Sorption of Cationic Surfactants to Artificial Cell Membranes: Comparing Phospholipid Bilayers with Monolayer Coatings and Molecular Simulations

Abstract

This study reports the distribution coefficient between phospholipid bilayer membranes and phosphate buffered saline (PBS) medium (D_MW,PBS) for 19 cationic surfactants. The method used a sorbent dilution series with solid supported lipid membranes (SSLMs). The existing SSLM protocol, applying a 96 well plate setup, was adapted to use 1.5 mL glass autosampler vials instead, which facilitated sampling and circumvented several confounding loss processes for some of the cationic surfactants. About 1% of the phospholipids were found to be detached from the SSLM beads, resulting in nonlinear sorption isotherms for compounds with log D_MW values above 4. Renewal of the medium resulted in linear sorption isotherms. D_MW values determined at pH 5.4 demonstrated that cationic surfactant species account for the observed D_MW,PBS. Log D_MW,PBS values above 5.5 are only experimentally feasible with lower LC-MS/MS detection limits and/or concentrated extracts of the aqueous samples. Based on the number of carbon atoms, dialkylamines showed a considerably lower sorption affinity than linear alkylamine analogues. These SSLM results closely overlapped with measurements on a chromatographic tool based on immobilized artificial membranes (IAM-HPLC) and with quantum-chemistry based calculations with COSMOmic. The SSLM data suggest that IAM-HPLC underestimates the D_MW of ionized primary and secondary alkylamines by 0.8 and 0.5 log units, respectively.

Introduction

Phospholipid membranes separate (sub)cellular compartments from surrounding conditions and play an important role in the uptake, distribution, and toxicity of xenobiotics in multicellular organisms. Traditionally, the octanol–water partitioning coefficient (K_OW) is used as a predictor for the uptake, distribution, and accumulation of organic chemicals in various organisms and their tissues. However, K_OW does not adequately represent the partitioning of ionogenic organic chemicals (IOCs) between water and phospholipid membranes, because the ionic solute’s interactions with octanol do not include ionic bonds that occur with the anionic phosphate groups and cationic choline groups in phosphatidylcholine phospholipids.¹⁻³ Cell membranes are expected to be the dominant sorption site for organic cations in tissue.^4,5 Just like sorption coefficients to individual soil components are much more relevant for IOCs than K_OW,⁶⁻¹⁰ the membrane–water distribution coefficient (D_MW) is a logical alternative for K_OW as the main model parameter to predict, for example, the tissue distribution and critical membrane burden in organisms.^4,11 Many IOCs are highly relevant from an ecotoxicological perspective because of designed bioactive properties and/or continuous input via wastewater streams.^12,13 Cationic surfactants are hydrophobic IOCs with a relatively high potential to disrupt cell membranes.¹⁴ Cationic surfactants are commonly used down-the-drain ingredients in personal care products, because of antistatic properties (hair conditioner, fabric softener). A variety of quaternary ammonium salt cationic surfactants are specifically used as biocide or disinfectants.^15,16 Several cationic surfactants are regularly detected in various aquatic environments, particularly sediments.¹⁶⁻²⁰ Quaternary dialkylamines are highly adsorptive and therefore less accessible for biodegradation processes under certain conditions, and the longest chain structures such as DODMAC have subsequently been prohibited in several countries for certain uses.²¹ This study aims to measure and model the D_MW for cationic surfactants with different alkyl chain lengths and head groups, in order to improve environmental and toxicological hazard/risk assessment for this class of IOCs. Since most common protocols to determine K_OW are considered to be impractible for surfactants, the assessment of D_MW would provide a representative alternative hydrophobicity parameter.

Several experimental methods exist to determine the D_MW of IOCs at physiological pH, such as liposome dispersions, specialized HPLC columns with phospholipid coatings, and solid supported phospholipid membranes (SSLM). Methods employing dispersions of freshly prepared liposomes are most realistic and accurate, but equilibrium dialysis requires considerable effort and long equilibration times.^2,22 Potentiometric titrations need substantial concentrations of chemical and phospholipids.²³ Recently, immobilized artificial membrane HPLC columns (IAM-HPLC) have been used to determine (relative) measures of lipophilicity in a number of frameworks.²⁴⁻²⁷ Confounding pH-dependent surface charges in IAM-HPLC have recently been recorded in detail.²⁸ These surface charges can considerably influence the retention capacity factors of IOCs on IAM-HPLC physiological pH.²⁸⁻³⁰ At least for cationic compounds, confounding surface charges can be avoided by testing at low pH and highly saline eluent medium, and therewith one can specifically determine the IAM phospholipid–water distribution coefficient for the ionic form (K_IAM,ion).²⁸ IAM-HPLC consists of an ordered monolayer of phospholipids covalently linked to a silica support,³¹ instead of a dispersed double layer of phospholipids. This might reduce its relevance as a surrogate for the lipid bilayer cellular membrane. Solid supported phospholipid membranes (SSLMs) are available with macroporous spherical supports (e.g., silica beads) which are readily separated from the aqueous phase by mild centrifugation.³²

