HIV-1 gp41 envelope residues 650-685 exposed on native virus act as a lectin to bind epithelial cell galactosyl ceramide.

The initial step in the interaction between human immunodeficiency virus (HIV-1) and epithelial cells is the binding of HIV-1 envelope glycoproteins to the epithelial cell galactosyl ceramide (GalCer). Here we show that HIV-1 envelope gp41 residues 650-685 bind GalCer in a galactose-specific manner. The gp41 residues that display this lectin activity are highly conserved among HIV-1 isolates and constitute three regions: residues 650-661, which encompass a charged helix; residues 662-667, referred to as the conserved epitope ELDKWA, the epitope recognized by antibodies that neutralize HIV-1 entry in epithelial and CD4(+)-mononucleated cells; and residues 668-685, a hydrophobic Trp-rich sequence that stabilizes the structure of the galactose binding site. Similar to other galactose-specific lectins, the gp41 lectin site is active only as an oligomer. Finally the orientation of the galactose toward the gp41 lectin site appears to be controlled by the lipid microenvironment of the epithelial membrane. From the experimental data we construct a theoretical model of the interaction between gp41 and GalCer based on thermodynamic considerations. This model integrates the dynamics and the spatial organization of the viral envelope glycoproteins, GalCer organized in raft microdomains in the apical region of the epithelial cell membrane and the interfacial water. Characterization of the minimal sequence and structure of gp41 in direct interaction with GalCer may help unravel the still unknown immunogenic determinant able to elicit antibodies against ELDKWA and target of one of the rare neutralizing antibodies against gp41.

The initial mucosal cells encountered by HIV-1 are either the epithelial cells in the monostratified epithelium of the gastrointestinal tract, endocervix, or rectum or the dendritic cells in the submucosa in the pluristratified mucosa of the vagina or the anus (1). The virus infects and replicates poorly in these initial target cells because fusion and integration are rare events. Instead, interaction between HIV-1 1 envelope glycoproteins and epithelial cells or dendritic cells initiates transcytosis or trapping, respectively, which are non-degradative, non-integrative processes that facilitate delivery of intact virus to CD4 ϩ target cells in the submucosa.
As the first step of HIV-1 interaction with monostratified epithelium, HIV-1 binds to the apical membrane of gastrointestinal, endocervical, or rectal epithelial cells. This membrane has a characteristic lipid composition with the external leaflet highly enriched in glycosphingolipids (2), including the glycosphingolipid galactosyl ceramide (GalCer). GalCer is a monohexosyl ceramide with a polar head composed of a single galactose linked to the ceramide through the sphingosine amide and alcohol groups. GalCer participates in the establishment of microdomains, referred to as rafts, which act as platforms for endocytosis (3,4) and transcytosis (5,6). GalCer is also highly expressed on the surface of immature dendritic cells. 2 Importantly, rafts were shown to mediate pathogen interaction with the target cell (for a review, see Ref. 7), for example respiratory syncytial virus endocytosis into dendritic cells (8) as well as SV40 entry in caveosomes (9).
HIV-1 envelope glycoprotein consists of two non-covalently associated subunits, the transmembrane gp41 and the surface gp120. At the surface of the virus, HIV-1 envelope glycoproteins oligomerize to form a spike in which gp120 forms a cap covering most of the gp41. Crystallographic studies of recombinant proteins suggest that the extracellular domain of gp41 contains at the N-terminal end a fusion peptide buried below gp120 in the spike and two helices referred to as the N-terminal and C-terminal helix (10,11). A study of interdomain interactions involved in the virus spike assembly shows that the region extending outside the gp120 cap and, therefore, accessible to the solvent comprises residues 650 -685, the last residues before the transmembrane domain (12). Residues 650 -685 cover the charged C-terminal helix enriched in amino acids containing free amino and carboxyl groups and a succession of aromatic amino acids (5 Trp, 1 Phe, and 1 Tyr). This region has rarely been studied due to its hydrophobicity despite the importance of the set of Trp in CD4 ϩ cell infection (13).
Until now, most structural and functional studies of gp41 domains have focused on the initial step involved in HIV-1 fusion with CD4 ϩ mononuclear cells (14) and on the structure of gp41 in its fusogenic conformation. In contrast, the native structure of gp41 on the virion, embedded in the viral membrane, before its interaction with the target cell remains poorly defined.
We have shown earlier (15) that HIV-1 crossed a monostratified epithelial barrier using transcytosis. HIV-1 transcytosis occurred without virus fusion with epithelial cells. A transcytosis process did not modify HIV-1 since the virus retained its infectious potential and could spread infection to CD4 ϩ -mononucleated cells. In our recent study of the initial steps in HIV-1 transcytosis across CD4 Ϫ epithelial cells (16), we demonstrate that gp41 binds to epithelial cell GalCer and that disruption of GalCer-containing raft microdomains inhibited HIV-1 transcytosis.
The phenotype of HIV-1 in chronically infected patients is a mixture of two virus subtypes, the R5 virus and the X4 virus, using CCR5 and CXCR4 as coreceptor for infection of CD4 ϩ cells, respectively. However, after acute infection, mainly R5 viruses are isolated and, therefore, are supposed to be the main vector of infection (17). This suggests that a selection process occurs at the site of HIV entry in the recipient from the HIV-1 mixture inoculated by the donor.
In human tissues, epithelial cells express in addition to Gal-Cer only CCR5 but not CXCR4 (18). Similarly, dendritic cells, the recognized primary HIV cell target in pluristratified mucosa, highly express CCR5 (19). Antibodies anti-CCR5 block HIV transfer from epithelial cells (18) or dendritic cells to CD4 ϩ cells, the next HIV cell target in the mucosa, but not HIV binding to epithelial 2 or dendritic cells (20). Thus, HIV binding to its initial target cells at mucosal sites appears independent of HIV coreceptor CCR5 or CXCR4.
In contrast, the binding of gp41 to GalCer involved the conserved ELDKWA region (residues 662-667) of gp41. This region is a potential vaccine target since antibodies to ELDKWA inhibit gp41 binding to epithelial GalCer (16), block HIV-1 transcytosis across a tight epithelial barrier (21), and neutralize infection by a large array of HIV-1 primary isolates both in vitro and in vivo (22,23). However, despite numerous attempts, neutralizing antibodies to ELDKWA could not be naturally elicited in vivo. Characterization of the immunogenic determinant able to elicit HIV-1-neutralizing antibodies against ELDKWA remains a challenge for the design of a vaccine against HIV-1.
