|Home | About | Journals | Submit | Contact Us | Français|
Retrovirus infection starts with the binding of envelope glycoproteins to host cell receptors. Subsequently, conformational changes in the glycoproteins trigger fusion of the viral and cellular membranes. Some retroviruses, such as avian sarcoma/leukosis virus (ASLV), employ a two-step mechanism in which receptor binding precedes low-pH activation and fusion. We used cryo-electron tomography to study virion/receptor/liposome complexes that simulate the interactions of ASLV virions with cells. Binding the soluble receptor at neutral pH resulted in virions capable of binding liposomes tightly enough to alter their curvature. At virion-liposome interfaces, the glycoproteins are ~3-fold more concentrated than elsewhere in the viral envelope, indicating specific recruitment to these sites. Subtomogram averaging showed that the oblate globular domain in the prehairpin intermediate (presumably the receptor-binding domain) is connected to both the target and the viral membrane by 2.5-nm-long stalks and is partially disordered, compared with its native conformation. Upon lowering the pH, fusion took place. Fusion is a stochastic process that, once initiated, must be rapid, as only final (postfusion) products were observed. These fusion products showed glycoprotein spikes on their surface, with their interiors occupied by patches of dense material but without capsids, implying their disassembly. In addition, some of the products presented a density layer underlying and resolved from the viral membrane, which may represent detachment of the matrix protein to facilitate the fusion process.
Entry of enveloped viruses into host cells occurs by means of fusion between the membranes of the virus and the host, an event mediated by one or more of the glycoproteins that stud the virion surface. Although viral fusion proteins fall into three structural classes (20, 53) and vary in their modes of activation, all are thought to undergo similar sets of conformational changes as fusion proceeds. Upon activation, fusion proteins first switch from their relatively compact native state into an elongated intermediate, in which a hydrophobic motif, the fusion peptide, is exposed and becomes embedded in the target membrane. Inferences concerning this “prehairpin” intermediate have mostly been indirect, based primarily on biochemical studies (17, 27, 30, 35, 42, 48). After engaging the target membrane, this intermediate is ready to transition into the postfusion state. In all classes of fusion proteins, the prehairpin conformer folds back on itself to form the hairpin. In class I proteins, which are trimeric, the predominant feature of this end state is a six-helix bundle composed of three hairpins. Formation of the hairpins is coupled with merging of the viral and target membranes and is thought to occur in a concerted process involving multiple glycoprotein spikes. First, a zone of hemifusion is created, i.e., a single bilayer comprising one leaflet from each membrane; ultimately, a fusion pore opens, allowing the viral nucleocapsid to pass into the host cell cytoplasm (20, 26, 53).
Several recent studies have used cryo-electron microscopy (cryo-EM) to examine intermediates in the fusion process mediated by the fusogenic proteins of Moloney murine leukemia virus (32, 55), herpes simplex virus 1 (36), influenza virus (15, 29), and vesicular stomatitis virus (31). However, no cryo-EM study has yet reported visualization of a prehairpin intermediate.
Avian sarcoma/leukosis virus (ASLV) is an alpharetrovirus that fuses with target cells via a two-step mechanism. Interaction of ASLV Env, a class I fusion protein, with its receptor, Tva, induces conformational changes in Env that expose its fusion loop (9, 21). Subsequent exposure to low pH induces further conformational changes in Env that mediate fusion (35, 38, 39, 51). Because interaction with the receptor induces formation of the inferred prehairpin intermediate without progressing to fusion, ASLV provides an excellent system for visualizing the target membrane-binding (prehairpin) stage of fusion, as well as the final fusion products, produced by subsequent exposure to low pH.
In this study, we analyzed successive stages of ASLV fusion with liposomes by using cryo-electron tomography (cryo-ET), a technique that allows visualization of individual pleiomorphic particles, such as ASLV virions, in three dimensions in their native state (19). This approach has already been used to characterize virion morphology and capsid polymorphism of the closely related Rous sarcoma virus (RSV) (2, 3). The experiments in this study were facilitated by the ability of sTVA, a soluble form of the receptor (which is normally membrane bound via either a glycosylphosphatidylinositol anchor, i.e., Tva800, or a transmembrane domain, i.e., Tva950), to bind to Env, preparing it for acid-inducible fusion. These observations provide new insights into the mechanism of membrane fusion and retroviral cell entry.