Recently, IAM-HPLC was used to determine intrinsic sorption affinities to the IAM phospholipid monolayer K_IAM,intr for 80 different hydrocarbon-based monoprotic cations (C_xH_yN⁺).²⁴ Remarkably, these K_IAM,ion values did not differ between analogue structures of primary, secondary, and tertiary alkylamines with the same alkyl chain length, and were marginally lower for quaternary ammonium chloride (QAC) analogues (∼0.2 log units). Quantum-chemistry based molecular calculations with a model DMPC bilayer (using the COSMOmic module within COSMOtherm software) of K_DMPC-W values for the ionic species closely aligned the full set of K_IAM,ion values but predicted a stepwise decrease in K_DMPC-W with each methylation of the charged N, with primary amines at a log unit higher affinity than analogous QACs. Droge et al.²⁴ stated that only measurements on phospholipid bilayers would clarify if either the effect of N-methylation is overpredicted by molecular simulations, or underpredicted by the IAM monolayer.

The main goal of the present study was to measure partitioning of several series of cationic surfactants with the molecular formula C_xH_yN⁺ to phospholipid bilayers using a commercially available SSLM assay, for comparison with IAM-HPLC results and COSMOmic simulations. We thereby focused on the influence of the alkyl chain length and different types of charged head groups but also on the difference between linear single chain alkylamines and dialkylamines. The applied SSLM assay (trademark TRANSIL) is sold as a standardized sorbent dilution series assay in a 96 well plate format³³ but was improved to facilitate measurements with hydrophobic organic cations.

Materials and Methods

Chemicals, Sorbent, and Solutions

Nineteen amine based cationic surfactants compounds were selected. Their molecular structures, physicochemical properties, purities, and suppliers are listed in the Supporting Information (SI, Table S1). Two secondary amines contained two linear alkyl chains of equal length (dihexylamine “S2–C₆”, dioctylamine “S2–C₈”). Other moieties besides linear alkyl chains include benzyl in three benzalkonium chloride compounds (BAC), and dodecylpyridinium (C₁₂-PYR) has the permanently charged nitrogen as part of an aromatic ring. All stock solutions were prepared as 100 ± 10 mM solutions in methanol, further diluted with methanol as necessary. All solutions in methanol were stored at −20 °C until use.

TRANSIL^XL Intestinal Absorption kits and TRANSIL^XL Intestinal Absorption kits for low affinity compounds were purchased from Sovicell GmbH (Leipzig, Germany). These kits consist of a 96 well plate made up of 12 strips with 8 wells each, individual strips containing two reference wells with phosphate buffered saline (PBS), and six wells with decreasing amounts (serial dilution factor of 1.8) of phosphatidylcholine coated macroporous silica beads (“beads”) suspended in phosphate buffered solution (PBS). The pore diameter of the silica beads has been specified as 4000 Å in the literature.⁴⁷ A 2012 paper by Hou et al. provides SEM images that give more insight into the three-dimensional structure of the beads.⁴⁸ PBS (Lonza BioWhittaker, Walkersville, USA; pH 7.4 ± 0.05, without Ca²⁺ and Mg²⁺) was used as the test medium for all experiments, unless noted otherwise. To assess the contribution to the observed D_MW of the small neutral fraction present at pH 7.4 for the ionizable amines (<1%), the D_MW was also determined at pH 5.2 (<0.01% neutral) for several alkylamines. A 10 mM acetate buffer was used with analytical grade acetic acid (2.0 mM) and sodium acetate (8.0 mM), dissolved in Milli-Q pure water (>18.2 MΩ·cm^–1, Millipore, Amsterdam, The Netherlands), and addition of 140 mM NaCl. Additions of all liquids were checked gravimetrically to record actual volumes.

Adapting the SSLM Test Protocol for Cationic Surfactants

Cationic surfactants are notoriously difficult to work with due to relatively high adsorptive properties and accumulation at the water–air interface. Concentration dependent sorption to polystyrene well plate material was expected for several of the tested surfactants, as well as sorption to pipet tips when transferring supernatant to autosampler vials.³⁴ To partially avoid these adsorptive challenges, the SSLM beads were transferred quantitatively to 2 mL glass autosampler vials. After equilibration of the chemicals with the beads on a roller bank, the vials were centrifuged and the supernatant in the autosampler vials could be directly injected by the stainless steel needle of the autosampler. This reduces pipetting steps and allows for testing in a larger aqueous volume, of which the composition can be customized. Whereas the 96 well plate assay applies two PBS reference wells (without SSLM) serving as 100% references of dissolved concentrations, it was expected that for several cationic surfactants substantial binding to the walls of the test vials would compromise a mass balance approach. We therefore applied quadruplicate methanol solutions as 100% reference (as calibration standards) and assume that losses due to sorption to the wall of the test vials in the vials that contain the lipids does not affect the estimated membrane sorption coefficients. The validity of this assumption has been validated.

