The complete influenza hemagglutinin fusion domain adopts a tight helical hairpin arrangement at the lipid:water interface

Abstract

All but five of the N-terminal 23 residues of the HA2 domain of the influenza virus glycoprotein hemagglutinin (HA) are strictly conserved across all 16 serotypes of HA genes. The structure and function of this HA2 fusion peptide (HAfp) continues to be the focus of extensive biophysical, computational, and functional analysis, but most of these analyses are of peptides that do not include the strictly conserved residues Trp²¹-Tyr²²-Gly²³. The heteronuclear triple resonance NMR study reported here of full length HAfp of sero subtype H1, solubilized in dodecylphosphatidyl choline, reveals a remarkably tight helical hairpin structure, with its N-terminal α-helix (Gly¹-Gly¹²) packed tightly against its second α-helix (Trp¹⁴-Gly²³), with six of the seven conserved Gly residues at the interhelical interface. The seventh conserved Gly residue in position 13 adopts a positive ϕ angle, enabling the hairpin turn that links the two helices. The structure is stabilized by multiple interhelical C^αH to C = O hydrogen bonds, characterized by strong interhelical H^N-H^α and H^α-H^α NOE contacts. Many of the previously identified mutations that make HA2 nonfusogenic are also incompatible with the tight antiparallel hairpin arrangement of the HAfp helices.¹⁵N relaxation analysis indicates the structure to be highly ordered on the nanosecond time scale, and NOE analysis indicates HAfp is located at the water-lipid interface, with its hydrophobic surface facing the lipid environment, and the Gly-rich side of the helix-helix interface exposed to solvent.

As exemplified by the 2009 outbreak of the swine flu pandemic, influenza remains a high priority health concern. Of the three different types of RNA influenza virus, the influenza A type is the most common and virulent human pathogen. Based on the antibody response to its viral surface proteins, hemagglutinin and neuraminidase, influenza type A is further subdivided into serotypes, ranging from subtypes H1-16 for hemagglutinin, and N1-9 for neuraminidase. Subtypes H1, H2, H3, N1, and N2 are the most common in human disease (1). Whereas neuraminidase is an enzyme responsible for cleavage of cellular sialic acid carbohydrates, a key step in viral exit, HA mediates cellular entry through binding to carbohydrates on epithelial cells and subsequent delivery of the viral contents to the cellular interior (2, 3).

Prior to infection of human tissue, the homotrimeric hemagglutinin, HA0, requires cleavage by one of several human proteases (4), resulting in a trimer of heterodimeric proteins, HA1 and HA2, covalently linked to one another by a single disulfide bridge. The atomic structures of the precursor and processed HA have been solved by X-ray crystallography (5–7) and reveal that the main rearrangement taking place upon cleavage is the relocation of the newly formed N-terminal HA2 tail from the surface of HA0 to the helical core of the HA2 ecto domain. Upon lowering the pH, the N-terminal HA2 fusion domain is extruded from the HA2 helical core to a position where it can target the endosomal membrane, while the C-terminal transmembrane helix keeps HA2 anchored to the viral membrane. These two membrane anchors are separated by the homotrimeric rod-shaped ecto domain which subsequently undergoes a drastic structural rearrangement, bringing the N- and C-terminal membrane anchors in close proximity (2, 3). During this rearrangement, the C-terminal transmembrane anchor of HA2 remains embedded in the viral membrane, but the structural rearrangement brings the N and C termini of HA2, and thereby their associated endosomal and viral membranes, in proximity of one another—a process preceding viral fusion (6, 8). Fusion then proceeds through a hemi-fusion intermediate (9), with mixing of the outer leaflets of the two membranes, followed by formation of a fusion pore (10–12).

