Cholesterol Interaction with the Trimeric HIV Fusion Protein gp41 in Lipid Bilayers Investigated by Solid-State NMR Spectroscopy and Molecular Dynamics Simulations

Abstract

HIV-1 entry into cells is mediated by the fusion protein gp41. Cholesterol plays an important role in this virus-cell fusion, but molecular structural information about cholesterol-gp41 interaction is so far absent. Here, we present experimental and computational data about cholesterol complexation with gp41 in lipid bilayers. We focus on the C-terminal region of the protein, which comprises a membrane-proximal external region (MPER) and the transmembrane domain (TMD). We measured peptide-cholesterol contacts in virus-mimetic lipid bilayers using solid-state NMR spectroscopy, and augmented these experimental data with all-atom molecular dynamics simulations. 2D ¹⁹F NMR spectra show correlation peaks between MPER residues and the cholesterol isooctyl tail, indicating that cholesterol is in molecular contact with the MPER-TMD trimer. ¹⁹F-¹³C distance measurements between the peptide and ¹³C-labeled cholesterol show that C17 on the D ring and C9 at the intersection of B and C rings are 7.0 Å from the F673 sidechain 4-¹⁹F. At high peptide concentrations in the membrane, the ¹⁹F-¹³C distance data indicate three cholesterol molecules bound near F673 in each trimer. Mutation of a cholesterol-recognition amino acid consensus (CRAC) motif did not change these distances, indicating that cholesterol binding does not require this sequence motif. Molecular dynamics simulations further identify two hotspots for cholesterol interactions. Taken together, these experimental data and simulations indicate that the helix-turn-helix conformation of the MPER-TMD is responsible for sequestering cholesterol. We propose that this gp41-cholesterol interaction mediates virus-cell fusion by recruiting gp41 to the boundary of the liquid-disordered and liquid-ordered phases to incur membrane curvature.

Introduction

Cholesterol is a central component of biological membranes. Cholesterol-rich regions of lipid bilayers provide sites of membrane protein interactions with each other and with intracellular proteins to carry out a myriad of biological functions [1, 2]. Many membrane proteins also specifically complex with cholesterol for function. For example, cholesterol binds G-protein coupled receptors (GPCRs) to affect their oligomerization [3], dynamics [4], and signaling [5]. Cholesterol binds PDZ domains of scaffold proteins to regulate protein networking and cell signaling [6, 7]. Cholesterol also interacts with the amyloid precursor protein and the Aβ peptide in lipid membranes [8–10], and these interactions have been suggested to be involved in the pathogenesis of Alzheimer’s disease [11].

The envelope glycoprotein (Env) of the human immunodeficiency virus 1 (HIV-1), the causative agent of the acquired immunodeficiency syndrome, mediates virus entry into cells. Env consists of a trimer of gp120 and gp41 heterodimers [12, 13]. Binding of gp120 to cell-surface receptors triggers large-scale conformational changes of gp41, which lead to fusion of the virus envelope with the plasma membrane or the endosomal membrane of the cell [14]. The gp41 conformational changes include the exposure and insertion of an N-terminal fusion peptide (FP) into the target cell membrane, while the protein remains anchored in the viral envelope by the C-terminal transmembrane domain (TMD). This extended conformation is metastable, and proceeds to bend into a hairpin, in so doing pulling the target cell membrane to the virus envelope. The trimer of hairpins, or six-helix bundle, marks the fully fused state [15–17]. The hydrophobic FP and TMD are structurally unknown in the post-fusion six-helix bundle; but they are believed to facilitate the formation of the six-helix bundle by disrupting the integrity of the target cell membrane and virus envelope [18, 19]. In addition to the FP and TMD, a membrane-proximal external region (MPER) N-terminal to the TMD also exerts a significant influence on membrane fusion and virus entry [20, 21]. The MPER harbors the epitopes of several broadly neutralizing antibodies [22–25] and binds to the lipid bilayer surface, approximately perpendicular to the TM helix [19].

Four lines of biochemical and biophysical evidence strongly suggest that cholesterol is important for gp41-mediated virus-cell fusion [26]. First, cholesterol depletion in the membrane severely impairs gp41-mediated virus entry [27, 28] as well as virus budding [29]. Second, peptide studies show that the FP, MPER, and TMD all have affinity for cholesterol-rich regions of the lipid membrane. Fluorescence microscopy and fusion assays show that the FP preferentially localizes to the edge of cholesterol-rich lipid domains, between liquid-ordered (L_o) and liquid-disordered (L_d) phases [30]. Similarly, MPER preferentially aggregates to cholesterol and sphingomyelin rich region of the membrane [31]. Third, a cholesterol-recognition amino acid consensus (CRAC) motif is present in the segment ⁶⁷⁹LWYIK⁶⁸³ between the MPER and the TMD [32]. The CRAC motif is characterized by a triad of a methyl-rich aliphatic Leu or Val residue, a central aromatic Tyr residue, and a basic Arg or Lys residue [33]. This motif runs from the N-terminus to the C-terminus of the protein sequence. Fluorescence spectroscopy, differential scanning calorimetry, and X-ray scattering data showed that this short peptide preferentially sorts to the cholesterol-rich region of the membrane [34–36]. Mutations of this CRAC motif retain fusion activity only if the mutant retains interactions with cholesterol [37]. Fourth, partition and leakage assays showed that a peptide containing the C-terminus of the MPER and the N-terminus of the TMD preferentially bound and destabilized cholesterol-rich membranes [38, 39].

Despite these biochemical data, direct molecular structural information about cholesterol interaction with gp41 has not been reported. In particular, the C-terminal segment of the protein, MPER-TMD, is anchored in the HIV virus envelope, which is well known to contain a high concentration of cholesterol [40]. Until recently, the MPER structure has largely come from solution NMR studies of the MPER peptide in cholesterol-free DPC micelles [24, 41]. Recently, the structure of the TMD and the C-terminal end of the MPER was determined in cholesterol-free lipid bicelles using solution NMR [42, 43]. However, these studies were inconclusive about whether the TMD is trimeric or monomeric in bicelles. We recently investigated the three-dimensional structural topology of the full-length MPER-TMD peptide (residues 665–704) in lipid bilayers using solid-state NMR [19]. By measuring both intermolecular and intramolecular distances, we found that MPER-TMD is trimeric in phospholipid bilayers, and the MPER lies on the membrane surface, bent from the TMD. This structural topology exists in a virus-mimetic membrane that contains POPC, POPE, POPS, sphingomyelin, and cholesterol.

