Rotational decoupling between the hydrophilic and hydrophobic regions in lipid membranes

Abstract

Cells use homeostatic mechanisms to ensure an optimal composition of distinct types of lipids in cellular membranes. The hydrophilic region of biological lipid membranes is mainly composed of several types of phospholipid headgroups that interact with incoming molecules, nanoparticles, and viruses, whereas the hydrophobic region consists of a distribution of acyl chains and sterols affecting membrane fluidity/rigidity related properties and forming an environment for membrane-bound molecules such as transmembrane proteins. A fundamental open question is to what extent the motions of these regions are coupled and, consequently, how strongly the interactions of phospholipid headgroups with other molecules depend on the properties and composition of the membrane hydrophobic core. We combine advanced solid-state nuclear magnetic resonance spectroscopy with high-fidelity molecular dynamics simulations to demonstrate how the rotational dynamics of choline headgroups remain nearly unchanged (slightly faster) with incorporation of cholesterol into a phospholipid membrane, contrasting the well-known extreme slowdown of the other phospholipid segments. Notably, our results suggest a new paradigm in which phospholipid dipole headgroups interact as quasi-freely rotating flexible dipoles at the interface, independent of the properties in the hydrophobic region.

Significance

The interactions of phospholipid headgroups at membrane interfaces are highly important in the context of molecular recognition, ion binding, and regulation of membrane protein function. Here, we demonstrate that the dynamics of phospholipid headgroups are not affected by either the structural or dynamic properties of the membrane hydrophobic core. This suggests a new paradigm in which the dynamic and structural properties of the hydrophilic and hydrophobic regions may be controlled independently.

Introduction

Of the plethora of lipids found in nature, the most ubiquitous are glycerophospholipids, which consist of a glycerol backbone attached to two hydrophobic fatty acid chains and a phosphodiester bridge connecting to a hydrophilic headgroup (1). Cells use vast amounts of energy, and rely on complex synthetic pathways, to adjust and maintain the specific composition of different types of phospholipid headgroups across cellular organelles (2). This chemical homeostasis implies that the headgroups play a key role in fundamental biological processes, and evidence for phospholipid-specific functionality concerning compartmentalization, signaling, transport, ion binding, peptide insertion, and regulation of membrane protein function has been found (2, 3, 4, 5). However, the molecular details on how lipid composition of a cellular membrane connects to its overall properties and to specific biological processes remain poorly understood.

A fundamental open question is to what degree the behavior of the water-facing headgroups in biological membranes correlate with the properties of the acyl chains in the membrane hydrophobic core (6, 7, 8, 9, 10, 11, 12, 13). Two limiting cases can be considered: 1) the conformational ensemble and dynamics of the headgroup, although positionally connected through the glycerol backbone, are uncoupled from the acyl chain region (freely rotating/weak coupling limit), or 2) the orientation and dynamics of the headgroup and hydrophobic acyl chains are strongly interdependent (strong coupling limit). These two limiting scenarios will give rise to very distinct biophysical behavior. In the case of strong coupling, the headgroups in lipid domains with ordered acyl chains and hindered dynamics, such as cholesterol-induced lipid rafts (14,15), would exhibit slower dynamics and possibly a different conformational ensemble. The acyl chain behavior would then indirectly affect the interactions between lipid headgroups and molecules in the aqueous media or within the membrane, such as proteins or drugs. Such scenario is implicit, for example, in the popular “umbrella model” for lipid-cholesterol interactions (4,7,9). In the weak coupling limit, the behavior of lipid headgroups is similar irrespective of the acyl chain structure, order, and dynamics, and consequently any cellular processes that depend on the conformation and dynamics of the headgroups are unaffected by the properties of the hydrophobic region.

We address the question of headgroup-tail decoupling using a phosphatidylcholine (PC)-cholesterol bilayer system (the most abundant phospholipid and sterol in eukaryotic cells), a model cellular membrane from which a wealth of both experimental and simulation data can be obtained. Cholesterol is known to drive lateral heterogeneity and make membranes more ordered (the so-called cholesterol condensing effect), which manifests as a substantial increase in hydrophobic acyl chain C–H bond order parameters ($S_{\text{CH}}$) in NMR experiments (16, 17, 18, 19, 20, 21, 22, 23, 24, 40).