Despite many years of intense study, the detailed atomic-level mechanism of membrane fusion remains elusive and challenges our understanding of protein-lipid interactions. The role of the fusion peptide has been investigated extensively by mutagenesis, carried out on the intact protein (13–19) and also on isolated HAfp peptides, which exhibit fusogenic activity. The latter studies investigated lipid-bound, fusion-active structures (20, 21), how they interact with the host membrane to promote fusion (22–25), and how conservative mutations have fatal consequences on fusion activity. Many of these studies focused on a truncated form of the HAfp subtype H3 peptide (20) and its various mutants in dodecylphosphatidyl choline (DPC) micelles at both neutral and fusogenic pH values (26, 27). This truncated form only contains the N-terminal 20 residues of HA2, followed by a Lys-rich heptapeptide for residues 21–27, enhancing its water solubility. Use of this so-called P20H7 host-guest system not only facilitated peptide synthesis and purification, it also enabled straightforward thermodynamic studies of its interaction energy with various lipid environments (28). However, the use of P20H7 is not without shortcomings. In particular, the absence of the strictly conserved residues, Trp²¹-Tyr²²-Gly²³, is a concern. Trp and Tyr often play a pivotal role in anchoring peptides and proteins at the membrane-water interface (29, 30), and conserved Gly residues in membrane proteins, in particular when part of a GXXXG or GXXG motif, as is Gly²³ in HAfp, can be important in helix stabilization and interhelical interaction (31, 32).

The P20H7 wild-type peptide was reported to have a boomerang shaped structure (20, 33), with its apex exposed to solvent, and with its positively charged N-terminal amino group (34) and negatively charged Asp¹⁹ side chain, more than 20 Å away from the N terminus, deeply embedded in the hydrophobic interior. Without any clear charge compensation, this raises the question of what drives the burial of these charged groups. A second question not resolved by the P20H7 structure relates to the impact of mutation of Gly⁸ to Ala, which abolishes fusion activity, without requiring disruption of the reported P20H7 backbone structure. A third question is whether the absence of the highly conserved residues 21–23 impacts the structure of the remainder of HAfp.

In order to address the above questions, we have isolated and characterized the full length HAfp for the H1 sero subtype, followed by a Lys-rich heptapeptide similar to that used for the shorter sequence, again solubilized in DPC micelles. The peptide was obtained using a bacterially expressed HAfp fusion protein, followed by proteolytic cleavage and purification. This approach allowed us to include ¹³C, ¹⁵N and ²H labeling, enabling use of the full battery of modern multidimensional isotope-based NMR methods, including the measurement of very extensive NOE, residual dipolar coupling (RDC), and J coupling restraints. The isotope labeling approach also facilitates evaluation of the pH dependence of the structure, its position at the water-lipid interface, and the protonation state of Gly¹. Our results show that the three conserved residues absent in P20H7 but present in our full length HAfp construct play a key role in stabilizing a highly ordered structure, where its N- and C-terminal helices pack tightly together in a hairpin-like structure. One side of the helical hairpin is completely hydrophobic and shows NOE interactions to the alkyl chains in the interior of the DPC micelle, whereas the other surface includes the polar side chains of Glu¹¹, Thr¹⁵, and Asp¹⁹, and the H^α2 hydrogens of five Gly residues located at the interhelical interface.

Results

Comparison of the ¹H-¹⁵N heteronuclear single quantum correlation (HSQC) NMR spectra of a truncated HAfp peptide containing only the first 20 residues, HAfp^1–20, with that of the full length HAfp^1–23 shows remarkably large shifts in resonance positions, suggestive of substantial structural differences (Fig. S1). Whereas narrower ¹⁵N line widths towards the amino- and carboxy-terminal ends of HAfp^1–20 are indicative of increased disorder, HAfp^1–23 shows highly homogeneous line widths with residues 21–23 protected from rapid exchange with solvent, even at the relatively high pH value of 7.4, indicative of stable hydrogen bonding. Comparison of the HAfp^1–23 HSQC spectra recorded at pH 7.4 and pH 4.0 shows only small changes in peak positions, with the exception of the backbone amides of Glu¹¹ and Asp¹⁹, which alter their side chain protonation state (Fig. S2). The similarity in the full length HAfp^1–23 HSQC spectra, recorded at high and low pH values, points to the absence of a major structural rearrangement. This conclusion is confirmed by the measurement of RDCs and NOE spectra (vide infra).