ASLV was produced by using DF-1 cells chronically infected with the RCASBP(A) vector system as described previously (13). Virus-containing supernatants were centrifuged at 2,500 rpm to clear debris and concentrated ~50× by centrifugation in Vivaspin-20 300-kDa cutoff concentrators (catalog number VS2052; Sartorius-Stedim). Concentrated supernatants were pelleted through a 1-ml cushion of 20% (wt/vol) sucrose by centrifugation at 175,000 × g in an SW55 rotor (Beckman) for 2 h. The viral pellets were resuspended in HM buffer (20 mM HEPES, 20 mM morpholineethanesulfonic acid, 130 mM NaCl; pH 7.5), layered on top of a 25%-to-50% (wt/vol) sucrose step gradient, and centrifuged at 50,000 × g in an SW55 rotor for 16 h. The virus-containing band at the 25%-50% sucrose interface was collected, resuspended in HM, and centrifuged at 100,000 × g for 2 h to remove residual sucrose. The final pellet was resuspended in ~200 μl of HM buffer. Its protein concentration was determined in a bicinchoninic acid assay (catalog number 23228; Thermoscientific). Soluble receptor (sTva) was produced as described previously (52). C-helix peptide R99 was produced as described in reference 13.
To produce fluorescent liposomes, lipid mixtures of phosphatidylcholine (PC; number catalog 830051P; Avanti), phosphatidylethanolamine (PE; catalog number 841118P; Avanti), sphingomyelin (SPH; 860061P; Avanti), cholesterol (CH; C-3137; Sigma), rhodamine phosphatidylethanolamine (RH-PE; 810146; Avanti), and (N-7-nitrobenz-2-oxa-1,3-diazol-4-yl)-1,2-dihexadecanoyl-sn-glycero-3-phosphatidylethanolamine (NBD-PE; N-360; Molecular Probes) at a ratio of 1:1:1:1.5:0.045:0.11 were dried with N2, lyophilized overnight, and resuspended in HM buffer to a 2 mM lipid concentration. The liposome suspension was freeze-thawed five times, extruded 30 times through a 100-nm filter, and stored at 4°C. Liposomes for electron microscopy were produced as described above, except that PC, PE, SPH, and CH were mixed in a 1:1:1:1.5 ratio and resuspended at a 20 mM lipid concentration.
Fusion was measured essentially as described in reference 10. For the standard assay, 20 μl reaction mixtures (in HM buffer) were set up in wells of a 384-well plate as follows: 20 to 30 μg of ASLV and sTva (at 1 μM) were mixed for 30 min at 4°C. Fluorescent liposomes (final concentration, 25 to 50 μM) were added, and the plates were incubated for an additional 5 min at 4°C. At this time the samples were incubated for 15 min at 37°C (to generate virus-liposome complexes bridged by ASLV Env in its prehairpin state). The plate was then placed in a fluorescent plate reader, and the baseline NBD fluorescence (excitation at 460 nm, emission at 540 nm) was measured for 5 min at 37°C. Unless otherwise specified, the samples were then acidified to pH 5.0 and the fluorescence was monitored for 10 min at 37°C. The maximum possible NBD fluorescence was determined by adding NP-40 to a final concentration of 1% and measuring fluorescence at 37°C for 15 min. The percent fusion was calculated as follows: (FpH − F0)/(FT − F0) × 100, where F0 is the baseline fluorescence (pH 7.5), FpH is the fluorescence value averaged over the plateau at pH 5.0, and FT is the fluorescence at infinite dilution (after disruption of the membranes with 1% NP-40).
For EM analysis, 50 μg of ASLV was incubated for 30 min on ice with sTva (0.3 μM). Liposomes were added (0.5 mM), and the samples were further incubated on ice for 20 min. The mixture was then incubated at pH 7.4 for 30 min at 37°C to allow formation of virus-liposome complexes. Where indicated, samples were then treated at pH 7.4, 6.1, or 5.0 for 5 min at 37°C. In the samples without receptor, HM buffer was used instead of sTva.
