Order and disorder control the functional rearrangement of influenza hemagglutinin

Significance

Influenza hemagglutinin (HA), a viral surface glycoprotein, undergoes a critical and large conformational rearrangement to promote fusion of the viral membrane with the host membrane. Unlike the variable receptor binding domain HA₁, the coiled-coil domain HA₂ is highly conserved, making HA₂ a promising target for therapeutics. Furthermore, the structural similarity between influenza HA₂ and other viral fusion proteins, including that of HIV, makes HA₂ a valuable model system. We build a model using information from only the prefusion and postfusion configurations of HA₂ and use molecular dynamics simulations to characterize the structural ensembles found during the conformational transition. We find that local unfolding facilitates interaction of HA₂ with the host membrane and enables a quasi-stable asymmetric intermediate during the transition.

Abstract

Influenza hemagglutinin (HA), a homotrimeric glycoprotein crucial for membrane fusion, undergoes a large-scale structural rearrangement during viral invasion. X-ray crystallography has shown that the pre- and postfusion configurations of HA₂, the membrane-fusion subunit of HA, have disparate secondary, tertiary, and quaternary structures, where some regions are displaced by more than 100 Å. To explore structural dynamics during the conformational transition, we studied simulations of a minimally frustrated model based on energy landscape theory. The model combines structural information from both the pre- and postfusion crystallographic configurations of HA₂. Rather than a downhill drive toward formation of the central coiled-coil, we discovered an order-disorder transition early in the conformational change as the mechanism for the release of the fusion peptides from their burial sites in the prefusion crystal structure. This disorder quickly leads to a metastable intermediate with a broken threefold symmetry. Finally, kinetic competition between the formation of the extended coiled-coil and C-terminal melting results in two routes from this intermediate to the postfusion structure. Our study reiterates the roles that cracking and disorder can play in functional molecular motions, in contrast to the downhill mechanical interpretations of the “spring-loaded” model proposed for the HA₂ conformational transition.

Hemagglutinin (HA) is a viral receptor-binding and membrane-fusion glycoprotein involved in the invasion of influenza virions into host cells (1). Structural rearrangements of HA during membrane fusion are crucial for the delivery of the viral genome. The postfusion conformation of HA shows considerable similarity to other viral fusion proteins and eukaryotic membrane receptors involved in intracellular vesicle trafficking (2), suggesting there may be common mechanisms in the function of these proteins. Therefore, HA may serve as a model system, allowing characterization of the molecular and energetic details that underlie its conformational transition to provide insights into general principles of membrane fusion (3).

HA is a homotrimer consisting of two domains connected by disulfide bonds (4): a globular receptor binding domain (HA₁), and a coiled-coil membrane-fusion domain anchored to the viral membrane (HA₂). Recognized by the sialic acid receptor of a host cell, the intact virus enters the cell via endocytosis. Low pH in a late endosome then induces the dissociation of HA₁ from HA₂ (1) and an irreversible conformational transition of HA₂. Experimentally, this conformational change can be triggered by either low pH, high temperature, or urea denaturation (5).

Structures of HA in pre- and postfusion pH conformations have been solved by X-ray crystallography. The structure of the prefusion ectodomain contains both HA₁ and HA₂, and was purified from influenza virions (4). A postfusion conformation of HA₁ and HA₂ were obtained from prefusion viral HA that was sequentially treated with low pH and trypsin (6, 7). Comparison of these two structures shows no structural changes in HA₁, but a major rearrangement in HA₂, including secondary, tertiary, and quaternary structural changes. The N-terminal domain of HA₂, initially adjacent to the transmembrane region in the prefusion configuration, undergoes a large movement (over 100 Å) during the transition. The C-terminal domain of HA₂ changes from a globular structure to three extended loops packed against the central coiled-coil.