Quantitative evaluation of the HAfp^1–23 backbone rigidity was carried out by analysis of amide ¹⁵N relaxation rates (35, 36). Both the longitudinal (R₁) and transverse (R₂) ¹⁵N relaxation rates at pH 7.4, as well as the heteronuclear ¹⁵N-{¹H} NOE, are remarkably uniform over the entire HAfp region, from residue Leu² to Gly²³ (Fig. 1). Note that the amino-terminal resonance of Gly¹ is not detectable due to the intrinsic fast exchange of hydrogens with solvent. Analysis of the amide ¹⁵N relaxation rates in terms of the standard model-free Lipari-Szabo formalism (37) yields rotational correlation times of 8.4 and 9.1 ns at pH 7.4 and 4.0, respectively, for the micelle-attached peptide, comparable to values measured for other peptides embedded in DPC micelles (38). Slightly elevated ¹⁵N R₂ rates for a subset of the backbone amides at pH 4.0 (Glu¹¹, Trp¹⁴, and Met¹⁷) suggest the presence of an exchange contribution to these rates at acidic pH. However, considering that the corresponding resonances remain sharp, with RDC and ¹H-¹H NOE values remaining fully consistent with those observed at high pH (Fig. 2D), the alternate conformation sampled by HAfp^1–23 must either have a very low population or be structurally very similar to the major conformer.

Fig. 1.

HAfp^1–23 backbone dynamics extracted from ¹⁵N relaxation rates at a magnetic field strength of 14.1 Tesla, pH 7.4 (black) and 4.0 (red). (A) Longitudinal ¹⁵N relaxation rates, R₁; (B) Transverse ¹⁵N relaxation rates, R₂; (C) ¹⁵N-{¹H} NOE; (D) Model-free generalized order parameters, S², derived from these data for a librationally corrected 1.04 Å bond length and an axially symmetric ¹⁵N Δσ value of -173 ppm applicable for α-helices (57), using an isotropic diffusion model.

Fig. 2.

Structure of HAfp^1–23. (A) Longitudinal view with the hydrophobic side chains drawn in yellow, polar side chains in green, acidic side chains in red, and Gly residues shown in white van der Waals representation. The interhelical angle between helix A (residues 2–12) and helix B (residues 14–22) is 158 °, and the interhelical distance calculated using the program interhlx (K. Yap, University of Toronto) is 6.3 Å. The amino terminus is marked N.(B) Lateral view. (C) Longitudinal view showing all atoms, and interhelical NOEs below 5 Å as black lines. (D) Singular value decomposition fit of the pH 4.0 and RDCs to the refined pH 7.4 structure. The couplings have been scaled by -0.484 relative to to account for the relative internuclear dipolar interaction strength. The Q-factor of the pH 4.0 data to the pH 7.4 structure is 22.3%.

The high degree of backbone order, as revealed by near-unity values of the generalized order parameter, S², across the entire length of HAfp^1–23 is comparable to that found in well-structured globular proteins. For such well-structured systems, interpretation of RDCs in terms of orientations of the internuclear vector is straightforward (39), and a large number of such couplings were therefore measured, including all and couplings, the geminal Gly values, and the intrinsically much smaller backbone and RDCs. In addition, more than 450 ¹H-¹H distance restraints were derived from recording two- and three-dimensional NMR spectra (Fig. S3). Both the short and medium range NOE data, the NMR chemical shifts, and the RDC data are fully compatible with α-helical structures extending from Gly¹ to Gly¹², and Trp¹⁴-Gly²³. A remarkable feature in these NOE spectra, however, is the presence of relatively strong , and NOE interactions between residues located in these separate α-helices, including interactions between Met¹⁷ and Phe⁹, Trp²¹ and Ala⁵, and Trp²¹ and Gly¹ (Fig. S3). Use of mixtures of ¹³C-labeled and unlabeled peptides combined with isotope editing and filtering procedures (40, 41), proved all these interhelical NOEs to be intramolecular, thereby excluding antiparallel homodimeric arrangements of fully α-helical HAfp^1–23 molecules.