In contrast, the headgroup and glycerol backbone order parameters are essentially unaffected up to the highest cholesterol concentrations possible to incorporate in PC membranes (25,26). The α-carbon order parameter of the choline headgroup, in particular, remains unchanged also upon other bilayer perturbations that significantly affect acyl chain order parameters, such as temperature, acyl chain composition, or membrane phase (Table S1 in the supporting material and references therein). On the other hand, the choline headgroup orientation (and consequently the headgroup order parameters) is highly sensitive to hydration level (27), hydrostatic pressure (28), and the inclusion of charges (29, 30, 31) or molecular dipoles in the membrane (32). Therefore, a picture of a rotationally decoupled headgroup, whose orientation is independent of the hydrophobic region but can be affected by the environment, emerges.

Although the C–H bond order parameters contain accurate information on the conformational ensemble, they do not convey how fast that ensemble is sampled (conformational dynamics). The motional timescales have been dominantly assessed through spin-lattice relaxation ($R_1$) and spin-lattice relaxation in the rotating frame ($R_{1\rho }$) NMR measurements (10,11,33, 34, 35, 36, 37, 38) which are sensitive to different (limited) timescales depending on experimental conditions and from which a physically meaningful change in the dynamics (speedup versus slowdown) can be challenging to interpret without multiple measurements under different magnetic fields. Such measurements demonstrate that the cholesterol-induced order in acyl chains is accompanied by slower rotational dynamics not only of the acyl chains but also of the glycerol backbone segments for which there is only a marginal conformational change (11,37).

However, the impact of cholesterol on the headgroup motional timescales, potentially occurring either via direct interaction or through the observed glycerol backbone slowdown, has remained unclear. An observation of increase of headgroup ¹³C $R_{1\rho }$ rates of DMPC upon incorporation of cholesterol (38) suggests that the cholesterol-induced slowdown of tail dynamics propagates to the phospholipid headgroups although neither the statistical significance nor the quantitative interpretation of the $R_{1\rho }$ increase in terms of physically meaningful correlation times was provided. In stark contrast, comparison of ¹³C cross-polarization (CP) and refocused insensitive nuclear enhanced polarization transfer (rINEPT) intensities suggest that the headgroup motional timescales remain unchanged even by the addition of 50% cholesterol (23). Very recently, Roberts and coworkers (39) reported that the choline moiety is significantly faster than the glycerol backbone on the tens of nanoseconds timescale in the presence of cholesterol.

Here, we show that the dynamics of the PC headgroup are unaffected by cholesterol, and consequently, that the motion of phospholipid headgroups is decoupled from the hydrophobic region (the freely rotating limit). To this end, we employ our novel NMR methodology (36) in which segmental effective correlation times (${\tau }_{\text{e}}$) are determined from solid-state NMR measurements of $R_1$, $R_{1\rho }$, and $S_{\text{CH}}$. The analysis of ${\tau }_{\text{e}}$ values enables us to interpret the relaxation rates in terms of a single, physically meaningful average timescale for each carbon where lowered and increased ${\tau }_{\text{e}}$ values denote speedup and slowdown of motion, respectively. In addition, we decipher the origin of the decoupled motion by analyzing distinct all-atom (CHARMM36 and Slipids) and united-atom (Berger) MD simulations that provide either realistic uncoupled (CHARMM36 and Slipids) or nonrealistic coupled (Berger) headgroup motions. The MD simulation models indicate that the decoupled motion originates from dihedral rotations that are present in all other glycerophospholipid types in addition to PCs. This suggests that biological membranes have independent rotational dynamics of the headgroups from the hydrophobic region, a feature that may be relevant in the machinery of biological cell membrane processes.

Materials and methods

Sample preparation

1-palmitoyl,2-oleoyl-$sn$-glycero-3-phosphocholine (POPC), cholesterol and chloroform were purchased from Sigma-Aldrich (St. Louis, MO). The samples were prepared by mixing the lipids with chloroform and rapidly evaporating the organic solvent under a nitrogen gas flow to obtain a homogeneous lipid film. Subsequently, the lipid film was dried under vacuum overnight. The film was then hydrated in a 0.5-mL tube by adding 50 %wt of water and manually mixing with a thin metal rod multiple times alternated by sample centrifugation until a homogeneous mixture was attained. The resulting mixture was then centrifuged into a KEL-F Bruker insert with a sample volume of approximately 25 μL specifically designed for solid-state NMR 4-mm rotors and left to equilibrate for at least 24 h at room temperature before measurements.

