Article pubs.acs.org/est

Experimental Determination of Polyparameter Linear Free Energy Relationship (pp-LFER) Substance Descriptors for Pesticides and Other Contaminants: New Measurements and Recommendations Angelika Stenzel,† Kai-Uwe Goss,†,‡ and Satoshi Endo*,† †

Helmholtz Centre for Environmental Research UFZ, Permoserstrasse 15, D-04318 Leipzig, Germany Institute of Chemistry, University of Halle Wittenberg, Kurt Mothes Strasse 2, D-06120 Halle, Germany

‡

S Supporting Information *

ABSTRACT: Well-calibrated polyparameter linear free energy relationships (pp-LFERs) are an accurate way to predict partition coefficients (K) for neutral organic chemicals. In this work, pp-LFER substance descriptors of 111 environmentally relevant substances, mainly pesticides, were determined experimentally using gas chromatographic (GC) retention times and liquid/ liquid partition coefficients. The complete set of descriptors for 50 compounds are being reported here for the first time. Validation of the measured substance descriptors was done by comparing predicted and experimental log K for the systems octanol/water (Kow), water/air (Kwa), and organic carbon/water (Koc), all of which indicated a high reliability of ppLFER predictions based on the determined descriptors (e.g., a root mean squared error of 0.39 for log Kow). The descriptors presented in this work in combination with existing pp-LFER system equations substantially extend (and in some cases correct) our knowledge on partition properties of these 111 chemicals. In addition, the results of this work provide insight on some general guidelines with respect to the method combination best suited for deriving descriptors for environmentally relevant compounds.

■

INTRODUCTION An increasingly large variety of pesticides are used by agriculture and industry. Applications of pesticides include protecting edible and commercial plants from weeds, insects, and pathogens; reducing disease transmitting animals such as mosquitoes, fleas, and rats; and preserving food through inhibiting the growth of bacteria and fungi. According to the United States Environmental Protection Agency, the world usage of pesticides amounted to approximately 2.4 million tons in 2007.1 Pesticides are biologically active compounds, and thus they are subject to a high level of regulations to protect nontarget organisms including humans. For instance, the European Commission recently decided to restrict the use of three bee-harming pesticides.2 Moreover, 15 out of the 22 substances currently listed under the Stockholm Convention on Persistent Organic Pollutants are pesticides.3,4 Effective regulations require the understanding of the environmental distribution and fate of pesticides. Therefore, a sound estimation method of partition coefficients (K) between environmental phases is essential. This is not an easy task, because many pesticides may contain several functional groups that may be very different in their chemical structure and may exhibit intramolecular interactions. These features may make pesticides unique to the compounds used to calibrate existing models to predict K, resulting in poor model performance.5 Simple correlations with the log of the octanol/water partition © 2013 American Chemical Society

coefficient (Kow) do not provide accurate estimations when diverse pesticides are considered.6,7 Moreover, available experimental log K for the same compound can disagree widely8 (e.g., up to 3.3 log units for Kow, up to 5.5 log units for water/air partition coefficients (Kwa) for the compounds considered in this study) and it is generally difficult to assess the accuracy of literature data. Instead of (or in addition to) experimental data, risk assessment often considers estimated log K. Frequently used estimation models are mostly based on group contribution methods. Such methods can predict accurate values if all molecular fragment values and correction factors necessary to describe the given chemical and partitioning system are available. Kow of many (but not all) chemicals may be predicted by such an approach, reflecting the abundance of data that can be used to calibrate the required fragment values. However, the model applicability is much more limited for less probed systems such as water/air and organic carbon/water partition coefficients (Koc). Furthermore, it is practically not possible to construct meaningful group contribution methods for other environmentally relevant Received: Revised: Accepted: Published: 14204

September 17, 2013 November 14, 2013 November 18, 2013 November 18, 2013 dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

The goals of this work are to extend the pp-LFER substance descriptor set for complex and multifunctional compounds and to improve the prediction accuracy of already available descriptors for different partition coefficients. In addition to pesticides, we determined substance descriptors for polychlorinated biphenyls (PCBs) and heterocyclic aromatic as well as nitroaromatic compounds. The selection of these chemicals is based on their environmental importance as well as structural diversity. We then compare the descriptors determined in this work and in the literature and discuss general schemes of how the descriptors for environmentally relevant chemicals should best be determined. Finally, suggestions for the best combination of experimental systems suited for descriptor determination are presented, offering high sensitivity (large system descriptors) as well as experimental ease, robustness, and accuracy.

partition coefficients such as lipid/water, protein/water, and aerosol/air partition coefficients because of limited data. The poly parameter linear free energy relationship (ppLFER) model (eq 1) provides an alternative approach to predict log K for simple as well as complex, neutral substances:9 log K = sS + aA + bB + vV + lL + c

(1)

The capital letters S, A, B, V, and L characterize the properties of the substance, whereas the small letters define the complementary properties of the partitioning system (e.g., octanol/water). L is the logarithm of the hexadecane/air partition coefficient, A defines the H-bond donor properties, B describes the H-bond acceptor properties, S represents additional specific interactions and is referred to as the dipolarity/polarizability parameter, and V is the characteristic volume.10 The pp-LFER is known to provide accurate fit for log K of diverse neutral organic substances (e.g., standard deviations (sd) for log K is 0.08−0.15 (water/air)11 and 0.23−0.24 (plant/air, plant/water)).12 The pp-LFER system parameters are published for a multitude of environmentally relevant systems, such as water/air, soil organic carbon/water, storage lipid/water, and aerosol/air.6,9,13−16 Note that other forms of pp-LFER equations also exist in which the E descriptor, the excess molar refraction, replaces the L or V descriptor of eq 1.9,17 We do not use the pp-LFER equations that involve E in this work for reasons stated in the Discussion section. The limiting factor to apply pp-LFER equations for pesticides is usually the availability of substance descriptors. Although descriptors for some pesticides have already been determined,18−21 their number is still small compared to the number of pesticides used now and in the past. The largest sources of descriptors for pesticides are the studies done by Tülp et al.18 and Bronner et al.19 They determined descriptors for 61−76 pesticides using slightly different experimental data sets (see below for details). The reported prediction accuracy of the available pesticide descriptors is however relatively low (e.g., root mean squared error (rmse) for log Kow is 0.72 in Tülp et al.18 and 0.84 in Bronner et al.)19 compared to the typical performance of pp-LFER predictions (the sd of estimates is typically 0.1−0.2 for partitioning to homogeneous solvents).9,22 The high rmse values may partially be due to inaccurate experimental log Kow from the literature. However, the low accuracy of the descriptors could also be a reason for the high rmse. The pp-LFER substance descriptors can be derived from a set of experimental partition coefficients as well as chromatographic retention times (the logarithmic net retention time (log tnet) simply replaces log K in eq 1) in systems whose system parameters are known. The accuracy of the determined descriptors thus depends on the types of partitioning systems and the quality of the experimental data used for the calibration. The data used by Tülp et al.18 and Bronner et al.19 are relatively limited, particularly with respect to accurate determination of S, A, and B (see the Results and Discussion section for details). Therefore, these descriptors could still be improved through the integration of other methods and systems in the descriptor determination. S, A, and B descriptors describe polar interactions (i.e., specific interactions) and thus are particularly important for modern pesticides. Polar interactions have tremendous influences on air−water partitioning and solubility in water, the most relevant environmental phase.

