ARTICLE pubs.acs.org/JPCB
Lipid Membranes with a Majority of Cholesterol: Applications to the Ocular Lens and Aquaporin 0 Joseph W. O’Connor and Jeffery B. Klauda* Department of Chemical and Biomolecular Engineering, University of Maryland, College Park, Maryland 20742, United States
bS Supporting Information ABSTRACT: Using molecular dynamics (MD) simulations, we studied the structure and dynamics of two dimyristoylphosphatidylcholine (DMPC):cholesterol bilayers at concentrations representative of the ocular lens (ratios of 1:1 and 1:2). These MD simulations agree well with experimental deuterium order parameters and bilayer peak-to-peak distances. Although it is known that the average surface area per lipid rapidly decreases from low to moderate levels of cholesterol, our simulations indicate that there is a relatively small change in the average lipid area from 50 to 66.7% cholesterol (40.5 ( 0.2 and 39.5 ( 0.1 Å2/lipid, respectively). Radial distribution functions for the hydroxyl group on cholesterol indicate the formation of cholesterol-only nanoscale domains for the membrane with 66.7% cholesterol but a uniform distribution of cholesterol and DMPC for the membrane with 50% cholesterol. These small domains form a single shell of hexagonally packed cholesterols that are interconnected in a web-like structure of cholesterol. Calculations of internal DMPC dynamics show that the relaxation times for carbonhydrogen reorientation of choline decrease with an increase in cholesterol, but the main body (carbonyl-glycerol to C11) is independent of cholesterol concentration. MD simulations of the aquaporin 0 tetramer show stabilization in its interactions with lipid membranes containing cholesterol by forming ringring stacking between surface aromatic residues of the protein and the rings of cholesterol. Moreover, there is an increase in hydrogen bonds with longer lifetimes in a mixed bilayer of DMPC and cholesterol.
’ INTRODUCTION The ocular lens of the eye is a transparent tissue that allows visible light to be transmitted unimpeded into the eye and focused to the retina. The structure of the lens resembles an onion because of the multiple layers of lens fiber membranes.1 Layers of fiber cells are deposited upon previous layers throughout the life of the lens.2 The lens nucleus is at the center of the ocular lens, and these membranes are the oldest. This is not a traditional cellular nucleus because fiber cells do not contain subcellular organelles that would normally scatter light and therefore hinder vision. Newer layers that include metabolically active fiber cells are part of the region named the lens cortex.2 The fiber cell membrane contains the highest relative concentration of cholesterol found in nature,3 where the cholesterol to phospholipid ratio ranges from 1 to 4.2 The concentration of cholesterol is the highest in the lens nucleus (ratio of 3 to 4) and lower in the cortex (ratio of 1 to 2).2 Eukaryotic cells on average have a cholesterol to phospholipid ratio from 0.5 to 1,2 but organelles typically range from 0.1 (endoplasmic reticulum) to nearly 1 for the plasma membrane.4 Cholesterol tends to aggregate into a cholesterol-rich domain (liquid ordered), and a resulting cholesterol-poor domain (liquid disordered) is formed.48 The temperature and compositions for this phase separation depend on the lipid types in the membrane with cholesterol, but the primary phase separation exists to about 40% cholesterol.4,8 The inclusion of cholesterol in membranes is r 2011 American Chemical Society
known to have many functions beyond domain formation. Cholesterol increases molecular order,9 reduces permeability,10 produces a condensing effect on membranes,11,12 and regulates membrane fluidity.12,13 In the ocular lens membranes, the phospholipid composition is over 50% sphingomyelin (SM) and sphingomyelin derivatives. Phosphatidylcholines and phosphatidylethanolamines constitute most of the remaining lipids.1,14 The cholesterol concentration in the membrane is disturbed in the development of cataracts; however, little is known about the structure of the membranes in cataracts.15 Although coarse-grained and all-atom simulations have been used to study membranes with no more than 50% cholesterol,16,17 to our knowledge, no simulations beyond conceptual models have been used to investigate membranes with a majority of cholesterol. There are several existing conceptual models for cholesterollipid interaction, such as the condensed complex model,18 umbrella model,19 and superlattice model.20 The condensed complex model states that certain cholesterol lipid complexes exist at various stoichiometric concentrations. These complexes form a condensed phase with small lateral areas. A mismatch in headgroup size (cholesterol vs phospholipid) proposed by the umbrella model results in preferred Received: September 10, 2010 Revised: April 8, 2011 Published: May 03, 2011 6455
dx.doi.org/10.1021/jp108650u | J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B
ARTICLE
Table 1. System Details for the Bilayer Simulationsa system
number of lipids
number of
total number
initial box length:
total simulation
time step
water molecules
of atoms
X Y Z (Å)
time (ns)
(fs)
1:1
516
15, 594
96, 318
100.3 100.3 90.0
123
2
1:2
558
16, 863
100, 065
100.2 100.2 94.0
123
2
AQP0/DMPC
370
24, 619
131, 875
125.1 125.1 81.9
25
1
AQP0/DMPCChol
L1, 127/132; L2, 118/134
24, 619
131, 875
125.1 125.1 81.9
100
1
a
The ratios given for the pure membrane simulations are (DMPC:cholesterol), and the AQP0 tetramer simulations were performed in a pure DMPC and DMPCcholesterol (48/52%) bilayer. The number of lipids (DMPC/cholesterol) for each leaflet (L1 and L2) is given for the AQP0 simulations with the DMPCcholesterol bilayer; other simulations have the same number per leaflet.