Although experiments have probed the fusion mechanism through mutation (8) and provided measures of fusion kinetics (9), there is a lack of structural information about how HA₂ transitions from the prefusion to postfusion conformations. Theoretical models have been suggested to describe the fusion mechanism based on the available experimental kinetic data (10–12). However, because of the large scale of the HA₂ rearrangement, only limited computational techniques, such as targeted molecular dynamics (13), have been applied to study the molecular details of the transition.

In this study we applied the principles of the energy landscape theory as developed in the context of protein folding (14–16) to examine structural details of the HA₂ conformational transition. We used a structure-based model (SBM) (17, 18) built with a dual-funneled landscape (19, 20) that has both the pre- and postfusion structures as explicit minima. The HA₂ landscape has at least two competing basins of attraction, corresponding to the pre- and postfusion structures of HA₂, respectively. HA₁ dissociation sterically enables HA₂ to explore beyond the prefusion local free-energy minimum and to diffuse toward the postfusion configuration. The long-length scale and extensive shuffling of secondary and tertiary structures is reminiscent of protein folding, but distinct in that both ends of the HA₂ transition can be described by ensembles of structurally similar configurations. Just as in protein folding, there may be free-energy barriers and structural intermediates along the HA₂ transition caused by the imperfect cancellation of energy and entropy. These intermediate ensembles may be interesting candidates for drug design to inhibit HA function.

Previously, a “spring-loaded” model has been applied to describe the mechanism for the HA₂ transition (21). This model suggested a downhill mechanical transition of the N-terminal region of HA₂ into an ordered helical structure that orients the fusion peptides away from the virus and toward the host membrane. Our simulations expand this view by showing that the conformational change of the N-terminal domains is associated with an entropic barrier and the unfolding of the C-terminal region is associated with the major energetic barrier during the HA₂ conformational transition. Kinetic competition between these two events creates a long-lived metastable intermediate that allows for two dominant routes. The first route (the “sequential route”) resembles the spring-loaded model, and the second route (the “cooperative route”) involves cooperative interactions between the N-terminal and C-terminal domains in forming the central coiled-coil. The presence of these distinct routes suggests multiple mechanisms for HA₂ rearrangement and membrane fusion.

Results

Subscripts I and F are used to denote either the prefusion (I, initial) or the postfusion (F, final) basin. For example, the prefusion and postfusion states of the full HA₂ trimer are denoted HA_2I and HA_2F throughout this paper, with similar notations being given to individual protein domains S_I and S_F. The HA₂ trimer is partitioned into five segments, S1 to S5 (Fig. 1), identified from these crystal structures as regions with distinct structural changes during the transition (6, 22). We also highlight Loop3-4, a subset of S4 that becomes a loop connecting S3 and S4 in the postfusion structure. S1 (residues 33–56 in each monomer) undergoes rearrangement of tertiary structure, which brings the fusion peptides (not represented in the simulation) away from the viral membrane. S2 (residues 57–74 in each monomer) has both secondary and tertiary changes, transitioning from extended loops to α-helices. S3 (residues 75–106 in each monomer), a triple-stranded coiled-coil, is the only region that is structurally conserved between the HA_2I and HA_2F crystal structures. S4 (residues 107–127 in each monomer) partially reorganizes to orient itself antiparallel to the central coiled-coil S3. Loop3-4 (residues 107–112 in each monomer) undergoes a helix-to-coil transition. S5 (residues 128–175 in each monomer), an initially globular region near the viral membrane, breaks into three loops that pack into the grooves of the central helices in the HA_2F configuration. Because of the trimeric nature of HA, the rearrangement also changes the quaternary structure of several sections. For example, S5 transitions from a configuration with many S5-S5 quaternary interactions in HA_2I to a configuration with none in HA_2F. The fusion peptides and their linkers to S1 (residues 1–32 in each monomer) initially have their N-termini buried in a cavity surrounded by S4. These residues do not exist in the postfusion structure, nor in our simulations.

Fig. 1.

