Capacities of Membrane Lipids to Accumulate Neutral Organic

CRITICAL REVIEW pubs.acs.org/est

Capacities of Membrane Lipids to Accumulate Neutral Organic Chemicals Satoshi Endo,*,† Beate I. Escher,‡ and Kai-Uwe Goss†,§ †

Department of Analytical Environmental Chemistry, UFZ
Helmholtz Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany ‡ National Research Centre for Environmental Toxicology (Entox), The University of Queensland, 39 Kessels Road, Brisbane, QLD 4108, Australia § Institute of Chemistry, University of Halle-Wittenberg, Kurt-Mothes-Strasse 2, D-06120 Halle, Germany

bS Supporting Information ABSTRACT: Lipids have been considered as the predominant components for bioaccumulation of organic chemicals. However, differences in accumulation properties between different types of lipid (e.g., storage and membrane lipids) have rarely been considered. Moreover, in view of toxic effects on organisms, chemical accumulation specifically in biological membranes is of particular importance. In this review article, partition coefficients of 240 neutral organic compounds between liposomes (phospholipid membrane vesicles) and water (Klipw), reported in the literature or measured additionally for this work, were evaluated. Values of log Klipw and log Kow (octanol
water partition coefficients) differ by 0.4 on average. Polyparameter linear free energy relationships (PP-LFERs) can describe the log Klipw data even better (standard deviations = 0.28
0.31) than the log Kow model. Recent experimental data for highly hydrophobic compounds fit well to the PP-LFERs and do not indicate the existence of a previously postulated “hydrophobicity cutoff”. Predictive approaches based only on the molecular structure (KOWWIN, SPARC, COSMOthermX, COSMOmic) were also evaluated for Klipw prediction. The PP-LFERs revealed that partition coefficients into membrane lipids can be two log units higher than those into storage lipids for H-bond donor compounds, suggesting that distinguishing between the two lipids is necessary to account for the bioaccumulation of these compounds, and that tissues rich in membrane lipids (e.g., kidneys, liver) instead of fat tissue can be the primary phase for accumulation.

’ INTRODUCTION Lipids are generally regarded as the predominant components for accumulation of organic chemicals in biota. Accordingly, measured chemical concentrations in an organism are normalized to the total lipid content of the organism to evaluate the extent of bioaccumulation. Similarly, concentration ratios of chemicals in organisms to various environmental phases (i.e., bioconcentration, bioaccumulation, biota-sediment accumulation factors) are calculated using the concentrations in organisms that are normalized to total lipid content. In these calculations, differences between lipids are not considered, and the accumulation capacity of “the lipid” is often assumed to be identical to that of n-octanol. This simplified view to the bioaccumulation of chemicals is frequently justified for convenience or for the reason of lack of detailed knowledge on accumulation capacities of individual biocomponents. In the past decades, however, a considerable number of studies have been devoted to the characterization of different lipids, in particular, membrane lipids,1
3 with regard to their interactions with chemicals. In this review article, we focus on the equilibrium partitioning behavior of neutral organic compounds into membrane lipids and evaluate the experimental data and estimation models for the respective partition coefficients. Note r 2011 American Chemical Society

that this article does not deal with ionic organic compounds, for which behavior different from neutral species is generally expected.4 Lipids existing abundantly in organisms are storage and membrane lipids. Storage lipids refer to triacylglycerides (esters of glycerol and three fatty acids). Storage lipids are the predominant component of fat tissue. For example, the nonaqueous fraction of the cells of human fat tissue consists of 95 vol % lipids, of which >99 vol % are storage lipids.5 Membrane lipids are the main components of biological membranes. The molecules typically contain two nonpolar acyl chains and one polar headgroup. Phospholipids, having substituted phosphate as the headgroup, are usually the major components in biological membranes.6 Among human tissues, kidneys, liver, and brain are relatively rich in membrane lipids (60
75 vol % of the total lipid5). Despite the obvious structural differences between storage and membrane lipids, their differences in accumulation properties have not been systematically addressed. Received: March 15, 2011 Accepted: June 14, 2011 Revised: June 9, 2011 Published: June 14, 2011 5912