Here, we investigate the intermolecular interactions between the MPER-TMD and cholesterol in virus-mimetic lipid membranes using solid-state NMR spectroscopy. Using fluorinated MPER-TMD and ¹³C-labeled cholesterol, we measured semi-quantitative distances between specific MPER residues and cholesterol atoms. We incorporated fluorine at two native MPER residues, F673 and W680. A mutant yeast strain was used to express 1-¹³C labeled cholesterol (1-¹³C CHOL) [44–46]. We find that cholesterol is in molecular contact with the MPER-TMD, but this complexation does not require the CRAC motif, because a mutant peptide without the motif exhibits the same peptide–CHOL distances. We also conducted molecular dynamics simulations to provide site-specific information about the hotspots in the peptide that interact with cholesterol, and to probe the stoichiometry of cholesterol binding over a wide range of protein concentrations.

Materials and Methods

Synthesis and purification of isotopically labeled gp41 MPER-TMD

The gp41(661–704) peptide used in this study corresponds to the MPER and TMD regions of HIV-1 clade D gp41. The amino acid sequence is LELDKWASL WNWFNITNWL WYIRLFIMIV GGLVGLRIVF AVLSI (Fig. 1a). This sequence differs slightly from the clade B gp41 sequence used in our recent study [19], by extending the N-terminus by four residues, and by having R683 instead of K683. The N-terminal extension is intended to allow future studies of antibody binding to MPER, while the use of R683 allows direct comparison of this study with previous solution NMR studies of clade D gp41 [42, 43]. Because of the common cationic nature of Arg and Lys, we do not expect any structural conclusion to be affected by this residue change. The 44-residue peptide was synthesized using a custom-designed flow synthesizer [47, 48] and Fmoc solid-phase peptide synthesis protocols. To investigate whether the CRAC motif is necessary for cholesterol binding, we also synthesized an AFI mutant peptide containing L679A, Y681F, R683I mutations. Both wild-type and mutant peptides incorporated 4-¹⁹F-F673 and 5-¹⁹F-W680 for distance measurements.

Figure 1.

(a) Amino acid sequence of gp41 MPER-TMD. Fluorinated F673 and W680 are shown in blue, while the CRAC residues L679, Y681, and R683 are underlined. The AFI mutant contains L679A, Y681F, R683I mutations. (b) Low-temperature 1D ¹⁹F NMR spectra of wild-type MPER-TMD in the VM+ membrane. Comparison of the 7 kHz and 10 kHz MAS spectra distinguishes the centerbands and sidebands. (c) ¹³C-labeled sites in 1-¹³C CHOL.[44] (d) 2D ¹³C-¹³C correlation spectra of 1-¹³C CHOL in gp41-containing VM+ membranes. The spectrum was measured at 278 K using a ¹³C spin diffusion mixing time of 300 ms.

The gp41(661–704) peptides were synthesized on the 0.05 mmol scale using H-Rink amide ChemMatrix resin (0.1 g at 0.5 mmol/g loading size). The resin was swelled in the reaction vessel for 5 min in ~5 mL of N,N-dimethylformamide (DMF) at 70°C. A 10-fold excess of unlabeled amino acids and 4-fold excess of fluorinated amino acids were singly- and triply-coupled using coupling times of 50 and 70 s, respectively. After the final coupling step, the peptide was deprotected and cleaved from the resin by addition of TFA/phenol/H₂O/TIPS solution (88 : 5 : 5 : 2 by volume) for 2.5 h. The resin was filtered off, and the crude peptide was precipitated and triturated three times using cold diethyl ether, then dried under vacuum overnight at room temperature. The dried crude peptide was dissolved in TFE (2,2,2-trifluoroethanol) and purified by preparative reverse-phase HPLC using a Vydac C18 column. A 10 min isocratic gradient (5% of channel A and 95% channel B) was followed by a linear gradient of 5–100% for channel A (channel A is 75 : 25 v/v acetonitrile : isopropyl alcohol and channel B is water) over 90 min at a flow rate of 10 mL/min. MALDI-MS analysis found the peptide mass to be 5317.9 Da for wild-type (WT) gp41(661–704) and 5215.3 Da for the AFI mutant, in good agreement with the calculated masses. The total yield of the synthesis and purification was ~7%.

Membrane sample preparation

1-¹³C CHOL was expressed from Saccharomyces cerevisiae strain RH6829 using 2-¹³C labeled acetate as the ¹³C source [44, 49, 50]. The sterol esters were saponified and the crude product was purified by HPLC, giving a final yield of ~30 mg per liter of culture. Heptafluorinated cholesterol (F7-CHOL) was purchased from Sigma. This cholesterol has the same interfacial area, molecular orientation, and phospholipid interaction as hydrogenated cholesterol based on pressure-area isotherm data [51].

The WT and mutant gp41 peptides were reconstituted into virus-mimetic phospholipid membranes (VM+), which consist of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE), sphingomyelin (SM), and cholesterol (CHOL) [52, 53]. Three membrane samples were prepared. Sample 1 contains POPC : POPE : SM : F7-CHOL at a molar ratio of 16 : 16 : 16 : 21, while the peptide : total phospholipid and SM : CHOL molar ratio (P : L : C) was 1 : 16 : 7. Sample 2 has the same peptide and lipid concentrations, but contains 1-¹³C CHOL to permit measurements of ¹³C-¹⁹F distances. Sample 3 contains POPC : POPE : SM : 1-¹³C-CHOL at a molar ratio of 32 : 32 : 32 : 21, while the P : L : C ratio was 1 : 32 : 7. Comparison of sample 2 and sample 3 allows us to probe whether higher lipid concentrations reduce the fraction of cholesterol bound to the peptide. The molar concentration of cholesterol is 30 mol% in samples 1 and 2 and 18 mol% in sample 3.

All lipids and cholesterol were dissolved in 600 μl chloroform and 4 μl methanol. The peptide was dissolved in 300 μl TFE and mixed with the lipid solution. The solvents were removed by nitrogen gas, then the samples were dried under vacuum overnight. The dried powder was resuspended in pH 7.5 HEPES buffer (10 mM HEPES/NaOH, 1 mM EDTA, 0.1 mM NaN₃) and dialyzed against pH 7.5 HEPES buffer for 3 days with six buffer changes to remove salt and residual TFA and TFE. The proteoliposomes were spun at 40,000 rpm using a Beckman SW60T rotor at 4°C for 6–10 h to obtain membrane pellets. The pellets were incubated in a desiccator until it reached a hydration level of ~40 wt% water and then spun into magic-angle spinning (MAS) rotors for solid-state NMR experiments.

Solid-state NMR spectroscopy

Solid-state NMR spectra were measured on a Bruker NMR spectrometer at 9.4 Tesla (400 MHz ¹H Larmor frequency) using a 4 mm ¹H/¹⁹F/¹³C MAS probe. ¹³C chemical shifts were externally referenced to the adamantane CH₂ signal at 38.48 ppm on the TMS scale. ¹⁹F chemical shifts were referenced to the ¹⁹F signal of Teflon at −122 ppm on the CFCl₃ scale. 1D ¹³C and ¹⁹F CP-MAS spectra were measured from 293 K to 238 K under 7 kHz to 10 kHz MAS. Typical radiofrequency (rf) field strengths were 62.5 kHz for ¹⁹F pulses, 50 kHz for ¹³C pulses, and 71.4 – 83.3 kHz for ¹H heteronuclear decoupling.