A Rainin Pipet-Lite XLS electronic multichannel pipet with adjustable spacer (Mettler Toledo BV, Tiel, The Netherlands) was used to transfer the contents of six wells from one strip with decreasing amounts of phospholipid coated beads to 2 mL autosampler vials. Wells were flushed six times with 50 μL of PBS, to transfer all beads to the respective autosampler vials. Initially, autosampler vials were then filled with additional 400 μL PBS, placed on a Stuart SRT9 roller mixer (Boom BV, Meppel, The Netherlands) for 15 min at 33 rpm, centrifuged at 750 g (20 °C) for 10 min. To these test vials, as well as to the four 500 μL methanol references and four additional 500 μL PBS controls (to verify the extent of glass binding), 20 μL of spike solution was added. Overall, our tests were performed in the 1–1000 nM concentration range. We tried to cover isotherms with concentrations spanning at least a factor of 10, ideally a factor of 100. After addition of the spike, vials were transferred to a roller mixer for 15 min, centrifuged at 750 g (20 °C) for 10 min, and stored at 4 °C until LC-MS/MS analysis.

Pilot experiments with longer chain surfactants showed distinct nonlinear sorption isotherms, while the shorter chain surfactants showed linear isotherms. Several tests were performed to evaluate effects of spiking with solvent and bead density. Since the phospholipid bilayer is not covalently bound to the beads, we considered that minute—but significant—fractions of phospholipids might leak from the beads into the test medium during storage and handling (e.g., thawing) and form small suspended liposomes. Especially for compounds with high D_MW, these liposomes could impact the D_MW measurements by artificially increasing measured concentrations in the aqueous phase being sampled. After transfer of the beads to the autosampler vials, we therefore added fresh PBS up to 1.8 mL, centrifuged, and carefully pipetted off 1.7 mL of supernatant to remove the majority of the detached phospholipids and added 400 μL of PBS and 20 μL spike solution. Equilibration times of 5, 30, 60, and 240 min were compared with N-methyldodecylamine (“S₁₂”). A test with dodecylpyridinium (“PYR₁₂”) was performed to assess the tendency of phospholipids to leak from centrifuged beads while in the autosampler at room temperature, by injecting the same test series of six vials with washed beads 3, 12, and 21 h after centrifugation.

LC-MS/MS Quantification and SSLM Data Analysis

All samples were analyzed using a PerkinElmer (Norwalk, USA) HPLC system with autosampler, coupled to an MDS Sciex API3000 triple quadrupole mass spectrometer (Applied Biosystems, Foster City, USA). Retention of test compounds from the saline test medium was achieved on a Kinetex 5 μm XB-C18 column (50 × 2.1 mm; 100 Å) with a C18 guard column. The mobile phase consisted of Milli-Q (pump A) and methanol (pump B), both containing 0.1% formic acid by volume. A solvent switch was employed to flush PBS salts into waste, at 10% B for 6 min, before eluting the surfactants from the column with 90% B. The autosampler needle depth was adjusted to prevent accidental injection of beads. External calibration standards in methanol had concentrations ranging from ∼2 nM up to ∼35 μM. Detailed LC and MS/MS parameter settings for each compound can be found in Table S2.

The total spiked amount (A_total) of surfactant in the autosampler will distribute between the aqueous phase (A_water), the phospholipid coating on the beads (A_lipid), and the glass/cap surfaces (A_glass). A_total is obtained from the average concentrations in the methanol controls (C_MeOH), and D_MW can then be calculated for each sample:

1

V_lipid is the volume of phospholipids on the beads, as provided by the supplier. Because A_glass is in equilibrium with A_water, then A_glass can be considered negligible for the calculation of both A_lipid and D_MW if A_lipid is >90% of A_total and if A_glass < A_water. A_glass was determined using the PBS reference samples, which demonstrated the level of equilibrium between glass sorbed fractions and dissolved fractions (at the spiked concentration). Samples for decylbenzyldimethylammonium (“BAC₁₀”) were emptied after analysis of C_water and flushed once with Milli-Q, and glass walls were extracted with 90% B/10% A eluent mixture. Concentration independent log D_MW values were obtained by fitting a linear curve on a double logarithmic plot with a forced slope of 1.

IAM-HPLC Measurements of the Phospholipid Monolayer–Water Distribution Coefficient K_IAM,intr for Cationic Surfactants

The IAM-HPLC procedure described for strongly sorbing C_xH_yN⁺ cationic amines without UV-absorbing moieties was followed as described previously.²⁴ Briefly, a solvent dilution series at pH 5 (10 mM acetate buffer) was tested in triplicate with LC-MS/MS detection. From a ∼1–5 mg/L surfactant sample in 10% acetonitrile, 5 μL was injected into an eluent mixtures of ≤30% acetonitrile, at flow rates of 1.0 mL/min. Multiplying the retention capacity factor k_IAM with the column’s phase ratio of 18.9 gives the apparent distribution coefficient to the IAM phospholipid phase (K_IAM,app) in each tested eluent mixture.³⁵ Extrapolation of the K_IAM,app values to fully aqueous medium buffered at pH 5 gives the intrinsic K_IAM,intr. For each surfactant at least six measurements were made.