Samples were mixed 1:1 with a suspension of 10-nm colloidal gold particles (Ted Pella, Redding, CA) to serve as fiducial markers, applied to Quantifoil R2/2 holey carbon grids (SPI, West Chester, PA) and plunge-frozen into liquid ethane by using a Vitrobot (FEI, Hillsboro, OR). Tilt series of projections were acquired on a Tecnai-12 (FEI) microscope, operating at 120 keV, using SerialEM (34). The microscope was equipped with an energy filter set in zero-energy-loss mode (slit width, 20 eV). Data were recorded on a charge-coupled-device camera, at 2,048 by 2,048 pixels (Gatan, Warrendale, PA). At least 500 viruses (~100 images) per condition were analyzed in two-dimensional projections. For tomography, a magnification of ×38,500 was employed, corresponding to a pixel size of 0.78 nm. Images were recorded at −4 μm and −6 μm nominal defocus, in 1° or 2° angular increments, over an angular range of ~−66° to 66°. The total electron dose for each tilt series was ~80 electrons/Å2 or lower. The numbers of tilt series acquired from the samples containing either isolated virions or virions mixed with receptors and liposomes at pH 7.4 and pH 5.0 were 4, 28, and 6, respectively.
Tilt series projections were aligned using the IMOD program (28), with the gold particles serving as fiducial markers. Tomograms were reconstructed by a weighted back-projection method (46) and then noise-filtered using two different procedures in Bsoft (22). For visualization, bilateral filtering (24) was applied to the reconstructions, with the following parameters: kernel size, 15 pixels; σ of spatial filter, 5 pixels; σ of the range filter, equal to the standard deviation (SD) of the tomogram. Stronger deionizing by nonlinear anisotropic diffusion (16) was applied to the subtomogram data extracted for the statistical analysis of the receptor-primed interaction between virions and liposomes at neutral pH. The edge-enhancing diffusion parameter was set to 1/100 of the standard deviation of the subtomograms, and 40 iterations were performed. Tomogram quality was assessed by using the NLOO approach (4) to estimate the in-plane resolution of three virions extracted from each tomogram. In each case the resolution was around 6 nm, based on a threshold criterion of 0.5. Gray-scale sections were prepared using the tools available in Bsoft (22), while segmentation was performed within Amira (Visage Imaging).
Glycoprotein spikes on 14 isolated virions and on 24 virions in receptor/virion/liposome complexes were located manually, using the 3dmod utility in IMOD (28), and their coordinates were recorded. Data for the junctional spikes and nonjunctional spikes were stored separately. For each spike, we measured the distances between it and its three nearest neighbors. The same data were used to calculate the nearest neighbor index (NNI) (5). Specifically, we used the K-order NNI, which is the ratio of observed k-th nearest neighbor distance to the average random distance, i.e., the distance that would be observed if the points were randomly distributed (6). Thus, a random distribution of points gives an NNI value close to 1, while values less than 1 are evidence of clustering and values greater than 1 are indicative of a uniform pattern. Our NNI was measured as the average of the K-order NNI, with K ranging from 1 to 3. The NNI for spikes located at the junctions was, on average, equal to 1.8, while the NNI for spikes distal to the junctions was equal to 1.1. The same value of 1.1 was obtained for the NNI measured on the spikes decorating isolated virions. All the analyses were performed with the statistical software R (45; http://www.R-project.org/) and its packages fractal (8), misc3d (14), geometry (1) and tripack (47), with custom scripts written in its programming language.
