INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

The binding of ethanol molecules to the lipid matrix of biomembranes is an important, and still not well understood, event in anesthesia. For example, Mitchell and Litman (2000) recently reported that under physiologically relevant conditions, ~90% of the effect of ethanol on the G-protein coupled membrane receptor rhodopsin appears to be the result of changes in the acyl chain packing of the lipid matrix. In contrast, other work has emphasized the direct interaction of ethanol with membrane proteins (Mihic et al., 1997). Perhaps ethanol acts at multiple sites with variable emphasis on indirect interaction via the lipid matrix or direct interaction with the protein, depending on the specific protein system involved. Ultimately, approaches that resolve ethanol interaction with the constituents of biomembranes at atomic resolution are required to provide answers about the location of ethanol, the degree of immobilization of bound ethanol molecules, and the functional perturbation that ethanol causes in the lipid matrix and in proteins.

We have shown previously (Holte and Gawrisch, 1997) that the location of ethanol in the lipid matrix of biomembranes can be studied by two-dimensional magic angle spinning (MAS) nuclear Overhauser enhancement spectroscopy (NOESY). This method is sensitive to interactions on the angstrom length scale and to motions on the pico- to microsecond time scale, thus providing data with extremely high temporal and spatial resolution. Since our first paper appeared, the processes of spectral acquisition and data analysis have been greatly refined (Huster and Gawrisch, 1999; Huster et al., 1999; Yau and Gawrisch, 2000), and the possibility of comparison with molecular dynamics computer simulations has developed (Feller et al., 1999).

Recently, our laboratories have collaborated to use NOESY cross-relaxation rates in phospholipid bilayers as a tool for the validation of molecular dynamics (MD) simulations and to use simulation trajectories to understand the factors influencing these rates (Feller et al., 1999). In this paper we extend these methods to the important problem of solute partitioning into bilayer membranes. Details of the experimental and simulation methodologies are presented, followed by experimental and computational results that provide a picture of ethanol-membrane interactions in unprecedented detail. We conclude with a summary of the biological implications of these results and a discussion of possible extensions of these methods to other systems of biochemical interest.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

Sample preparation

The phospholipids 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine-d4 (POPC-d4), and 1-palmitoyl-d31-2-oleoyl-sn-glycero-3-phosphocholine (POPC-d31) were obtained from Avanti Polar Lipids (Alabaster, AL). Lipids were received as a powder and dried to a monohydrate in the presence of phosphorous pentoxide under a vacuum of 50 µm Hg for 24 h. Multilamellar vesicles were prepared by adding a mixture of ethanol in D₂O to ~50 mg of the dry lipid. The appropriate amount of ethanol solution was added to establish a molar ratio of lipid/ethanol/water of 1:1:10. The samples were sealed and subjected to centrifugation and temperature cycling between 20 and 50°C over 6 h to achieve homogeneity. Ten milligrams of sample were then transferred to an MAS insert and sealed inside a 4-mm rotor for magic angle spinning.

Nuclear magnetic resonance measurements

²H Nuclear magnetic resonance (NMR) powder spectra were acquired on a Bruker DMX300 spectrometer at a resonance frequency of 46.1 MHz using a stationary dual resonance probe with a 4-mm solenoidal sample coil. Temperature was maintained at 10°C. A quadrupolar echo sequence (Davis et al., 1976) was used with two 1.8-µs 90° pulses and an interpulse delay of 50 µs. For the lipid resonances, 4096 scans with at a spectral width of 200 kHz and a delay time of 0.5 s were acquired. Deuterated ethanol and water required fewer scans. ²H NMR powder pattern spectra were dePaked (Sternin et al., 1983; McCabe and Wassall, 1995) to give spectra that correspond to the 0° orientation of the bilayer normal with respect to the external magnetic field. Smoothed order parameter profiles of the palmitic chain of POPC were calculated according to Lafleur et al. (1989) and Holte et al. (1995).

