The Orientation and Dynamics of Estradiol and Estradiol Oleate in Lipid Membranes and HDL Disc Models

Abstract

Estradiol (E₂) and E₂ oleate associate with high-density lipoproteins (HDLs). Their orientation in HDLs is unknown. We studied the orientation of E₂ and E₂ oleate in membranes and reconstituted HDLs, finding that E₂ and E₂ oleate are membrane-associated and highly mobile. Our combination of NMR measurements, molecular dynamics simulation, and analytic theory identifies three major conformations where the long axis of E₂ assumes a parallel, perpendicular, or antiparallel orientation relative to the membrane’s z-direction. The perpendicular orientation is preferred, and furthermore, in this orientation, E₂ strongly favors a particular roll angle, facing the membrane with carbons 6, 7, 15, and 16, whereas carbons 1, 2, 11, and 12 point toward the aqueous phase. In contrast, the long axis of E₂ oleate is almost exclusively oriented at an angle of ∼60° to the z-direction. In such an orientation, the oleoyl chain is firmly inserted into the membrane. Thus, both E₂ and E₂ oleate have a preference for interface localization in the membrane. These orientations were also found in HDL discs, suggesting that only lipid-E₂ interactions determine the localization of the molecule. The structural mapping of E₂ and E₂ oleate may provide a design platform for specific E₂-HDL-targeted pharmacological therapies.

Introduction

In Western societies, atherosclerotic cardiovascular diseases (ACDs) are the most common causes of death and several factors are increasing the risk for ACD, such as increases in the prevalence of obesity, metabolic syndrome, and type 2 diabetes. Premenopausal women clearly have a lower risk than men, but the incidence of ACD in women increases after natural or surgical menopause (1,2), and this has been associated with a reduction of circulating estrogen. Estrogens display several antiatherogenic functions, such as vasodilatation, inhibition of smooth muscle cell proliferation, and decrease in vascular endothelial permeability (3). Estrogens also affect biosynthesis of high-density lipoproteins (HDLs) and elevate the plasma level of these proteins while reducing those of low-density lipoproteins (LDLs) (4). In addition, HDL has inherent antioxidative properties, which attenuate lipoprotein oxidation, and HDL-associated estrogens may play a role in this function (5–7).

Oxidation of LDL and accumulation of LDL in the subendothelial intimal space of arteries are crucial steps in the progression of atherosclerosis (8,9). 17β-estradiol (E₂) is transported by lipoproteins, esterified, to their tissue targets (10). Of total serum estradiol, a mean of 0.7% (549 pmol/L) is in the form of fatty acyl esters (11). The majority of lipoprotein estradiol fatty acyl esters, 54%, is recovered in HDL, and 28% in LDL, whereas nonesterified, free estradiol is not detectable. Previous reports demonstrate that in circulation, E₂ is esterified by lecithin-cholesterol acyltransferase in HDL particles (12–14). Instead of using the C-3 position in the A-ring of the molecule, as in the case of cholesterol, E₂ is esterified only at the 17β hydroxyl group in the D-ring of the molecule (15). After generation of E₂ fatty acyl esters in HDL, similar to cholesteryl esters, these fatty acyl derivatives are transferred into apoB-100-containing particles consisting primarily of LDL and VLDL via the function of cholesterol ester transfer protein (16,17). Vihma and Tikkanen (18) analyzed the distribution of E₂ fatty acyl esters between circulating lipoproteins, finding that the major portion of E₂ fatty acyl esters was recovered in HDL and LDL fractions (54% and 28%, respectively).

Although the function of E₂ fatty acyl esters is not clear, they are known to have rather long half-lives and to express prolonged hormonal activity, suggesting that they may function as a reservoir for the release of active E₂. Kinetically, the liberation of E₂ from its fatty acyl ester form would also be a rapid method for generating active hormone compared with biosynthesis starting from cholesterol. Therefore, it is important to understand the interaction of E₂ and its derivatives with lipoproteins such as HDL, since this association directly regulates hormone release and activation processes in vivo. Reports on the association and steric orientation of estrogens on lipoproteins are scarce. Current models of HDL particles do not consider estradiols and their esters (19). E₂ fatty acyl esters could become orientated on a lipoprotein surface such as HDL in a way analogous to that observed for free cholesterol, i.e., with the lipophilic 17β fatty acyl chain of the E₂ sticking into the lipoprotein core with the A ring 3β-hydroxyl group directing on the surface (7).

Structural data can contribute to understanding such complicated biological processes. Solid-state NMR spectroscopy has great potential for studying the structure and dynamics, as well as the membrane orientation, of small molecules in lipid assemblies (21–24). In particular, ²H NMR has proven useful to determine the orientation and dynamics of sterols in lipid membranes (25–27). Here, we used ²H NMR of deuterated E₂ and E₂ oleate to determine their orientation and dynamic distribution in lipid membranes and HDL discs. NMR measurements and analysis were assisted by molecular dynamics (MD) simulations of E₂ in 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) bilayers to develop a model of the orientation and molecular dynamics of E₂ and E₂ oleate in lipid bilayers and HDL discs.

Materials and Methods

Materials

POPC was purchased from Avanti Polar Lipids (Alabaster, AL). 3,4-¹³C-estradiol (3,4-¹³C-E₂) and 2,4,16,16-²H-estradiol (E₂-d₄) were purchased from Euriso-Top (Saarbrücken, Germany). The chemical structure of E₂ is shown in Fig. 1 A.

Figure 1.