The position coordinates of the glycoproteins used for the statistical analysis (see above) were also used for subtomogram averaging, after an initial screening. Junctional (n = 672) and nonjunctional (n = 2,392) spikes were extracted from raw tomograms in volumes of 42 by 42 by 42 voxels and 34 by 34 by 34 voxels, respectively. A total of 1,197 Env spikes in the native conformation were located manually on a tomogram of isolated virions without receptors and then extracted in volumes of 44 by 44 by 44 voxels. For each virion, an estimate of its center was used to determine the initial orientations of all of its spikes. These subtomograms, each containing one spike and the adjacent regions, were subjected to multiple (up to 8) iterations of alignment and averaging. As initial templates, a computationally generated cylinder on a plane or a cylinder between two planes was used for nonjunctional spikes or junctional spikes, respectively. The correlation metric used to determine the spike orientations was compensated for the missing wedge, and after two iterations the noise in real space was masked from both the template and the raw particles, using their most recent solutions to properly orient the mask. Alignment and averaging procedures were performed with a set of Python programs that were interfaced to routines from Bsoft. Three-fold symmetry was imposed on the averages after each iteration. After completing the iterations, the aligned particles were classified using a maximum likelihood approach, as implemented in xmipp (50), to discard less consistent particles, and the final averages were then calculated. The numbers of contributing particles were 513 (junctional spikes), 2,000 (nonjunctional spikes), and 680 (native Env spikes). Resolution was estimated by calculating the Fourier shell correlation curve between the reconstructions from two half-data sets (49), resulting in values between 4.0 and 4.5 nm at a cutoff of 0.5.
Incubation with sTva at neutral pH and temperatures above 25°C induces Env on ASLV virions to change their conformation (11, 18, 35) and bind to liposomes (9, 21). Subsequent shifting to a low pH triggers fusion (35, 39, 51). It has also been shown that ASLV that has become endocytosed into cells in the presence of NH4Cl (via interaction with the receptor, Tva 800) traffics into endosomes, but after 6 h it can still proceed to infect these cells when the NH4Cl is washed out (40). This suggests that Env can persist in its receptor-triggered state for extended periods.
In order to determine how long receptor-triggered virions remain fusion competent, we used a fluorescence-based fusion assay (Fig. 1). ASLV was incubated with sTva in the presence or absence of liposomes containing a pair of lipids capable of FRET (RH-PE and NBD-PE), at neutral pH at 37°C for various times and then switched to pH 5.0. Fusion was measured from the gain in NBD emission, as the FRET lipids were diluted upon merging of the liposomes with viral membranes. We found that ASLV/sTva could be maintained in the presence of liposomes for at least 8 h at 37°C without a significant loss in fusion activity (Fig. 1A, black columns). If the C-helix peptide analog R99 was added just prior to acidification, fusion was strongly inhibited (Fig. 1A, hatched columns), even if the ASLV-liposome complexes had been incubated for 8 h at 37°C. (C-helix peptides are derived from Env and inhibit fusion by binding to the N-terminal heptad repeats of Env in its extended prehairpin conformation, thereby blocking formation of the six-helix bundle [13, 17, 23, 37, 42, 54].)
The latter observation indicated that Env had remained in the prehairpin conformation throughout. However, if liposomes were not added to sTva-ASLV complexes incubated for various times (at pH 7.5 and 37°C) until 5 min prior to acidification, a progressive loss of fusion activity was observed (Fig. 1A, gray columns). This inactivation might reflect aggregation of Env (in virions) or insertion of the fusion loop into the viral membrane, as has been observed for influenza virus hemagglutinin and paramyxovirus F proteins (7, 44). We conclude that sTva-ASLV-liposome fusion intermediates remained stable for at least 8 h at 37°C (Fig. 1A). This time could be increased to at least 32 h if the prefusion incubation took place at 4°C (Fig. 1B).
The observed longevity of the virion-liposome complexes encouraged us to examine them by cryo-EM. After incubation with sTva and liposomes at 37°C and neutral pH, more than 90% of the virions were observed to be in contact with two or more liposomes (Fig. 2A). These interactions were accompanied by a change in membrane curvature in some liposomes, reflecting their greater malleability. In contrast, in the absence of sTva, fewer than 5% of virions were seen to be in contact with two or more liposomes (Fig. 2B).