Close contacts between H^α protons which are distant in the amino acid sequence are commonly observed in β-sheet structures (42). The presence of such contacts between residues in neighboring α-helices is rare, but not unprecedented. For example, short (≤ 3 Å) interhelical H^α-H^α distances are seen in the classic glycophorin A (GpA) dimer structure (43), even though the corresponding NOEs were not reported at that time, and in a substantial number of X-ray structures (44). Multiple interhelical H^α-H^α NOEs are reported in a recent NMR study of the BNIP3 transmembrane domain (45). Both GpA and BNIP3 structures are homodimeric and stabilized by intermolecular H-bonds between C^αH and backbone carbonyl oxygens, with the H-bond-linked transmembrane helices in a parallel orientation. Our NMR structure, derived from the NOE, RDC, and three-bond and coupling restraints, shows a tightly packed helical hairpin (Fig. 2), where the helices are antiparallel to one another but also exhibit the characteristic interhelical hydrogen bonds between C^αH and backbone carbonyl oxygens. For each of these four H-bonds (Trp²¹ C^αH to Gly¹ C = O; Met¹⁷ C^αH to Ala⁵ C = O; Ala⁵ C^αH to Met¹⁷ C = O; Phe⁹ C^αH to Gly¹³ C = O), the carbonyl oxygen of residue i accepts the regular helical H-bond from its as well as from the atom of the other helix, putting in close proximity these hydrogens from separate helices. Therefore, such interhelical H-bonds predict strong interhelical NOE interactions between and , as indeed observed for all four above listed H-bonds (Fig. S3D). Furthermore, the proximity and geometry between the Gly¹ protons and the carbonyl of Gly²⁰ produces another hydrogen bond, as identified by the potential-of-mean-force hydrogen bond database (HBDB) module (46) of the XPLOR-NIH structure calculation program (47). Inclusion of these hydrogen bonds as fixed restraints in the refinement did not adversely affect structural statistics and improved the quality validation factors, and (Table S1).

In the resulting structures, one side of the helical hairpin is completely covered by hydrophobic side chains (Fig. 2B), whereas the other side exposes the polar side chains of Glu¹¹, Thr¹⁵, and Asp¹⁹, as well as the H^α2 protons of the α-helical Gly residues. The side chains of the majority of hydrophobic residues are locked into a single χ₁ conformer, as judged by and values (Table S2), and confirmed by intraresidue and sequential NOEs. A positive ϕ angle on Gly¹³ reverses the backbone chain, leading into the second helix, and a total of 66 long-range NOEs (Fig. 2C) lock in the tertiary contacts between the two helices. Small regions taken from the two-dimensional and three-dimensional NOESY spectra illustrate the unambiguous nature of such interactions when spectra are recorded at high field (900 MHz) and at high spectral resolution (Fig. S3).

The close similarity between the HSQC spectra recorded at pH 7.4 and 4.0 (Fig. S2) strongly suggests the absence of major structural rearrangement between the two pH values. Indeed, the interhelical contacts found at pH 7.4 remain present at pH 4.0 (Fig. S3 B, C). The RDC values recorded at pH 4.0 differ substantially from those at pH 7.4 (Fig. S4), which could be caused by a change in structure or by a different orientation of HAfp^1–23 relative to the magnetic field. The finding that the RDCs measured at pH 4.0 fit well to the structure calculated at pH 7.4 (Fig. 2D) indicates that it is the change in alignment, presumably reflecting the change in electrostatic repulsion between the alignment medium (negatively charged dGpG (48), or negatively charged polyacrylamide gel (49, 50)) and the HAfp^1–23 upon protonation of Glu¹¹ and Asp¹⁹, which dominates the change in RDCs, and not a change in HAfp^1–23 structure.