(A) Chemical structure and numbering of the individual atoms of E₂. (B) To describe the orientation of the molecule within the bilayer, an internal coordinate system ${\overrightarrow{x}}_{\mathrm{int}},{\overrightarrow{y}}_{\mathrm{int}},{\overrightarrow{z}}_{\mathrm{int}}$ was defined (blue arrows). The z axis of the coordinate system points from C-16 to C-3, whereas the x axis points from the z axis toward C-12. With these two axes, the internal y axis is defined via ${\overrightarrow{z}}_{\mathrm{int}}\times {\overrightarrow{x}}_{\mathrm{int}}={\overrightarrow{y}}_{\mathrm{int}}$. Two angles were used that describe the orientation of this internal coordinate system within the bilayer: the roll angle, α_IM, and the pitch angle, β_IM, which are defined via the Euler rotations for a coordinate transformation from the internal coordinate system (I) to the membrane coordinate system (M). This transformation can be visualized as three subsequent rotations. The first rotation is about the internal z axis with the roll angle, α_IM, and the result is the new coordinate system ${\overrightarrow{x}}_{\mathrm{int}}^{\prime },{\overrightarrow{y}}_{\mathrm{int}}^{\prime },{\overrightarrow{z}}_{\mathrm{int}}^{\prime }$ (green). The magnitude of α_IM is such that the new ${\overrightarrow{y}}_{\mathrm{int}}^{\prime }$-axis is perpendicular to the membrane normal (large red arrow). Therefore, with the second rotation (which creates the new coordinate system ${\overrightarrow{x}}_{\mathrm{int}}^{″},{\overrightarrow{y}}_{\mathrm{int}}^{″},{\overrightarrow{z}}_{\mathrm{int}}^{″}$ (red)) about this ${\overrightarrow{y}}_{\mathrm{int}}^{\prime }$-axis the ${\overrightarrow{z}}_{\mathrm{int}}^{\prime }$-axis (which is the same as the ${\overrightarrow{z}}_{\mathrm{int}}$-axis) can be made identical with the membrane normal. Therefore, the pitch angle, β_IM, simply is the angle between ${\overrightarrow{z}}_{\mathrm{int}}$ and the membrane normal. A third rotation about γ_IM is not necessary, as it is averaged out due to the axial symmetry of the membrane about its normal. The angle α_IM can have values between 0° and 360°, whereas β_IM has values between 0° and 180°. To see this figure in color, go online.

E₂ oleate synthesis

A total of 18 mg (0.065 mmol) of 2,4,16,16,17-²H-estradiol produced as described in Kiuru and Wähälä (28) was placed in a microwave tube with 0.1 mL pyridine, 1 mg (dimethylamino) pyridine, 2.4 molar equivalent of oleic acid chloride, and 67 mg and 65 mg of the ionic liquid [Bmim]Cl. The mixture was microwave-irradiated at 40°C with 20 W power for 1 min. Then, 67 mg of KOH 1.2 mmol and 1 mL toluene was added to the vial and microwave-irradiated at 80°C for 1 min. The reaction mixture was poured into ice-cold water neutralized with 1 N HCl (to pH 7) and extracted three times with 7 mL ethyl acetate. Combined organic phase was washed with NaHCO₃ solution and brine, and dried over Na₂SO₄. The crude product was purified by flash chromatography on a silica gel with n-hexane/ethyl acetate 3:1 giving >80% yield of 2,4,16,16,17-²H-estradiol-17β oleate (E₂ oleate-d₅) by ¹H and ¹³C NMR and mass spectrometry analysis (spectra were in accordance with that of the unlabeled compound (29)).

Preparation of PC vesicles

Phospholipid (PC) (P2772, egg yolk PC, Type XI-E, Sigma-Aldrich, St. Louis, MO) stock solutions were made in chloroform. PC vesicles containing either deuterated free E₂-d₄ or E₂ oleate-d₅ were prepared as follows: egg PC alone (34.5 mg) or containing 1 mg of E₂-d₄ or E₂ oleate-d₅ and 100 nmol of butylhydroxytoluene antioxidant (stock 1 nmol/μL in chloroform) were pipetted into glass tubes on ice. Organic solvent was evaporated under nitrogen until dry, and residual organic solvent was removed by lyophilization for 30 min. Onto the dry lipid film, 2 mL of Tris-buffer (10 mM Tris-HCl, 150 mM NaCl, and 1 mM EDTA, pH 7.4) was added and vortexed vigorously to solubilize lipophilic molecules. The suspension was probe-sonicated for 3 × 5 min (12 ampl.microns, SoniPrep 150, MSE, London, United Kingdom), carefully keeping the tube in the ice bath. After each 5 min sonication, the tubes were left to stand on ice for 1 min, and then sonication was continued, avoiding foaming. After this protocol, the mixture became slightly opaque. The sonicated material was subsequently transferred into Eppendorf tubes (2.5 mL) and centrifuged for 10 min at 15,000 rpm at room temperature to pellet particulate material and titanium residual released from the sonicator probe. After centrifugation, vesicles were dialyzed overnight against phosphate-buffered saline (PBS) at pH 7.4. The opalescent PC vesicles and vesicles with deuterated hormone molecules were transferred to a new 2.5 mL Eppendorf tube and the tube was flushed with nitrogen and kept at 4°C, wrapped in aluminum foil, until use.

Preparation of HDL discs containing E₂-d₄ and E₂ oleate-d₅

Reconstituted HDL discs containing apolipoprotein A-I (apoA-I) as the main apolipoprotein, egg PC, free cholesterol (Sigma-Aldrich) and either deuterated E₂-d₄ or E₂ oleate-d₅ were prepared by the cholate dialysis method, as described (30,31). The resulting reconstituted HDL particles were specifically prepared (molar ratios): 1), apoA-I/PC/chol, 1:250:12.5; 2), apoA-I/PC/chol/E₂-d₄, 1:250:12.5:20; 3), apoA-I/PC/chol/E₂ oleate-d₅, 1:250:12.5:10. In brief, egg PC (10 mg/mL stock in chloroform/methanol, 9:1 v/v), cholesterol (1 mg/mL stock in benzene), E₂-d₄, or E₂ oleate-d₅ were added to a glass vial in the molar ratios given above and the organic solvents were evaporated under nitrogen at room temperature. Afterward, the mixture was lyophilized for 30 min to remove all traces of solvent. After lyophilization, 3 mL of Tris-buffered saline (TBS; 10 mM Tris-HCl, 1 mM EDTA, and 140 mM NaCl, pH 7.4) were added and the mixture was vigorously vortexed. Finally, purified apoA-I was added. Sodium cholate (stock solution 0.725 M in TBS) was then added (55 mM final concentration) and the mixture was gently mixed on vortex (4 × 15 s), avoiding foaming. The mixture was next incubated for 20 min at 24°C in a shaking water bath and then dialyzed against TBS for ∼72 h at 4°C using 3500 cutoff dialysis tubing. After dialysis, the final volume was adjusted to 2.0 mL. Before NMR experiments, the suspensions were dialyzed overnight against a PBS buffer solution containing 50 wt % poly(ethylene glycol) 40,000 to remove excess water.