Lectins are glycoproteins that bind sugar with a high specificity. Numerous studies (24 -26) on the lectin carbohydrate binding properties demonstrate that this interaction is dependent on charge transfer processes facilitated by hydrogen bond formation. Two kinds of functional groups, free amino and carboxyl groups, and two aromatic structures, Trp and Tyr, could participate in the formation of hydrogen bonds with the surrounding water and may form charge transfer complexes with carbohydrates. As discussed above, residues of the solvent-exposed portion of gp41 contain these two types of functional groups.
Here, we sought to determine the minimal sequence and the structure and the oligomerization state of gp41 including ELDKWA able to interact with epithelial GalCer. Such a domain may provide a clue on the immunogenic determinant able to elicit HIV-1-neutralizing antibodies against ELDKWA. We also sought to determine whether this interaction between gp41 and epithelial GalCer is of the lectin-carbohydrate type. In addition, we investigated the role of the raft lipid microdomain in the orientation of the galactose head of GalCer toward its binding site on gp41. Using the present experimental data and thermodynamic computation of the affinity constant of the gp41 peptide for the GalCer, we have constructed a theoretical model of HIV-1 lectin interaction with glycosphingolipids organized in an epithelial cell membrane. According to this model, the dynamics and the spatial organization of the viral spike lectin and the epithelial cell membrane play an essential role in the gp41 lectin-carbohydrate interaction.

Epithelial Cell Binding Assay of Peptides
Peptide binding to HT29 epithelial cells was performed as previously described (16) using biotinylated ELDKWA or P1 at the indicated concentrations (25-500 M).

Fluorescent Glycosphingolipids
Bodipy-galactosylceramide (BpGalCer*), a C12-coupled bodipy, Bodipy-glucosylceramide (BpGluCer*), and Bodipy-sphingomyelin were purchased from Molecular Probes (Eugene, OR) and stored as an ethanolic solution. To study the interaction of Bp-glycolipids* and peptides, the ethanolic solution of Bp-glycolipid* was directly injected in the peptide-containing solution of PBS at a final concentration of 1-6 M Bp-glycolipid*. At these concentrations, Ͻ5 mol %, we verify that no collisional quenching due to any form of aggregation was observed. This is in agreement with the characteristics of the probe, as indicated by the manufacturer.
Single bilayer lipid vesicles were prepared according to the ethanol injection procedure (27). Aliquots of PC, unlabeled GalCer or GM1, and cholesterol, each dissolved in chloroform, were mixed at the following ratio. GalCer⅐PC (4:1) with or without cholesterol is referred to as "raft lipids," and GM1⅐PC (1:4) with or without cholesterol is referred to as "non-raft lipids." The solvent was removed under nitrogen. Dry lipids were dissolved in ethanol at a concentration of 1-5 mM.
The fluorescent Bp-glycolipid* in ethanol was added to the ethanolic solution at concentrations of 10 -40 M and rapidly injected through a needle with continuous and vigorous stirring into the peptide-containing solution to obtain the desired final concentration. The final proportion of ethanol in the working solution was always Ͻ0.5%.

Spectroscopy
Circular Dichroism Measurements-The circular dichroic spectra of peptides in the far UV (, 200 -260 nm) was obtained on a spectrometer Mark V dichrograph Jobin et Yvon equipped with software for the acquisition of the data. The concentration of peptides varied from 10 to 500 M using a different path length from 0.01 to 2 cm. Blanks for buffer were subtracted, and spectra were analyzed based on the characteristic spectra of ␣-helix, ␤-pleated sheet and random coil (28) and using the equation () ϭ 3300⌬⑀ /m ϭ ()␣ P␣ ϩ ()␤ P␤ ϩ () P, with P␣ ϩ P␤ ϩ P ϭ 1 for each pair of wavelengths between 210 and 222 and between 222 and 231; ⌬⑀ is the molar dichroic absorption coefficient (M Ϫ1 cm Ϫ1 ) at ; m is the number of residues per polypeptide chain; () is the molar ellipticity per residue; ()␣ , ()␤ , and () , are the characteristic molar ellipticities per residue in each structure; and P␣, P␤, and P are the different proportions of each secondary structure.
Absorption Spectra-Absorption spectra were recorded on a UVvisible Varian DMS 70 spectrophotometer.
Fluorescence Measurements-Steady state fluorescence emission and excitation spectra were obtained on a spectrofluorimeter with the experimental arrangement (29,30) equipped with MAC II Ci and Labview 3.1 software for the acquisition of data and Kaleidagraph 3.0 for analysis of the data. The entrance and exit slit widths were 2 mm for the McPherson excitation monochromator and 1.8 and 1 mm for the Baush & Lomb emission monochromator.
Quartz cuvets of 2 ϫ 10 mm or 3 ϫ 3 mm were positioned with 2 perpendicular micrometers. The absorbance peaks (⑀/cm) of the samples were less than 0.1 to eliminate inner filter effect in all measurements with the peptide and the fluorescent lipid alone. In the experiments with the fluorescent probe incorporated in lipid vesicles, corrections have been made to account for the fractional absorbance of the solution using the formula B() ϭ A()/A() ϩ A s (){1 Ϫ 10 Ϫ(A() ϩ As() }.
All spectra were made at 22°C. Fluorescence quantum yield (Q d ) of the donor peptides were determined from a solution of Trp in PBS using a value of Q o ϭ 0.13 for Trp, as given by Chen et al. (31). Calculation of the energy transfer efficiency from the P1 Trp residues to the Bodipy on C12 of the ceramide chain of BpGalCer* were made using the decrease in the value of the maximum I f in arbitrary units at 350 nm of the peptide bound to BpGalCer*, I fa /I fo .
