The Bilayer Collective Properties Govern the Interaction of an HIV-1 Antibody with the Viral Membrane

Abstract

Efficient engagement with the envelope glycoprotein membrane-proximal external region (MPER) results in robust blocking of viral infection by a class of broadly neutralizing antibodies (bnAbs) against human immunodeficiency virus (HIV). Developing an accommodation surface that engages with the viral lipid envelope appears to correlate with the neutralizing potency displayed by these bnAbs. The nature of the interactions established between the antibody and the lipid is nonetheless a matter of debate, with some authors arguing that anti-MPER specificity arises only under pathological conditions in autoantibodies endowed with stereospecific binding sites for phospholipids. However, bnAb-lipid interactions are often studied in systems that do not fully preserve the biophysical properties of lipid bilayers, and therefore, questions on binding specificity and the effect of collective membrane properties on the interaction are still open. Here, to evaluate the specificity of lipid interactions of an anti-MPER bnAb (4E10) in an intact membrane context, we determine quantitatively its association with lipid bilayers by means of scanning fluorescence correlation spectroscopy and all-atom molecular dynamic simulations. Our data support that 4E10 establishes electrostatic and hydrophobic interactions with the viral membrane surface and that the collective physical properties of the lipid bilayer influence 4E10 dynamics therein. We conclude that establishment of peripheral, nonspecific electrostatic interactions with the viral membrane through accommodation surfaces may assist high-affinity binding of HIV-1 MPER epitope at membrane interfaces. These findings highlight the importance of considering antibody-lipid interactions in the design of antibody-based anti-HIV strategies.

Significance

The viral membrane plays a crucial role in human immunodeficiency virus (HIV) neutralization by many anti-MPER antibodies. Understanding the mechanics of membrane-antibody interaction is key to engineering anti-MPER-based vaccines. We describe the interaction of the anti-HIV-1 broadly neutralizing antibody 4E10 with a membrane, correlating membrane-antibody association with the bilayer collective properties. To do so, we have applied for the first time to our knowledge, a methodology based on measuring antibody diffusion at the bilayer to describe the nature (hydrophobic, electrostatic, or lipid-specific) of the antibody-membrane association. Our findings are important to understand the membrane-mediated events leading to HIV-1 neutralization by anti-MPER antibodies. Besides, the methodology we have used will contribute to describing the mechanistic basis of antibody-epitope recognition in a membrane context.

Introduction

The human immunodeficiency virus (HIV) deploys several strategies to evade immune neutralization. Among them, the rapid sequence variation of the envelope glycoprotein (Env) with successive replication cycles facilitates escape from the adaptive immune response. Access to antibodies (Abs) is hindered further because of the scarce number of copies of this protein present on the virion surface (1). Anti-HIV broadly neutralizing Abs (bnAbs) are capable of neutralizing genetically diverse HIV strains, usually targeting conserved regions of the Env protein, the only viral protein found on the outside of the virus (1).

Among all known bnAbs, those with the largest breadth (such as 4E10 and 10E8) bind to a highly conserved region close to the transmembrane domain of the gp41 Env subunit, termed the membrane-proximal external region (MPER) (Fig. 1 A; (1,2)). The MPER is responsible for the disruption of the HIV membrane during fusion of the cell plasma membrane and the viral envelope (3, 4, 5).

Figure 1.

Lipid contribution to 4E10 recognition of HIV Env. (A) Presumed interacting geometry of 4E10 Fab (gray) with the Env glycoprotein (light green) MPER region (residues 671–693, dark green) and the HIV membrane is shown. This model is based on PDB: 4XBG for 4E10 and PDB: 2MG2 for MPER. Env glycoprotein trimer contour has been adapted from the cryo-electron microscopy structure with EMDB: EMD-3308. The gray dotted line indicates the approximate level of the membrane interface. Fab hydrophobic sites W100 and W100b at the HCDR3 loop (top inset) and cationic patches (in blue, bottom inset) are known to be involved in gp41 recognition. (B) A cartoon illustrating the different Ab-membrane interaction modes interrogated in this work is given. From left to right: lipid-specific interaction, represented by a prominent positive lobe and a network of stereospecific hydrogen-bonding interactions that bind the protein to a single lipid copy; nonspecific charge-driven attraction through polycationic patches at the protein surface; and hydrophobic protein-membrane association. Red indicates anionic lipid species.

Another peculiarity of anti-MPER bnAbs is their functional association with the viral membrane: epitope binding occurs at the interface of the viral lipid envelope and relies on Ab-lipid interactions (6). Mutation of lipid-interacting residues prevents Env binding and subsequent viral neutralization (6, 7, 8, 9). This idea was further supported by the recently resolved crystal structure of the anti-MPER bnAbs 4E10 and 10E8 in complex with lipids, revealing a surface on which interactions between backbone atoms of the protein and glycerol groups, phosphate moieties, and the upper section of the lipid tails of phospholipids can be observed (10,11).

This functional and structural evidence stresses the need for understanding the mechanisms governing the dynamic Ab-lipid-binding process at the membrane interface. Despite the vast available knowledge on the molecular basis governing Ab-epitope recognition, little is known on the contribution of membrane interactions to this phenomenon in some relevant instances, such as bnAbs against the HIV MPER.

In this work, we focused on the interaction of 4E10, a well-characterized pan-neutralizing anti-MPER bnAb. 4E10-membrane interaction is thought to occur through the concerted interplay of 1) an electrostatic attraction between the membrane-associated paratope area (MAPA) (Fig. 1 A, bottom view) and the lipid polar heads (7,12) and 2) a deeper hydrophobic interaction between the anti-MPER Ab heavy-chain complementary determining region 3 (HCDR3) and the lipid bilayer (Fig. 1 A; (10,13)). In that regard, 4E10 resembles the features of peripheral membrane proteins (7).