NOESY NMR experiments with magic angle spinning were conducted on a Bruker DMX500 widebore spectrometer at a resonance frequency of 500.13 MHz. Sample spinning at 10 kHz was accomplished in a Bruker double gas bearing MAS probehead for 4-mm rotors. Compressed air was passed through a copper coil immersed in a cooling bath to maintain sample temperature at 10 ± 1°C. Temperature inside the spinning rotor was calibrated by recording the proton spectra of 1-stearoyl-2-oleoyl-sn-glyero-3-phosphocholine as a function of temperature to detect the gel-to-liquid crystalline phase transition at 6.6°C (Boggs and Tummler, 1993). Two-dimensional NOESY was carried out in the phase-sensitive mode. The pulse sequence (90°  t₁  90°  _m  90°  acquire [t₂])_n was used with 512 t₁ increments, mixing times, _m, from 5 to 800 ms, 2048 t₂, increments, and 16 scans per t₁-increment. A 3.7-µs 90° pulse and 3.3-kHz spectral width were used. The magnetic field strength was locked to the internal D₂O resonance of the sample, using the third channel of the MAS probe.

Intensity of diagonal- and cross-peaks was recorded as a function of mixing time and data were analyzed with XWIN-NMR software (Bruker Instruments, Billerica, MA). To more accurately measure volumes of cross-peaks between lipid and ethanol resonances, we performed signal deconvolution of the two-dimensional NOESY spectra using the curve fitting routine in SigmaPlot 5 (SPSS Inc., Chicago, IL). The resonances are reasonably well approximated by Gaussian peaks with peak position, peak width in x- and y-dimensions, and peak height as fitting parameters. Precision of the deconvolution procedure depends mostly on experimental artifacts, such as residual t₁-noise and deviation of line shapes from a Gaussian form, but also on the ratio of superimposed peak intensities. Typical experimental errors of fitted peak volumes are 10 to 20%. Cross-relaxation rates were calculated by a matrix algorithm that corrects for mild to moderate influences from spin diffusion and spin-lattice relaxation (Huster et al., 1999). No significant contributions from spin diffusion were detected for mixing times up to 400 ms.

Molecular dynamics simulations

The simulation protocol has been successfully applied to the simulation of numerous saturated and unsaturated phosphatidylcholine bilayers in the recent past (Feller et al., 1997a, 1999; Armen et al., 1998; Schneider and Feller, 2001) and is described here for completeness. The periodic simulation cell contained 72 lipids (36 per monolayer), 72 ethanol, and 720 water molecules, corresponding to the mole ratios of samples prepared for NMR investigations (see above). A partially flexible simulation cell was used with the z dimension (i.e., the bilayer normal) adjusted to maintain the P_zz = 1 atm, and the x and y dimensions fixed to maintain a surface area of 65 Å²/molecule, as estimated from x-ray diffraction experiments on similar lipids (Koenig et al., 1997) after considering corrections from changes in sn-1 chain order due to differences in the level of hydration and the presence of ethanol. Initial coordinates for the POPC molecules were taken from a previously published POPC simulation (Armen et al., 1998).

The program CHARMM (chemistry at Harvard molecular mechanics) (Brooks et al., 1983) was used with the PARM22b4b all-atom parameter set (Schlenkrich et al., 1996; Feller and MacKerell, 2000) and its extension to unsaturated lipids (Feller et al., 1997b). The CHARMM potential contains internal energy terms for bond lengths, bond angles, torsional angles, and improper torsional angles. The interactions between nonbonded atoms are described by Coulombic interactions between partial point charges on the atomic centers and a Lennard-Jones 6-to-12 potential. The Lennard-Jones potential was switched smoothly to zero over the region from 10 to 12 Å. Electrostatic interactions were included via the particle mesh Ewald summation (Essmann and Berkowitz, 1999). All bonds involving hydrogen were fixed at their equilibrium distances using the SHAKE algorithm (Ryckaert et al., 1977). A time step of 2 fs was used with a modified leap-frog Verlet integration scheme. A neighbor list, used for calculating the Lennard-Jones potential and the real space portion of the Ewald sum, was kept to 14 Å and updated every 20 fs. A variant of the extended system formalism, the Langevin Piston algorithm (Feller et al., 1995), was used to control the normal pressure. The temperature was maintained at 10°C by means of the Hoover thermostat (Hoover, 1985). Coordinates sets were saved every 1 ps for subsequent analysis. Simulations were carried out using four or eight processors on a Beowulf-type parallel computer with each nanosecond of simulation taking ~5 days of wall time. The total simulation length was 11 ns. For data analysis, the first nanosecond of motions was discarded as equilibration time.