Q o , the quantum yield of P1, was equal to 0.078. The efficiency of the transfer is given by T ef ϭ Q a /Q o , where Q a is the quantum yield of P1 in the presence of the acceptor obtained from I fa /I fo. The Förster radius, calculated for such a transfer from the Förster equation, is R o ϭ (JQ o K 2 )/ 4 ) 1/6 (9.79 ϫ 10 3 Å), where K 2 is the dipole-dipole orientation factor, taken as 2/3 for randomly oriented donor and acceptor molecules, and is the wave number. J, the donor-acceptor spectral overlap integral in cm 3 M Ϫ1 , is calculated from the emission spectrum of the P1 absorption spectrum of BpGalCer* overlap between 300 and 475 nm, ⑀M 280 nm (P1) ϭ 29.73 ϫ 10 3 , ⑀M 360 nm (BpGalCer*) ϭ 28.8 ϫ 10 3 , and ⑀M 505 nm (BpGalCer*) ϭ 85 ϫ 10 3 . For measurements in the presence of lipids, the fractional molar absorption for P1 in the lipid medium is ⑀ ϭ 23.0 ϫ 10 3 . Each value on the curves is the representative result of more than 10 reproducible experiments for each set of data in different conditions obtained after standard control for each day.

RESULTS
The Charged C-terminal Helix and Trp-rich Region Surrounding ELDKWA on HIV-1 gp41 Are Required for gp41 to Bind Epithelial Cell GalCer-We previously showed that the conserved hexapeptide ELDKWA in gp41 was the determinant for gp41 binding to epithelial GalCer (16). Gp41 binds epithelial cell GalCer, and this interaction is inhibited by an excess of ELDKWA peptide. The ELDKWA peptide appears to be a key component of the exposed sequence of gp41 capable of binding the target cell. Therefore, we determined whether ELDKWA binds directly to the epithelial cell GalCer using our in vitro model (15). As shown on Fig. 1 (first row), ELDKWA (25-500 M) did not bind to epithelial cells.
In the gp41 molecule, ELDKWA (gp41 residues 662-667) is surrounded by a charged helix at the N terminus and a hydrophobic Trp-rich region at the C terminus. These two regions may provide a structure to ELDKWA (32). Therefore, we synthesized a longer peptide P1 of 35 amino acids (gp41 residues 650 -685) encompassing part of the helix, ELDKWA, and the Trp-rich region. As mentioned above, P1 covers the part of gp41 exposed at the surface of the viral particle before the virus interacts with target cells (12).
In contrast to the inability of ELDKWA alone to bind epithelial cell surface, P1 bound the epithelial surface in a concentration-dependent manner ( Fig. 1, second row). The concentration-dependent binding of P1 to GalCer suggests that P1 adopts a structured, concentration-dependent oligomeric state. Such a result would be in agreement with the well recognized oligomeric structure of the intact virion envelope spike (33-36) that interacts with epithelial cells of the mucosa. In gp41, LZ peptide (residues 586 -608) has been shown to interact with the gp41 C-terminal helix, which share residues 650 -664 with P1, as a necessary step in the fusion process. Preincubation of biotinylated P1 with the LZ in stoichiometric amount (100 M) prevented biotinylated P1 binding to the epithelial cell surface (Fig. 1, third row). The present results suggest that gp41 in the fusogenic state (when both the C-terminal helix and LZ interacts tightly) cannot interact with GalCer (see also Fig. 4d).
Amino Acids 650 -685 of gp41 Oligomerize and Exhibit a Partial Helical Structure-We next determined the structure and oligomerization state of P1 by circular dichroism (CD) in the far UV and measured P1 molar ellipticity as a function of concentration. As shown in Fig. 2a, below 50 M the CD spec- and P2 (residues 660 -679) containing the Trp-rich region reaches a maximum emission at 348 nm, characteristic of Trp in a relatively hydrophobic environment. P3 (residues 650 -669), containing only 1 Trp in a more polar environment, exhibits a maximum at 353 nm with a lower relative trum of P1 indicates a poorly structured molecule, as shown for related peptide DP178/T20 (residues 643-678) at 10 M (14). At a concentration above 50 M, the CD spectra of P1 exhibit well defined extrema at 207 and 222 nm, which is characteristic of an ␣-helical structure. With a value of the molar ellipticity of 222 ϭ Ϫ7.76 ϫ 10 3 and 207 ϭ Ϫ 8.2 ϫ 10 3 calculated for 35 residues and given a value of Ϫ 22 ϫ 10 3 for 100% helix (37), the estimated percent of helicity for P1 is ϳ35%. The CD spectra obtained at higher concentrations, namely 250 and 500 M, also exhibited negative peaks at 207 and 222 nm, thus confirming the helical structure of the peptide (data not shown).
However, the mean residue ellipticity value at 220 nm ( 220 ) exhibited a direct linear relationship with the P1 concentration ( Fig. 2b), displaying a slope of 0.045 for P1 between 25 and 100 M. For P1 between 100 and 500 M, the slope is 10 times lower (0.0045) even if no aggregation was observed from the absorbance value. Such a P1 concentration dependence of 220 indicates on the one hand interchain interaction, as shown for poly-lysine peptides (38), and on the other hand secondary structure changes, as shown for staphylococcal-D-toxin (39).
The Charged C-terminal Helix and Trp-rich Regions Bracketing ELDKWA Together with ELDKWA Are Required for gp41 to Bind GalCer-We next used fluorescence resonance energy transfer (FRET) between the Trp present on P1 and Bodipylabeled GalCer (BpGalCer*) to determine the exact sequence of gp41 peptide required for HIV-1 gp41 to bind to GalCer. P1 contains three modules, 1) the charged C-terminal helix, 2) ELDKWA, and 3) the Trp-rich region of gp41. Besides P1, three additional peptides were synthesized; they are P2 (residues 660 -679), including ELDKWA and the Trp-rich region; P3 (residues 650 -669), including the charged C-terminal helix and ELDKWA; and P4 (amino acid 640 -659), including the charged C-terminal helix. The fluorescence emission spectra of the P1-4 peptides at 50 M as a result of Trp excitation at 280 nm are shown on Fig. 3a. P4 without Trp served as a negative control.
For FRET, the peptide Trp residues were used as the donor, and BpGalCer* was used as the acceptor. Fig. 3b shows that the probes are compatible for FRET. Indeed, the fluorescence emission spectrum of the Trp-containing peptides (P1, P2, and P3) overlaps the BpGalCer* absorption spectrum in the wavelength range of 300 -475 nm.