■

MATERIALS AND METHODS Determination of Substance Descriptors. The method for descriptor determination is identical to the one used previously by us to determine the substance descriptors for flame retardants.23 Briefly, for each chemical, gas chromatographic (GC) retention times were measured on the columns SPB Octyl and DB-200 as well as the partition coefficients in three liquid/condensed phase systems: heptane/propylene carbonate; ethylene glycol/1,2-dichloroethane; poly(dimethyl siloxane) (PDMS)/water. A GC coupled to a mass spectrometer was used for both retention time and partition coefficient measurements. L, A, B, and S descriptors were determined simultaneously by minimizing the rmse between experimental and pp-LFER predicted log K/log tnet using Excel Solver (V was calculated from the molecular structure). System parameters for the calibration systems mentioned above were taken from our previous work.23 Validation of the determined substance descriptors was done by checking for internal and external consistency. Internal consistency control is based on the rmse (later called “internal rmse”) between pp-LFER predicted and experimental log K (or log tnet) for the calibration systems: rmse =

∑ (log Kexp − log K ppLFER )2 n

(2)

Kexp is the partition coefficient (or tnet) measured in our experiments, KppLFER the partition coefficient (or tnet) resulting from the pp-LFER calculation for the calibration systems, and n gives the number of calibration systems (number of available log K/log tnet). External consistency was checked through the comparison of pp-LFER predicted and experimental log K collected from the literature for Kow, Kwa, and Koc measured for soil/sediment organic matter. We use these validation systems because they are virtually the only partitioning systems for which enough literature data are available to perform a thorough validation. The validated descriptors can be used to predict partition coefficients for a number of other systems including environmentally important ones. The experimental methods have been reported previously;23,24 and a short description is given in the Supporting Information (SI). The only modification of the method is the determination of log KPDMS/water < 4 using the depletion method based on the method by ter Laak et al.25 In this method, KPDMS/water is derived from the dependence of the analyte depletion in the PDMS phase on the PDMS-to-water volume ratio. The detailed 14205

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

Table 1. pp-LFER Substance Descriptors and rmse Values for the 111 Target Compounds CAS

compound

S

Aa

B

Vb

L

rmse

3424−82−6 72−55−9 50−29−3 15972−60−8 309−00−2 71626−11−4 82657−04−3 4824−78−6 2104−96−3 786−19−6 5103−71−9 470−90−6 470−90−6 1897−45−6 2921−88−2 5598−13−0 1861−32−1 60238−56−4 60238−56−4 56−72−4 1194−65−6 97−17−6 1085−98−9 62−73−7 60−57−1 87674−68−8 298−04−4 115−29−7 1031−07−8 72−20−8 66230−04−4 563−12−2 299−84−3 122−14−5 39515−41−8 55−38−9 62924−70−3 85509−19−9 944−22−9 1024−57−3 23560−59−0 118−74−1 121−75−5 67129−08−2 51218−45−2 15299−99−7 1836−75−5 26530−20−1 34622−58−7 56−38−2 1825−21−4 52645−53−1 52645−53−1 298−02−2 2310−17−0 32809−16−8 2312−35−8 13457−18−6 3689−24−5 117−18−0 13071−79−9

2,4′-DDE 4,4′-DDE 4,4′-DDT alachlor aldrin benalaxyl bifenthrin bromophos-ethyl bromophos-methyl carbophenothion cis-chlordane chlorfenvinfos 1c chlorfenvinfos 2c chlorothalonil chlorpyrifos-ethyl chlorpyrifos-methyl chlorthal-dimethyl chlorthiophos 1c chlorthiophos 2c coumaphos dichlobenil dichlofenthion dichlofluanid dichlorvos dieldrin dimethenamid disulfoton endosulfan endosulfan-sulfate endrin esfenvalerate ethion fenchlorphos fenitrothion fenpropathrin fenthion flumetralin flusilazole fonofos heptachlor epoxide heptenophos hexachlorobenzene malathion metazachlor metolachlor napropamide nitrofen octhilinone orbencarb parathion-ethyl pentachloroanisole permethrin 1c permethrin 2c phorate phosalone procymidone propargite pyrazophos sulfotep tecnazene terbufos

1.38 1.36 1.78 1.60 1.03 1.88 1.36 1.26 1.51 1.62 1.49 1.56 1.70 2.08 1.36 1.50 1.48 1.91 1.65 2.44 1.40 1.25 2.03 1.61 1.59 1.35 1.29 1.39 1.91 1.53 2.48 1.66 1.48 2.17 1.91 1.73 1.65 2.22 1.35 1.49 1.53 0.85 1.84 1.73 1.46 1.71 1.73 1.53 1.29 1.49 0.96 1.51 1.53 1.19 2.32 1.71 1.53 2.01 1.36 1.21 1.11

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.21 0.16 0.24 1.07 0.26 1.18 0.95 0.71 0.46 0.94 0.29 0.99 0.99 0.31 0.61 0.50 0.67 0.68 0.87 1.01 0.32 0.74 0.65 0.27 0.33 1.29 0.84 0.65 0.98 0.46 0.81 1.25 0.47 0.51 0.68 0.67 0.86 0.88 0.69 0.38 0.79 0.17 1.18 1.23 1.27 1.17 0.71 1.04 0.85 0.88 0.35 0.98 0.98 0.80 0.91 0.91 1.16 1.25 0.89 0.28 0.95

2.05 2.05 2.22 2.14 2.01 2.58 3.03 2.24 1.96 2.36 2.13 2.33 2.33 1.52 2.15 1.87 1.92 2.37 2.37 2.43 1.12 2.07 2.07 1.31 2.01 2.06 2.05 2.08 2.14 2.01 3.19 2.67 1.91 1.86 2.77 1.99 2.61 2.31 1.87 1.96 1.64 1.45 2.32 2.09 2.28 2.25 1.80 1.79 1.96 2.00 1.53 2.82 2.82 1.91 2.40 1.86 2.75 2.61 2.27 1.38 2.19

9.09 9.42 9.98 8.24 8.69 10.05 10.85 9.30 8.65 10.12 9.10 8.91 8.90 7.58 8.64 8.03 8.58 9.59 9.85 11.36 5.63 8.25 8.22 4.84 9.23 8.27 7.84 9.22 10.10 9.48 12.60 9.99 8.20 7.98 10.80 8.35 9.65 9.19 7.75 8.70 6.58 7.64 8.32 8.81 8.75 9.31 9.40 7.87 8.42 8.67 7.63 11.67 11.76 7.41 10.64 9.00 10.54 11.10 7.30 6.95 7.92

0.06 0.06 0.11 0.06 0.01 0.07 0.08 0.00 0.16 0.23 0.10 0.10 0.13 0.08 0.04 0.05 0.06 0.11 0.13 0.01 0.07 0.14 0.14 0.10 0.03 0.04 0.13 0.01 0.04 0.00 0.16 0.05 0.05 0.10 0.07 0.23 0.17 0.13 0.21 0.08 0.00 0.04 0.09 0.02 0.00 0.02 0.07 0.07 0.10 0.11 0.05 0.05 0.04 0.18 0.06 0.12 0.05 0.02 0.10 0.05 0.04