Influenza hemagglutinin X-ray crystal structures. (Left) Receptor binding domain HA₁ and membrane fusion domain HA₂ oriented with respect to a cartoon membrane. A pH drop after endocytosis dissociates HA₁ from the HA₂ trimer, sterically allowing the conformational transition. (Right) The crystal structures of prefusion and postfusion HA₂ (PDB IDs 2HMG and 1QU1) sectioned and colored by different regions from S1 to S5 (defined in the text). Open and filled circles denote the N and C termini, respectively.

In the case of protein folding, the number of atom pairs that form native contacts Q is often used as a reaction coordinate to characterize the folding dynamics (23, 24). Here, an atom pair was considered to be in contact if the pair existed as a native contact in one of the two crystal structures and the spatial distance between the atoms was less than 1.5 times their distance in the crystal structure (17). The reaction coordinate Q_X/Y was introduced to quantify the number of interface contacts between section X and Y. Because contacts unique to one conformation could be broken without a compensating contact unique to the other formed, the Q for either HA_2F or HA_2I individually cannot serve as a reaction coordinate to monitor the overall transition progress. Rather, we chose a difference of Q between HA_2F and HA_2I configurations: ΔQ_X/Y = Q_X/Y(HA_2F) − Q_X/Y(HA_2I), where Q(x) means Q relative to structure x. Accordingly, three reaction coordinates were selected: ΔQ_S123/S45 quantifies the number of interfacial contacts between subunits S1-3 and S4-5 and is a measure of global progression of the transition, whereas ΔQ_S1/S1 and ΔQ_S5/S5 measure both interfacial and intramonomer contacts within S1 and S5 and serve to distinguish between the two major routes observed in this transition. Finally, ΔQ_trans = ΔQ_S123/S45 + ΔQ_S1/S1 was used to avoid a spurious backtracking along the coordinate ΔQ_S123/S45 (SI Appendix). All reaction coordinates were normalized to range from 0 to 1 (Fig. 2).

Fig. 2.

Two-dimensional kinetic probability distribution of the HA₂ structural transition at T = 0.8T_m, the melting temperature of HA₂ (Top). ΔQ_trans = Q_S123–S45 + Q_S1–S1, ΔQ_S1–S1, and ΔQ_S5–S5 are used as the reaction coordinates and combine data from 1,000 independent trajectories. The deepest local minima are labeled as basins A–H and representative structures for each are shown below. (A) Thermal fluctuations about the HA_2I crystal structure. (B) Arms (S1-2) of HA₂ come off and fusion peptides are exposed. (C) HA₂ bends resulting in a broken trimeric symmetry. (D and E) Same as B and C except that two helices in S1 have associated with each other. (F) All three helices in S1 associated. Note that by the time all three helices in S1 associate, HA has already bent in over 99% of the simulations, so the basin to the left of F is scarcely populated. (G) An intermediate state where the overall structural transition is complete except for one unit of S5, which is likely caused by asymmetry in the HA_2F crystal structure. (H) Thermal fluctuations around the HA_2F structure. Basins A–E represent the common “fast phase,” and the bifurcation from basin E distinguishes two transition routes. The Inset shows complete sampling for a single-basin HA_2I model of HA₂. (Middle) Average Q_S1–S1 (black) and Q_S5–S5 (red) for the two routes bifurcating from basin E. The two coordinates increase concommitantly in the cooperative route.

Experimentally, the HA₂ transition has been found to be irreversible (1), implying that HA_2F is much more stable than HA_2I and only the I-to-F transition is of interest. The greater energetic stability of HA_2F is captured within this model by the higher number of atom-atom native contacts in the HA_2F structure (∼4,500) than in the HA_2I structure (∼3,000). Accordingly, we characterized the kinetic intermediates along the transition using simulations that were initiated in the HA_2I conformation and were terminated once they reached the HA_2F ensemble (Fig. 2). In these simulations, we find that the dynamics within any HA₂ transition can be separated into a fast and a slow phase based on the rate of events relative to that of S5_I unfolding. Once S5_I unfolds, the protein rapidly adopts the HA_2F configuration.