dx.doi.org/10.1021/es200855w | Environ. Sci. Technol. 2011, 45, 5912–5921

Environmental Science & Technology In pharmacology, it is becoming common to consider individual biocomponents (e.g., storage lipid, membrane lipid, protein, water) to have different partitioning capacities in order to better simulate the body distribution of administered drugs.5,7,8 A similar concept was proposed in the environmental field, too.9,10 Such descriptive modeling would be beneficial to both estimation and mechanistic understanding of chemicals’ accumulation in organisms. However, limited availability of the required parameters, such as the partition coefficients into each biocomponent, often restricts the application of such models to a narrow range of compounds. For the investigation on chemical partitioning into membranes, liposomes (artificial membrane vesicles made up of phospholipids) have been extensively used as model phases. Phospholipid liposomes have been shown to mimic accumulation in biological membranes well. Thus, the liposome
water partition coefficient (Klipw) was suggested to be a more accurate descriptor to estimate membrane affinity of chemicals and membrane-related processes such as bioconcentration in aquatic organisms11,12 and baseline toxicity (refs 13, 14, and 15 (p 376)) than the octanol
water partition coefficient (Kow). Contradicting this fact, Kow is still the sole parameter used in most bioaccumulation models. One reason is the far larger availability of experimental data and estimation methods for Kow compared to Klipw. It is thus sensible to assemble a set of experimental data and establish estimation methods for Klipw. In the literature, it has been reported that for hydrophobic compounds a cutoff log Kow value (or a cutoff molecular size) exists above which log Klipw does not increase any more or even starts to decrease with increasing log Kow (or molecular size).16,17 A proposed explanation for these observations is that a bulky solute molecule needs a relatively large free energy for cavity formation in the structured membrane phase, rendering the relationship between the molecular size and the cavity formation energy nonlinear in the membrane, whereas the size
energy relationship is linear in octanol.16 A recent study,11 however, demonstrated that the reported cutoff phenomena can largely be explained by experimental artifacts such as nonequilibrium and overestimation of the aqueous phase concentration. This argument points toward the necessity of a careful and systematic reevaluation on reported experimental Klipw values. The goals of this review are 4-fold: (I) experimental Klipw values for diverse types of neutral organic compounds reported in the literature and measured in this study are evaluated, followed by a literature review on the factors that influence the values of Klipw; (II) estimation models are calibrated and evaluated for accuracy in calculating Klipw values and are used to find data that are substantially inconsistent with the other data; (III) partition coefficients into membrane lipids are compared to those of storage lipids using the calibrated models; and finally, (IV) implications for the properties of membranes and the bioaccumulation behavior of organic chemicals are discussed.

’ COLLECTION AND EVALUATION OF EXPERIMENTAL Klipw DATA Klipw Data Compilation. Klipw data are available mostly for liposomes made up of phosphatidylcholine (PC) only or PC mixed with other membrane lipids. Hence, this work mainly uses data for PC (Table S1, Supporting Information). PC is among the major lipids in biological membranes.6 Moreover, PC is an experimentally convenient material, as it spontaneously forms