NMR experiments

The solid-state NMR experiments to measure $R_1$ and $R_{1\rho }$ were performed on a Bruker Avance II-500 NMR spectrometer operating at a ¹³C Larmor frequency of 125.78 MHz equipped with an E-free CP-magic-angle spinning (MAS) 4 mm (¹³C/³¹P/¹H). The R-PDLF measurements were performed on a Bruker Avance III 400 spectrometer operating at a ¹H Larmor frequency of 400.03 MHz equipped with a standard 4-mm CP-MAS HXY probe. All experiments were performed under MAS conditions at a rate of 5 kHz and a temperature of 298 K. The R-PDLF, $R_1$, and $R_{1\rho }$ experiments were performed as previously described in (36). More details on experimental set up are given in the supporting material.

MD simulations

We performed MD simulations for two systems: a pure POPC bilayer and a bilayer containing additional 50% cholesterol, both accompanied by enough water molecules per lipid to result in fully hydrated bilayers. We used three lipid MD models (force fields): CHARMM36 (41), Slipids (42), and Berger (43)/Höltje (23,44) together with either TIP3P (45,46) or SPC (47) water. The choice of force fields was based on previous works in which their ability to capture structure (48) and dynamics (49) of lipid headgroup and glycerol backbone was assessed against NMR measurables. The simulations were performed using the GPU-version of Gromacs2020 (50) MD engine, with sampling rate of 10 ps and maintaining 303 K temperature. The list of all simulated systems, along with the trajectory lengths and links to the freely available simulation data, is given in Table 1. Further details of the simulations are presented in the supporting material.

Table 1.

Summary of Simulated Systems: The Force Fields Used, Numbers of POPC, Cholesterol, and Water Molecules, Trajectory Lengths, and Access Links to the Simulation Files

Force field POPC/water + cholesterol	POPC/chol	Water	Length (ns)	Files
CHARMM36 (41)/TIP3P (46)	122/0	4480	840	(51)
Slipids (42)/TIP3P (45)	122/0	4480	1200	(52)
Berger-POPC-07(43)/SPC (47)	256/0	10,342	1200	(53)
CHARMM36 (41)/TIP3P (46)+CHARMM36 (54)	122/122	9760	1240	(51)
Slipids (42)/TIP3P (45)+Slipids (55)	122/122	9760	1200	(56)
Berger-POPC-07 (43)/SPC (47)+Höltje-CHOL-13 (23,44)	256/256	20,480	1200	(53)

Results

Experimental demonstration of the uncoupled motion

Fig. 1 shows the effect of cholesterol on both the dynamics and structure of POPC headgroup and glycerol backbone. Chemical shift resolution for all the distinct carbons in the refocused-INEPT spectra displayed in Fig. 1 A is enabled by simultaneous MAS and heteronuclear decoupling. The effect of cholesterol on the phospholipid dynamics is assessed by measuring the $R_1$ and $R_{1\rho }$ values from POPC and POPC/cholesterol (1:1) multilamellar vesicles and the effect on phospholipid structure by R-PDLF spectroscopy. The relaxation rates for the headgroup and glycerol backbone are shown in Fig. 1 B. The complete set of decays and relaxation rates measured for all the phospholipid segments resolved in the ¹³C spectrum is given in the supporting material in Fig. S1 (headgroup and glycerol backbone) and Fig. S2 (acyl chains). The dipolar splittings used to calculate the $S_{\text{CH}}$ order parameters of headgroup and glycerol backbone are presented in Fig. 1 C. The resulting order parameters confirm the previously reported values (23).

Figure 1.