association of membrane lipids; i.e., the phospholipid head groups act as an umbrella to protect the hydroxyl group of cholesterol. Finally, the superlattice model states that long-range repulsive interaction between cholesterol is the key driving force for cholesterollipid interactions. Of the three conceptual models described above, only the umbrella model21 agrees with experimental observables for all cholesterol concentrations up to the solubility limit of 66.7% for saturated chains with phosphatidycholine head groups.22 Although the superlattice model can be used to describe superlattices below 50% cholesterol, it cannot predict these above 50% due to the inherent cholesterol repulsion.21 Some evidence supports a single liquid ordered phase that changes an ordered pattern at 57.1 and 66.7% cholesterol,8,16,21 while other evidence supports nanoscale domains of cholesterol.17,23,24 These domains, also known as immiscible cholesterol domains (ICDs), have been observed using F€orster resonance energy transfer (FRET) measurements on dimyristoylphosphatidylcholine (DMPC)cholesterol membranes.23,24 Previous united-atom simulations of DMPCcholesterol (80/20%) did not result in such ICD formation, but coarse-grained simulations up to an equal molar composition suggest the existence of these nanoscale ICD domains.17 Additional molecular simulations at cholesterol concentrations at and above 50% may better probe possible nanoscale inhomogeneities. The lens fiber membrane also contains a high concentration of the major intrinsic protein or aquaporin 0 (AQP0). AQP0 is a water transporter and accounts for 60% of proteins in the lens fiber cells.25 Unlike other aquaporins, AQP0 mediates adhesive contacts between cells26,27 and transports water approximately 10-fold more slowly than the other aquaporins.28 In order to maintain transparency, the water content of the fiber cells must be regulated and AQP0 enables such water circulation.29 Without this process, the cells will become opaque.30 X-ray diffraction structures have shown that the protein exists as a tetramer where each monomer transports water through its lumen.29,31 The biological significance of the low water permeability is currently not well understood. Also, AQP0 mutations cause the loss of transparency in the ocular lens.32 It is therefore important to further study the structure and dynamics of AQP0 and membranes that represent that of the ocular lens. Our study will use molecular simulations of DMPC/cholesterol bilayers to study the influence of cholesterol concentration on various lipid properties. Although SM is the major lipid besides cholesterol in ocular lens membranes, all-atom force field parameters are not currently available for this lipid with the all-atom CHARMM force field.33 The structures of SM and DMPC are similar, except that SM has sphingosine and DMPC palmitoyl chains.34 To our knowledge, simulations of lipid
membranes with cholesterol concentrations greater than 50% have not been previously studied and will be the focus of this work. Binary membranes with cholesterol and DMPC or dipalmitoylphosphatidylcholine (DPPC) have complex phase behavior, as described above. Therefore, one focus of our DMPCcholesterol simulations will be the possible existence of nanoscale inhomogeneities or ICDs. Our simulations will be compared to experimental measurements with 50% cholesterol and offer additional information about the properties of phospholipid membranes in the presence of an elevated cholesterol concentration. Moreover, we will also present simulations of AQP0 with pure DMPC and DMPC/cholesterol (48/52%) membranes and suggest how this protein has evolved to prefer membranes with high levels of cholesterol.
’ METHODS Two bilayers, with a DMPC:cholesterol ratio of 1:1 and 1:2, were created using CHARMM-GUI (Chemistry at HARvard Macromolecular Mechanics-Graphical User Interface: charmmgui.org),35,36 and the number of molecules in each of these systems is shown in Table 1. The lipids are placed in pseudorandom positions to form a bilayer that equally distributes lipid molecules based on their headgroup volume.36,37 This results in a bilayer with minimal preference for aggregation of similar lipids. The ratio values reported in the remainder of the paper are DMPC to cholesterol unless otherwise specified. When building the bilayers, the length of the axis perpendicular to the membrane surface (Z) was based on a water thickness of 12 Å, which resulted in approximately 30 waters per lipid for each system. The length of X and Y were based on the ratios of lipid components and an initial guess of 100 Å. After building, standard CHARMM-GUI scripts36 were used with the CHARMM38,39 program to equilibrate the membrane simulations with a constant area ensemble (NPAT) based on the initial estimate for the area from CHARMM-GUI. After this initial equilibration, simulations of the lipid membranes were performed at a constant pressure of 1 bar and at a temperature of 30 C in the NPT (constant number of molecules, pressure, and temperature) ensemble. Langevin dynamics and the Langevin-piston algorithm were used to maintain constant temperature and pressure, respectively. A tetragonal unit cell was used to maintain X = Y but not equal to the Z-dimension. The CHARMM36 (C36)33 force field was used because of its ability to match experiment in the NPT ensemble with the TIP3P40 water model. All simulations were performed using the NAMD41 simulation program. A potential switching function was used to smooth the van der Waals interactions to zero from 810 Å. Particle-mesh Ewald42 was used to include long-range 6456
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B
ARTICLE
Figure 1. Snapshots for three of the simulated systems: (a) 1:1 (DMPC:cholesterol), (b) 1:2, and (c) a top view of the AQP0 tetramer in a pure DMPC bilayer. The red molecules are water, blue molecules are cholesterol, and green molecules are DMPC. The different colors of the AQP0 tetramer denote the different monomers. The water molecules in part c were deleted from the snapshot for clarity. (d) Structure and nomenclature for a DMPC molecule.
