The Influenza Hemagglutinin Fusion Domain Is an Amphipathic Helical Hairpin That Functions by Inducing Membrane Curvature*

Background: The influenza hemagglutinin fusion peptide is responsible for fusion of viral and endosomal membranes upon infection. Results: Negatively induced curvature and fusion activity were significantly reduced by changing the shape of the fusion peptide. Conclusion: The hemagglutinin fusion peptide structure promotes negative membrane curvature. Significance: This work enables elucidates how a crucial component of influenza infection functions and can be targeted. The highly conserved N-terminal 23 residues of the hemagglutinin glycoprotein, known as the fusion peptide domain (HAfp23), is vital to the membrane fusion and infection mechanism of the influenza virus. HAfp23 has a helical hairpin structure consisting of two tightly packed amphiphilic helices that rest on the membrane surface. We demonstrate that HAfp23 is a new class of amphipathic helix that functions by leveraging the negative curvature induced by two tightly packed helices on membranes. The helical hairpin structure has an inverted wedge shape characteristic of negative curvature lipids, with a bulky hydrophobic region and a relatively small hydrophilic head region. The F3G mutation reduces this inverted wedge shape by reducing the volume of its hydrophobic base. We show that despite maintaining identical backbone structures and dynamics as the wild type HAfp23, the F3G mutant has an attenuated fusion activity that is correlated to its reduced ability to induce negative membrane curvature. The inverted wedge shape of HAfp23 is likely to play a crucial role in the initial stages of membrane fusion by stabilizing negative curvature in the fusion stalk.

The influenza virus is a leading cause of mortality by infectious diseases worldwide. The emergence of highly virulent strains, such as the highly pathogenic influenza A (H5N1), have reported mortality rates above 50% (1), and the understanding and treatment of influenza infection remain a high priority health concern. The influenza virus has an RNA genome and a capsid coated by a phospholipid membrane (2). The virus infects host cells by a two-stage process that involves cellular endocytosis into the endosome followed by the transfer of viral contents accomplished by the fusion between viral and endosomal membranes (3). Both stages of viral entry are dependent on the hemagglutinin surface glycoprotein (HA) of the virus, consisting of a homotrimer of HA1 and HA2 subunits generated by the proteolytic cleavage and activation of the HA0 precursor (4 -6). The HA1 subunit is responsible for the initial internalization by binding to the sialic acid receptor before transfer of the virus to the endosome (7)(8)(9). Acidification of the late endosome triggers a conformational change in HA, exposing the highly conserved and hydrophobic 23-residue fusion peptide domain (HAfp23) from the HA2 subunit. HAfp23 attaches to the endosomal membrane and catalyzes the membrane fusion process. It is strictly required for membrane fusion such that conservative mutations or truncations in this sequence abolish infectivity (10). Lear and DeGrado (11) first demonstrated that a short peptide with a minimum sequence of 20 residues (HAfp20) from the fusion domain exhibits lipid-mixing fusion activity. The full-length HAfp23 has a higher fusion activity (12), and it extends an additional 3 completely conserved, C-terminal hydrophobic residues (10,13,14). HAfp23 adopts a tight helical hairpin structure with well defined amphipathic ␣-helices from residues 1-12 and 14 -23 that reside at the lipid-water interface of the membrane (15). The amphipathic structure is stabilized by four C ␣ H ␣ -O hydrogen bonds and a charge-dipole interaction at the N terminus (15,16). The two helices lie approximately in the plane of the membrane with a hydrophobic face oriented toward the membrane and a hydrophilic face exposed to the aqueous solvent (15). By contrast, the truncated HAfp20 predominantly adopts an open "boomerang" structure (17,18). Further insight has shown that the truncated HAfp20 is highly dynamic and populates a partially unfolded "open" structure as well as an 11% population of a "closed" structure that resembles the helical hairpin structure of the full-length HAfp23 (19), which could account for this peptide's fusion activity.