Calculation of NOESY cross-relaxation rates

The NOESY experiment probes the transfer of magnetization, occurring over a time interval referred to as the mixing time, from a set of magnetically equivalent protons (producing resonance i) to a second set (producing resonance j). The calculation of NOESY cross-relaxation rates from protein models and simulations has been described in the literature (Brüschweiler and Wright, 1994; Brüschweiler et al., 1992) and is briefly reviewed here. The connection between the experimental NOESY cross-relaxation rates and the molecular dynamics simulations is established through the magnetic dipolar interaction correlation function

(1)

in which Y₂₀ = [5/16]^1/2(3 cos²  1), and  is the angle between the internuclear vector, _ij, and the z axis (normal to the membrane). The summations over i and j include all magnetically equivalent protons of each resonance. From the MD simulation, the correlation function defined by Eq. 1 is calculated directly from the trajectory. It should be noted that the number of individual pair correlation functions is large, however, only those pairs with small internuclear separations contribute to the summation. Therefore, the finite size of the simulation cell does not affect the results. The cross-relaxation rates are calculated from the spectral density functions J_ij(), which are the Fourier transform of the correlation functions. Cross-relaxation rates depend on J_ij(2₀) and J_ij(0) according to

(2)

in which, ₀ is the proton Larmor frequency, and  = (2/5)⁴²(µ₀/4)², in which  is the gyromagnetic ratio for protons. The correlation function defined by Eq. 1 was calculated from the methylene protons of each ethanol to each of the unique ¹H lipid resonances observed experimentally. The raw correlation functions were fit to a sum of four exponentials as described previously (Feller et al., 1999), C(t) = a_ie^1/t_i, allowing the Fourier transform to be performed analytically and the contribution from each exponential to the overall cross-relaxation rate to be quantified.

When interpreting the _ij it is important to emphasize that these cross-relaxation rates are determined both by structural considerations (the probability of close contact between protons) and dynamic factors (the relaxation of the proton-proton internuclear vector). The MD simulation results, by providing a direct picture of the underlying correlation function, C_ij(t), can be used to quantify these contributions. The probability of close contact is measured by the magnitude of C_ij(0), which is proportional to 1/r and thus extremely sensitive to the number of very close contacts. In terms of the fit parameters, this quantity is given by the sum of the intensity factors, a_i, and will be denoted a_s. The dependence of _ij on relaxation time is seen most easily by considering a simple single exponential decay of the correlation function:

(3)

Due to the difference in sign between the contributions of J_ij(2₀) and J_ij(0) to _ij (see Eq. 2), correlation times, , longer than ~400 ps lead to negative cross-relaxation rates for a proton Larmor frequency of 500 MHz.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

Characterization of the state of POPC, ethanol, and water by ²H NMR

As visible from the deuterium spectra and chain order parameter profiles (Figs. 1 and 2), POPC is in the lamellar state. The effects of dehydration and ethanol addition on chain order have opposite sign and similar magnitude. Dehydration from excess water to 10 waters per lipid increased chain order parameters, whereas addition of one ethanol per lipid lowered them (Fig. 2). Both ethanol and water interact with the lipid, leading to anisotropic solvent motions and quadrupolar interactions of the ²H nuclei with the electric field gradients within these molecules that only partially average out. The order parameters of the ethanol C---D₂ bonds, ethanol C---D₃ bonds, and water O---D bonds, are 0.064, 0.004, and 0.005, respectively. Such values are characteristic for temporary association of solvent molecules with the lipid/water interface (Barry and Gawrisch, 1994; Gawrisch et al., 1992). The MD simulation gave similarly low values for the order parameters of 0.039, 0.010, and 0.008 with standard deviations of ~0.020.