electrostatics with an interpolation order of 4, a direct space tolerance of 106, and a grid size of 108 108 96 Å. A time step of 2 fs was used in these simulations, and coordinates were saved every picosecond. The RATTLE algorithm43 was implemented to maintain a constant heavy atomhydrogen distance for all molecules. A pure DMPC bilayer with an AQP0 tetramer embedded in the middle was also built with CHARMM-GUI and simulated in the same manner as described above (equilibration and production). AQP0 was generated as a biologically functional unit (a tetramer) based on the crystal structure from Harries et al.29 The original structure contained a ligand (nonyl β-Dglucopyranoside), but the ligand was deleted for our study. The protein was oriented along the Z-axis using Ala-45 and Gly-165 from the first monomer as vectors of reference. The tetramer was then translated 4 Å along the Z-axis so that the protein was at the center of the membrane. The length of Z is based on a water thickness of 12 Å, and the length of X and Y was based on an initial guess of 125 Å. For the simulation of AQP0 in a cholesterol-containing bilayer, a 20 ns snapshot of the 1:1 DMPCcholesterol bilayer was used to build the system with the protein. Additional lipids were added (based on unit cell dimensions) so that initial box dimensions were greater than 125 Å. Lipids and water that were within 2 Å of the protein were deleted. Snapshots of the lipid membrane systems and that with AQP0 are shown in Figure 1AC. The figure also labels the important DMPC atoms relevant to our analysis of these membranes (Figure 1D). In addition to the number of molecules, simulation times and other conditions are listed in Table 1. The surface area per lipid was used as a metric to determine if our simulations were equilibrated, and this as a function of time is shown in Figure S1 of the Supporting Information for both of the
DMPC:cholesterol simulations. The area converges after ∼17 ns for the 1:1 simulation and ∼10 ns for the 1:2 simulation. All reported calculations for structural and dynamical properties were found using data after the area per lipid equilibrated. The membrane simulations with AQP0 demonstrate equilibration by the x-dimension of the unit cell after 20 ns (Figure S2, Supporting Information), and analysis is based on data after 20 ns. In order to evaluate the structure and dynamics of the systems, calculations were undertaken using CHARMM.38,39 The deuterium order parameters (SCD) can be calculated by 3 2 1 cos θ SCD ¼ ð1Þ 2 2 where θ is the angle between a CH vector and the bilayer normal.44 Reported values of SCD in this paper are always positive; i.e., the absolute value of eq 1 is shown. 13C spinlattice relaxation (1/T1) rates were calculated to understand the lipid dynamics in these membranes. The method used to find the rates is similar to other molecular dynamics studies.4547 Assuming pure dipolar relaxation between the 13C nucleus and its attached protons48 1 1 pγc γh 2 ¼ ½ JðωH ωC Þ þ 3JðωC Þ þ 6JðωH þ ωC Þ NT1 10 rCH 3 ð2Þ where N is the number of attached protons, rCH is the effective CH bond length, γH and γC are gyromagnetic ratios, and ωH and ωC are Larmor frequencies. J(ω) is the spectral density given by Z ¥ JðωÞ ¼ C2 ðtÞ cosðωtÞ dt ð3Þ 0
6457
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B
ARTICLE
Table 2. Average Surface Area per Lipid (Å2) for Cholesterol and DMPC and the Corresponding Standard Errors system
cholesterol
DMPC
1:1
29.40 ( 0.05
51.29 ( 0.05
1:2
32.41 ( 0.02
53.59 ( 0.06
^ ðtÞÞæ C2 ðtÞ ¼ ÆP2 ð^ μ ð0Þ 3 μ
ð4Þ
where C2(t) is the second rank reorientational correlation function, P2 is the second-order Legendre polynomial, and μ^(t) is the unit vector along the CH bond direction at time t. The reorientational correlation function was fit to a multiexponential function to tease out relaxation times: C2 ðtÞ ¼ a0 þ
3
∑ ai et=τ i¼1
i
ð5Þ
The spectral density is the Fourier transform of eq 5 JðωÞ ¼
3
aτ
i i ∑ 2 i ¼ 1 1 þ ðωτ i Þ
ð6Þ
The average surface area for each lipid type (DMPC and cholesterol) was calculated for using Voronoi tessellation.49,50 Radial distribution functions, g(r), were also calculated between the oxygen atoms on cholesterol and include periodic boundary conditions with a Δr value of 0.02 Å. For the hydrogen bond analysis of AQP0 with lipids, the COOR HBOND command was used in CHARMM based on a maximum hydrogenheavy atom distance of 2.2 Å.
’ RESULTS Area per Lipid. The simplest measure for the average size of lipids in a membrane is commonly reported as the surface area per lipid. Moreover, this property describes the effective lateral density of membranes and can depend on lipid type51 and concentration.12 The average surface area per lipid is 40.5 ( 0.2 and 39.5 ( 0.1 Å2 for the 1:1 and 1:2 MD simulations, respectively. These values are very similar; however, there is a slight decrease in surface area as the cholesterol concentration increases. As a comparison, the experimental surface area per lipid for a pure DMPC bilayer at 30 C is much greater (60.6 Å2).52 X-ray scattering experiments also show that increasing the mole fraction of cholesterol from 0 to 0.4 decreases the area per lipid from about 60.6 to 46.4 Å2.53 Our MD simulations also follow this trend when you compare values in this study with simulations with C3633 of a pure DMPC bilayer with a surface area of 60.8 Å2; i.e., there is a significant decrease in surface area per lipid from 0 to 0.5 mol fraction of cholesterol. The individual lipid surface areas were also calculated and demonstrate a slight increase in surface areas for the lipid types as cholesterol concentration increases (Table 2). Cholesterol has a maximum condensation effect on DMPC (8.7 Å2 decrease in area) at an equal molar composition. An increase in cholesterol concentration to 66.7% results in less contact with the acyl chains of DMPC and 3 Å2 increase in the cholesterol surface area. Deuterium Order Parameters. In order to describe the structure of the alkyl chains, the deuterium order parameters (SCD) were calculated for each carbon on the chain. Figure 2 shows the magnitude of the order parameters of the DMPC/
Figure 2. Deuterium order parameters (SCD's) for the DMPC sn-1 (top) and sn-2 tails (bottom) for (DMPC:cholesterol) 1:1, 1:2, pure experimental DMPC (sn-154 and sn-255), 1:1 experimental DMPC56 membranes, and pure simulation DMPC.33 The sn-1 experiments were measured at 40 C and 46.1 MHz.
cholesterol simulations for both alkyl chains (sn-1 and sn-2). The experimental SCD values for the pure DMPC bilayer (sn-154 and sn-255) and the 1:1 bilayer follow the same general trends;56 i.e., the largest values are the carbons near the headgroup, and they decrease toward the end of the chain. Comparing our MD simulations with experiment, the 1:1 values for the middle and end of the chain (carbons 514) agree well with experiment. However, there is a decrease in order toward the headgroup for carbons 24 on sn-1, which is not experimentally observed.56 A decrease is measured for the 1:1 and pure DMPC bilayer experiments5456 for carbon 2 on sn-2, and this is consistent with our calculations. The disagreement between experiment and simulation for most of the upper chain may be the result of the experimental spectral assignment for membranes that had all chain hydrogens replaced with deuterium. Selective replacement of hydrogen at each carbon position would better quantify if there is a reduction of order for both chains toward the carbonyl of the headgroup. However, the simulations do agree with experiment for the relative order of each aliphatic chain; i.e., the order for sn-2 is greater than that for sn-1. Although order dramatically increases from 0 to 50% cholesterol (Figure 2), the SCD's for the simulations with cholesterol are virtually the same, especially from C2 to C8. There appears to be a saturation limit to the order that any more cholesterol will not influence the order of the lipid chain. However, a slight decrease in order for the end carbons (C9C14) with an increase in cholesterol is calculated for both of the simulations. This dependence on sterol concentration is similar to that measured with lower concentrations of cholesterol (2550%).9 Moreover, this agrees with the slight increase in surface area for DMPC from 6458
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B 1:1 to 1:2, suggesting a decrease in order results in an increase in lipid area. Electron Density. Another way to describe the structure of a bilayer is by evaluating electron density profiles for the entire bilayer, individual components to the membrane, and different moieties of the lipids. The peak-to-peak distance for the lipid bilayer density, as shown in Figure 3, is 43.52 ( 0.05 and 42.94 ( 0.05 Å for the 1:1 and 1:2 simulations, respectively. The 1:1 value for simulations was close to the experimental value of 42 Å for 50% cholesterol at 10 C.57 The experimental value for the pure DMPC bilayer is 35.5 Å at 30 C.52 Our results show that, at high cholesterol concentrations, the bilayer thickness shrinks as the cholesterol concentration increases. However, according to experimental results, at lower cholesterol concentrations (00.5 mol fraction of cholesterol), the bilayer thickness expands as the cholesterol concentration increases.57,58 The bilayer height shrinking may be the result of having a minority of the phospholipid compared to cholesterol that allows for more
Figure 3. Electron density profiles for 1:1 and 1:2 DMPC:cholesterol membranes.