Competing Interactions Drive the HA₂ Transition.

The biological function of HA₂ is to introduce and insert the fusion peptides into the host membrane. Fusion peptide insertion must occur before the C-terminal domains (S5) melt from the HA_2I configuration and adopt the HA_2F configuration (Discussion). Without a local energetic bias toward the HA_2I configuration, S5_I would immediately melt. Therefore, any model of functional HA₂ must contain stabilizing interactions from both the HA_2I and HA_2F crystal conformations, necessitating at minimum a two-basin representation (Materials and Methods).

The dual-basin implementation sheds insights on the driving force for the HA₂ conformational transition. Simulations using a single-basin model for HA_2I exhibit local fluctuations about Basin A (Fig. 2, Inset), but unfolding is never observed. The dual-basin potential destablizes Loop3-4 (Fig. 3) through local interactions with the HA_2F Hamiltonian and allows HA₂ to leave basin A. Furthermore, in a dual-basin model (upon removal of HA₁) the HA_2I configuration is no longer a stable minimum on the free-energy surface, and the mean time spent in basin A τ_A is only 1% of the overall mean transition time τ_trans (i.e., τ_A/τ_trans = 0.011).

Fig. 3.

Disorder during the HA₂ conformational change. (A) Initial cracking event (basins A and B) in the HA₂ S3/S4 interface (Loop3-4). Red curves show RMSF for each residue averaged over the three monomers. RMSF values are calculated as an average over each structure within a basin in all trajectories. The RMSF of S3 is virtually equivalent with respect to either reference. Black curves show a local dihedral disorder measure. This disorder measure quantifies the similarity between the dihedral angle distribution for a particular dihedral and a uniform distribution (fully disordered state) relative to fluctuations about HA_2I, and is bounded between −1 and 1. A larger value means more overlap with a uniform distribution and thus more disorder, with a value of zero indicating that dihedral fluctuations are equivalent to those about the single-basin HA_2I structure (SI Appendix). Notably, Loop3-4 undergoes an order to disorder transition from basin A to B, coinciding with the location that the N termini of the fusion peptides bury in the prefusion structure. (B) Representative structural picture showing that the disorder in Loop3-4 destroys the burial site for the fusion peptides. (C) RMSF and dihedral disorder data for all basins A–H. RMSF fluctuations are given relative to HA_2I where the background is white, HA_2F where shaded (minimum of the two fits is used). Threefold symmetry breaking in the fast phase (basins B to C, D to E) returns some local dihedral order in Loop3-4.

Fast Phase: Disorder, Cracking, and Threefold Symmetry Breaking.

In all simulations, the HA₂ transition initiates with two rapid events. The helical residues in Loop3-4 locally disorder (Fig. 3B) from a combination of low helical propensity (effected from HA_2F Hamiltonian) and an inherent lack of stability because of the cavity left behind when the fusion peptides move from their initial burial sites (Fig. 3A). Additionally, global torsion of HA₂ helps the “arms” (S1 and S2) to break their native interdomain contacts (Fig. 2, basin B). This torsion is consistent with previous predictions by normal mode analysis (25). This combination of events allows the arms (and their attached fusion peptides) to extend away from their previously restrained positions. For the remainder of the fast phase (Fig. 2, basins B–E), the arms broadly sweep out the configuration space, reflected in their large root mean-square fluctuation (RMSF) with respect to either reference crystal structure (Fig. 3C).

The local disorder transition in Loop3-4 has additional consequences beyond allowing the arms to dissociate. One monomer of S4 bends antiparallel to the central trimeric coil as it refolds to its HA_2F structure (Fig. 2, basins C and E). S5_I is a relatively stable trimeric unit, and although fully folded, geometrically precludes two of the three S4 monomers from adopting their HA_2F conformation. This leads to an asymmetric HA₂ configuration, where the central coil is bent sideways and packed against one side of S5. The S4 monomer that bends antiparallel forms most of its HA_2F contacts, causing a concurrent increase in Loop3-4 dihedral order as the symmetry is broken (Fig. 3, basins B to C, D to E).

