VISUALIZING INTERMEDIATE STAGES OF VIRAL MEMBRANE FUSION BY CRYO-ELECTRON TOMOGRAPHY

Abstract

Protein-mediated membrane fusion is the dynamic process where specialized protein machinery undergoes dramatic conformational changes that drive two membrane bilayers together, leading to lipid mixing and opening of a fusion pore between previously separate membrane-bound compartments. Membrane fusion is an essential stage of enveloped virus entry that results in viral genome delivery into host cells. Recent studies applying cryo-electron microscopy techniques in a time-resolved fashion provide unprecedented glimpses into the interaction of viral fusion proteins and membranes, revealing fusion intermediate states from the initiation of fusion to release of the viral genome. In combination with complementary structural, biophysical, and computation modeling approaches, these advances are shedding new light on the mechanics and dynamics of protein-mediated membrane fusion.

PROTEIN-MEDIATED MEMBRANE FUSION DURING ENVELOPED VIRUS ENTRY

Membrane fusion is the dynamic process of merging two lipid membranes and opening a conduit that allows cargo to be transferred between previously separate membrane-bound compartments. This process underlies many essential biological processes including synaptic vesicle signaling, cell-cell fusion during placenta and myoblast formation, multi-cellular organismal development, regulation of mitochondrial activity, and gamete fertilization, to name a few. Additionally, membrane fusion has been implicated in pathological events including various aspects of cancer progression and enveloped virus entry into host cells.

In this review, we examine recent advances in understanding the function of fusion proteins on the surface of enveloped viruses that mediate membrane fusion to initiate infection of host cells. Fusion proteins on enveloped viruses are canonically grouped into three structural classes I-III (Figure 1) [1-3]. While the number of fusion protein complexes, their organization, and their structures vary dramatically among viruses, they perform the common function of driving the merging of virus and host cell membranes to open a pore through which the viral genome is deposited into the host cell’s cytosol. A longstanding goal has been to characterize the structural transitions that fusion proteins undergo and determine how they are coupled to membrane remodeling, lipid mixing, and fusion pore (see Glossary) formation (Box 1); however, until recently, this has been challenging to visualize in intact fusion systems by most experimental approaches (Box 2). Understanding these processes at a structural and mechanistic level can inform strategies to block virus infection or design and engineer novel membrane-manipulating machines [4], as well as advanced vaccines [5].

Figure 1. Fusion proteins in enveloped viruses are grouped into three classes.

(A) Class I exemplified by HIV-1 Env glycoprotein, a homotrimer of gp120/gp41 receptor binding/fusion subunit heterodimers, which sparsely decorates the surface of immature virions shown here and interacts via its TMD and cytoplasmic tail with the internal Gag polyprotein lattice [11,119]. (B) Class II represented by Chikungunya alphavirus with homotrimers of E1/E2 fusion protein/accessory protein dimers arrayed with icosahedral symmetry [11,45,119]. Alphavirus E1 and E2 have transmembrane domains anchoring them to the virus membrane, while the E2 accessory protein cytoplasmic tail also binds to the nucleocapsid. (C) Class III herpes simplex virus-1 gB fusion protein trimers coexist on the herpesvirus surface with several other glycoproteins including those that bind to receptors and engage with gB to activate it (adapted from [65]). A composite model for gB is shown, composed of prefusion gB ectodomain and TMD and cytoplasmic domain observed in a postfusion structure, which has been speculated to also exist in some form in prefusion gB (PDB 6Z9M and 5V2S [54,56,65]). Cryo-electron tomography enables analysis of intact, integrated viral fusion machinery and virus ultrastructure. (D) Cryo-ET section of filamentous influenza A virus (IAV) show a dense array of class I hemagglutinin (HA) fusion glycoproteins in stark contrast to the low copy number of Env on HIV-1. (E) Cryo-ET provides 3-dimensional reconstructions revealing virus ultrastructure as shown in tomographic sections (Movie S1). (F) Individual viral fusion and receptor binding proteins (hemagglutinin for IAV shown here), membrane fine structure, internal matrix proteins and nucleoprotein complexes can be resolved in tomograms.

Box 1. Classical pathway for protein-mediated membrane fusion.

All viral fusion proteins employ two hydrophobic motifs: a transmembrane domain (TMD), which anchors the protein to the viral membrane, and a fusion peptide or fusion loop, which is released from a sequestered site under activating conditions, allowing it to insert into a target host membrane [95]. Following activation, the fusion proteins undergo large-scale conformational changes from prefusion to an extended “prehairpin” form that bridges the two membranes, followed by a critical step in which the proteins fold back upon themselves into postfusion “hairpin” configurations, drawing the membranes together [1] (Figure I). The conformational changes accompanying fusion protein activation and refolding are instigated by environmental signals such as exposure to acidic pH in endosomes, receptor binding, and proteolytic cleavage [1]. Many viral fusion proteins also combine cellular receptor binding functions into the fusion protein assembly, while others engage with separate receptor binding proteins on the virus surface.

