Enveloped viruses initiate infection by fusing their membrane with the cell membrane and thereby depositing their genome into the cytosol. This membrane merger is catalyzed by specialized viral proteins referred to as fusion proteins. When activated via interactions with cellular receptors and/or by acidic endosomal pH, these proteins promote membrane merger by undergoing complex conformational changes (reviewed in 12). The principal challenges facing researchers studying molecular details of this process are: (i) limited structural information about fusion proteins and their refolding pathways; (ii) transient and generally irreversible nature of conformational changes; and (iii) often redundant number of proteins the majority of which may undergo off-pathway refolding. In spite of these obstacles, considerable progress has been made towards understanding viral fusion, as discussed in a number of excellent reviews 123456. The emerging picture is that disparate enveloped viruses have adapted a common strategy to fuse membranes. This review will discuss the general principles by which viral proteins promote fusion, focusing on the retroviral envelope (Env) glycoproteins exemplified by HIV-1 Env.

Whereas viral proteins regulate and promote the merger of biological membranes, complete fusion occurs when lipids from two distinct bilayers rearrange to form a continuous membrane. Thus, to elucidate the principles of protein-mediated fusion, it is essential to understand the mechanism of lipid bilayer fusion. The most prominent model for membrane fusion (Fig. 1A), referred to as the "stalk-pore" model 7, posits that contacting monolayers of two membranes are initially joined via a local saddle-shaped connection referred to as a "stalk" 89. Lateral expansion of the lipid stalk permits the distal monolayers to come into direct contact and form a shared hemifusion diaphragm. Accumulated evidence suggests that hemifusion is a common intermediate in a variety of protein-mediated fusion reactions (for review, see 10). The subsequent rupture of a hemifusion diaphragm results in the formation of a fusion pore through which both membrane and content markers redistribute 1112.

Generally, fusion proteins of enveloped viruses are type I integral membrane proteins expressed as trimers or dimers 12356. With a few exceptions, these proteins are rendered fusion-competent upon post-translational cleavage by cellular proteases of either the protein itself or of an associated regulatory protein 1213. A salient feature of viral proteins is a highly conserved, functionally important stretch of hydrophobic residues referred to as the fusion peptide or the fusion domain 11314. In their mature, proteolytically cleaved form viral fusion proteins are thought to exist in a meta-stable, "spring-loaded" conformation 15, capable of releasing the energy as they transition to final conformation. While it is likely that this conformational energy drives fusion, the exact mechanism of coupling between protein refolding and membrane rearrangements is not fully understood.

Based on the structure of extracellular domains, viral fusion proteins are currently categorized into three classes. Fusion proteins of retroviruses, filoviruses, coronaviruses, ortho- and paramyxoviruses displaying a prevalent α-helical motif belong to the class I proteins 11617. In an initial conformation, the N-terminal or N-proximal hydrophobic fusion peptides of the TM subunit (Fig. 1B) are usually sequestered at the trimer interface. Perhaps the best studied representatives of the class I proteins are influenza hemagglutinin and HIV-1 envelope (Env) glycoprotein (reviewed in 1819). The defining feature of the class II fusion proteins of flaviviruses and togaviruses is the predominant β-sheet motif 13. These fusogens are expressed as homo-dimers (tick-borne encephalitis virus E protein) or hetero-dimers (Semliki Forest Virus E1/E2 proteins) with their hydrophobic fusion domains sequestered from solution at the dimer interface (Fig. 1C). The newly identified class III viral proteins (rhabdoviruses and herpesviruses) exhibit both α-helical and β-sheet elements and thus appear to combine the structural features of first two classes 156. Interestingly, fusion proteins of rhabdoviruses exemplified by the G protein of Vesicular Stomatitis Virus (VSV) undergo low pH-dependent transition from a pre-fusion to a post-fusion form, but, unlike other viral proteins, return to their initial conformation at neutral pH 2021. This unique reversibility implies that the difference in free energy of pre- and post-fusion conformations of G proteins is relatively small. Thus, the pre-fusion structure of this protein may not be viewed as meta-stable, suggesting that the "spring-loaded" mechanism 15 that relies on large changes in the protein's free energy may not be operational here 20.