Despite the numerous studies that have elucidated the structural biology of the full-length HAfp23, the function and mechanism of action of this domain remain unclear. Several mechanisms have been proposed, including the disruption of membranes (18,20,21), the promotion of lipid tail protrusions (22), the control of line tension within the membrane (23), the aggregation of fusion proteins and dehydration of membranes (24,25), the induction of positive membrane curvature (26), and the induction of negative membrane curvature (27)(28)(29). The remodeling of two planar membranes requires high curvature membrane intermediates. Several results support a "fusion stalk" mechanism in which the two membranes must first join in a stalk between contacting monolayers before proceeding to a hemifusion diaphragm and fusion pore structure (30 -32). The fusion stalk is a high energy intermediate with an hourglass shape and a high degree of concave (negative) curvature along one axis and convex (positive) curvature along the other. In support of this model, studies on foreign lipids have demonstrated that the inverted wedge shape of the dioleoylphosphatidylethanolamine (DOPE) 2 lipid stabilizes negative curvature and promotes membrane fusion when placed in the contacting monolayers between two fusing membranes (33,34). Conversely, the wedge-shaped lysophosphatidylcholine lipid inhibits fusion and destabilizes negative curvature (35,36). Detailed crystallographic and differential scanning calorimetry (DSC) experiments have further demonstrated that HAfp20 promotes the negative curvature of hexagonal inverted (H II ) and cubic phases of lipids (28,37,38), which is consistent with the fusion domain's stabilization of the fusion stalk structure. However, these mechanistic studies have largely focused on the truncated HAfp20 domain, and a clear relationship between the structure and function of the full-length domain has remained elusive.
Like the DOPE lipid, the amphipathic structure of the fulllength HAfp23 has an inverted wedge shape with a larger hydrophobic base in proportion to the hydrophilic head region. In this report, we demonstrate how changes in this domain's shape and structure are correlated to its lipid-mixing fusion activity and ability to stabilize negative curvature in membranes. We have investigated and compared the structure and activities of the wild type HAfp23 with the fusion-deficient HAfp23-F3G mutation (40). The HAfp23-F3G mutant maintains the backbone helical hairpin structure and affinity for membranes of the wild type domain, yet its inverted wedge shape is significantly reduced by the replacement of a phenylalanine residue for a glycine at the hydrophobic base of the structure. Despite the structural similarities in the backbone of both constructs, morphological differences resulting from the reduction in the hydrophobic volume at the base of the structure result in lowered lipid-mixing fusion activity consistent with its reduced activity in stabilizing negative membrane curvature. These results demonstrate that the HAfp23 domain is an amphipathic helical hairpin structure that functions by inducing negative membrane curvature.

EXPERIMENTAL PROCEDURES
Protein Expression-Protein purification closely followed the previously published protocol for the HAfp23 influenza fusion domain (15). The wild type HAfp23 and mutant HAfp23-F3G fusion peptides (H1 subtype) were expressed in Escherichia coli BL21(DE3) as a fusion protein with the IgG-binding domain B1 of streptococcal protein G (GB1; Protein Data Bank code 3GB1) located at the N terminus and a solubility tag, GSKKKKD, at the C terminus to assist in the purification. The 15 N isotope labeling was achieved by supplementing M9 minimal medium with 15 NH 4 Cl (41), and natural abundance proteins were prepared by growing the E. coli in LB medium. Cells were lysed by sonication, and proteins were purified from lysate by histidine affinity chromatography using a HisTrap HP column (GE Healthcare) on Aktä PrimePlus and Aktä Purifier FPLC systems. Further purification was achieved by size exclusion chromatography through a Superdex 75 26/600 PG (GE Healthcare) in the presence of 25 mM Tris, pH 7.4. The GB1 portion was removed by Factor Xa proteolytic cleavage, separated by reverse phase chromatography on a Resource RPC 3-ml column (GE Healthcare), and eluted using an acetonitrile gradient in 0.1% TFA. The final peptide HAfp23 and HAfp23-F3G sequences were GLFGAIAGFIEGGWTGMIDGWYGGSKKKKD (3148 Da) and GLGGAIAGFIEGGWTGMIDGWYGGSKKKKD (3057 Da), respectively. SDS-PAGE analysis, HPLC/MS experiments, size exclusion chromatography, and NMR spectra confirmed the molecular weight and homogeneous purity of the protein samples.
NMR Sample Preparation-The 15 N-labeled HAfp23 and HAfp23-F3G samples were prepared at a concentration of 0.3-0.9 mM in a 25 mM Tris, pH 7.4, buffer with 5-7% 2 H 2 O and 140 -160 mM dodecylphosphocholine (DPC; Anatrace). Samples with an acidic pH were prepared by titrating with 25 mM citric acid to a final pH of 4.2. Protein concentrations were quantitated from the integrals of a 1 H one-dimensional NMR spectrum.
Anisotropic alignment of protein samples for residual dipolar coupling (RDC) measurements was achieved with 5.0% stretched acrylamide gel (Bio-Rad) supplemented with 1.43% 2acrylamido-2-methylpropane sulfonic acid (Sigma). A 5.4-mm acrylamide gel was stretched to a diameter of 4.0 mm. Anisotropic alignment was confirmed from the residual quadrupolar coupling (RQC) of 2 H 2 O, which was 1.92 Hz at pH 7.4 and 1.87 Hz at pH 4.2.