COSMOmic Calculations of the K_DMPC-W for Cationic Surfactants

COSMOmic was run within COSMOtherm Version C30_1501, as described in the previous comparison between IAM-HPLC and COSMOmic.²⁴ However, instead of using COSMOmic’s DMPC example micelle (1,2-dimyristoyl-sn-glycero-3-phosphocholine), we now used the same time averaged DMPC micelle file and TZVP-optimized structure of DMPC as used by Bittermann et al.³⁶ Briefly, the input file to represent a hydrated phospholipid bilayer is obtained with a molecular dynamics (MD) run, using 128 DMPC molecules equilibrated with thousands of water molecules. COSMOmic divides the average atomic distribution in the MD simulated DMPC bilayer into 30 layers for one-half of the hydrated bilayer and uses the lowest free energy for each surfactant structure at 162 orientations at each layer to calculate the weighted DMPC–water partition coefficient (K_DMPC-W). The three-dimensional input structures for each cationic surfactant were quantum-chemically optimized for calculations at TZVP level with COSMOmic, including different conformers (see Table S3 for information on conformers). COSMOconf version 3.0 was used to create up to six of the most relevant conformers for all charged surfactants.

Results and Discussion

Measurements of D_MW with adapted SSLM Assays

Measured concentrations of the quadruplicate methanol control samples differed by less than 3.9% for all compounds tested. For C₈-alkylamines and C₁₀-alkylamines, the PBS controls showed 0–30% lower concentrations than the methanol controls, with exception of the larger C₁₀-benzalkonium (“C₁₀-BAC” 39%). For C₁₂ surfactants, losses to autosampler surfaces were between 20 and 60% in PBS. If PBS references would have been used for C₁₀-BAC as if they represented 100% of the available compound D_MW would have been 0.2 log units lower than with methanol control samples. Using methanol controls, measured extracts of the glass walls in vials with beads showed that the residual impact of glass binding on D_MW calculations was insignificant (0.011 log units).

Concentrations of test compound were aimed at keeping the phospholipid/sorbed compound molar ratio above 60 to prevent possible electrostatic effects due to the accumulated charge in the membranes.^37,38 As shown in the full matrix of the final sorption isotherms for all tested compounds in the SI Figure S8, for nearly all of the selected surfactants, we have tested up to this maximum sorbed concentration in the membranes of ∼40 mmol/kg to avoid electrostatic effects. Although the corresponding dissolved concentrations are orders of magnitude below the critical micelle concentrations (CMC, Table S1), the tested concentrations are most likely still well above highest environmental concentrations but may be in the range of adverse effect concentrations. Concentrations of phospholipid in the test vial should result in sorbed fractions of at least 30%, to minimize effects of analytical uncertainties of C_water on the mass balance calculations. The adapted TRANSIL^XL Intestinal Absorption kits allows for measuring D_MW values above 1000 with buffer volumes of ∼525 μL in the autosampler vials. The TRANSIL^XL Intestinal Absorption kits for low affinity compounds contain a factor of ∼20 higher levels of beads per well, thereby allowing for measurements of D_MW ≥ 50. The results from these two kits show overlapping sorption isotherms and concentration independent D_MW, as shown for dihexylamine (“S2-C₆”) in Figure 1. The sorption isotherms for C₈- and C₁₀-alkylamines also showed concentration independent D_MW values, and overlapping sorption data with and without PBS renewal (Figure S1).

Figure 1.

Sorption data for dihexylamine (S2-C6) obtained with two different sorbent dilution series and fit of a linear sorption isotherm (slope of 1), resulting in a log D_MW,PBS of 3.15 (95% CI 3.10–3.20).