A transient increase in local disorder that facilitates a conformational transition (Fig. 3, basins A to B), known as “cracking,” has been previously shown in other protein systems (26–28). Here, the conformational change from basin A to B has features similar to cracking; local disorder in S4 lowers the barrier of the transition by breaking S1_I interfacial contacts. Although this mechanism seems unlike previously studied examples of cracking because S4 returns to a configuration that is structurally dissimilar to S4_I, the configurations at either end of the transition are more ordered than the intermediate ensemble. Thus, we do not consider this to be a distinct mechanism from cracking.

Slow Phase: Conformational Transition Follows Two Dominant Routes.

From basin E, the protein proceeds via one of two routes to complete the transition (shown by the two arrows leaving basin E in Fig. 2). The first is a sequential route, where the S1 monomers associate before the S5_I trimeric interface breaks. This route may appear to resemble the spring-loaded intermediate, but remains distinct in that the threefold symmetry of HA₂ is broken and S2 often does not form a well-packed helical bundle before S5_I breaks. The second is a cooperative route, where the S5_I trimeric interface is broken before S1 domains overcome the entropic barrier to association. As S5 folds into its postfusion conformation, it forms interdomain contacts with S1-S4, cooperatively pushing the arms together. The unfolding of S5_I, which is common to both routes, is the rate-limiting step in the HA₂ transition.

Comparison of RMSF between basins E and F (Fig. 3) shows that S1 undergoes a disorder-to-order transition along the sequential route, suggesting that there may be a substantial entropic barrier to the formation of the helical trimeric interface. Additionally, comparisons of RMSF values between basins F and G show that S2 can remain disordered until tertiary contacts with S5 are formed, even if S1 has fully adopted the trimeric postfusion conformation. In the sequential route, S1 and S2 must overcome this entropic barrier with a conformational search, reflected in their high RMSF in basin E (Fig. 3). However, in the cooperative route, local guidance from interactions with S5 lowers the free-energy barrier to formation of the S1-S2_F coiled-coil.

The two dominant routes share similar kinetic characteristics, except for the relative order of events between them (Fig. 2, Middle). The sequential route spends a relatively longer time at around Q_trans = 0.4, corresponding to basin F in Fig. 2. Here, the association of S1_F happens before S5_I breaks apart, whereas in the cooperative route, S1 and S5 move to their final conformations concurrently. In both the sequential and cooperative route, the breaking of S5_I and subsequent formation of S5_F interface contacts are the final events in the HA₂ conformational transition.

Qualitative Transition Features Are Robust to Parameterization.

We calculated the folding routes and kinetics for four different Hamiltonians at varying temperatures (Table 1). Although in every case the conformational transition follows one of the same two previously discussed folding routes, the population of each route depends on the rate of the central coiled-coil formation (S1_F association) relative to the rate of S5_I melting. In particular, the type of threefold symmetry breaking observed in the fast phase of the vanilla model is extremely robust to changes in the Hamiltonian and is always observed. The melting of S5_I is similar to unfolding a globular protein, where increasing temperature lowers the barrier to unfolding and thus lowers the unfolding time for S5_I ${\overline{\tau }}_{\mathrm{S}5}$. Because the cooperative route is defined by S5_I melting before S1_F formation, the population of transitions through the cooperative route increases with temperature. Also, because S5_I melting is the rate-limiting step in the transition, the overall rate of the transition increases with temperature.

Table 1.