View larger version (12K): [in this window] [in a new window]   FIGURE 1   2H NMR spectra for POPC-d31 (A), ethanol-d5 (B), and D2O (C) in POPC/ethanol/water (1:1:10, mol/mol) samples recorded at 10°C. (A) Superposition of quadrupole splittings of all sn-1 chain deuterons in POPC-d31; (B) superposition of the methylene and methyl group quadrupole splittings of ethanol; (C) single averaged quadruple splitting of all deuterons in water.

View larger version (19K): [in this window] [in a new window]   FIGURE 2   Influence of dehydration and ethanol on POPC sn-1 chain order parameter profiles recorded at 10°C. Dehydration of POPC bilayers from excess water to 10 waters per lipid increases chain order, whereas addition of one ethanol molecule per lipid decreases chain order.

¹H MAS NOESY NMR

The resolution of resonance lines allows detection of 13 proton signals, 10 from POPC, two from ethanol, and one from water (Fig. 3). Improvements in the design of MAS rotor inserts and MAS probeheads resulted in two-dimensional NOESY spectra (Fig. 4) with lower t₁-noise compared with our previous investigation (Holte and Gawrisch, 1997). Resonance signals of ethanol are partially superimposed on lipid resonances. To reduce the influence from signal superposition on cross-peak intensities, we also studied POPC-d4 with two deuterated methylene groups in the choline headgroup, and POPC-d31 with a perdeuterated sn-1 palmitic acid chain. The chain-deuterated lipid only reduced signal superposition, whereas the headgroup deuterated POPC allowed unperturbed observation of ethanol-CH₂ cross-relaxation with lipid resonances. Intensities of partially superimposed cross-peaks were determined by deconvolution as described under Materials and Methods.

View larger version (22K): [in this window] [in a new window]   FIGURE 3   The 500-MHz magic-angle spinning 1H NMR spectrum of a POPC/ethanol/D2O sample (1:1:10, mol/mol) with peak assignments, recorded at a spinning speed of 10 kHz and a temperature of 10°C.

View larger version (41K): [in this window] [in a new window]   FIGURE 4   Two-dimensional NOESY contour plot of the sample shown in Fig. 3, recorded at a mixing time of 400 ms. Sets of cross-peaks between ethanol and lipid resonances have been detected. The cross-peak volume as a function of mixing time was determined by integration. Two-dimensional deconvolution procedures were applied to determine cross-peak volumes of partially superimposed resonances.

Fig. 5 gives the per-proton cross-relaxation rates between ethanol and lipid resonances. Measured cross-relaxation rates are highest for interaction of ethanol with interface lipid protons, particularly with those in the glycerol region. There is an intriguing difference in cross-relaxation rates between ethanol-methyl and ethanol-methylene groups. Whereas methyl groups have stronger cross-relaxation with chain signals, methylene groups interact more strongly with headgroup resonances, suggesting that the ethanol molecule is oriented with its methyl group toward the hydrophobic bilayer core. Additional evidence for this orientation comes from the MD simulation. We found that the orientation of ethanol is such that the average distance of the methyl group to the bilayer center is 0.65 Å less than the distance to the hydroxyl oxygen.

View larger version (23K): [in this window] [in a new window]   FIGURE 5   Experimental lipid-ethanol per-proton cross-relaxation rates as a function of position along the membrane normal. Filled and unfilled bars give the methyl and methylene results, respectively. The abbreviations , ,  indicate choline methylene, and methyl groups, sn-1 and sn-3, are the corresponding glycerol methylene groups, C2 and C3 are the chain methylene groups near the glycerol, CH2---C = the two methylene groups near the double bond in the sn-2 chain, and CH3 the terminal methyl groups of palmitic- and oleic acid chains.