ARTICLE
conformational flexibility of the headgroup and reduction of chain order (Figure 2). This results in a small shift in carbonylglycerol and phosphate distribution of ∼0.5 Å closer to the center of the bilayer (Figure S3, Supporting Information). The plots in Figure 3 have the typical characteristics seen in diffraction experiments.52 The symmetrized plots in Figure S3 of the Supporting Information contain the peak around 21.5 Å, which represents the phosphate groups of the lipid head and a minimum at the center of the bilayer. As opposed to pure phospholipid membranes, there is also a second peak around 10 Å that is a combination of the density from the cholesterol ring and lipid methylenes. Consequently, the peak around 21.5 Å shrinks and the peak around 10 Å rises as the amount of cholesterol increases. Figure S3 of the Supporting Information depicts how, as the cholesterol concentration increases, the Gaussian-like peaks for the carbonyl-glycerol and phosphatidylcholine groups shrink. The methyl distributions for the end carbons of DMPC are non-Gaussian and do not have a maximum at the center of the bilayer, which is common for pure bilayers.59 Radial Distribution Functions (RDFs). Although the abovementioned structure-based analysis gives an overall description of the bilayer, local order may exist in these bilayers that are known to form domains at lower concentrations of cholesterol. Cholesterolcholesterol RDFs demonstrate changes from the initial and final conditions in these long atomistic simulations (Figure 4 and Figure S4 of the Supporting Information). There is a slight increase (first vs last 5 ns) in the density of lipids in the first four peaks for the 1:2 simulation and for the second to fourth peak in the 1:1 simulation. Moreover, there is a slight decrease in the position of the second peak, which is the result of area equilibration. Both simulations show at least four peaks for the cholesterolcholesterol RDFs, but the position of the peaks beyond the first shell differs between the 1:1 and 1:2 simulations (Figure S4, Supporting Information). The distance between the central molecule and its first shell neighbors is 5.55.6 Å and is
Figure 4. Two-dimensional radial distribution functions for cholesterolcholesterol defined by the hydroxyl oxygen. Averages for the first and last 5 ns are shown. 6459
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B
ARTICLE
Figure 5. T1 for aliphatic carbons of (DMPC:cholesterol) 1:1, 1:2, and MD simulations of pure DPPC at 75.4 MHz.33 For proper comparison, Æ411æ, 12, and 13 are Æ413æ, 14, and 15 for DPPC. This way, the second and third from last carbon is compared directly from DPPC and the DMPCcholesterol membranes.
invariant to cholesterol concentration. The second shell is nearly twice that of the first shell for the 1:2 simulation (11.2 Å) and slightly less for 1:1 (10.8 Å). The small tertiary peak is at 17.0 and 15.2 Å for 1:2 and 1:1, respectively. The peaks that are near multiples of the first indicate clustering of cholesterol in the 1:2 simulation up to four solvation shells (Figure S4, Supporting Information), but this does not exist in the 1:1 simulation. For some leaflets, the tertiary peak does not exist initially and occurs after 10 and 66 ns for leaflet 1 and 2 of the 1:2 simulation, suggesting that simulations on the order of 50 ns or greater are needed to form cholesterol domains with our starting conditions. A further description on how this may pertain to DMPC/ cholesterol models for interaction is presented in the Discussion section. CH Reorientational Dynamics and SpinLattice Relaxation Rates. The dynamics of lipid intramolecular motions was probed by calculating carbonhydrogen reorientational correlation functions for various positions on the chain and headgroup. These were fit to three term exponentials (Methods), and parameters can be found in Table S1. For the two simulations, the fast and intermediate time constants (τ1 and τ2) for the alkyl chain decrease with carbon number. The slow relaxation time constant (τ3) follows a similar trend with a few exceptions. There is a slight increase in τ3 as you increase the cholesterol concentration (Table S1, Supporting Information), but larger differences exist for carbons near the headgroup (C2 and C3). For example, τ3 increases from 3.8 to 7.5 ns as the cholesterol concentration increases from 50 to 67%. The τ3 constant for the glycerol carbons also increases with cholesterol concentration. For example, the value of τ3 for CG1 increases from 6.2 to 8.6 ns and the value for CR decreases from 1.2 to 0.8 ns as the cholesterol concentration increases from 50 to 67%. Figure 5 contains the T1 values for the DMPC carbons for a carbon frequency of 75.4 MHz and compares them to a DPPC MD simulation.33 The pure DPPC MD simulation was run with the C36 lipid FF and based on analysis of a simulation previously reported.33 To compare the carbons for the end of the chain, the DMPC relaxation times for carbons Æ411æ, 12, and 13 are compared to DPPC Æ413æ, 14, and 15. The DMPC T1 values are invariant to changes in cholesterol for the aliphatic carbons 211. The DMPC:cholesterol simulations follow the same general trend as the DPPC simulations, and the T1 values in the lipid chain region of C2C11 are nearly identical. Therefore,
Figure 6. The root mean squared deviation (rmsd) for each monomer (AD) of AQP0 as a function of simulation time. The top panel is for the simulation in pure DMPC and the bottom for DMPCcholesterol.