Effects of changing model parameters on the transition route

Model	T ∼ 0.8Tm	T ∼ 0.9Tm	T ∼ Tm
Homogeneous*
% Seq	27	11	0
${\overline{\tau }}_{\text{S}1}$	14	5.8	3.4
${\overline{\tau }}_{\text{S}1}^{\text{ind}}$	42	19	55
${\overline{\tau }}_{\text{S}5}$	17	4.7	1.0
Increased S2 helicity†
% Seq	83	38	0
${\overline{\tau }}_{\text{S}1}$	5.5	3.1	2.4
${\overline{\tau }}_{\text{S}1}^{\text{ind}}$	5.2	4.2	23
${\overline{\tau }}_{\text{S}5}$	17	3.6	1.3

% Seq is the percentage of trajectories that go through the sequential route. ${\overline{\tau }}_{\text{S}1}$ and ${\overline{\tau }}_{\text{S}5}$ give the mean first passage times for the formation of S1_F and the breaking of S5_I, respectively. ${\overline{\tau }}_{\mathrm{S}1}^{\mathrm{ind}}$ measures the first passage time for S1_F formation independent of S5 cooperation (i.e., with immobilized S5_I). Times are measured in units of ${\overline{\tau }}_{\text{S}5}$ at T_m. Values are averages over ≥100 independent trajectories for each parameter set.

^*The dual-basin SBM with both structures equally weighted.

^†Single-basin HA_2F dihedral potential for S2.

Previous experimental studies (21) and secondary structure predictors (29) have both suggested that the S2 monomer has high helical propensity. Because this propensity may be underestimated by the dual-basin modeling of S2 that gives equal weight to backbone torsions of the loop configuration, we tested a Hamiltonian that has secondary-structure propensity only to the S2_F configuration (fully helical). This perturbation reduces the time until S1_F formation ${\overline{\tau }}_{\mathrm{S}1}^{\mathrm{ind}}$ by limiting the conformational search time of S1 monomers. Smaller ${\overline{\tau }}_{\mathrm{S}1}^{\mathrm{ind}}$ with a constant ${\overline{\tau }}_{\text{S}5}$ also means higher sequential route population. Finally, when HA₂ is represented in its biological context of being anchored to the viral membrane, the transition shows qualitatively similar features (SI Appendix).

Discussion and Conclusion

A key step in the biological function of HA is insertion of fusion peptides into the host membrane (30). Our simulations focused on the conformational transition of HA₂ that facilitates the interaction between fusion peptides and the host membrane leading to subsequent pore formation. Based on the kinetic intermediates observed in the HA₂ simulations, a possible membrane-fusion mechanism is summarized in Fig. 4. Endocytosis is followed by a drop in pH, which causes HA₁ to dissociate (31), allowing HA₂ to subsequently undergo four refolding stages. (i) Loop3-4 cracks and the arms (S1 and S2) detach from the central helices of HA₂, releasing the fusion peptides. (ii) One S4 monomer forms an antiparallel structure with respect to the central helices (S3), causing HA₂ to bend toward the viral membrane and breaking the threefold symmetry. (iii) After reaching this asymmetric intermediate ensemble, the dynamics of HA₂ can diverge into one of two routes, both of which may be energetically accessible. In the sequential route, S1 first adopts its final trimeric arrangement, leading to a configuration where all fusion peptides can be inserted into the host membrane. In the cooperative route, one or two fusion peptides may insert into the viral membrane and the remaining into the host membrane. (iv) In the last stage of both routes, the C-terminal domain (S5_I) melts and wraps HA₂ into its final compact configuration. This step is likely very important for fusion because the viral transmembrane domains are attached to S5. Thus, in the HA_2F configuration the fusion peptides are in the same (presumably fused) membrane as the transmembrane domains. If at any point S5_I breaks before at least one fusion peptide is inserted in the host membrane, the conformational change may sterically restrict the fusion peptides near the viral surface, likely leading to an inactive HA₂. The dual-funneled landscape of HA₂ has evolved to avoid this fate (Fig. 5). Influence from the HA_2F funnel encourages cracking that expedites HA₂ reaching a fusogenic configuration. The HA_2I funnel kinetically traps S5_I to provide sufficient time delay for insertion of the fusion peptides into a membrane.