CRITICAL REVIEW

stable liposome vesicles in aqueous solutions and is overall neutral (zwitterionic) over a wide range of aqueous pH.18,19 Recently, Spycher et al.20 conducted a large literature survey for Klipw on PC-liposomes. This work formed the basis for our data collection. In this study, the original publications cited therein18,21
34 were all re-evaluated and additional studies11,12,16,17,35
45 were considered. In the literature, there exist Klipw data for a variety of aromatic compounds, whereas only limited data are available for aliphatic compounds. To mitigate this gap, we measured Klipw values for 14 volatile aliphatic compounds with liposome suspension using a headspace sampling method. In addition, Klipw for 16 hydrophobic chlorinated compounds were determined by a solid-phase dosing method. The procedures of both experiments are described in the Supporting Information, SI-1. In the literature, there are many data available for membrane affinity measured using immobilized artificial membrane (IAM) columns (reviewed by refs 46 and 47). These data were not used in our discussions, as there appear to be differences in interaction properties between liposomes and IAM (see SI-2). Klipw has been reported in various units, e.g., Lwater/Lliposome, Lwater/kgliposome, and kgwater/kgliposome. Fortunately, since the density of both hydrated liposome membrane and water is 1.0 kg/L,48,49 the value of Klipw does not significantly depend on which unit is used. Factors Influencing Klipw of Neutral Compounds. Only experimental data that are obtained under comparable conditions should be used to calibrate and evaluate models. In this section, factors that may affect the values of Klipw are overviewed. Membrane Fluidity and Types of PC. The most important factor that influences Klipw is the fluidity of the membrane. On heating through the main phase transition temperature (Tc), the membrane undergoes a transition from a low-fluidity “gel phase”, where acyl chains are aligned in order, to a high-fluidity “liquidcrystalline phase”, where acyl chains have a higher degree of freedom for movement.49,50 Liposomes composed of di(saturated acyl)-PC exhibit relatively high values of Tc and the value increases with chain length; e.g., dimyristoyl-PC (DMPC; 14:0/14:0; Tc = 23.5 °C), dipalmitoyl-PC (DPPC; 16:0/16:0; Tc = 41.5 °C), distearoyl-PC (DSPC; 18:0/18:0; Tc = 55.5 °C).1 The presence of unsaturated acyl groups lowers Tc; e.g., palmitoyl-oleoyl-PC (POPC; 16:0/18:1; Tc =
3 °C), dioleoyl-PC (DOPC; 18:1/ 18:1; Tc =
21 °C), egg yolk PC (Tc =
10 ( 5 °C).1 Biological membranes in natural conditions are generally in the liquid-crystalline state, reflecting their relatively high contents of unsaturated chains. On cooling below Tc, Klipw exhibits a sharp decrease. For example, Klipw values above and below Tc differ by a factor of 1
3 for substituted phenols in DMPC,36 ∼4 for 1-hexanol in DMPC,42 ∼8 for n-hexane in DMPC,43 and 10
25 for chlorobenzenes in DPPC.45 Similarly, at a constant temperature, liposome membranes in the gel state exhibit considerably smaller values of Klipw than liposomes in the liquid-crystalline state. The differences reported are up to ∼20 times38,39 or can be even ∼100 times.45 Thus, to obtain Klipw values representative for membranes in natural conditions, we considered only data measured above Tc. At temperatures higher than Tc, Klipw values for different types of PC fall within a fairly narrow range. The collected data exhibit mostly up to (0.2 log-unit variations for each compound (Table S1). Note that these variations also include method differences (see SI-3 regarding Klipw determination methods), interlaboratory differences, and minor temperature effects (see below), and random experimental variations. Exceptions are highly hydrophobic compounds, as will be discussed below. 5913

dx.doi.org/10.1021/es200855w |Environ. Sci. Technol. 2011, 45, 5912–5921

Environmental Science & Technology

Figure 1. Comparison of the temperature dependence between liposome
water and octanol
water partition coefficients. Values of ΔHow were calculated from the PP-LFERs.51 ΔHlipw values cited were measured above Tc.

Temperature Dependence of Klipw in the Liquid Crystalline State. Temperature effects on Klipw above Tc generally follow the van’t Hoff equation, i.e., linearity between log Klipw and 1/T. The experimental data available12,35,39,42
45 suggest that the enthalpy change of the liposome
water partitioning process, ΔHlipw, is roughly comparable to that of octanol
water partitioning, ΔHow (Figure 1; data are given in Table S2). Here, values of ΔHow had to be estimated from the reported PP-LFER equations51 due to the lack of experimental values. The differences between ΔHlipw and ΔHow are mostly within (10 kJ/mol. Thus, although tentatively, the temperature dependence of Kow can be applied to that of Klipw. That is to say, for most compounds |ΔHlipw| < 30 kJ/mol, which corresponds to only a 7 from refs 16 and 17 deviate particularly from the linearity between log Klipw and log Kow. The log Klipw values for PCBs and dioxins were measured with different methods, but even those measured in this study using a polymer-dosing method are somewhat lower than the values for PAHs. Generally, inconsistency in the log Klipw
log Kow plot can arise from experimental artifacts in Klipw,11 but it can also result from inaccurate Kow values and/or different interaction properties of liposome membrane and octanol. Therefore, more robust models that do not rely on Kow may be useful for further data evaluation. Multiple Regression with PP-LFER Descriptors. Polyparameter linear free energy relationships (PP-LFERs) are a multiple linear regression model and have been used to characterize partitioning into various environmental and biological phases.60
63 PP-LFERs take into account all relevant molecular interactions between the neutral solute and the phases. Thus, they are robust and applicable to compounds from different chemical classes. In this work, two types of PP-LFERs64,65 are used to model log Klipw log Klipw ¼ c + eE + sS + aA + bB + vV