NMR Experiments and Data Analysis-NMR experiments were collected on 500-, 600-, and 800-MHz Bruker Avance I and Avance III HD NMR spectrometers. The 500-and 800-MHz NMR spectrometers were equipped with a 1 H/ 13 C/ 15 N/ 31 P QXI and 1 H/ 13 C/ 15 N TCI room temperature probes, respectively, with three-axis gradients. The 600-MHz spectrometer was equipped with a cryogenically cooled 1 H/ 13 C/ 15 N TCI probe with single-axis gradient. Spectra were recorded at a temperature of 32.0°C.
Chemical shifts were measured and assigned using a combination of a 15 N HSQC experiment with the Rance-Kay detection scheme and a NOESY-HMQC three-dimensional experiment. R 1 , R 1 , and 15 N-{ 1 H} NOE experiments were used to measure the backbone dynamics of the peptides in DPC at 500 MHz (42,43). The 15 N R 1 rates were measured with seven delay points between 5 ms and 1.5 s. Cross-correlated relaxation was suppressed with a WALTZ decoupling scheme during 15  measured with a 1500-Hz spin-lock field using delays of 0.005-1.5 ms. Cross-correlated relaxation was suppressed using 1 H 180°decoupling pulses with a 21.8-kHz field strength every 2 ms during the spin-lock period. The R 2 rates were calculated from the measured R 1 rates using the effective field and spin-lock tilt for each 15 N resonance and correction for the contribution from the R 1 rate (44). The R 1 and R 1 rates were calculated from fitting the intensity decay profiles to a monoexponential using Sparky (45). A 4-s relaxation delay was used between experiments. The 15 N-{ 1 H} NOE rates were measured as described previously (43), using a relaxation delay of 9 s between experiments. Relaxation rates were fit using MODELFREE and the model-free spectral density formalism (46,47). The 1 H-15 N J-couplings and RDCs were measured using an IPAP-WATERGATE sequence (48), and data were analyzed using NMRPipe and Sparky (45,49). The structure of the wild type HAfp23 (15) peptide was used as a reference for the fitting of RDCs using a singular value decomposition analysis (50). The reported Q-factors were calculated using NMRPipe (49,51).
Hydrogen exchange rates were measured using a WEX-III TROSY experiment (52). Mixing times allowing for exchange of water protons with amide protons were varied using five points from 10 to 320 ms. A reference spectrum was also collected under identical conditions with the exclusion of a mixing period. Longitudinal relaxation of water protons was measured using a saturation-recovery 1 H T 1 experiment.
It has previously been demonstrated that the time scale of exchange between the "open" and "closed" state of HAfp23 is ϳ28 Ϯ 1 s, and therefore, a chemical shift frequency representative of the population-weighted average is evolved during the time scale of the NMR acquisition. Chemical shift measurements of a truncated HAfp14 indicate that it maintains its secondary ␣-helical chemical shift structure and can therefore be used to represent the fully open structure for the first 14 residues (53). Plotting the HAfp23-F3G chemical shift differences from those of HAfp23, the fully closed state at high pH, against the differences between the open and closed state can therefore yield the population of the closed state, p closed , as follows (53), where ␦* is the scaled chemical shift for a given nucleus. Because the range of chemical shifts varies between nuclei, normalization scaling factors of 1.0 and 0.181 for 1 H and 15 N, respectively, were used, where the scaling factors are proportional to the inverse width of their distributions in the protein chemical shift database (53).
Translational Diffusion and Bicelle-induced Curvature and Sorting (BICS) Measurements-Translational diffusion rates for HAfp23 and HAfp23-F3G in the presence of dihexanoylphosphatidylcholine/dimyristoylphosphatidylcholine (DHPC/ DMPC) bicelles were measured by solution NMR. Temperature and viscosity were corrected to 32°C and a composition of 10% 2 H 2 O and 90% H 2 O. Samples were prepared for NMR consisting of 77 mM DHPC, 24 mM DMPC, 25 mM MOPS, and a 430 M concentration of either 15 N-HAfp23 or 15 N-HAfp23-F3G at pH 7.4 as well as an empty bicelle control for use in BICS anal-ysis, discussed below (54). Experiments were carried out at 500 MHz using a 1 H longitudinal eddy-current delay and bipolar pulse pair (LED-BPP) diffusion experiment with a WATER-GATE readout (55). Translational diffusion constants, D t , were measured from a Gaussian fit of the intensity decay profiles, I(f), as a function of the fractional gradient strength, f.