Fig. 4.

Cartoon illustrating the HA₂ conformational transition with respect to viral and host membranes. The sections and color scheme applied here is the same as that in Fig. 1. Three black curves represent linkers connecting S5 to transmembrane anchors, depicted as three thicker black lines in the viral membrane. After dissociation of HA₁, S4 cracks, releasing S1 and S2 from the central stalk. This is followed by a break in the threefold symmetry of HA₂. The stochastic search of the arms can lead to all three fusion peptides inside the host membrane. However, because of the broken symmetry, they may also be closely apposed to and, as a result, inserted into the viral membrane. During the entire process, if S5 breaks before any fusion peptide reaches the host membrane, the conformational change will pin the fusion peptides to the viral membrane and invert HA₂ with respect to its initial orientation, thus rendering it unable to participate in fusion.

Fig. 5.

Schematic picture of the HA dual-funneled energy (Upper, solvent-averaged energy) and free-energy (Lower, solvent-averaged energy plus conformational entropy) landscape. Microscopic Hamiltonian mixing means equally considering all contacts from HA_2I and HA_2F at all times. This has two main consequences: (i) the energy landscape contains a local energetic minimum with S5_I and S1_F formed (basin F in Fig. 2), and (ii) the kinetic barrier to leaving the HA_2I basin is reduced by cracking induced by competition between HA_2I and HA_2F contacts (detailed in Fig. 3). On such a landscape, the unfolding of S5_I becomes decoupled from forming the HA_2F coiled-coil, allowing the S5_I unfolding barrier to delay the transition. We hypothesize that a delay is functionally important because the fusion peptides require sufficient time to interact with and insert into a membrane.

A previously suggested “spring-loaded” model of HA function explains that HA₁ removal releases HA₂ from a metastable compact fold and that formation of the extended coiled-coil enthalpically drives HA₂ toward a conformation that facilitates interaction with the host (21). Our presented mechanism shares similarities with this picture. First, the HA_2I conformation is observed to be metastable because HA₁ removal causes an immediate structural rearrangement of HA₂. Second, an extended coiled-coil kinetic intermediate has significant population (basin F in Fig. 2). The key difference lies in the molecular mechanism of the transition. Instead of the coil-to-helix transition of S2 mechanically driving the molecule away from the HA_2I structure in the “spring-loaded” model, our simulations suggest that the disorder in Loop3-4 (Fig. 3) and subsequent symmetry breaking of S4 (Fig. 2, basins C, E, and F) disrupts the binding sites for the fusion peptides and expedite their release. Here, the dual-basin nature of S2 does not provide a strong bias toward the coiled-coil S2_F structure, leading to a kinetic intermediate where the HA₂ arms (S1 and S2) are disordered (basin E in Fig. 2). In this intermediate, the symmetry breaking in S4 bends HA₂ toward the viral membrane, localizing fusion peptides more closely to the viral membrane. The threefold symmetry breaking has been previously proposed as the last stage of conformational transition for HA₂ and other retrovirus fusion proteins, which serves to bring the host and viral membranes closer together (3, 32). However, our results indicate that this event occurs earlier in the conformational transition than previously suggested. Coupled with previous work, which suggests that the fusion peptides are able to insert into both the host and viral membrane (11, 33, 34), this allows a fusogenic configuration of HA₂ where the three fusion peptides are divided between the viral and host membranes. When S5_I breaks, the refolding of S5_F cooperatively aids in formation of the HA_2F coiled-coil. This cooperative route is especially interesting when considering that the free energy released during the HA₂ conformational transition can be used to assist in the subsequent membrane fusion (22). Because the HA_2F coiled-coil is not formed when S5_I breaks in the cooperative route, the free energy gained by forming its well-packed hydrophobic core can still be available during fusion. This is why we distinguish between the sequential and cooperative route: any free energy gained by refolding S5 is always available for fusion, but it can also be cooperatively added to the free energy gained by formation of the coiled-coil.

