Rotation-Activated and Cooperative Zipping Characterize Class I Viral Fusion Protein Dynamics

Abstract

Class I viral fusion proteins are α-helical proteins that facilitate membrane fusion between viral and host membranes through large conformational transitions. Although prefusion and postfusion crystal structures have been solved for many of these proteins, details about how they transition between these states have remained elusive. This work presents the first, to our knowledge, computational survey of transitions between pre- and postfusion configurations for several class I viral fusion proteins using structure-based models to analyze their dynamics. As suggested by their structural similarities, all proteins share common mechanistic features during their transitions that can be characterized by a diffusive rotational search followed by cooperative N- and C-terminal zipping. Instead of predicting a stable spring-loaded intermediate, our model suggests that helical bundle formation is mediated by N- and C-terminal interactions late in the transition. Shared transition features suggest a global mechanism in which fusion is activated by slow protein-core rotation.

Introduction

Viruses continue to pose an important biomedical problem because of the perennial health hazards they present to humans. Although we have made significant progress vaccinating and treating many viral diseases, others remain leading causes of death around the world despite extensive research. Development of effective treatments has remained challenging, in part because of the fact that underlying biophysical principles of viral infection remain poorly understood (1). Although the life cycle of a typical virus contains many complex steps, ranging from cell entry and replication to eventual lysis and release of progeny virions (2), this study focuses only on the physical processes by which a virus recognizes and enters its host cell.

Many envelope viruses are structurally comprised of similar components (3). The interior of the virus contains RNA/DNA, nucleoproteins, and polymerases required for replication inside the host cell. These are surrounded by matrix proteins that form a protective capsid around the genetic material. The capsid is covered by a lipid bilayer and a set of surface proteins, which perform a range of functions from host cell recognition to receptor cleavage. For all envelope viruses, one of these proteins is responsible for fusing the viral and cell membranes (4), allowing the viral genetic material to enter the cell. These proteins are critical for the virus’s ability to infect the host (5), and they are promising targets for inhibitor design and therapeutic development.

Viral fusion proteins are categorized into one of three classes, which typically differ in their secondary structure composition (6) and can usually be grouped in the following way: class I proteins are mostly α-helical, class II are mostly β-sheet-like, and class III proteins contain a balance of both helices and β-sheets. Well-known examples of viruses with class I proteins include influenza and human immunodeficiency virus (HIV), class II includes dengue and zika viruses, and class III includes the vesicular stomatitis virus and herpesvirus. The most biophysically studied of these proteins is influenza hemagglutinin, a system for which there have been many mutational (7, 8) and kinetic measurements performed (9, 10, 11).

Class I viral fusion proteins are transmembrane homotrimers that are formed as a single chain and later cleaved into three sets of two domains. One of these trimeric domains is typically responsible for host cell recognition and specificity (the HA1 domain in influenza and the F2 domain in paramyxovirus), whereas the other is responsible for mechanically fusing the viral and host lipid membranes (12) (the HA2 domain in influenza and the F1 domain in paramyxovirus). The N-terminal residues of the fusion domain, created by domain cleavage (13), mark the so-called fusion peptides, highly conserved hydrophobic peptides whose purpose is to embed in the host membrane (14). The C-terminal residues form a transmembrane domain that spans across the viral membrane.

Each protein initially exists in a metastable prefusion structure (5, 6) that is much higher in free energy than the postfusion ensemble. An initial triggering event, ranging from changes in pH to interactions with other proteins, then releases the fusion peptides from their initially protected position (15). This chemical trigger causes the protein to refold into a fusogenic configuration by destabilizing the initial structure (e.g., by changing charge states of certain titratable residues), allowing the protein to diffuse out of its metastable basin. Regardless of differences in sequence and structure, all viral fusion proteins undergo massive conformational changes (16, 17) to promote fusion between the viral and host membranes, in many cases almost completely refolding (10). Studies have shown that interrupting this conformational change either greatly inhibits or abolishes the virus’s ability to fuse with the host membrane (8, 18, 19).

Despite the importance of these proteins in the viral life cycle, there is precious little structural information available about the majority of them. In the cases for which there exist nearly complete prefusion and postfusion structures, little is known about the mechanisms or pathways which connect them. Furthermore, the detailed coupling between these conformational motions and the bilayer fusion itself remains a mystery.