Effect of cholesterol on the dynamics (B) and structure (C) of headgroup (α, β and γ) and glycerol backbone ($g_1$, $g_2$, and $g_3$) carbons in POPC lipid membranes. (A) Chemical structure of POPC with carbon labels and refocused-INEPT spectra from POPC membranes with (red) and without (black) cholesterol. (B) ¹³C spin-lattice relaxation rates, $R_1$, and spin-lattice relaxation in the rotating frame rates, $R_{1\rho }$, showing the independent motion of the choline headgroup and slowdown of the glycerol backbone upon cholesterol incorporation. A Larmor frequency of 500 MHz for ¹H nuclei and a spin lock field equal to 50 kHz were used. The corresponding experimental decays for each data value are shown in Fig. S1. The definition of the error bars is described in the supporting material. (C) Dipolar recoupling profiles acquired with R-PDLF spectroscopy from POPC membranes with (red) and without (black) cholesterol. Similar magnitudes of the splittings indicate structural independence of both the choline headgroup and glycerol backbone on cholesterol incorporation. Note that the line shapes are highly sensitive to the experimental setup and that the relevant information on conformations is in the splittings, which are proportional to $\left|S_{\text{CH}}\right|$. (D) Overlayed lipid conformations in MD simulations with (right) and without (left) cholesterol (yellow), illustrating the experimental observations.

$R_1$ rates remain constant for both the glycerol backbone and the headgroup, showing that the C–H bond motions with timescales close to ns are not affected by cholesterol for these carbons. On the other hand, the $R_{1\rho }$ rates in the glycerol backbone (carbons $g_1$, $g_2$, and $g_3$) increase by approximately a factor of two upon cholesterol addition. In sharp contrast, the $R_{1\rho }$ values for the choline headgroup (α, β, and γ segments) are unaffected by cholesterol and significantly lower than in the glycerol backbone. The invariance of both the headgroup carbon dipolar couplings (Fig. 1 C) and the relaxation rates (Fig. 1 B) upon cholesterol incorporation implies that the conformational ensemble and the time required to span all the available headgroup conformations is the same irrespective of the glycerol backbone slowdown induced by cholesterol. For acyl chains, both structural ($S_{\text{CH}}$) and dynamic observables ($R_1$ and $R_{1\rho }$) vary with incorporation of cholesterol (Figs. S2 and S3), as expected from the ordering and slowdown of acyl chains in line with previously reported results (11,23).

The $R_1$ and $R_{1\rho }$ rates are linear combinations of spectral density terms determined from the Fourier transform, $\mathcal{F}\left\{g\left(\tau \right)\right\}$, of the autocorrelation function for C–H bond reorientation (36)

$$g\left(\tau \right)=⟨P_2\left[\overrightarrow{\mu }\left(t\right)\cdot \overrightarrow{\mu }(t+\tau )\right]⟩,$$ (1)

where $P_2$ denotes the second Legendre polynomial, $P_2\left(\xi \right)=1/2\left(3{\xi }^2-1\right)$, $\overrightarrow{\mu }\left(t\right)$ is the unitary vector having the direction of the C–H bond at time t, and where the angular brackets denote a time average. For large lipid vesicles, $g\left(\tau \right)$ first decays to a plateau equal to $S_{\text{CH}}^2$ and then decays to zero at longer times due to the isotropic diffusion around the lipid vesicle. The initial decay toward the plateau can be written as a multi-exponential decay

$$g_f\left(\tau \right)=\left(1-S_{\text{CH}}^2\right){\int }_0^{+\infty }p\left({\tau }_f\right)\text{exp}\left(-\tau /{\tau }_f\right)d{\tau }_f+S_{\text{CH}}^2,$$ (2)

where $p\left({\tau }_f\right)$ is a weight function that depends only on the specific motion of the C–H bond. $g_f\left(\tau \right)$ is accessible within the timescales covered by all-atom MD simulations presently. From this expression an effective correlation time can be defined as follows:

$${\tau }_e:={\int }_0^{+\infty }p\left({\tau }_f\right){\tau }_{f}d{\tau }_f={\int }_0^{+\infty }\frac{g_f\left(\tau \right)-S_{\text{CH}}^2}{1-S_{\text{CH}}^2}d\tau$$ (3)

which is a quantitative measure of the time needed for the C–H bond to sample all its possible orientations; ${\tau }_e$ may be estimated from the experimental data presented in panels B and C of Fig. 1 through (36)

$${\tau }_e=\frac{5R_{1\rho }-3.82R_1}{4{\pi }^2{d}_{\text{CH}}^{2}N\left(1-S_{\text{CH}}^2\right)},$$ (4)

where N denotes the number of protons covalently bound to the carbon and the coupling constant $d_{\text{CH}}$ is approximately −22 kHz.