Cross-relaxation rates from MD simulations

Fig. 6 shows examples of correlation functions calculated from the MD trajectory. Correlation functions were fit to a sum of four superimposed exponentials as described in Materials and Methods. Fit parameters are reported in Table 1. Based on previous analysis of the time scales of lipid dynamics (Pastor and Feller, 1996), the four relaxation rates can be attributed to various motions in the lipid matrix. Librational motions of ethanol and lipid chemical bonds cause the very fast initial decay of correlation functions with correlation times of a few picoseconds. The decay with correlation times in the range of 100 ps is the result of gauche-trans isomerization, correlation times from 500 to 1000 ps are caused by molecular rotation and wobble, and finally, correlation times in the multinanosecond range are related to the diffusive motions of the smaller ethanol molecules within the lipid matrix.

View larger version (17K): [in this window] [in a new window]   FIGURE 6   Dipolar correlation function for interactions between the ethanol methylene group with choline methyl-, glycerol sn-1-, and chain methyl groups. Correlation functions can be approximated by a superposition of four exponentials with correlation times corresponding to librational motions of ethanol and lipid chemical bonds (5 ps), gauche-trans isomerization (100 ps), molecular rotation and wobble (500-1000 ps), and the diffusive motions of the smaller ethanol molecules within the lipid matrix (3-11 ns).

Table 1 also provides the experimental and simulation results for the cross-relaxation rates between ethanol methylene protons and each resolved peak in the POPC spectrum. Both experiment and simulation give the same trend of rates of magnetization transfer between ethanol and lipid hydrocarbon-, glycerol-, and headgroup resonances (compare Fig. 5 and 7), i.e., the strongest magnetization transfer occurs to lipid resonances at or near the lipid/water interface, including upper segments of lipid hydrocarbon chains, glycerol, and lipid headgroups. Cross-relaxation rates decrease both going into the choline segments and moving further down the lipid chains to the center of the hydrophobic core. Considering the high sensitivity of cross-relaxation rates to experimental conditions such as temperature and water content, comparison of measured and calculated rates resulted in very good agreement.

View larger version (34K): [in this window] [in a new window]   FIGURE 7   Calculated lipid-ethanol per-proton cross-relaxation rates as a function of position along the membrane normal. The bars represent the rates, ij, as calculated by Eq. 4. The solid line represents the contribution from the spectral density function Jij(0) to ij. The dashed line is a scaled presentation of the intensity of correlation functions at time 0, C(0) (equivalent to as in Table 1). Cross-relaxation rates are dominated by contributions from spectral density function Jij(0).

Dynamic factors influencing the magnitude of NOESY cross-relaxation rates

We have calculated the contributions to the cross-relaxation rates from the spectral density functions J_ij(0) and J_ij(2₀). Results are summarized in Fig. 7. The spectral contributions J_ij(0) from motions with the correlation time, ₄, and the amplitude, a_4, (diffusive motions of ethanol molecules within the lipid/water interface) are dominating the cross-relaxation rates with their large negative values. The fast motions with correlation times of less than 400 ps contribute to J_ij(2₀) and result in a small positive correction to the overall negative cross-relaxation rates. At 10°C the correction is of the order of 5 to 20% (the bars in Fig. 7 represent the total cross-relaxation rates _ij, the solid line is the J_ij(0)-contribution only). Nonetheless, the ethanol/lipid cross-relaxation rates are such that they are sensitive to the competing contributions of J_ij(0) and J_if(2₀). Not only does J_ij(0) decline with increasing temperature because ₄ becomes smaller, but the contribution from J_if(2₀) (with opposite sign) simultaneously increases. We propose that this explains the tremendous sensitivity of measured cross-relaxation rates to temperature and water content of samples.

Fig. 7 also shows the importance of taking into account differences in relaxation rates between the correlation functions for each lipid segment. The scaled intensity of the correlation function at time 0, C(0) (dashed line in Fig. 7), poorly predicts differences between cross-relaxation rates, demonstrating that dynamic factors play an important role in the relative cross-relaxation rates. Differences in the longest correlation time, _4, have the most significant influence on these rates. This correlation time represents an effective lifetime of temporary associations between lipid and ethanol. The lifetime of these associations with functional groups at the hydrophobic/hydrophillic interface is increased compared with those that occur in the aqueous layer (Table 1).