ð1Þ

log Klipw ¼ c + lL + sS + aA + bB + vV

ð2Þ

The following notations are used for compound descriptors: E, excess molar refraction; S, dipolarity/polarizability parameter;

A, solute H-bond acidity; B, solute H-bond basicity; V, molar volume; L, the logarithm of the hexadecane
air partition coefficient. The lower cases are fitting coefficients and are to be determined by multiple linear regression analysis against experimental log Klipw. The S, A, and B terms are common in eqs 1 and 2 and are representative for polar interactions between the solute and the phases. To describe the nonspecific interactions, eq 1 uses E and V, whereas eq 2 uses L and V descriptors. In general, there is no significant difference in overall fitting quality between the two equations.65
68 Values of the compound descriptors have been reported for several thousands of compounds. Values of all six descriptors were available for 146 compounds in our data set (Table S3). Concerning hydrophobic compounds, descriptor values were available for 1,3,5-tribromobenzene, octachloronaphthalene, chlorobenzenes, PCBs, and PAHs in the data set. Descriptors were not available for hexabromobenzene, bromobiphenyls, and dioxins, the data of which thus could not be evaluated by PP-LFERs. Using the descriptor values available, the fitting coefficients in eqs 1 and 2 were calibrated against experimental log Klipw. In the preliminary calculations, it became apparent that the measured values for all PCBs from this study and ref 12 (both measured with polymer-mediated methods) and relatively small PCBs from ref 16 fit well to the PP-LFERs, whereas the other experimental values for PCBs16,17 and the value for octachloronaphthalene,16 which apparently show the “cut-off” behavior, are consistently too small and are systematic outliers in our whole data set (see SI4 for details). Hence, for PCBs only those data measured by polymer-mediated methods (i.e., data from this study and ref 12) were considered in the following discussions. After removing the other strong outlier (urea; error >3σ) from the data set, we 5915

dx.doi.org/10.1021/es200855w |Environ. Sci. Technol. 2011, 45, 5912–5921

Environmental Science & Technology

CRITICAL REVIEW

Figure 3. Fitting of the PP-LFER models to experimental log Klipw. The resulting regression equations are eqs 3 and 4 in Table 1. Only the selected data were used to calibrate the equations (see text).

refitted the PP-LFER equations. The final data set used for calibration is given in Table S3. The equations obtained from the calibration are eqs 3 and 4 in Table 1, and the experimental and fitted values are compared in Figure 3. The one-parameter log Kow model was also recalibrated after removing the data for hydrophobic compounds that are not verified by the PP-LFERs (i.e., data for hexabromobenzene, bromobiphenyls, dioxins, octachloronaphthalene, and data for PCBs from refs 16 and 17; a list of the data used for this calibration is given in Table S4). The result is shown in Table 1 and Figure S2. Fitting with the PPLFERs resulted in a standard deviation (SD) of 0.28
0.31 and was better than the simple log Kow regression (SD = 0.43). The qualities of overall fitting were comparable between eqs 1 and 2. However, eq 2 fit the Klipw values of PCBs generally better than eq 1 (fitted
measured = 0.08 ( 0.25 with eq 2 and 0.42 ( 0.26 with eq 1). This observation corroborates the recent finding by van Noort et al. 69 that the V descriptor in combination with the E descriptor does not properly account for the varying nonspecific interaction properties of PCBs. Thus, eq 2, which uses V and L to describe the nonspecific interactions, can be regarded as a more robust model than eq 1 (see SI-4 for further discussions). Additionally, PP-LFER equations specific to 25 and 37 °C were derived using the temperature-corrected Klipw data (eqs 5
8 in Table 1). ΔHow values calculated according to ref 51 were used for the temperature corrections. There is no difference in either coefficients or statistics between the PP-LFERs based on the raw data and on the data corrected to 25 °C, because many raw data were measured around 25 °C. The PP-LFERs for 37 °C appear slightly different, and, as expected, they calculate slightly lower Klipw values for highly hydrophobic compounds.