Preparation of MLVs containing E₂-d₄

For the E₂/POPC mulilamellar vesicles (MLV), mixtures of POPC and either E₂-d₄ (for ²H NMR) or 3,4-¹³C-E₂ (for ¹H-¹³C dipolar coupling and chemical shift (DIPSHIFT) experiments) were codissolved in chloroform/methanol at molar ratio 4:1. After evaporating the solvent, the samples were redissolved in cyclohexane and lyophilized at high vacuum to obtain a fluffy powder. After hydration with 40 wt % H₂O, the samples were equilibrated by freeze-thaw cycles and gentle centrifugation.

Solid-state NMR measurements

The ²H NMR spectra were measured on a Bruker Avance 750 MHz NMR spectrometer (Bruker Biospin, Rheinstetten, Germany) operating at a resonance frequency of 115.1 MHz for ²H. A single-channel probe equipped with a 5 mm solenoid coil was used. The typical length of the 90° pulses was 3 μs in a phase-cycled quadrupolar echo sequence (32). A relaxation delay of 0.75 s was applied.

The ¹³C magic-angle-spinning (MAS) NMR experiments were conducted on a Bruker Avance III 600 MHz spectrometer (resonance frequency 600.1 MHz for ¹H and 150.9 MHz for ¹³C) using a 4 mm double-resonance MAS probe. The cross-polarization contact time was 700 μs, typical lengths of the 90° pulses were 5 μs for ¹³C and 4 μs for ¹H. For heteronuclear two-pulse phase modulation decoupling, a ¹H radio-frequency field of 65 kHz was applied. ¹³C chemical shifts were referenced externally relative to trimethylsilane. To measure ¹H-¹³C dipolar couplings, constant-time DIPSHIFT experiments using frequency-switched Lee Goldberg for homonuclear decoupling (80 kHz decoupling field) were carried out (33). The order parameter was derived by dividing the determined coupling by the known rigid limit, after simulating the dephasing curve over one rotor period. Reference values for the rigid limit ¹H-¹³C dipolar couplings were obtained from the literature (34). Ultrafast dynamical modes leading to alterations of the C-H bond lengths, which may result in an ∼10% variation of the rigid-limit dipolar coupling, were not considered here (35). All MAS experiments were carried out at a MAS frequency of 2.5 kHz. All NMR experiments were conducted at a temperature of 30°C.

MD simulation

Two all-atom MD simulations were carried out: one reference simulation of a pure POPC membrane and one simulation of a POPC membrane in the presence of E₂. The first simulation consisted of 50 POPC and 1372 H₂O molecules and was run for 200 ns. Setup of the membrane was conducted according to published procedures (36). The second simulation consisted of 50 POPC, 10 E₂, and 1800 H₂O molecules and was run for 1582 ns. For the setup of this simulation, the final coordinates of the pure POPC membrane simulation were used and 10 E₂ molecules added to the membrane. Particular attention was paid to avoid any molecule becoming threaded through one of the E₂ rings. All E2s were spaced evenly within the membrane and oriented with their long axis parallel to the membrane normal so that O-3 faced the lipid/water interface whereas O-17 was close to the terminal methyl groups of the POPC chains. A temperature of 30°C was used for both simulations. The program NAMD (37) was employed for the simulation under conditions of normal pressure (1.013 bar), using the CHARMM all-H CMAP protein force field (38,39) with the most recent all-H C36 lipid force field (40), allowing simulation with flexible surface area and therefore adaptation of the system to the presence of the E₂ (no additional force was applied in the plane of the membrane). The E₂ force field values were based on previous reports (41). The smooth particle-mesh Ewald algorithm was used to compute the electrostatic forces (42) and the SHAKE algorithm was used to keep rigid all bonds involving hydrogen atoms, allowing a 2 fs time step (43).

Spartan quantum mechanical calculations

A quantum mechanical model of E₂ was built in Spartan’04. The equilibrium structure in vacuum was calculated using the Hartree-Fock 3-21G model. Further, an electron density surface with an isovalue of 0.002 electrons/au³ was calculated, and the strength of the electrostatic potential mapped on this surface.

Results

²H NMR investigations of E₂ and E₂ oleate

We acquired ²H NMR spectra of E₂-d₄ in PC vesicles and HDL discs and, for comparison, also in POPC MLVs. NMR spectra are shown in Fig. 2, A–C. The ²H NMR spectra consist of a superposition of Pake doublets where each ²H label corresponds to one doublet. Each doublet has a typical symmetrical lineshape: the Pake pattern in which two distinct maxima exist. The distance of these maxima is called the quadrupolar splitting. Note that the low hydration level of 40 wt % prevents HDL discs from overall tumbling and thus, NMR powder spectra are obtained. The best signal/noise ratio was achieved for the reference sample of E₂-d₄ in POPC MLVs, but the overall shape of the spectra of E₂-d₄ in PC vesicles and HDL discs is very similar. All spectra could be explained by the superposition of four individual Pake doublets of identical area and a small isotropic component, which is also visible in all three NMR spectra. The extracted quadrupolar splittings were very similar in all samples and showed values of 2.6, 17.3, 18.5, and 23.4 kHz for E₂-d₄ in POPC MLVs. The small isotropic contribution most likely originated from a small fraction of highly mobile structures.

Figure 2.

Experimental ²H solid-state NMR spectra of E₂-d₄/PC vesicles (A), E₂-d₄/HDL discs (B), E₂-d₄/POPC MLV (C), E₂ oleate-d₅/PC vesicles (D), and E₂ oleate-d₅/HDL discs (E) are shown in black. All spectra were acquired at a temperature of 30°C. The best simulation of these spectra obtained from our model of the orientation and dynamics of E2/E2 oleate are shown in red. All samples were investigated at a hydration level of 40 wt %, which abolishes the overall tumbling of the HDL discs. To see this figure in color, go online.

For E₂ oleate-d₅, the NMR spectra looked different (Fig. 2, D and E). However, the NMR spectra of E₂ oleate in PC vesicles and HDL discs are again similar. Both spectra could be explained by the superposition of five Pake spectra. Again, the extracted quadrupolar splittings were very similar between PC vesicles and HDL discs and had values of 2.5, 6.2, 14.2, 16.1, and 34.5 kHz for E₂ oleate-d₅ in PC vesicles. This indicates differences in the molecular orientation and/or fluctuations between E₂ oleate and E₂, whereas the bilayer environment of these molecules seemed to play a minor role.

