorine compounds between the gasphase and atmospheric aerosols on glass-fiber filters. Environ. Sci. Technol. 1992, 26 (3), 469–478.

(28) Chen, L. G.; Mai, B. X.; Bi, X. H.; Chen, S. J.; Wang, X. M.; Ran, Y.; Luo, X. J.; Sheng, G. Y.; Fu, J. M.; Zeng, E. Y. Concentration levels, compositional profiles, and gas-particle partitioning of polybrominated diphenyl ethers in the atmosphere of an urban city in South China. Environ. Sci. Technol. 2006, 40 (4), 1190– 1196. (29) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry, 2nd ed.; Wiley & Sons: Hoboken, NJ, 2002. (30) Kalberer, M.; Paulsen, D.; Sax, M.; Steinbacher, M.; Dommen, J.; Prevot, A. S. H.; Fisseha, R.; Weingartner, E.; Frankevich, V.; Zenobi, R.; Baltensperger, U. Identification of polymers as major components of atmospheric organic aerosols. Science 2004, 303 (5664), 1659–1662. (31) Hueglin, C.; Gehrig, R.; Baltensperger, U.; Gysel, M.; Monn, C.; Vonmont, H. Chemical characterisation of PM2.5, PM10 and coarse particles at urban, near-city and rural sites in Switzerland. Atmos. Environ. 2005, 39 (4), 637–651. (32) Krivacsy, Z.; Gelencser, A.; Kiss, G.; Meszaros, E.; Molnar, A.; Hoffer, A.; Meszaros, T.; Sarvari, Z.; Temesi, D.; Varga, B.; Baltensperger, U.; Nyeki, S.; Weingartner, E. Study on the chemical character of water soluble organic compounds in fine atmospheric aerosol at the Jungfraujoch. J. Atmos. Chem. 2001, 39 (3), 235–259. (33) Gysel, M.; Weingartner, E.; Nyeki, S.; Paulsen, D.; Baltensperger, U.; Galambos, I.; Kiss, G. Hygroscopic properties of water-soluble matter and humic-like organics in atmospheric fine aerosol. Atmos. Chem. Phys. 2004, 4, 35–50. (34) Goss, K.-U.; Arp, H. P. H.; Bronner, G.; Niederer, C. Partition behavior of hexachlorocyclohexane-isomers. J. Chem. Eng. Data 2008, 53 (3), 750–754. (35) Tittlemier, S. A.; Halldorson, T.; Stern, G. A.; Tomy, G. T. Vapor pressures, aqueous solubilities, and Henry’s law constants of some brominated flame retardants. Environ. Toxicol. Chem. 2002, 21 (9), 1804–1810. (36) Mader, B. T.; Pankow, J. F. Vapor pressures of the polychlorinated dibenzodioxins (PCDDs) and the polychlorinated dibenzofurans (PCDFs). Atmos. Environ. 2003, 37 (22), 3103–3114.

ES8025165

VOL. 43, NO. 6, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

1929