In this study, we use molecular dynamics simulations to elucidate broad features of how several class I fusion proteins transition from their prefusion to postfusion configurations and how these motions may contribute to membrane fusion between the virus and host. The timescales involved during viral infection are relatively long by molecular dynamics standards. Obtaining thermodynamic sampling of the full viral-mediated membrane fusion is currently not computationally feasible with detailed fully atomistic simulations. Although membrane simulations using smaller peptides have been successfully carried out (20), these proteins are spatially extended trimers with hundreds of residues per monomer; even the isolated conformational transitions with no lipids present are intractable by standard all-atom sampling techniques. Instead, we use a structure-based model to achieve the timescales necessary for these biological processes.

Previously, structure-based models (21, 22) (SBMs) developed using the principles of energy landscape theory (23, 24, 25) have been used extensively in understanding protein folding (26). Protein sequences have evolved to be minimally frustrated (25); although these minimally frustrated models do not contain explicit sequence information, they have been shown to reproduce transition state ensembles and folding mechanisms well (27, 28) and are computationally much less expensive than approaches using fully atomistic force fields. The largest drawback to this modeling approach is that information about the native structure of the protein is required to construct the potential. This restricts the systems that can be studied to those for which we have experimental structural information. Values for numerical constants and further details about the potential can be found in the Supporting Material.

SBMs have also been used to analyze transitions between different folded configurations, including previous examples in kinases (29), molecular motors (30, 31), and the ribosome (32, 33). Motivated by early models of influenza A (34), we employ a similar coarse-grained approach to study conformational dynamics of the whole family of class I viral fusion proteins for which there are overlapping prefusion and postfusion structures: influenza A and B hemagglutinin, parainfluenza (PIV) F, and respiratory syncytial virus (RSV) F. We construct a potential that contains both the prefusion and postfusion structures as explicit energy minima and neglect nonnative interactions. Under such a potential, the postfusion structures are naturally lower in energy than the prefusion because they are better packed and have a larger number of native contacts. Fusogenic conditions such as low pH or hyperosmality can influence factors, including helical propensity or interactions with titratable residues; these effects are implicitly captured by the SBM because they are reflected in the native potential of the postfusion structures themselves. Previous work in influenza hemagglutinin (34) has demonstrated that modulating similar parameters can impact local kinetics slightly but does not change the global transition mechanism. We initialize each simulation in a state representative of a time very shortly after the trigger to begin fusion destabilizes the proteins from their metastable configurations and observe their time evolution.

It has been suggested that because members of this protein class all have similar structural motifs in their postfusion structures (i.e., the so-called “trimer of hairpins” (35)), they likely also share some mechanistic features. Therefore, a challenge in this area is to uncover the underlying mechanism by which they transition and fuse the virus and host. By analyzing the structural ensembles resulting from our simulations, we can discover these common motifs and determine which features are generic and/or robust across the entire class. Although we build an individual model for each different viral protein, we find that there is one dominant global transition mechanism that is robust across several distinct viruses and is likely to extend to many other structurally similar fusion machines.

Methods

For each member of the protein family, we created a dual-basin SBM with energy minima at both the experimentally determined prefusion and postfusion configurations. Prefusion/postfusion crystal structures were taken from Protein Data Bank IDs 2HMG/1QU1 (17, 36), 3BT6/4NKJ (37, 38), 4GIP/1ZTM (39, 40), and 4JHW/3RRR (41, 42) for influenza A hemagglutinin (HA), influenza B HA, PIV F, and RSV F, respectively (Fig. 1). Multiple sequence alignment (43) was performed for each set of structures to determine the maximal overlapping set of residues for which there were coordinates recorded. Because the fusion peptides are not present in any postfusion crystal structure and are likely disordered in water, they are necessarily absent from our model. Any missing loops were structurally modeled using the SWISS-MODEL server (Biozentrum, University of Basel, Basel, Switzerland) (44, 45). We performed a steepest-descents energy minimization to remove steric clashes and placed a single bead per amino acid at the α-carbon position, as described elsewhere (21).

Figure 1.