To quantitatively interrogate the nature of 4E10 interaction with the membrane, we correlated the diffusional behavior of the 4E10 Fab on controlled model-membrane systems with their biophysical properties such as surface membrane potential and lipid packing. To do so, we used scanning fluorescence correlation spectroscopy (sFCS) to accurately measure diffusion coefficients at the surface of a membrane (14,15), as well as two-photon generalized polarization (GP) imaging to quantify the membrane lipid packing (16,17). In addition, molecular dynamics (MD) simulations with a total accumulated time of 1 μs were used to obtain atomic-level insights on the interaction of 4E10 with the bilayer, complementing the experimental results.

Our quantitative findings support that 4E10 dynamics on virus-like lipid membranes are governed by collective biophysical properties such as surface membrane potential and molecular order and argue against a specific docking interaction between 4E10 and phosphatidylserine (PS) (Fig. 1 B), the most abundant anionic lipid in the viral membrane (18). The three known 4E10-virus interaction sites (the MPER specificity pocket, HCDR3, and MAPA) condition the diffusion of 4E10 Fab on virus-like membranes, suggesting that these interactions overall contribute to the formation and stability of the 4E10-HIV complex. Thus, we hypothesize that 4E10-membrane interaction sites evolved to increase affinity for its primary epitope MPER through the formation of secondary Ab-lipid interactions. These observations are key to the successful engineering of MPER-targeting immunogens, as well as to defining the structural components that must be preserved when developing Abs as immunotherapeutic agents.

Materials and Methods

Lipids and MPER

1,2-dioleoylphosphatidylcholine (DOPC), 1,2-dioleoylphosphatidylserine (DOPS), cholesterol (Chol) and egg sphingomyelin (SM, containing ∼86% N-palmitoyl SM) were purchased from Avanti Polar Lipids (Birmingham, AL). 6-dodecanoyl-2-dimethylaminonaphthalene (Laurdan) was obtained from Molecular Probes (Eugene, OR). Fluorescently labeled 1,2-dioleoylphosphatidylethanolamine (DOPE-STAR RED) was purchased from Abberior (Abberior, Göttingen, Germany). Phospholipid stock concentrations were determined by phosphate assay. The MPER peptide (KKDKWASLWNWFDITNWLWYIKLFIMIVGKKK) was synthesized in C-terminal carboxamide form by solid-phase methods using Fmoc chemistry, purified by reverse-phase high-performance liquid chromatography, and characterized by matrix-assisted laser desorption ionization-time of flight mass spectrometry (purity >95%).

4E10 Fab purification and labeling

The 4E10 Fab sequence was cloned in pColaDuet plasmid and expressed in Escherichia coli T7-shuffle strain. Recombinant expression was induced at 18°C overnight with 0.4 mM isopropyl-D-thiogalactopyranoside when the culture reached an optical density of 0.8. Cells were harvested and centrifuged at 8000 × g, after which they were resuspended in a buffer containing 50 mM HEPES (pH 7.5), 500 mM NaCl, 35 mM imidazole, DNase (Sigma-Aldrich, St. Louis, MO), and an EDTA-free protease inhibitor mixture (Roche, Spain). Cell lysis was performed using an Avestin Emulsiflex C5 homogenizer. Cell debris was removed by centrifugation, and the supernatant was loaded onto a nickel-nitrilotriacetic acid affinity column (GE Healthcare, Chicago, IL). Elution was performed with 500 mM imidazole, and the fractions containing the His-tagged proteins were pooled, concentrated, and dialyzed against 50 mM sodium phosphate (pH 8.0), 300 mM NaCl, 1 mM dithiothreitol, and 0.3 mM EDTA in the presence of purified protease tobacco etch virus. Fabs were separated from the tobacco etch virus and cleaved peptides containing the His6× tag by an additional step in a nickel-nitrilotriacetic acid column. The flow-through fraction containing the Ab was dialyzed overnight at 4°C against sodium acetate (pH 5.6) supplemented with 10% glycerol and subsequently loaded onto a MonoS ion exchange chromatography column (GE Healthcare). Elution was carried out with a gradient of potassium chloride, and the fractions containing the purified Fab were concentrated and dialyzed against a buffer containing 10 mM sodium phosphate (pH 7.5), 150 mM NaCl, and 10% glycerol. For the preparation of the Δloop mutant Fab (W100-G100a-W100b-L100c residues substituted with an S-G dipeptide), the KOD-Plus mutagenesis kit (Toyobo, Osaka, Japan) was employed, following the instructions of the manufacturer. Fab labeling was attained by first introducing titratable Cys residues at position Cys228HC of the 4E10 sequence and then by modifying those with a sulfhydryl-specific iodacetamide derivative of the Abberior STAR RED (KK114) probe (Abberior). After purification, fluorescence emission measured after sodium dodecyl sulfate–polyacrylamide gel electrophoresis and absorbance measurements confirmed almost total titration of the single free Cys residues in the Fabs.

Preparation of giant unilamellar vesicles

Giant unilamellar vesicles (GUVs) were produced by spontaneous swelling, following the procedures described in (17,19). For preparation of MPER-containing GUVs, vesicles and peptides were mixed at a 1:1000 peptide:lipid molar ratio for 15 min (25°C) before incubation with silica beads. Dried silica beads covered with lipid-peptide mixtures were collected and transferred to an 85 g/L sucrose buffer to induce spontaneous swelling of GUVs. Formed vesicles were transferred to a bovine-serum-albumin-blocked observation dish in an isosmotic 10 mM HEPES, 150 mM KCl (pH 7.4) buffer that already included 4E10 Fab to a final concentration of 3–20 nM for sFCS experiments and 200 nM for imaging experiments. GUVs and Abs were incubated for 15 min and subsequently imaged.

Atto 647-labeled Annexin A5 was purchased from Adipogen (Liestal, Switzerland). For the Annexin A5 experiments, GUVs were incubated with this protein in a 2 mM CaCl₂-HBS (HEPES-buffered saline) buffer.