BpGalCer* excited at 488 nm fluoresced with a peak at 525 nm only when the probe was in a hydrophobic medium such as organic solvents, lipids, or P1, a peptide with hydrophobic residues, namely five Trp, one Phe, and one Tyr, but did not fluoresce in the aqueous buffer PBS (Fig. 3c, inset). The maximum intensity of intrinsic fluorescence (I f ) of BpGalCer* excited at 488 nm was dependent on the P1 concentration (Fig.  3c). In the presence of only P1 without additional lipids, the relationship between I f and P1 concentration is linear. However, a sharp transition occurred at 50 M P1 when BpGalCer* is embedded in P1 and raft lipids. The BpGalCer* is then in the hydrophobic environment of the lipid fatty acyl chain, protect-ing the fluorescent Bodipy from radiation-less energy loss interaction with the aqueous solvent and, therefore, increasing the quantum yield of the fluorescent probe. Therefore, on the one hand BpGalCer* adopts a specific raft-induced orientation due to its hydrophobic part embedded among the fatty acid tails of the surrounding lipids, and on the other hand, the interaction of the galactosyl moiety with the P1 lectin site stabilizes this orientation at the concentrations where the lectin site is efficient. Such characteristics could explain the highest intrinsic fluorescence of the probe, evidenced by the transition in the I f value, in the presence of 50 -100 M P1.
A significant amount of energy transferred from P1 Trp residues to BpGalCer* when P1 (50 M) was excited at 280 nm (Fig. 4a). The efficiency of this transfer was T ef ϭ 0.58 ϫ 0.078 ϭ 0.045, as calculated with I fa /I fo ϭ 0.58. The Förster radius (R o ), the distance at which energy transfer is 50%, calculated with a spectral overlap value of J ϭ 5.9 ϫ 10 Ϫ16 cm 3 M Ϫ1 (Fig. 3b), was 28 Å, which corresponds to the distance between the Bodipy on the ceramide chain and the carbohydrate binding site in the peptide. Given the distance between the C12 atom of the ceramide, to which the Bodipy probe is conjugated, to the galactose ring as 12-15 Å, one can calculate the distance between the galactose and the galactose binding site in P1 (at 50 M) to be less than 15 Å.
FRET from P1 is sugar-specific as no transfer occurred to Bodipy-glucosylceramide (BpGlucCer*) (Fig. 4b); a similar absence of transfer was observed from P1 to Bodipy-sphingomyelin and -ceramide (data not shown). These results show that P1 at 50 M presents a galactosyl-specific binding site.
From P1-4, only P1 was able to transfer energy to BpGal-Cer* (Fig. 4a). The importance of the charges on the helical part of the peptide in FRET is revealed by the absence of FRET from P2 to BpGalCer*. The absence of FRET from P3, containing the charged helix and ELDKWA but not the stacked hydrophobic Trp aromatic rings, provides evidence for a hydrophobic region requirement for gp41 binding to GalCer (Fig. 4a).
Carbohydrate sites of galactose-specific lectins have been shown to arrange as a pocket containing non-contiguous charged Glu and Asp and hydrophobic Trp amino acids. Point mutations on analogous amino acids were performed on P1; i.e. Glu-662 to Ala and Asp-664 to Ile in the one hand and Trp to Gly. FRET to BpGalCer* could not be detected with either of P1 mutants (Fig. 4c).
In summary, all three portions of P1, the charged C-terminal helix, ELDKWA, and the highly hydrophobic Trp-containing sequence, are necessary and sufficient for highly efficient binding of gp41 to GalCer. Amino acids Glu-662, Asp-664, and Trp-666 are essential for binding GalCer. Altogether, P1 exhibited the characteristics of a galactose-specific lectin.
The Efficiency of Energy Transfer from P1 to GalCer Is a Function of the P1 Concentration-Because of P1 structural change as a function of its concentration (Fig. 2), FRET experiments were conducted with P1 concentrations between 10 and 100 M. The transfer efficiency is directly proportional to the reciprocal of the sixth power of the linear distance between  4. Gp41 residues 650 -685 interaction with GalCer. a, BpGalCer* interacts with P1, not with P2 or P3. BpGalCer* (6 M) and P1, P2, P3, or P4 (50 M) in PBS were excited at 280 nm, and fluorescence emission spectra were recorded. A maximum emission peak at 525 nm due to BpGalCer* fluorescence together with a decrease in the P1 I f at 280 nm are characteristic of FRET between P1 and BpGalCer* and indicate the interaction between the two probes. No FRET was detected from P2, P3, or P4. b, P1 interaction with BpGalCer* is specific for galactose. P1 (50 M) and BpGalCer* (6 M) or BpGluCer* (6 M) in PBS were excited at 280 nm. FRET occurred only between P1 and BpGalCer*, but not donor and acceptor but also is strongly dependent on their mutual orientation. Therefore, the variation in I f of BpGalCer* in the FRET experiment (520 nm when excited at 280 nm) as a function of P1 concentration provides an indication of the distance and the relative orientation of the acceptor toward the donor. As shown in Fig. 4e, both in the presence and absence of unlabeled lipids and at a constant concentration of BpGalCer* for P1 between 10 and 25 M, the transfer efficiency is not significant; transfer is only efficient for P1 above 25 M, with a direct linear relationship between I f and P1 concentration. A sharp transition in the transfer efficiency occurred for P1 at 50 M with a slope ratio between 25-50 and 50 -100 M of 4. This concentration dependence of I f (in the same concentration range as for CD experiments) indicated that the formation of the carbohydrate binding sites specific for galactose on P1 depends on a structural rearrangement of the peptide related to a non-ideal self-association equilibrium. Together with a similar concentration dependence of P1 binding to epithelial cells (Fig. 1), namely above 25 M when P1 adopts an oligomeric conformation as shown by CD, these results confirm that the carbohydrate binding on P1 is dependent on the oligomeric state of the peptide.