The gyromagnetic ratio, ␥, of 2.675 ϫ 10 8 s Ϫ1 tesla Ϫ1 for 1 H was used. Diffusion gradients were varied from 0 to 60% for the xand z-gradients together, producing a combined G max of 73.4 G/cm. A total of 16 gradient points were measured. A gradient pulse length, ␦, of 2 ms, and delays ⌬ and of 500 ms and 200 s, respectively, were used for the bicelles/peptide samples. The diffusion of DHPC and DMPC was followed by their resolved terminal 1 H 3 C resonances, and the rate of diffusion for the peptide and bicelles was assumed to match that of DMPC. The free and unbound concentration of peptide was determined using the following equation.
The rate of translational diffusion for the free peptide was estimated from the molecular mass of the peptide (3147 and 3057 kDa for HAfp23 and HAfp23-F3G, respectively) and calculated from the diffusion rate and molecular mass of lysozyme (14.3 kDa, 11.54 ϫ 10 Ϫ11 m 2 /s at 32°C).
The BICS analysis was used as an independent measure of the curvature induction by HAfp23 and HAfp23-F3G peptides (54). This technique measures the ratio in molecular weights, denoted the B T ratio, for bicelles with an average of 0.2 molecules of peptide per bicelle and reference bicelles without peptide. We have previously found that this technique can be used to measure both positive and negative curvature and to estimate ensemble-averaged curvature parameters. Translational diffusion rates were measured as part of a temperature cycling schedule from 15 to 40°C over 8 h. B T ratios are reported at 35°C, near the physiologically active temperature of the influenza virus. The temperature is about 10°C above the phase transition of DMPC in the central region of the bicelles. Diffusion rates were measured in quadruplicate, and it was found that at least one temperature cycle was needed to achieve consistent diffusion rates. The molecular weight contribution of the peptide has been corrected in the final B T ratio.
Lipid-mixing Fluorescence Fusion Assay-1,2-Dioleoylphosphocholine (DOPC, Avanti Polar Lipids) and DOPE (Avanti Polar Lipids) were mixed in a 1:1 molar ratio in chloroform, dried into lipid films overnight at 37°C, and lyophilized for 1 day to remove residual chloroform. The lipid films were resuspended in 20 mM acetate buffer, pH 4.9, and given 30 -60 min to rehydrate on ice. The multilamellar vesicles were then extruded 20 times through a polycarbonate membrane with a 0.1-m diameter pore size (Whatman) to generate large unila-mellar vesicles (LUVs). Fluorescent donor and acceptor lipid pairs, N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)-1,2-dihexadecanoyl-sn-glycero-3-phosphoethanolamine, triethylammonium salt (Invitrogen) and rhodamine B 1,2-dihexadecanoylsn-glycero-3-phosphoethanolamine, triethylammonium salt (Invitrogen) were each added to a final molar concentration of 0.6% to the LUVs. Fluorescently labeled and unlabeled LUV mixtures were combined in a 1:4 ratio to a final lipid concentration of 3.8 mM in 2.0 ml. The mixture was equilibrated at 37°C with magnetic mixing for 10 min before the addition of peptide. Peptides were added from a solution in DMSO to a final concentration of 15 or 45 M. Lipid mixing was monitored through donor excitation at 465 nm and a decrease in acceptor fluorescence at 590 nm. Triton X-100 was added to a final concentration of 2.5 mM to complete lipid mixing.
Differential Scanning Calorimetry-DSC experiments used 1,2-dipalmitoleoyl-sn-glycero-3-phosphocholine vesicles (DPoPE; Avanti Polar Lipids) to study the phase transition from lamellar to hexagonal inverted phase (29,56). DPoPE lipids were used without further purification. DPoPE lipids were dissolved in a 1:2 mixture of chloroform to methanol and dried to a lipid film under a stream of nitrogen gas. Lipid films were further dried by overnight lyophilization. Lipid films were hydrated by resuspending in a 10 mM acetate buffer, pH 5.0, with 150 mM NaCl, 1 mM EDTA, and vortexed for 5 min. Hydrated lipid films were used within 24 h of preparation (38). Peptides lyophilized from acetonitrile reverse-phase chromatography elution were added directly to the samples in a peptide/lipid molar ratio of either 1:1600 or 3:1600. Differential scanning calorimetric measurements were conducted using a MicroCal VP-DSC instrument with a deionized water reference. Data were collected with a pre-equilibration time of 60 -160 min at the starting temperature. Calorimetric scans were conducted from 10 to 70°C at a rate of 37°C/h. Baseline correction was achieved by fitting the thermograms with a cubic function and fast Fourier transform smoothing over 25 points. All data analysis was performed using Origin software.