measurements at 75.4 MHz would only show differences between pure and mixed bilayers for headgroup carbons and C12 and C13 of the chain. For the choline section, it is found that, as the cholesterol concentration increases, T1 increases as well. AQP0 and Lipid Membrane. The water transport protein, AQP0, exists in the unique ocular lens membranes with elevated levels of cholesterol. Two simulations (Table 1) were run to probe the influence of membrane on membraneprotein interactions. Both MD simulations resulted in a minimal rmsd for each of the AQP0 monomers (Figure 6). For all but one monomer, the average rmsd was below 1.3 Å. The rmsd for the inner loop connecting monomer D with the other three monomers in the DMPC/cholesterol bilayer (Figure 7A) was the main cause for this elevated rmsd in the bottom panel of Figure 6. X-ray diffraction on AQP0 in a DMPC and E. coli polar lipid membrane has been used to determine important lipidprotein interactions.31,60 Although certain residues indicate lipidprotein interaction via charged and polar residues (Arg-11, Tyr-105, Ser-106, Arg-113, Ser-126, Gln-128, Gln-129, Arg-196, and Lys238), these may not constitute a true lipid-binding motif.60 Since no structure has been determined with cholesterol, our simulations demonstrate that polar interactions via hydrogen bonding result in longer hydrogen bond occupancies and lifetimes with the cholesterol-containing membrane (Table 3). Although proteinlipid interactions agree with most of those observed from experiment, the long-lived interactions (greater than 10% occupancy) are shown in Table 3. Overall, there are on average 4.9 lipidprotein contacts via hydrogen bonding to the hydroxyl group of cholesterol or the glycerol of DMPC that have a lifetime of 31.3 ps. However, for the DMPC-only bilayer, this is reduced to 3.8 lipidprotein contacts at any given moment. Of the polar and charged residues, Ser-106 and Arg-196 (Table 3 and Figure 7D) interact with the membrane to the greatest extent independent of the presence of cholesterol. The occurrence of AQP0 and its interactions with DMPC or cholesterol is similar (2.5 and 2.4 interactions at any given time, respectively), but 6460
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B
ARTICLE
Figure 7. (A) 99 ns snapshot of the simulation system of AQP0 with a DMPC and cholesterol (yellow) bilayer. Water is shown in red. (B) 99 ns snapshot of cholesterol forming ringring stacking interaction with Trp-10 and Phe-9 and a hydrogen bond with Ser-6. (C) 99 ns snapshot of cholesterol forming ringring stacking interaction with Phe-189 and Phe-198. (D) 99.7 ns snapshot of two cholesterols. One cholesterol forms ringring stacking interactions with Phe-189 and Trp-202 and another a hydrogen bond with Arg-196.
Table 3. Hydrogen Bond Average Occupancy (Æoccæ) between All Lipids and Specified Residues of the AQP0 Tetramera pure DPPC
1:1
Æoccæ
ÆHB lifetimeæ (ps)
Trp-10
0.36
4.2
0.40
24.4
Ser-106
0.54
16.3
0.98
62.3
Gln-129
0.42
5.3
0.26
18.0
Arg-196 Lys-238
0.95 0.27
7.9 9.9
1.96 0.51
43.2 20.7
residue Ser-6
Æoccæ
ÆHB lifetimeæ (ps)
0.26
28.2
a
If a specified residue on each tetramer interacts with a lipid 100% of the time, then Æoccæ = 4. The average lifetime of these hydrogen bonds (ÆHB lifetimeæ) is also listed.
the lifetime of hydrogen bonds is greater for DMPC (38.9 vs 24.2 ps). Although lipid acyl chains form important interactions with AQP0,31,60 the rings in cholesterol offer the ability to form stacking interactions (faceface and edgeface). There are six residues that form such interactions on leaflet 1 (Phe-18, Phe136, Phe-189, Trp-198, Trp-202, and Trp-205) and four on leaflet 2 (Phe-9, Trp-10, Phe-73, and Phe-221). A sample of these cholesterolprotein conformations are shown in Figure 7BD. Although faceface stacking is more common, edgeface interaction also exists, as shown in Figure 7D with Phe-189.
’ DISCUSSION In the present study, MD was used to characterize the structure and dynamics of DMPC/cholesterol membranes (50 and 66.7% cholesterol) and membrane simulations with the integral protein of the ocular lens, AQP0. Molecular dynamics has been used frequently to study cholesterol-containing bilayers,16,17,6167 but to our knowledge, this is the first study that has systematically investigated the influence of cholesterol in bilayers with a majority of this sterol. The C36 force field was used because the tensionless NPT ensemble results in accurate surface areas per lipid, order parameters, and 13C relaxation rates.33 Our bilayer simulations of 50 and 66.7% cholesterol resulted in calculations for the SCD, electron density, and T1, which all followed published trends based on experiments and previous simulations. Moreover, the peak-to-peak distance of the 1:1 bilayer also agreed with the experimental value with a percent error of less than 4%. Our simulations were able to determine how incorporating high cholesterol concentrations into a DMPC bilayer influence various structural properties of bilayers. It is well-known that cholesterol can condense lipid bilayers (reduce surface areas per lipid),11,12 but this influence is also nonuniversal and depends on chain type.12 However, this dramatic increase in the lipid lateral density occurs when cholesterol is the minority compound in a bilayer. The average surface area per lipid continues to decrease when cholesterol is the majority lipid, but the rate of decrease is relatively small. For example, the rate of change in surface area from 0 to 20% cholesterol (DPPC)68 compared to our 6461
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B simulations is 87 and 6 Å2/mol fraction of cholesterol, respectively. Although the surface area per lipid decreases at elevated cholesterol levels, the effect is small compared to lower concentrations of cholesterol. As shown in Figure 1, there exists some minor undulation in our bilayers that might result in 25% change69 in the real vs projected surface area. However, the individual lipid areas slightly increase from the 1:1 to 1:2 bilayers (Table 2). Modeling results on DPPC/cholesterol bilayers suggest that the partial specific area of DPPC reaches a minimum at 50% cholesterol.68 On the basis of our MD simulations, the specific area of the phospholipid then slowly increases when the cholesterol concentration is increased beyond 50%. This suggests that the condensing effect of cholesterol is maximized at 50% for lipids with saturated acyl chains. The slight increase in the specific surface area is likely the result of cholesterol-only domain formation (see detailed discussion below), where selfpacking of cholesterol dominates over the cholesterolphospholipid chain packing. In addition to affecting the overall structure, our simulations indicate that increasing the amount of cholesterol at high concentrations does not affect the SCD's of the aliphatic chain for C2C8 but does influence the order at the end of the chain (Figure 2). Specifically, as you increase the amount of cholesterol, the order of the end carbons (C9C14) decreases, which is consistent with experiments on similar bilayers at lower concentrations (3:1 to 1:1).9 This is opposite to the general trend of increasing order with increasing surface area for pure bilayers,70 but as DMPC becomes more isolated in a sea of cholesterol, this likely results in more conformational flexibility for the end chain carbons. The density profiles for these bilayers change from a profile that looks fairly similar to other lipid bilayers with a single maximum52,71 to one that has a global and local maximum (Figure 3). As the concentration of DMPC decreases, the density of the headgroup peak also decreases and a local density maximum appears due to the increased density of the sterol