The Efficiency of the Energy Transfer Depends on the Lipid Microenvironment of BpGalCer*-In the epithelial cell membrane, GalCer is arranged in lipid raft microdomains (2). Therefore, we examined the influence of the lipid environment of the acceptor BpGalCer* during FRET from P1. BpGalCer* was incorporated at the same molar ratio BpGalCer*/L (1/200) in liposomes composed of different proportions of unlabeled GalCer, PC, ganglioside (GM1)-mimicking raft microdomains (GalCer/PC, 4:1), or mimicking membrane regions distinct from raft microdomains, poor in glycolipids, and referred to as non-raft (GM1/PC, 1:4). The transfer efficiency between Trp in P1 (50 M) and BpGalCer* in a raft environment is T ef ϭ 0.0435, as compared with T ef ϭ 0.045 (see above), slightly more efficient than for the BpGalCer* without additional lipids in PBS. In contrast, when BpGalCer* is in a non-raft environment, no FRET occurred (Fig. 4d). In membranes, cholesterol is required to stabilize the lipids raft microdomains. We therefore next included cholesterol in the various liposomes. FRETs were not modified upon addition of cholesterol either to the raft or to the non-raft liposomes (not shown), suggesting that small-sized liposomes mimic uniform microdomains that do not need cholesterol to be stabilized, in contrast to a biological membrane.
To confirm the specificity of P1 interaction with GalCer, similar experiments were conducted replacing Bp-GalCer* by Bp-SM* or Bp-GlcCer* in the various liposomes. No FRET could be detected in these conditions (not shown).
Leucine Zipper Interacting with P1 Inhibits the Binding of P1 to GalCer-Using solubilized synthetic peptides (34,40) and x-ray crystallographic analysis (10,41), the N-terminal helix (amino acids 540 -590), also referred to as a leucine zipper (LZ), has been shown to interact with the C-terminal helix (amino acids 620 -664). This interaction mimics the fusogenic conformation that gp41 is thought to adopt after interaction with CD4 ϩ T cells. Therefore, because P1 shares amino acids 650 -661 with the C-terminal helix peptide, we investigated the effect of this LZ on the P1 interaction with GalCer. As shown on Fig. 4f, when an LZ peptide (residues 586 -608) was added in stoichiometric amount to P1, regardless of the peptide concentration, FRET to BpGalCer* in raft liposomes is completely inhibited. Similarly, preincubation of P1 with LZ in a equimolar ratio strongly inhibited P1 binding to epithelial cells (Fig. 1,  third row). This LZ effect was specific, since preincubation of P1 with controlled (Scrambled) peptide had no inhibitory effect (Fig. 1, third row), further suggesting that LZ inhibition occurred by blocking the P1 galactose binding site. The inhibition of P1 binding to GalCer by P1 preincubation with gp41 LZ indicates that gp41 in the fusogenic state (when both the Cterminal helix and LZ interacts tightly) cannot interact with GalCer.
Thermodynamic of the Interaction between gp41 and the Epithelial Cell Membrane-To obtain quantitative data on the different forces involved in the P1 lectin interaction with Gal-Cer, we calculated the thermodynamic parameters of that interaction, namely to obtain a value of the affinity constant. We used the data computed for the interaction of the lectin BSL-I (Bandeiraea simplicifolia) and its specific carbohydrate ligand, the ␣-anomer of the galactose (26), as a paradigm. From the experimental data presented above, we proposed that three Glu residues (657, 659, 662) and one Asp residue (664) are the acceptors of the C6, C4, C3, and the C2 hydroxyl groups of the galactose, respectively, with a computed ⌬G value f ϳϪ14 kJ mol Ϫ1 . The ␣-1 anomeric oxygen of the GalCer could act as a hydrogen bond acceptor from a neutral polar group of gp41 lectin site P1, likely one of the glutamines (Gln-650, -652, -653, or -658), with a ⌬G value of ϳϪ3.5 kJ mol Ϫ1 . If each one of the hydroxyls is hydrogen-bound, the free energy of binding/mol of GalCer would approximately be ⌬G ϳ Ϫ60 kJ mol Ϫ1 at 25°C.
From the equation ⌬G°ϭ ϪRT ln K, the affinity constant K a for the binding of one galactosyl moiety of GalCer to the amino acids proposed as hydrogen bond acceptors or donors in the lectin binding site of the P1 peptide, would have a value of ϳ53 ϫ 10 9 M Ϫ1 . Such a high affinity value is similar to the value calculated for an antigen-antibody interaction (42) and for the monoclonal antibody 2F5 reactivity toward a complex of two peptides from the N-and C-terminal domains of gp41 (33). However, it is important to underline that the accessibility to the epithelial cell surface of the gp41 lectin site within the HIV envelope spike is more restricted than the one of the free peptide P1. The K a value calculated for P1 and GalCer is certainly not directly valid for the virion spike. Nevertheless the value of K a for the intact virion spike should be of the same order of magnitude, and it implies that the virion enters the cell still bound to the GalCer of the cell membrane. Dissociation could take place in an endosomal compartment.
Computation of the Free Energy of Activation of the Electron Transfer between gp41 Peptide and GalCer-The experimental data obtained by FRET between P1 and GalCer establishes the electrochemical nature of the interaction between P1 and Gal-BpGluCer*, indicating that P1 binds specifically the galactose ring of the glycolipid. c, Glu-662, Asp-664, and Trp-666 are essential for P1 interaction with BpGalCer*. P1 or mutated P1 (P1 E662A,D664I or P1 W666G (50 M)) and BpGalCer* (6 M) in PBS were excited at 280 nm. FRET occurred only between P1 and BpGalCer* but not the mutated peptides, indicating that Glu-662, Asp-664, and Trp-666 are part of the P1 lectin site that interacts with the glycolipid. d, P1 interaction with BpGalCer* occurs in raft lipid but not in non-raft lipid environment. P1 (50 M) and BpGalCer* (6 M) in raft or non-raft lipids were excited at 280 nm. FRET between P1 and BpGalCer* occurs when the probes are in raft lipids with a maximum at 525 nm, in contrast to an absence of FRET when the probes are in non-raft lipids. e, FRET depends on P1 concentration. BpGalCer* (6 M) and P1 (from 10 to 500 M) in PBS or in raft lipids were excited at 280 nm. Fluorescence emission spectra were recorded. The plot of the maximum intensity of the fluorescence emission peak at 525 nm (I f ) as a function of P1 concentration exhibits a sharp transition between P1 concentration of 25 and 50 M in both raft lipids and PBS environment, indicating that the GalCer* binding affinity to the P1 depends strongly on P1 concentration and, therefore, on P1 secondary structure (see Fig. 2). f, P1 interaction with BpGalCer* was inhibited by gp41 LZ. BpGalCer* (6 M) and P1 (50 M) or P1 ϩ LZ (1:1) at 50 or 100 M in PBS were excited at 280 nm. FRET from P1 to BpGalCer* was inhibited when P1 is in stoichiometric interaction with LZ (50 or 100 M).