2D ¹³C-¹³C correlation spectra were measured to confirm ¹³C-labeled sites in 1-¹³C CHOL. 2D ¹⁹F-¹⁹F correlation spectra of fluorinated gp41 and F7-CHOL were measured to investigate peptide-peptide and peptide-CHOL interactions. We used the CORD dipolar recoupling sequence [54] for ¹³C and ¹⁹F spin diffusion. The mixing times ranged from 100 ms to 750 ms. These 2D spectra were measured at 238 K under 10 kHz MAS.

¹⁹F-detected and ¹³C-detected ¹³C-¹⁹F REDOR experiments were conducted to measure gp41-CHOL spatial proximities. F673, W680-fluorinated peptide was combined with 1-¹³C CHOL in VM+ membranes. For each mixing time, a pair of spectra were measured, with and without inversion pulses on the unobserved channel [55]. The control experiment (S₀) does not have inversion pulses on the unobserved channel and gives the full echo intensity, which controls for T₂ relaxation of the observed nuclear spin. The dephased experiment (S) has inversion pulses on the unobserved channel to induce dipolar dephasing. The ¹⁹F-detected ¹³C-dephased REDOR experiments used 90°180°90° composite pulses on the unobserved ¹³C channel, whereas the ¹³C-detected REDOR experiments used 90°180°90° composite pulses on the unobserved ¹⁹F channel. These composite pulses compensate for rf field inhomogeneity [50, 56]. Both the ¹⁹F-detected and ¹³C-detected REDOR experiments were conducted at 238 K under 10 kHz MAS. This low temperature was necessary to immobilize the translational motion of the peptide and CHOL so that peptide-cholesterol distances can be measured. REDOR mixing times ranged from 0.6 ms to 8 ms for the ¹⁹F-detected REDOR experiments and 5 ms to 8 ms for the ¹³C-detected REDOR experiments.

The REDOR dephasing values, S/S₀, were obtained from the ratio of the peak heights in the dephased and control spectra. This intensity ratio depends on dipolar coupling through the equation [55, 57]

$$S\left(t_{mix}\right)/S_0\left(t_{mix}\right)=\underset{\beta }{\int }\underset{\alpha }{\int }d\alpha \text{cos}\Delta \Phi \left(\alpha ,\beta ,t_{mix}\right)\text{sin}\beta d\beta /\underset{\beta }{\int }\underset{\alpha }{\int }d\alpha \text{sin}\beta d\beta$$ (1)

where the phase ΔΦ of the recoupled dipolar interaction is proportional to the distance-dependent dipolar coupling constant D_IS and the mixing time t_mix, $\Delta \Phi \left(\alpha ,\beta ,t_{mix}\right)\propto D_{IS}\cdot t_{mix}$. The dipolar coupling constant is inversely proportional to the internuclear distance r_IS to the third power and proportional to the gyromagnetic ratios γ of the two spins, $D_{IS}\propto {\gamma }_I{\gamma }_S/r_{IS}^3$. In Equation 1, β and α are the polar angle and azimuthal angle of the internuclear vector with respect to a rotor-fixed coordinate system whose z axis is along the rotor axis.

Uncertainties in the S/S₀ values were propagated from the experimental uncertainties ${\epsilon }_{S_0}$ and ε_S of each REDOR spectrum according to the equation ${\epsilon }_{S/S_0}=\left(S/S_0\right)\sqrt{{\left({\epsilon }_{S_0}\right)}^2+{\left({\epsilon }_S\right)}^2}$. The values of ${\epsilon }_{S_0}$ and ε_S are the inverse of the signal-to-noise ratios of the peaks.

The ¹³C-detected ¹³C-¹⁹F dephasing curves were fit using the SIMPSON program [58] assuming a two-spin system. An intensity scaling factor of 0.85 was applied to the simulated curves to compensate for ¹³C and ¹⁹F pulse imperfections. Best-fit distances were obtained by minimizing the root-mean-square deviations between the measured and simulated curves. The total measuring time of the ¹⁹F-detected ¹³C dephased REDOR spectra was 35 days to include multiple samples and multiple mixing times, while the total measuring time for the ¹³C-tectected REDOR experiments was 38 days.

Molecular dynamics simulations

We used the solid-state NMR structural model of gp41-TMD (PDB: 6DLN) [19] as the input structure for MD simulations. The peptide was embedded in a lipid bilayer using CHARMM-GUI [59, 60]. The bilayer consists of POPC, POPE, SM and CHOL at a molar ratio of 64 : 64 : 64 : 84, to which we inserted four gp41 trimers. Thus, the P : L : C molar ratio of the system is 1 : 16 : 7, which is identical to the composition of the high peptide-concentration samples in the NMR experiments.

The peptide-lipid system was equilibrated using the highly mobile membrane mimetic (HMMM) model [61] to allow efficient sampling of lipid distribution around gp41. The system size was 10.5 × 10.5 × 7.3 nm³. 36 Cl⁻ ions were added to reach charge neutrality, then an additional 6 Na⁺ and Cl⁻ ions were added to reach a salt concentration of 10 mM as used in the SSNMR samples. Protein, lipids, cholesterol and ions were described using the CHARMM36 force field [62], and water was described with TIP3P [63]. The protein structure model was energy minimized using the conjugate gradient method to remove any bad contacts between solvent and solute atoms. This was followed by a 200 ns constant-pressure constant-temperature simulation in which all heavy atoms of the protein were restrained. Three randomly equilibrated structures were then converted to full lipid models using CHARMM-GUI. These three full lipid systems were simulated for 500 ns in independent restraint-free simulations at constant pressure and constant temperature. The hydrogen atoms were constrained using the LINCS algorithm [64] to allow a 2 fs time step for integration. The temperature and pressure of the system in both the HMMM and full lipid models were controlled using the Nosè-Hoover thermostat [65] and Parrinello-Rahman barostat [66], respectively. All simulations were performed using the GROMACS-2018.3 package [67].

We also simulated a lower peptide concentration, where a single gp41 trimer was inserted into the lipid bilayer. This composition corresponds to a P : L : C molar ratio of 1 : 64 : 28. To ensure consistent sampling of cholesterol distributions in the two sets of simulations, the initial structure of the lower peptide concentration simulation was taken from a fully equilibrated snapshot of the higher-peptide concentration condition, but with three out of four gp41 trimers removed. Three independent simulations of at least 500 ns each were carried out. Results for a set of low peptide concentration simulations prepared independently of the higher-peptide concentration simulations are discussed in the Supporting Information.