Structure of the Mutant HAfp23-F3G-The wild type
HAfp23 influenza fusion domain has a tight and stable helical hairpin structure in DPC micelles and in bicelles composed of DHPC/DMPC (15,57). The structure is stabilized by four ali-phatic C ␣H␣-O hydrogen bonds and a charge-dipole interaction between the strictly conserved Gly-1 residue and the C-terminal helix (16). The backbone amide chemical shifts are very sensitive to changes in structure, including small populations of open structures (53). Between pH 4 and 7, the wild type HAfp23 15 N HSQC spectrum remains relatively unchanged except for the peaks associated with residues Glu-11 and Asp-19, which become protonated at the lower pH (16).
The F3G, I6G, F9G, W14A, and W21A mutants change the hydrophobicity of the fusion domain and have previously been demonstrated to have reduced fusion activities (10,21,40). We have measured the 15 N HSQC spectra for these mutants, and the I6G, F9G, W14A, and W21A mutants show large changes in chemical shifts, consistent with a large structural change and partial unfolding in the peptide structure. This change in structure is consistent with the loss in fusion activity produced by disrupting the helical hairpin structure, as observed previously for the truncated HAfp20 and HAfp23-G8A mutants (15,19). By contrast, the F3G mutant has 1 H N -15 N chemical shifts that are nearly identical to those of the wild type HAfp23, indicating that these two peptides share the same backbone structure in DPC micelles. We therefore selected the HAfp23-F3G mutant for this study because it maintains the helical hairpin structure of the wild type peptide, yet this mutation produces a perturbation that cleanly isolates the contribution of hydrophobic residues to the function of the fusion domain's structure.
The 15 N HSQC spectrum of the mutant HAfp23-F3G at pH 7.4 and 4.2 is largely superimposable with that of the wild type HAfp23 (Fig. 1). The site of mutation, F3G, shows significant changes in chemical shift, consistent with the difference in bonding topology between these two types of residues. Likewise, an observable change in chemical shift is measured for residues Glu-11 and Asp-19 of HAfp23-F3G, indicating a change in protonation state for these residues between pH 7 and 4. Both HAfp23 and HAfp23-F3G have additional sharp resonances corresponding to the polylysine C-terminal solubilization tail, which remain highly dynamic and disordered according to chemical shift and hydrogen exchange experiments.
We used RDCs to further compare the backbone structure of HAfp23 and HAfp23-F3G. RDCs are a high resolution measure of the orientation of bonds in a molecular structure (58). The  Fig. 2 demonstrate the nearly identical backbone structure between wild type HAfp23 and the mutant HAfp23-F3G at both pH 7.4 and 4.2. The high resolution structure of the HAfp23 (Protein Data Bank code 2KXA) was refined with 138 RDCs using two alignment media and 449 NOEs (15). Like the 1 H 15 N RDCs of the wild type HAfp23, the RDCs for the mutant HAfp23-F3G have an oscillatory pattern that shows the two distinct and well defined ␣-helices between residues 3-12 and 14 -23 in the helical hairpin structure. A singular value decomposition analysis of these RDCs to the HAfp23 structure displays a high degree of agreement. The corresponding quality factors (51, 58) of the fit are 11.1% at pH 7.4 and 9.3% at pH 4.2 (Fig. 2). These results indicate that the mutant HAfp23-F3G maintains the same high resolution helical hairpin backbone structure of the wild type HAfp23 in DPC micelles.

H 15 N RDCs presented in
Rigidity of the Mutant HAfp23-F3G Structure-The mutant HAfp23-F3G backbone 15 N R 1 , R 2 , and 15 N-{ 1 H} NOE rates at 500 MHz are plotted in Fig. 3 at pH 7.4 and 4.2. These rates are sensitive to motions on the time scale of the overall rotational diffusion for the peptide-micelle complex as well as to the internal motions of the 1 H-15 N bonds faster than this time scale. The rates for the N-terminal residue Gly-1 and Leu-2 could not be measured due to an elevated rate of exchange with water. The spectroscopic absence of these residues was also observed for the wild type HAfp23.
The 15 N rates are highly uniform from residue 3 to 23, and the backbone structure is well folded at both high and low pH. The amide 15 N relaxation rates were analyzed using the Lipari-Szabo formalism (46). The rotational correlation times of the peptide-micelle complex are 9.2 and 9.4 ns at pH 7.4 and 4.2, respectively. These are comparable with the rotational correlation times of the wild type HAfp23 in DPC, which are 8.4 and 9.1 ns at pH 7.4 and 4.0, respectively (15). The backbone order parameters show a high degree of order in the backbone with an average of 0.84 Ϯ 0.02 (mean Ϯ S.D.) for residues 3-22. Residue Gly-23 at the C terminus shows a small increase in fast time scale dynamics consistent with partial fraying in the last turn of the helix. This modest increase in dynamics at residue Gly-23 was also observed in the wild type HAfp23.