The general pathway of membrane remodeling pictured here has been previously proposed based upon fluorophore-monitored lipid and content mixing assays, as well as from theoretical considerations, simulations, and modeling [1,24,70,71,73]. The pathway depicts stages of viral fusion protein activation leading to attachment, initial stages of membrane deformation, usually depicted as symmetrical, localized deformation of virus and target membranes into dimples that are drawn together. Next, a state in which the two proximal leaflets have merged, giving rise to an hourglass shaped hemifusion “stalk” has been considered the most likely lipidic intermediate state. This then expands into a bilayer called a hemifusion “diaphragm” formed by the distal leaflets. Lastly the diaphragm transitions to open a pore that allows content transfer between virus and cytosol. Recent studies of viral membrane fusion suggest a reexamination of the classical pathway describing protein-mediated membrane fusion with a particular focus on observed intermediate stages of membrane remodeling.

Box 2. Biophysical and structural approaches for analyzing membrane fusion.

Protein-mediated viral membrane fusion is a dynamic process that involves many interacting parts. Powerful biophysical approaches for probing highly dynamic intermediate states of fusion proteins and reactions can provide complementary insights into the machinery, membranes, and processes.

• Cryo-electron microscopy (cryo-EM) with single-particle analysis can generate high-resolution (<4 Å) structures of isolated macromolecular assemblies [6]. A flash-frozen specimen embedded in vitreous ice is imaged using a transmitted electron beam, which compresses 3-dimensional information into a projection on the detector. To obtain high-resolution 3-dimensional structures extensive averaging, alignment, and classification of images of the specimen is required. Variable, dynamic intermediate complexes involving whole viruses undergoing fusion can be challenging to characterize by this approach.

• Cryo-electron tomography (cryo-ET) is a cryo-EM approach where 3-dimensional information is gathered by taking a series of projection images of a specimen as it is tilted through an angular range, followed by computational reconstruction of the tomographic volume from the images (Figure I) [7,102]. Using cryo-ET, one can directly image membrane fusion reactions with resolution of individual fusion proteins and membrane leaflets.

• Hydrogen/deuterium exchange-mass spectrometry (HDX-MS) measures changes in backbone amide accessibility to solvent [114]. The relative deuterium exchange levels relate to protein structure and local dynamics. HDX-MS provides peptide-level resolution and reports on conformational transitions on a broad range of timescales from tens of milliseconds to hours.

• Molecular dynamics simulations span a wide range of spatial and temporal resolutions from nanoseconds-millisecond. Atomistic and course-grain simulations have been applied to study viral fusion [115,116].

• Nuclear magnetic resonance (NMR) spectroscopy is a versatile technique that has been used to study protein structure, dynamics and interactions relevant to fusion [117]. NMR signals are sensitive to changes in chemical environment and have been used to determine fusion peptide structure and interactions with membranes as well as to probe lipid ordering. New approaches such as magic angle spinning NMR can be applied to investigate structure and dynamics in large assemblies as well [117].

• Single-molecule Förster resonance energy transfer (smFRET) involves labeling a protein with a pair of donor and acceptor fluorophores and using a fluorescence microscope to image and monitor energy transfer between single FRET pairs, typically with millisecond resolution [118]. Energy transfer efficiency is a function of relative distance and orientation of the fluorophores. In most cases, FRET is used to probe distances of 1-10 nm. Intrinsic dynamics reflecting structural fluctuations and conformational transitions can thus be tracked within individual fusion proteins.

Cryo-EM has revolutionized the study of biological structure and function including dynamic processes such as membrane fusion [6]. Rapid cryogenic freezing of samples in buffer without requiring fixation allows macromolecules and even whole viruses to be imaged at high-resolution in their native hydration state (Box 2). Fusion complexes between whole viruses and target membranes tend to be somewhat variable in organization and are not amenable to the standard single particle cryo-EM approaches for 3-D reconstruction. Fortunately, cryo-electron tomography (cryo-ET), which reconstructs 3-dimensional structure from projection images of a specimen gathered over a wide angular range (Box 2), enables membrane fusion intermediates to be captured and the organization of complex fusion sites to be imaged with resolution of individual fusion proteins, internal virus architecture, as well as integrity, curvature, and organization of membrane leaflets (Figure 1 and Movie S1) [7].

Some of the first high-resolution glimpses of viral protein-mediated membrane fusion were captured using cryo-ET to track herpes simplex virus-1 (HSV-1) infection of neurons [8]. Although studying fusion in an authentic viral infection in this manner is ideal for maintaining native membrane and receptor composition as well as cellular architecture, analyzing such a dynamic process in cells poses a challenge due to the difficulty of imaging large numbers of events and synchronizing the activity across a population of virions. By studying membrane fusion in an in vitro reaction between virus and liposomes or cell-derived membrane vesicles, one can set up a time course after triggering the fusion reaction, then flash freeze the specimen at specific time points, capturing hundreds of fusion events at intermediate stages [9-18]. Since each individual membrane fusion event occurs rapidly, on the order of milliseconds-to-seconds as measured by fluorescence microscopy [19-22], the states observed on the timescale of cryo-ET experiments (seconds-minutes) primarily reflect significantly populated metastable stages along the fusion pathway rather than highly transient states. By following the rise and fall of intermediate states as the reaction progresses through a time course, the sequence of reaction intermediates and general pathways can be inferred from direct visualization of individual fusion reactions. With these new approaches, we can begin to integrate the available information and observed snapshots of fusion reactions and from their population kinetics determine how the data informs our understanding of the fusion process. In interpreting these dynamic processes, we note that it is important to keep in mind that fusion reactions may follow more than one possible path to achieve fusion pore opening [23,24].