Imaging lipid packing at the membrane: GP

Lipid-packing imaging and quantification through the ratiometric quantity GP (20) was performed as previously described in detail (16,17). Briefly, images were acquired on a Leica TCS SP5 II microscope (Leica Microsystems, Wetzlar, Germany). Laurdan-stained GUVs (1:100 mol fraction) were excited at 780 nm using a 63× water-immersion objective (numerical aperture = 1.2) and 512 × 512 pixel images. The fluorescence emission was simultaneously imaged at 435 ± 20 nm and at 500 ± 10 nm. GP images were computed for every pixel in the image (Eq. 1), where I_B is the intensity in the blue channel and I_R the intensity in the green channel:

$$GP=\frac{I_B-G\times I_R}{I_B+G\times I_R}.$$ (1)

The G factor accounts for the relative sensitivity of the two channels, calibrated using a 5 μM Laurdan solution in pure DMSO (Dimethyl sulfoxide) (GP_cal = 0.207 at 22°C) (21):

$$G=\frac{G{P}_{cal}+G{P}_{cal}\times G{P}_{obs}-1-G{P}_{obs}}{G{P}_{obs}+G{P}_{cal}\times G{P}_{obs}-1-G{P}_{cal}},$$ (2)

where GP_obs is the GP value determined for the standard Laurdan-DMSO solution before calibration (G = 1).

The uncertainty of the GP values in the text is given by the experiment SEM (standard error of the mean); the edges of the boxes in the GP box-and-whiskers plot are the 25^th and 75^th percentiles, and whiskers span to ±1.5 times standard deviation (SD).

Diffusion at the membrane: sFCS

sFCS was used to study the diffusional behavior and intermolecular interactions of fluorescently labeled proteins bound to GUVs (Fig. S1). Excitation and detection were performed through a water-immersion objective HCX PL APO 63×/1.20 W CORR Lbd Bl (Leica) on a confocal Leica SP5 using the avalanche photodiodes (APDs) (Excelitas Technologies SPCM-AQRH-W3; Waltham, MA) fitted on the microscope DD X1 port. Abberior STAR RED-tagged Fabs were excited at 633 nm, and the fluorescent emission was detected through a BP647-703 nm filter. The average power density at the sample plane was 50 kW/cm². Photon arrival times were recorded using an SPC830 TCSPC card (Becker & Hickl, Berlin, Germany), which also registered the pixel, line, and frame signals from the scanner to track the beam position at the sample. An HRT-82 Eight Channel Router (Becker & Hickl) for the APDs was used to spectrally tag the APD output for the SPC830. The whole system was externally clocked at 20 MHz.

The imaging mode at the microscope was set to xt (line-scanning) at 1400 Hz scanning frequency, and the SPC830 detection mode was set to FIFO image. The scanning length spanned over 16.64 μm, binned in 64 pixels of 260 nm each. GUV membranes were scanned for 90–120 s. The SPC830 binary output files were decoded using an in-house-developed MATLAB (The MathWorks, Natick, MA) routine. The xt photon trace was corrected for the vesicle drift and subsequently autocorrelated with a minimal lag time fixed by the inverse of the scanning frequency. The resulting curve G(τ) was finally fitted to a two-dimensional diffusion model:

$$G\left(\tau \right)=\frac{1}{N}{\left(1+\frac{\tau }{{\tau }_D}\right)}^{-1/2}{\left(1+\frac{\tau }{S^2{\tau }_D}\right)}^{-1/2},$$ (3)

where τ is the lag time, τ_D the diffusion time, N the average number of molecules in the focal volume, and S the ratio of the axial (z_o)/radial (ω_o) 1/e² dimensions of the confocal volume. ω_o and z_o were obtained after calibrating the volume using dyes with known diffusion (Alexa 633, D = 305 μm²/s in HEPES buffer at 21°C). Finally, the diffusion coefficient is D = ω_o²/(4τ_D).

Determination of the average diffusion coefficient and experiment SD for every sample and membrane composition comprised at least three independent experiments and more than nine measurements (typically 19 measurements per experiment). The shortest time lag for the calculation of the autocorrelation function in sFCS experiments is fixed by the excitation beam scanning frequency, hindering the observation of autocorrelation characteristics at shorter time lags. This, in turn, may affect the uncertainty of the fitted parameters. To ensure the reliability of the fitted parameters, we analyzed the uncertainty due to the fitting procedure (fitting uncertainty) and that given by the overall experimental variability (experiment SD). Support plane analysis of the diffusion coefficient ${\chi }_R^2$ surface (Fig. S2; (22)) typically showed a steep minimum at the value of D that minimized ${\chi }_R^2$ and a fitting uncertainty at the 68% confidence level that was always smaller than the experiment SD. This indicated that variability in sample preparation and related experimental procedures, rather than fitting, was the main source of uncertainty, thus guaranteeing the reliability of the retrieved parameters.

The uncertainty of the diffusion coefficients given in the text is the experiment SEM unless otherwise stated; whiskers in the box-and-whiskers plots span to ±1.5 times SD. Autocorrelation curves shown in Figs. 2, 3, and 4 were normalized after fitting to allow easier visual comparison between samples with different D. All in-house-developed software tools are available upon request.

Figure 2.

Electrostatic and hydrophobic interactions play a critical role in efficient lipid-mediated antigen binding. (A) Association of the 4E10 Fab to the virus-like negatively charged membrane at physiological ionic strength is shown. A neutral surface or deletion of the hydrophobic residues at 4E10 Fab HCDR3 thwarted the association. (B) 4E10 antigen (MPER peptide) markedly enhances Fab binding to the vesicle-antigen ensemble. In the presence of the MPER, electrostatic and hydrophobic interactions between the lipid and the 4E10 Fab promote binding to the vesicle-antigen system. (C) The bnAb PGT145 does not associate to the viral-like membranes, confirming the nonrandom association of 4E10. (D) Representative correlation curves of 4E10 Fab in solution (gray) and the membrane of virus-like GUVs (colors as in E) are shown. Curves were normalized to their value at G(0) after fitting for easier comparison. (E) The diffusion of the 4E10 Fab on the virus-like vesicle is slowed down by its interaction with the negatively charged lipids at the membrane, the MPER peptide loaded on the membrane, and the 4E10 Fab HCDR3 loop. (A and B) Scale bars, 4 μm. (E) The edges of the boxes are the 25^th and 75^th percentiles. Whiskers indicate ±1.5 × SD.