Slightly elevated 15 N R 2 relaxation rates are observed for a small number of residues at pH 4.2. These were likewise observed in the wild type HAfp23. At high pH, HAfp23 folds in the helical hairpin structure, yet at low pH, the wild type HAfp23 partially opens on a microsecond time scale (53). The opening motion occurs on a fast time scale relative to the evolution of the chemical shifts, and population-weighted chemical shifts and NOEs revealed that the wild type HAfp23 populates the open conformer to 11% at pH 4.2 (53). Chemical shifts in fast exchange are also observed for HAfp23-F3G, and the corresponding open conformer populations are 6 Ϯ 11% at pH 7.4 and 15 Ϯ 6% at pH 4.2 (Fig. 4), which are in close agreement with the wild type HAfp23.
Membrane Interaction and Affinity-Rates of exchange for the backbone amide protons with water were measured for the  mutant HAfp23-F3G at pH 7.4 (52). These were compared with the predicted intrinsic exchange rates of a disordered peptide and plotted as a protection factor in Fig. 5 (59).
Protection factors follow an ␣-helical arrangement. Hydrophobic residues in contact with the micelle have high protection factors, and hydrophilic residues oriented to the aqueous solvent have reduced protection factors. The mutant HAfp23-F3G maintains the amphipathic profile of the wild type HAfp23 (15).
The difference in functional activity between HAfp23 and HAfp23-F3G could, in principle, be attributable to a difference in affinity of the two peptides to membranes because the mutant peptide has lost a bulky hydrophobic residue (Phe-3). To test this possibility, we measured the affinity of both peptides to DHPC/DMPC bicelles using diffusion ordered spectroscopy experiments (Fig. 6). The wild type HAfp23 and mutant HAfp23-F3G are 81 Ϯ 11 and 88 Ϯ 12% bound to the DHPC/DMPC bicelles, respectively, and no substantial differences are observed in the affinity of both peptides to mem-branes. The 15-20% unbound population is not unexpected because both peptides were prepared with a polyionic C-terminal tail (K 4 D) to aid in the purification, which is expected to contribute to the partial solubilization of the peptides in water.
Lipid-mixing Fusion Activity-Heteropolykaryon assays first demonstrated that the F3G mutant in the fusion domain of the hemagglutinin glycoprotein was deficient in fusion activity (40). We used an in vitro lipid mixing fusion assay to quantitate activity differences between wild type HAfp23 and the mutant HAfp23-F3G. This assay measures the redistribution of fluorescently tagged lipids by fluorescence resonance energy transfer (FRET) (60). Promotion of lipid-mixing membrane fusion drives the redistribution of lipids from LUVs containing both donor-and acceptor-labeled lipids to unlabeled LUVs, thereby decreasing the acceptor fluorescence. Lipid mixing fusion assays for HAfp23 and HAfp23-F3G are shown in Fig. 7.
The fusion assay was conducted at pH 4.9 near the functional maximum of the hemagglutinin fusion protein (61). The wild type HAfp23 is significantly more fusiogenic than the mutant HAfp23-F3G. The mutant HAfp23-F3G has a diminished initial rate of fusion, and the extent of fusion is reduced by ϳ70% for an equimolar amount of peptide.
Curvature-inducing Properties-DSC experiments on aqueous vesicles of DPoPE lipid were used to measure the propensities for stabilizing membrane curvature of the wild type and mutant fusion domains. DPoPE vesicles have a phase transition from the lamellar liquid crystalline (L ␣ ) state to the H II state near ϳ40 -42°C (62,63). This transition involves a rearrangement of lipids from a planar arrangement to inverted tubules with an aqueous core and negative membrane curvature perpendicular to the axis of the tubule. The phase transition of DPoPE is conveniently near physiological temperatures, and the fusion domains remain folded at this temperature. Protein or lipid additives that have an inverted wedge shape stabilize negative membrane curvature and reduce the phase transition temperature of DPoPE to the H II state (64). Conversely, lipids or proteins that have a wedge shape stabilize positive membrane curvature and increase the H II phase transition temperature. The magnitude of the change in transition temperature depends on the molar ratio of the protein or lipid additive to DPoPE (65).  