Analysis of the E₂²H NMR spectra

To analyze the ²H NMR spectra, it is crucial to know the static quadrupolar coupling constant to extract correct information about the orientation and motion of E₂ that scale the quadrupolar splitting. In aliphatic carbon deuterium bonds, this constant has a value of 167 kHz, which results in a full quadrupolar splitting of a C-D bond in the rigid limit of 125.25 kHz (44). However, it is known that the quadrupolar coupling constant is slightly higher for aromatic carbons (45). Therefore, we measured a ²H NMR spectrum of the E₂-d₄ powder at −27°C, which provided the full rigid quadrupolar splitting of the ring ²H (Fig. S1 in the Supporting Material). The ²H NMR spectrum was fitted using four Pake doublets of identical intensity. Two showed a quadrupolar coupling constant of 167 kHz (for the aliphatic deuterons attached to C-16) and two gave rise to a slightly larger quadrupolar coupling constant of 181.2 kHz for the aromatic deuterons at C-2 and C-4, in good agreement with the literature (45).

In the ²H NMR spectra of E₂ or E₂ oleate, such large quadrupolar splittings were clearly not observed, and the narrowing of the quadrupolar couplings could be the result of a specific orientation, or distribution of orientations, of E₂ in the membrane. This orientation is described by two angles: the pitch angle, β_IM, and the roll angle, α_IM, as depicted in Fig. 1. Considering the orientation of the C-D bonds with respect to the tetracyclic ring system, these two angles can be used to calculate the quadrupolar splitting of each deuterium nucleus under the assumptions of a rigid molecule with a single fixed orientation, i.e., no fluctuations in the orientation with respect to the membrane. Fitting the four (for E₂-d₄) or five (for E₂ oleate-d₅) values (one for each deuteron) to the experimental quadrupolar splitting, we could reproduce all ²H NMR spectra within this rigid model. The fit allowed for all possible assignments of the individual Pake doublets to the different deuterium positions, since those were not known. Best fits for the orientation of E₂ in POPC MLVs were obtained for a pitch angle of β_IM = 15.3° and a roll angle of α_IM = 220.0°. For E₂ oleate in PC vesicles, the best fit was obtained for a pitch angle of β_IM = 88.7° and a roll angle of α_IM = 53.4°. For both molecules, the angles for the other membrane environments were very similar.

Although this fixed molecular orientation of E₂ in the lipid environment reproduces the experimental ²H NMR spectra, it is doubtful that the E₂ would adopt a specific and rigid orientation in the lipid membrane, which is highly fluid and characterized by large-amplitude motions of the lipids that are interchanging among a multitude of lipid conformations (46,47). For instance, typical order parameters are 0.03–0.25 for the phospholipid chains (48) and 0.1–0.4 for cholesterol (49). It is likely that dynamical averaging resulting from molecular fluctuations also affects the spectral shape of the NMR spectra of E₂ due to averaging of the EFG tensor. Since ²H NMR quadrupolar splittings are sensitive to both orientation and dynamics of the labeled molecule, the effects of the two cannot be separated in the ²H experiments. Therefore, we conducted additional ¹³C MAS NMR experiments employing 3,4-¹³C-E₂ in POPC MLVs (Fig. S2). The effect of motional averaging on the ¹H-¹³C dipolar coupling on C-4 provides a measure of the dynamics of E₂ irrespective of the mean orientation, allowing the separation of orientation and dynamics. From the DIPSHIFT dephasing curve, a molecular order parameter of 0.25 ± 0.03 for the ¹H-¹³C-4 bond was determined. This low order parameter suggests that a significant portion of the narrowing of the ²H NMR spectra of E₂-d₄ is caused by molecular fluctuations. Further, if it is assumed that the tetracyclic ring system of E₂ moves as a whole unit, this value indicates that E₂ undergoes large-amplitude motions. Therefore, we began to develop models that considered simultaneously the orientational fluctuations of E₂.

It should be stressed again that the order parameters measured in static ²H NMR experiments contain the effects of both molecular dynamics and orientation of the ²H-C bond with respect to the external magnetic field, whereas the ¹H-¹³C order parameter, measured for the same molecular segment in DIPSHIFT experiments under MAS conditions solely reports the molecular dynamics of the bond vector and does not contain geometrical effects (50).

In these models, several assumptions had to be made to keep the number of fitting parameters feasible. The simplest assumption to explain the low DIPSHIFT order parameter is motional averaging due to an axially symmetric rotation about the long axis of the molecule, such that all roll angles α_IM are equally probable. Unfortunately, simple uniform rotation about a long axis of fixed orientation is inconsistent with the experimental ²H NMR spectra. In such a model, the ratio of the quadrupolar splittings from the four E₂ sites would be fixed with respect to each other and only scaling of the overall width of the ²H NMR spectra would occur. However, we observed quadrupolar splittings that did not follow such a ratio for any tilt angle. Therefore, a simple uniform rotation of the E₂ around its molecular axis cannot describe the orientation and dynamics of the molecule. Although more elaborate models, including, for example, hindered rotation and wobbling of the long axis, could be constructed, we instead chose to first carry out MD simulations to provide additional insight into the molecular dynamics of E₂ without assuming any particular model.

MD simulations of E₂ in POPC

For the MD simulation, we chose to model the E₂/POPC mixture, since it had a clearly defined molecular composition and its ²H NMR spectrum had the best signal/noise ratio. Furthermore, this spectrum was very similar to the spectra of E₂ in PC vesicles and HDL discs, indicating that E₂ behaves similarly in all three environments. In the starting configuration, the E₂ molecules were spaced evenly within the membrane and oriented parallel to the membrane normal, with the O-3 oxygen atom close to the lipid/water interface and O-17 close to the terminal methyl groups of the POPC chains (Fig. 3 A). This orientation was chosen because the structurally related cholesterol exhibits the same orientation (26), and previous NMR data showed that H1, H2, and H4 of E₂ are localized in the vicinity of the upper chain segments of POPC (51). However, after 100 ns of simulation, several E₂ molecules had already changed their orientation dramatically, e.g., flipped the orientation of the long axis with respect to the membrane, and at the end of the simulation, each E₂ had changed its orientation multiple times (Fig. 3 B). All E₂s completely reversed their orientation at least four times (overall, 89 events were observed where β_IM changed from ∼0° to ∼180° and vice versa). Furthermore, a range of long-axis orientations were sampled. Not only were orientations parallel to the membrane normal stable, but E₂ molecules often were found in ∼90° orientations such that they were parallel to the membrane surface.

Figure 3.