Here, we highlight recent work that reveals general aspects about the fusion process and identify plausible pathways traversed by systems where the broadest knowledge, most complete set of membrane fusion snapshots, and clearest mechanistic details are currently available. Specifically, we focus on influenza A virus (IAV) hemagglutinin (HA)-mediated fusion [13-15,25-28] and the Chikungunya alphavirus as the prime examples of class I and II fusion systems respectively; fusion intermediate states have been most completely mapped by cryo-ET and complementary methods [11,21,29]. Class III fusion systems offer an intriguing parallel case we discuss, but investigations have been more limited to date. We then describe the membrane remodeling that occurs during fusion and discuss the role fusion proteins play in regulating these processes. Collectively these findings encourage a revisitation of the membrane fusion pathway and suggest future experimental, theoretical, and computational modeling studies to test the updated models. Finally, we consider newly discovered fusion systems and point to powerful methods that are enabling viral-host membrane fusion to be imaged in situ.

FUSION PROTEIN INTERMEDIATES AND MEMBRANE REMODELING DURING FUSION

Class I viral fusion systems

To date, the most complete set of snapshots of class I membrane fusion has been gathered for influenza A virus (IAV) hemagglutinin (HA)-mediated fusion (Figure 2) [13-15,25-28]. These studies revealed that membrane remodeling under endosomal low pH conditions commences with activation of HA producing a prefusion-like conformation that attaches to the target membrane as the fusion peptides are deployed (Figure 2A). This HA state may relate to the extended, prehairpin conformations recently revealed by single-particle cryo-EM analysis of soluble, isolated HA (Figure 2B) [30]. Evidence of HA subsequently folding back on itself into the hairpin configuration is indicated by “V” or “” shaped densities (Figure 2C) that bracket regions of localized membrane dimpling [13-15,18,31]. Notably, early stages of membrane deformation with IAV are almost entirely focused on the target membrane since the internal matrix layer within IAV particles remains intact, preventing the viral membrane from bending [13,18,26,32,33].

Figure 2. Example of a class I fusion pathway.

(A) Recent biophysical studies of influenza A virus (IAV) hemagglutinin (HA)-mediated fusion have revealed that this largely alpha-helical pH-triggered class I fusion protein, a trimer of HA1 receptor binding (grey)/HA2 fusion subunit (blue) heterodimers, transitions through highly dynamic intermediate conformations in which the fusion peptides (red) become exposed and membrane-active [25,31,120]. Initial stages of the pathway show partial reversibility [120], and dynamic intermediate states persist for longer when HA is present on the virus surface where it interacts with the membrane and the internal M1 matrix layer [25]. (B) Single particle cryo-EM analysis [30] has revealed HA intermediate conformations (PDB 6Y5G, 6Y5I, 6Y5J, 6Y5K [30]; postfusion HA2 PDB 1QU1 [98]), however dynamic intermediate states that engage target membranes may be too flexible to resolve by this approach. HA1 head domain dilation allows HA2 to refold into intermediate prehairpin and postfusion hairpin conformations. (C) Snapshots from time-resolved cryo-ET analysis of IAV fusing with liposomes have revealed the interplay of HA and membranes from fusion peptide-mediated target membrane attachment all the way to completing fusion [13-15,18,25]. (D) Based upon these studies, IAV fusion appears to begin with attachment, followed by localized dimple formation as HA begins to refold. Subsequently, tight membrane docking across large areas is observed with HA in intermediate configuration at the periphery of the contact zones (C, blue arrows). Once the internal matrix layer dissociates from the viral membrane, the reaction can proceed beyond tightly docked contacts to achieve pore opening.

The next membrane configuration observed with IAV fusion involves tightly docked membrane contact zones where the two proximal membrane leaflets are, at the resolution of the cryo-electron tomograms (<1nm), virtually indistinguishable (Figure 2C) [13,15]. Fusion proteins are excluded from the contact zone and arrayed instead along the periphery. Interpretation of the tightly docked contacts will be discussed below alongside findings from recent modeling and simulations. These membrane contacts subsequently appear to resolve to a “hemifusion” bilayer referred to as a diaphragm [13-15]. In the case of influenza HA-mediated fusion, the stability and abundance of the hemifusion diaphragms are highly dependent on lipid and sialic acid receptor composition in the target membranes [13-15,34]. Cryo-ET time course experiments have shown that these intermediates lead to postfusion complexes in which the viral RNP is transferred into the lumen of the liposomes [13,15], however, the structural transition from the hemifusion diaphragm to a fusion pore has yet to be directly visualized.