14206

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

Table 1. continued CAS

compound

S

Aa

B

Vb

L

rmse

640−15−3 731−27−1 50471−44−8 2051−60−7 2051−61−8 2051−62−9 13029−08−8 2050−68−2 7012−37−5 15862−07−4 35693−99−3 37680−73−2 35065−27−1 486−25−9 119−65−3 91−22−5 92−83−1 86−57−7 121−14−2 91−23−6 555−03−3 100−17−4 602−60−8 81−15−2 57−83−0 53−19−0 834−12−8 1912−24−9 314−40−9 18181−80−1 63−25−2 1563−66−2 510−15−6 101−21−3 99−30−9 122−39−4 319−84−6 319−85−7 58−89−9 319−86−8 21087−64−9 40487−42−1 527−20−8 7287−19−6 122−42−9 23950−58−5 886−50−0 86−74−8 120−72−9 3380−34−5

thiometon tolylfluanid vinclozolin PCB 1 PCB 2 PCB 3 PCB 4 PCB 15 PCB 28 PCB 29 PCB 52 PCB 101 PCB 153 9-fluorenone isoquinoline quinoline xanthene 1-nitronaphthalene 2,4-dinitrotoluene 2-nitroanisole 3-nitroanisole 4-nitroanisole 9-nitroanthracene musk xylene progesterone 2,4′-DDD ametryn atrazine bromacil bromopropylate carbaryl carbofuran chlorobenzilate chlorpropham dicloran diphenylamine α-HCH β-HCH γ-HCH (lindane) δ-HCH metribuzin pendimethalin pentachloroaniline prometryn propham propyzamide terbutryn carbazole indole triclosan

1.42 1.77 1.78 1.07 1.10 1.08 1.19 1.24 1.17 1.08 1.23 1.19 1.17 1.55 1.25 1.11 1.17 1.49 1.84 1.57 1.22 1.47 1.73 1.47 1.97 1.61 1.02 0.97 1.41 1.83 1.54 1.37 1.00 1.25 1.66 0.96 1.31 1.52 1.35 1.61 0.71 1.39 1.12 1.11 1.01 1.61 1.20 1.91 1.26 1.11

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.14 0.33 0.49 0.40 0.00 0.00 0.11 0.65 0.29 0.27 0.61 0.15 0.40 0.17 0.27 0.65 0.16 0.23 0.33 0.27 0.24 0.30 0.44 0.40 0.42

0.78 0.90 0.80 0.23 0.15 0.33 0.12 0.13 0.09 0.13 0.13 0.15 0.24 0.37 0.36 0.54 0.37 0.46 0.56 0.61 0.48 0.46 0.53 0.65 1.51 0.19 1.16 1.05 0.92 1.00 0.93 1.04 0.82 0.52 0.36 0.20 0.50 0.51 0.47 0.44 1.35 0.71 0.40 0.87 0.68 0.89 0.83 0.15 0.32 0.63

1.77 2.21 1.84 1.45 1.45 1.45 1.57 1.57 1.69 1.69 1.81 1.94 2.06 1.37 1.04 1.04 1.42 1.26 1.21 1.09 1.09 1.09 1.63 2.08 2.62 2.10 1.80 1.62 1.63 2.51 1.54 1.69 2.27 1.58 1.24 1.42 1.58 1.58 1.58 1.58 1.62 2.15 1.43 1.94 1.45 1.84 1.94 1.32 0.95 1.81

7.32 8.86 8.11 6.43 6.81 6.87 6.90 7.75 8.14 8.07 8.38 9.21 10.11 7.41 5.31 5.27 7.16 6.79 6.27 5.41 5.59 5.74 8.98 8.25 12.47 9.35 8.36 7.60 8.23 10.58 8.03 7.29 9.95 6.97 7.10 6.89 7.35 7.47 7.59 7.56 8.14 9.00 7.95 8.36 6.17 7.57 8.41 7.33 5.08 9.06

0.24 0.02 0.01 0.07 0.04 0.08 0.14 0.15 0.02 0.02 0.04 0.01 0.01 0.07 0.03 0.00 0.10 0.12 0.09 0.04 0.13 0.13 0.15 0.07 0.12 0.08 0.07 0.08 0.15 0.11 0.22 0.04 0.04 0.00 0.02 0.01 0.05 0.07 0.04 0.07 0.16 0.05 0.05 0.01 0.01 0.11 0.00 0.10 0.03 0.06

Due to the nonexistence of any H-bond donor functional group, A has been set to zero for the first half of the chemicals ranging from 2,4′-DDE up to progesterone. bMolar volume in (cm3/mol)/100. cIsomers in the order of peak appearance in the GC.

a

reliability of the results. Additionaly, some other log KPDMS/water measurements were not successful because the concentration of the commercially available stock solution was too low or water solubility of the analyte was too low for a reliable measurement. In the following, we only consider compounds for which data for KPDMS/water are available, because PDMS/water is in our set the only system that is highly sensitive to B and is thus

method is explained in the SI. Some compounds that we initially included had to be removed from consideration because of difficulty in measuring log KPDMS/water. First, log KPDMS/water of very hydrophilic compounds was not measurable in our experimental setting because the concentrations in the PDMS phase were too low to measure (i.e., log KPDMS/water < 0). Moreover, we decided not to measure compounds possessing an expected log KPDMS/water > 6 due to the reduced 14207

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

absolutely necessary for accurate determination of B (see the SI for further information). Literature Data. The following databases were considered for experimental log K to be used in external validation: (a) the experimental databases stored in the EPISuite (version 4.0) for log Kow, log Kwa, and log Koc (U.S. Environmental Protection Agency’s Office of Pollution Prevention and Toxics and Syracuse Research Corporation), (b) The Pesticide Properties Database (PPDB 2.0, version: June 2013, the Agriculture & Environment Research Unit (AERU), University of Hertfordshire, UK; accessed June 2013, http://sitem.herts.ac.uk/aeru/ footprint/en/index.htm) for log Kow, log Kwa, and log Koc, c) LOGKOW© - A databank of evaluated octanol−water partition coefficients (Log P) (version 2013−06−06, provided by Sangster Research Laboratories, Montreal, Canada; accessed June 2013, http://logkow.cisti.nrc.ca/logkow/index.jsp) for log Kow. The rmse values resulting from the comparison between predictions and literature log K for each database are given in SI Table S5. For some substances, additional literature data were included such as the log Koc from Bronner and Goss,6 the log Kow from Hawker and Connell,26 or the log Kow and log Kwa from Xiao et al.27 All references and literature log K are provided in SI Table S4. If more than one literature value was available, the mean of the log values was calculated. If one log K value differed substantially (i.e., > 1 log unit) from the remaining log K values, the outlier value was removed before calculating the mean.