Fig. 2 shows ${\tau }_e$ calculated using Eq. 4 for both the headgroup and glycerol backbone segments. The ${\tau }_e$ values of the choline headgroup are within 0.1 to 0.5 ns and remain constant within the experimental accuracy upon addition of 50% cholesterol, while the ${\tau }_{\text{e}}$ values for the glycerol backbone show a slowdown with a factor of approximately two, and are one order of magnitude higher than in the headgroup.

Figure 2.

Impact of cholesterol (open bars: pure POPC, solid bars: 50% POPC+50% cholesterol) on the effective correlation times, ${\tau }_{\text{e}}$, of different carbons in the headgroup and glycerol backbone of POPC quantified experimentally using Eq. 4 and from lipid bilayer MD simulations with the CHARMM36, Slipids, and Berger force fields using Eq. 3. Note the different y-scales used on the left and right plots to appreciate the significant difference of effective correlation times for the choline headgroup (0.1–0.5 ns) and glycerol backbone segments (2–5 ns). The experimental error bars are the cummulative error from using Eq. 4 with the R₁ and R_1ρ values in Fig. 1. The simulation error bars are as described in reference (49).

Comparison of experiments with MD simulations

Also included in Fig. 2 are the ${\tau }_{\text{e}}$ values calculated from three sets of MD simulations using the CHARMM36, Slipids, and Berger force fields. Notably, CHARMM36 simulations reproduce the experimental ${\tau }_{\text{e}}$ values of the choline headgroup, both with and without cholesterol, almost flawlessly within experimental uncertainty. CHARMM36 also captures well the experimental choline C–H bond order parameters which remain the same with the presence of the sterol (Fig. 3). For the glycerol backbone, CHARMM36 simulations slightly overestimate the slowdown of the effective correlation times, but give the best structural model among the three force fields used. Slipids simulations show the best agreement with experiments for the effect of cholesterol on the glycerol backbone ${\tau }_{\text{e}}$ values, although they fail to capture the glycerol backbone structure, and overestimate the effective correlation time for the γ carbon.

Figure 3.

Effect of cholesterol (open symbols: pure POPC, solid symbols: 50% POPC+50% cholesterol) on the POPC headgroup and glycerol backbone C–H bond order parameters. The experimental values were determined by R-PDLF spectroscopy. The gray areas show the range of $S_{\text{CH}}$ values for PC bilayer systems with and without cholesterol reported to date (see, e.g., (22,23)). For comparing with the effect on the acyl chains, see Fig. S3. The experimental error bars were defined as ±0.02 to account for the maximum mismatch reported between ¹H-¹³C dipolar recoupling measurements and ²H NMR measurements.

The Berger force field clearly produces the least realistic dynamics, giving a significant overestimation of ${\tau }_{\text{e}}$ for the choline headgroup segments (both with and without cholesterol) and predicts an erroneous, large (approximately fivefold) cholesterol-induced slowdown of the choline headgroup dynamics (see Fig. S4 for ${\tau }_e$ from the Berger simulations drawn to scale). Both the structural and dynamic force field properties observed here are in line with the previously reported (48,49).

In contrast to the Berger model, both the CHARMM36 and Slipids models, although not perfect, capture the key experimental observation in this work: The structure and dynamics of the choline headgroup are not affected by incorporation of cholesterol despite the increased acyl chain order and hindered dynamics in the acyl chains and glycerol backbone, i.e., the headgroup is uncoupled from the glycerol backbone in these models.

