The Atomic Structure of the HIV-1 gp41 Transmembrane Domain and Its Connection to the Immunogenic Membrane-proximal External Region*♦

Background: The structure of the HIV glycoprotein transmembrane anchor is unknown. Results: NMR spectroscopy reveals two helices connected by a flexible segment. The N-terminal helix constitutes a scaffold for neutralizing antibodies. Conclusion: The HIV transmembrane sequence combines two subdomains involved in fusion and immune response modulation during infection. Significance: These data may guide the rational design of vaccines and inhibitors. The membrane-proximal external region (MPER) C-terminal segment and the transmembrane domain (TMD) of gp41 are involved in HIV-1 envelope glycoprotein-mediated fusion and modulation of immune responses during viral infection. However, the atomic structure of this functional region remains unsolved. Here, based on the high resolution NMR data obtained for peptides spanning the C-terminal segment of MPER and the TMD, we report two main findings: (i) the conformational variability of the TMD helix at a membrane-buried position; and (ii) the existence of an uninterrupted α-helix spanning MPER and the N-terminal region of the TMD. Thus, our structural data provide evidence for the bipartite organization of TMD predicted by previous molecular dynamics simulations and functional studies, but they do not support the breaking of the helix at Lys-683, as was suggested by some models to mark the initiation of the TMD anchor. Antibody binding energetics examined with isothermal titration calorimetry and humoral responses elicited in rabbits by peptide-based vaccines further support the relevance of a continuous MPER-TMD helix for immune recognition. We conclude that the transmembrane anchor of HIV-1 envelope is composed of two distinct subdomains: 1) an immunogenic helix at the N terminus also involved in promoting membrane fusion; and 2) an immunosuppressive helix at the C terminus, which might also contribute to the late stages of the fusion process. The unprecedented high resolution structural data reported here may guide future vaccine and inhibitor developments.

Computational techniques have provided evidence for TMD self-oligomerization in the membrane milieu (12,13). These studies suggest that inter-chain hydrogen bonds may stabilize membrane-embedded trimers. Molecular dynamics simulation (MDS) studies on monomers revealed potential kinks and/or metastable hinges at or around Arg-696, a residue located near the center of TMD (14,15). According to these calculations, the most stable conformation for TMD would be a complete ␣-helical conformation. This continuous helix can nevertheless bend and acquire metastable conformations in the membrane (14). In line with a potential conformational plasticity, several mutagenesis studies support an active role for the TMD in gp41-mediated fusion during viral entry (9, 16 -18). Particularly, the specific sequence of the highly conserved N-terminal region (residues 684 LFIMIVGGLVGL 695 ), the presence of basic residues flanking or within the TMD, and the spacing between these charged residues at the C terminus of the TMD appear to be critical for the fusogenicity of Env complexes and infectivity of HIV-1 (9,16,18).
In addition, the TMD and the adjacent membrane-proximal external region (MPER) C-terminal segment have been involved in modulation of innate and adaptive immune responses during HIV infection. The boundary between MPER and TMD constitutes a site of vulnerability on the Env glycoprotein, which is targeted by broadly neutralizing antibodies (bNAb-s) 4E10 and 10E8 (19 -21). The peptide epitopes are bound to these Fabs in a highly unusual helical conformation (22,23). Consistent with the existence of a defined MPER-TMD structural motif, the gp41 TMD orients the epitope recognized by the 4E10 bNAb for optimal binding and overall modulates MPER epitope exposure (13). However, most of the current models do not include the MPER C-terminal 4E10/10E8 epitopes as constituents of the TMD anchor (24 -29). Those models further posit that the TMD starts at Lys-683, where a sharp bend is predicted to occur in the helix backbone (Fig. 1B,  left panel).
In addition, the C-terminal section of the TMD comprises a distinct immunosuppressive domain (30,31). Studies motivated by the similarity between the sequence 694 GLRIVFAV 701 and a transmembrane section of the T-cell receptor ␣ subunit demonstrated the interaction of the gp41 TMD with CD3 partners in the TCR complex, resulting in the inhibition of T-cell proliferation (30). More recently, it has been reported that the gp41 TMD may also associate with the Toll-like receptor-2 TMD in the membrane, interfering with the receptor-mediated signaling and decreasing pro-inflammatory cytokine secretion (32). Collectively, these findings suggest that the sequence of the HIV-1 TMD plays critical functions beyond that of anchoring the Env complex to membranes (8 -11), although the structural requirements underlying these distinct functional roles remain to be determined.
Here, we have undertaken the structural characterization of the gp41 TMD and its connection to the upstream MPER C-terminal section by using two overlapping synthetic peptides, CpreTM and TMDp, that together cover the HIV-1 Env(671-704) sequence (HXB2c numbering) (Fig. 1B, right  panel). The CpreTM peptide represents the MPER-TMD boundary and is delimited between the hinge segment at position 671-674 (24,26,33,34) and the membrane-buried kink around position 693, as predicted by hydrophobicity (35) and MDS (14) analyses. Thus, NMR measurements of CpreTM in membrane mimetics were expected to prove the existence of the Lys-683 kink in the middle of its sequence. The sequence and range of the TMDp peptide corresponded to that recently defined as the full-length gp41 TMD by Cohen et al. (30). To our knowledge, NMR studies on gp41 TMD and its connection to MPER have not been reported to date.