Prefusion and postfusion crystal structures of the influenza A and B hemagglutinin (HA A/B), parainfluenza F (PIV F), and respiratory syncytial virus F (RSV F). Each protein is partitioned into several subdomains based upon changes in local structure over the conformational transition. S1 (yellow) marks N-terminal regions that remain helical during the transition but eventually form tertiary contacts to form a coiled coil. The fusion peptides (not pictured) are attached to the N-terminal of S1. S2 (blue) marks a loop-to-helix region. S3 (red) is structurally conserved during the transition. S4 (green) marks a region that inverts to become a turn in the postfusion structure. S5 (orange) is the C-terminal domain that is attached to the viral transmembrane peptides. In PIV F and RSV F, S6 (gray) is a portion of the F2 chain, the analog of which in the influenza viruses (HA1, not pictured) is thought to dissociate as a precursor to the conformational transition.

Both the prefusion and postfusion configurations must be thermodynamically stable and kinetically accessible at different stages in the viral life cycle. The prefusion structure is thought to be metastable (6) after a trigger that initiates the conformational change. It is known that the transitions are irreversible (5, 46) and the postfusion structures are lower in free energy. This is inherently captured in our model because the postfusion structures contain many more native contacts.

In a dual-basin SBM, every interaction stabilizes one of the two native folded states. Any energetic frustration in our model therefore arises as a result from competing interactions between the prefusion and postfusion landscapes. We construct an ordinary structure-based potential for both the prefusion and postfusion crystal structures of each protein. We then include both sets of native minima simultaneously for all angles, dihedrals, and native contacts. Angles and dihedrals are given a fourth order polynomial to smoothly interpolate between the minima. Bonds and angles for each structure are given by quadratic functions,

$$V_{bond}=\frac{1}{2}{k}_{bond}{\left(r-r_0\right)}^2,V_{angle}=\frac{1}{2}{k}_{angle}{\left(\theta -{\theta }_0\right)}^2,$$

whereas dihedral angles are given by

$$V_{dihedral}=\left[1-cos\left(\varphi -{\varphi }_0\right)\right]+\frac{1}{2}\left[1-cos\left(3\left(\varphi -{\varphi }_0\right)\right)\right].$$

Native contacts are defined using the Shadow Algorithm (47) with a 1.2 nm cutoff and are given attractive gaussian potentials (48)

$$V_{contact}=\mathit{ϵ}\left(1+{\left(\frac{{\sigma }_{NC}}{r}\right)}^{12}\right)\left(1-exp\left(-{\left(r-r_0\right)}^2\right)/\left(2{\sigma }^2\right)\right)-1.$$

The contact maps are symmetrized such that each monomer in a given protein trimer has the same set of native contacts. Contacts which exist in both the prefusion and postfusion structures, but with different native distances in each, are given a double well potential which allows them to adopt either configuration. Additionally, a planar harmonic restraint is placed on the C-terminal atoms in each monomer,

$$V_{membrane}=\frac{1}{2}{k}_{mem}{\left(z-z_0\right)}^2,$$

to mimic the effects of being tethered to the viral membrane by transmembrane domains. This model serves as a limiting case in which lateral diffusion through the viral membrane is free (high lipid diffusivity), but large scale bending motions of the membrane are energetically prohibitive. In addition, the Hamiltonian neglects any energetic contributions because of interactions with a host membrane (absent from the model).

Results

The postfusion ensembles should be thermodynamically stable to facilitate membrane fusion between the virus and host, i.e., the postfusion state should not contain significant unfolded content under physiological transition conditions. However, the simulation temperature should not be greatly below the folding temperature to avoid local traps and glassy dynamics. From previous experience (34), we find that about 80% of the melting temperature of the postfusion state is a reasonable choice that satisfies these biological constraints. We simulate kinetic transitions instead of a full thermodynamic analysis of each landscape because the conformational transitions are inherently kinetic in vivo. Backward transitions never occur biologically, so any such trajectories are irrelevant for virus-host membrane fusion.