Figure 3.

Cholesterol content modulates 4E10 Fab binding to the membrane. (A) 4E10 Fab diffusion at the membrane is slowed down when Chol content in the membrane increases. (B) Representative correlation curves of 4E10 Fab diffusing on the vesicles in (A) are shown. (C) Cholesterol content in the membrane has a direct effect in lipid packing, quantified by GP measurements. Top: representative false-color GP images of the GUVs used to quantify lipid packing. (A–C) The edges of the boxes are the 25^th and 75^th percentiles.Whiskers indicate ±1.5 × SD. (C) Scale bars represent 5 μm in all cases.

Figure 4.

4E10 Fab binding to the membrane is driven by negative charges. (A) 4E10 Fab diffusion at the membrane is significantly slowed down with increasing PS content. Inset: membrane surface potential with increasing PS content at physiological ionic strength (150 mM). (B) Representative correlation traces of the Fab diffusing in (A) are shown. (C). Diffusion of a DOPE-Star Red molecule (1:1000 mol fraction) in the DOPS:Chol:DOPC mixture does not change with increasing PS content. (D) The diffusion of annexin A5 on a charged bilayer in the presence of Ca²⁺ is not dependent on the membrane surface potential and is consistent with individual lipid diffusion shown in (C). (A, C, and D) The edges of the boxes are the 25^th and 75^th percentiles. Whiskers indicate ±1.5 × SD.

Estimation of molecular mean area and membrane surface potential

The surface potential (ψ_s) of a charged membrane immersed in a monovalent electrolyte bath where only one species of ions in solution neutralizing the charged surface is considered is (23)

$${\psi }_s=-\frac{4K_{B}T}{e}\arctan \left(-\frac{b}{{\lambda }_D}+\sqrt{{\left(\frac{b}{{\lambda }_D}\right)}^2+1}\right),$$ (4)

where K_B is the Boltzmann constant, T the temperature, e the electron charge, b = e(2πλ_B|σ|)⁻¹, with λ_B the Bjerrum length, σ the surface charge density, and λ_D the Debye-Hückel screening length (λ_D = 0.304n₀^−0.5 in nm, where n₀ is the electrolyte strength in molar units).

The surface potential of the DOPS:Chol:DOPC membranes was calculated using a 60 ± 1 Å² mean molecular area of 18-carbon DOPS in 0.150 M NaCL solution at 300 K, determined by MD simulations.

Molecular modeling

The structure of the 4E10 Fab (heavy (H) and light (L) chains only) in the apo form were obtained from the crystal structure deposited with Protein Data Bank (PDB): 5CIP. A short missing loop in chain H (residues 128–135) was reconstructed using the homologous loop present in a holo form of the protein (PDB: 2FX7). Residues protonation states were obtained using H++ at neutral pH (24, 25, 26). A DOPS:Chol:DOPC (0.45:0.35.0.20) lipid bilayer was generated using CHARMM-GUI (27, 28, 29, 30, 31). The protein was placed on the surface of the membrane oriented so that residues S74, S30, and G27 were close to the membrane surface (orthogonal distance between the bilayer headgroups and the geometric centers of S74, S30, and G27 at ∼9 Å). The system was centered in a simulation box of dimensions 134 × 135 × 157 Å³ and solvated, and Na⁺/Cl⁻ ions were added to neutralize the system and to achieve a bulk ionic concentration of 150 mM. The CHARMM36m force field (32) was used to represent the protein, CHARMM36 (33) was used for the lipids, the TIP3P model (34) for water, and the standard CHARMM and NBFIX parameters were used for ions (35,36). The system was energy-minimized in 5000 steps of geometry optimization, and equilibration at 300 K and 1.01325 bar was achieved in 1 ns of MD in the NpT ensemble. Position restraints on the protein backbone atoms were applied during equilibration to maintain the initial structure. Semi-isotropic pressure coupling at 1.01325 bar was accomplished using the Nose-Hoover Langevin piston (37,38) while temperature was maintained at 300 K by means of the Langevin thermostat (39). The particle mesh Ewald algorithm (40) with grid spacing of 1 Å was used for long-range electrostatic interactions, and van der Waals forces were smoothly switched off between 10 and 12 Å. A Verlet neighbor list with pairlist distance of 16 Å was used. Equations of motions were integrated using the multi-time-step algorithm Verlet-I/r-RESPA (41). The RATTLE algorithm (42) was used to constrain bonds involving hydrogen atoms allowing for 2 fs time steps. Long-range electrostatic forces were updated every time step. Production MD simulations were carried out in two replicas, MD replica I (750 ns) and MD replica II (250 ns). The MD production run was carried out for 250 ns. All calculations were performed using the NAMD 2.10 software package (43).

Results

Electrostatic and hydrophobic association to the viral membrane are essential for efficient epitope binding

With the aim of understanding the role of Ab-membrane interactions in the 4E10-mediated neutralization, we studied the 4E10 Fab behavior on model GUVs that were prepared from a five-lipid mixture derived from the HIV lipidome (virus-like mixture: PC:Chol:SM:PE:PS 14:46:17:16:7) (16,18). To address the factors governing the interaction of 4E10 with membrane lipids, we included the unsaturated lipids DOPC, DOPE, and DOPS (DO denotes dioleoyl), which facilitate association of the Ab to the bare lipid bilayer (7,8). We fluorescently tagged the 4E10 Fab with Abberior STAR RED and studied its binding and diffusive behavior on naked or MPER-peptide-containing virus-like GUVs. Pure lipid GUVs constitute a controlled environment to study the membrane-dependent steps that contribute to epitope recognition, whereas MPER-loaded GUVs constitute an adequate model for Ab-epitope recognition in a lipid bilayer environment.