The high resolution NMR data confirmed that CpreTM and TMDp adopt ␣-helical structures in membrane-mimicking environments, as well as conformational flexibility around the transmembrane sequence 690 GGLV 693 . In contrast, no evidence was found for significant changes in the orientation of the backbone of the helix at, or close to, position Lys-683, which had been proposed to mark the initiation of the TMD at the membrane interface. Supporting the relevance of a continuously helical MPER-TMD connection for immune recognition, the CpreTM peptide based on this sequence was bound with nanomolar affinity by the anti-MPER 4E10 bNAb and raised IgG in rabbits that inhibited Env-mediated cell entry. Thus, according to the high resolution data reported here, the HIV-1 Env(671-704) sequence can be described as a two-subdomain helical motif separated by a membrane-embedded flexible hinge. The N-terminal helix, longer than expected, would be involved in membrane fusion and elicitation of neutralizing antibodies, whereas the C-terminal helix would take part in modulation of T-cell immune responses and in the last stages of the fusion process.

Experimental Procedures
Materials-The peptides used in the structural and immunological studies were synthesized in C-terminal carboxamide form by solid-phase methods using Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry, purified by reverse phase HPLC, and characterized by matrix-assisted time-of-flight (MALDI-TOF) mass spectrometry (purity Ͼ95%). Peptides were routinely dissolved in dimethyl sulfoxide (DMSO, spectroscopy grade), and their concentration was determined by the bicinchoninic acid microassay (Pierce). MAb4E10 was a gift of Dietmar Katinger (Polymun Scientific, Klosterneuburg, Austria). The Experimental Procedures described in Refs. 36, 37 were followed for the production and purification of Fabs.
The NMR spectra were acquired on a Bruker Avance-600 spectrometer operating at a proton frequency of 600.13 MHz and equipped with a cryoprobe. One-dimensional spectra were acquired using 32 K data points, which were zero-filled to 64 K data points before performing the Fourier transformation. Phase-sensitive two-dimensional correlated spectroscopy (COSY), total correlated spectroscopy (TOCSY), and nuclear Overhauser enhancement spectroscopy (NOESY) spectra were recorded by standard techniques using presaturation of the water signal and the time-proportional phase incrementation mode, as reported previously (39). NOESY mixing times were 150 ms, and TOCSY spectra were recorded using 60-ms DIPSI2 with z filter spin-lock sequence. For the samples in 25% HFIP, 1 H-13 C heteronuclear single quantum coherence (HSQC) spectra were acquired at 13 C natural abundance using standard pulse sequences (39). Acquisition data matrices were defined by 2048 ϫ 512 points in t 2 and t 1 , respectively. Data were processed using the TOPSPIN program (Bruker Biospin, Karlruhe, Germany). The two-dimensional data matrix was multiplied by either a square-sine-bell or a sine-bell window function with the corresponding shift optimized for every spectrum and zerofilled to a 2048 ϫ 1024 complex matrix prior to Fourier transformation. Baseline correction was applied in both dimensions. Two-dimensional spectra were analyzed using the SPARKY software (T. D. Goddard and D. G. Kneller, SPARKY3, University of California, San Francisco).
Structure Calculation-Structures for peptide CpreTM in DPC micelles and in 25% HFIP and for TMDp in 25% HFIP were calculated from distance and dihedral angle constraints derived from NMR parameters by following the three-step protocol previously described for MPERp (34). Distance constraints were obtained from the cross-peaks present in 150-ms two-dimensional 1 H-1 H NOESY spectra, which were integrated using the standard SPARKY integration subroutine (SPARKY3), and the dihedral angle restraints for and angles were derived from 1 H ␣ and 13 C ␣ chemical shifts using the program TALOS (40). First, we applied the standard iterative procedure for automatic NOE assignment of the program CYANA 2.1 (41), which consists of seven cycles of combined automated NOE assignment and structure calculation of 100 conformers per cycle (42). The list of distance constraints resulting from the last automatic cycle was checked by inspection of the corresponding NOESY spectra, and ambiguous constraints were removed or relaxed to generate the final list used as input for a standard simulated annealing CYANA 2.1 calculation of 100 conformers. The 20 conformers with the lowest target function values were selected and subjected to 2,000 steps of energy minimization using the generalized Born continuum solvation model with a nonbonded cutoff of 10 Å as implemented in the program AMBER9 (D. A. Case, T. A. Darden, and T. E. Cheatham, III, University of California, San Francisco). The quality of these final structures was assessed using PROCHECK/NMR (43) as implemented at the Protein Structure Validation Suite server. All of the residues were either in the most favored or allowed regions of the Ramachandran map ( Table 1). The structural ensembles calculated for CpreTM and TMDp have been deposited in the PDB data bank with accession codes 2MG1 (TMDp in HFIP), 2MG2 (CpreTM in HFIP), and 2MG3 (CpreTM in DPC). The structural statistics data for these structures are provided in Table 1. The structures were visualized and examined using the programs MOLMOL (44) and Swiss-Pdb viewer (45).
Crystallization of Peptide in Complex with Recombinant Fab 4E10 -Co-crystals of Fab 4E10 in complex with the peptide 4E10ep (NWFDITNWLWYIK-KKK) were prepared by the hanging drop method. The Fab-peptide complex in the presence of detergent dihexanoyl-phosphocholine at 3 mM was concentrated to 2.5 mg/ml and mixed with a reservoir solution composed of 200 mM sodium acetate, 30% PEG 8,000, and 100 mM sodium cacodylate (pH 6.5). Suitable crystals grew to full size within a few days at 20°C, after which they were transferred to mother liquor supplemented with 20% glycerol, and subsequently frozen and stored in a vessel containing liquid N 2 .