Results

MPER-TMD contacts with cholesterol in lipid bilayers

To investigate cholesterol interaction with gp41, we reconstituted fluorinated MPER-TMD and 1-¹³C CHOL in the virus-mimetic (VM+) membrane. The cholesterol concentrations used in our samples were 18 mol% to 30 mol%, which are similar to the cholesterol concentrations in cell membranes [68] but lower than those in virus lipid envelopes [40]. Two peptide: lipid molar ratios, 1 : 16 and 1 : 32, were used in our study. These high peptide concentrations are required to provide sufficient sensitivity for the NMR spectra. Each HIV virion contains only about ten Env trimers; however, the spatial distribution of these trimers is not random, with local clustering observed in cryoelectron tomography data [69, 70]. Moreover, whether gp41 trimers cluster to mediate virus-cell fusion is not yet known. The equivalent fusion protein in the influenza virus, hemagglutinin, has been shown to cluster to at least three trimers to cause fusion [14, 71]. Similar clustering of multiple trimers has been suggested for the parainfluenza virus fusion protein F based on analytical ultracentrifugation data [72]. Therefore, the high peptide concentrations used in the current NMR experiments may be relevant for the local gp41 densities in the virus during fusion.

For the two MPER residues, the ¹⁹F isotropic chemical shift of 4F-F673 at −115 ppm is well resolved from the chemical shift of 5F-W680 at −125 ppm (Fig. 1b). The ¹⁹F NMR spectra measured under 7 kHz MAS gave multiple spinning sidebands, which fit to chemical shift anisotropies (CSAs) of δ = 57 ppm for 4F-F673 and δ = 46 ppm for 5F-W680. These CSA values are near the rigid limit for amino acid 4F-Phe (δ = 59 ppm) and 5F-Trp (δ = 51 ppm), indicating that the MPER is immobilized in the membrane at 243 K. The ¹³C spectrum of 1-¹³C CHOL (Fig. 1c) at 278 K shows narrow linewidths of 0.5 ppm, and all ¹³C chemical shifts are readily assigned using the 2D ¹³C-¹³C correlation spectrum (Fig. 1d) [44].

¹⁹F is an excellent probe of long-range distances up to ~2 nm due to its large gyromagnetic ratio [73, 74]. If the fluorinated F673 and W680 lie within ~2 nm of each other, then ¹⁹F–¹⁹F correlation peaks should be detectable in the 2D ¹⁹F-¹⁹F correlation spectra. Fig. 2a shows the 500 ms 2D ¹⁹F spin diffusion spectrum of MPER-TMD. A correlation peak between F673 and W680 is observed, indicating that the two fluorine atoms are within ~2 nm of each other. The cross-peak intensity is asymmetric, with the W680 5-¹⁹F peak in the ω₁ dimension showing higher intensity than the F673 4-¹⁹F peak in the ω₁ dimension (Fig. 2c). We tentatively attribute this asymmetry to longer T₂ relaxation time of 5-¹⁹F Trp (~2 ms) than 4-¹⁹F Phe (~1.3 ms). The presence of a W680-F673 cross peak is consistent with the helical conformation of MPER. The solid-state NMR structural model of gp41(665–704) in lipid bilayers [19] gave an average intrahelical distance of 2.0 nm between the two fluorines (Fig. 2d), whereas the interhelical distances range from 1.9 to 2.6 nm. The DPC-micelle bound MPER has an intramolecular F673-W680 distance of 1.6 nm [24] (Fig. 2e). Since the intermolecular distances between the two residues are on average longer than the intramolecular distance, we attribute the cross peak mainly to intramolecular contact between the two sidechains.

Figure 2.

2D ¹⁹F-¹⁹F correlation spectra of membrane-bound gp41 with 500 ms mixing. (a) 2D spectrum of fluorinated gp41 in membranes containing 1-¹³C CHOL. A W680-F673 cross peak is observed. Spinning sidebands are denoted by SSB. (b) 2D ¹⁹F-¹⁹F correlation spectrum of gp41 in membranes containing F7-CHOL. Cross peaks between W680 and CHOL CF₃ and between F673 and CHOL are observed. (c) 1D cross sections from the W680 and F673 rows in the 2D spectra of (a) and (b), showing protein-protein and protein-CHOL cross peaks. (d) Solid-state NMR structural model of trimeric MPER-TMD in lipid bilayers.[19] Intrahelical (left) and interhelical (right) distances between 4F-F673 and 5F-W680 are indicated. (e) Solution NMR structural model of MPER in DPC micelles, [24] showing a 1.6 nm distance between F673 Hζ and W680 Hζ3.

We next measured the 2D ¹⁹F-¹⁹F correlation spectrum of MPER-TMD in the VM+ membrane containing F7-CHOL, to investigate whether the fluorinated peptide are in molecular contact with cholesterol. With 500 ms mixing, the 2D spectrum showed clear correlation peaks between the cholesterol CF₃ at −77 ppm and the 4F-F673 and 5F-W680 peaks (Fig. 2b, c). Therefore, the cholesterol tail is within nanometer contact with both MPER residues. Again, W680 has a higher cross peak with cholesterol than F673. These peptide-cholesterol correlation peaks set the stage for a more quantitative study of the distances and stoichiometry of cholesterol binding to gp41 using REDOR.

Using fluorinated MPER-TMD and 1-¹³C CHOL, we measured distance-dependent ¹⁹F-¹³C dipolar couplings. The REDOR experiments were conducted with both ¹⁹F detection of the peptide and ¹³C detection of cholesterol. Each 1-¹³C CHOL contains 15 ¹³C-labeled sites whereas each peptide contains only two fluorine atoms (Fig. 3a). We first measured ¹⁹F-detected REDOR spectra, because the high sensitivity of ¹⁹F NMR and the simultaneous dephasing of each ¹⁹F by multiple ¹³C spins facilitate the observation of dipolar dephasing. We then switched to ¹³C-detected and ¹⁹F-dephased REDOR experiments to obtain more site-specific information about which CHOL carbons are close to the fluorinated residues. ¹³C-¹⁹F dipolar couplings are sensitive to distances up to ~1.0 nm [75, 76].

Figure 3.

¹⁹F-detected and ¹³C-detected ¹³C-¹⁹F REDOR spectra to determine cholesterol-gp41 proximity in the VM+ membrane. (a) Schematic of fluorinated MPER-TMD trimer in VM+ membranes. 1-¹³C CHOL with ¹³C-labeled sites is illustrated. Each fluorine is coupled to multiple CHOL carbons. (b) Representative ¹⁹F-detected REDOR spectra, measured at 238 K under 10 kHz MAS. At 5 ms mixing, significant intensity differences are observed between the control (S₀) and dephased (S) spectra, indicating that F673 and W680 are close to 1-¹³C CHOL. Two samples with protein/lipid molar ratios of 1 : 16 and 1 : 32 gave similar results. The S/S₀ values are indicated for the two ¹⁹F peaks. (c) Representative ¹³C-detected REDOR spectra, measured at 238 K under 10 kHz MAS. At 8 ms mixing, clear intensity differences between the S₀ and S spectra are observed, indicating that some of the cholesterol molecules are in molecular contact with the peptide. The dephasing values S/S₀ for C17 and C9 are indicated. Asterisks indicate lipid natural abundance signals, some of which partially overlap with the cholesterol peaks.