We then performed cryo-ET on receptor-virion-liposome complexes formed at neutral pH, whose susceptibility to fusion blocking by R99 (see above) indicated that Env was in the prehairpin state. From a total of 11 tomograms, we identified 59 virions (out of 101 total) that were clearly engaged with one or more liposomes (Fig. 2C to toI).I). There were a total of 134 such interactions, and the highest number of interacting liposomes per virion was 7. The interactions involved extensive interfaces, where multiple Env spikes were observed to be present, as illustrated by the 8-nm-thick slice through a tomogram in Fig. 2C (arrowheads). The spikes appeared closely packed, giving almost continuous ribbons of density. This clustering was not observed for Env spikes outside the virion-liposome interfaces (Fig. 2E and andF),F), nor on control virions that had not been mixed with receptors and liposomes (Fig. 2J and andK).K). As already observed from the cryo-electron micrographs (above), the interfaces exhibited alterations in liposome curvature. In most cases (~80%), the liposome membrane was visibly flattened where it bound to the virion (Fig. 2D). In a few cases (~8%) where the interface was larger, the liposome was molded around the virion (Fig. 2H and and33).
Next, we analyzed in greater detail the most clearly defined junctions (n = 39, involving 24 virions). The intermembrane spacing (virus to liposome) was uniformly 16 nm, center to center. Env spikes in these junctions appeared as globular densities, approximately midway between the membranes. Env spikes in nonjunctional regions had much the same appearance, although they appeared slightly larger. The number of spikes per junction varied from 4 to 47 (median, 15), roughly in proportion to the size (area) of the junction. (The total number of spikes per ASLV virion in these preparations was determined to be 111 ± 33 [SD], more than for the related Rous sarcoma virus . The spread represents almost entirely particle-to-particle variability, not experimental error.) Of note, the concentrations of spikes in junctions appeared higher than elsewhere in the viral envelope (Fig. 3; see also Movies S1 and S2 in the supplemental material). To examine this observation on a quantitative basis, we determined the distributions of interspike spacings in both junctional and nonjunctional regions (Fig. 4). This analysis yielded average interspike spacings of 11 ± 3 nm and 19 ± 6 nm, respectively. For comparison, we repeated the analysis on the glycoproteins identified on 14 virions in the absence of receptors and liposomes. In this case, the interspike spacing was 19 ± 7 nm. Use of the nearest neighbor index (see Materials and Methods), which measures the degree of spatial dispersion, confirmed that the spikes tended to be more closely packed in junctions than elsewhere. These data implied that Env is able to diffuse within the viral membrane and be recruited to the junctions.
In order to compare native Env with the receptor-bound and membrane-embedded prehairpin conformations, we performed subtomogram averaging for each state. For native Env, we averaged spikes taken from virions that had not been mixed with receptors or liposomes. Thus visualized (Fig. 5A), the molecule appeared as an oblate density, 8 nm in diameter and 5 nm thick, and significantly offset from the membrane. The observed offset implied that the globular domain is connected to the membrane by a thin stalk that was not directly visualized at the current resolution (~4 nm) (see Discussion). Then we analyzed the spikes in receptor-virion-liposome complexes formed at neutral pH. The overall shape and dimensions of Tva-bound spikes not interacting with liposomes (Fig. 5B) were similar to those of the native glycoprotein, except that they appeared slightly less wide (diameter, 6.5 nm; thickness, 5 nm) and less defined, despite the expected additional mass of three receptors (50 kDa for three sTVAs, added to the 150 kDa of three glycosylated Env ectodomains). This suggests that the conformational change triggered by receptor binding involves some disordering.
In junctions, the Env spikes are assumed to be in the prehairpin state. Here, a globular domain remained visible, although it was smaller still (diameter, ~6 nm; thickness, ~4 nm) and apparently embedded in a continuous sheet of density that presumably arose from averaging the close-packed Env ectodomains. This domain was centered between the two membranes (Fig. 5C), which are 16 nm apart (see above). Also, the liposome membrane density was markedly lower than that of the virion envelope, presumably reflecting that the latter lipid bilayer is backed by the closely apposed matrix protein layer.