Data Collection and Refinement-Diffraction data from a single crystal was collected on beamline BL5A of the Photon Factory (Tsukuba, Japan) under cryogenic conditions (100 K). Diffraction images were processed with the program MOSFLM and merged and scaled with the program SCALA of the CCP4 suite (46). The three-dimensional structure was determined by molecular replacement using the coordinates of 4E10 (PDB code, 2FX7) with the program PHASER (47). The initial model was further refined with the programs REFMAC5 (48) and COOT (49). Validation was carried out with PROCHECK (50). Data collection and refinement statistics are given in Table 2.
Isothermal Titration Calorimetry-Calorimetric titration experiments were performed with a VP-ITC microcalorimeter (MicroCal, Northampton, MA) at 25°C. Prior to the experiment, proteins were dialyzed overnight at 4°C against 10 mM sodium phosphate, 150 mM NaCl, and 10% glycerol at pH 7.5, and degassed just before the measurement. The buffer was supplemented with 5 mM DPC. Fab4E10 (3 M) was titrated with 40 M peptide (dissolved in the dialysis buffer). The volume of each injection was 10 l. The heat corresponding to the dilution of the peptide was subtracted. The binding isotherms were fitted to a one-site binding model using the program ORIGIN 7.0. The fitting yields the stoichiometry (n), the binding constant (K D ), and the enthalpy (⌬H 0 ) of the binding reaction (Table 3).
Membrane Binding Assays-Vesicle flotation experiments in sucrose gradients were performed following the method described by Yethon et al. (51). In brief, 100 l of a sample containing rhodamine-labeled liposomes (1.5 mM lipid concentration) was adjusted to a sucrose concentration of 1.4 M in a final volume of 300 l, and subsequently overlaid with 400-and 300-l layers of 0.8 and 0.5 M sucrose, respectively. The gradient was centrifuged at 436,000 ϫ g for 3 h in a TLA 120.2 rotor (Beckman Coulter, Brea CA). After centrifugation, four 250-l fractions were collected. Material adhered to tubes was collected into a 5th fraction by washing with 250 l of hot (100°C) 1% (w/v) SDS.
Rabbit Immunization and Response Analyses-New Zealand White rabbits were immunized at the antibody production service from the CID-CSIC (Barcelona, Spain). For immunization in Freund's Complete (FC) adjuvant, the peptide was dissolved in 0.5 ml of PBS and mixed with an equal volume of adjuvant (Sigma). Liposome-based formulations were prepared following the methods described by Dreesman et al. (52) and Maeso et al. (53) and included MPL as adjuvant (54). Peptide dissolved in DMSO was added at a final peptide-to-lipid ratio of 1:50 (mol/ mol) to a swirling solution of freeze-thaw 1-palmitoyl-2-oleoylphosphatidylcholine/cholesterol/phosphatidic acid/MPL (2.0: 1.5:0.2:0.01 molar ratio) vesicles dispersed in PBS. After incubation for 1 h, the samples were lyophilized. Rabbits were inoculated intradermally at multiple sites on day 0 with 1 ml of sample reconstituted in pure water, which contained 0.5 mg of peptide. For subsequent boosting injections, 1 ml of the reconstituted liposome formulation containing 0.3 mg of peptide was used on day 21, and 0.2 mg of liposomal peptide were injected on days 42, 63, and 84. Specific antibodies against the TMD N-terminal helix were recovered from sera through affinity purification. To that end the KKK-NWFDITNWLWYIKLFI-KKK-C peptide was immobilized onto a beaded agarose support using the Sulfolink immobilization kit for peptides (Thermo Scientific, Rockford, IL). The remaining nonspecific binding sites in columns were blocked adding L-cysteine⅐HCl at 50 mM. Every analyzed serum was loaded on the columns after diluting and filtering it to remove the particulate material. Filtered samples were let flow through the columns five times to allow the binding of the serum antibodies that recognized specifically the immobilized peptide. After washing the columns with at least 10 bed volumes of 500 mM NaCl containing buffer to remove nonspecifically bound antibodies and most serum proteins, the fraction enriched in the specific antibodies was eluted using 100 mM glycine buffer at pH 2.5. The fraction that was not recovered using acidic pH was eluted using freshly prepared 100 mM triethylamine buffer at pH 11.5. IgG recovery was routinely determined by SDS-PAGE and Western blot analyses of the column-bound fractions.
For the cell infection inhibition assays (34,37), HIV-1 pseudoviruses were produced by transfection of human kidney HEK293T cells with the full-length env clone JRCSF (kindly provided by Jamie K. Scott and Naveed Gulzar, Simon Fraser University, British Columbia, Canada) using calcium phosphate. Cells were co-transfected with vectors pWXLP-GFP and pCMV8.91, encoding a green fluorescent protein and an envdeficient HIV-1 genome, respectively (provided by Patricia Villace, CSIC, Madrid, Spain). After 24 h, the medium was replaced with Opti-MEM GlutaMAX II (Invitrogen) without serum. Three days after transfection, the pseudovirus particles were harvested, passed through 0.45-m pore sterile filters (Millex HV, Millipore NV, Brussels, Belgium), and finally concentrated by ultracentrifugation in a sucrose gradient. Neutralization was determined using TZM-bl target cells (AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, National Institutes of Health, and contributed by J. Kappes). Samples were set up in duplicate in 96-well plates and incubated for 1.5 h at 37°C with a 10 -15% tissue culture infectious dose of pseudovirus. After antibody-pseudovirus co-incu-bation, 11,000 target cells were added in the presence of 30 g/ml DEAE-dextran (Sigma). Neutralization levels after 72 h were inferred from the reduction in the number of GFPpositive cells as determined by flow cytometry using a BD FACSCalibur Flow Cytometer (BD Biosciences).