One other key aspect of the fusion process underscored by cryo-ET analysis is the role of the internal matrix protein layer that lines the viral membrane. Two recent studies reported the architecture and structure of M1 as it exists in prefusion IAV [35,36]. The IAV M1 layer exhibits interactions with the membrane that change under endocytic low pH conditions [13,18,26,32,33]; these interactions initially are maintained while HA grapples to the target membrane and draws it into apposition with the virus surface (Figure 2C). Dissociation of the M1 layer, which occurs at a lower pH than HA activation, increases the pliability of the viral envelope, which enables the fusion reaction to progress beyond the tightly docked membrane stage [13,18,26]. Similar functions of analogous matrix proteins in regulating membrane fusion have been proposed in filoviruses such as Ebola, which also employ a class I fusion protein [37]. Matrix layers and their interaction with fusion proteins thus appear to be key regulators of fusion protein and membrane remodeling activities.

Intermediate stages of activation and fusion of other class I systems have recently been visualized for SARS-CoV-2 S [10,38], Lassa arenavirus GP [39], human parainfluenza virus (hPIV) F [40], murine leukemia virus (MLV) Env [41], HIV-1 [9,12,42], and avian sarcoma/leukosis virus (ASLV) Env [17]. Many of these studies focused on initial stages of fusion protein-receptor interactions or capturing extended prehairpin-like intermediates using fusion inhibitors [10,12,42], and key intermediate stages remain to be observed. To gain further insights into the generality of the membrane remodeling and fusion process across the class I systems, more extensive time-dependent sampling is needed for these viruses.

Class II viral fusion systems

Class II fusion proteins are rich in beta-sheets and comprised of three globular domains I-III (Figure 3). The membrane-active fusion loop is at the distal tip of domain II. In the prefusion conformation it is sequestered either underneath an accessory protein or at an interface with adjacent fusion proteins in a lattice [43]. In some viruses the fusion protein assemblies are organized with icosahedral symmetry, such as is the case with flavi-, alpha-, and phlebo- viruses (Figure 1E) [44-46], while in others, such as orthobunya- and hanta- viruses, fusion protein-accessory protein heterodimers are organized in discrete tripods or tetramers, but those assemblies are not arranged symmetrically on the virions [47,48].

Figure 3. Example of a class II fusion pathway based upon the Chikungunya alphavirus.

(A) Class II viral fusion proteins in many cases form heterodimers with an accessory protein, which helps cap the fusion protein’s hydrophobic fusion loop (pink). In response to low pH, the fusion protein E1 (domain I, II, III: red, yellow, blue) and the accessory protein E2 (light blue) reorganize. Monomeric E1 grapples with the target membrane via its fusion loop then trimerizes into extended, prehairpin homotrimers. Subsequently, domain III (blue) folds back against the trimerized domain II (yellow) central core forming the final hairpin conformation. This draws the membranes together and induces them to fuse. (B) Cryo-ET was used to capture snapshots of class II E1-mediated fusion between CHIKV and liposomes. The resulting snapshots revealed stages of protein-membrane interaction and membrane remodeling, which showed both similarities (e.g. tightly docked membrane contacts and hemifusion diaphragms) and differences (lack of localized, high-curvature dimples) with class I fusion. (C) One can integrate the information provided from cryo-ET imaging of this class II fusion system (top) along with other available protein and virus structural information and knowledge of lipid bilayer organization to animate the entire fusion process (bottom, Movie S2). Stills from animation by Drs. Margot Riggi and Janet Iwasa, University of Utah. Synthesizing the information in this way allows one to test whether the pathway we presume is traversed is reasonable, or whether the proposed interpretation may need to be revised.

By applying a similar time-resolved cryo-ET approach as used for IAV fusion, the pathway of class II Chikungunya alphavirus (CHIKV) membrane fusion was recently visualized (Figure 3 and Movie S2) [11]. Soon after activation by acidic pH, the fusion loop on the CHIKV fusion protein, E1, becomes exposed from underneath the E2 accessory protein [49] and E1 engages the target membrane as a monomer (Figure 3A, B) [11]. Remarkably, membrane binding by the fusion loop appears to be so strong that E1 monomers can be hyperextended as virus particles grapple to the target membrane [11,50]. Over time, multiple copies of E1 monomer bridge across to the target membrane before associating into extended prehairpin homotrimers with trimerized domain II cores while domain III remains oriented towards the virus surface (Figure 3A).

Notably, no sign of localized dimples, which are common for class I HA-mediated fusion, was observed during CHIKV fusion. Instead, virions appear to be brought into direct contact with the target membrane as additional E1 homotrimers adjacent to the initial contact site grapple onto the target membrane (Figure 3B) [11]. In alphaviruses such as CHIKV, the cytoplasmic tail of the E2 protein binds to the icosahedral nucleocapsid, holding it close to the inner leaflet of the viral membrane, (Figure 1E). Yet just as M1 eventually dissociates from the IAV membrane, in CHIKV, E2 releases from the internal nucleocapsid at intermediate stages of fusion. This step is necessary to unmoor E1 and E2, permitting them lateral mobility. Once this is achieved, the surface proteins are pushed to the periphery of the membrane contact zones, and the two membranes form a tightly docked interface, as seen with influenza virus (Figure 3B). The tightly docked contacts thus appear to be a common intermediate during fusion of these divergent systems. Based upon cryo-ET time courses, the tightly docked membranes next transition to hemifusion diaphragms that nearly span the width of the CHIKV particle [11]. Densities consistent with postfusion E1 homotrimers are observed fringing the perimeter of the diaphragm, suggesting that hairpin formation may be a prerequisite for the two membranes to attain the hemifused state. Finally, the diaphragm appears to disintegrate to open a pore, leading to nucleocapsid release.