Fig. 3b, c show representative ¹⁹F-detected and ¹³C-detected REDOR spectra. The spectra were measured at low temperature (238 K) in order to immobilize the peptide and cholesterol. The ¹⁹F-detected spectra show significant intensity differences between the control (S₀) and dephased (S) spectra (Fig. 3b). With 5 ms mixing, the S/S₀ intensity ratios are 0.75 for 4F-F673 and 0.79 for 5F-W680, indicating that both fluorinated sidechains are within a nanometer of CHOL carbons. Doubling the lipid concentration relative to the peptide did not change the REDOR dephasing, indicating that cholesterol has sufficient affinity to remain bound to the peptide even with two-fold dilution of the peptide. At P : L : C molar ratios of 1 : 16 : 7 and 1 : 32 : 7, the cholesterol concentrations are 30 mol% and 18 mol%, respectively. The former mimics the liquid-ordered (L_o) region of biological membranes whereas the latter mimics the boundary between the L_o and liquid-disordered (L_d) phases. Thus, the ¹⁹F-detected REDOR spectra indicate that gp41 interacts with cholesterol similarly in both micro-environments.

The ¹³C-detected REDOR spectra (Fig. 3c) gave more site-specific information about ¹³C-¹⁹F distances, because each cholesterol ¹³C signal is dephased by only two ¹⁹F spins. Compared to the ¹⁹F-detected spectra, the intensity difference (ΔS) between the control and dephased spectra is much smaller, corresponding to S/S₀ values of 0.92 and higher. The high values are not surprising because cholesterol is in 7-fold excess to peptide monomers in the membrane, thus only a fraction of all cholesterol molecules is expected to be sufficiently close to the peptide to experience dipolar dephasing by the ¹⁹F spins. Despite this small fraction, because the ¹³C control spectrum has extremely high sensitivity, the difference intensity is precise. We focus on the S/S₀ values of two cholesterol carbons, C17 and C9, because these two peaks are well resolved from lipid natural abundance ¹³C signals (Fig. S1) even under the line broadened conditions at low temperature. The S/S₀ values are 0.95 for C17 and 0.96 for C9 at 8 ms REDOR mixing. Several other cholesterol ¹³C signals such as C3 and C18 show higher S/S₀ values; however, their chemical shifts are very close to the chemical shifts of lipid peaks such as glycerol G2 and acyl chain ω, thus the S/S₀ values are not exclusively due to cholesterol.

To investigate whether the CRAC motif is required for gp41-cholesterol interaction, we measured ¹⁹F-detected ¹³C-dephased REDOR spectra of the AFI mutant. The mutant peptide exhibits the same ¹⁹F-¹³C REDOR dephasing as the WT peptide (Fig. 4), thus indicating that the CRAC motif is not required for cholesterol complexation with gp41.

Figure 4.

Cholesterol binding stoichiometry to trimeric MPER-TMD from ¹³C-¹⁹F REDOR analysis. (a) ¹³C-detected REDOR dephasing of C17 and C9. Results of the WT and mutant peptides are compared, and the C17 and C9 dephasing results are compared. (b) ¹⁹F-detected REDOR dephasing of 4F-F673. Results of the WT and mutant peptide at P:L = 1:16 and the WT peptide at P:L = 1:32 are compared. (c) ¹⁹F-detected REDOR dephasing of 5F-W680. Top row shows best-fit simulations using a model of three CHOL molecules bound per trimer. At a P : C ratio of 1 : 7, this number corresponds to 14.3% of all CHOL bound per trimer. Middle row shows best-fit simulations using a model of two CHOL molecules bound per trimer. This stoichiometry corresponds to an intensity scaling factor of 9.5% (= 2/21). Bottom row shows best-fit simulations using a model of one CHOL bound to each trimer. This stoichiometry corresponds to an intensity scaling of 4.8% (= 1/21).

Number of cholesterol molecules associated with the MPER-TMD trimer

We next sought to determine the number of CHOL molecules bound per trimer and the distances of the bound cholesterol to the peptide. We extract this information by analyzing the ¹³C-detected and ¹⁹F-detected ¹³C-¹⁹F REDOR spectra. Each ¹³C spin of 1-¹³C CHOL is dephased by two ¹⁹F spins, whose distances are not known simultaneously. However, we can approximate this three-spin system using a two-spin model, because dipolar couplings scale with internuclear distances r according to 1/r³, thus the ¹⁹F spin closer to the carbon will dominate the REDOR dephasing. Each cholesterol molecule is unlikely to be equidistant to both F673 and W680. This assumption is borne out by MD simulations (vide infra). Once the ¹³C-detected REDOR dephasing is fit to an effective two-spin distance, we then consider the ¹⁹F-detected REDOR dephasing. Here the dipolar coupling network is more complex, because each ¹⁹F is simultaneously dephased by multiple ¹³C spins. Each 1-¹³C CHOL contains 15 ¹³C spins, many of which may be similarly close to a fluorine atom (Fig. 3a). Therefore, we consider the ¹⁹F-detected REDOR dephasing qualitatively, by requiring that the two-spin distance obtained from the ¹³C-detected REDOR data must be longer than the effective distance manifested in the ¹⁹F-detected REDOR dephasing.

Fig. 4a shows the ¹³C-¹⁹F dephasing curves of CHOL C17 and C9 atoms. Since the cholesterol concentration is 7-fold higher than peptide monomers, only a fraction of cholesterol molecules is expected to be within molecular contact with the MPER-TMD. Thus, fitting these REDOR curves requires two parameters, the number of cholesterol molecules in close proximity to the peptide and the ¹³C-¹⁹F distance. We first investigated the model of three cholesterol molecules tightly bound to each trimer. At the P : L : C molar ratio of 1 : 16 : 7, this stoichiometry corresponds to 3 out of 21, or 14.3%, of all cholesterol molecules bound. Thus, we fit the C17-¹⁹F REDOR dephasing using an intensity scaling factor of 14.3%. The best-fit distance for C17-¹⁹F dephasing is 7.0 Å. We then calculated the two-spin REDOR dephasing for this distance (Fig. 4b, c) without the intensity scaling factor, since all peptide ¹⁹F spins should be in close contact with some cholesterol molecules. The simulated dephasing curves decay more slowly than the measured ¹⁹F-detected REDOR dephasing. This is consistent with the fact that each ¹⁹F spin is dephased by multiple ¹³C spins. Therefore, the three-CHOL binding model is in good agreement with the experimental data. Mutating the CRAC residues caused minimal change to the dipolar dephasing. Similarly, diluting the peptide concentration two-fold did not significantly change the dephasing (Fig. 4b, c). Therefore, the gp41-cholesterol interaction is sufficiently specific for the peptide and cholesterol concentration ranges used here.