Models for TMD and Its Connection to MPER in Membranes-
The high hydrophobicity-at-interface of the MPER C-terminal region is consistent with its insertion into the external leaflet of the viral membrane (27,55,56). Current models (Fig. 1B, left panel) depict the axis of the MPER C-terminal helix inserted in parallel to the membrane plane and the 4E10 epitope residues facing the membrane interface (26,27). It is postulated that this interfacial helix is followed by a kink of ϳ90°at Lys-683 to allow insertion of the TMD in a perpendicular orientation. In addition, MDS studies predict conformational flexibility for a membrane-embedded TMD stretch close to the polar Arg-696 (14,15). However, high resolution structural data supporting these models have not been reported yet.
To obtain structural information at the atomic level on the membrane-inserting sequences of gp41, we used CD and NMR to characterize the CpreTM and TMDp synthetic peptides, encompassing the MPER C-terminal region plus the N-terminal TMD section and the full TMD, respectively (Fig. 1B, right  panel). In a membrane-mimicking low polarity environment, these two peptides displayed circular dichroism minima at 208 and 222 nm (Fig. 1C), and [] 222 /[] 208 Ͻ1 ellipticity ratios, consistent with monomeric, mainly helical structures (57). Deconvolution of the spectra using the CDPro software package (58) confirmed that the helical content in these samples amounts to ϳ70%, whereas contribution of ␤-structures is negligible (Ͻ5%). These observations underpinned subsequent structural characterization by solution NMR. Table 1 summarize the NMR results obtained for the partially overlapping CpreTM and TMDp sequences in DPC micelles and in 25% HFIP at pH 7.0 and 25°C. Based on analyses of two-dimensional COSY, TOCSY, and NOESY spectra by standard sequential methods (59, 60), we completely assigned the 1 H chemical shifts for peptide CpreTM in the two solvent conditions and for peptide TMDp in 25% HFIP. Most 13 C chemical shifts were also assigned for the two peptides in 25% HFIP by analyzing the 1 H, 13 C-HSQC spectra acquired at natural 13 C abundance. However, because of signal broadening, we could not assign TMDp in micellar media. NMR signals of CpreTM were also slightly broader in DPC than in HFIP, but in this case only the quality of the less sensitive natural abundance 1 H, 13 C-HSQC was greatly affected. Therefore, 13 C chemical shifts for CpreTM in DPC were not assigned. The assigned chemical shifts have been deposited at BioMagResBank with accession codes 19583 (CpreTM in DPC), 19582 (CpreTM in HFIP), and 19581 (TMDp in HFIP).

NMR Structures of CpreTM and TMDp Peptides-Figs. 2-3 and
Having assigned the NMR spectra, we performed a qualitative analysis of the NMR parameters. First, we examined the conformational shift ⌬␦ H␣ values, which are the differences between the ␦ values observed for the H␣ protons and those typical of random coil (RC) peptides (⌬␦ H␣ ϭ ␦ H␣ observed Ϫ ␦ H␣ RC , ppm). As seen in Fig. 2 (left panels), most ⌬␦ H␣ values in peptide CpreTM in the two examined solvent conditions and in peptide TMDp in HFIP lie outside the range typical of random coil peptides (͉⌬␦ H␣ ͉ Ͼ0.05 ppm), which is an indication that the two peptides adopt ordered structures. It is empirically well established that the ⌬␦ H␣ values are related to the and dihedral angles (61), with positive values being characteristic of ␤-strands and negative values of ␣-helices. Hence, the negative ⌬␦ H␣ values shown by most residues in CpreTM and TMDp (Fig. 2, left panels) are indicative of helical structure, which is highly populated according to the values' large magnitudes. The populations of helix estimated from the averaged ⌬␦ H␣ value (62) (60% for CpreTM and 70% for TMDp in HFIP) are in accordance with the helix contents deduced from CD data (see above). The positive and large ⌬␦ C␣ displayed by the two peptides in HFIP are also in agreement with the formation of helical structures (data not shown). It is interesting to note the similitude of the profiles observed for CpreTM in DPC and in HFIP. In both conditions, Trp-678 exhibits a positive ⌬␦ H␣ value. This positive conformational shift could indicate some irregularity in the helical structure or could be explained by the anisotropy effects of the aromatic rings. Analogously, the almost null ⌬␦ H␣ values observed in HFIP for Gly-691 in both peptides, CpreTM and TMDp, is compatible with the distortion of the helix around this residue (see below).
The strongest evidence on the formation of helical structures comes from the set of observed NOEs, because only pairs of spatially closed protons (approximately at a distance shorter than 5 Å) give rise to NOE cross-peaks. Thus, CpreTM in both DPC and HFIP and TMDp in HFIP showed nonsequential (Fig. 2, right FIGURE 1. Designation of peptides covering the HIV-1 gp41 TMD sequence and its connection to MPER. A, diagram illustrating the most important structural constituents of membrane-anchored gp41 ectodomain. FP, fusion peptide; NHR and CHR, N-and C-terminal helical regions, respectively. The same color code has been used to depict the gp41 ectodomain pre-fusion structure below (PDB accession code 4TVP). B, left panel, membrane topology predicted for the TMD and its connection to MPER. Right panel, designation of the overlapping CpreTM and TMDp peptides. C, circular dichroism spectra of CpreTM and TMDp peptides in the low polarity medium provided by 25% HFIP.
panel, and supplemental spectra), and medium range NOEs (i,i ϩ 3 and i,i ϩ 4) involving side-chain protons, for example between the aromatic protons of Phe-685 and the methyl groups of Val-689, which are observed in the two peptides, and between the Trp-678 aromatic protons and the ␦ methyl group of Ile-682 (supplemental spectra).