Cryo-ET-based snapshots from the CHIKV study were rendered with protein and lipid structures based upon high-resolution data for components and brought to life in a 3-D animation (Figure 3C and Movie S2). The practice of animating this dynamic biological process is helpful for testing hypotheses and visualizing how large-scale changes in molecular organization occur as the system traverses the pathway inferred from the tomography data. While this animation only accounts for physical forces at the level of steric hinderance and spatial occupancy, modeling and simulation that take into account other forces and energetics governing the reaction are needed to gain physical insight into detailed mechanisms of protein-mediated membrane fusion [51].

Whether other class II systems follow a similar trajectory and populate similar intermediates is not yet known. For one, the prefusion organization of class II fusion proteins and their interaction with accessory proteins varies dramatically between different virus families. Beyond CHIKV, other individual stages of class II membrane fusion have been characterized for the Sindbis and Eastern Equine Encephalitis alphaviruses [49,50], Uukuniemi [16] and Rift Valley Fever Viruses [52], phleboviruses, and orthobunyavirus [48]; however, as with class I fusion systems, additional time-resolved analysis of intermediates with an eye towards membrane remodeling is needed for these systems to identify the fusion pathways they traverse.

Class III viral fusion systems

Recent studies have begun to make headway in revealing conformational states and activation of class III fusion proteins including gB from herpesviruses (Figure 1C) [53-56]; however, herpesvirus gB activity is regulated by interaction with several other surface glycoproteins, which poses a challenge for reconstituting gB fusion activity in vitro [57]. Likewise endosomal pH-induced conformational changes in the G fusion protein from the rabies rhabdovirus have been described recently [58,59]. Examination of another rhabdovirus, vesicular stomatitis virus (VSV), G-mediated fusion indicate that class III proteins, like class II proteins, appear to populate a monomeric intermediate [60] prior to adopting a stable postfusion trimer [61]. Interestingly, in class III systems, bipartite fusion loops are initially oriented towards the viral surface where contacts with the membrane are believed to help stabilize the prefusion trimer conformation (Figure 1C) [54,62-64]. It is not clear whether the viral membrane-bound fusion loops result in deformation of the viral membrane itself over the course of the fusion reaction. Indeed, to date relatively few studies imaging fusion for class III systems have been reported. In the in situ HSV-1 studies reported by Maurer et al., acute membrane dimples were observed extending between both HSV-1 and cellular membranes [8]. These sites were bracketed by protein density consistent with gB refolding from an extended prehairpin to a hairpin conformation. The pliability of the HSV-1 membrane, in contrast to IAV, may relate to HSV-1 having a membrane-associated tegument layer, which is less ordered than a typical membrane-bound matrix layer [65]. Other viruses, such as the rhabdoviruses, with class III fusion proteins exhibit highly ordered internal architecture, which may stabilize the viral membrane and maintain virus structure during fusion [66]. Electron microscopy imaging of fusion between VSV and liposomes provided a first glimpse of G-mediated fusion [67], while more recent studies have investigated biophysical and functional aspects of VSV and herpesvirus fusion reactions [57,59,68,69].

LIPID AND MEMBRANE REMODELING PATHWAYS

Nature of the initial membrane contacts and the elusive hemifusion stalk

The application of cryo-ET to capture and visualize membrane fusion reactions in progress as described above has made it possible to not only image what is happening to the viruses and their fusion proteins, but to directly image membrane remodeling. Such experiments provide unique insight into mechanisms of fusion protein-driven induction of curvature, membrane docking, perturbation of membrane leaflets, production of hemifused intermediates, and pore opening.

Most prevailing models of membrane fusion depict initial membrane contacts occurring at localized dimples, which would minimize the area that needs to be dehydrated in order for the proximal leaflets to make contact (Box 1 and Figure 4A, path i). Tight curvature at the tip of dimples has been suggested to facilitate exposure of lipid tails and enable lipid exchange across the interface [70]. The classical model hypothesizes that localized dimpled contacts would transition to a narrow hemifusion stalk, which classically has been described from continuum (non-atomistic) models as an hourglass-shaped neck formed by proximal leaflets merging [70-73]. In the stalk configuration, hydrated polar headgroups line the exterior of the neck, and apolar tails are sequestered in the interior (Figure 4A, path i). Such structures have been experimentally observed under highly controlled conditions in lipidic systems [74,75], but thus far have not been definitively visualized in viral fusion systems. In some computational modeling and theoretical studies, a broader range of lipid organizations are sometimes referred to as “stalks”; in these cases the operable definition appears to primarily involve acyls tail from opposing membranes co-mingling, even across closely apposed bilayers, but local leaflet curvature and hydration state often differ dramatically from the classical model [24,51,72-74,76-81]. In our view it is important to be specific about this terminology and architecture as the energetics and forces at play will differ substantially from one configuration to the next. In this review, by “stalk” we refer to the classical hourglass-shaped neck [73].