Snapshots taken at the beginning of the MD simulation (A) and after 1 μs simulation time (B) are shown. The membrane consisted of 10 E₂ and 50 POPC molecules and was simulated at a temperature of 30°C. All molecules are shown in a stick representation, where POPC is transparent. In addition O-3 and O-17 are shown as van der Waals spheres and colored red and blue, respectively. To see this figure in color, go online.

Assuming that the MD simulation samples the potential configurations of E₂ with the correct probability distribution, the ²H NMR spectra and DIPSHIFT order parameters of the molecule can be calculated from the trajectory and directly compared to the experiments. From the simulated order parameters, quadrupolar splittings were calculated and used to simulate a ²H NMR spectrum of E₂, which is shown in Fig. 4 A compared to the experimental spectrum. Overall, the agreement is poor. This indicates that the distribution of orientations for the labeled E2 C-H bonds observed in the simulations does not faithfully represent the real system. There are many possible reasons for such a discrepancy, such as shortcomings in the relatively simple force field employed. For example, the internal structure of E₂ could have deviations such that even when the correct molecular orientation is achieved, the C-H vectors are in error, or there could exist subtle errors in the lipid-sterol-water interactions. Beyond issues related to the force field, likely the greatest challenge facing atomistic MD simulation is obtaining sufficiently long trajectories that the equilibrium distribution of states, E₂ orientations in this case, is achieved. The calculation of NMR observables, which depends on the entire distribution of conformers rather than simply the mean value, is especially sensitive to inadequate sampling. Fortunately, by having a data set that includes both DIPSHIFT order parameters, which report on fluctuations with respect to the ¹H-¹³C bond, and ²H quadrupolar splittings, which are sensitive to orientational fluctuations with respect to the membrane normal, we are able to shed light on the discrepancy between the experimental and simulated spectra. Specifically, the DIPSHIFT order parameters, extracted from the simulation using published procedures (50) and shown in Fig. 4 B, are uniformly below the experimental value, and the average of all order parameters is approximately one third of the experimental value. This suggests that the magnitude of fluctuations in the molecular frame is too large, potentially from insufficiently strong interactions with the membrane, i.e., a force field error, or from a failure to reach the correct equilibrium distribution of conformers, i.e., a shortfall in sampling. Turning to the ²H NMR spectrum (Fig. 4 A), all quadrupolar splittings extracted from the simulation are larger than the smallest splitting observed experimentally. Therefore, at least one quadrupolar splitting in the MD simulation is too large. Splittings that are too large can be explained by either insufficiently large fluctuations or incorrect average orientations. Based on analysis of the DIPSHIFT order parameters, which show that the fluctuations in the simulation are actually too large, we can reject the former and conclude that the simulation fails to sample with correct probability at least some of the orientations of the E₂ with respect to the membrane bilayer.

Figure 4.

(A) Calculated ²H NMR spectrum of E₂-d₄ in POPC MLV determined from the orientation and dynamics of the molecule according to the MD simulation. Orientation and order parameters were extracted from the MD trajectory and used to calculate the resulting ²H NMR spectrum (red line), which is plotted with the experimental spectrum (black line). For the simulation of the spectrum it was assumed that all four Pake doublets have the same intensity and line width and their chemical shifts were the same as observed experimentally. (B) DIPSHIFT order parameter of each E₂ molecule in the simulation, as well as their average (dashed line) in comparison to the experimental value (dotted line). To see this figure in color, go online.

Furthermore, ²H NMR order parameters for the palmitoyl chain of POPC were calculated from the MD simulation and compared to a reference simulation of POPC without E₂ (Fig. S3). The obtained order parameters for the pure POPC bilayer agree well with experimental literature data acquired at 27°C and 30°C, respectively (51,52). However, a considerable increase in order due to the presence of E₂ is observed over the whole chain, and the mean order parameter is increased by 28.6%. This results in only a 4.8% increase in the area of the simulation cell despite the addition of 10 E₂s. Therefore, in the MD simulation, E₂ exhibits a condensation effect on POPC. This is similar to the effect of cholesterol, which is known to increase the mean order parameter of POPC at 30°C by 30.3% (15 mol % cholesterol) and 38.8% (20 mol % cholesterol) (53). However, experimental data for E₂ showed only a very small increase in order due to the presence of 20 mol % E₂ (51) and, therefore, virtually no condensation effect (54). In summary, the MD simulation failed to reproduce the experimental observables. Comparison of the sterol order parameters suggests that the distribution of molecular orientations is not correct. This is perhaps not surprising given that multiple orientations were observed during the course of the simulation, whereas all sterols were initially set to a single (cholesterol-like) orientation. It is thus unlikely that an equilibrium distribution can be obtained, even from microsecond-scale simulations, when the lifetime of conformational states is on the order of 100–200 ns. Furthermore, the lipid order parameters, which indicate a significant condensation effect, suggest that the simulation oversampled the cholesterol-like initial orientation.

Analysis of the E₂²H NMR spectra and DIPSHIFT order parameter using an analytical model

Given the failure of the simple, i.e., single-orientation, analytical models and the detailed MD simulation to reproduce the experimental data, a more complex model based on some observations from the MD simulation was developed. In general, the quadrupolar splitting Δν_Q for a ²H labeled position is expressed by

$$\text{Δ}{\nu }_{\text{Q}}=\frac{3}{2}{\text{χ}}_{\text{Q}}\langle D_{00}^{\left(2\right)}\left({\text{Ω}}_{\text{PL}}\left(t\right)\right)\rangle ,$$ (1)

where χ_Q is the quadrupolar coupling constant, which has values of 167 kHz and 181.2 kHz for the aliphatic (C-16) and aromatic (C-2 and C-4) deuterons, respectively. The element $D_{00}^{\left(2\right)}\left({\text{Ω}}_{\text{PL}}\left(t\right)\right)$ is the Wigner rotation matrix that describes the coordinate transformation of the electric field gradient (EFG) of the C-²H bond from its principal axis system (P) to the laboratory frame (L). The Euler angles α_PL, β_PL, and γ_PL that describe the rotations of this coordinate transformation are summarized in the angle Ω_PL(t). The angular brackets indicate a time average of the coordinate transformation, since it is time-dependent if the EFG, which is attached to the E₂ molecule, undergoes motion with respect to the laboratory frame.