We performed 100 kinetic simulations for each of HA A/B, PIV F, and RSV F, which were initiated in the prefusion crystal structure and were terminated when they reached the postfusion ensemble. 100 runs were chosen as a reasonable number for which removing half the dataset and recalculating observables did not significantly impact global probability distributions. After thermal fluctuations rapidly move the protein away from the metastable prefusion basin, refolding proceeds roughly downhill to the postfusion structure. Analysis of these trajectories shows that the overall order of events, as parameterized by relative local structure formation, is preserved across all proteins studied (Transition Mechanism between Prefusion and Postfusion Structures Is Conserved across Proteins). Initially, the N-terminal domains lose their interface contacts, freeing the fusion peptides to search for a membrane. This is followed by prefusion C-terminal unfolding. A structurally conserved core observed in all viral fusion proteins then slowly rotates as the termini remain unfolded and must reach a critical rotation angle of roughly 90°. Once this rotation criterion has been met, the protein reaches an activated state in which late N- and C-terminal interactions are easily accessible. This diffusive rotational search also causes a time delay between fusion peptide release and final refolding into the postfusion structure, crucially allowing the fusion peptides sufficient search time to find a membrane before the transition completes (Discussion). We show that these long-distance N- and C-terminal interactions are critical to the stability of the full postfusion structure, and we observe no long-lived (49) coiled-coil structure in their absence (C-Terminal Interactions Stabilize the Spring-Loaded Loop-Helix Coiled-Coil). When this rotation has been satisfied, the termini can cooperatively zip together to form a stable helical bundle (Rotation of the Structurally Conserved Domain Is Required before N- and C-Terminal Zipping), bringing the viral and host membranes together and leading to fusion. Each of these features is described in detail below.

Transition mechanism between prefusion and postfusion structures is conserved across proteins

For analysis, we partition each protein into several domains based on local structural changes between the prefusion and postfusion structures (Fig. 1), labeled S1–S5 and colored yellow, blue, red, green, and orange for visualization. S1 corresponds to N-terminal helical domains that come together to form part of a trimeric three-stranded coiled-coil but initially have no tertiary interactions with each other. The fusion peptides, hydrophobic regions that attach to the host membrane and are required for successful membrane fusion, are connected to the N-termini of S1. S2 is a region that begins as a disordered loop and undergoes a loop-helix transition to form most of the central helical coiled-coil. This region has been traditionally called the “spring-loaded” domain in studies of influenza HA. S3 is defined as the portion of the protein that is structurally conserved during the transition. S4 forms a turn region where each chain refolds and reverses direction along the central axis relative to its initial orientation. S5 is connected to the C-terminal viral transmembrane region and eventually wraps antiparallel along the grooves of the coiled-coil in the postfusion structure. The RSV F and PIV F structures also have a sixth domain, S6, which is absent in the influenza proteins; this is due to the fact that the HA1 chain dissociates and is thought to not play a role in the conformational transition, but the F2 chain in RSV and PIV remains locally bound to the F1 fusion chain.

In traditional protein-folding studies, the fraction of native contacts Q has been shown to be a powerful reaction coordinate that can clearly distinguish between transition state ensembles and reaction endpoints (50). It is frequently used to measure overall folding progress. To get more detailed information about local changes in structure as the transitions progress, we define the fraction of native contacts Q of each subdomain S1–S5 for both configurations of each protein, $Q_{S1}-Q_{S5}\left(prefusion\right)$ and $Q_{S1}-Q_{S5}\left(postfusion\right)$. We also define a global measure of progress along any given transition, $Q_{transition}=a{Q}_{postfusion}-b{Q}_{postfusion}$, where $Q_{pre/postfusion}$ are the total fraction of native contacts for their respective structures and a/b are constants chosen such that the postfusion and prefusion crystal structures have values of $Q_{transition}=\mathrm{\pm }1$, respectively. We then plot these coordinates against each other to evaluate formation and loss of local structural elements as a function of overall transition progress.

A typical transition trajectory for any protein begins in the prefusion crystal structure and proceeds through the following steps. First, the N-terminal domains S1 and S2 lose their prefusion interface contacts with the rest of the protein. Loss of these contacts pulls S1 and S2 away from the center of the protein and releases the hydrophobic fusion peptides connected to the N-terminal of S1, which allows them to begin a diffusive search for a bilayer. Next, the C-terminal domains S4 and S5 unfold. This leads to a transient structural ensemble in which S1, S2, S4, and S5 are all relatively disordered and contain no significant structure compared to either endpoint state. Only the conserved domain S3 remains folded, allowing the rest of the protein to begin refolding around it.