According to FRET-based fusion assay results, lipid mixing between virions and liposomes proceeds in two stages as the pH is lowered: the first stage occurs at only a slightly acidic pH, starting at pH 6.7, and the second stage occurs when the pH falls below 5.8 (10). With cryo-EM, we detected some fusion when the complexes that formed at neutral pH were shifted to pH 6.1, but the incidence was low (5 out of ~500 observed). Despite their rarity, fusion products were readily recognizable as such (Fig. 6A, right insert), because their membranes were fringed with glycoprotein spikes (like virions) and their interiors were partially filled with dense material (as for virions) and partially void (as for liposomes). Fusion was more prolific when the pH was lowered to pH 5.0. Under this condition (Fig. 6C), approximately equal numbers of intact virions and fusion products were observed; from this we inferred that ~50% of the virions underwent fusion. Fusion products tended to be larger than virions and to have a greater fraction of their interiors as voids (Fig. 6C; see also Movie S3 in the supplemental material). If receptors (sTva) were omitted, no fusion was observed at pH 6.1 or 5.0, and virions and liposomes showed little tendency to interact, as at neutral pH (Fig. 6B and andDD).
The differences between virions and fusion products were better observed in tomographic slices (Fig. 6E to toK).K). The fusion products analyzed (n = 19) had an average diameter of 146 ± 20 nm, consistent with fusion of an average-sized virus (diameter, ~130 nm) with a smaller liposome. All postfusion complexes had glycoproteins on their surface and patchy internal densities, which we took to be the contents, somewhat redistributed, of the fused virions (Fig. 6I and andJ,J, arrows). We did not observe more-advanced fusion intermediates or membrane dimpling at virus-liposome interfaces. As such, our data concurred with a recent report that fusion follows rapidly upon acidification of viruses whose glycoproteins are trapped in the prehairpin conformation (43).
A notable feature of fusion products is the absence of intact capsids. In unfused virions, capsids are clearly visible in the tomographic slices (Fig. 2D and andEE and and3A).3A). These observations suggested that conditions in the postfusion lumen favor disassembly.
Another property of the postfusion products was that most of their envelopes appeared thinner than in virions (Fig. 6F to toK;K; cf. Fig. 2D to toK)K) and had essentially the same thickness as liposomes (Fig. 2C). (In native ASLV virions, the envelope appears thicker because the matrix layer is closely apposed to the bilayer and resolved only marginally if at all in the tomograms.) Also of note, about 25% of the postfusion complexes (n = 5) showed a layer of density that tracked but was clearly resolved from the interior surface of the membrane (Fig. 6J and andK,K, arrowheads). This layer covered from ~25 to 50% of the interior surface of the postfusion particle.
Viral glycoproteins undergo conformational changes in order to induce fusion of viral and cellular membranes (20, 53). One of the proposed intermediates in this process is the preharpin conformation; however, although many reports point to its existence, based on biochemical approaches (17, 35, 42, 48) and by negative-staining EM (27, 30), a prehairpin has never been visualized directly by cryo-EM. Subtomogram averaging (Fig. 5) conveyed this state for ASLV-Env spikes that we know, based on the inhibitory effect of R99, to be in it. Although these data were of limited resolution (to ~4 nm), they yielded two incisive observations.
The mass of Env-sTva should be ~50 kDa larger than that of the Env ectodomains alone, although sTva has been observed to dissociate after ASLV binds to membranes (9, 21). Whether or not sTva is retained, a definite size disparity is observed, and it appears to arise from partial disordering of Env upon its binding to the receptor. This interpretation is consistent with the observation by Delos et al. (11) that sTva binding by SU (the receptor-binding portion of Env) induces a conformational change in SU that involves a loss of the α-helical structure, occlusion of the hydrophobic surface(s), and occlusion of tryptophan residues.
By analogy with other class I fusion proteins, in particular with the Ebola virus GP2 subunit (12), we expected transmembrane domains in the prehairpin conformation to adopt a trimeric coiled-coil stalk made up of the N-heptad, chain reversal region, and the C-heptad. Accordingly, the observed globular density (Fig. 5C) should have been contributed by the SU domains, which appeared to retain some organized structure in the prehairpin conformation. The function of the SU domains after receptor binding remains unclear, although they could be involved in SU-SU lateral interactions that recruit more Env spikes into a growing junction and/or facilitate the coordinated action of the multiple trimers in producing an effective fusion pore (see below).