high internal consistency of our descriptors. To externally validate the determined substance descriptors, they were used to predict K for three partitioning systems relevant for environmental behavior of chemicals: Kow, Kwa, and Koc. The rmse values derived from the differences between predicted and mean literature log K are shown in Table 2. Table 2. Rmse Values Resulting from the Comparison between Experimental and Predicted Log K Using Substance Descriptors from This Study As Well As Previous Literature Sources log Kow

log Kwa

log Koc

substance descriptors from

rmse

n

rmse

n

rmse

n

this work (complete data set) Tülp et al.18 this work (restricted data set)a Bronner et al.19 this work (restricted data set)b other sourcesc,36−45 this work (restricted data set)d

0.39 0.48 0.41

111 29 29

1.19 1.39 0.89

99 26 26

0.71 0.43 0.59

94 29 29

0.60 0.40

30 30

1.00 0.89

26 26

0.48 0.58

30 30

0.45 0.37

36 36

0.54 0.67

32 32

0.63 0.59

28 28

a

Restriction to compound set of Tülp et al.18 bRestriction to compound set of Bronner et al.19 cCollected using UFZ-LSER database;61 descriptors from Tülp et al.18 as well as Bronner et al.19 were excluded. dRestriction to compound set from other sources36−45

■

RESULTS AND DISCUSSION Experimental Results. In total, pp-LFER substance descriptors were determined for 111 substances (isomers are always counted as single substances): 86 pesticides and pesticide transformation products, 10 PCBs, 6 heterocyclic aromatic compounds containing either N or O, 7 nitroaromatic compounds, 1 hormone, and 1 antibacterial. All 111 substances are listed in Table 1. The measured retention times on SPB Octyl and DB-200 are provided in SI Table S1. The partition coefficients for the systems heptane/propylene carbonate, ethylene glycol/1,2-dichloroethane, and PDMS/water are provided in SI Table S2. Ethylene glycol/1,2-dichloroethane partition coefficients are missing for some non-H-bond donor compounds (i.e., A = 0), but this is not a required system for these compounds because this system is only necessary for determination of A. For the calibration of descriptors, the PDMS/water partition coefficients from our own experiments were complemented by partition coefficients collected from DiFilippo and Eganhouse.28 The authors assessed the quality of the available literature values based on the experimental method used (i.e., ratio of water volume to PDMS volume). Only log KPDMS/water satisfying their criteria were taken into account. If more than one value passed, a mean value was calculated and used. The log KPDMS/water from the literature are marked in SI Table S2 (in total 28 values). Comparability of the log KPDMS/water from this study and from DiFilippo and Eganhouse28 was evaluated for those compounds for which data are available in both sources. This comparison showed a reasonable agreement of the log KPDMS/water (rmse is 0.44 for 17 substances; comparison is shown in SI Table S3). pp-LFER Substance Descriptors of the Target Substances. The determined substance descriptors and internal rmse values for each substance are shown in Table 1. The mean internal rmse is 0.07 (range 0.00−0.24) indicating an overall

The comparison between predicted and experimental literature log Kow is displayed in Figure 1. The rmse resulting

Figure 1. Comparison between experimental (x-axis) and pp-LFER predicted (y-axis) logarithmic partition coefficients for the system octanol/water (Kow). The continuous line is the 1:1 line; the two dashed lines mark a difference of ±1 log unit from the 1:1 line.

from the differences between experimental and predicted log Kow is 0.39 and the predicted values are evenly distributed around the 1:1 line (mean log Kow difference: 0.01). This agreement is regarded as high, considering the fact that the descriptors have been calibrated independently from Kow, and the experimental Kow data are from diverse sources with a varying degree of uncertainty. The predicted log Kow for only one compound shows a deviation of more than 1 log unit from the experimental value (chlorobenzilate, deviation 1.07 log 14208

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

dichlofenthion (exp. log Kwa: 1.43) has a structural analogue bromophos-ethyl (with one Br added to the phenyl ring of dichlofenthion), of which the experimental log Kwa is 3.19. A difference of 1.76 log units in Kwa just by substitution of one Br is unlikely (see, e.g., data cited in ref 23), suggesting that the given literature log Kwa for dichlofenthion could be too low. In contrast, Kwa of bromopropylate is comparable to that of chlorobenzilate, providing no clear indication of an experimental error. There are no similar compounds for the remaining compounds. The rmse obtained from comparing experimental and ppLFER predicted log Koc is 0.71 (comparison in Figure 3). This

units). Although the descriptors of chlorobenzilate are internally consistent (internal rmse 0.04), the deviations for Kwa and Koc predictions are comparably high as well (1.57 for log Kwa, 0.79 for log Koc). The experimental Kow of chlorobenzilate appears not to be extremely inaccurate, as it is lower than Kow of bromopropylate, a chlorobenzilatehomologue having two Br instead of two Cl and one more methyl group (SI Table S6), which is consistent with the expectation. By now, the descriptors of chlorobenzilate should thus be considered tentative. It should be noted that slight system-dependence of the B descriptor could contribute to the prediction accuracy for Kow. It was shown before29 that, for some compounds such as sulfoxides, anilines, and Nheterocyclic compounds, B values optimal for systems consisting of water and a phase partially miscible with water (such as octanol/water) are slightly different from B values for other solvent−water systems (including PDMS/water). This difference could explain some of the observed discrepancy between experimental and predicted Kow. The results of the comparison between experimental and ppLFER predicted log Kwa displayed in Figure 2 scatter more, with

Figure 3. Comparison between experimental (x-axis) and pp-LFER predicted (y-axis) logarithmic partition coefficients for the system organic carbon/water (Koc). The continuous line is the 1:1 line; the two dashed lines mark a difference of ±1 log unit from the 1:1 line.

rmse is higher than that for the log Kow comparison but is still considered reasonable, because organic carbon is not a uniquely defined phase like octanol or water. That is, there are different forms of organic matter contributing to the organic carbon content. Natural variability of log Koc for different soils and sediments is about ±0.3 log units sd.32 In addition, pp-LFER predictions appear to be less accurate for Koc than for welldefined organic phases because of heterogeneous sorption sites on organic matter molecules.6,33 The two compounds possessing the highest deviations are bromophos-methyl (difference between predicted and experimental log Koc is 2.78) and dichlofenthion (difference 2.02). The experimental log Koc of these chemicals (1.23 and 1.26, respectively) are much lower than the experimental log Kow values (5.05 and 5.14, respectively). Hence, experimental errors or additional interactions beyond the pp-LFER theory are possible reasons for the large errors. An additional factor that can influence prediction accuracy is the extrapolation of the pp-LFER equations. The substance descriptors of the target chemicals vary widely and can be out of the calibration domains of the pp-LFER equations used. To evaluate the influences of model extrapolation on the prediction accuracy, we calculated the leverages34,35 of our target compounds for each validation system and compared those to the prediction error (i.e., log K predicted − log K experimental). This comparison, shown in SI Figure S1, shows that most of the target chemicals are within the

Figure 2. Comparison between experimental (x-axis) and pp-LFER predicted (y-axis) logarithmic partition coefficients for the system water/air (Kwa). The continuous line is the 1:1 line; the two dashed lines mark a difference of ±1 log unit from the 1:1 line.