We next tested the two-CHOL binding model. Using an intensity scaling factor of 2/21 = 9.5%, we obtained a best-fit distance of 6.3 Å for C17-detected REDOR dephasing. Applying this distance to the ¹⁹F-detected F673 and W680 curves resulted in good apparent agreement with the measured dipolar dephasing. However, this apparent agreement is inconsistent with the fact that each ¹⁹F is coupled to multiple ¹³C spins. Thus, the two-CHOL stoichiometry is not consistent with the experimental data. Finally, we tested the single-CHOL binding model, using an intensity scaling factor of 1/21 = 4.8%. The best-fit distance from the C17-detected REDOR dephasing is 5.4 Å. This short distance resulted in a simulated ¹⁹F-¹³C REDOR curve that decays much faster than the experimental data, thus ruling out the single-CHOL binding stoichiometry.

Molecular dynamics simulations of cholesterol interactions with gp41

To obtain more site-specific information about the distance distribution, binding stoichiometry, and concentration dependence of gp41-cholesterol interactions, we performed all-atom molecular dynamics simulations. We took the solid-state NMR structural model of MPER-TMD [19] (PDB: 6DLN) as the initial structure, and constructed a molecular model of the peptide-containing virus-mimetic membrane using the experimental molar ratios of 3 : 16 : 16 : 16 : 21 for peptide : POPC : POPE : SM : cholesterol (Fig. S2). To efficiently sample the lipid and cholesterol distribution around the peptide, we used the HMMM model [61] where part of the POPC, POPE and SM lipid tails were removed to enhance their diffusion. The center of the lipid bilayer was replaced with an organic solvent 1,1-dicholorethane (DCLE). The peptide-containing HMMM system was equilibrated for 200 ns, then the truncated lipid tails were regrown to their full length. The full system was simulated for another 500 ns in three independent runs.

Fig. 5a shows a snapshot of the lipid membrane at the end of the 500 ns run for the high-concentration membrane (P : L : C = 1 : 16 : 7). Cholesterol is homogeneously distributed in the membrane, with multiple molecules in close contact with the MPER-TMD trimers. To quantify the peptide-cholesterol contacts, we calculated the number of cholesterol molecules that have any atom within 3 Å of any atom in the MPER-TMD. The value of 3 Å was chosen because it is the typical distance for both polar (e.g. hydrogen bonding) and hydrophobic interactions. With this criterion, we found approximately 8 cholesterol molecules associated with each gp41 trimer (Fig. S2b). Some of these cholesterol molecules are near the MPER while others pack close to the TMD (Fig. 5b). Thus, the simulations show a larger number of bound cholesterol than the experimental data. However, the REDOR data report cholesterol carbons near two specific MPER residues, F673 and W680, while simulations search for cholesterol molecules near both the MPER and the TMD. Thus, the REDOR data should represent a lower bound to the number of bound cholesterols to each trimer.

Figure 5:

All-atom molecular dynamics simulation of CHOL-gp41 interactions at P : L : C = 1 : 16 : 7. (a) Top view of the equilibrated lipid membrane, showing protein (red) and cholesterol (green) molecules in the upper leaflet. (b) A representative structure showing cholesterol molecules within 3 Å of a trimer. (c) A representative snapshot showing the positions of F673 Cζ and W680 Cζ3 (orange spheres) in the MPER-TMD trimer. Three bound cholesterols whose C9 atoms are within 7.0 Å from F673 Cζ are shown. These sub-nanometer contacts can be detected by ¹³C-¹⁹F REDOR dipolar coupling measurements. (d) Radial distribution function (RDF) and cumulative sums (dotted lines) of cholesterol C9 and C17 from F673 Cζ. The value of the cumulative RDF up to the first minimum (dashed lines) for C9 and C17 is averaged to give the number of bound cholesterols per gp41 monomer, which is 0.75. Thus, 2.3 cholesterols are bound near the MPER per trimer. (e) Number of cholesterol atoms in close contact with each MPER-TMD residue.

To further compare the simulated and the experimental data, we calculated the radial distribution functions (RDFs) of C17 and C9 from F673 Cζ and W680 Cζ3 (Fig. 5c), which mimic the 4-¹⁹F and 5-¹⁹F atoms of the two residues. The results show that C9 and C17 are located at more than 1.0 nm away from W680 Cζ3 (Fig. S2c), suggesting that 5F-W680 contributes less to C9 and C17 detected REDOR dephasing than 4F-F673. In comparison, the C9 and C17 distances to F673 Cζ peak at 6.1 and 6.5 Å (Fig. 5d), respectively, in good agreement with the measured distances of 7.0 Å by REDOR. Integrating the RDFs up to the first minimum yielded a value of 1.0 for C9 and 0.5 for C17. Taking the average of the two values, and multiplying it by 3 for the three gp41 monomers, we obtained a stoichiometry of 2.3 cholesterols near MPER per trimer as measured from the F673 sidechain. This result provides support to the conclusion from the analysis of the REDOR data (Fig. 4).

To identify important peptide residues for cholesterol interaction, we computed the number of cholesterol atoms that are within 3 Å of each peptide residue. Fig. 5e shows a small number of contacts for the CRAC segment ⁶⁷⁹LWYIK⁶⁸³, while the highest numbers of atomic contacts are observed for the MPER segment ⁶⁷³FNITN⁶⁷⁷ and the TMD segment ⁶⁸⁴LFIMI⁶⁸⁸. This finding is consistent with the experimental data that the triple CRAC mutations (L679A, Y681F, and K683I) did not perturb cholesterol binding (Fig. 4). Interestingly, the TMD segment ⁶⁸³RLFIMIV⁶⁸⁹, which encompasses the second hotspot of cholesterol interaction, fulfills the so-called CARC motif [33] by having R683, F685, and V689. This CARC motif was predicted by molecular docking [77] to have affinity for cholesterol because the N-terminal Arg or Lys interact with the cholesterol hydroxyl group, the central aromatic residue interact with cholesterol via CH-π interactions, while the C-terminal Leu or Val interact with the isooctyl tail.