We first evaluated 4E10 binding to the virus-like GUVs using intensity-based confocal microscopy (Fig. 2, A and B). 4E10 Fab readily partitioned to virus-like vesicles that did not contain the MPER (naked) (Fig. 2 A). The reason can be ascribed to the electrostatic attraction between the Fab and the membrane and the interaction between the hydrophobic residues in the HCDR3 loop and the bilayer. As a negative control for this interaction, we used another STAR RED-tagged bnAb, PGT145 (44). PGT145 is known to bind to the Env apex (farther apart from the viral membrane) and did not exhibit any partition to the membrane (Fig. 2 C), confirming the non-random association of 4E10 to the membrane. To identify the contribution of each type of interaction, we independently prevented first, the hydrophobic contact of the Fab with the bilayer through the ablation of the hydrophobic residues at the HCDR3 loop tip (4E10 Δloop), and second, the electrostatic attraction between 4E10 MAPA and the membrane, removing the negatively charged PS lipids from the bilayer (0% PS, PC:Chol:SM:PE 17:46:17:20). In the absence of its epitope, 4E10 Fab binding to the naked GUVs was not observed in either case (Fig. 2 A). Both types of interactions are therefore necessary for 4E10 Fab binding to the naked GUVs. When presented to MPER-loaded virus-like GUVs (Fig. 2 B), vesicles appeared brighter because of the Ab affinity for the peptide. In contrast to naked vesicles, we found that deletion of the hydrophobic residues or removal of the charged lipids from the membrane only partially disrupted binding to the vesicles. Even the simultaneous arrest of both electrostatic and hydrophobic interactions did not totally prevent Ab binding to the MPER (Fig. 2 B).

A quantitative description of the molecular contact of 4E10 Fab with the bilayer was obtained through the study of the diffusional regime of the Fab on the membrane using sFCS (Fig. S1). Point-FCS (pFCS) and sFCS are single-molecule sensitivity techniques that allow the diffusion coefficient of a molecule to be determined in solution (pFCS) or as the membrane-bound molecule diffuses on the surface of a vesicle (sFCS) (14,15). The diffusion coefficient of the 4E10 Fab in the aqueous buffer was 95 ± 1 μm² s⁻¹ (measured by pFCS), as expected given its molecular weight (47.8 kDa) (Fig. 2 D). However, sFCS measurements revealed that the diffusion coefficient of the 4E10 Fab on the surface of naked virus-like vesicles decreased to 8.3 ± 0.5 μm² s⁻¹ (Fig. 2, D and E), caused by the slower dynamics of membrane lipid phases.

The molecular model for epitope recognition postulates that the interaction of the Fab 4E10 heavy chain with the membrane occurs through a surface patch with net positive charge (Fig. 1 A; see also MD simulations below). Consistent with this possibility, the theoretical pH(I) of the 4E10 heavy chain is 9.34. To probe the effect of the electrostatic attraction, rather than removing the charged components—which would prevent Ab association—we increased the membrane net negative surface charge by increasing the anionic PS lipid concentration up to 37 mol % (Chol:SM:PS 46:17:37). This resulted in a significant decrease of the diffusion coefficient to 5.0 ± 0.4 μm² s⁻¹, further confirming the effect of the electrostatic contribution to the 4E10-membrane interaction (Fig. 2, D and E).

We next addressed the effect of the membrane on 4E10 Fab binding to the MPER. The Fab-epitope interaction could be restricted to the epitope binding site, that is, a binary interaction, or be mediated or modulated by the membrane through a ternary interaction. Compared to the naked GUVs, 4E10 Fab diffusion further decreased to 2.6 ± 0.2 μm² s⁻¹ on MPER-loaded virus-like GUVs (Fig. 2, D and E). Interestingly, the Δloop variant, which lacks the HCDR3 hydrophobic loop that inserts in the membrane, showed a diffusion three times faster (7.2 ± 0.8 μm² s⁻¹) than the wild-type Fab. This result demonstrated that 4E10 Fab dynamics are mainly governed by Ab-lipid interactions because the wild-type and Δloop versions of 4E10 Fab diffused differently upon epitope binding. Finally, the electrostatic interaction in the presence of the MPER could not be reliably probed because of the patchy distribution of the Fab on MPER-containing GUVs made out of neutral lipids (Fig. 2 B).

Membrane lipid packing governs 4E10 Fab diffusion

Our results in the previous section support that 4E10-envelope interactions rely on the electrostatic attraction between the Ab residues and the negatively charged membrane together with the anchoring of the Ab HCDR3 loop hydrophobic groups in the bilayer. These results are consistent with previous reports (7,45). We therefore set to evaluate systematically the contribution of collective biophysical properties of the bilayer, namely lipid packing and surface charge density, to 4E10 Fab interaction with membranes.

To this aim, we prepared PS:Chol:PC vesicles with increasing Chol content while keeping PS content constant (30%, estimated surface potential −68 mV). Adding Chol to a fluid bilayer rigidifies the membrane, resulting in increased lipid packing. To quantify membrane packing, we used two-photon GP imaging of Laurdan-stained GUVs. Laurdan is a hydrophobic polarity probe; when incorporated into a bilayer, its fluorescence emission is a function of the hydration and viscosity of the membrane, reflecting its phase state (20). Laurdan fluorescence emission in loosely packed membranes, which accommodate more interstitial water molecules, will be shifted to the red edge of the spectrum, whereas in tightly packed and poorly hydrated membranes, it will be shifted to the blue. A wavelength-ratiometric parameter, termed GP (20), quantifies Laurdan’s spectral shift, providing an indirect measurement of membrane order through packing (16).