However, for the C₁₂-chain surfactants series measured without flushing off leaked phospholipids showed distinctly nonlinear isotherms, with higher D_MW values for the highest concentrations in a series (samples with lowest amounts of SSLM beads), and no correspondence between two series spiked at different initial concentrations (Figure S1). For each individual series, the slope of a linear isotherms on the double logarithmic plots was >1. This result would cause doubt on the resulting K_MW from the SSLM assay, and an apparent concentration dependent sorption affinity over such a narrow tested concentration range would have considerable deviations of the K_MW at considerably lower (e.g., most environmental) and possibly higher concentrations (e.g., as applied, or at adverse effect levels). We found no evidence of influence of solvent from spiking solution, as data for secondary N-methyldodecylamine (“S₁₂”) from methanol stock solutions overlapped with nonlinear results from stocks dissolved in water (Figure S2). Instead of a sorbent dilution series, we then tested primary dodecylamine (“P₁₂”) with two series with constant concentration of SSLM material, spiked at six different concentrations (accompanied by six sets of methanol controls). Now, each series indicated concentration independent D_MW (slope of 1 on logarithmic plot), but again the series with higher amount of SSLM material showed a lower sorption affinity (Figure S3). Evidently, higher SSLM material resulted in a higher detached amount of phospholipids from the beads leaking into the aqueous phase. If the sorbed amount of cationic surfactants to this phase significantly increases the measured C_water, this leads to underestimation of the D_MW. Using a common extension of eq 1 for third phase systems,³⁹ we modeled this effect by assuming a constantly leaked fraction (f_leak) of the total amount of lipids coated on the beads, i.e., where the amount of lipids dispersed in the medium equals V_lipid·f_leak:

2

Experimental and modeled results are plotted for dodecylpyridinium (PYR₁₂) in Figure 2, combining data for “unwashed” samples (still including medium from the well plate, thus with f_leak still present) with “washed” samples (with f_leak mostly removed). As shown in Figure 2, the curve representing 1% of the phospholipids leaking from bilayers into the test medium (f_leak = 0.01) approximated the observed experimental values that show a nonlinear curve. A modeled curve for f_leak = 0.02 overestimated the observations. More plots comparing simulations for unwashed and washed beads for compounds with log D_MW of 4.0, 4.5, and 5.0 can be found in the Supporting Information (Figure S4). All experimental data and modeling output suggest a phospholipid leakage of approximately 1% from the bilayers on the beads into the medium, which–if not removed from the test medium–will influence the sorption isotherms for compounds with a D_MW > log 4.0.

Figure 2.

Experimental sorption data and simulated sorption data for dodecylpyridinium (PYR12) obtained with “unwashed” and “washed” sorbent dilution series, where unwashed still contains the medium from the well plates, whereas the medium was replenished with fresh PBS in “washed” samples. The simulated series show curves of a lipid leakage fraction of 0%, 1%, and 2%. The fitted linear curve for the experimental data indicates a log D_MW,PBS of 4.89 (95% c.i. 4.84–4.95).

Varying the incubation time on the roller mixer (5–240 min) did not have a significant impact for fully dispersed SSLM solutions (Figure S5); 30 min was kept as standard. After centrifugation, the autosampler vials may stand for several hours in the autosampler before injection. The results of analysis after standing for 3, 12, or 21 h indicated no further leakage of phospholipids, as the measured surfactant concentrations and resulting sorption isotherm showed excellent overlap, whereas a significant fraction of leaked phospholipids would have increased the apparent concentrations in the medium (Figure S6). In test solution of pH 5.4 (washed), the D_MW values of an ionizable tertiary N,N-dimethyldecylamine (“T₁₀”) as well as of a permanently charged QAC N,N,N-trimethyldecylammonium (“Q₁₀”) were not statistically different from the D_MW in PBS (pH7.4), indicating that the <0.5% neutral fraction of T₁₀ is not contributing to the measured D_MW in PBS and that there were no confounding pH-dependent surface charge effects of the SSLM material. In contrast, pH-dependent surface charge effects in IAM-HPLC confound the D_MW of organic cations in saline medium by ∼0.7 log units between pH 5 and 7.4.²⁸ As a result, the measurements of D_MW in PBS of cationic surfactants all relate to the partition coefficient of the ionic form (K_MW,ion) and can thus be directly compared to COSMOmic simulations with the ionized structures and IAM-HPLC measurements made at pH 5.

Modeling the D_MW by the Structure of Single Chain Cationic Surfactants

Isotherms were fitted with a fixed slope of 1 for all data points from tests with washed medium, where f_sorbed ≥ 0.3 and the phospholipid/sorbate ratio was higher than 60 (Figure S8); an overview of all isotherm details can be found in Table S3. Standard errors were <0.05 log units, and 95% confidence intervals were <0.17 log units for all tested D_MW values. The resulting D_MW values (Table 1) were used to derive a simple quantitative structure activity relationship (QSAR) based on binary values for headgroup types and the length of the alkyl chain:

3

Where n_carbon denotes the number of carbon atoms in the alkyl chain, N^Benz is a binary value indicating the presence (1) or absence (0) of a complete benzalkonium headgroup (three compounds included), and N^CC (two compounds included), N^CCC (three compounds included), and N^CCCC (four compounds included) are binary values indicating the degree of N-alkylation. The N^CCCC value for benzalkonium compounds should be 0 by default. C₁₂-pyridinium was omitted as there was only one compound with this headgroup. Dialkylamines were omitted since they are expected to behave differently because the dual alkyl chain influences their orientation in the phospholipid membrane.²⁴ Both types are discussed further. As shown in Figure 3, the regression model of eq 3 fits all input compounds within a factor 3 of the experimental values (root-mean-square error of 0.20). Using binary values on only 3 or 4 compounds for each parameter in the QSARs of eq 3 results in a fairly low level of redundancy in the data. However, our main aim of this excercise was to obtain insight in these simplified headgroup properties (N-methylation) and not to provide a functional well validated QSAR to predict K_MW values for cationic surfactants. Considering that this data set of cationic surfactants is still structurally relatively nondiverse and relatively small, no further efforts were made to refine a QSAR based on other physicochemical or quantum-chemically derived parameters. Obviously, D_MW increases with longer alkyl chains, with 0.59 ± 0.03 log units per CH₂ unit, which can be used to extrapolate D_MW predictions to longer chain surfactants. Comparing this fragment value with the tabulated D_MW values (Table 1) it seems to be slightly higher than expected based on the experimental values obtained for the quaternary ammonium compounds, and slightly lower than expected based on the secondary and tertiary C₁₂ compounds. Although we expect a constant CH₂ unit contribution for all single linear alkyl chains, the contributions of CH₂ unit in the chain near the charged amine may be slightly lower, as these may not all reside in the hydrophobic core, and this is not defined in eq 3.²⁴ Effects of pH and possible neutral fraction for ionizable amines were ruled out based on additional tests at pH 5.4 (above and Figure S7). A consistent trend of decreasing D_MW with increasing methylation of the N atom is observed for the three analogue series of C₈-, C₁₀-, and C₁₂-amines. Primary amines have 0.28 ± 0.16 log unit higher D_MW than secondary amines, which are 0.28 ± 0.16 log units higher than tertiary amines, which are 0.53 ± 16 log units higher than QACs (excluding benzalkonium compounds). Taking the average over all the C₈-, C₁₀-, and C₁₂-amines, primary amines sorbed 1.06 log units stronger than the QAC analogues. Considering that the quaternary amine has three more CH₂ units than the primary analogue, this is a remarkable feature of IOCs. The benzalkonium compounds have a D_MW of 1.0 ± 0.17 log units higher than trimethylalkylammonium compounds, which reflects the effect of an additional benzyl moiety. With an experimental log D_MW of 4.89, PYR₁₂ positions in between BAC₁₂ and Q₁₂ compounds, corresponding to the molecular volume differences. For 16 out of 19 compounds, two or more conjoined series were tested, demonstrating both consistency of experimental results and steadiness over multiple orders of magnitude of the D_MW estimates.

Table 1. Log D_MW Values for All Compounds Tested with the SSLM Assay,^a As Well As Uncorrected IAM-HPLC log K_IAM,intr Measurements and COSMOmic Predicted log K_DMPC-W Values (No Offset Correction).

alkyl chain	method	primary amine (P)	secondary amines (S)	tertiary amines (T)	trimethyl ammonium (Q)	benzalkonium cations (BAC)	pyridinium cations (PYR)	secondary dialkyl-amines (S2)	quaternary dialkyl-ammonium (Q2)
C6	SSLM					2.12		3.15
	IAM	1.32				2.42		2.88
	COSMO	2.63	1.92	1.45	1.08	1.87		2.77
C8	SSLM	3.10	2.76	2.35	2.18	3.11		4.65
	IAM	2.30	2.33	2.35	2.18	3.30		5.04
	COSMO	3.71	3.01	2.53	2.18	2.96		4.34
C10	SSLM	4.30	3.98	3.65	3.34	4.01
	IAM	3.55	3.59	3.59	3.44	4.47
	COSMO	4.83	4.14	3.70	3.30	4.05		6.26	7.53
C12	SSLM	5.58	5.39	5.30	4.35		4.89
	IAM		4.81		4.57
	COSMO	6.01	5.28	4.76	4.47	5.14	4.77	8.40	9.79
C14	SSLM				5.46
	COSMO	7.15	6.42	5.89	5.61	6.65	5.90		12.06
C16	COSMO	8.31	7.56	7.07	6.72	7.90	7.03		C16/C18
C18	COSMO	9.43	8.74	8.29	7.94	9.01			15.51
C22	COSMO				10.1

^anr of data used, standard errors, and 95% confidence intervals are presented in SI Table S3. δ_IAM-SSLM corrected K_MW(IAM) values are presented in SI Table S3.

Figure 3.

Observed and predicted log D_MW,PBS values for single linear chain cationic surfactants using eq 3 (P = primary amines, S = secondary amines, T = tertiary amines, Q = trimethylalkylammonium compounds, BAC = benzalkonium chloride compounds).