Figure 4. Alternative membrane remodeling pathways leading to lipid mixing and hemifusion.

Transitioning from two separate lipid bilayers to a fusion pore involves application of mechanical force by fusion proteins on the membranes to draw them into apposition as well as perturbation of membrane hydration (blue layer) and lipid packing to facilitate mixing of lipid tails. (A) Comparison of membrane remodeling intermediates. According to the classical pathway i (see Box 1), membrane interactions start with localized, high curvature dimples from one or both membranes approaching each other. A hemifusion stalk forms when polar headgroups move aside, allowing apolar lipid tails to intermingle at the point of contact. The hourglass-shaped stalk formed by merged proximal leaflets then transitions to a hemifusion diaphragm with this new bilayer formed by distal leaflets. An alternative model, following pathway ii, based largely on cryo-ET observations, involves the membranes forming tightly docked areas with extensively dehydrated polar headgroups from proximal leaflets intimately apposed. Polar headgroup packing fluctuations allow the apolar tails to splay between membranes. The headgroup boundary then ruptures, allowing significant lipid mixing and formation of the hemifusion diaphragm composed of distal leaflets. (B) How might fusion proteins drive lipid mixing? After membrane apposition, fusion proteins are located at the periphery of tightly docked membrane contacts. The membrane transitions depicted in (A) may initiate at the edges of the contact zones as fusion peptides and transmembrane domains disrupt lipid order and headgroup packing while also being drawn together by hairpin formation and direct fusion peptide-TMD interactions.

It is conceivable that such classical stalks have been elusive in experiments because they rapidly transition to open fusion pores on a time scale faster than can be captured by cryo-ET, as has been proposed to occur with cellular SNARE protein-mediated fusion [23,82]. However, we note that in both class I and II systems described above, rather than seeing an early accumulation of completed fusion complexes or even hemifusion diaphragms immediately following initial membrane apposition, the state that is most commonly populated are the extended, tightly docked membrane-membrane contact zones (Figure 4A, path ii), and only at later time points do hemifusion diaphragms and subsequently open pores and postfusion complexes increase in abundance [11,13-15]. Indeed, even when fusion begins with point contact-like dimples such as with influenza A HA-mediated fusion, it appears that the membranes next form tightly docked contacts between proximal leaflets rather than stalks, hemifusion diaphragms, or open pores (Figure 2) [13].

In contrast to a localized dimple-like membrane contact (Figure 4A, path i), the tightly docked interfaces that are observed extend over tens to hundreds of square nanometers and involve extensive dehydration and close packing of phospholipid headgroups (Figure 4A, path ii) [83]. A number of studies suggest that the energetic cost of dehydration can be offset by favorable dipolar interactions between headgroups and an increase in entropy upon release of bound water [83-85]. Enrichment of target membranes with fusogenic components such as cholesterol or the endosomal lipid bis(monoacylglycero)phosphate (BMP) promote tight membrane docking [13] and increase the fraction of complexes that reach complete fusion [86]. This may be due to properties such as cholesterol’s hydroxyl headgroup being readily dehydrated or a lipid’s structure facilitating tail exposure; alternatively, the fusogenic lipids may disfavor off-pathway dead-end lipidic states. Tightly docked membrane-membrane configurations have also been observed in SNARE- [23,83,87] and mitofusin-mediated fusion [88] and thus appear to be fairly universal states populated during protein-mediated membrane fusion.

Polar boundary rupture, hemifusion, and pore opening

In order for lipid mixing to commence, polar boundaries composed of closely packed headgroups must be breached to allow the apolar tails to start intermingling. Modeling and simulations have shown that thermal fluctuations of lipids from two closely apposed membranes can lead to tail exposure, producing nascent bridging connections as they splay between the apolar zones of the bilayers [77,81,89-91].

If a fusion reaction follows pathway ii in Figure 4A, as is consistent with cryo-ET time course observations, once lipid tail splaying and lipid mixing commences at the tightly docked interface, hydrophobic interactions between tails would come to dominate over headgroup interactions, rupturing the polar boundary and allowing the distal leaflets to approach to form a hemifusion diaphragm. Hemifusion diaphragms are commonly observed in viral fusion experiments, though some studies of SNARE-mediated fusion have suggested they may represent a kinetically trapped state on a pathway that is secondary to more efficient fusion via localized “point” contacts [23]. A recent all-atom simulation of SNARE-mediated fusion indicated that small vesicle fusion occurs through a localized, dehydrated contact zone where the TMD promoted acyl tail fluctuations across the polar boundary leading to formation of a somewhat heterogeneous stalk-like organization that degenerated quickly to an open pore, without populating a distinct hemifusion diaphragm [82].