S4 then folds backward and forms the characteristic “trimer of hairpins” structure, found in many viral fusion proteins, by packing against the outside of S3. On average, S1, S2, and S5 still have a low fraction of native contacts formed (Fig. 2), signifying that the central coiled-coil does not often form before hairpin formation. Finally, S5 refolds, guiding S1–S2 into their native postfusion structures by forming interfacial contacts with them and completing the conformational transition. This nucleation of the central coiled-coil by interface contacts with the C-terminal domain is consistent with results from a previous model of influenza HA and appears to be a robust phenomenon across many class I viral fusion proteins.

Figure 2.

Local fraction of native contacts $Q_{subdomain}$ as a function of global transition progress, $Q_{transition}=a{Q}_{postfusion}-b{Q}_{postfusion}$. $Q_{pre/postfusion}$ are the total fraction of native contacts for their respective structures and a/b are normalization constants such that $Q_{transition}=\mathrm{\pm }1$ in the postfusion and prefusion crystal structures, respectively. Statistics for each protein are taken from 100 kinetic simulations that are initialized with each protein in the prefusion crystal structure and terminated when the protein reaches the postfusion basin. Dashed lines correspond to values of $Q_{transition}$ where the gap between the amount of local structure formation between different domains is largest. 1) For each protein, we find that the overall order of events is qualitatively preserved. S1–S2 lose their prefusion contacts, many of which are interface contacts with S3. 2) This corresponds with the release of the fusion peptides, attached to the N-terminal of S1. S4–S5 unfold from their prefusion configurations. 3) This creates an ensemble of intermediate configurations in which S1, S2, S4, and S5 have very little of either their pre- or postfusion structure formed. 4) S4 refolds, forming postfusion contacts with S3 and creating a trimeric hairpin. After this hairpin is established, S5 begins refolding and eventually cooperatively forms native postfusion contacts with S1–S2, completing the transition.

C-terminal interactions stabilize the spring-loaded loop-helix coiled-coil

It has been suggested (49, 51) that the transition pathway for class I fusion proteins proceeds through a set of structured intermediates in which the full postfusion central coiled-coil $\left({\text{S}1-\text{S}3}_{postfusion}\right)$ is formed first and the C-terminal loops $\left({\text{S}5}_{postfusion}\right)$ later fill in the grooves of this coiled-coil to bring the two termini (and thereby two membranes) together. The reaction coordinates used to track the global transition shown above are too general to clearly distinguish this hypothesis from any alternate mechanisms because they do not distinguish between S2–S2 and S2–S5 interactions. This prompts the creation of more detailed coordinates to probe the interplay between the coiled-coil and C-terminal domains: $Q\left(S2-S5Interface\right)$ measures the fraction of postfusion contacts formed between S2 and S5, whereas $Q\left(S2Coiled-Coil\right)$ measures only the fraction of helical bundle contacts formed inside the spring-loaded region S2. The probability distributions of these two variables show that all four proteins have similar features under these coordinates (Fig. 3).

Figure 3.

Probability distributions of S2 loop-helix coiled-coil formation versus interfacial contacts formed between S2 and C-terminal domain S5. (A) 2-D histograms giving $-ln\left(P\right)$ of native contacts formed in the postfusion three-helix bundle versus N- and C-terminal S2–S5 interface contacts are shown. X marks on each plot denote starting and ending points in each trajectory (prefusion bottom left, postfusion top right). In all cases, the central coiled-coil fluctuates around an ensemble of structures with a low fraction of helical contacts until there are sufficient C-terminal interactions to lock the helical bundle in place. Some backtracking along $Q\left(S2Coiled-Coil\right)$ may be required to proceed with the transition if too much of the coiled-coil forms early relative to S5 refolding. (B) Probability distributions of $Q\left(S2Coiled-Coil\right)$ at $Q\left(S2-S5Interface\right)=0$ are given. None of the proteins ever reach a stable and fully native-like helical bundle structure without interfacial interactions with S5. The influenza proteins show a broader spread in S2 native contact formation than either paramyxovirus, suggesting that a more fluid unfolding/refolding is occurring for their S2 domains.