To further probe cholesterol-gp41 interactions, we repeated the simulations using a four-fold lower peptide to lipid ratio (P : L : C = 1 : 64 : 28). This reduction of protein concentration decreased the number of close contacts with cholesterol (Fig. 6a). Within 3 Å of any peptide atoms, there are 6 cholesterols (Fig. 6b, c). By integrating the RDF up to the first minimum (Fig. 6d), we obtained a value of 0.36 for C9 and 0.67 for C17. Taking the average of the two, and multiplying it by 3 to account for the three gp41 monomers, we obtained a stoichiometry of 1.5 cholesterol near MPER per gp41 trimer. Despite the smaller number of bound cholesterol, ⁶⁷³FNITN⁶⁷⁷ and ⁶⁸⁴LFIMI⁶⁸⁸ remain the two hotspots of cholesterol interaction with gp41 (Fig. 6e). We note that for the low peptide concentration, care needs to be exercised to properly sample the cholesterol distribution to avoid artificial clustering. Simulations prepared with a different protocol led to a slightly lower stoichiometry of 1.1 cholesterol near MPER per gp41 trimer (Fig. S3).

Figure 6:

All-atom MD simulation of cholesterol-gp41 contacts at low protein concentrations (P : L : C = 1 : 64 : 28) (a) Top view of the equilibrated lipid membrane, showing only protein (red) and cholesterol (green) molecules in the upper leaflet. (b) A representative structure of cholesterol molecules within 3 Å of the MPER-TMD trimer. (c) A representative snapshot showing the location of W680 Cζ3 and F673 Cζ (orange spheres) in the MPER-TMD structure. Also shown is a bound cholesterol whose C9 atom is within 7.5 Å of F697 Cζ. (d) Radial distribution functions (solid lines) and cumulative sums (dotted lines) of cholesterol C9 and C17 from F673 Cζ. The value of the cumulative RDF up to the first minimum (dashed lines) for C9 and C17 is averaged to give the number of bound cholesterols per gp41 monomer, which is 0.52. Multiplied by a factor of 3, this gives a stoichiometry of 1.5 cholesterols near MPER for each gp41 trimer. (e) Number of cholesterol atoms in close contact with each MPER-TMD residue.

Fig. S4a shows a representative binding mode of cholesterol with only the MPER. The polar W666 and hydrophobic F673 of MPER bind to the head and tail of the cholesterol, respectively. Another representative binding mode (Fig. S4b) shows that the MPER and TMD can cooperatively bind a single cholesterol. In one case, the polar T676 of MPER and hydrophobic L684 of TMD sequester a cholesterol, and in another case, multiple hydrophobic residues of TMD bind the hydrophobic tail while the cholesterol headgroup is stabilized by the polar N677 of MPER. As expected, the cholesterols from the lower leaflet mostly bind to TMD through hydrophobic interactions. A representative snapshot (Fig. S4c) shows the tail of one cholesterol packed against V701 and I697 and the tail of another cholesterol packed with A700 and V693, while their headgroups remain in the lipid headgroup region of the membrane. For a discussion of cholesterol orientation upon interacting with different regions of the peptide, see Supporting Information.

Discussion

These solid-state NMR data and molecular dynamics simulations give molecular structural evidence about cholesterol complexation with the gp41 C-terminus in a native-like lipid bilayer. The two fluorine atoms provide nanometer distance rulers between MPER and cholesterol. Compared to radio, spin, photoaffinity, and fluorescently labeled cholesterol [78–80], ¹⁹F NMR of sparsely fluorinated peptides is much less perturbing. The ¹³C-¹⁹F distance data were measured at low temperature on frozen membranes. This low temperature is necessary to immobilize the rotational and translational diffusion of the peptide and cholesterol, which will average the dipolar couplings and prevent the measurement of intermolecular distances. The distance information obtained from these frozen membranes represents the equilibrium average of peptide-cholesterol proximity, without kinetic information about the on- and off-rates of cholesterol binding.

The peptide-cholesterol ¹⁹F-¹⁹F cross peaks (Fig. 2b) and ¹⁹F-¹³C REDOR data (Fig. 3) unambiguously show that cholesterol is in molecular contact with the MPER-TMD. 2D ¹⁹F-¹⁹F correlation spectra of model compounds [73, 74] indicate that ¹⁹F spin diffusion can detect internuclear distances of ~2 nm. Thus, the isooctyl tail of cholesterol is within ~2 nm of the MPER residues. Since the hydrophobic chain of cholesterol is embedded in the middle of the lipid bilayer, about 2 nm from the membrane surface, this result implies that cholesterol is sequestered directly under the “canopy” of the MPER helix, instead of being displaced horizontally from the MPER helix. This conclusion is supported by the ¹³C-¹⁹F REDOR data (Fig. 4a), which show that the C17-bearing D ring and the C9-bearing B/C ring junction of cholesterol are ~7.0 Å from F673 and W680 sidechains. Although the REDOR data do not directly indicate which of the two fluorines contributed more to the dipolar dephasing, MPER structural data so far show that F673 faces the membrane interior whereas W680 faces water [24]. Molecular dynamics simulations support this model, by finding a larger number of contacts between F673 and cholesterol than between W680 and cholesterol (Fig. 5e). Therefore, we tentatively attribute the ¹⁹F-¹³C REDOR dephasing predominantly to 4-¹⁹F F673. On the other hand, the ¹⁹F-detected ¹⁹F-¹³C REDOR data show only slightly higher S/S₀ values for W680 than for F673 (Fig. 3b), suggesting that the average distance of cholesterol to W680 is not much longer than the distance to F673. Whether this similar average distance results from the F673 and W680 rotameric structures or due to the cholesterol distribution relative to the MPER will require future studies. Regardless of which residue accounts more for dipolar interaction with ¹³C-labeled cholesterol, the ~7.0 Å distance constrains the cholesterol C17 and C9 to be directly under the MPER helix, near a TMD helix.

The stoichiometry of CHOL binding to the MPER-TMD trimer depends on the distance range used to define the bound cholesterol. With ¹⁹F-¹³C REDOR experiments that detect the F673 sidechain, the data indicate three bound cholesterol molecules per trimer near F673. This stoichiometry was found in both the 30% CHOL membrane (P : L : C = 1 : 16 : 7) and the 18 mol% CHOL membrane (P : L : C = 1 : 32 : 7) (Fig. 4). This cholesterol concentration range spans the biologically relevant concentrations of cholesterol in L_d and L_o phases. The fact that cholesterol dilution did not change cholesterol complexation with the MPER-TMD suggests that the affinity of the peptide for cholesterol is stronger than the affinity of lipids for cholesterol.