To visualize and get detail into the features of the helices, we performed structural calculations (see under "Experimental Procedures"). As reported previously, for the MPERp peptide (34), the superposition of CpreTM and TMDp structures show that the conformations are well defined (Fig. 3, top and bottom  panels, respectively). The pairwise root mean square deviations of the 20 lowest energy conformers calculated for each peptide and condition are also indicative of well defined structures (Fig.  3, inset). Not only the backbone, but also most side chains can be considered ordered according to the usual criterion of the 1 angle variation being less than Ϯ30°.
The CpreTM structure solved in DPC displayed a distinct pattern. At the N terminus, the hinge segment 671 NWFD 674 (33) is partially extended, consistent with the conformational flexibility put forward by previous NMR studies (24,26,29,33,34). This element is followed by a continuous ␣-helix extending to Gly-690. This residue exhibits a positive dihedral angle (48 Ϯ 1°), typical of residues at position CЈ in C-terminal helix-capping motifs (63). At this position, a kink induces a sharp change in the direction of the helix. The following Gly-691 displays again a negative angle (Ϫ72 Ϯ 2°) and helical conformation.
The main axes of 675-689 and 691-693 helices form an angle of 133 Ϯ 2°, based on calculations using the 20 models and the CalcHelix routine of the MOLMOL program (44). This helix-kink-helix motif appears to be stabilized by nonpolar interactions between side chains of Phe-685, Ile-686, Val-689, and Val-693 that cluster at the inner side of the elbow and are mostly buried from the solvent (mean accessible solvent surface areas are 31.5, 27.4, 7.9, and 25.7%, respectively).
A less abrupt, but significant change in the helix direction is also observed around the Gly-690/Gly-691 dipeptide in the CpreTM and TMDp structures solved in HFIP. In this medium, the helical axes form angles of 52 Ϯ 3 and 34 Ϯ 6°in CpreTM and TMDp, respectively. In line with the observed bending, the helix H-bond networks are also disrupted around Gly-690/ Gly-691. Thus, instead of bonding to Met-687, the Gly-691 amide is H-bonded to Ile-688 CO in CpreTM, whereas Gly- In the case of the sequential NOEs, the intensities classified as strong, medium, and weak are indicated by the thickness of the lines. The Lys residues added to the wild type sequence to improve peptide solubility are shown in italics. theless, in both HFIP structures, the angles of Gly-690/ Gly-691 residues remained negative (Ϫ56 Ϯ 1°/Ϫ107 Ϯ 3°in CpreTM-HFIP, and Ϫ61 Ϯ 3°/Ϫ66 Ϯ 4°in TMDp), indicating that bending occurs without local unfolding of the helical conformation. As compared with the residue burial observed for the kinked structure in CpreTM-DPC, the more exposed side chains of residues Ile-686, Val-689, and Val-693 further supported preservation of the overall secondary structure in the bent HFIP structures.     4). The overarching structure resulting from this reconstruction process reveals novel information on the organization of the gp41 TMD anchor and its connection to the upstream MPER region. Overall, the data confirm the adoption of chiefly ␣-helical conformations for these sequences in a membrane milieu. However, the first-time high resolution data on the stretches covered by CpreTM and TMDp reveal the presence of a kink or the bending of the helix in a membrane-buried position. Unexpectedly, no change in the direction of the helix axis is observed at position Lys-683, which was previously proposed to initiate the TMD domain (26,27,64).
Thus, the structurally variable segments at positions 671 NWFD 674 and 690 GGLV 693 delimit an unforeseen continuous helix at the N-terminal section of the TMD, spanning residues 675 ITNWLWYIKLFIMIV 689 (Fig. 4). The presence of a membrane-buried flexible element further underpins the existence of the C-terminal TMD section (Fig. 4). We propose that the TMD segments uncovered by the superposition of the NMR structures represent structurally defined functional subdomains, and we have designated them as MPER-N-TMD and C-TMD (Fig. 4).

Effects of MPER-N-TMD Residues on the Energetics of 4E10
Binding-The NMR structures determined above reveal the presence of an uninterrupted N-terminal ␣-helix that encompasses neutralizing epitopes located at MPER C-terminal region. As noted above, in membrane-inserted constructs, gp41 TMD residues specifically enhance the binding of the bNAb 4E10 to the epitope (13). Together, both observations suggest that the continuous MPER-N-TMD helix might embody the natural scaffold structure for the helical 4E10 epitope in the membrane environment. To support this hypothesis, we determined the energetic basis of antibody-peptide binding using peptide epitopes progressively elongated at the C terminus in the presence of the membrane mimetic detergent DPC (Fig. 5).   (Fig. 3), only the first coordinate set of the NMR ensemble obtained in DPC was used.

Atomic Structure of HIV-1 gp41 Transmembrane Domain
To examine the binding thermodynamics, we employed isothermal titration calorimetry (ITC) using a monovalent 4E10 Fab expressed in Escherichia coli. To validate this recombinant Fab, we crystallized it in complex with the peptide NWFDIT-NWLWYIK-KKK (4E10ep) ( Table 2). The crystal structure of the recombinant 4E10 Fab-peptide complex determined at 2.10 Å was essentially indistinguishable from that of the Fab obtained by papain digestion of the recombinant IgG (root mean square deviation ϭ 0.20 Å) (22).