Cer. Therefore, we sought to determine the different components of the free energy of activation of the electron transfer, the basis for signaling between gp41 and the epithelial cell membrane.
The 1:1 electron transfer that will occur between gp41 and GalCer can be computed as a sequence involving the formation of a precursor complex as described by the following equations, gp41 ox Ϯ GalCer red 7 gp41 ox ⅐GalCer red (Eq. 1) where K 0 is the equilibrium constant, gp41 ox ⅐GalCer red ¡ gp41 red ⅐GalCer ox ¡ products (Eq. 2) where k et (s Ϫ1 ) is the electron transfer rate constant within the assembly of the precursor. In such an expression (43), the specificity of the participants of the bimolecular interaction is expressed by the K 0 term, the equilibrium constant for the formation of the gp41⅐GalCer precursor complex. The observed second order rate constant k 12 may be expressed as, For an electron transfer reaction described by Equations 1 and 2, the equilibrium constant for the formation of the complex gp41⅐GalCer may be described by the following equation, Using the coulombic term calculated above, W coul ϭ Ϫ12.37 kJ⅐mol Ϫ1 , and taking a value of -21 kJ⅐mol Ϫ1 for the noncoulombic term (hydrogen bonds and salt bridging) (44), the equilibrium constant for the formation of the gp41⅐GalCer complex would have the value of K 0 ϭ 13.7 ϫ10 9 , which is in the same range as the value calculated for the affinity constant K a of the GalCer for the lectin as described above.
The unimolecular rate constant for elementary electron transfer may be expressed as follows, Here k B is the Boltzmann constant, k el is an electronic transmission coefficient, and ⌬G* is the free energy of activation of electron transfer within the precursor complex. The term ⌬G* comprises contributions from aqueous solvent hydration shell reorganization energies. k el varies exponentially with r(Å), the distance separating the donor and acceptor molecules; for r ϭ 12 Å, k el has been predicted to be 6 ϫ 10 Ϫ6 (43). From the FRET experiments reported above, the experimental value for the distance between the galactosyl moiety of GalCer and the lectin binding site of gp41 is Ͻ15 Å. Therefore, from the value of K 0 ϭ exp(Ϫ⌬G*/RT) ϭ 13.7 ϫ 10 9 , as given above, and k el ϭ 6 ϫ 10 Ϫ6 , the value of k et , calculated from Equation 6, will be of k et ϭ 182 s Ϫ1 and ⌬G* ϭ Ϫ11 kJ⅐mol Ϫ1 .

DISCUSSION
In this study, we determined that HIV-1 gp41 residues 650 -685 are the minimal region required for gp41 to bind GalCer, the epithelial cell receptor for HIV-1. This interaction exhibits all the characteristics of lectin-carbohydrate binding based on charge transfer processes facilitated by hydrogen bond formation (24 -26). Such a role of glycoconjugates covalently linked to cell membrane lipids as recognition sites for toxins and viruses has been well established (45) (or review, see Refs. 7 and 32).
The characteristics of P1 established above, namely the absolute requirement for GalCer binding of the charged helical part, ELDKWA, specifically Glu-662, Asp-664, and Trp-666, and the following hydrophobic sequence (residues 668 -685) allow the development of a scheme of the carbohydrate binding site of P1 (Fig. 5). This site comprises three glutamic acids, E-657, -659, and -662, and the aspartic acid D-664, acting as H ϩ acceptors, from the hydroxyl group of the C6-C4-C3-C2 in the galactose ring. Glutamine 658 would act as a H ϩ donor to the oxygen 1 from the galactosyl ring of GalCer either directly or through some structured water molecules at the interface of the membrane and the peptide. The hydrophobic Trp (residue 666) of ELDKWA faces the hydrophobic region of the galactose and should stabilize the orientation of the carbohydrate in its gp41 binding site. The interaction between the hydrophobic segment of P1 (residues 666 -685) appears as primarily responsible for stabilizing an oligomeric state. The interface between two hydrophilic charged C helices with ELDKWA (residues 650 -666) brought together by the oligomerization would form the carbohydrate binding site.
Two other functional groups, the free amino and carboxyl groups, and the aromatic structures Trp and Tyr groups could participate in the formation of hydrogen bonds between the peptide and the surrounding water and may form charge transfer complexes with carbohydrates. The stack of hydrophobic Trp and Phe in P1, which are exposed to the surrounding aqueous medium, modify the water-water interactions. Indeed, the hydrophobic stack increases the degree and strength of hydrogen bonding between water molecules by shifting their spatial and orientational distribution, forming the so called "structured water" (46). This structured water participates by an important entropy contribution (47) to the charge transfer from the polar, neutral, and charged amino acid residues of P1 to the galactosyl ring of GalCer in the epithelial cell membrane. Thus, the P1 galactose binding site presents all the characteristics of the galactose binding animal C-lectins, including the dependence of the galactose binding on the oligomerization of the lectin (32, 48 -50).