By contrast, the viral fusion studies described above indicate that following emergence of tightly docked membrane contacts, hemifusion diaphragms are on-pathway, transitioning over time to productive fusion pores [11,13,15]. It remains to be determined whether this transition between the tightly docked state to a hemifused diaphragm bilayer initiates at some site or sites in the central region of the contact zone or at the edge where fusion proteins undergoing refolding to hairpin conformations with membrane-embedded fusion peptides/loops and TMD are localized; notably the membranes also exhibit acute discontinuity in local curvature at these positions, which may facilitate exposure of lipid tails and lipid mixing (Figure 4B). Likewise, the steps that lead a hemifusion diaphragm to transition to an open pore remain poorly understood. Simulations suggest these types of leaflet disrupting transitions may commence at the edges where fusion peptide and TMD are concentrated (Figure 4B) [51,92-94]. Modeling (reviewed in [51,95]) and recent NMR studies [96] of class I fusion proteins have shown that fusion peptides promote local lipid dehydration, membrane perturbation, and acyl tail exposure, which can lead to lipid translocation between apposed membranes. One analysis reported that transbilayer fusion peptide interactions with phospholipid headgroups on the distal leaflet generate intense local stresses that facilitate pore opening [93]. Whether a similar type of transbilayer interaction can occur for class II and III fusion loops or whether they induce local perturbations through other mechanisms remains to be seen.

A fusion protein fully attaining the postfusion hairpin conformation in some cases leads the fusion peptide and TMD to form a complex [97,98]. A recent cryo-EM structure of the class I SARS-CoV2 S2 fusion subunit bound to a lipid nanodisc exhibited an intertwined wedge of fusion peptide and TMD [97]. A similar cap formed by N- and C-terminal segments of influenza HA had been previously observed in a crystal structure for the postfusion HA2 fusion subunit [98]. Thus, it seems likely that a combination of fusion peptide and TMD domain perturbation of the lipid ordering as well as direct coupling of these membrane interactive components may drive the destabilization of the hemifusion diaphragm, opening a pore between viral lumen and cytosol.

The experimental observation of membrane fusion intermediate states by cryo-EM/ET provides quantitative observables documenting bilayer thickness, curvature, contact areas, and leaflet integrity, which can help constrain theory, computational modeling, and simulations [83,99-101]. Modeling on the other hand can fill in molecular detail that cryo-ET does not resolve, especially for illuminating the role of water and hydration, lipid structure including head group and tail behavior, and the action of motifs such as fusion peptides/loops and TMD.

Future combined cryo-ET and modeling studies— where the effect of lipids with different propensities to flip between membranes, as well as differences in polarity, shape, tail saturation and length, membrane thickness, and ability to form lipidic domains are compared— can advance our fundamental understanding of the mechanics and physical determinants of protein-mediated membrane fusion [74,76,79,83,91,99-101].

CONCLUDING REMARKS

As informative as in vitro membrane fusion studies have been, ideally one would like to image virus entry and membrane fusion events in cells; however, particularly for viruses that enter through endocytosis, the thickness of the cell body poses a challenge for transmission electron microscopy due to beam attenuation. For such specimens it can be impossible to image fusion events with the clarity sufficient to resolve individual fusion proteins and membrane fine structure. Recent advances using focused ion beams to mill cellular specimens followed by cryo-ET imaging of the resulting thin lamella have made it possible to obtain 3-dimensional tomographic sections from deep inside cells [102]. In one recent study for example, the effect of the innate immunity restriction factor IFITM3 on inhibition of IAV fusion within endosomes was analyzed using this powerful new approach [103].

Imaging viruses in situ using cellular tomography, even without ion beam thinning, has also been key to capturing native viral morphology, which otherwise can be distorted during purification [104]. Since morphology may be a key determinant of virus entry into cells, the ability to study native virions in cell culture without destructive purification steps may improve the authenticity and completeness of our understanding of cell entry and membrane fusion for many enveloped viruses. Cell surface viral fusion proteins also mediate cell-cell fusion and formation of syncytia as an alternative mode of viral spread to infection by free viral particles [105]. While the fusion proteins that mediate the process are the same as in particles, much less is understood about the organization of the fusion proteins on the cell surface and their interaction with receptors on target cells. Cellular tomography can resolve viral fusion proteins on the cell surface [106]. Thus, using such approaches, it will soon be possible to image cell-cell fusion as well as virus transfer across virological synapses and other modes of inter-cellular transmission [105].

Beyond the familiar three classes of viral fusion proteins, recent structural insights into the glycoproteins used by hepaciviruses, such as hepatitis C virus, indicate that these, as well as related pesti- and pegi-viruses, employ alternative fusion machinery distinct from other viral fusion proteins [107]. Fusion proteins with novel folds have even been described in archaeal viruses [108]. The activity of these intriguing fusion protein families that lack structural similarity with known fusion proteins are just beginning to be probed but may offer new insights into mechanisms of viral membrane fusion. It will also be informative to compare pathways and fusion intermediates with cellular fusion systems such as SNARE- and mitofusin-mediated processes [109,110]. Likewise, an expanding menagerie of gamete fusion protein based upon the same class II fold found in viruses have been described [111-113]. Among these wide-ranging systems, already some generalities are observed such as the presence of tightly docked membrane contacts, but the extent of other parallels and differences remains to be determined. Future comparative studies may point to universal aspects of protein-mediated membrane fusion or possibly reveal alternative pathways.