To facilitate the development of a model, it is often useful to introduce intermediate coordinate transformations into the PL coordinate transformation (55). One can depict this as doing one coordinate transformation from the principal axis system (P) to a temporary coordinate system (T) and a subsequent second coordinate transformation from the temporary coordinate system (T) to the laboratory frame (L):

$$D_{sm}^{\left(2\right)}\left({\text{Ω}}_{\text{PL}}\left(t\right)\right)=\sum _{r=-2}^2{D}_{sr}^{\left(2\right)}\left({\text{Ω}}_{\text{PT}}\left(t\right)\right)D_{rm}^{\left(2\right)}\left({\text{Ω}}_{\text{TL}}\left(t\right)\right).$$ (2)

The development of the model is simplified considerably by careful choice of the temporary coordinate systems. In the case of E₂, two intermediate coordinate systems are useful. The first is an internal coordinate system (I), which is attached to the molecule (see Fig. 1 for an exact definition). The second temporary coordinate system is the membrane coordinate system (M) that is aligned along the membrane normal. This allows Eq. 1 to be rewritten as (55)

$$\text{Δ}{\nu }_{\text{Q}}=\frac{3}{2}{\chi }_{\text{Q}}\sum _{s=-2}^2\sum _{r=-2}^2{D}_{0s}^{\left(2\right)}\left({\text{Ω}}_{\text{PI}}\right)\langle D_{sr}^{\left(2\right)}\left({\text{Ω}}_{\text{IM}}\left(t\right)\right)\rangle D_{r0}^{\left(2\right)}\left({\text{Ω}}_{\text{ML}}\right).$$ (3)

The first coordinate transformation (PI) can now be calculated using the known orientation of the C-²H bond within the molecule and is time-independent, assuming that there are no internal motions of the molecule. The second coordinate transformation (IM) describes the orientation and dynamics of the molecule in the membrane and is therefore time-dependent. The model of the orientation and dynamics of E₂ will be used to calculate this transformation. The third coordinate transformation (ML) is time-independent again as it describes the known orientation of the membrane normal within the magnetic field (powder distribution). Assuming that the membrane is axially symmetric around its normal, the second summation in Eq. 3 vanishes and the quadrupolar splitting, Δν_Q, simplifies to

$$\text{Δ}{\nu }_{\text{Q}}=\frac{3}{2}{\chi }_{\text{Q}}\sum _{s=-2}^2{D}_{0s}^{\left(2\right)}\left({\text{Ω}}_{\text{PI}}\right)\langle D_{s0}^{\left(2\right)}\left({\text{Ω}}_{\text{IM}}\left(t\right)\right)\rangle D_{00}^{\left(2\right)}\left({\text{Ω}}_{\text{ML}}\right).$$ (4)

In this expression, the first and last coordinate transformations are easy to calculate, since the angles Ω_PI and Ω_ML are known from the geometry of the system. The most interesting coordinate transformation is the second, for which a model had to be developed. This model describes the probability of each pair of angles (α_IM,β_IM), (p(α_IM,β_IM)) depending on a set of fitting parameters, as explained in the next section. All Δν_Q were then weighted with this probability and summed over all α_IM and β_IM such that the final result of the model for one ²H position is

$$\overline{\text{Δ}{\nu }_{\text{Q}}}=\frac{3}{2}{\chi }_{\text{Q}}\underset{0}{\overset{2\pi }{\int }}\underset{0}{\overset{\pi }{\int }}p\left({\alpha }_{\text{IM}},{\beta }_{\text{IM}}\right)\sum _{s=-2}^2{D}_{0s}^{\left(2\right)}\left({\text{Ω}}_{\text{PI}}\right)D_{s0}^{\left(2\right)}\left({\text{Ω}}_{\text{IM}}\right)D_{00}^{\left(2\right)}\left({\text{Ω}}_{\text{ML}}\right)d{\beta }_{\text{IM}}d{\alpha }_{\text{IM}}.$$ (5)

In this equation, the time average of the second coordinate transformation (IM) is replaced by an ensemble average, assuming that E₂ completely samples all possible (α_IM,β_IM) combinations during the quadrupolar interaction time, which should be very reasonable considering the low DIPSHIFT order parameter that was measured. The resulting averaged $\overline{\text{Δ}{\nu }_{\text{Q}}}$ is compared with the experimental result and their difference is minimized by variation of the fitting parameters that modify p(α_IM,β_IM).

The probability p(α_IM,β_IM) is calculated from the Boltzmann distribution. For this, an energy function, E, is needed. Therefore, potential functions of the roll angle, α_IM, and the pitch angle, β_IM, were created that describe the orientation and dynamics of E₂ in the membrane. The definition of the potentials was based on the observations made in the MD simulation. In Fig. S4 A, the histogram of the pitch angle, β_IM, as observed in the MD simulation is shown. Three major orientations of β_IM are observed: 0°, 90°, and 180°. Therefore, the potential for the angle β_IM was chosen such that this shape could be reproduced (Fig. S4 A, red line):

$$E_{\beta }\left({\beta }_{\text{IM}}\right)=-I_{\beta }\left(r_{\beta 1}\cos \left(4{\beta }_{\text{IM}}\right)+r_{\beta 2}\cos \left(2{\beta }_{\text{IM}}\right)+\left(1-r_{\beta 1}-r_{\beta 2}\right)\cos \left({\beta }_{\text{IM}}\right)\right).$$ (6)

Fit parameters of this equation are I_β, r_β1, and r_β2. In contrast, the histogram of the roll angle, α_IM, as observed in the MD simulation shows only two major orientations, 0° and 180° (Fig. S4 C). Therefore, the potential for α_IM was created differently to allow reproduction of this shape (Fig. S4 C, red line):

$$E\alpha \left({\alpha }_{\text{IM}}\right)=-I_{\alpha }\left(r_{\alpha }\cos \left(2({\alpha }_{\text{IM}}+{\alpha }_{\text{offset}})\right)+\left(1-r\alpha \right)\cos ({\alpha }_{\text{IM}}+{\alpha }_{\text{offset}})\right).$$ (7)

Fit parameters of this equation are I_α, r_α, and α_offset. Further, several additional effects have to be considered due to the axial symmetry around the membrane normal. First, E₂ orientations with β_IM close to 0° or 180° are less likely than 90° orientations, and therefore, the whole probability distribution has to be scaled by sin(β_IM). Further, the intensity of the potential for α_IM depends on β_IM. If β_IM is 90°, a change of α_IM might expose certain atoms that were embedded in the membrane to water and therefore change their energy. However, if β_IM is 0° (Fig. S4 B) or 180° (Fig. S4 D), all angles α_IM should be equivalent due to the axial symmetry, as is observed in the MD simulations. Therefore, the depth of the potential for α_IM is further scaled with the factor sin(β_IM). The resulting 2D potential function for each (α_IM,β_IM) pair was fitted to the distribution observed in the MD simulations (Fig. S5 A). The obtained fit (Fig. S5 B) showed that the created potential is sufficiently flexible to reproduce the main features of the distribution observed in the MD simulation with a reasonable number of fitting parameters (summarized in Table 1). It is important to note that in the following, the probability distribution employed was motivated by the MD simulation results but was determined independently by fitting to the experimental data, i.e., we assumed that the MD simulation identified the energetically favorable orientations but did not converge to the equilibrium probabilities.