Molecular dynamics simulations provided additional insights into the gp41-cholesterol interaction, by considering a much lower protein concentration than accessible by NMR experiments. At a P : L : C molar ratio of 1 : 64 : 28, the hotspots for cholesterol interactions remained the same, suggesting that the affinity of the ⁶⁷³FNITN⁶⁷⁷ and ⁶⁸⁴LFIMI⁶⁸⁸ segments for cholesterol is intrinsic to these amino acid sequences. However, the average number of cholesterols near MPER decreased to 1.5 per trimer (Fig. 6d), despite the fact that the cholesterol concentration is in 28-fold excess to the protein. To rationalize this changing cholesterol stoichiometries for the different protein concentrations, we note that at the high protein concentration used here, the gp41 trimers are spatially close to each other (Fig. 5a), thus they are expected to confine cholesterols and enhance their binding interactions through configurational entropy effects [81]. A previous coarse-grained simulations found that hemagglutinin clustering in multi-component lipid bilayers led to local enhancement of saturated lipids and cholesterols [82]. While a quantitative analysis will require a separate study, the different cholesterol binding stoichiometries observed at different protein concentrations suggest that elevated cholesterol binding in vivo may be correlated with MPER-TMD trimer clustering in the membrane. The number of Env trimers in each virion is 8–10, and the diameters of the HIV virus are 110–140 nm [40, 69]. These values translate to a biological P : L : C molar ratio of about 1 : 5500 : 4500. This extremely low protein concentration would suggest a bound cholesterol number of 1 per trimer; however, it is not known whether gp41 clusters locally in the virus envelope to carry out virus-cell fusion. If gp41 forms clusters, like hemagglutinin [71], then cholesterol may bind with the higher stoichiometry determined here.

Fig. 7 shows the structural model of trimeric MPER-TMD complexed with the three nearest cholesterol molecules obtained from MD simulations. The MPER segment ⁶⁷³FNITN⁶⁷⁷ and the TMD segment ⁶⁸⁴LFIMI⁶⁸⁸ before and after the CRAC segment together sequester a cholesterol molecule. These results indicate that the requirement for cholesterol binding to the C-terminus of gp41 is not the CRAC motif, but the helix-turn-helix topology of this C-terminal domain. Within this fold, a number of atomic interactions are potentially responsible for stabilizing the peptide-cholesterol complex. These include 1) hydrophobic interactions between the methyl-rich Leu and Ile residues in the TM helix and the methyl-rich rough face of cholesterol [50]; 2) aromatic interactions between the lipid-facing polar residues of MPER [24] and the cholesterol sterol rings; and 3) hydrogen bonding between Trp residues in the MPER and the cholesterol C3 hydroxyl group. Molecular dynamics simulations suggest that the TMD binds cholesterol mostly through hydrophobic interactions, whereas the MPER interacts with cholesterols through both polar and hydrophobic motifs. Although the TMD hotspot is part of a CARC motif when combined with K683 and V689, our simulations show that K683 and V689 themselves have only a small number of contacts with cholesterol (Fig. 5e). Future mutagenesis-based structural measurements will be necessary to fully understand whether this CARC motif plays a role in cholesterol binding to MPER-TMD. We also do not know the equilibrium dissociation constant of cholesterol to gp41. Measurement of this binding affinity by NMR requires low-concentration samples, which will have prohibitively low sensitivity. However, the affinity to the MPER region is likely weak, based on the simulated radial distribution functions. As shown in Figs. 5d and 6d, the peak value of the cholesterol-F673 radial distribution functions is in the range of 3–6, which corresponds to an attractive potential of mean force in the range of 1–2 k_BT (recall that $g\left(r\right)=e^{-w\left(r\right)/k_{B}T}$, in which w(r) is the potential of mean force). The modest strength of the interaction is likely a consequence of the competition between MPER-cholesterol and lipid-cholesterol interactions.

Figure 7.

Solid-state NMR structural model (PDB: 6DLN) of trimeric MPER-TMD complexed with cholesterol molecules (yellow) in the upper leaflet of a lipid bilayer. (a) Side view. Residues with the largest numbers of contacts with cholesterol are shown in green for ⁶⁷³FNITN⁶⁷⁷ and cyan for ⁶⁸⁴LFIMI⁶⁸⁸. The CRAC residues L679, Y681 and K683 are shown in magenta. Fluorinated (orange spheres) F673 points to the membrane interior, close to a cholesterol, whereas 5F-W680 points to water, away from cholesterol. (b) Top view, showing the three nearest cholesterols bound to the MPER-TMD trimer. Two of the cholesterol molecules are associated with the middle of the MPER helix, in close proximity to F673, while the third cholesterol is associated with the N-terminal end of one MPER helix.

The MPER Trp residues are important for virus-cell fusion: their mutations severely reduce virus entry and membrane fusion [21, 23]. Our data suggest that this importance is not only due to Trp residues’ well known ability to tether membrane proteins to the membrane-water interface [83, 84], but also due to the involvement of Trp in recruiting cholesterol to the MPER helix. Indirect support for the requirement of both the MPER and TMD for complexing cholesterol is provided by differential scanning calorimetry data, which showed that full-length MPER (residues 665–683) did not cause cholesterol-rich domains, while the ⁶⁷⁹LWYIK⁶⁸³ peptide did [85]. Moreover, both the MPER and TMD are necessary for membrane fusion, while each helix alone has little ability to cause vesicle leakage at physiologically relevant cholesterol concentrations [38]. Even a truncated construct containing the C-terminal half of the MPER, the N-terminal half of the TMD, and the intervening LWYIR, is able to cause vesicle leakage at high cholesterol concentrations [39, 86]. These biochemical data thus indicate that both membrane fusion and cholesterol sequestration require joining part of the MPER and part of the TMD through the ⁶⁷⁹LWYIK⁶⁸³ turn.

The finding that gp41 complexation with cholesterol requires a helix-turn-helix fold is in excellent agreement with our recent results about the influenza M2 protein [44, 50]. Mutation of M2’s CRAC motif in a membrane-surface amphipathic helix did not change cholesterol binding. Instead, removal of the amphipathic helix abolished cholesterol contact with the TM helix. Therefore, the HIV-1 gp41 and influenza M2 studies suggest a general principle for cholesterol binding to membrane proteins, which is the combination of a membrane-spanning helix and a membrane-surface amphipathic helix. This combination is effective for sequestering cholesterol, because of the combined contributions from hydrophobic residues in the TM helix and polar residues in the membrane-surface amphipathic helix to stabilize the non-polar and polar groups of cholesterol, respectively.

How does cholesterol binding to gp41 mediate virus-cell fusion? We hypothesize that the extended membrane-surface MPER helices, together with the trimeric TMD stalk, create a strongly asymmetric wedge shape that should incur local membrane curvature [18, 87]. This local curvature and disorder prime the virus envelope to bend and merge with the opposing target cell membrane, which is bound by the N-terminal FP and fusion-peptide proximal region [88]. Second, cholesterol binding of gp41 may act as a mechanism for lateral partitioning of gp41 to the edge of the cholesterol-rich microdomain. Total internal reflection fluorescence microscopy and content mixing assays showed that the N-terminal FP of gp41 is preferentially located at the edge of L_d and cholesterol-rich L_o domains of lipid vesicles [30]. This boundary location of FP was hypothesized to be juxtaposed with the L_d-L_o boundary of the virus lipid envelope where the MPER-TMD resides. Such boundary matching between the two membranes may facilitate fusion, because the boundaries present fault lines for membrane deformation and line tension that can only be relieved by membrane fusion.