OUTSTANDING QUESTIONS:

• What additional information can be extracted from cryo-electron tomograms to inform modeling and simulations? From tomographic data, more quantitative analysis of the membrane lipid densities may offer insights that qualitative observations have overlooked, such as intensity of the lipidic density at contact zones and variation in bilayer thickness at sites of membrane remodeling [83]. Recent studies using cryo-EM/ET have resolved membrane microdomains based upon bilayer thickness and lipid headgroup charge [83,99-101]. Likewise careful analysis of continuity of leaflet density may offer insights into whether breaches in the polar boundaries can be resolved.

• What is the impact of specific lipid types and properties on the fusion pathway and membrane remodeling intermediates? Past studies have investigated the role that lipid composition has on the fusion process, but in general their effect has been interpreted in terms of membrane curvature and how they may promote fusion stalk formation. The effect of specific lipids on apposition and dehydration leading to formation and transition from tightly docked membrane states can now be visualized directly by cryo-ET (e.g. [79,91]). Such experimental studies combined with computational modeling will aid in improving our mechanistic understanding of membrane remodeling.

• At which stage and through what type of membrane deformation or defect does lipid mixing begin? Under what conditions can or does a hemifusion stalk form, and if so, what is its specific geometry and architecture? Understanding aspects of curvature, hydration, and involvement of fusion peptides and transmembrane domains at the critical sites where lipid mixing may commence will be important for resolving this critical stage of the fusion process. It also remains to be understood how transitions between the metastable states that have been imaged take place such as from tightly docked bilayers to hemifusion diaphragms and from diaphragms to open pores.

• What pathways do other viral fusion systems follow, such as the class III fusogens and systems where receptor-binding and/or proteolytic cleavage are the key triggers of viral fusion machinery.

HIGHLIGHTS.

• Cryo-electron tomography enables capture of 3-dimensional images of membrane fusion intermediate states with resolution of fusion proteins and membrane leaflet remodeling

• Studies of Class I, II, and III viral membrane fusion proteins are revealing general features as well as differences in stages of membrane reorganization leading to complete fusion and pore formation

• Viral fusion machinery is comprised of fusion proteins, viral membrane, and internal structural components such as matrix proteins that regulate fusion protein conformation and function

• Tightly docked membrane contact zones with closely apposed proximal headgroup layers appear to be a near universal intermediate state along the fusion pathway

• Computational modeling that builds upon observed structural information from cryo-electron tomography may offer a path towards a detailed mechanistic understanding of the forces and energetics that drive protein-mediated membrane fusion

GLOSSARY

Activation

When viral fusion proteins are exposed to specific triggers, they become more structurally dynamic with exposure of fusion peptide or fusion loops. This primes the machinery for large-scale structural changes

Cryo-electron tomography (cryo-ET)

A cryogenic transmission electron microscopy method in which 3-dimensional data is gathered by imaging the specimen over a range of tilt angles followed by reconstruction of the 3-D volume (tomogram) from the series of 2-D images

Distal leaflet

The leaflets of two apposed membrane bilayers that are farthest apart, facing the virus lumen, cytosol or vesicle lumen

Fusion peptide/fusion loop

Membrane-binding moiety of fusion proteins that embeds in target membrane. Usually hidden in prefusion state and exposed upon activation

Fusion pore

Aqueous conduit connecting the lumen of virus and cytosol or vesicle, formed following fusion of the previously separate membranes

Fusogenic

Fusion promoting, usually referring to a property of specific lipids

Hairpin conformation

Postfusion conformation of fusion proteins in which they are folded back upon themselves and the two membrane-active moieties (fusion peptide/loop and transmembrane domain) are colocalized

Hemifusion

Intermediate stage of membrane fusion with partially merged lipid bilayers. The proximal leaflets have merged, while distal leaflets form a new bilayer that still separates the aqueous compartments enclosed by the two original membranes

Hemifusion diaphragm

Fusion intermediate configuration in which a lipid bilayer is formed by distal leaflets of fusing membranes, following merging of proximal leaflets. The diaphragm separates the aqueous compartments of virus lumen and cytosol or vesicle lumen

Leaflet

One layer of membrane bilayer composed of lipids with polar headgroups oriented towards aqueous solvent and apolar tails sequestered in hydrophobic membrane center

Lipid bilayer

Composed of two lipid leaflets with their hydrophobic tails positioned towards each other in membrane center and hydrophilic headgroups oriented towards aqueous solvent

Matrix protein

An internal structural protein found in many viruses that often associates with the inner leaflet of the viral membrane, conferring structural stability and morphology to the virus

Nucleocapsid

Viral structural protein that binds to encapsulated nucleic acid such as the viral genome

Postfusion state

Conformational state of fusion proteins following activation and refolding to the hairpin configuration

Prefusion state

Conformational state of fusion proteins prior to activation

Prehairpin state

Extended intermediate configuration of fusion proteins where they bridge virus and host target membranes.

Proximal leaflet

The leaflets of two apposed membrane bilayers that will make initial contact with each other

Stalk

A narrow hourglass-shaped lipidic neck formed by merging of proximal leaflets between two apposed bilayers with lipid tails positioned inside and hydrophilic head groups are positioned outward towards aqueous solvent

Syncytia

Multinucleated cells formed by cell-cell fusion

Virological synapse

A junction between two cells across which viral particles can be efficiently transferred