Table 1.

Summary of the fitting parameters

	Iα × 1020/J	rα	αoffset/°	Iβ × 1020/J	rβ1	rβ2
Fit of results from MD simulation	0.967	0.735	−6.869	1.083	0.133	0.548
E2/PC vesicles	3.021	0.463	22.273	0.099	−0.343	−0.005
E2/HDL discs	1.412	0.452	23.897	0.099	0.416	0.197
E2/POPC MLV	1.288	0.535	24.760	0.130	0.412	0.038
E2 oleate/PC vesicles	0.509	0.226	−2.021	2.684	0.008	−0.598
E2 oleate/HDL discs	0.511	0.201	−0.604	2.491	0.008	−0.590

Fitting parameters for all samples as well as from the MD simulation were obtained using Eqs. 6 and 7. Parameters I_α and I_β describe the overall intensity of the potentials for α_IM and β_IM, whereas r_α, r_β1, and r_β2 describe the ratio of the individual cosines that constitute the potential functions. The parameter α_offset describes how far the 0° maximum of the overall potential for α_IM deviates from the ${\overrightarrow{y}}_{\mathrm{int}}$ axis of the internal coordinate system. The data from the MD simulation were determined by fitting Eqs. 6 and 7 to the histograms of α_IM and β_IM obtained from the simulation.

After all these factors have been considered, the averaged $\overline{\text{Δ}{\nu }_{\text{Q}}}$ can be calculated for a given set of fit parameters. Further, the DIPSHIFT order parameter is calculated for all possible (α_IM, β_IM) combinations, which are also weighted with the probability p(α_IM,β_IM) from the model. For the definition of DIPSHIFT order parameters, the reader is referred to deAzevedo et al. (56), and a detailed description of their calculation from a set of given coordinates is given in Vogel et al. (50). Here, it is important to note that DIPSHIFT order parameters are defined differently from ²H NMR order parameters. Whereas ²H NMR order parameters are determined with respect to the membrane normal, DIPSHIFT order parameters are determined with respect to the ¹H-¹³C bond orientation. Therefore, ²H NMR order parameters contain information on the orientation of the molecule within the membrane and its dynamics, whereas DIPSHIFT order parameters contain information on molecule dynamics only. We exploit this difference and use the DIPSHIFT order parameter to discriminate between the two ways in which ²H NMR lineshapes can be narrowed.

By fitting this model to the experimental DIPSHIFT order parameter and to the quadrupolar splittings, very good fits of all spectra could be obtained, as shown, overlaid with the experimental spectra, in Fig. 2. The resulting fit parameters of all samples are summarized in Table 1. For the two samples of E₂ and E₂ oleate with the best signal/noise ratio in the ²H NMR spectra, the shapes of the potentials for α_IM and β_IM and their correlation in a 2D representation are shown in Fig. 5. The same plots for the other samples are shown in Fig. S6.

Figure 5.

Dependence of probability function for orientation of E₂ in bilayers on pitch angle, β_IM, and roll angle, α_IM. From the fits of the final model (Eqs. 6 and 7), the resulting histograms for α_IM and β_IM were calculated. The probability of each (α_IM, β_IM) pair is shown in a 2D plot, where the color changes from blue to red with increasing probability of the corresponding pair. Data are shown for E₂-d₄/POPC MLV (A) and E₂ oleate-d₅/PC vesicles (B). The corresponding plots for the other samples are shown in Fig. S6. Also shown are the 1D histograms for β_IM (upper graphs) and α_IM at an angle of β_IM = 90 ± 15° (right graphs). Please note that the histogram for α_IM depends on β_IM and would be a straight line for β_IM = 0° or β_IM = 180° (see Fig. S4). To see this figure in color, go online.

Fig. 5 A shows the results for E₂ in POPC MLVs, which deviate considerably from the distributions observed in the MD simulation. For β_IM, all three possible conformations—0°, 90°, and 180°—exist, but the 90° orientation is the most abundant. For α_IM, the 0° orientation is the most often observed, whereas the 180° orientation is unlikely. The results for E₂ oleate in PC vesicles differ, as shown in Fig. 5 B. For β_IM a single ∼60° orientation is observed almost exclusively. For α_IM, however, the distribution becomes very wide, and although the 0° orientation is the most abundant, other orientations are also observed in considerable amounts. Since the spectra in the other membrane environments were very similar, the resulting fits have the same general appearance for E₂ and E₂ oleate, respectively (Fig. S6).

We also tested whether the results vary when we add the effect of internal mobility of E₂ to our model. With this internal mobility, the result of the fit was very similar to the model without internal mobility and the internal order parameter was ∼0.9. Therefore, we decided to use the simpler model without internal mobility of E₂ since it contains one less fitting parameter and gives very similar results.

The assignment of the four Pake doublets in the experimental spectra to the four deuterations in E₂ also was a result of our final model. For the E₂-d₄/POPC MLV sample, the quadrupolar splitting of the deuteration on C-4 was 18.5 kHz, which corresponds to a ²H NMR order parameter of 0.148. This is significantly lower than the DIPSHIFT order parameter of 0.25 measured for the same bond. Since both types of order parameters are sensitive to dynamics, this difference originates from the mean orientation of this bond, to which only the ²H NMR order parameter is sensitive.

Discussion

We investigated the orientation and dynamics of E₂ and E₂ oleate in several lipid membrane environments and HDL discs. ²H NMR spectra of deuterated E₂ and E₂ oleate were recorded for all samples and a ¹H-¹³C order parameter was measured for E₂ in POPC MLVs. The results were fitted to a model of the orientation and fluctuations of E₂ and E₂ oleate developed on the basis of an all-atom MD simulation that provided information about the dynamic reorientation of the molecule in a phospholipid bilayer.