an rmse of 1.19. The difference between experimental and predicted log Kwa is more than 1 log unit for 30 out of the 99 compounds for which Kwa values were found in the literature. In general, there is a slight trend toward overprediction of log Kwa by the pp-LFER (the mean log Kwa difference is 0.50). As was also mentioned by Tülp et al.,18 the strong scattering is likely due to a high uncertainty in experimental data of Kwa. Often, the log Kwa are derived from water solubility and saturation vapor pressure, and the measurement of a small value of either of the properties is prone to large errors.30,31 For instance, Pontolillo and Eganhouse found that the literature values for water solubility of DDT span up to four orders of magnitude.31 In this study six compounds exhibit a deviation between 2 and 3 log units from the experimental log Kwa (coumaphos, dichlofenthion, endosulfan sulfate, ethion, procymidone, bromopropylate) and tecnazene even has a deviation of 4.76 log units. Such large errors are unlikely caused only by the descriptor values that at the same time provide reasonable predictions for log Kow. Among these chemicals, 14209

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

calibration domain of the Koc equation, whereas at least a quarter of our target chemicals are out of the calibration domains for Kow and Kwa. Nonetheless, the prediction error does not increase with increasing leverage, suggesting that the model extrapolation does not significantly influence the prediction accuracy in this case. The leverage calculation also proves that this work significantly extended the descriptor availability toward the types of chemicals that are different from typical neutral organic chemicals in terms of solvation properties. In addition to the above-mentioned validation systems, we also compared experimental log K for the system heptane/ methanol and log tnet on eight high performance liquid chromatography (HPLC) columns determined by Tülp et al.18 to the pp-LFER predicted values. The rmse values, presented in SI Table S7, are relatively small being between 0.12 and 0.50. However, the system parameters of these systems are generally small, for example, b is −1.09 for heptane/methanol whereas it is −3.45 for the octanol/water system. Small system parameters result in similar log K even when rather different substance descriptors are used. Therefore, these systems are not sensitive enough to serve as robust validation systems. We also attempted to include the retention times measured by Tülp et al.18 and Bronner et al.19 in the substance descriptor determination. This recalculation was possible for 30 of our target compounds. The recalculated descriptor values exhibit small differences compared to the original ones (SI Table S11). Prediction accuracy for these 30 compounds is improved through the recalculation for Kow (rmse (original) = 0.40, rmse (recalculated) = 0.32, n = 30), Kwa (rmse (original) = 0.89, rmse (recalculated) = 0.79, n = 26), and Koc (rmse (original) = 0.58, rmse (recalculated) = 0.47, n = 30). In the following discussions we use the original descriptors as shown in Table 1 because these descriptors are based on a consistent data set. Comparison of Descriptors from this Study and the Literature. Substance descriptors were available in the literature only for about half of our target compounds (61 out of 111). Thus, the descriptors presented in this work significantly extend the set of available pp-LFER substance descriptors especially for complex and multifunctional compounds. To compare our descriptors with previously measured ones, experimental log Kow, log Kwa, and log Koc were compared to the pp-LFER predicted values. Predictions using our descriptors were compared to predictions from descriptors determined by Tülp et al.18 (descriptors were available for 29 of our substances), Bronner et al.19 (for 30 substances) or other sources (mainly from Abraham or Poole and co-workers,36−45 for 36 substances). The resulting rmse values are displayed in Table 2. Rmse values were calculated for subsets of our substance list as well to allow a direct comparison of rmse values based on the descriptors of this study and the literature. Note that Tülp et al.18 did not report L but E. Therefore, partition coefficients were predicted using S, A, B, V, and E in the case of the Tülp descriptors. In comparison to the descriptor sets from Tülp et al.18 as well as Bronner et al.,19 the descriptors from the present study resulted in lower rmse values, that is, improvement in prediction accuracy for Kow and Kwa. This improvement can be explained by the different experimental methods used to determine the descriptors. The descriptors from Tülp et al.18 are only derived from retention times on HPLC columns. In general, the HPLC system parameters are rather small in

comparison to two-phase partitioning systems and they are further decreased by high organic contents in the mobile phase.36 In the study of Tülp et al.,18 up to 75% v/v organic solvent had to be added to the mobile phase to obtain reasonable retention times. As a result, the maximum absolute values of system parameters are 0.64 (s), 1.33 (a), and 1.59 (b) in ref 18, which are much smaller than our calibration system: 2.12 (s), 2.85 (a), and 3.84 (b). Bronner et al.19 included GC retention times to determine L. However, they derived their S, A, and B values based on the HPLC data from Tülp et al.18 without any additional measurements. Therefore, the same shortcomings apply to their calibration system. An additional difference between our predictions and those from Tülp et al.18 is the application of a slightly different ppLFER equation type (in which L is replaced through E), as explained above. As discussed already by Bronner et al.19 and Atapattu and Poole,46 the E descriptor of solid chemicals has to be estimated, and the estimated values strongly depend on the estimation tools used. Bronner et al.19 showed that the use of experimentally derived L instead of predicted E significantly improves the prediction of Kwa, and the results of our study confirm their finding. In predicting Koc, our descriptors are less accurate than those from Tülp et al.18 and Bronner et al.19 We are unable to give a clear reason why the predictions for Koc did not improve while the predictions for Kow and Kwa did improve, other than soil organic carbon being a complex medium6,32,33 (see also the Koc discussion in the previous section). This type of pp-LFER validation should work best for homogeneous phases, with consistent partitioning system descriptors. Therefore, we surmise that rmse of 0.6−0.7 log units is the best achievable accuracy for predictions of Koc using pp-LFER models even with accurately calibrated descriptors when a very diverse set of chemicals is considered. Of course, even more careful selection of Koc data used for validation could potentially reduce the rmse, which however is beyond the aim of this work. The comparison of the descriptors from this study and those from sources other than Tülp et al and Bronner et al. (i.e., refs 36−45) reveals comparable prediction accuracy for all of Kow, Kwa, and Koc (Table 2). As the calibration methods differ in different sources and are sometimes unknown, it is difficult to interpret these results. Our approach is minimalistic in terms of input data, using five selected partition systems for determination of four descriptors, while more data have usually been used in the cited literature to determine the substance descriptors. Therefore, we consider the good overall agreement between our and the literature descriptor predictions as a support for our approach to reduce the set of calibration systems to the necessary minimum. Relatively large descriptor differences were found for dieldrin (especially L, S, and B), endrin (L and B), pentachloroanisole (B), progesterone (L and S), atrazine (S and A), diphenylamine (S, B, and A), prometryn and terbutryn (both L), and carbazole (S) (SI Table S8). In addition, we want to compare our and literature descriptors for two specific chemical groups, hexachlorocyclohexanes (HCHs) and PCBs, because their physicochemical properties and substance descriptors are a controversial subject in the literature.21,37,47 For HCHs (α, β, γ, and δ isomers), descriptor sets have been reported by Abraham et al.38 and Goss et al.21 (descriptors shown in SI Table S9). The values of the descriptors from both of the studies are somewhat different from those determined in this work (i.e., our S and A values are slightly higher than the 14210