For all four viral proteins, the loop-helix region S2 has no long-lived three-helical bundle structure in the absence of C-terminal (S5) interactions. This provides evidence that these S2–S5 interfacial interactions are crucial in nucleating the formation of a stable, central coiled-coil, which carries implications for the overall mechanism of membrane fusion (see Discussion). Further, we show that within our model some backtracking may be required if too much of the coil is formed too early in the transition, particularly in the influenza A and RSV fusion proteins. This backtracking is likely a result of the geometric constraint imposed because the C-termini are anchored to the viral plane. We have also analyzed the distribution of coiled-coil contacts formed at $Q\left(S2-S5Interface\right)=0$ (Fig. 3 B). Three main features emerge: 1) none of the S2 domains maintain a postfusion native-like helical bundle structure in the absence of S2–S5 interface interactions; 2) the two influenza proteins have a broader range of helical content transiently forming and unforming than the paramyxovirus proteins; and 3) the RSV distribution peaks around a noticeably higher $Q\left(S2Coiled-Coil\right)$ value than the other proteins studied. Analyzing the structures within this peak shows a small structural domain at the base of the bundle in RSV that has a high propensity to form a stable helical bundle at an early stage in the transition when compared with the other fusion proteins.

Rotation of the structurally conserved domain is required before N- and C-terminal zipping

To emulate the presence of the viral transmembrane domains, which serve as anchors in vivo but are structurally absent from our simulations, we restrain the C-termini of each protein to a plane by a harmonic potential (See Methods). In the postfusion structure of every protein, the N- and C-termini are adjacent to each other. To bring the termini together while accommodating this anchoring transmembrane constraint, each molecule must undergo some rotation about its center of mass as it proceeds through the conformational transition (Fig. 4). We introduce two more reaction coordinates to track how terminal separation changes with respect to this necessary rotation: $R_{N-C}$ is given by the mean difference between the N- and C-terminal heights from the plane of the viral membrane ($Z_N-Z_C$), and ${\theta }_{S3}$ is given by the angle between the cylindrical axis of S3 and the viral membrane plane. Because the fusion peptides eventually embed in the host, $R_{N-C}$ serves as a rough proxy for viral-host membrane separation in a successful fusion event. S3 is chosen as the region to calculate this rotation angle with respect to the viral membrane because it is the only domain which stays fully folded throughout the entire transition and is the only region for which the prefusion and postfusion structures are identical.

Figure 4.

Rotation of the structurally conserved domain is required for N- and C-terminal zipping. (A) The choice of reaction coordinates is as follows: ${\theta }_{S3}$ gives the angle between the axis of S3 and the plane of the viral membrane. $R_{N-C}$ gives the difference in height between the N- and viral-anchored C-termini. (B) Probability distributions of ${\theta }_{S3}$ and $R_{N-C}$ are given. X marks denote the prefusion and postfusion basins, with each trajectory starting middle left and ending in the bottom right. The black contour shows the highest probability value of $R_{N-C}$ for each ${\theta }_{S3}$ with arrows showing direction. The N- and C-terminal separation becomes maximal before S3 rotates significantly, corresponding to the fusion peptides projecting away from the viral membrane and toward the host. S3 then gradually rotates toward the plane of the virus, whereas N- and C-terminal separation does not change substantially. After a critical rotation angle has been reached, the termini can “zip” together with minor change in angle required. (C) Representative typical structures for each protein with slightly sub-critical ${\theta }_{S3}$ just before the termini come together and complete the transition are shown. C-termini are highlighted with purple and N-termini are highlighted with black dots. The S4 hairpin structure is always formed in these structures (green-red interaction), but S1, S2, and S5 (yellow, blue, and orange) are frequently disordered and there is no long helical bundle formed.