It is generally held that multiple copies of Env are required for fusion, as with other viral fusion proteins, although the exact numbers required remain unclear (20, 33, 53). Our data indeed pointed in this direction. We observed a 3-fold increase in the concentration of Env spikes in virus-liposome junctions compared with nonjunctional regions. Moreover, statistical analysis pointed to a tighter packing of spikes within junctions. We can imagine two scenarios that would have this concentrating effect. Importantly, both require Env to be mobile within the viral membrane. First, the buildup of Env in junctions could be due to trapping of the spikes once their fusion loops have inserted in the liposome membrane, causing a decrease in their lateral mobility. Second, the close packing of Env in junctions could be due to attractive intertrimer SU-SU interactions. Such interactions could help to organize spikes into higher-order assemblies, which appear to be needed to generate an effective fusion pore.
Although the nominal resolution of the averaged tomographs was modest, their high signal-to-noise ratio and a priori information about the components allowed certain more-detailed parameters to be extracted. The thickness of the target membrane should be 4.5 nm, typical of a lipid bilayer, and the position of its center is specified quite precisely by the tomogram (Fig. 5C). Similarly, we estimated the viral envelope thickness as 7 nm (4.5 nm for the bilayer plus an estimated 2.5 nm for the closely associated matrix protein layer). As the centers of the two membranes are 16 nm apart, there should be about 10 nm between the surfaces of the respective bilayers. The thickness of the globular Env domain is not known exactly, but 5 nm would seem a reasonable approximation (an oblate ellipsoid with a 7-nm major axis and 5-nm minor axis; the density of protein would have a mass of about 160 kDa). This leaves 5 nm for the combined separations of this domain from the two membranes, and as the domain is centered between them, the connections should have the same length, about 2.5 nm. We performed computational simulations with model structures (Fig. 7) that suggested that these conclusions are fairly robust to variations in the actual thickness of the globular domain, supporting the supposition that two narrow connections, most likely both coiled coils, connect the remaining globular domain to the viral membrane and to the fusion peptides embedded in the target membrane. Conversely, no placement of a larger globular domain closer to the target membrane could reproduce the observed structures.
The ALSV-liposome fusion products provided some insight into this process. On addressing virus-liposome interactions at neutral pH, we observed that only liposome membranes were deformed. Therefore, the viral envelope is more resistant to deformation than the liposome membrane, probably on account of its backing layer of matrix protein. (These observations also agree with what has been seen for influenza virus under similar conditions .) It follows that, for fusion to take place, the matrix protein layer should at some point lose its rigidity or detach from the viral membrane. In this context, we found the membranes of postfusion vesicles to be thinner than viral envelopes, suggesting that the matrix protein had already detached. If not solubilized, this protein should form part of the patchy internal density seen in these particles (Fig. 6F to toKK).
Approximately 25% of these membranes showed a layer of density tracking the membrane on the inside and resolved from it. It is plausible that this layer represents matrix protein at an intermediate stage of detachment. This could be an effect of pH alone, but if host lipids were to have a lower affinity for matrix protein than viral membrane lipids, detachment would be promoted by the outward diffusion of host lipids from an incipient fusion pore. On the other hand, this layer of density is unlikely to represent capsid protein, as its shape does not resemble capsids as originally assembled, and conditions inside postfusion products appear not to be conducive to capsid assembly.
We thank Sue Delos for helpful discussions during the initial stages of this work and Bryan Fellman at the M.D. Anderson Cancer Center for help with the statistical analysis.
This research was supported in part by the Intramural Research Program of the National Institute of Arthritis and Musculoskeletal and Skin Diseases. Work in the White laboratory was supported by NIH AI22470. M.B. was supported in part by an Infectious Disease Training Grant (5T32 AI07046-27) from the NIAID.
Published ahead of print 29 August 2012
Supplemental material for this article may be found at http://jvi.asm.org/.