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

literature values and our B values are slightly smaller). Although our descriptors do not improve the prediction accuracy, they confirm the finding from Goss et al.21 that β-HCH is significantly different from α- and γ-HCH in terms of the Hbonding properties. The PCB descriptors determined by us were compared to descriptors from Abraham and Hussaini37 and van Noort et al.47 (descriptors in SI Table S10). Our L values increase with an increasing substituents number, whereas there is no clear trend for the S and B descriptor (A is always zero). Comparing our descriptors to the literature descriptors shows larger differences for PCBs possessing a higher number of Cl substituents. The L values determined by us are slightly higher than the values from both of the literature sources and our S values are smaller. Our B values are overall similar to the B values by Abraham and Hussaini37 and they are higher than those by van Noort et al.47 The overall prediction accuracy of our descriptors is less good than the accuracy of the literature descriptors. We tested whether our descriptors accuracy would be improved through incorporating van Noort’s V, which provides a correction for increasing ortho-chlorination. However, using van Noort’s V in combination with our experimental log K/log tnet (recalculation of L, S, and B) caused, if any, only marginal improvements in our descriptor accuracy. Experimental Determination of pp-LFER Substance Descriptors: Summary and Recommendations. Based on our experience in determining the descriptors of pesticides and flame retardants23 as well as experience from others (Poole et al.33,36 and Abraham et al.17), we propose the following discussion points and recommendations for determining ppLFER descriptors for environmentally relevant compounds. The suitability of the partitioning systems to be used for descriptor calibration can be evaluated based on three criteria: (a) sensitivity to the descriptors, (b) accuracy of the model fit, and (c) experimental applicability. Sensitivity can be quantitatively assessed with the magnitude of the system parameters (i.e., s, a, b, l), and this is directly related to the accuracy of the determined substance descriptors (as mentioned above, liquid−liquid partitioning systems are often better than HPLC systems for determining S, A, and B). Ideally, there should be one or more partitioning systems that are sensitive to each of the descriptors. High accuracy of the ppLFER model fit is another important factor for selecting partitioning systems to be used for descriptor determination. Partitioning between homogeneous phases usually gives an excellent fit, while partitioning to complex materials (like humic matter, aerosols, mineral surfaces) may not be modeled that well with pp-LFER. The latter materials are, therefore, not as suitable as the former for descriptor calibration. Experimental considerations include the possibility of being able to measure the analyte (explained more in SI SI-2), the accuracy and precision of measurements, through-put, and cost. In the following, we evaluate commonly used methods for descriptor determination, namely, GC, HPLC, and two-phase partitioning. The GC retention times can generally be measured accurately and precisely with a high through-put, and ppLFER models for many GC columns are of excellent quality (typical sd 0.01−0.04).36 Of all systems, GC offers the highest sensitivity to L. GC retention measurements on a completely nonpolar column, such as SBP Octyl19,24,48 is particularly useful to improve the accuracy of L, because they are sensitive only to L and V (the latter can be derived from the molecular

structure). Apolane 87 (Alltech) also has negligible polarity, though its temperature limit is relatively low. Commonly used poly(dimethylsiloxane) columns do possess slight polarity and thus are less suited.48 Another column type being recommendable is poly(trifluoromethylsiloxane) columns, such as DB-200 (Agilent), characterized by comparably large s values, whereas a and l are relatively small. Therefore, this column is suitable to determine S. Highly polar columns such as poly(ethylene glycol) (e.g., HP-INNOWax (Agilent)) and poly(biscyanopropylsiloxane) (e.g., SP-2340 (Supelco)) have high s and a values according to the system parameters published by Poole and Poole48 and thus may be useful for determination of S and A. However, for many environmentally relevant chemicals including pesticides and flame retardants, one cannot obtain a chromatographic peak within a reasonable experimental time because these columns strongly retain the chemicals whereas their maximal operating temperatures are relatively low. In addition, we found that either the retention times were not reproducible (INNOWax) or the calibration was not consistent (SP-2340) in our experiments at relatively high temperatures.23 The largest shortcoming has been that no conventional GC columns exhibit H-bond donor properties (i.e., b is zero) and thus incorporation of other methods is imperative. Recently, it was indicated that some ionic liquid columns exhibit a significant H-bond donor capacity and could thus be considered as an alternative option for B determination.49−56 Further research on selecting and optimizing ionic liquid columns for this purpose may be useful. HPLC retention times can also be measured repeatably and accurately. The through-put is also high, though the fitting of pp-LFER models are less precise compared to GC systems (typical sd 0.02−0.07).36 HPLC has not been used in this work because it suffers from the drawbacks mentioned above (i.e., small system parameters in combination with comparably high sd, leading to high uncertainties especially for S and A). A reversed phase column with a low content of organic modifier in the mobile phase could possibly be of use, because of relatively high b. For many environmental chemicals, however, one would have to use a very short column (shorter than commercially available columns) to obtain a peak within a reasonable experimental time. Two-phase partitioning systems are generally characterized by large partitioning system parameters for s, a, and b, though the actual values can vary substantially depending on the system. Compared to chromatographic retention time measurements, two-phase partitioning experiments generally are more labor intensive and somewhat less repeatable and accurate (typical model sd 0.1−0.5).36 Water containing systems always exhibit high b values.36 Therefore, any aqueous system is well suited for improved determination of B. However, low solubility or instability of the target chemical in water can make accurate determination of the partition coefficient difficult. In addition, aqueous partitioning systems exhibit large v values resulting in high K for large molecules which hampers experimental determination. Partitioning systems with two immiscible organic solvents, which in contrast to water offer greater solubility, more stability, and small v values, are useful alternatives.57−59 We decided on heptane/propylene carbonate (high s and a) as well as ethylene glycol/1,2dichloroethane (high a) because of their favorable system parameters. According to Qian and Poole,60 it should also be possible to use an organic system such as heptane/2,2,2trifluoroethanol or heptane/1,1,1,3,3,3-hexafluoroisopropanol 14211

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

to determine B because these systems possess high b values. Our experiments with these systems confirm this in general, but we found that for some substances, B calibrated on such fluoroalcohol systems do not accurately predict partition coefficients of water-containing systems such as Kow.23 We consider water containing systems as necessary for determining the B descriptors for environmental systems, because many important environmental partition systems contain the water phase and because it is not possible to predict a priori which compounds would exhibit system-dependent B values. To summarize, our recommendations for a minimal set of systems to determine pp-LFER substance descriptors are: GCSPB Octyl (for L), GC-DB-200 (for S), heptane/propylene carbonate (for S and A), ethylene glycol/dichloroethane (for A), and a water containing system such as PDMS/water (for B). Our current and previous studies show that substance descriptors derived from this minimal set can predict log Kow with an rmse of 0.4 in comparison to experimental data. Obviously, any additional GC, LC, or partitioning data would improve the accuracy of determined descriptors and it would help to check internal consistency, though at the cost of more experimental work. The main limiting factor in our procedure is the measurement of K PDMS/water . Although measuring KPDMS/water is more practical than any solvent/water partition coefficient, due to easy separation of the PDMS phase from water, it is still much more time-consuming and subject to a greater error than measurements for GC, LC, and fully organic solvent systems. For relatively hydrophilic compounds, reversed phase HPLC may be an alternative option to determine B. Ionic liquid columns might become an option for some compounds, though experiences are still limited. Particularly, for highly hydrophobic compounds (log KPDMS/water > 6), determination of any solvent−water and sorbent-water partition coefficients is challenging. Such hydrophobic compounds, however, are often of high environmental relevance. Therefore, finding a suitable system that is sensitive to B, reflects water-B, and is experimentally practicable for hydrophobic compounds is an important topic of future research.