rings (Figure S3, Supporting Information). Although the umbrella model for phospholipidcholesterol interaction can be used to describe observed experimental measurements of cholesterol activity up to the solubility limit,21 the model suggests that these form regular patterned distributions of lipids rather than domains. At 50 and 66.7%, cholesterol lattice-based Monte Carlo simulations suggest a regular hexagonal and maze pattern, respectively. This conceptually is thought to be the result of the PC head groups forming lattice-like structures to prevent unfavorable solvent exposure to cholesterol that naturally forms a crystal monohydrate at cholesterol concentrations above 66.7%. However, it appears that this model may not be valid for elevated cholesterol concentrations based on our MD simulations and that by de Meyer et al.17 Instead of forming patterned regular structures, our 1:2 MD simulations demonstrate the existence of cholesterol-only and equimolar domains. The RDFs for the 1:2 simulation (Figure 4 and Figure S4 of the Supporting Information) contain peaks only at multiples of the first cholesterolcholesterol peak (up to three additional solvation shells). However, the position of the secondary peak in the 1:1 MD simulation is not a multiple of the primary peak. This analysis suggests that there exists an isotropic uniformity in cholesterol distances for the 1:2 simulation, i.e., a nanoscale domain of only cholesterol. This is clearer in Figure S5 of the Supporting Information, where small, nearly hexagonally packed domains (green highlight forming the secondary peak in the RDFs) are connected by neighboring lines of cholesterol
ARTICLE
(yellow highlight forming the tertiary peak in the RDFs). If a more uniform regular distribution were to exist as predicted by the umbrella model (a maze-like structure for 66.7% cholesterol), then the RDFs and snapshots indicating a cholesterol-only phase would not exist. Our simulations are in better agreement with the formation of ICDs proposed by Troup et al. in their DMPCcholesterol experiments,23,24 but they do have some maze-like characteristics connecting these domains (yellow highlight in Figure S5 of the Supporting Information). These hexagonally packed domains are interconnected with a web-like structure of 12 cholesterol molecules thick. Although our MD simulations do form these ICDs at the 1066 ns time scale and are likely not the result of our initial conditions, these may be metastable states that could eventually form regular patterns or larger hexagonal domains. The time scales for our all-atom simulations prohibit longer μs simulations that may be required to determine the stability of these domains. Aside from structure, the internal dynamics of DMPC in these lipid membranes based on CH relaxation times suggests that a PC lipid can be separated into three regions, i.e., choline, main body (carbonyl-glycerol to C11), and end of a chain. This is similar to the decoupled headgroup from the lipid body model presented previously.72 The choline T1 relaxation times increase with the concentration of cholesterol, and the 1:2 simulation is higher than pure DPPC. The environment seen by choline is changed when DMPC is a minor component, and consequently, all relaxation times (slow to fast) decrease (Table S1, Supporting Information). With a lower concentration of DMPC, the choline is freer to move and has less interaction with neighboring DMPC molecules approaching that of choline freely moving in water. The lipid main body relaxation rates are independent of cholesterol concentration (Figure 5) at 75.4 MHz, which probes motions on the order of 0.510 ns. These slower motions are likely the result of lipid axial motions known as wobbling in a cone.72 Although our results indicate that axial relaxation may be somewhat invariant of cholesterol concentration, further analysis is needed and beyond the scope of this work. It is clear that the end of the chain does behave quite differently from a pure lipid bilayer. Since these DMPC/cholesterol membranes are in a liquid ordered state, only carbons at the end have some freedom to relax their conformation but not to the extent as seen with liquid crystalline DPPC bilayers (Figure 5). It appears that the slower motions of the lipid body (lipid wobble) propagate further down the chain and influence the relaxation of these carbons toward the end of the chain. The integral membrane protein to the ocular lens, AQP0, has specifically evolved to function in the unusual environment similar to the DMPC/cholesterol membranes discussed above. However, previous simulation studies on AQP0 have involved lipid membranes without cholesterol and with only one lipid, such as POPE73 or POPC.74 In addition, an experimental twodimensional X-ray structure was solved for AQP0 in DMPC31,60 and E. coli polar lipid60 membranes, both without cholesterol. On the basis of the fundamental properties that change dramatically for membranes with a majority of cholesterol, such as density, lipid order parameters, and head-to-head spacing, simulations at nonphysiological conditions will at least perturb the membrane environment near a protein that is known to locate in lipid raft domains.75 From X-ray diffraction with membranes containing only phospholipids, the main membraneprotein interaction is van der Waals (vdW) contacts of the acyl chains with nonpolar and 6462
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B aromatic residues.31,60 Hydrogen bonding and electrostatic interactions with the glycerol and choline moieties exist but do not appear to form a lipid binding motif.60 Although vdW contacts do occur between the surface residues of the AQP0 tetramer and DMPC, the aromatic residues such as Trp and Phe allow for stronger interactions with the rings on cholesterol (Figure 7BD). A majority of these stacking interactions exist in the lower leaflet in Figure 7A (six vs four) with the most stable contacts for Phe189, Trp-198, Trp-202, and Trp-205. The most stable protein cholesterol vdW contacts with the upper leaflet are Trp-10 and Phe-221. There appears to be a preference to forming cholesterol protein interactions in the lower leaflet half of AQP0. In addition to vdW contacts, the majority of cholesterol in these bilayers promotes hydrogen bonds and their stability or lifetimes (Table 3). On average, there is 0.95 hydrogen bonds formed between each AQP0 monomer and the lipid membrane in the pure DMPC membrane but this increases to 1.25 for the 1:1 membrane. In addition to occupancy, the less fluid 1:1 membrane has longer hydrogen bond lifetimes compared to the DMPC-only membrane (12.5 vs 31.3 ps). Ser-106 and Arg-196 are the two most common residues to form hydrogen bonds with the carbonyl of DMPC or the hydroxyl group of cholesterol. For DMPC membranes with cholesterol, cholesterol also has the advantage as a hydrogen bond donor or acceptor with its interactions with AQP0, where DMPC can only accept hydrogen bonds. Considering the combination of polar and ringring stacking interactions, AQP0 forms more contacts with the surrounding membrane for a longer duration when the membrane contains cholesterol. It should also be noted that membranes with sphingomyelin, a lipid commonly found in the ocular lens, offer the ability to be a hydrogen bond donor and acceptor as cholesterol. Therefore, SM might interact more favorably with AQP0 than compared to other lipids, such as DMPC, POPC, or POPE, and our DMPC/cholesterol studies are a first step in understanding AQP0lipid membrane interactions with the ocular lens. Future studies of this integral membrane protein in a more realistic model of the ocular lens will be performed once CHARMM parameters for sphingomyelin are finalized.