Effect of cholesterol on the timescale of internal motions of the headgroup and glycerol backbone

Knowing their ability to reproduce the NMR measurables, we proceeded to exploit the temporal and spatial resolution in the distinct MD models to gain insight on the rotational dynamics of specific sites in the molecules as well as the origin and degree of (un)coupling between the headgroup and the rest of the phospholipid. To this end, Fig. 4 shows the distributions of selected headgroup and glycerol dihedral angles, ϕ, and the corresponding dihedral effective correlation times, ${\tau }_c^{\varphi }$, which are extracted from autocorrelation functions

$$G^{\varphi }\left(\tau \right)=⟨\cos \varphi \left(t\right)\cos \varphi \left(t+\tau \right)⟩,$$ (5)

where the angular brackets denote an average over time and over the number of molecules in the system. We define the dihedral effective correlation time (${\tau }_c^{\varphi }$) as simply the area under the reduced normalized autocorrelation function

$$g^{\varphi }\left(\tau \right)=\frac{G^{\varphi }\left(\tau \right)/G^{\varphi }\left(0\right)-⟨\cos \varphi {⟩}^2}{1-⟨\cos \varphi {⟩}^2},$$ (6)

where $⟨\text{cos}\varphi {⟩}^2$ is the value of $G^{\varphi }\left(\tau \right)$ at infinitely long τ. ${\tau }_c^{\varphi }$ provides a measure of how much time is needed for dihedral motion to sample its angle distribution.

Figure 4.

Effect of cholesterol on internal structure and dynamics of POPC. (A) Dihedral angle distributions for pure POPC membranes (darker thick lines) and POPC/cholesterol membranes (lighter thin lines) from MD simulations using the CHARMM36 (red), Slipids (blue), and Berger (green) force fields. (B) Reduced and normalized dihedral torsion autocorrelation functions (see Eq. 6) showing their corresponding dihedral effective correlation times (${\tau }_c^{\varphi }$). Note that the extremely short correlation times for the outermost right column are due to the very narrow angle range accessible for this torsion dihedral.

For the more realistic force fields (CHARMM36 and Slipids), cholesterol does not affect these dihedral distributions as expected from the $S_{\text{CH}}$ data alone. More interestingly, for both of these force fields, the dihedral angles over the phosphate linkage between the glycerol and the alpha carbon, $\left(g_3\right)$O–P and P–O(α), and the glycerol backbone dihedral $g_2$–$g_3$, have no angles that are unavailable, in contrast to the other dihedrals analyzed. Furthermore, the ($g_3$)O–P and P–O(α) exhibit fast correlation times (${\tau }_c^{\varphi }\le$ 0.5 ns) in CHARMM36, which are very close to the effective correlation times in Fig. 2.

The most notable differences between the more realistic simulations and Berger simulations are the slower ${\tau }_c^{\varphi }$ in Berger, and how these are affected by cholesterol. While in CHARMM36 and Slipids only minor changes are observed, with a slight speedup of the sampling by cholesterol, in the Berger model these internal dynamics slow down considerably.

Polar and azimuthal motion of the choline dipole and of the glycerol backbone

To investigate the correlation between motions of headgroup and other parts of lipid molecules, we quantified the autocorrelation functions of the polar and azimuthal angles, θ and φ (coordinate system where the z-direction coincides with membrane normal), for a number of selected vectors between intramolecular atomic pairs. The definition of the autocorrelation functions $G^{\theta }\left(\tau \right)$ and $G^{\phi }\left(\tau \right)$ are the same as in Eq. 5 but using θ and φ as angles, respectively. Note that for $G^{\theta }\left(\tau \right)$ a nonzero plateau at the long τ is expected because the different θ angles are not equally likely to occur. On the other hand, $G^{\phi }\left(\tau \right)$ is always zero at long τ due to the lipid uniaxial motion.

In Fig. 5, we show these correlation functions for the choline dipole orientation, P$\to$N, and for the interatomic vector connecting the carbonyl carbons in the sn-1 and sn-2 positions, calculated from the most realistic force field (CHARMM36). The φ autocorrelation functions clearly show the contrasting effects of cholesterol on these vectors. Although for the choline dipole a speedup of reorientational motion is observed, the φ dynamics of the vector connecting the carbonyl carbons become slower by almost an order of magnitude. To extract effective correlation times ${\tau }_c^{\theta }$ and ${\tau }_c^{\phi }$, we again integrate the reduced and normalized autocorrelation functions. Although cholesterol induces more than a fourfold slowdown for the φ dynamics of the vector connecting carbonyl carbons, the correlation times of the P$\to$N orientation remain essentially the same. The complete set of reduced autocorrelations functions analyzed is given in Figs. S7–S13 together with θ distributions, and ${\tau }_c^{\theta }$ and ${\tau }_c^{\phi }$ values.