■

Networking Fund through Helmholtz Interdisciplinary Graduate School for Environmental Research (HIGRADE).

■

(1) Grube, A.; Donaldson, D.; Kiely, T.; Wu, L., Pesticides Industry Sales and Usage - 2006 and 2007 Market Estimates, EPA No. 733-R-11001; United States Environmental Protection Agency: Washington, DC, 2011. (2) Commission Implementing Regulation (EU) No 485/2013; European Commission: Brussels, Belgium, 2013. (3) Stockholm Convention on Persistent Organic Pollutants (POPs) as amended in 2009 (Reference: C.N.524.2009.TREATIES-4), Stockholm Convention, 2009. (4) An amendment to Annex A adopted by the Conference of the Parties to the Stockholm Convention on Persistent Organic Pollutants at its fifth meeting (Reference: C.N.703.2011.TREATIES-8), Stockholm Convention, 2011. (5) Goss, K. U.; Arp, H. P. H.; Bronner, G.; Niederer, C. Nonadditive effects in the partitioning behavior of various aliphatic and aromatic molecules. Environ. Toxicol. Chem. 2009, 28, 52−60. (6) Bronner, G.; Goss, K. U. Predicting sorption of pesticides and other multifunctional organic chemicals to soil organic carbon. Environ. Sci. Technol. 2011, 45, 1313−1319. (7) Harris, A.; Dangerfield, N.; Woudneh, M.; Brown, T.; Verrin, S.; Ross, P. Partitioning of current-use and legacy pesticides in salmon habitat in British Columbia, Canada. Environ. Toxicol. Chem. 2008, 27, 2253−2262. (8) Benfenati, E.; Gini, G.; Piclin, N.; Roncaglioni, A.; Vari, M. R. Predicting log P of pesticides using different software. Chemosphere 2003, 53, 1155−1164. (9) Goss, K.-U. Predicting the equilibrium partitioning of organic compounds using just one linear solvation energy relationship (LSER). Fluid Phase Equilib. 2005, 233, 19−22. (10) Abraham, M. H.; McGowan, J. C. The use of characteristic volumes to measure cavity terms in reversed phase liquid chromatography. Chromatographia 1987, 23, 243−246. (11) 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, 8, 1777− 1791. (12) Platts, J. A.; Abraham, M. H. Partition of volatile organic compounds from air and from water into plant cuticular matrix: An LFER analysis. Environ. Sci. Technol. 2000, 34, 318−323. (13) 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, 9519− 9524. (14) Endo, S.; Escher, B.; Goss, K. U. Capacities of membrane lipids to accumulate neutral organic chemicals. Environ. Sci. Technol. 2011, 45, 5912−5921. (15) Endo, S.; Bauerfeind, J.; Goss, K. U. Partitioning of neutral organic compounds to structural proteins. Environ. Sci. Technol. 2012, 46, 12697−12703. (16) Arp, H. P. H.; Schwarzenbach, R. P.; Goss, K.-U. Ambient gas/ particle partitioning II: The influence of particle source and temperature on sorption to dry terrestrial aerosols. Environ. Sci. Technol. 2008, 42, 5951−5957. (17) Abraham, M. H.; Ibrahim, A.; Zissimos, A. M. Determination of sets of solute descriptors from chromatographic measurements. J. Chromatogr., A 2004, 1037 (1−2), 29−47. (18) Tülp, H. C.; Goss, K.-U.; Schwarzenbach, R. P.; Fenner, K. Experimental determination of LSER parameters for a set of 76 diverse pesticides and pharmaceuticals. Environ. Sci. Technol. 2008, 42, 2034− 2040. (19) Bronner, G.; Fenner, K.; Goss, K. U. Hexadecane/air partitioning coefficients of multifunctional compounds: Experimental data and modeling. Fluid Phase Equilib. 2010, 299, 207−215.

ASSOCIATED CONTENT

S Supporting Information *

Experimental method descriptions, further information on the influence of the PDMS/water system on B, additional details for the method comparison, experimentally determined retention times and partition coefficients for the target compounds, literature partition coefficients used for validation, a comparison between our and literature descriptors, and recalculated descriptors can be found in the Supporting Information. This material is available free of charge via the Internet at http://pubs.acs.org.

■

REFERENCES

AUTHOR INFORMATION

Corresponding Author

*(S.E.) Phone: +49 341 235 1818; e-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We thank Andrea Pfennigsdorff for her assistance in the lab and Hans Peter Arp for his useful comments on the manuscript. This work was supported by Helmholtz Impulse and 14212