Implicit in the segregated distribution of the side chains according to hydrophobicity (Fig. 2B) is the orientation of HAfp^1–23 relative to a phospholipid surface, with the aromatic and hydrophobic side chains pointing downward, towards the hydrophobic interior, and the top of the molecule exposed to solvent. The intermolecular NOE contacts between the micelle and HAfp^1–23 provide additional support for this interfacial location (Fig. 3). NOEs to the alkyl chains of DPC are more than twofold weaker for the amide hydrogens of Gly⁴, Gly⁸, Gly¹⁶, and Gly²⁰, located at the top of the molecule, than for amides on the opposing face, despite the Gly amide protons not being shielded from close intermolecular contacts by the side chains of the perdeuterated HAfp^1–23, used for these measurements. Moreover, the amides of Ala⁵, Ile⁶, and Ile¹⁸ on the hydrophobic face are the only ones showing detectable NOE interactions with the terminal methyl group of DPC. Intermolecular NOEs between the DPC choline methyl resonance and the Trp¹⁴ and Trp²¹ H^ε1 indole protons are also observed (Fig. S5). Trp H^ε1 and Tyr OH groups in membrane proteins frequently form H-bonding interactions to the polar atoms at the water-lipid interface of membrane lipids (29, 51), and the observation of these interactions is therefore not surprising.

Fig. 3.

HAfp interactions with the micelle and with water at pH 7.4. (A) Strength of NOE interactions between perdeuterated, amide-protonated HAfp^1–23 and DPC micelle protons. Amides with cross peaks to choline methyl protons (Hγ) are at the lipid-water interface. Residues with stronger intensity to the DPC methylene (H3-H11) and methyl (H12) protons are closer to the lipid core of the micelle. A NOE mixing time of 100 ms was used, and the plotted intensities have been scaled by the intensity of the diagonal amide resonance. DPC/H^N NOEs at pH 7.4 and 4.0 are compared in Fig. S5C. (B) NOE intensity between the DPC H3-H11 methylene protons and the backbone H^N protons of the fusion peptide mapped on the structure, with red corresponding to strongest NOEs. Residues in gray could not be detected due to fast exchange of the amide/amino protons with water. The amino terminus is marked N. (C) Solvent hydrogen exchange rates of the backbone amide protons. (D) Hydrogen bonding pattern in the HAfp^1–23 structure. Residues Gly¹ to Gly¹² and Trp¹⁴ to Gly²³ have well satisfied αhelical hydrogen bonds, marked by black dashed lines. C^αH-O = C hydrogen bonds are shown as green dashed lines. Hydrogen bonding between the of the N-terminal Gly¹ and the carbonyl oxygen of Gly²⁰ is also suggested by the structure. Inclusion of the N-terminal hydrogen bond and the C^αH-O = C hydrogen bonds does not adversely affect experimental statistics of the refined structure (SI Text). Blue residue labels indicate proximity to the DPC choline; red labels are for residues with NOEs to the DPC methyl groups.

Amide hydrogen exchange rates indicate that all backbone amides engaged in α-helical hydrogen bonds are strongly protected from exchange with solvent (Fig. 3C), but the amides at the N-terminal ends of the two helices, which lack their C = O H-bond partner (preceding it by four residues) show less protection. Nevertheless, these hydrogen exchange rates are significantly (> 10-fold) slower than expected for random coil peptides dissolved in water at pH 7.4, 33 °C (52). This latter observation suggests that these amides are not fully accessible to solvent and/or engaged in hydrogen bonds with phospholipid headgroups. For the two Gly residues in the turn region of the hairpin, Gly¹²-NH makes a slightly distorted H-bond to the Gly⁸-C = O, while Gly¹³-NH donates a H-bond to the C = O of Phe⁹, resulting in relatively high hydrogen exchange protection factors of > 100. The structures calculated also suggest the presence of a hydrogen bond, with O-H distance of ≤ 2.5 Å and C-O-H angle of ≥120°, between the of Gly¹ and the carbonyl oxygen of Gly²⁰, and simultaneously for a subset of the calculated structures, between and either Trp²¹ C = O or Gly²³ C = O. The N-terminal Gly¹ amino group of HAfp^1–23 is protonated, as indicated by its ¹⁵N chemical shift of 27.4 ppm. Despite the long-range H-bond interactions of the Gly¹ amino hydrogens, their exchange with solvent remains too rapid for direct observation of their ¹H NMR signal. Similarly, the amide signal of Leu² at pH 7.4 also exchanges too rapidly with solvent to permit observation of its ¹H-¹⁵N correlation in the two-dimensional HSQC spectrum, although it yields a visible resonance at pH 4.0 (Fig. S2), where hydrogen exchange is intrinsically much slower.