Because the S1–S2 prefusion contacts are the first ones to be lost for each protein (Fig. 2), N- and C-terminal separation grows rapidly before S3 significantly changes orientation. These contacts initially keep the N-termini near the center of the protein to protect the fusion peptides. This initial growth in N- and C-terminal separation is followed by a gradual rotation in the conserved domain S3 (Fig. 4), bringing the central axis of the protein slowly closer to parallel with the viral membrane. During this gradual rotation, the remaining domains of the protein remain unstructured and contain little native content from either the prefusion or postfusion crystal structure (Fig. 2). N- and C-terminal separation decreases slightly as S3 rotates toward the viral membrane because of steric restrictions, and, on rare excursions, the N-termini may even stochastically approach the viral surface. We find that until a certain rotation angle in S3 has been satisfied, the termini cannot come together, and the transition cannot complete. The value of ${\theta }_{S3}$ before N- and C-terminal apposition can occur is always roughly ${\theta }_{S3}\approx 90\mathrm{°}$, with the exception of influenza B, in which it is $\approx 60\mathrm{°}$. This difference is attributed to the fact that the crystal structure used for influenza B HA has a relatively smaller S5 domain than the other studied systems. This is likely to be an artifact of these missing residues instead of a real difference in biological mechanism. After this critical rotation has been satisfied, the termini come together quickly with very little change in ${\theta }_{S3}$. This coincides with S5 forming its postfusion contacts with S1–S2, nucleating the central helical bundle and finishing the transition (Fig. 3). It is only at this stage in a typical transition that the loop to helix region S2 maintains a stable coiled-coil structure. After reaching this structure, each protein undergoes thermal fluctuations about the postfusion structure but never transitions backward from the postfusion ensemble. This indicates that our model correctly predicts that the postfusion structure is much deeper in free energy, although we cannot precisely quantify its value.

Discussion

Common transition features resulting from our model suggest a general mechanism for viral-host fusion mediated by any class I viral fusion protein (Fig. 5). Because the model does not contain any explicit details about the sequence, the dominant transition features appear to be driven by overall topological effects and not by sequence-specific details. Fusion is initiated by a triggering event that signals the beginning of the conformational transition and allows the protein to leave its metastable prefusion basin. Refolding from the prefusion to postfusion structure then proceeds through the following stages. First, the N-terminal regions (S1–S2) break their interface contacts with the rest of the protein and rapidly extend away from the viral surface, allowing the hydrophobic fusion peptides to diffusively search for a membrane. After this rapid N- and C-terminal separation, the viral-anchored C-terminal domains (S5) unfold and start to refold into their postfusion structures. The conserved core of the protein (S3) begins a gradual rotation toward the viral membrane. This diffusive rotational search provides a crucial separation in timescales between fusion peptide release and eventual nucleation of the central postfusion coiled-coil, allowing the peptides to find a membrane much earlier than the typical transition completion time. During this phase, the proteins explore a set of disordered configurations in which some fusion peptides are inserted in the host but both the N- and C-terminal regions remain mostly unstructured. This ensemble of structures effectively serves as a “waiting” state in which the N- and C-termini (and host/viral membranes) cannot be brought together until the protein core rotates beyond a required threshold. When this rotation activation is satisfied, the global conformational transition can be completed as the C-terminal domain S5 nucleates the formation of the central coiled-coil and rapidly “zips up” the protein into its final bundled configuration. The free energy released during the final N- and C-terminal zipping drives the two membranes together when the transition completes. Lipid and eventual content mixing can proceed from here.

Figure 5.

Schematic mechanism for class I viral membrane fusion. The protein in the prefusion structure at the start of the transition is shown. The N-terminal domains lose their contacts and project away from the viral membrane. The N-terminal-attached fusion peptides begin searching for a bilayer to embed in. C-terminal domains unfold, and the conserved core of the protein begins to rotate toward the virus. The characteristic hairpin structure is formed, and gradual rotation of the central domain continues. The central coiled-coil is not formed and only contains transient helical content, and the C-terminal domain remains mostly disordered. After a critical rotation ($\approx 90\mathrm{°}$) of the conserved domain is reached, the protein reaches an activated state in which the C-terminal domains quickly nucleate formation of the coiled-coil. This nucleation zips the N- and C-termini and their attached membranes together, completing the conformational transition.