The effect of increasing Chol on 4E10 Fab diffusion on the membrane was to decrease the diffusion coefficient of the Fab from 8.3 ± 0.7 μm² s⁻¹ for 20% Chol mixtures to about half this value (4.1 ± 0.3 μm² s⁻¹) when Chol content was increased to 50% (Fig. 3, A and B). This reduction correlated with the increase in lipid packing, as quantified by GP, from 0.128 ± 0.006 to 0.392 ± 0.003, with Chol increasing in an approximate linearly fashion (Fig. 3 C).

Increasing membrane lipid packing also resulted in a concomitant slowdown of membrane lipid mobility (Fig. S3). To rule out that the reduction of the Fab diffusion coefficient with increasing lipid packing was due to individual lipid deceleration, we compared the Fab diffusion and the diffusion of a fluorescently tagged lipid (DOPE-STAR RED at 1:1000 mol fraction) in otherwise same-composition GUVs (Fig. S3). At low Chol content (20%), the diffusion coefficient of the Fab is higher than that of individual lipids (8.3 ± 0.7 and 5.9 ± 0.6 μm² s⁻¹, respectively), whereas at high Chol content (50%), the Fab and the lipids diffuse at similar rate (4.1 ± 0.3 and 4.2 ± 0.5 μm² s⁻¹, respectively). The reduction rate of the Fab diffusion coefficient with increasing Chol content is therefore much more pronounced than that of the lipids, supporting that Fab slowdown with increasing Chol is not due to individual lipid deceleration but the result of a change in a collective biophysical property such as lipid packing.

4E10 Fab diffusion regime supports a nonspecific electrostatic interaction with anionic membranes

sFCS experiments of 4E10 Fab on virus-like vesicles had shown that the increase of the overall bilayer charge through an increase of PS content triggered a significant reduction of the 4E10 Fab diffusion coefficient (Fig. 2 E). The negatively charged lipid PS is exposed on the viral membrane external leaflet (8,46) and accounts for nearly 10% of the total HIV lipid content (18,47). PS is, therefore, the major contributor to the envelope net charge. For this reason, we set to get further insight into the Fab electrostatic association to the HIV lipid envelope. We studied the 4E10 Fab diffusion on a series of model PS:Chol:PC GUVs with increasing content of PS while keeping Chol content constant.

Increasing PS had the effect of increasing the vesicle overall negative charge and, in turn, the membrane surface potential (ψ_s). ψ_s was calculated to be −40 mV for 15% PS and increased, in the negative sense, to −86 mV for 45% PS (Fig. 4 A, inset). 4E10 Fab diffusion significantly decreased with increasing PS content, from 12.2 ± 0.8 μm² s⁻¹ for the 15% PS mixture to 5.5 ± 0.4 μm² s⁻¹ for the 45% PS mixture (Fig. 4, A and B). In contrast, the diffusion of individual lipids within the bilayer, which is essentially dependent on membrane lipid packing, was not affected by ψ_s, averaging to 4.8 ± 0.2 μm² s⁻¹ at 35% Chol irrespective of PS content (Fig. 4 C).

4E10 Fab electrostatic association with a PS-containing membrane may either involve specific molecular recognition of the PS polar head or nonspecific charge-driven interactions established through the Fab basic-residue containing MAPA surface (Fig. 1 B). To address this question, we first tested sFCS’s ability to recognize the different nature of the lipid-protein electrostatic association, comparing the diffusion of 4E10 Fab to that of annexin A5 at PS:Chol:PC vesicles of the same composition. Annexin A5 is a peripheral protein known to specifically dock onto single PS headgroups with high affinity (K_d ≈ 20 nM (48)). Previous studies on supported bilayers using fluorescence recovery after photobleaching and single-particle tracking have shown that proteins tightly bound to a single lipid in the bilayer diffuse at the same rate as the free lipid because the main component of the frictional drag is the viscosity of the bilayer through the lipid they are attached to (49,50). Consequently, the diffusion coefficient of proteins bound to individual lipids does not depend on the relative concentration of the specific lipid ligands. On the other hand, proteins associated to the bilayer through less specific electrostatic interactions are expected to diffuse faster than individual lipids in the bilayer and to display dependence on the fraction of anionic lipids (51).

Annexin A5 diffusion was strikingly different to 4E10 Fab diffusional behavior. Whereas the Fab diffusion slowed down with increasing PS molar fraction and, as a consequence, ψ_s (Fig. 4, A and B), the annexin diffusion coefficient was not dependent on ψ_s, averaging to 3.2 ± 0.2 μm² s⁻¹ (Fig. 4 D). This behavior is analogous to that of individual lipids (Fig. 4 C), reflecting that annexin A5 tightly binds to PS molecules diffusing along with them. Taking into account that annexin A5 can self-assemble in two-dimensional trimeric arrays (52), the diffusion coefficient we have obtained is likely to be an average of individual and trimeric annexin repeats (associating to one and three PS copies, respectively) and, as a result, slightly lower than that observed for individual lipids.

This experiment comparing 4E10 and annexin A5 diffusional behaviors reflected the potential of sFCS measurements to discern the nature of protein-lipid interactions. Thus, the results demonstrated that the membrane-bound 4E10 Fab does not dock onto individual PS molecules diffusing along with them, but rather, slides over the bilayer surface, presumably subject to intermittent adsorption events due to nonspecific charge-driven attractive forces.

To further confirm this result, we studied the diffusional behavior of 4E10 at anionic membranes with fixed PS content while increasing the buffer ionic strength (Fig. S4; (53)). Increasing the ionic strength increases the electrolyte concentration, decreasing, in turn, vesicles’ ψ_s (Fig. S4, inset). Because membrane lipid mobility is largely unaffected by buffer ionic strength, a protein associated through a stereospecific interaction to the headgroup of a lipid would diffuse at the same rate, irrespective of the electrolyte concentration. We observed, however, that 4E10 diffused faster with increasing ionic strength (Fig. S4), reflecting that the electrostatic attraction between the anionic membrane and the Fab is more effectively screened by the growing number of electrolyte ions surrounding the Fab. This supported that the electrostatic contribution to Fab association with the bilayer is governed by a nonspecific charge-driven interaction with the anionic membrane.