Discussion

The structure reported here for full length HAfp^1–23 differs substantially from results reported for the construct that lacks the strictly conserved residues Trp²¹-Gly²³. Nevertheless, initial NMR measurements by us on these shorter versions of the fusion peptides also exhibited long-range H^α-H^α NOE contacts between Phe⁹ and Met¹⁷, and Ala⁵ and Ser²¹ (with Ser²¹ being the first residue of the carboxy-terminal solubilization tag). However, these NOE interactions were quite weak, and ¹⁵N relaxation measurements indicated significant disorder at both the N- and C-terminal ends of this HAfp^1–20 fusion peptide. The presence of these weak long-range NOEs suggest that HAfp^1–20 also populates the helical hairpin conformation, but only transiently. Indeed, we were unable to fit the NOEs and RDCs observed for HAfp^1–20 to a single static structure. These observations prompted us to study constructs containing all strictly conserved residues, encompassing residues 1–23.

The helical hairpin structure observed in our study is stabilized by interactions between residues Trp²¹-Gly²³ and the N-terminal residues of HAfp, explaining the high degree of backbone order observed for these residues (Fig. 1D). In retrospect, the tight antiparallel helical packing of the two helices in HAfp^1–23 is perhaps not surprising, considering that both the N- and C-terminal helix contain the classic GXXXG motifs that are the hallmark of many tight transmembrane helix-helix associations (31, 32). In most, but not all of these associations, the helical orientations are parallel, not antiparallel. However, the Protein Data Bank does not contain a single case where the antiparallel, aliphatic hydrogen-bond linked helices are connected by a tight hairpin turn, as found here for HAfp. The close packing is permitted by Gly residues that line the inner faces of the two helices (Fig. 4A), consisting of Gly¹, Gly⁴, and Gly⁸ on one side, and Gly¹⁶, Gly²⁰, and Gly²³ on the other, all of which are completely conserved (8, 53). Sequence conservation of the HA2 fusion domain has been shown to be important for burial of the HAfp in the hydrophobic interior of the postcleavage but prefusion state of HA, playing a critical role in tuning the pH where membrane fusion occurs (8, 15, 54). Our structural results on HAfp^1–23 after the peptide has been extruded from the hydrophobic core highlight a second, structural role for this high degree of sequence conservation. Positioning of the hydrophobic and hydrophilic residues relative to the hairpin surface appears key for defining its location at the water-lipid interface (Fig. 2, 3). Strict sequence conservation of Gly¹³ permits the positive ϕ angle, allowing formation of the exceptionally tight helical hairpin arrangement. The Gly⁸ H^α3 atom is fully buried at the interhelical interface (Fig. 4A), and modeling indicates that even a Gly → Ala mutation at this position is incompatible with the hairpin structure. This potentially explains the strict conservation of this residue, and the finding that a G8A mutation abrogates fusogenicity (14). In fact, six Gly residues in HAfp are found at the helical interface, and the remaining two are at the center of the hairpin turn. The only HAfp residue that lacks strong sequence conservation is Thr¹⁵, for which mutations to Glu, Gln, Ser, and even Pro are observed in the various HA sero subtypes. Residue 15 is located on the solvent-side of the hairpin. The position of the molecule at the N-terminal end of helix 2 causes it to lack the intrahelical hydrogen bond for its amide proton, and mutation to Pro is therefore less disruptive. Indeed, Pro is commonly found as one of the first residues in an α-helix.

Fig. 4.