HIV-1 interaction with the epithelial cell GalCer involves actually three partners, the viral envelope spike, including gp120 (15,51)  Spatial Component of the Interaction-HIV-1-envelope protein complex is formed by oligomers of the two-envelope glycoprotein subunits gp120 and gp41 and is referred to as viral spikes. Spikes are inserted in the viral membrane through the transmembrane sequence of gp41, suggesting that the selfassociation equilibrium of gp41 is dependent on the dynamics of the viral membrane. The native state of gp41, i.e. the prefusogenic form of gp41, is thought to be an oligomer of gp41 in which the N-terminal helix, namely LZ, of three or four molecules self-associates but is prevented from forming a complex with the C-terminal helix perhaps by interaction with native gp120 (33,36). Furthermore, earlier biochemical studies on intact virions also suggest a dimer/tetramer organization of HIV-1 envelope (34,35). In contrast a trimeric three-dimensional organization of gp41 relieved of gp120 interaction and representing gp41 in a fusogenic conformation has been proposed. Indeed, upon the interaction of gp120 with CD4 and the specific coreceptor, in the context of additional molecular interactions between the virus envelope and CD4 ϩ cell membrane which increase the avidity of the virion for the cell (52)(53)(54), it is thought that gp120 undergoes conformational changes that allow the pre-fusogenic form of gp41 to assume the fusogenic coiled coil configuration that brings the viral and cell membrane into juxtaposition, leading to virus cell fusion. Such a trimeric coiled coil structure of LZ and the C-terminal helix has been confirmed by crystallographic studies using gp41 noncontiguous sub-domains (10,11,55,56).
The present results showed an analogy with the results of a study on T20/DP178, a synthetic peptide part of the C terminus of gp41 that shares with P1 residues 650 -678 (57). This study investigates the oligomerization state of T20/DP178 as a function of its concentration (1-100 M) by analytical ultracentrifugation. As does T20/DP178 in the same range of concentration, P1 exhibited a concentration-dependent helicoidal structure, adopting a non-ideal self-association monomerdimer between 25 and 100 M and a monomer-tetramer equilibrium at higher concentrations. The non-ideal self-association equilibrium could be due, as for T20, to the presence of incompetent monomer incapable of self-association and the presence of metastable conformers of the self-associating solute that are not in equilibrium with each other. The FRET (Fig. 4) and peptide binding onto epithelial cell (Fig. 1) experiments confirm that P1 interacts with GalCer in the same P1 concentration range where P1 adopts the self monomer-dimer association equilibrium (Fig. 4). Only the dimeric form of P1 (at concentrations between 25 and 125 M) binds the galactosyl ceramide, in contrast to the monomeric form (at low concentration). For P1 concentrations above 125 M, most of the peptide would be in a higher "incompetent" oligomeric state, most likely a tetramer, as evidenced by the biophysical studies (Fig. 2), and the carbohydrate binding site would not be available for the galactosyl ligand.
T20, a synthetic peptide of 36 amino acids which shares with P1 residues 650 -678 and possesses additional N-terminal residues 643-649 but lacks of the C-terminal residues 679 -685 of P1, exhibits a monomer-tetramer non-ideal self-equilibrium (57) and is a potent inhibitor of the entry of HIV-1 into host cells most likely by interfering with the transition of gp41 into its fusion active state (58). This set of data suggest that the state of the spike on the intact virion that will be transcytosed through the epithelial cell is different from its fusion state after interacting with CD4 ϩ target cells. HIV-1 interaction with and transcytosis across epithelial cells depends on gp120 interaction with GalCer (15,16), and ␣-1 GalCer has been identified as an alternative receptor to CD4 for HIV-1 gp120 envelope glycoprotein (59).
In contrast to the interaction between gp41 and GalCer, the interaction between gp120 and GalCer is not of the lectincarbohydrate type. In this regard, the various sites of gp120 described to interact with GalCer, namely residues 206 -275 (60), a part of V3 (GRAP) (61), and a conformational site of 193 amino acids, including the Val-3, -4, and -5 regions (62), do not contain aromatic amino acids and charged helical structures. Because gp120 is heavily glycosylated, gp120 interaction with GalCer could rather be of the carbohydrate-carbohydrate type as shown for the integrin ␣ 5 ␤ 1 in interaction with the glycosphingolipids (63). One can estimate that in an envelope spike the GalCer binding site of gp120, localized (64) on the outer part of the molecule just above and/or in the V3 domain, is at a distance of Ͻ7 nm from the GalCer binding site on gp41. The area of the lectin site of the spike in interaction with GalCer can be calculated from the distance between the gp120 and gp41 GalCer binding sites, Ͻ7 nm, the number of residues involved per gp41 lectin GalCer binding site arranged in a dimer, ϳ30 residues, and the minimum distance between each dimer in the spike, ϳ10 Å. Thus, the calculated area has a value of ϳ200 Å 2 .
In the apical membrane of the epithelial cell, H ϩ bonds may form from the hydroxyl groups in the galactosyl moieties of the GalCer to the surrounding water molecules, the amide, and the hydroxyl groups of the sphingosine base and the hydroxy fatty acid of neighboring glycosphingolipids and cholesterol. This net of H ϩ bonds participate in the establishment of lipid microdomains in a membrane (65). Such transient lateral assemblies of glycosphingolipids and cholesterol, referred to as raft microdomains (4,66), would function in epithelial cells as platforms for traffic toward the apical membrane in clathrin-coated pits as well as in clathrin-independent endocytosis. The size of these microdomains, with the relative proportion of the components 4 GalCer:1 phospholipid:1 cholesterol (7, 67) has been estimated between 7 and 50 nm, which is in the same range as the distance between the two binding sites of gp120 and gp41 (see "Spacial Component of the Interaction"). Using a value of 30 Å 2 for the cross-section area of the galactosyl moiety of GalCer, a value of 36 Å 2 for the phosphocholin head of PC (taken as the prototype of phospholipid present on the membrane extracellular leaflet), and accounting for the cholesterol planar ring interacting within the fatty acyl chains, the area of a raft approximates 200 Å 2 . Such an area corresponds to the area that can be covered by the two GalCer binding sites in the heterotetramer of gp120-gp41 described above and derived from the experimental data on P1.
The cholesterol could function to fill the "voids" formed beneath large head groups of the sphingolipid between the saturated hydrocarbon chains of the ceramide moieties. Cholesterol also increases the order parameter, mainly at the 7-9 C atom level of the fatty acyl chains (68,69), and stabilizes raft microdomains. This role is in agreement with our previous observation (16) on living cells in which cholesterol depletion from the epithelial cell surface inhibited HIV-1 transcytosis. In the present study we used small-sized liposomes with a defined composition corresponding to that of the "raft microdomain" (Fig. 4). Our data show that in these conditions cholesterol does not influence gp41 peptide binding to GalCer, in agreement with the role of cholesterol in the lipid cell membranes dynamics, to stabilize raft microdomains (by increasing the order parameter).