The titration of Fab with peptides gave rise to well defined exothermic peaks that, after integration, defined classical sigmoidal binding isotherms (Fig. 5A). The fitting of the binding isotherms to a one-site model yielded three key thermodynamic parameters in a single experiment (dissociation constant (K D ), change of enthalpy (⌬H 0 ), and stoichiometry (n), see Table 3). The titration experiments demonstrate that in the presence of DPC all peptides engage with the Fab with high affinity (nanomolar range) driven by favorable noncovalent interactions (⌬H 0 Ͻ0). However, significant differences are observed in the relative contribution of the enthalpy (⌬H 0 ) and entropy (ϪT⌬S 0 ) terms to the binding free energy (⌬G 0 ) of each peptide to the Fab (Table 3). Specifically, the entropy loss of the peptide CpreTM upon binding to the antibody is greatly reduced in comparison with that of the peptides with shorter C-terminal tails. This observation suggests that preservation of the full MPER-N-TMD helix results in the optimization of hydrophobic surfaces in contact with Fab and/or in lower levels of conformational and roto-translational restrictions upon binding. The thermodynamic signature of CpreTM (high affinity and pre-organization of the helical structure in a membranemimetic environment) could be a desirable property for the design of better vaccines, as previously discussed for the case of alternative 4E10 epitope scaffolds (65) or stapled peptides (66).
To obtain a molecular model of the MPER-N-TMD subdomain bound to the Fab, the constant helical region (residues 674 -683) was employed to superpose the coordinates of the CpreTM peptide in DPC (determined by NMR above) with those of the shorter 4E10ep peptide-epitope obtained from the crystal structure with Fab (Fig. 5B, left panel). The fact that, according to the 1 criterion (see above), the side chains of the considered residues are ordered in the NMR structure supports the validity of this structural comparison. As shown above, the coordinates of the CpreTM in DPC show a disorganized hinge segment at its N-terminal region (Fig. 3). Therefore, the superposition of CpreTM structure with 4E10ep peptide in the Fabbound structure suggests that the segment 671 NWF 673 of CpreTM must reorient to bind adequately to the antibody. Conversely, the other residues (674 -683) superpose with those of the Fab-bound peptide with remarkable exactitude. It is also observed that the additional residues at the C terminus of CpreTM and the long CDR-H3 loop protrude from the same side of the Fab (Fig. 5B, right panel). This interaction mode might enable the peptide to engage in additional noncovalent interactions with the antibody. Thus, our data suggest that a preformed MPER-N-TMD helix might primarily promote antibody binding by reducing the entropic cost, while the flexible 671 NWF 673 segment would adjust to engage the paratope through stereospecific interactions.
Immunogenicity of the MPER-N-TMD in Rabbits-To further corroborate the implication of the MPER-N-TMD subdomain in gp41 immunogenicity, we carried out rabbit immunizations using CpreTM peptide as a vaccine. CpreTM was formulated in the nonpolar media provided by the standard FC adjuvant or liposomes. Fig. 6 displays experiments conducted to establish the effective insertion of CpreTM into the liposomes and its antigenicity at the membrane surface. Following a strategy similar to that described in our previous study (34), those experiments included a Fab4E10 mutant with the CDR-H3 loop tip ablated as a negative control. The Fab4E10-⌬loop mutant binds peptide epitopes in solution with comparable affinity, but it is not neutralizing (36), and therefore allows discriminating functional versus nonfunctional binding.
In these experiments, CpreTM (4 kDa) and liposomes (Rho emission) were recovered together from the floating fractions, demonstrating quantitative and irreversible CpreTM binding to membranes (Fig. 6, A and B, left panels). Consistent with   efficient binding to the peptide epitope inserted into the membrane, the Fab4E10-WT (28 kDa) co-floated with the liposomes that contained the CpreTM peptide (Fig. 6A, left panel). In contrast, the Fab4E10-⌬loop was mainly recovered from pellets under these conditions (Fig. 6B, left panel), indicating that this nonfunctional mutant Fab has lost the capacity to bind to the membrane-inserted epitope. Finally, none of the Fabs seemed to co-float appreciably with control liposomes lacking peptide under otherwise similar experimental conditions (Fig. 6, A and   B, right panels). Thus, the existence of an antibody-binding function correlation in the case of the liposome-CpreTM vaccine suggests that this formulation could be relevant to target 4E10-like B-cell receptors. In Fig. 7 the immunogenicity of CpreTM-FC and CpreTMliposome-MPL formulations was compared (FC1/FC2 and LM1/LM2, respectively). Antigen-specific IgGs could be recovered upon immunization of rabbits with both formulations. The midpoint titers in the FC sera were approximately an order of magnitude higher than in the Liposome-MPL sera (Fig. 7A,  left panel). However, measurements of cell entry inhibition by CpreTM-specific antibodies isolated from the sera of the four rabbits resulted in a different pattern (Fig. 7A, right panel). The most effective antibodies in these experiments were those isolated from LM sera. Thus, when grouped according to the adjuvant (Fig.  7B), the inhibitory capacity of LM antibodies was significantly higher than that displayed by the FC antibodies. Moreover, the inhibition levels of HIV-Env-mediated cell entry were significant as compared with those of the negative controls (VSV-GP pseudotypes). However, the observation that a fraction of the pseudovirus population used as negative control is indeed inhibited might indicate certain level of nonspecific reactivity of the raised antibodies with lipids or unrelated proteins.