Correlation of SSLM Data with IAM-HPLC Measurements and COSMOmic Predictions

The available K_IAM values obtained with IAM-HPLC (see detailed solvent series data in Figure S9, Table S3) and K_DMPC-W values from COSMOmic (Table S3) are plotted against the experimental D_MW results from the SSLM assay in Figure 4A. Overall, the alternative data sets and the SSLM data align reasonably well, with RMSE of 0.39 and 0.27 log units, for IAM-HPLC and COSMOmic, respectively. Instead of eq 3, now also dialkylamines and pyridinium compounds can be included. As discussed in Droge et al.,²⁴ IAM-HPLC data and COSMOmic predictions differed in the contributions of the N-methylations to the sorption affinity. The SSLM D_MW values confirm the ordering of primary > secondary > tertiary > quaternary amine analogues as predicted by COSMOmic and indicate that IAM-HPLC accounts insufficiently for effects of N-methylation. Considering that the SSLM data are obtained with relatively fluid phospholipid bilayers, and IAM-HPLC applies a covalently bound monolayer, we suggest applying corrective increments for the N-methyl headgroup contributions to K_IAM values of IAM-HPLC, compared against the SSLM D_MW values (δ_IAM-SSLM). Multiple linear regression results in δ_IAM-SSLM of 0.78 ± 0.07 (se) log units for primary alkylamines (i.e., K_IAM values underestimate the K_MW in bilayers), and 0.47 ± 0.09 log units for secondary amines. The δ_IAM-SSLM for tertiary amines is insignificant (−0.03 ± 0.10 log units), and very small for quaternary alkyltrimethylamines −0.11 ± 0.09. For COSMOmic δ_DMPC-SSLM amine type corrective increments were derived similarly. Interestingly, primary amines tended to be slightly overestimated by COSMOmic (0.17 ± 0.17 log units), while the other amine types were slightly underestimated (0.2–0.4 ± 0.16 log units). COSMOmic seems to be capable of predicting the effect of N-methylation in phospholipid bilayers. Contrary to suggested correction of COSMOmic values reported previously,²⁴ the SSLM data support the notion that the IAM-HPLC monolayer is the most probable source of inconsistency, creating scatter between IAM-HPLC and COSMOtherm. Using these δ_IAM-SSLM and δ_DMPC-SSLM corrective increments, there is strong correspondence between SSLM D_MW values, the corrected IAM-HPLC K_MW(IAM) values, and corrected COSMOmic K_DMPC-W values, as shown in Figure 4B. All values are well within a factor 2 of the 1:1 line and the RMSE improved from 0.39 to 0.21 (IAM-HPLC) and from 0.27 to 0.18 (COSMOmic).

Figure 4.

Experimental log D_MW,PBS values with TRANSIL bilayers plotted against experimental log K_IAM results from the IAM-HPLC monolayer and simulated log K_DMPC-W values with DMPC bilayers using COSMOmic. In the graph on the right, IAM-HPLC values are corrected for 1° amines with +0.8 log units and 2° amines with +0.5 log units. COSMOmic values are corrected for 1° amines with −0.5 log units.

The three methods attribute similar effects of alkyl chain length, but also display a striking consistency in the difference between secondary dialkylamines and linear chain secondary alkylamine analogues. As discussed in Droge et al.,²⁴ COSMOmic provides a mechanistic explanation for the relatively lower contribution of CH₂ units to the D_MW compared to single chain surfactants. Their design with two alkyl chains creates a steric effect, where the most favorable molecular position and orientation is mostly within the headgroup area of the phospholipid bilayer where—as a compromise—neither alkyl chain is aligned most favorably into the hydrophobic core region. The SSLM results confirm the difference in D_MW between secondary linear amine N-methyldodecylamine S₁₂ and secondary dihexylamine S2-C₆ by 2.24 log units, as was observed by Droge et al.²⁴ for IAM-HPLC (1.93 log units) and COSMOmic (2.43 log units). For all three methods the difference observed is much larger than expected with one extra CH₂ fragment in S₁₂. Between S2-C₆ and dioctylamine (S2-C₈), the SSLM D_MW increased by 1.5 log units, while the regression model in eq 3 would predict a 2.36 log unit increase based on the addition of four CH₂ units. COSMOmic also predicts a smaller value (1.73 log units) for the CH₂ increment between S2-C₆ and S2-C₈, while IAM-HPLC showed a 2.16 log unit difference (Table S3).

D_MW Values for Cationic Surfactants Compared to K_OW Predictions

The most recent analysis of K_MW values of neutral compounds,⁴⁰ mostly obtained with liposomes, showed a strong correlation with K_OW:

4

There is only a poor correlation (R² = 0.49) between the SSLM measured log D_MW at pH 7.4 and the log K_OW of the neutral primary, secondary, and tertiary amines, as shown in SI Figure S10. The influence of the methyl units on the charged nitrogen are reversed for the two distribution coefficients. Each N-methyl unit increases the log K_OW while it reduces the log D_MW. Also, even though the sorption affinity of the protonated amines to the SSLM bilayer increases with linear alkyl chain length as the K_OW predicts, the log D_MW of the dialkylamines is orders of magnitude lower than the log K_OW (for S2-C₈ 4.61 and 7.01, respectively). Typically for studies on the toxicokinetic properties of ionizable chemicals, a “log D approach” is followed, correcting the log K_OW for speciation of the neutral amines at the tested pH 7.4 (99.9% ionic for primary and secondary amines, 99.7% for tertiary amines). To derive a log D, however, the affinity of the ionic species for octanol needs to be known or predicted. In the absence of measurements, often either a constant factor of ∼3 log units lower than log K_OW is applied, or the affinity of the ionic species is ignored and the K_OW is multiplied by the fraction of neutral species. As shown in Figure S10, both such log D approaches underestimate the sorption affinity to membranes by a up to a factor of 1000 and do not solve the poor correlation between log D_MW and log K_OW. Instead, log K_OW could still be used to identify specific scaling factors to the difference in sorption affinity to a bilayer between neutral and ionic species, with the neutral affinity still based on eq 4. Table S3 lists the K_MW values for the neutral primary, secondary and tertiary amine species calculated via eq 4 (log K_OW predicted by ACD/Laboratories). Accordingly, the average difference between charged D_MW,PBS and neutral K_MW species (Δ_MW) for primary amines is −0.05, so the affinity of charged primary amines is larger than the neutral species. For the three linear single chain N-methylalkylamines Δ_MW is 0.44, for the linear N,N-dimethylalkylamines, the Δ_MW is 1.25. These values closely correspond to the scaling factors suggested for the bioaccumulation model for ionogenic compounds (BIONIC) proposed by Armitage et al.,⁴ which were 0.3, 0.5, and 1.25 for primary, secondary and tertiary amines, respectively, based on data sets of measured K_MW values for both ionic and neutral species. However, the Δ_MW is 1.90 and 2.55 for the two dialkylamines S2-C₆ and S2-C₈, respectively, much higher than the 0.44 derived with the other secondary amines. The examples of the dialkylamines show that applying a single Δ_MW scaling factor for all secondary amines to calculate the D_MW,PBS from the K_OW relationship in eq 4, can lead to erroneous values. Similarly, this exercise shows that K_OW is not an adequate single descriptor to model the D_MW values for ionized compounds. Measured K_MW values with TRANSIL, K_PLIPW values with IAM-HPLC, and even simulated K_DMPC values with COSMOmic, provide much more accurate and more mechanistically sound estimates compared to K_OW-based regressions.

Perspective on SSLM Assay Measurements and Associated D_MW Estimates

This study showed that the adapted SSLM protocol, with SSLM beads transferred from a well plate to autosampler vials, facilitated the analysis, improved recovery in methanol reference vials, and gained experimental control over the aqueous phase. The medium renewal removed third phase liposome artifacts, and allowed for altered pH of the test medium. The problem of detached phospholipids in the original SSLM medium became significant for all compounds with a D_MW higher than log 4, while using methanol controls instead of PBS controls seems mostly important for hydrophobic organic cations, and surfactants in general.⁴¹ The experimental determination of D_MW becomes problematic above log 5.5, because with the required phospholipid/sorbate molar ratio >60, the aqueous phase concentrations obtained directly from the autosampler vials, are nearing LC-MS/MS detection limits. This means that to measure D_MW values for cationic surfactants with alkyl chains longer than C₁₂, and dialkylamines with alkyl chains longer than C₈, either cumbersome solid phase extraction steps from larger test volumes are required—which may include uncontrolled adsorption losses—or sorption affinities need to be extrapolated with the model, or from series of smaller analogues. Alternatively, COSMOmic seems to provide a realistic and accurate predictive tool for cationic surfactants, which allows for extrapolations to longer chain cationic surfactants (Table 1) and slight alterations of the head groups. For example, the log K_DMPC-W for didecyldimethylammonium chloride (DDAC), a commonly used biocide,⁴² and cetylpyridinium, a commonly used antiseptic, are 7.53 and 7.03 (Table 1), which are both experimentally not feasible to measure with currently applied IAM-HPLC and SSLM assays. The log K_DMPC-W for behentrimonium, a trimethylalkylammonium compound with a chain length of C₂₂ which is used in many hair cair products and which has been detected in marine sediments,⁴³ is 10.3. The dialkylquat DODMAC (mixed chain length of C₁₆/C₁₈), banned for certain uses in the EU,²¹ has a predicted log K_DMPC-W of 15.5. For these examples, the very high predicted sorption affinities to cell membrane should be considered in risk assessment models (e.g., for bioaccumulation⁴) alongside strong sorption affinities to environmental particles,⁶⁻⁹ which strongly reduces the bioavailability, and thus result in relatively low accumulation from the environment into tissues.^34,44−46

Supporting Information Available

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.6b05662.

• Additional tables with information on chemicals, LC-MS/MS settings, and sorption isotherm details, as well as figures with detailed sorption data under various conditions (PDF)

Author Present Address

^† S.T.J.D.: UvA-IBED, Postbus 94248, 1090 GE Amsterdam, The Netherlands.

The authors declare no competing financial interest.