As it has been mentioned above, water molecules surrounding hydrophobic molecules organize themselves in a network of structured water different from the bulk of dense water, which is characterized by stochastic molecular diffusion (46). In the most stable arrangement of water, a single linear hydrogen bond links two water molecules with a length of ϳ3 Å and a strength of about 21 kJ⅐mol Ϫ1 (44). The molecular size of water estimated from theoretical calculations is such that in a water molecule, a proton acceptor can additionally accept a second proton from another donating neighbor water molecule. The water dimer can simultaneously donate its own two protons to two other neighbor water molecules. This set of hydrogen bonds completes an organized tetrad of linear hydrogen bonds able to naturally build a network of structured water. The potential energy of this tetrad depends on the close groupings of the four water molecules and exhibits an overall increase in binding energy as compared with the isolate water molecule together with a reduction in hydrogen bond length. Furthermore, such a tetrad adopts a spatial organization similar in size and shape to a galactosyl ring (70).
The interaction between HIV-1 envelope and epithelial cells is envisioned to be a cooperative one. The primary contact between gp120 with GalCer at the epithelial cell surface would stabilize a raft microdomain. This is in agreement with the most recent demonstration using model lipids (71) that thermodynamic coupling between lipids and proteins is a mechanism for both lipid and protein clustering on a fluid bilayer, domains that could be precursors of rafts. The stabilization of the raft microdomain would in turn allow the galactosyl ring of GalCer to orient favorably toward gp41, resulting in the interaction between the gp41 lectin site and GalCer. The interfacial structured water molecules contributes to this interaction. Indeed, the interaction of water with sugar solute, favored by the equatorial position of the C-hydroxyls and the ␣-anomer of the carbohydrate, fit the particular hydration layer of the carbohydrate without modification of the hydrogen-bonded solvent structure (70).
Temporal Component of the Interaction-The raft microdomains are transient structures in the membrane. Such microdomains provide evidence of an increased order parameter and an increased relaxation time (rotational and translational diffusion) as compared with the bulk lipids of the membrane bilayer due to interactions with cholesterol and proteins in the bilayer (72). In addition, the lifetime of these transient microdomains is increased when they interact with extracellular ligand here the HIV-1 glycoproteins. Such microdomains are stable in a time scale of 10 Ϫ6 s (72,73).
The relaxation time of structured water in the hydration shell of the gp41 lectin site P1 is characterized by a reduced relaxation time (10 Ϫ12 s) compared with that of bulk water (10 Ϫ9 s). Therefore, the electrochemical signal provided by gp41 and transduced to the epithelial cell membrane through the structured water shell takes place in 10 Ϫ6 s, in agreement with the rate of electron transfer calculated above, i.e. during the lifetime of the lipid microdomain.
Finally, our results may shed light on the characterization of the immunogenic determinant able to elicit HIV-1-neutralizing antibodies against ELDKWA. The antibodies to the ELDKWA sequence (662-667) at the hinge of the C helix and of the hydrophobic terminal segment have been shown to inhibit gp41 binding to epithelial GalCer (16), block HIV-1 transcytosis across a tight epithelial barrier (15), and neutralize infection by a large array of HIV-1 primary isolates both in vitro and in vivo (21)(22)(23). However, as shown here, ELDKWA is not able to bind GalCer in the absence of the two complementary sequences, residues 650 -661 and 667-685. It confirms that these gp41conserved sequences are required to structure ELDKWA and to build the lectin binding site for the GalCer.
Two very recent papers (74,75) confirm our findings on the complex structure of the ELDKWA-containing epitope in the native form of gp41. The authors have extended their previous work on the HIV-neutralizing IgG 2F5 specificity to ELDKWA (amino acids 662-667) and report now that the 2F5 epitope is constituted by residues 658 -671, including in addition to ELDKWA, the N-terminal acid residues 656 -661 and the Cterminal Trp. Furthermore, screening for neutralizing antibodies from an IgG library construct from an HIV-seropositive individual result in the authors concluding the importance of the four C-terminal Trp residues close to the transmembrane domain of gp41. The importance of unraveling the precise structure and function of this region of gp41 to define the best structure of the immunogenic determinant to be presented to the immune system is underlined.
Dendritic cells, one of the first players in the antibody induction mechanism, highly express galactosyl ceramide. 2 Because P1 binds to GalCer, one may imagine that P1 is internalized specifically by dendritic cells to be efficiently presented to the immune system. ELDKWA itself could not follow this route due to its inability to bind GalCer. P1 might be the best structure including ELDKWA to induce 2F5-like HIV-neutralizing antibodies. Along this line, 2F5 IgG exhibit a higher specificity to P1 than ELDKWA. Similarly, IgA from a seropositive patient specific to ELDKWA (16) exhibit a 6 -10-fold higher specific activity for P1 than ELDKWA. 2 Immunization protocols with P1 are in progress in the laboratory.
In summary, we have shown that gp41 residues 650 -685 act as a lectin to bind GalCer, the epithelial cell receptor for HIV-1. The gp41 lectin site encompasses the gp41 C-terminal charged helix, the immunodominant hexapeptide ELDKWA, and the hydrophobic stack of aromatic residues preceding the transmembrane domain of the gp41 molecule. In addition, we show that the oligomerization of gp41 residues 650 -685 controls gp41 lectin activity and, hence, binding to GalCer. The thermodynamic computations give compelling evidence that the specificity of the gp41 lectin interaction with epithelial cell is governed by the dynamics of the partners of the interaction, namely the viral spike and the epithelial cell membrane bathed by the interfacial aqueous medium. The overall spatio-temporal parameters of this lectin-carbohydrate interaction provide evidence of a fine tuning of the space and time domain of each of the interacting partners. Furthermore, the organization of glycosphingolipids in raft microdomains with their characteristic sugar moiety in interaction with the interfacial water is a fundamental mediator of cell pathogen signal recognition. Residues 650 -685 could provide the best structure to be presented to the immune system to elicit HIV-1-neutralizing antibodies.