Discussion
Recent crystallographic studies have shed light on the structural basis sustaining gp41 ectodomain metastability in the prefusion state (5,7,67). Remarkably, the gp41 native-like structure is preserved in the absence of the membrane anchors, which appear to contribute only modestly to the overall stability of the trimeric Env complex (68,69). Thus, the gp41 membrane regions may fold independently from the ectodomain to exert their anchoring function. However, the Env(671-704) sequence covering the entire gp41 TMD and a fraction of the adjacent MPER shows a high degree of conservation (18,26,36), consistent with the additional functional roles ascribed to this region during the infectious cycle, i.e. membrane fusion promotion (17,18,36) and immune response modulation (30 -32). Despite these important functional roles, high resolution structural data on the gp41 TMD and its junction to MPER were lacking ( Table 4).
The NMR structures calculated herein for the overlapping CpreTM and TMDp peptides exhibited remarkably well defined main-chain conformations and little variability among residue side-chain orientations (Fig. 3). The fact that overlapping helical regions display comparable conformational shift values strongly suggests that the two peptides adopt similar structures and hence warranted the superposition of the calculated structures (Fig. 4). This reconstruction process, which included the previously solved MPERp peptide structure (34), provides insights into the HIV-1 gp41 TMD structure and its connection to the immunogenic MPER. Most notably, the structures calculated for the CpreTM peptide in membranemimetic environments do not show evidence of the predicted helix kink at position Lys-683 (Figs. 3 and 4). On the contrary, CpreTM structures display a continuous helix that extends into the TMD to Gly-690. Furthermore, the structures calculated for the TMDp peptide corroborated the existence of a membrane-buried stretch that presents conformational variability. Thus, the structures solved here support the existence of two functional TMD subdomains as follows: MPER-N-TMD, which encompasses the N-terminal TMD helix, is longer than expected; and the shorter helical C-TMD, which is immersed in the inner leaflet of the membrane (Fig. 8).
Although the molecular mechanism remains unsolved, MPER-N-TMD promotes membrane fusion as evidenced by mutagenesis studies and peptide-based assays (18,36,64,70). Consistent with a pivotal role in the fusion process, bNAbs raised FIGURE 7. Immune response after rabbit vaccination with peptide-based formulations. A, left panel, midpoint IgG titers (negative log EC 50 dilution values). Sera from rabbits immunized with FC or LM adjuvants were titrated in ELISA using 1.4 M CpreTM. Right panel, inhibition of cell entry by MPER-N-TMD-specific antibodies in the sera of the different rabbits. HIV JRCSF (tier-2) pseudoviruses were preincubated with antibodies purified from sera of rabbits immunized with the CpreTM-based formulations. Infection of TZM-bl target cells was subsequently monitored by flow cytometry as described previously (34,37). Purified polyclonal IgGs were used at 30 g/ml in these assays. B, levels of inhibition of cell entry mediated by HIV-1-Env or VSV-GP controls. Rabbits were grouped according to the employed adjuvant. Significant inhibition of Env-HIV-1 versus VSV-GP was observed for both groups, although LM IgG-mediated inhibition was significantly higher than that mediated by FC IgG (***, p Ͻ 0.0005; **, p Ͻ 0.005). MAb4E10 (2 g/ml) was included as positive control. Means Ϯ S.D. of six determinations are shown.

TABLE 4 Atomic structures of peptides that include MPER-N-TMD and C-TMD residues
Residues outside MPER-N-TMD and C-TMD subdomains are in italics.

Method
Peptide sequence a PDB accession codes b against this element in response to infection exhibit in vitro and in vivo efficacy and exceptional neutralization breadth (21,71,72). The existence of a continuous MPER-N-TMD helix further implies that peptides ending at Lys-683 (Table 4), such as MPERp (see also Fig. 4), contain a truncated structural motif at their C termini. This truncation might explain the variety of conformations observed for the MPER C terminus (24,26,29,34), as well as the failure of the MPERp-like peptides to direct humoral responses to the MPER C-terminal epitopes (34,53,54,(73)(74)(75). Consistent with the existence of a longer than expected natural scaffold structure, ITC assays revealed that pre-formation of the MPER-N-TMD helix in a membrane mimetic medium may be required for optimal 4E10 antibody binding to its helical epitope (Fig. 5). In addition, the peptide encompassing this ele- FIGURE 8. Models for gp41 TMD organization in membranes as inferred from the calculated NMR structures. A, disparate structures adopted by the 671 NWFD 674 and 690 GGLV 693 hinges, which flank the MPER-N-TMD helix at the N and C terminus, respectively. The side chains of Trp-672 (green) and Leu-692 (red) are displayed to facilitate comparison. Side chain of Lys-683 residue is depicted in space-filling representation (yellow). B, bipartite TMD organization as inferred from the combination of CpreTM and TMDp structures measured in HFIP. C, models for TMD insertion into membranes (see text for details). ment was immunogenic in rabbits when formulated in the nonpolar environments provided by FC (reverse micelles) or LM (lipid bilayers) adjuvants, and it elicited antibodies with the capacity of blocking cell infection (Fig. 7). These results support the importance of the MPER-N-TMD helix for functional antibody and B-cell receptor binding, and therefore, preservation of this element should be considered in the future design of anti-MPER immunogens.