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

(20) Green, C. E.; Abraham, M. H.; Acree, W. E.; De Fina, K. M.; Sharp, T. L. Solvation descriptors for pesticides from the solubility of solids: Diuron as an example. Pestic. Manag. Sci. 2000, 56, 1043−1053. (21) Goss, K.-U.; Arp, H. P.; Bronner, G.; Niederer, C. Partition behavior of hexachlorocyclohexane isomers. J. Chem. Eng. Data 2008, 53, 750−754. (22) Abraham, M. H.; Gola, J. M. R.; Cometto-Muniz, J.; Cain, W. S. Solvation properties of refrigerants, and the estimation of their watersolvent and gas-solvent partitions. Fluid Phase Equilib. 2001, 180, 41− 58. (23) Stenzel, A.; Goss, K. U.; Endo, S. Determination of polyparameter linear free energy relationship (pp-LFER) substance descriptors for established and alternative flame retardants. Environ. Sci. Technol. 2013, 47, 1399−1406. (24) Stenzel, A.; Endo, S.; Goss, K. U. Measurements and predictions of hexadecane/air partition coefficients for 387 environmentally relevant compounds. J. Chromatogr., A 2012, 1220, 132−142. (25) ter Laak, T. L.; Durjava, M.; Struijs, J.; Hermens, J. Solid phase dosing and sampling technique to determine partition coefficients of hydrophobic chemicals in complex matrixes. Environ. Sci. Technol. 2005, 39, 3736−3742. (26) Hawker, D. W.; Connell, D. W. Octanol-water partition coefficients of polychlorinated biphenyl congeners. Environ. Sci. Technol. 1988, 22, 382−387. (27) Xiao, H.; Li, N. Q.; Wania, F. Compilation, evaluation, and selection of physical-chemical property data for alpha-, beta-, and gamma- hexachlorocyclohexane. J. Chem. Eng. Data 2004, 49, 173− 185. (28) DiFilippo, E. L.; Eganhouse, R. P. Assessment of PDMS-water partition coefficients: Implications for passive environmental sampling of hydrophobic compounds. Environ. Sci. Technol. 2010, 44, 6917− 6925. (29) Abraham, M. H. Hydrogen Bonding. 31. Construction of a scale of solute effective or summation hydrogen-bond basicity. J. Phys. Org. Chem. 1993, 6, 660−684. (30) Delle Site, A. The vapor pressure of environmentally significant organic chemicals: A review of methods and data at ambient temperatures. J. Phys. Chem. Ref. Data 1997, 26, 157−193. (31) Pontolillo, J.; Eganhouse, R. P., The Search for Reliable Aqueous Solubility (Sw) and Octanol-Water Partition Coefficient (Kow) Data for Hydrophobic Organic Compounds: DDT and DDE as a Case Study, Water-Resources Investigations Report 01-4201; U.S. Department of the Interior, U.S. Geological Survey: Reston, Virginia, 2001. (32) Schwarzenbach, R. P.; Gschwend, P. M.; Imboden, D. M. Environmental Organic Chemistry, 2nd ed.; John Wiley & Sons: Hoboken, 2003. (33) 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., A 2013, 1317, 85−104. (34) Jouan-Rimbaud, D.; Bouveresse, E.; Massart, D. L.; de Noord, O. E. Detection of prediction outliers and inliers in multivariate calibration. Anal. Chim. Acta 1999, 388, 283−301. (35) Harrell, F. E. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, And Survival Analysis; Springer Verlag: New York, 2001. (36) Poole, C. F.; Atapattu, S. N.; Poole, S. K.; Bell, A. K. Determination of solute descriptors by chromatographic methods. Anal. Chim. Acta 2009, 652 (1−2), 32−53. (37) Abraham, M. H.; Al-Hussaini, J. M. Solvation parameters for the 209 PCBs: Calculation of physicochemical properties. J. Environ. Monit. 2005, 7, 295−301. (38) Abraham, M. H.; Enomoto, K.; Clarke, E. D.; Roses, M.; Rafols, C.; Fuguet, E. Henrỳs law constants or air to water coefficients for 1,3,5-triazines by an LFER method. J. Environ. Monit. 2007, 9, 234− 239. (39) Sprunger, L. M.; Achi, S. S.; Acree, W. E.; Abraham, M. H. Development of correlations for describing solute transfer into acyclic

alcohol solvents based on the Abraham model and fragment-specific equation coefficients. Fluid Phase Equilib. 2010, 288, 139−144. (40) Abraham, M. H.; Ibrahim, A.; Acree, W. E. Partition of compounds from gas to water and from gas to physiological saline at 310 K: Linear free energy relationships. Fluid Phase Equilib. 2007, 251 (2), 93−109. (41) Karunasekara, T.; Poole, C. F. Compounds for expanding the descriptor space for characterizing separation systems. J. Chromatogr., A 2012, 1266, 124−130. (42) Abraham, M. H. Hydrogen Bonding XXVII. Solvation parameters for functionally substituted aromatic compounds and heterocyclic compounds, from gas-liquid chromatographic data. J. Chromatogr. 1993, 644, 95−139. (43) Abraham, M. H.; Acree, W. E. Characterisation of the waterisopropyl myristate system. Int. J. Pharm. 2005, 294 (1−2), 121−128. (44) Abraham, M. H.; Sanchez-Moreno, R.; Gil-Lostes, J.; Acree, W. E.; Cometto-Muniz, J.; Cain, W. S. The biological and toxicological activity of gases and vapors. Toxicol. In Vitro 2010, 24, 357−362. (45) Houser, E. J.; Mlsna, T. E.; Nguyen, V. K.; Chung, R.; Mowery, R. L.; McGill, R. A. Rational materials design of sorbent coatings for explosives: Applications with chemical sensors. Talanta 2001, 54 (3), 469−485. (46) Atapattu, S. N.; Poole, C. F. Solute descriptors for characterizing retention properties of open-tubular columns of different selectivity in gas chromatography at intermediate temperatures. J. Chromatogr., A 2008, 1195 (1−2), 136−145. (47) 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. (48) Poole, C. F.; Poole, S. K. Separation characteristics of wallcoated open-tubular columns for gas chromatography. J. Chromatogr., A 2008, 1184 (1−2), 254−280. (49) Anderson, J. L.; Ding, J.; Welton, T.; Armstrong, D. W. Characterizing ionic liquids on the basis of multiple solvation interactions. J. Am. Chem. Soc. 2002, 124, 14247−14254. (50) Poole, S. K.; Poole, C. F. Chemical interactions as a possible limitation on the useful solvent properties of liquid alkylammonium salts. J. Chromatogr. 1988, 435, 17−28. (51) Zhao, Q. C.; Eichhorn, J.; Pitner, W. R.; Anderson, J. L. Using the solvation parameter model to characterize functionalized ionic liquids containing the tris(pentafluoroethyl)trifluorophosphate (FAP) anion. Anal. Bioanal. Chem. 2009, 395, 225−234. (52) Anderson, J. L.; Armstrong, D. W. Immobilized ionic liquids as high-selectivity/high-temperature/high-stability gas chromatography stationary phases. Anal. Chem. 2005, 77, 6453−6462. (53) Anderson, J. L.; Ding, J.; Ellern, A.; Armstrong, D. W. Structure and properties of high stability geminal dicationic ionic liquids. J. Am. Chem. Soc. 2005, 127, 593−604. (54) Huang, K.; Han, X.; Zhang, X.; Armstrong, D. W. PEG-linked geminal dicationic ionic liquids as selective, high-stability gas chromatographic stationary phases. Anal. Bioanal. Chem. 2007, 389, 2265−2275. (55) Payagala, T.; Zhang, Y.; Wanigasekara, E.; Huang, K.; Breitbach, Z. S.; Sharma, P. S.; Sidisky, L. M.; Armstrong, D. W. Trigonal tricationic ionic liquids: A generation of gas chromatographic stationary phases. Anal. Chem. 2009, 81, 160−173. (56) Zhao, Q.; Anderson, J. L. Highly selective GC stationary phases consisting of binary mixtures of polymeric ionic liquids. J. Sep. Sci. 2010, 33, 79−87. (57) Karunasekara, T.; Poole, C. F. Models for liquid-liquid partition in the system ethylene glycol-organic solvent and their use for estimating descriptors for organic compounds. Chromatographia. 2011, 73, 941−951. (58) Karunasekara, T.; Poole, C. F. Models for liquid-liquid partition in the system propylene carbonate-organic solvent and their use for estimating descriptors for organic compounds. J. Chromatogr., A 2011, 1218, 809−816. 14213

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214

Environmental Science & Technology

Article

(59) Karunasekara, T.; Poole, C. F. Model for the partition of neutral compounds between n-heptane and formamide. J. Sep. Sci. 2010, 33, 1167−1173. (60) Qian, J.; Poole, C. F. Distribution of neutral organic compounds between n-heptane and fluorine-containing alcohols. J. Chromatogr., A 2007, 1143, 276−283. (61) UFZ-LSER database. Version: 1.0, June 2013. http://www.ufz. de/index.php?en=31698

14214

dx.doi.org/10.1021/es404150e | Environ. Sci. Technol. 2013, 47, 14204−14214