HCDR3 tryptophan residues and a group of basic residues orient 4E10 paratope on the membrane surface

To acquire atomistic insight into the orientation and interactions of 4E10 with the membrane, we ran MD simulations of the 4E10 Fab in the presence of a DOPS:Chol:DOPC (45:35:20) bilayer. We had observed that 4E10 Fab diffusion on this membrane was the slowest of the lipid mixtures experimentally studied, suggesting strong Ab-membrane interactions. During 750 ns, the overall 4E10 Fab structure remained stable, as well as that of individual H and L chains and the constant (C) and variable (V) domains (Fig. 5 A). This was further confirmed by clustering analysis of the 4E10 Fab coordinates, which showed only two rather similar 4E10 Fab conformations (Fig. S5 A), indicating that the structural integrity of the 4E10 Fab was preserved. During the simulation, the HCDR3 loop only sampled a stable conformation (Fig. 5 A; Fig. S5 B). The HCDR3 loop maintained a specific three-dimensional arrangement, characterized by having the W100 and W100b residues exposed (Fig. S5 B) and prone to membrane insertion. Consistent with such disposition, the 4E10 Fab readily approached the membrane at the beginning of the simulation, inserting W100 and W100b into the bilayer and remaining anchored to the membrane for the rest of the simulation (Fig. 5 B). The conservation of the HCDR3 conformation and its insertion into the lipid bilayer membrane correlated with the stabilization of a specific orientation of the H-chain-V-domain, which enables the MAPA patch to associate with the membrane surface (Fig. 5 C). Overall, during the simulation the MPER binding pocket retained an orientation that would allow Fab docking onto Env on the viral membrane surface (compare Fig. 1 A to Fig. 5 D).

Figure 5.

Atomistic insight into 4E10-membrane interactions (MD replica I: 750 ns). (A) Cα-root mean-square deviation of the complete 4E10 Fab, H and L chains, constant domain, variable domain, and the HCDR3 loop is shown, measured during the MD simulation. Flexible loops were not considered in the root mean-square deviation measurement except for HCDR3. (B) z Axis position of P and N lipid atoms in the membrane leaflet closer to 4E10 Fab and of the geometric center of protein residues S74, S30, G27, W100, and W100b is shown. The bottom color bar matches the simulation time points in (C). (C) Sampling of the φ/ψ space of the 4E10 Fab during the MD simulation is shown. Dot color indicates the time as in (B): red, shorter simulation times; blue, longer times. φ corresponds to the angle between the protein first principal axis (v1) and the simulation box z axis (normal to the membrane), and ψ corresponds to the angle between the second principal axis (v2) of the protein and the x axis of the simulation box (see bottom right inset in C). Isodensity contour lines are shown to display density values: mean + 1 SD (light gray), mean + 2 SD (gray), and mean + 4 SD (dark gray). The black cross indicates the orientation of the starting configuration, illustrated in the top-left inset. (D) Representative structure taken from the high-density region of the φ/ψ space is shown. A cholesterol molecule is shown interacting with residues W100 and W100b. (E) Top: 4E10 Fab residues with high probability of contact (setting 5 Å as the cutoff distance between any lipid heavy atom and any protein heavy atom) with the membrane are indicated (basic: blue; acidic: red; polar: green; hydrophobic: white). Bottom: same as in the top panel but discriminating according to the specific lipid in contact with the protein (see Fig. S6 for the corresponding probability of contact distributions). The 4E10 Fab surface is colored by chain (H: gray, L: blue).

Further analysis of the interactions between the V-domain residues and the membrane lipids identified different types of contacts with varying probability (Fig. S6). A summary is shown in Fig. 5 E, in which V-domain residues with a high probability of interaction with PC, PS, and Chol have been highlighted. In line with previous crystallographic studies (10), these results confirmed high-probability contacts between phospholipids PC/PS and Fab residues W100/W100b and S25/G26/G27/S28/F29/S30 from HCDR3 and HCDR1, respectively. In addition, they also allowed the identification of a group of basic residues that have a high probability of interaction with PS molecules (K100E, R54, R77, R73, K23, R45, and R61) and cooperate with W100 and W100b to stabilize the Fab orientation (Figs. 5 E and S6). Notably, Chol interacts with the HCDR3 apex residues W100/G100a/W100b (Fig. 5, D and E).

To provide further robustness to the MD simulation results, an independent 250 ns simulation (MD replica II) was carried out, starting from the same system configuration as that used for the original 750 ns simulation (MD replica I). The results obtained are in full agreement with those from MD replica I (Figs. S7 and S8).

Discussion

Identification and characterization of Ab-pathogen interactions is critical for vaccine design and therapy. A relatively new concept has been introduced in this area upon the discovery of anti-MPER Abs against HIV: the establishment of Ab-membrane interactions for effective engagement with antigens. The case of the anti-HIV 4E10 bnAb is paradigmatic in this sense because Ab-lipid interactions are essential for its neutralizing activity (6,7,9). For this reason, determining the degree of specificity of Ab-lipid interactions in a membrane context has relevant and crucial functional consequences for vaccine design.

Among anti-MPER bnAbs, 4E10 shows the strongest membrane partitioning capacity. MPER recognition by 4E10 is thought to be favored by the interplay of the hydrophobic residues W100 and W100b at the Ab HCDR3 with the membrane and its epitope (7,9,12). Besides, 4E10 has been shown to partition to PS-containing vesicles with the same lipid composition as used in this work (K_x = 0.6 × 10⁻⁶ M) (7) and to weakly bind phospholipids immobilized on ELISA plates, including PS, PE, phosphatidic acid, phosphatidylglycerol, and phosphatidylinositol (54). This observation prompted subsequent crystallographic analyses of Fab 4E10 in complex with water-soluble, short-acyl-chain derivatives of those phospholipids (10). Upon crystallization, Fab residues from HCDR1 and HCDR3 were indeed found establishing interactions with glycerophosphate moieties, supporting the accommodation of phospholipid interfacial groups onto the MAPA during the process of antigen binding. However, whether 4E10 specifically binds all or some of these lipids in a membrane context and how the process may contribute to Ab-MPER binding are unclear.