N-HAfp23-F3G and a "closed" HAfp23 at pH 7.4 (A) and pH 4.2 (B). The chemical shift differences of an "open" HAfp14 and closed HAfp23 yield a slope of 94 Ϯ 11 and 85 Ϯ 6% for HAfp23-F3G at pH 7.4 and 4.2, respectively, corresponding to the population of the open conformer for this peptide at these two pH values. Residues Gly-1 and Leu-2 are missing in the HSQC experiments due to elevated hydrogen exchange rates. Gly-3 and Trp-14 were omitted because these are at or near the site of mutation for HAfp23-F3G and HAfp14, respectively. Glu-11 was omitted from the pH 4.2 data set because this resonance changes chemical shifts predominantly due to a change in protonation state. FIGURE 5. The HAfp23-F3G backbone amide proton water exchange protection factors for HAfp23-F3G in DPC micelles at pH 7.4. Residue-specific exchange rates with water were measured using a WEX-III pulse sequence (52) and compared with the predicted intrinsic exchange rates for an unstructured peptide using SPHERE (87). DSC thermograms of DPoPE with HAfp23 and HAfp23-F3G at the fusion-active pH 5.2 are presented in Fig. 8. Control DPoPE vesicles have a transition temperature of 38.9°C. The addition of 1:1600 and 3:1600 molar ratio of HAfp23 to DPoPE produces a change in phase transition temperature, ⌬T, of Ϫ5.4 and Ϫ8.2°C, respectively. A second transition at 20.6°C is observed at the higher HAfp23 concentration, which possibly correlates with the introduction of a cubic membrane phase (4). The reduction in phase transition temperature indicates that HAfp23 stabilizes the negative curvature H II state. The DSC thermogram of the mutant HAfp23-F3G at the higher molar concentration of 3:1600 demonstrates only a modest reduction in the phase transition temperature, with a ⌬T of Ϫ2.2°C. The reduced propensity for the mutant HAfp23-F3G to stabilize negative membrane curvature is consistent with its reduced lipid mixing fusion activity (10).
We have additionally used BICS experiments to measure the curvature induced by the wild type HAfp23 and mutant HAfp23-F3G on bicelles (54). This technique measures changes in bicelle sizes from the curvature induced by foreign lipids and proteins. Changes are quantified with a B T ratio from NMR DOSY translational diffusion experiments, which measures the relative increase or decrease in average bicelle molecular weight induced by lipids or proteins. For bicelles with an average of 0.2 molecules of wild type HAfp23 or mutant HAfp23-F3G, we have found that the bicelles become larger by 66% (B T ϭ 1.659 Ϯ 0.050) and 8% (B T ϭ 1.078 Ϯ 0.020) at 35°C, respectively. The corresponding estimates for the ensemble-averaged radii of curvature are Ϫ26 Ϯ 2 Å for HAfp23 and Ϫ187 Ϯ 45 Å for HAfp23-F3G, which indicates that the wild type HAfp23 induces significantly more membrane curvature than the mutant HAfp23-F3G.

DISCUSSION
NMR structural and dynamic experiments show that the wild type HAfp23 and mutant HAfp23-F3G both have the same helical hairpin structure, mode of interaction with membranes, and binding affinity to membranes. Backbone relaxation experiments confirm that this structure is stable at high and low pH. Despite these shared features, HAfp23-F3G exhibits significantly reduced fusion activity. Previous reports have shown that this mutation ablates fusion activity in the formation of polykaryon cells by the full-length hemagglutinin (40). We have likewise demonstrated that this mutation is compromised in its lipid mixing fusion activity.
Together with a reduced functional activity, the HAfp23-F3G exhibits a reduced propensity for stabilizing negative membrane curvature. The fusion of two membranes requires the remodeling of lipids between fusing bilayers, introducing high energy bending deformations in the membranes. Experimental and theoretical models support the fusion stalk mecha-  nism of membrane fusion with a hemifusion stage preceding formation of a pore. In hemifusion, lipids in the outer, contacting monolayers redistribute between fusing membranes through a "fusion stalk" in which the two bilayers must form an hourglass-shaped stalk with catenoidal topology before proceeding to a transmembrane-contacting structure and finally a fusion pore (32,66,67). The stalk has a high degree of negative curvature along one axis and positive curvature along the other, and its structure is comparable with the bicontinuous cubic phases of lipids (28). In support of this model, experiments incorporating negative curvature lipids with an inverted wedge shape, such as DOPE and arachidonic acid, promote the rate of membrane fusion when placed in the contacting monolayers of fusing membranes by stabilizing the high energy negative curvature of the stalk (35, 68 -71). Likewise, positive curvature lipids with a wedge shape, such as the micelle-forming lysophosphatidylcholines, reduce the rate of membrane fusion when placed in the contacting monolayers by destabilizing the stalk. These same lipids exhibit the opposite effect, by inhibiting fusion in the former case and promoting fusion in the latter, when placed in the distal monolayers of two fusing membranes, thereby promoting the positive curvature of the pore (35,72).