While the structures of ectodomains (or their core fragments) have been solved for several viral proteins, information regarding intermediate conformations of full-length viral proteins in the context of fusing membranes is not available. Complementary functional assays are thus important for gaining insight into the refolding pathways of viral proteins. Mechanistic studies of viral fusion have been primarily carried out using a cell-cell fusion model 112223. Cell-cell fusion assays adequately reflect the activity of viral proteins, especially when early manifestations of fusion, such as small pore formation, are being monitored. Further, this model is ideally suited for manipulating experimental conditions and for convenient and reliable quantification of fusion products. However, there is increasing awareness of the fact that not all features of virus-cell fusion can be faithfully reproduced in this model. For instance, murine leukemia virus (MLV) undergoes receptor-mediated translocation ("surfing") along microvilli to a cell body before fusing to a plasma membrane 24. An example of cellular compartment-specific entry is Ebola virus fusion that occurs after the cleavage of its glycoprotein by the lysosome-resident cathepsin B 2526. This intracellular activation of the fusion protein makes the cell-cell fusion model unsuitable for functional studies. The use of cell-cell fusion assays is also limited when surface expression of viral fusion proteins is low due to an endoplasmic reticulum retention signal. Examples of such glycoproteins include the Dengue Virus E 27 and Hepatitis C Virus E1/E2 28 glycoproteins.

Until recently, direct techniques to measure virus-cell fusion were not available, and most functional studies employed infectivity assays to evaluate fusion 29303132. However, measuring the levels of infection that rely on successful completion of viral replication steps downstream of fusion may underestimate the efficacy of fusion 3334. Novel techniques monitoring the delivery of viral core-associated enzyme into a host cell permit direct assessment of the extent and kinetics of virus-cell fusion 3334353637, but these assays have limited sensitivity and temporal resolution. A powerful approach to study virus-cell fusion that circumvents fundamental limitations imposed by the heterogeneity of virus population is time-resolved imaging of single viral particles (e.g., 383940414243). Using this technique, important advances have been made towards understanding the mechanisms of receptor-mediated virus uptake, endosomal sorting, and towards identifying the preferred sites of virus entry 44454647. Time-resolved imaging of viral lipid and content redistribution permitted visualization of intermediate steps of fusion between single HIV-1 and Avian Sarcoma and Leukosis Virus (ASLV) particles and target cells 4849.

Viral proteins are activated through various mechanisms principally determined by the virus entry pathway 122394150. Viruses that do not rely on low pH for entry are activated by binding to their cognate receptor(s) 5152 and are thought to fuse directly with a plasma membrane. Fusion proteins of viruses entering cells via an endocytic pathway are mainly triggered by acidic pH in endosomes 139. These viruses often use cellular receptors as attachment factors to facilitate their internalization. Interestingly, ASLV Env is activated via the two-step mechanism that involves binding the cognate receptor that renders Env competent to undergo conformational changes upon subsequent exposure to low pH in endosomes 53545556575859. The two-step activation of viral fusogens is not uncommon. HIV Env is rendered fusogenic through sequential interactions with CD4 and a coreceptor 5160. Following receptor-mediated endocytosis, the Ebola virus glycoprotein is activated by proteolytic cleavage in lysosomes 2526. These multiple triggering steps may help sequester the conserved functional domains of viral fusion proteins from immune surveillance and/or ensure the release of the viral genome at preferred cellular sites.

In spite of structural differences, different classes of fusion proteins appear to promote membrane merger through a common "cast-and-fold" mechanism (reviewed in 1234561116222361). The critical evidence supporting this universal fusion mechanism is the conserved trimeric hairpin (or 6-helix bundle, 6HB) motif shared by post-fusion conformations of disparate viral proteins 161617. For class I fusion proteins, this structure is formed by antiparallel assembly of the central N-terminal trimeric coiled coil (or heptad repeat 1, HR1 domain) and three peripheral C-terminal helices (HR2 domains), as depicted in Fig. 1B. The antiparallel orientation of the C-terminal and N-terminal segments of ectodomains of class II and III viral proteins indicates that these proteins also form trimeric hairpin structures (Fig. 1C). An important implication of a hairpin structure is that, in the final conformation, the membrane-spanning domains (MSDs) and the hydrophobic fusion peptides, which are not a part of crystal structure, are positioned close to each other.

The following consensus model for viral protein-mediated fusion has emerged from the implicit proximity of the MSDs and fusion peptides in the conserved hairpin structures and from extensive biochemical and functional data (Fig. 1B, C). When triggered by receptor binding and/or by low pH, viral proteins insert their fusion peptides into a target membrane 6263646566. At this point, the initially dimeric class II proteins convert to fusion-competent homotrimers 3613. In addition to anchoring the viral proteins to the target membrane, the fusion peptides appear to destabilize lipid bilayers by promoting the formation of non-lamellar structures 14676869. Next, the extended trimeric conformation bridging the viral and target membranes drives membrane merger by folding back on itself and forming a hairpin structure. Several lines of genetic and functional evidence support this model. First, mutations in the conserved fusion peptides 7071727374757677 and those destabilizing the trimeric hairpin 7879808182 attenuate or abrogate fusion. Second, peptides derived from the HR1 and HR2 regions of class I proteins (referred to as C- and N-peptides, respectively) inhibit fusion by binding to their complementary domains on the fusion protein and preventing 6HB formation (reviewed in 16). Likewise, soluble fragments of class II fusogens also block fusion 83, apparently by preventing the formation of trimeric hairpins.