One suitable methodology to estimate the contribution of specific lipids to protein-protein interactions that evolve in the membrane interface environment is the study of proteins’ diffusional regime. Diffusion of soluble proteins in solution is dependent on their hydrodynamic radius, solvent temperature, and viscosity. In contrast, lateral translation of molecules fully embedded in a membrane is weakly dependent on their radius but strongly dependent on the bilayer thickness (55,56), in turn primarily determined by lipid packing. Peripheral proteins represent an intermediate scenario because they cannot freely diffuse, but they are not deeply embedded in the membrane, either. They are, instead, associated to the bilayer in a shallow position through attractive electrostatic forces and/or hydrophobic anchoring (Fig. 1 B). Each of these interactions contributes to the protein frictional drag at the membrane in an additive fashion (50,57), defining the protein diffusion rate and therefore allowing the investigation of the nature of protein-lipid contacts.

Our results, based on sFCS diffusion measurements, demonstrate that long-range electrostatic and hydrophobic interactions cooperate to promote 4E10 association to the membrane (Fig. 2, A and B). We found that both interactions slow down 4E10 Fab dynamics at the membrane (Fig. 2, D and E). Diffusion experiments of the Fab on model membranes with increasing cholesterol (Fig. 3) or PS content (Fig. 4) show an inverse relationship between the concentration of these lipids and 4E10 Fab diffusivity, suggesting that hydrophobic and electrostatic interactions sustain insertion of the Ab in the membrane. Because the intercalation of hydrophobic residues in the bilayer contributes to total friction (57), 4E10 Fab slowdown at the membrane with increasing lipid packing (Fig. 3 A) due to increased Chol content is consistent with 4E10 Fab hydrophobic anchoring at the membrane.

The nonspecific nature of the electrostatic interaction of 4E10 with the membrane was supported by the different diffusion patterns on PS:Chol:PC bilayers of the 4E10 Fab and those of a protein known to bind PS in a lipid-specific manner (annexin A5) (Fig. 4). Dependence of membrane diffusion rates on ψ_s (through a change of PS mole fraction and ionic strength) ruled out the existence of stereospecific binding of this anionic phospholipid to 4E10. Increased ψ_s may either cause an increase in the frequency of protein adsorption events or a stronger, deeper protein binding in the membrane (or both). Although both effects would result in the slowdown of a protein, a deeper 4E10 Fab insertion in the bilayer (through HCDR3 W100 and W100b residues) is not expected to be altered by surface charge density because tryptophan residues accommodate in the vicinity of the phospholipid glycerol groups (58). For this reason, although we cannot fully rule out that surface charge affected 4E10 hydrophobic association, we expect this to be of small or unappreciable effect.

This conclusion, drawn from diffusion sFCS experiments, was further supported by MD simulations (Fig. 5). In line with previous crystallographic work (10), MD simulations of the Fab at the membrane suggest high-probability contacts between PC/PS and residues from HCDR1 and HCDR3 (Figs. 5 E and S6) that would be responsible for 4E10 hydrophobic association to the membrane. Moreover, PS high-probability contacts with the Fab involved several basic residues (Fig. 5 E; Table S1), suggesting a nonspecific charge-driven attraction between the negatively charged PS and the basic residues on the protein surface. This conclusion is also consistent with previous vesicle-binding experiments performed by our group (7) and others (45).

Our MD simulations also showed that the aromatic HCDR3 residues W100 and W100b, which were associated with lipid tails in Fab-phospholipid cocrystals (10), can favorably stack with Chol. Interestingly, acyl-chain interactions seem to be accompanied by a conformational change that reorients the HCDR3 loop (10). This raises the possibility of loop conformations stabilized in the membrane through interaction with different lipid species and/or the MPER helix.

Altogether, we envision a multiple-lipid interaction model in which all phospholipid molecules in the viral membrane interact intermittently but favorably with the lipid-binding subsites on 4E10 MAPA, anionic PS molecules establish additional electrostatic interactions with positively charged residues, and sterol rings and phospholipid acyl chains alternate occupancy of the membrane integral groove in between W100 and W100b rings.

Conclusion

We have investigated 4E10-membrane interaction, studying the diffusion regime of 4E10 Fab at the bilayer and correlating it to the collective membrane biophysical properties lipid packing and membrane surface potential. MD simulations have provided further insight into the 4E10-membrane interaction. Our experiments provide unprecedented and rigorous information about the dynamics of the Ab in a virus-like membrane environment.

Our data support that, because 4E10 does not spontaneously bind to the virus membrane in the absence of the viral protein Env (8), nonspecific electrostatic Ab-lipid interactions increase 4E10 affinity for Env by providing extra contact sites on the viral surface, enlarging the interacting area, and/or facilitating the insertion of the Ab in the membrane after MPER engagement, thus stabilizing the 4E10-Env complex. The orientation of the 4E10 Fab at the membrane as observed during MD simulations reinforces this idea. We anticipate that these results will help in the design of the next generation of MPER-based immunogens.

Author Contributions

P.C. performed experiments and contributed to the experimental design, discussion, and manuscript writing. L.D. performed the MD simulations. I.R.O.-A. performed experiments. E.R. produced Abs and designed the labeling strategy. A.G.V. performed experiments. G.d.l.H.-M. set sFCS methodology up. C.D. supervised the MD simulations and contributed to the discussion. J.L.N. formulated the experiment and contributed to the discussion and manuscript writing. J.R.-I. supervised the experimental design and contributed to the discussion and manuscript writing.

Editor: Alemayehu Gorfe.