Fusion domains analogously demonstrate a promotion in the rate of membrane fusion by stabilizing negative curvature on the outer membranes of vesicles. At low peptide/lipid ratios (ϳ0.1-0.3%), the truncated HAfp20 as well as the fusion domains of the feline leukemia virus SIV decrease the thermal transition temperature from the DPoPE L ␣ phase to stabilize the H II phase with negative curvature (37,73,74). Mutant fusion domains that exhibit reduced membrane fusion activity are compromised in their ability to lower the H II phase transition temperature and show a reduced propensity for negative membrane curvature structures (74,75). At higher molar ratios (ϳ1-3%), fusion domains further introduce a bicontinuous cubic phase with a structure that resembles the stalk (28,38,76). Our DSC experiments support these results and show that the wild type HAfp23 stabilizes the H II phase of DPoPE. By contrast, the mutant HAfp23-F3G demonstrates a reduced stabilization of negative curvature of the H II phase that correlates with its reduced lipid mixing fusion activity. Although DSC remains a conventional method to measure free energy transfer of phase transition, a secondary method of BICS was applied to isotropically tumbling bicelles. Results clearly indicate that HAfp23 induces significant negative curvature, whereas the mutant HAfp23-F3G only induced modest curvature.
The change in the functional behavior of the mutant HAfp23-F3G agrees with the structural features of the HAfp23 helical hairpin. From a lateral view, the wild type HAfp23 helical hairpin structure has a significant inverted wedge shape along one axis and a more cylindrical shape along the other (Fig.  9). The inverted wedge shape of the amphipathic helical hairpin HAfp23 is analogous to that of negative curvature lipid. The hydrophobic residues on the lipid face comprise 70% of the protein's volume (2514 Å 3 ), whereas the hydrophilic residues on the aqueous side only comprise 30% (1077 Å 3 ). Integration of the inverted wedge-shaped helical hairpin at the membrane surface induces negative membrane curvature. Mutation of Phe-3 to a Gly residue reduces the base volume to 63% and changes the inverted wedge shape of the structure to a more cylindrical shape.
The impact of molecular shape is well documented for lipids and amphipathic helices (77)(78)(79). Class A amphipathic helices have a wedge shape and induce positive membrane curvature, whereas class L amphipathic helices have an inverted wedge shape and induce negative membrane curvature (78). For example, the ENTH domain of epsin and the N-terminal region (known as H0) of N-BAR domains are unstructured segments that fold into amphipathic ␣-helices when inserted in the membrane headgroup region and induce positive membrane curvature (80,81). These amphipathic helices have a wedge shape with a larger cross-sectional area in the hydrophilic face in relation to the hydrophobic groups (82). The synthetic 18L peptide is a class L amphipathic helix, and, like the helical hairpin of HAfp23, its hydrophobic base has a larger cross-sectional area than its hydrophilic face (64). The 18L peptide stabilizes negative membrane curvature to a lesser extent than HAfp23, and it promotes membrane fusion (64). Furthermore, a reduction in the volume in the hydrophobic base with the W10G mutation reduces this peptide's propensity for stabilizing negative curvature and fusing membranes. Likewise, pardaxin and the islet amyloid polypeptide are short sequences that induce negative curvature (83,84). The HAfp23 structure is distinct from these amphipathic helices because it utilizes a tight helical hairpin structure to combine the inverted wedge shape of two amphipathic helices to induce curvature. Structures are based on the helical hairpin structure of the wild type HAfp23 (Protein Data Bank code 2KXA), and the mutant Phe-3 has been replaced by a Gly residue (15). Structures are shown in space-filling representations with hydrophobic (yellow), hydrophilic (green), acidic (red), and basic ␣-amino group of Gly-1 (blue). The remaining glycine residues are colored in gray. Hydrogen exchange experiments show that the hydrophobic phase is oriented to the membrane core, and the hydrophilic face is oriented to the aqueous solvent at the surface of membranes.
The shape and stabilization of membrane curvature are critical features in the fusion domain's function in promoting early fusion intermediates, yet this domain's function may not be limited to the initial stages of membrane fusion. Other reports have indicated that viral fusion domains may participate in the aggregation of hemagglutinin proteins, in the dehydration of membranes, and in formation of the final pore (19,85). In summary, our results support the functional mechanism for the hemagglutinin fusion domain that relies on its amphipathic inverted wedge shape to stabilize negative membrane curvature and promote the initial stages of membrane fusion. The HAfp23-F3G mutant has the same backbone helical hairpin structure and membrane interaction as the wild type HAfp23, yet it has significantly reduced activities to fuse and induce negative curvature in membranes. The functional role of stabilizing early fusion intermediates is an important step in influenza infection, and these results could have further implications for the infection processes of other viruses that are believed to use a similar mechanism, including HIV and Ebola.