’ ASSOCIATED CONTENT
bS
Supporting Information. Figures of surface area vs time, x-dimension vs time of the proteinlipid simulation, component electron density of lipid bilayers, final XY-plane snapshots of the lipid membranes, and a table of fitted constants for the reorientational correlation function. This material is available free of charge via the Internet at http://pubs.acs.org.
’ AUTHOR INFORMATION Corresponding Author
*E-mail:
[email protected]. Phone: (301) 405-1320. Fax: (301) 314-9126.
’ ACKNOWLEDGMENT This work was supported by institutional funding from the University of Maryland. This research was also supported in part by the National Science Foundation through TeraGrid resources provided by National Institute for Computational Sciences (Kraken) under grant number TG-MCB100139 and the High Performance Computing Cluster at the University of Maryland.
ARTICLE
’ REFERENCES (1) Yappert, M. C.; Borchman, D. Chem. Phys. Lipids 2004, 129, 1. (2) Jacob, R. F.; Cenedella, R. J.; Mason, R. P. J. Biol. Chem. 1999, 274. (3) Zelenka, P. S. Curr. Eye Res. 1984, 3, 1337. (4) van Meer, G.; Voelker, D. R.; Feigenson, G. W. Nat. Rev. Mol. Cell Biol. 2008, 9, 112. (5) Engelman, D. M.; Rothman, J. E. J. Biol. Chem. 1972, 247, 3694. (6) McMullen, T. P. W.; Lewis, R. N. A. H.; McElhaney, R. N. Curr. Opin. Colloid Interface Sci. 2004, 8, 459. (7) Vist, M. R.; Davis, J. H. Biochemistry 1990, 29, 451. (8) Feigenson, G. W. Annu. Rev. Biophys. Biomol. Struct. 2007, 36, 63. (9) Sankaram, M. B.; Thompson, T. E. Biochemistry 1990, 29, 10676. (10) Mathai, J. C.; Tristram-Nagle, S.; Nagle, J. F.; Zeidel, M. L. J. Gen. Physiol. 2008, 131, 69. (11) Hung, W. C.; Lee, M. T.; Chen, F. Y.; Huang, H. W. Biophys. J. 2007, 92, 3960. (12) Pan, J.; Mills, T. T.; Tristram-Nagle, S.; Nagle, J. F. Phys. Rev. Lett. 2008, 100, 198103. (13) McMullen, T. P. W.; McElhaney, R. N. Curr. Opin. Colloid Interface Sci. 1996, 1, 83. (14) Epand, R. M. Biophys. J. 2003, 84, 3102. (15) Jacob, R. F.; Cenedella, R. J.; Mason, P. R. J. Biol. Chem. 2001, 276, 13573. (16) Dai, J.; Alwarawrah, M.; Huang, J. J. Phys. Chem. B 2010, 114, 840. (17) de Meyer, F. J.-M.; Benjamini, A.; Rodgers, J. M.; Misteli, Y.; Smit, B. J. Phys. Chem. B 2010, 114, 10451. (18) Radhakrishnan, A.; McConnell, H. M. Biophys. J. 1999, 77, 1507. (19) Huang, J.; Feigenson, G. W. Biophys. J. 1999, 76, 2142. (20) Chong, P. L. Proc. Natl. Acad. Sci. U.S.A. 1994, 91, 10069. (21) Ali, M. R.; Cheng, K. H.; Huang, J. Proc. Natl. Acad. Sci. U.S.A. 2007, 104, 5372. (22) Huang, J.; Buboltz, J. T.; Feigenson, G. W. Biochim. Biophys. Acta, Biomembr. 1999, 1417, 89. (23) Troup, G. M.; Tulenko, T. N.; Lee, S. P.; Wrenn, S. P. Colloids Surf., B 2003, 29, 217. (24) Troup, G. M.; Tulenko, T. N.; Lee, S. P.; Wrenn, S. P. Colloids Surf., B 2004, 33, 57. (25) Bloemendal, H. Science 1977, 197, 127. (26) Gonen, T.; Cheng, T.; Kistler, J.; Walz, T. J. Mol. Biol. 2004, 342, 1337. (27) Zampighi, G. A.; Hall, J. E.; Ehring, G. R.; Simon, S. A. J. Cell Biol. 1989, 108, 2255. (28) Yang, B. X.; Verkman, A. S. J. Biol. Chem. 1997, 272, 16140. (29) Harries, W. E. C.; Akhavan, D.; Miercke, L. J. W.; Khademi, S.; Stroud, R. M. Proc. Natl. Acad. Sci. U.S.A. 2004, 101, 14045. (30) Nemeth-Cahalan, K. L.; Hall, J. E. J. Biol. Chem. 2000, 275, 6777. (31) Gonen, T.; Cheng, Y.; Sliz, P.; Hiroaki, Y.; Fujiyoshi, Y.; Harrison, S. C.; Walz, T. Nature 2005, 438, 633. (32) Chepelinsky, A. B. J. Exp. Zool., Part A 2003, 300A. (33) Klauda, J. B.; Venable, R. M.; Freites, J. A.; O’Connor, J. W.; Tobias, D. J.; Mondragon-Ramirez, C.; Vorobyov, I.; MacKerell, A. D., Jr.; Pastor, R. W. J. Phys. Chem. B 2010, 114, 7830. (34) Khare, R. S.; Worthington, C. R. Biochim. Biophys. Acta 1978, 514. (35) Jo, S.; Kim, T.; Iyer, V. G.; Im, W. J. Comput. Chem. 2008, 29, 1859. (36) Jo, S.; Lim, J. B.; Klauda, J. B.; Im, W. Biophys. J. 2009, 97, 50. (37) Jo, S.; Kim, T.; Im, W. PLoS One 2007, 2, e880. (38) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; Swaminathan, S.; Karplus, M. J. Comput. Chem. 1983, 4, 187. (39) Brooks, B. R.; Brooks, C. L.; Mackerell, A. D.; Nilsson, L.; Petrella, R. J.; Roux, B.; Won, Y.; Archontis, G.; Bartels, C.; Boresch, S.; Caflisch, A.; Caves, L.; Cui, Q.; Dinner, A. R.; Feig, M.; Fischer, S.; Gao, 6463