Ribbon structures showing the orientation of Gly residues and various side chains. (A) The Gly H^α3 protons, which would be replaced by a side chain C^β atom if mutated, are displayed. (B) Side chains of Ala⁵, Ala⁷, and Trp²¹. Mutation of Ala⁵ to Val would require rearrangement of the Trp²¹ side chain.

The two H^α protons of Gly¹ are highly nonequivalent and differ in ¹H chemical shift by 0.7 ppm, confirming the high degree of ordering for this residue. Its position is defined both by NOEs and by and RDCs, orienting it such that its amino group donates a H-bond to the carbonyl of Gly²⁰, while its carbonyl accepts an α-helical H-bond from Ala⁵. The N-terminal Gly is highly conserved, and it has long been known that fusion activity is sensitive to its mutation. Not unexpectedly, deletions of G1 and L2 abolish fusion (14) as removal of Gly¹ or Leu² abrogates these interactions, which appear key in stabilizing the hairpin configuration. In contrast to the bulky G1V mutation, substitutions of Gly¹ by a non-β-branched polar residue (G1E, G1Q, and G1K) appear possible without disruption of the backbone or its H-bond interactions, but these mutations have been reported to lack fusion activity (19). Instead, the more conservative mutation to a small hydrophobic side chain, G1A, does not abolish fusogenic activity (14, 19). Interestingly, the G1S mutation is fusogenic, but activity is arrested at the hemi-fusion state (19), highlighting the critical role of the N-terminal residue in transitioning from the hemi-fused state to channel formation.

Strict sequence conservation is found for the two Trp residues in HAfp, at positions 14 and 21. Even though mutations to residues with smaller side chains are structurally feasible without disrupting the backbone, Trp residues are known to be key in positioning proteins at the membrane-water interface. It therefore is likely that the conserved nature of these two residues relates to their function in interacting with the membrane, rather than being dictated by conserving the backbone structure. Interestingly, the inactive A5V mutation (24) would cause a steric clash with the Trp²¹ side chain, requiring its rearrangement (Fig. 4B). In contrast, the A7V mutation is fully active and fusogenic (24). When retaining the backbone structure of HAfp^1–23, the Val⁷ side chain is located in the plane of the hairpin, pointing sideways at the water-lipid interface, without causing any steric interference with the interhelical interaction.

Even with the current structure of HAfp in hand, the mechanism by which the peptide mediates viral fusion is not immediately obvious. Single cell fusion studies indicate the formation of fusion pores, forming narrow aqueous channels, as an early intermediate in the fusion process (11, 12), with models suggesting six HA trimers, i.e., 18 HAfp peptides participating in the formation of a single pore (55). The length of the HAfp hairpin structure appears too short to span a lipid bilayer, and it is conceivable that during the process of pore formation the peptide rearranges from the helical hairpin to a contiguous α-helical state upon forming oligomeric pores. Although highly speculative, it is possible that the small degree of conformational exchange broadening observed in HAfp^1–23 at low pH, but absent at high pH (Fig. 1C), is related to such a structural transition. In this respect, it is interesting to note that another class I viral protein fusion domain, gp41 of HIV-1, adopts a contiguous α-helical conformation when studied in detergent micelles (56).

Materials and Methods

HAfp^1–23 of sero type H1 was expressed as a fusion protein, flanked by the residues SGKKKKD at its C terminus and by the Igg-binding domain B1 of streptococcal protein G (GB1; PDB entry 3GB1) at its N terminus. Cleavage by factor Xa protease resulted in the peptide with sequence GLFGAIAGFI EGGWTGMIDG WYGSGKKKKD (Fig. S6). Details regarding expression and purification are included as SI Text.

NMR measurements were carried out at 600 and 900 MHz on uniformly ¹⁵N/¹³C- and ²H/¹⁵N/¹³C-enriched samples at peptide concentrations of ca 0.6 mM, in 130–180 mM DPC. NMR structures were calculated using semiquantitative NOE distance restraints (Fig. S7), starting from a fully extended chain, using the XPLOR-NIH software (47). For details, see SI Text.