Although the experimentally detected ²H NMR spectra of E₂ could be reproduced by a fixed orientation of the molecule in the membrane with a pitch angle of β_IM = 15.3° and a roll angle of α_IM = 220.0°, our analysis suggests that this model gives an inaccurate picture, and that in reality E₂ is distributed in the membrane in a highly dynamic fashion. This was confirmed by a relatively low ¹H-¹³C order parameter of 0.25 and by an ∼1.5 μs all-atom MD simulation that showed E₂ to undergo significant fluctuations in tilt and roll angles. Our results for E₂ are in sharp contrast to those for other sterols like cholesterol, which is known to wobble around an upright orientation (near 0°) in the membrane (26). Our model predicts that the long axis of E₂ can assume a parallel (0° orientation), perpendicular (90° orientation), or antiparallel (180° orientation) relative to the bilayer membrane normal. However, the perpendicular orientation, where E₂ is located in the membrane interface, is the most populated. Furthermore, when E₂ assumes this tilt-angle orientation, it strongly favors a particular roll angle, facing the membrane with carbons 6, 7, 15, and 16, whereas carbons 1, 2, 11, and 12 point toward aqueous phase (Fig. 1 A). This result is in agreement with previous data, where the hydrogen attached to position 4 is more deeply embedded in the membrane than the hydrogens attached to positions 1 and 2 of the molecule (51). This agrees with experimental observations that incorporation of E₂ into HDL is limited due to the hydrophilic nature of the molecule (18).

It was observed that E₂, in contrast to the related cholesterol, exerts no condensation effect on lipids (51). These observations can be understood in terms of the high fraction of perpendicular orientations we suggest for E₂. Since E₂ is only rarely deeply embedded among the lipid acyl chains, it cannot condense lipids. In contrast to cholesterol, it is mostly located in the lipid-water interface, where it occupies space and restricts the mobility of the upper lipid acyl chain segments, resulting in a slightly increased order parameter (51). This is confirmed by studies of cholesterol analogs that show that even putatively minor alterations of the sterol structure have dramatic effects on the orientation in the membrane and its ability to induce phospholipid chain condensation (53,58).

The molecular origin of such an orientation of E₂ is most likely the two polar OH groups that are attached to positions 3 and 17. These hydroxyl groups can form favorable hydrogen bonds with lipid phosphate and carbonyl groups in the lipid-water interface of the membrane. In contrast, cholesterol only features a single OH group at carbon 3 while a hydrophobic acyl chain is attached to carbon 17, which keeps the molecule in its upright position. Since in E₂, the hydrophobic acyl chain is replaced by a polar OH group, this orientation is unlikely, because the energy cost for its transfer into the membrane center is high. Instead, both OH groups prefer to reside close to the lipid-water interface. For the roll angle, α_IM, the preferred orientations are significantly influenced by the flat shape of E₂. Any orientation in which the plane of the E₂ rings is parallel to the membrane surface is relatively unlikely, since this would require more interfacial area within the membrane; thus, two orientations are preferred. In addition, the region of carbons 6, 7, 15, and 16 is more apolar than the remaining molecule, as implied by a quantum chemical calculation of the surface potential (Fig. 6). Consequently, E₂ is mostly oriented in such a fashion that these positions face the membrane interior.

Figure 6.

Electrostatic surface potential of E₂. Quantum chemical calculations were used to obtain the magnitude of the electrostatic potential on the surface of the E₂ molecule. The color scale reaches from red (−146.44 kJ/mol) over green (0 kJ/mol) to blue (+146.44 kJ/mol). The calculations and the graphic were carried out in Spartan’04. To see this figure in color, go online.

For E₂ oleate, this mix of interactions is altered because of the lipid chain attached to carbon 17. Our analysis shows that the long axis of E₂ oleate assumes a single orientation of ∼60°. Further, for the roll angle, α_IM, orientations are now observed in which carbons 6, 7, 15, and 16 do not face the membrane interior (Fig. 5 B). The reason for this altered behavior is the added oleoyl chain, which now adds a second hydrophobic moiety to the molecule, enabling additional possible orientations of the molecule. It is likely that the hydrophobic oleoyl chain is inserted into the bilayer interior (18).

Our MD results provide important examples of both the possibilities and the shortcomings of the method. The challenge of generating an equilibrium distribution from initial conditions based on intuition derived from a related system was clearly shown. In the starting configuration, all E₂ were oriented parallel to the membrane normal in the same fashion as cholesterol and were thus very far from an equilibrium state, which turned out to be mostly the 90° orientation of the molecule. Although the simulation failed to provide a quantitatively accurate description of the system (e.g., a direct comparison of observables calculated from the MD simulation with the experimental values showed large differences), the simulation nonetheless was valuable in that it provided a qualitative picture of the accessible conformations. Thus, the MD simulation enhanced our understanding of the behavior of E₂, since orientations that correspond to our final model showed up very quickly and were critically important in constructing the detailed analytical model that successfully described the system. The skewed relative probability of the different orientations and the many transitions between them led to the lower order parameters of E₂ observed in the simulation. Furthermore, the overrepresentation of the 0° and 180° orientations impacted the membrane structure as well, producing a moderate condensation effect of POPC that was not observed experimentally (51). In summary, the MD simulation was able to populate the important orientations of E₂ within a lipid bilayer but oversampled the parallel orientations due to the biasing effect of the initial conditions, even after >1.5 μs of simulation.

In this study, we have determined the orientations of E₂ and E₂ oleate in lipid membranes and HDL discs. Taken together, E₂ has a preference for an interface localization in the membrane with the long axis of the molecule perpendicular to the membrane normal. Interestingly, this orientation was also found in HDL discs, suggesting that only lipid-E₂ interactions determine the localization of the molecule and that contacts to the apoA-I protein are not relevant for E₂ orientation. Such interface localization appears to be a common structural motif for many aromatic compounds, and in this case it is stabilized by hydrogen bonds, cation-π interactions, and dipole-dipole interactions (59–61). The attachment of the oleic acid chain alters the distribution of the roll angle, but the general interface localization of the molecule is retained. These data provide to our knowledge new biophysical evidence about how E₂ and its storage form, E₂ fatty acyl esters, orient into lipid membranes and HDL particles. Since E₂ is physiologically a potent hormone, pharmacological applications incorporating E₂ into lipid membranes and HDL could be developed to transfer active hormones or their derivatives to their target tissues.