These simulations provide direct evidence that varied examples of class I viral fusion proteins all share common dynamical features during their structural transitions. In this new mechanistic view, the transiently disordered N- and C-terminal intermediate states that we predict contrast with the standard spring-loaded picture (49) of class I membrane fusion. In that view, the loop-helix transition in S2 happens very early in the fusion process and remains stable indefinitely as the C-terminal domains slowly refold around a rigid rod. The proposed mechanism for fusion via late nucleation of the coiled-coil is appealing because the free energy released in forming the central helical bundle can be used toward bringing the viral and host membranes together. Alternatively, if this bundle is formed early in the transition, as in the spring-loaded model, the free energy of formation cannot impact any membrane dynamics and is effectively wasted. This helical bundle was thought to be crucial for initially projecting the fusion peptides away from the virus. We have shown that even in the absence of stable helical content (Fig. 3), the fusion peptides can go over 100 Å away from the viral membrane (Fig. 4) with very high probability. It is known that more than one protein is typically required for successful fusion (52, 53). Therefore, the flexibility afforded by the disordered collapsed states found in our model allows for several proteins to aggregate in a small region. These intermediate waiting states also suggest that a promising general therapeutic avenue may be to disrupt interactions between the C-terminal domain (S5) and loop-to-helix region (S2). Disruptions of this region have already been shown to successfully inhibit HA-mediated fusion (54).

One important caveat to the ensembles resulting from these simulations is that they result from transitions in which we have neglected the presence of any host cell. Therefore, our model cannot currently account for any effects that the fusion protein-host bilayer interactions may create on the structural pathways. In particular, there have been some recent cryo-electron microscopy results that show HA molecules bridging the gap between virus and host in extended intermediates of various lengths (55). Our results are consistent with such a picture for several reasons. The current resolution on these imaging techniques is such that local secondary structure formation at any given moment is difficult to determine. Therefore, these extended stalks could contain any nondetermined amount of nonhelical content that is nucleated later by C-terminal tertiary interactions. Another possibility is that tension imparted by the host membrane pulling on the protein causes stable helical conformations to become energetically favorable, an effect which would inherently be missing from our current simulations. Teasing out the interplay between the structure of the protein and virus-host interactions remains an interesting area of future research and can help resolve some of these questions.

The fusion peptides themselves play no essential role in the protein’s global conformational transition. Therefore, if the irreversible transition is initiated but the fusion peptides never find a host bilayer, we expect that reaching the postfusion configuration is still possible and even favorable in free energy. This is suggestive of a common inactivation pathway for these proteins. It is known that incubating influenza HA at low pH for too long in the absence of a host abolishes its ability to cause membrane fusion (56). This observation appears to be consistent with our model. After the initial trigger (such as the pH drop in influenza), we see that each protein has a natural tendency to rotate toward the viral membrane over time to satisfy native postfusion contacts (Fig. 4). If the rotation activation criterion gets satisfied but no fusion peptide has come in contact with the host, the C-terminal domains should still nucleate the coiled-coil and conclude the transition. This implies a final configuration in which the protein has adopted its postfusion structure but is inverted with respect to its initial orientation on the viral surface (34, 57, 58). This appears to be the dominant configuration that our model predicts in the absence of a host bilayer. If the conformational transition ends with the protein in such a state, all fusion peptides are effectively pinned to the viral surface. These functional motions are both required for membrane fusion and irreversible. Therefore, such a “pinned” protein could never actively contribute to infection.

We have explored conformational dynamics of several different viral fusion proteins using coarse-grained models. Our results demonstrate that there are common structural mechanisms which are shared by all proteins during transitions from their respective prefusion to postfusion states. These theoretical studies present a general mechanism for class I viral membrane fusion mediated by surface proteins. In this model, late cooperative N- and C-terminal zipping is the primary driving force behind fusion. This computational study includes a diverse set of viral proteins and provides insight on the mechanistic commonalities between them, a picture that differs from the classical spring-loaded image of the class I viral fusion mechanism. Other viruses have different fusion proteins that can differ significantly in structure and overall topology (5, 6). We expect some features of this mechanism to hold generally across more viral systems than the examples studied here. However, we do not yet know which aspects are robust from virus to virus and which are specific to members of class I. Although our model uses a minimal approach to describe the viral membrane and ignores host interactions, it appears to be sufficient to extract the global mechanism during these structural transitions. In principle, these interactions could impart forces that modulate the dynamics of the viral proteins themselves, and quantifying these forces could address questions such as the number of proteins required at a fusion site to successfully open a pore. It remains an interesting open challenge to rigorously examine how these conformational changes couple with detailed lipid dynamics and obtain a complete picture of viral mediated bilayer fusion.

Author Contributions

N.R.E. and J.N.O. designed research. N.R.E. performed research. N.R.E. analyzed data. N.R.E. and J.N.O. wrote the manuscript.

Editor: Elizabeth Komives.