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
The Journal of Physical Chemistry B J.; Hodoscek, M.; Im, W.; Kuczera, K.; Lazaridis, T.; Ma, J.; Ovchinnikov, V.; Paci, E.; Pastor, R. W.; Post, C. B.; Pu, J. Z.; Schaefer, M.; Tidor, B.; Venable, R. M.; Woodcock, H. L.; Wu, X.; Yang, W.; York, D. M.; Karplus, M. J. Comput. Chem. 2009, 30, 1545. (40) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L. J. Chem. Phys. 1983, 79, 926. (41) Phillips, J. C.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhorshid, E.; Villa, E.; Chipot, C.; Skeel, R. D.; Kale, L.; Schulten, K. J. Comput. Chem. 2005, 26, 1781. (42) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.; Pedersen, L. C. J. Chem. Phys. 1995, 103, 8577. (43) Andersen, H. C. J. Comput. Phys. 1983, 52, 24. (44) Feller, S. E.; Pastor, R. W. J. Chem. Phys. 1999, 111, 1281. (45) Klauda, J. B.; Roberts, M. F.; Redfield, A. G.; Brooks, B. R.; Pastor, R. W. Biophys. J. 2008, 94, 3074. (46) Klauda, J. B.; Eldho, N. V.; Gawrisch, K.; Brooks, B. R.; Pastor, R. W. J. Phys. Chem. B 2008, 112, 5924. (47) Pastor, R. W.; Venable, R. M.; Feller, S. E. Acc. Chem. Res. 2002, 35, 438. (48) Lipad, G.; Szabo, A. J. Am. Chem. Soc. 1982, 104, 4546. (49) Pandit, S. A.; Vasudevan, S.; Chiu, S. W.; Mashl, R. J.; Jakobsson, E.; Scott, H. L. Biophys. J. 2004, 87, 1092. (50) Shinoda, W.; Okazaki, S. J. Chem. Phys. 1998, 109, 1517. (51) Nagle, J. F.; Tristram-Nagle, S. Biochim. Biophys. Acta, Rev. Biomembr. 2000, 1469, 159. (52) Kucerka, N.; Liu, Y. F.; Chu, N. J.; Petrache, H. I.; TristramNagle, S. T.; Nagle, J. F. Biophys. J. 2005, 88, 2626. (53) Pan, J.; Tristram-Nagle, S.; Nagle, J. F. Phys. Rev. E: Stat., Nonlinear, Soft Matter Phys. 2009, 80, 021931. (54) Nevzorov, A. A.; Trouard, T. P.; Brown, M. F. Phys. Rev. E 1998, 58, 2259. (55) Douliez, J. P.; Leonard, A.; Dufourc, E. J. Biophys. J. 1995, 68, 1727. (56) Trouard, T. P.; Nevzorov, A. A.; Alam, T. M.; Job, C.; Zajicek, J.; Brown, M. F. J. Chem. Phys. 1999, 110, 8802. (57) Needham, D.; McIntosh, T. J.; Evans, E. Biochemistry 1988, 27, 4668. (58) Hodzic, A.; Rappolt, M.; Amenitsch, H.; Laggner, P.; Pabst, G. Biophys. J. 2008, 94, 3935. (59) Klauda, J. B.; Kucerka, N.; Brooks, B. R.; Pastor, R. W.; Nagle, J. F. Biophys. J. 2006, 90, 2796. (60) Hite, R. K.; Li, Z.; Walz, T. EMBO J. 2010, 29, 1652. (61) Bennett, W. F. D.; MacCallum, J. L.; Tieleman, D. P. J. Am. Chem. Soc. 2009, 131, 1972. (62) Carrillo-Tripp, M.; Feller, S. E. Biochemistry 2005, 44, 10164. (63) Cournia, Z.; Ullmann, G. M.; Smith, J. C. J. Phys. Chem. B 2007, 111, 1786. (64) Risselada, H. J.; Marrink, S. J. Proc. Natl. Acad. Sci. U.S.A. 2008, 105, 17367. (65) Zhang, Z.; Lu, L.; Berkowitz, M. L. J. Phys. Chem. B 2008, 112, 3807. (66) Khelashvili, G.; Pabst, G.; Harries, D. J. Phys. Chem. B 2010, 114, 7524. (67) Marrink, S. J.; deVries, A. H.; Harroun, T. A.; Katsaras, J.; Wassall, S. R. J. Am. Chem. Soc. 2008, 130, 10. (68) Edholm, O.; Nagle, J. F. Biophys. J. 2005, 89, 1827. (69) Lindahl, E.; Edholm, O. Biophys. J. 2000, 79, 426. (70) Feller, S. E.; Venable, R. M.; Pastor, R. W. Langmuir 1997, 13, 6555. (71) Kucerka, N.; Tristram-Nagle, S.; Nagle, J. F. Biophys. J. 2006, 90, L83. (72) Klauda, J. B.; Pastor, R. W.; Brooks, B. R.; Roberts, M. F.; Redfield., A. G. Biophys. J. 2008, 94, 3074. (73) Jensen, M. Ø.; Dror, R. O.; Xu, H.; Borhani, D. W.; Arkin, I. T.; Eastwood, M. P.; Shaw, D. E. Proc. Natl. Acad. Sci. U.S.A. 2008, 105, 14430. (74) Hashido, M.; Ikeguchi, M.; Kidera, A. FEBS Lett. 2005, 579, 5549.
ARTICLE
(75) Tong, J.; Briggs, M. M.; Mlaver, D.; Vidal, A.; McIntosh, T. J. Biophys. J. 2009, 97, 2493.
6464
dx.doi.org/10.1021/jp108650u |J. Phys. Chem. B 2011, 115, 6455–6464