The general principles by which viral proteins cause membrane fusion are likely dictated by the physical properties of lipid bilayers which must form highly curved and thus energetically unfavorable intermediate structures (e.g., a stalk and a fusion pore). Accumulating evidence that fusion induced by distinct classes of viral proteins converges to a common hemifusion intermediate 4956848586878889 further supports the universal mechanism of membrane merger.

While it is widely accepted that the transition from an initial conformation to a final hairpin drives fusion, the refolding pathways of viral proteins are poorly characterized. In discussing the conformational intermediates of class I viral proteins, this review will focus primarily on fusion induced by HIV-1 Env. Numerous antibodies to HIV-1 Env and entry inhibitors targeting the receptor binding and fusion steps are available for mechanistic studies of Env-mediated fusion. Recent functional work using various HIV fusion inhibitors provided new clues regarding the HIV entry process.

It is worth pointing out that 6HBs are only a part of the trimeric hairpin motif of class I proteins. There is evidence that regions outside the HR1/HR2 domains play a role in fusion. For instance, the membrane-proximal external region (MPER) and residues adjacent to the fusion peptide are essential for the formation and growth of a fusion pore mediated by HIV-1 Env and influenza hemagglutinin 78142143. Interestingly, ASLV Env appears to form 6HBs at low pH prior to membrane merger, as evidenced by resistance of fusion to the inhibitory HR2-derived peptide added at a lipid-arrested stage 144. This finding suggests that, unlike the HIV-1 Env 123 and paramyxovirus F 145 glycoproteins, interactions between residues outside the ASLV heptad repeat domains are responsible for hemifusion and fusion. The degree of coupling between bundle formation and membrane merger may depend on the length and/or flexibility of a region between the HR2 and MSD. It thus appears that, in order to induce fusion, viral proteins must zipper completely and bring their membrane-anchored regions (MSDs and fusion peptides) into close proximity. Interactions between HR1 and HR2 domains within the 6HB may or may not provide the main driving force for a fully zippered structure. We and others 1161 have hypothesized that fully assembled hairpins permit direct interactions between MSDs and fusion peptides, which may destabilize a hemifusion diaphragm and promote opening of a fusion pore (Fig. 1B).

Dilation of fusion pores to sizes that permit viral nucleocapsid delivery (~50 nm) is critical for infection, yet the mechanism of pore enlargement is not understood. Studies of influenza hemagglutinin and HIV Env-induced cell-cell fusion showed that nascent pores are reversible structures sustained by fusion proteins 123. Several lines of evidence suggest that the reliance on energy provided by viral proteins increases as fusion progresses from hemifusion to pore opening and pore enlargement steps 7884123146147148149150. First, the GPI-anchored ectodomain of influenza hemagglutinin is capable of promoting hemifusion and, with much lower probability, small non-enlarging pores 148151. In other words, lipid mixing can be readily achieved by the ectodomain anchored to the external leaflet of a plasma membrane, whereas a full-length protein is required to form expanding pores. Second, complete fusion (content mixing) appears to require a greater density of active proteins compared to hemifusion (lipid mixing) 4884147150. Third, the number of cell pairs exhibiting lipid mixing is usually greater than those forming small fusion pores, and only a minor fraction of nascent pores enlarge 148152. These observations support the notion that formation, and especially dilation, of small pores is energetically unfavorable compared to hemifusion. Thus, a greater number of active fusion proteins is required to form and sustain functional pores.

The above considerations and several lines of functional evidence 20153154155156 indicate that successful fusion is achieved through the concerted action of several viral proteins. For those class I proteins that exhibit strict coupling between 6HB formation and membrane merger 123130157, pore growth could occur through recruiting additional proteins into its edge 123. The ability to form the lowest energy 6HB structure at the pore perimeter, but not at sites of membrane apposition, would drive the pre-bundle incorporation into a nascent pore (Fig. 3). The limitation of this model is that it requires a large number of activated fusion proteins in the vicinity of a pore and is applicable only to proteins that cannot form 6HBs prior to membrane merger.