Location of ethanol molecules in the lipid bilayer

The significant difference in ethanol-lipid cross-relaxation rates (Fig. 5), and in the magnitude of the choline and glycerol correlation functions (Fig. 6), suggests that the density of ethanol in the vicinity of the glyercol segment is higher than near the choline. To address the relationship between local ethanol density and NOESY cross-relaxation rates, the concentration of ethanol through the lipid bilayer, derived from the MD simulation, has been plotted in Fig. 8. Ethanol prefers the aqueous phase and the lipid water interface but can also penetrate into the lipid bilayer to the region of upper chain segments. Notice that the location of highest ethanol density is between the carbonyl and phosphate groups, i.e., coincident with the glycerol segment that is found, from both experiment and simulation, to have the highest rates of cross-relaxation. Due to its partial hydrophobicity, penetration into the hydrophobic regions of the bilayer is significantly deeper compared with water. The relatively low energetic cost of ethanol penetration into the interface region is quantified by the potential of mean force calculations presented in Fig. 9. Ethanol-free energy near the lipid phosphate groups is the same as in water but increases exponentially with decreasing distance from the bilayer center. Locating ethanol or water in the center of the hydrophobic core (z = 0) is energetically very unfavorable with the probability of finding an ethanol molecule in the bilayer center ~1000 times lower than finding ethanol in the water phase.

View larger version (23K): [in this window] [in a new window]   FIGURE 8   Probability distribution for ethanol and water as a function of position within the membrane. The location of the phosphate and carbonyl groups are shown for reference. Ethanol density has a local maximum near the location of lipid phosphate groups. Ethanol penetrates deeper into the lipid/water interface region compared with water molecules.

View larger version (15K): [in this window] [in a new window]   FIGURE 9   Potential of mean force for water (solid line) and ethanol (dashed line) as a function of position along the bilayer normal. The increase of ethanol-free energy by 4 kcal/mol corresponds to an ~1000-fold lower probability of finding ethanol in the bilayer center compared with the water phase or the lipid/water interface region near the phosphate groups.

In Figs. 5, 6, and 8, we have demonstrated that the highest NOESY cross-relaxation rates correspond to those segments of the lipid molecule that reside in the region of highest ethanol density. The origin of this high ethanol density near  z  = 17.5 Å is the strong interaction between ethanol and the lipid phosphate and carbonyl groups. This is shown clearly in Fig. 10 where the lipid-ethanol interaction energies from the MD simulation have been decomposed into contributions from submolecular fragments of the lipid molecule. The strongest ethanol interactions are observed with the phosphate and carbonyl groups, thus placing the ethanol methyl and methylene protons in close proximity to lipid protons of the glycerol and upper chain segments of the molecule and causing the high rates of cross-relaxation. A typical lipid/ethanol conformation is displayed in Fig. 11, showing a hydrogen bond between the polar ethanol hydrogen and a phosphate oxygen. The strong interactions between ethanol and lipid groups further modulate the observed cross-relaxation rates by increasing the lifetime of ethanol-lipid contacts. This is seen in Table 1 where the correlation functions for the glycerol segments have the longest mean relaxation times of any segments. Additionally, the dipolar correlation function depends not on r_ij, i.e., the simple distance between protons, but is instead a function of r. Thus, a small number of close contacts with long residence times can outweigh more frequent contacts at longer distances and/or shorter duration.

View larger version (32K): [in this window] [in a new window]   FIGURE 10   Average interaction energy between ethanol molecules and the various submolecular fragments of POPC. Attractive interaction is strongest with lipid phosphate groups, followed by hydrocarbon chain carbonyls, choline groups, hydrocarbon chains, and glycerol. The strong interaction with phosphate groups is the result of formation of hydrogen bonds.

View larger version (28K): [in this window] [in a new window]   FIGURE 11   Representative conformation of a temporary association of ethanol with a phospholipid molecule. Ethanol prefers to form hydrogen bonds with the lipid phosphate group whereas the ethyl residue is directed toward the bilayer hydrophobic core.