’ PREDICTIONS OF Klipw FROM MOLECULAR STRUCTURES Predictions of Klipw from molecular structure without using any solute descriptor are useful to screen a large number of chemicals. Modeling of molecular interactions also provides mechanistic

insights into the membrane partitioning processes. In this study, three predictive approaches were evaluated. Approach 1: Predictions via log Kow. First, log Kow values were predicted using KOWWIN v1.67a71 implemented in EPI Suite 4.0, SPARC v4.5 (http://archemcalc.com/sparc),72,73 or COSMOthermX (version C21_0111, COSMOlogic GmbH & Co. KG).74 The principles on which these calculation tools are based are described in the cited references. Then, Klipw values were calculated from eq 11 in Table 1. In KOWWIN and SPARC, the 2D-molecular structure of the solute (entered as a SMILES string) is the only input. COSMOthermX is based on quantum chemical calculations using the 3D structure of molecules. It can thus also take different conformers into account. COSMOconf was used for structure generation and optimization for possible conformers. Approach 2: Solvent Model. Klipw was calculated as a solvent
water partition coefficient using a hypothetical PC “solvent” phase. Thus, the membrane is modeled as a homogeneous, unstructured phase consisting of PC molecules. SPARC and COSMOthermX were used to run this calculation. The inputs for the solutes are the same as in Approach 1. Molecular structures of POPC, DMPC, and DPPC were tested as the input solvent structure, but the calculation results from both SPARC and COSMOthermX were not sensitive to the types of PC. Thus, only the calculations with DMPC were considered in the following discussions. Approach 3: COSMOmic. Klipw was calculated using COSMOmic, a subprogram of COSMOthermX. COSMOmic considers a structured bilayer membrane and calculates partition coefficients in many (here 40) sublayers of the membrane, which can be integrated with respect to the membrane depth to give Klipw. The details of the program have been described by the developers.75 A structure of the bilayer DMPC membrane constructed by molecular dynamics calculations76 was used as the input structure. Membrane structures from other sources (refs 77 and 78; pbd files downloaded at http://www.apmaths.uwo.ca/∼mkarttu/ and http://moose.bio.ucalgary.ca/, respectively) were also used, but no significant differences were found. 5916

dx.doi.org/10.1021/es200855w |Environ. Sci. Technol. 2011, 45, 5912–5921

Environmental Science & Technology

CRITICAL REVIEW

Figure 4. Predictions of log Klipw by (A) SPARC and (B) COSMOthermX using a hypothetical homogeneous, unstructured solvent consisting of DMPC molecules (Approach 2), and (C) COSMOmic using the bilayer DMPC membrane structure as model membrane (Approach 3). Some examples of outlier compounds are marked with open symbols explained in the figure. AE: average error ( standard deviation. RMSE: root mean squared errors.

Figure 5. Comparison of partition coefficients into membrane and storage lipids. Partition coefficients were calculated from eqs 7 and 9. Pesticides and pharmaceuticals are those compounds listed in refs 82 and 83, respectively, and also include some compounds not used as pesticides or pharmaceuticals.

All calculations were performed for 25 °C. All compounds in the data set were used as test compounds, but for hydrophobic compounds only the data qualified with the PP-LFER models above were used. All the calculation results are given in Table S4. Approach 1 provided predictions of Klipw that are mostly within 1 log unit from the measured values (Figure S3). The overall accuracy of predictions was similar for KOWWIN and SPARC with the root mean squared errors (RMSE) of 0.56 and 0.62, respectively. Predictions by COSMOthermX were somewhat less accurate (RMSE = 0.78) and PAHs appear to be systematic outliers. For Approach 2, the agreement between the experimental and predicted values highly depended on the compound type and the calculation software used (Figure 4A, B). SPARC fairly

accurately predicted Klipw of nonpolar compounds with errors typically