Figure 5.

The effect of cholesterol on the timescales for the reorientations of P$\to$N and (sn-1)O=C$\to$C=O(sn-2) vectors over the spherical angles φ and θ. (A) Definition of the angles and the vectors connecting the atoms. (B and C) Autocorrelation functions for angles (B) θ and (C) φ both with and without cholesterol. The autocorrelation functions shown here are described in Eq. 5.

Discussion

Our experimental and MD simulation results show that the phospholipid headgroup conformational ensemble and dynamics remain unaffected by addition of 50% cholesterol to the lipid membrane, despite the significant acyl chain ordering and reduction of both the acyl chain and glycerol backbone dynamics. Therefore, our results do not support models that contain interdependence between the structure or dynamics of the hydrophobic and hydrophilic regions of cellular or model phospholipid membranes.

The observed slowdown of the glycerol backbone and acyl chains upon addition of cholesterol (Figs. 2 and S5) arises from the longer timescales to which $R_{1\rho }$ is sensitive to. The internal motions of the glycerol backbone are not affected by cholesterol in the most realistic MD simulations used (CHARMM36 and Slipids, see Fig. 4) in line with the invariance of the glycerol backbone $R_1$ values (Fig. 1). Therefore, the slower glycerol backbone dynamics induced by cholesterol most likely arise from a slowdown of the rotational diffusion of the phospholipid main body (the acyl chains and glycerol backbone), as previously suggested by Roberts et al. (11,37), rather than restrictions in internal dynamics.

The independence of headgroup dynamics from the dynamics of the glycerol backbone and acyl chains must result from a set of fast internal rotations around phospholipid bonds with specific orientations that decouple these motions. The MD simulations in best agreement with the NMR experiments show a high flexibility for dihedral angles in the headgroup region with a wide range of accessible conformations (Figs. 4, S6 as well as (57)). From the set of dihedral distribution functions, one clearly observes the highly flexible nature of the ($g_3$)O–P, P–O(α) and $g_2$-$g_3$ dihedral angles with all angles over a complete dihedral rotation having nonzero probability in contrast to the remaining torsions. This applies for all three force fields, although Berger has a less even distribution than CHARMM36 and Slipids. For CHARMM36, the correlation times for the rotations around the ($g_3$)O–P and P–O(α) bonds are lower than 0.5 ns (Fig. 4) and very close to the effective correlation times measured for the C–H bonds from the α and β carbons. These ${\tau }_c^{\varphi }$ values slightly decrease with the addition of cholesterol, i.e., cholesterol induces a slight speedup of the torsion dynamics for these particular dihedrals, most likely because fewer steric hindrances are present due to an increased average distance between headgroups.

The highly flexible dihedral rotations around the phosphate P–O bonds, as well as the $g_2$–$g_3$ torsion, are much faster than the transverse and longitudinal rotational diffusion of the molecular frame whose correlation times have been estimated to be 10 to 20 ns and 100 ns, respectively (10). The most probable θ angle of the ($g_3$)O–P and $g_2$–$g_3$ bonds is less than 10° (Figs. S8 and S11), which indicates that these bonds are very close to being parallel with the bilayer normal axis. A flexible, fast-rotating dihedral aligned with the membrane normal leads to decoupling of headgroup from the longitudinal rotational diffusion of the molecular frame. Decoupling from the transverse rotational diffusion is mostly enabled by the fast motion over the P–O(α).

The torsions around the phosphate group and $g_2$–$g_3$ dihedrals act analogously to a “frictionless spherical-joint”, which decouples the choline headgroup structure and dynamics from the glycerol backbone. This is in line with the previous analysis (10) in which a partial decoupling of the headgroup from the main phospholipid body due to the rotation of a phosphate dihedral was suggested based on a comparison of CHARMM C27r MD simulation of pure dipalmitoylphosphatidylcholine bilayers to ³¹P-NMR $R_1$ data under several magnetic fields. Here, we demonstrate that such decoupling is strong enough to prevent the propagation of the slowdown effect of cholesterol from the acyl chains and glycerol backbone to the headgroup.