It should be emphasized that, due to the inherent disorder of lipids within the bilayer, ethanol-lipid contacts do not necessarily occur at the average location of the lipid segments. For example, there is a small but measurable cross-relaxation between the ethanol methyl group and the terminal methyl group of hydrocarbon chains. According to Fig. 8, the probability of ethanol penetration into the membranes hydrophobic core is extremely low. This raises the question of where ethanol and hydrocarbon chain methyl groups meet. We divided the simulated membrane into slabs of 5 Å thickness and estimated the spatial origin of magnetization transfer between these methyl protons (Fig. 12). The black bars represent the distribution of chain methyl groups, whereas the gray bars are the distribution of contacts between chain and ethanol methyl groups, defined as approach to a distance of 6 Å or less. Probability of contact between ethanol and lipid methyl groups is highest in the hydrocarbon chain region near the lipid/water interface and in the interface region. Therefore the contacts between methyl groups of ethanol and lipid hydrocarbon chains are both a reflection of lipid hydrocarbon chain upturns and excursions of ethanol molecules into the upper hydrophobic chain region of membranes.

View larger version (25K): [in this window] [in a new window]   FIGURE 12   Probabilities of chain methyl location (black bars) and ethanol-chain methyl contacts defined as approach to a distance of 6 Å or less (gray bars) as a function of distance from bilayer center. Intensities were averaged over slabs with a thickness of 5 Å. The probability of contacts between ethanol and chain methyl groups is highest in the hydrocarbon chain region near the interface, indicating infrequent chain upturns and brief excursions of ethanol molecules into regions of upper chain segments.

Concluding remarks

The quantitative analysis of two-dimensional NOESY spectra from lipid membranes has greatly increased the number of NMR parameters of membranes. For example, the total number of order- and relaxation parameters that were measured for the ethanol-POPC samples exceeds 100. In contrast to NOESY cross-relaxation rates in proteins, cross-relaxation in the lipid matrix does not report a fixed geometry of biomolecules, but reflects the distribution function of lipid- and solvent segments along the bilayer normal, the probability of short-lived associations, and the presence of motions within membranes covering correlation times in the range from pico- to microseconds. The good agreement between simulation and experiment serves to validate the model and computational protocol. Furthermore, the NMR parameters reflect intricate details on membrane structure, dynamics, and solvent interactions that can be interpreted quantitatively when the experiments are combined with MD simulations. Thus, the present study lays a foundation for further studies of membrane solute interactions via NOESY spectroscopy and MD simulation.

The combination of spectroscopic and simulation results unambiguously places ethanol at the membrane/water interface, specifically in the chemically heterogeneous region between the phosphate and carbonyl groups. Ethanol interacts with the lipid matrix primarily via hydrophilic interactions like hydrogen bonds and gains only modestly from hydrophobic interactions with upper chain segments. Although an interface location of ethanol was predicted previously on the basis of Fourier transform infrared (Chiou et al., 1990; Klemm, 1990), NMR (Holte and Gawrisch, 1997; Barry and Gawrisch, 1994, 1995; Klemm and Williams, 1996), and fluorescence studies (Slater et al., 1993), the current experiments excel in temporal and spatial resolution of ethanol membrane interaction. Ethanol molecules prefer the lipid/water interface but can penetrate into the region of upper chain segments. This location is primarily facilitated by hydrogen bonding between ethanol and the lipid phosphate groups, but also by hydrogen bonding to carbonyls and by weak hydrophobic van der Waals attraction between the short ethyl group and upper chain segments. The hydrophobic contributions to this interaction are unable to compete with the strong desire of ethanol to form hydrogen bonds, thus preventing high concentrations of ethanol within the hydrophobic core.

We speculate that interaction of ethanol with the interface region of proteins is governed by similar needs to satisfy hydrophilic interactions, aided by small gains from inserting the ethyl groups into a hydrophobic environment. We expect that ethanol interacts with the same functional groups on protein surfaces as in the lipid/water interface region. Interestingly, phosphorylation of proteins is a common mechanism to control activity. Perhaps, ethanol binding to phosphorylation sites of proteins is one of the mechanisms by which ethanol influences protein function.