Our description leads to predictions that can be experimentally tested. For example, the increase of temperature will contribute to increased fusion activity until a point where the timescale of S5_I dissociation ${\overline{\tau }}_{\text{S}5}$ is comparable to the timescale of fusion peptide insertion into the host membrane. As previously mentioned, S5_I breaking apart before any fusion peptide has inserted into the host can lead to an HA₂ with its fusion peptides pinned to the viral membrane, which is likely inactive. Therefore, we expect fusion activity to level off—or even decrease—as the temperature is increased beyond this threshold. The stability of S5 can also be tuned with denaturants, and addition of denaturant may be a more accessible experimental parameter. Mutagenesis of S5 to yield S5_I of different stabilities can also be used to modulate the rate of S5_I unfolding. The simulations highlight a subset of contacts that, when broken, are causally related to the unfolding of S5_I (SI Appendix). Additionally, either single molecule or ensemble FRET could be used to examine the kinetics of S1 coiled-coil formation. In our suggested scenario, there would be a double exponential distribution for the rate of S1 association, one representing the S1 trimer forming in a spring-loaded manner, and another representing S1 association assisted by interactions with S5.

In conclusion, we have presented a minimally frustrated all-atom model to explore the extensive conformational rearrangement of influenza HA₂. The dual-basin landscape of HA₂ shows that native interactions from the HA_2F crystal structure cause metastability of the HA_2I conformation. Local disorder transitions in Loop3-4 present a mechanism for the release of the fusion peptides, which shows HA₂ to be another example where local disorder enables a biomolecular functional motion (28). The interplay between the energetic stability of the S5_I trimer and the entropic barrier toward S1_F creates a quasi-stable kinetic intermediate, which in turn allows for parallel routes in the HA₂ conformational transition: a sequential route and a cooperative route, where the latter suggests a similar role for the HA₂ coiled-coil as in membrane fusion by the SNARE complexes. This understanding provides new insight into the function of influenza HA and other class I viral fusion proteins, which are thought to operate by a similar mechanism (35).

Materials and Methods

To study the transition, the principles of energy landscape theory were implemented through a structure-based model (SBM) (i.e., explicitly encoding the HA_2I and HA_2F crystal structures as deep energetic minima in the simulation Hamiltonian). An SBM reduces the complexity of molecular interactions by setting interactions between atoms that are close in the crystal structure as attractive and ignoring all others. The Hamiltonians for the two structures can be combined into a multibasin SBM through a global (macroscopic) mixing function (26) or combined locally (microscopic) by including all interactions present in both structures simultaneously (19, 36, 37). We adopted the local model for two reasons: (i) the endpoints are so disparate that in a macroscopic mixing model the barrier would be insurmountable, and (ii) we find that locally competing contacts serve as the driving force to facilitate the transition. The set of native contacts, atom pairs that are in proximity in the native structure, is called a contact map and uniquely defines the atom pairs given attractive interactions. Here, we used the “Shadow” criterion to define native contact maps for both the HA_2I and HA_2F structures (38). The dual-basin SBM contact map was created by combining all native contacts from both conformations and giving each contact equal strength (19). All other pair interactions were modeled as hard spheres to maintain excluded volume. Langevin dynamics was used to sample the SBM. A complete description of the dual-basin SBM and simulation protocols is given in the SI Appendix.

HA₁ was not represented in the simulations, despite the disulfide bond that anchors it to HA₂. Our rationale is based on the observation that the primary role of HA₁ during the conformational transition is simply to prevent HA₂ from undergoing premature conformational change. It is known that the HA₂ trimer without HA₁ forms the postfusion structure (22, 39) and that HA₁ undergoes no conformational change at low pH (7). The tendency for HA₁ to dissociate at low pH can be easily understood from the high isoelectric point of the HA₁ (pI∼8) compared with HA₂ (pI∼5) (40). Therefore, for simplicity, we removed HA₁ from the model.