We base our molecular interpretation of the decoupled motion on the CHARMM36 force field, which gives the best overall description for the headgroup and glycerol backbone structure and dynamics among the available models (Figs. 2 and 3, and (48,49)). However, not all the NMR observables calculated are within experimental errors even in CHARMM36 simulations, and we cannot exclude the possibility that improved future models correctly capturing the decoupling effect, effective correlation times, and structural order parameters may give an alternative molecular interpretation.

The headgroup decoupling is not observed in the MD simulations based on the Berger force field (although it also provides fast dynamics over the phosphate group dihedrals, as shown in Fig. 4). This is most likely due to an overestimation of the attractive interaction between cholesterol and the choline group. Such interaction has been interpreted previously as a consequence of the so-called “umbrella effect” in which a reorientation of the headgroup due to presence of cholesterol is often assumed (58). However, the combination of MD simulations and experiments presented here indicates that such interaction is artifactual and that both the orientation and dynamics of the headgroup are unaffected by the sterol presence. Therefore, the implicit assumption in the “umbrella model” of a cholesterol effect on headgroup reorientation, either through a change of the conformational ensemble or a change of dynamics, is not supported by our experimental results or the more realistic MD models.

Correlation functions of P$\to$N and other intramolecular vectors, calculated from CHARMM36 simulations, further support the idea of decoupled dynamics of headgroup and acyl chains. Cholesterol induces a significant slowdown of the reorientation of the interatomic vectors between atoms belonging to the glycerol backbone and acyl chain segments (Figs. S7–S13), while its effect on the P$\to$N vector, representing the choline dipole, is negligible (with only a slight speedup of the dynamics most likely due to the increase of the distance between phospholipids headgroups).

Conclusion

The molecular description suggested here has rather strong implications for membrane biophysics and should motivate a number of additional experiments and simulations. It implies that the dipolar surface of glycerophospholipid bilayers consists of freely rotating dipoles with timescales faster than 2 ns that do not depend on the dynamics of the acyl chains or glycerol backbone. The timescale of reorientation of the dipoles is expected to influence the interaction of the headgroups with charged molecules, e.g., proteins, that approach the lipid biomembrane. For instance, it is known that the tilt of the headgroup dipole is highly sensitive to membrane surface charge (30). Under a positive surface charge, the headgroups tilt to a more upright orientation (increase of the α and β $S_{\text{CH}}$ values) and vice-versa for a negative surface charge due to the charge-dipole electrostatic interactions. The results presented here suggest that the phosphate and the $g_2$–$g_3$ dihedrals enable an unconstrained response of headgroups to the electrostatic field and effectively uncouple the interactions occurring in the membrane surface from the hydrophobic region. Although we only investigate PC headgroups here, it is foreseeable that the decoupling applies to all other glycerophospholipids because the same “molecular bearings” are present irrespectively of the substituent headgroup (1).

All the observations presented and discussed in this article pertain only to rotational diffusion. It is known from diffusion experiments that the lateral translational diffusion of the phospholipid molecule is affected by cholesterol (1). Less is known about how cholesterol affects the translational fluctuations of the phospholipids in the direction of the bilayer normal, although molecular protrusions have been linked to bending rigidity (59); we refrain from investigating such effect here, but it is worth pointing out that all the MD simulation data produced in this work, as well as MD simulation data from many other systems, are now publicly available in a format that enables analysis and comparison of any user-defined property over a large set of trajectories (www.nmrlipids.fi).

In summary, our results suggest that for describing the dipolar interactions at the surface of membranes, the hydrophobic structure may be neglected to a good approximation and that the relevant headgroup physics lie on electrostatic interactions—which is remarkably useful considering the complex molecular arrangement in the hydrophobic region of biological membranes.

Author contributions

T.M.F. coordinated the research, and designed and supervised the experimental work. T.M.F. performed and analyzed the NMR experiments. H.A. designed and performed the MD simulations. T.M.F., A.W., and H.A. analyzed the MD simulations. T.M.F. wrote the first draft of the paper. All authors revised and rewrote the manuscript.

Editor: Siewert-Jan Marrink.