Recent work has challenged a common view that several proteins are required to form a functional fusion pore. Based on measurements of infectivity as a function of the ratio of the wild-type to a dominant-negative mutant of HIV-1 Env incorporated into virions, Yang and co-authors concluded that a single Env may mediate productive entry 32. However, this conclusion is model-dependent. The more rigorous theoretical analysis of the above data yielded a greater number of HIV-1 Env (between 5 and 8) in a fusion complex 158159. Can a single viral protein store sufficient conformational energy to cause fusion? While estimates of the energy required for pore formation are available 160161162, the energy released upon refolding into a complete trimeric hairpin (including possible interactions between MSDs and fusion peptides) has not been determined. It is also not known how efficiently this conformational energy is utilized to restructure lipid bilayers. Regardless of the energy stored in fusion proteins, a single protein might not be able to exert a force to reshape and rupture fluid membranes. There is evidence that, in order to destabilize and merge two bilayers, fusion proteins must first form an oligomeric "fence" that restricts the lateral diffusion of lipids 84.

The controversy around the stoichiometry of fusion complexes suggests that perhaps this problem should be considered in a different context. Viruses often rely on cellular signaling and actin remodeling to enhance infection 163164. For instance, HIV Env-mediated signaling via CD4 and/or coreceptors has been implicated in productive entry 183950165166167168169170 and Env-mediated fusion 131165168171. It is thus tempting to speculate that viruses may accomplish the formidable task of creating and expanding a fusion pore by hijacking the cellular machinery. In other words, viral proteins could utilize their conformational energy to promote hemifusion and to create a small pore while relying on a host cell to carry out the energetically costly step of pore dilation. For instance, VSV may undergo low pH-dependent fusion with intralumenal vesicles of early/intermediate endosomes and release its capsid into the cytosol via the constitutive "back-fusion" reaction between intralumenal vesicles and the limiting membrane of a late endosome 42. However, this two-step fusion entry model for VSV has recently been challenged 172. Thus, the role of cellular processes in the dilation of viral fusion pores has yet to be unambiguously determined.

The cytoskeleton may facilitate retrovirus entry not only by promoting receptor clustering on the cell surface 131173174175 or transport of bound viruses along microvilli to the cell body 24, but also by augmenting the fusion and early post-fusion steps (174176 and references therein). The exploitation of cellular processes to drive the energetically costly step of pore dilation could explain the ability of a few (perhaps even a single 32177) retroviral Env to initiate infection. Once a hemifusion intermediate or a small fusion pore is formed, viral capsid delivery might be augmented by cytoskeleton rearrangements and/or by membrane trafficking machinery.

Recent studies of viral fusogens revealed that structurally diverse proteins may have adopted a common "cast-and-fold" mechanism to merge membranes. Moreover, the general principles of viral fusion could be shared by proteins responsible for intracellular and developmental fusion 178179. This common mechanism is likely dictated by the physical properties of lipid bilayers and by the necessity to follow the least energy-costly membrane restructuring pathway leading to fusion without disrupting the membrane barrier function. While structures of the ectodomains or the core fragments of viral proteins showed that these proteins undergo major restructuring that culminates in formation of a trimeric hairpin, the actual refolding pathways remained conjectural. Functional studies demonstrated that viral fusion progresses through a number of distinct, reversible and increasingly unfavorable steps. The notion that formation, and especially enlargement of fusion pores, is an uphill process changes our views on how viral proteins may function. The increasing cost of forming and enlarging fusion pores indicates that viral fusogens should form oligomeric complexes capable of exerting an increasing force as fusion progresses to completion. In addition, viruses may rely on cellular machinery to enlarge fusion pores and release their capsid into the cytosol. Advances in understanding both the molecular details and unifying principles of viral protein-mediated fusion should help identify new targets for antiviral therapy and vaccine development.

6HB: six-helix bundle structure; ASLV: Avian Sarcoma and Leukosis Virus; Env: envelope glycoprotein; GPI: glycosylphosphatidylinositol; HR1 and HR2: helical heptad repeat 1 and 2 domains of class I viral fusion proteins; LAS: a lipid-arrested stage of fusion; MLV: Murine Leukemia Virus; MPER: membrane-proximal external domain of a fusion protein; MSD: membrane-spanning domain; SU and TM: surface and transmembrane subunits, respectively, of a fusion protein; TAS: a temperature-arrested stage of fusion; VSV: Vesicular Stomatitis Virus.

The author declares that they have no competing interests.

The author would like to thank Dr. Kosuke Miyauchi for critical reading of the manuscript and stimulating discussions. This work was supported by NIH R01 grants GM054787 and AI053668.