In the CpreTM structure solved in DPC, two flexible elements flanked the MPER-N-TMD helix (Fig. 8A). The variety of conformations reported for the N-terminal 671 NWFD 674 segment supports its structural metastability (Fig. 8A, left panel, and Table  4) (33). Our findings further suggest that the C-terminal 690 GGLV 693 residues might function as a second articulated joint embedded in the membrane core (Fig. 8A, right panel). The CpreTM structure in DPC displayed a helix kinked at position Gly-690 (Fig. 8A, C-DPC). This finding is consistent with Gly being the most common residue at position CЈ in capping motifs found at the helix C termini (63). In addition, it gives structural support to previous MDS of monomers immersed into the lipid bilayer milieu, which suggested that the midspan polar Arg and the Gly residues might provoke an inherent metastability and be capable of adopting alternating kinked forms (14).
The CpreTM and TMDp structures solved in HFIP also exhibited a change in the orientation of the helix axis at position 690 GGLV 693 (Fig. 8A, C-HFIP and T-HFIP, respectively). In this case, although the calculated angles indicated a significant degree of helix bending, the overall helical conformation is preserved in both peptides.
Our determination of the high resolution structure of TMDp confirms the predominantly helical conformation of the TMD sequence in a low polarity environment (Fig. 8B), in agreement with previous MDS and molecular modeling reports (13,14). Helix bending around position 690 GGLV 693 also supports the existence of the C-TMD subdomain. The C-TMD includes the immunosuppressive 694 GLRIVFAVL 702 motif, which resembles part of the TCR␣ TMD (30). The gp41 TMD co-localizes with CD3 within the TCR complex and functions by inhibiting CD3 antibody-stimulated in vitro T-cell proliferation (30,31). Our data here emphasize the uniqueness of this motif within TMD and provide the first detailed view of its helical structure at atomic resolution.
The segmentation into MPER-N-TMD and C-TMD subdomains and the existence of a membrane-embedded flexible site suggest different models for MPER-TMD organization in membranes (Fig. 8C). In principle, a flexible joint may allow alternating angles for MPER-N-TMD helix insertion (Fig. 8C, blue cylinders). Tilted insertion in the "a" model would adjust the length of the TMD anchor to the membrane thickness and satisfy most of the aromatic interactions with the membrane interface at its N terminus. The helix is proposed to twist within the membrane at position 691 GL 692 and then continue in a direction parallel to the bilayer normal. In the pre-fusion state, the Asp-664 residues at the C terminus of the C-terminal helical region (CHR in Fig. 1A) are splayed apart by ϳ30 Å at the base of the trimeric gp41 ectodomain (5,7). Thus, a tilted MPER-N-TMD might be required to link the TMD with the native ectodomain. A continuous helical segment comprising MPER and TMD segments was indeed suggested by a cryo-EM study (76). In addition, this arrangement might habilitate the C-TMD (red, Fig. 8B) for functional interaction with the TCR complex (30). Finally, a wedge between MPER-N-TMD and C-TMD helices might enable the formation of water defects in the bilayer as suggested by recent MDS studies (15).
The "b" model postulates MPER-N-TMD helix insertion occurring perpendicular to the membrane plane. Following the molecular modeling reported by Montero et al. (13), this arrangement would be stabilized through more extensive oligomeric contacts. We have recently proposed that, in this configuration, the MPER-N-TMD would be membrane-active and capable of extracting phospholipid from the viral envelope (36). Thus, we speculate that this arrangement might be related to fusion intermediates, emerging upon activation of the process, and proposed to promote the initial lipid exchange between viral and cell membranes (2,77).
Finally, the "c" model incorporates the sharp kink observed in the CpreTM-DPC structure. The possibility for occurrence of kinked metastable membrane structures was previously suggested by MDS analyses (14). We infer that such metastable arrangements could be disruptive for the lipid bilayer organization, and thus, we hypothesize that its occurrence might be also confined to some of the structural intermediates emerging along the fusion pathway (2,3). Mutagenic analysis revealed that the Env(696 -707) stretch encompassing the C-TMD subdomain is important for fusion (16). Thus, we speculate that within a kinked structure, the C-TMD residing in the inner leaflet of the viral membrane might gain access to the hemifusion diaphragm (2, 3) and cause its destabilization. In this way, the flexible joint between MPER-N-TMD and C-TMD may facilitate the formation of a fusion pore within a hemifusion diaphragm.
Concluding Remarks-The NMR structures determined in this and the previous study (34) suggest that gp41 MPER-TMD region is organized in contact with the viral membrane as a continuous, but distinctly jointed, helical structure (Fig. 4). Most importantly, the NMR data reveal the functional segmentation of the gp41 membrane anchor into two subdomains as follows: MPER-N-TMD, encompassing a helix, longer than expected and flanked by two hinges, which is involved in fusion and elicitation of humoral responses; and C-TMD, which includes a C-terminal helix involved in down-regulation of the TCR-mediated signaling and possibly in the promotion of the last steps of the fusion process. Thus, the membrane interaction model emerging from these studies diverges from the current view that considers MPER as an interfacially adsorbed helix, bent at position Lys-683 and followed by the TMD inserting perpendicularly to the viral membrane plane. Moreover, the experimental results indicate that the MPER-N-TMD may function as a helical scaffold to increase affinity of anti-MPER antibody in membrane mimetics and suggest that its preservation may be a prerequisite for the immunogenic efficacy of peptide vaccines targeting the MPER C-terminal region.
Acknowledgments-Access to beamline BL5A was granted by the Photon Factory Advisory Committee (Proposal Number 2013G738). Technical assistance by Miguel García is greatly appreciated. Critical reading by Dr. Douglas V. Laurents is also acknowledged.