The Dynamic Envelope of a Fusion Class II Virus

Semliki Forest virus is among the prototypes for Class II virus fusion and targets the endosomal membrane. Fusion protein E1 and its envelope companion E2 are both anchored in the viral membrane and form an external shell with protruding spikes. In acid environments, mimicking the early endosomal milieu, surface epitopes in the virus rearrange along with exposure of the fusion loop. To visualize this transformation into a fusogenic stage, we determined the structure of the virus at gradually lower pH values. The results show that while the fusion loop is available for external interaction and the shell and stalk domains of the spike begin to deteriorate, the E1 and E2 remain in close contact in the spike head. This unexpected observation points to E1 and E2 cooperation beyond the fusion loop exposure stage and implies a more prominent role for E2 in guiding membrane close encounter than has been earlier anticipated.

Membrane fusion is important in biology to allow exchange of materials between compartments enclosed by lipid bilayer membranes. Naturally, viruses utilize such ways of genome delivery. Two different mechanisms of virus fusion are observed: the Class I mechanism, seen in viruses where the fusion protein structure is dominated by ␣-helix folding, like the influenza and the HIV, and the Class II mechanism, represented by viruses where the envelope proteins are dominated by ␤-sheet structures like the Tick Borne Encephalitis virus and the Semliki Forest virus (SFV) 2 in the Flavi and Toga virus families, respectively (1,2).
SFV infects its host cells by receptor-mediated endocytosis and fusion within acidic endosomes. It has been demonstrated that the membrane fusion of SFV is strictly dependent on the exposure of the virus to a low pH (3)(4)(5) and on the presence of cholesterol in the target membrane (6). Experimental mimicking of the endosomal acidification has shown that low pH triggers a series of conformational changes in the virus spikes (7)(8)(9)(10). Mild acidification leads to exposure of the fusion peptide at the virion surface (11). The moderately hydrophobic fusion peptide loop can then interact with the target membrane and initiate fusion (12,13). Optimal fusion kinetics of SFV has been observed at pH 5.5, while the pH threshold for fusion activation is reported to be around pH 6.2 (3,5).
Early data on the alphavirus imply that the fusion process involves a low pH-induced conformational change in the envelope whereby homotrimers of E1 are formed (9,10,14). This is supported by recent studies on the soluble ectodomain of the fusion protein, E1*. The crystal structure of E1* reveals an elongated molecule, mainly folded into ␤-sheets and interconnecting loops (15,16). The structure resembles that of the TBE glycoprotein E ectodomain (17). The molecule comprises three domains; a ␤-sheet-folded domain I (DI) located in the central part of the molecule inbetween the projecting domain II (DII), carrying the putative fusion peptide loop (18,19) at the distal end, and the immunoglobulin-like domain III (DIII) (15,16). The C terminus of DIII in the full-length E1 protein connects to the transmembrane and submembrane domains of the molecule. From fitting of three E1* molecules per spike into the Cryo-EM density map of SFV, it is implied that DI and DIII constitute the major portion of the shell region. Here the DIII spreads around the spikes and align with those from neighboring spikes, leaving openings at the 5-and 2-fold axes. In the model the DII rises at a slant from the shell forming the sides of the spike and the lower parts of its bulky head (15).
Exposure of the soluble E1* to a low pH in the presence of a target membrane results in a membrane-inserted, vertically oriented homotrimer (E1*HT) (12) with the fusion loops inserted into the target membrane. In the E1*HT, the core is formed by DI and DII, while the DIII is translocated to partly cover the hinge region in DII of a neighboring monomer. The C-terminal sequence is following the cleft between the monomers pointing toward the membrane-inserted fusion peptide. This fold-back generates a hairpin-like configuration with the fusion peptide and envisioned transmembrane region at the same end of the molecule (2). An E1 homotrimer has earlier been observed to form in virions under similar acidic condi-tions (9,10,14), why the E1* trimerization would reflect a native phenomenon.
To reveal stages of transformation of the SFV into a fusogenic configuration, we have exposed samples of the virus to a series of decreasing pH. The acid-treated SFV particles were subsequently plunged frozen in vitreous ice, and micrographs acquired. The micrographs were digitized, and the virus structures reconstructed. We earlier reported a gradual increase of the virion diameter after short incubations at pH 7.4, over pH 6.2, and down to pH 5.9 (8). The expansion occurred concomitant with a reciprocal relocation of assumed E1 and E2 densities in the shell and lower part of the spike region. In the present study, we determine the SFV structure at pH 5.8, and improve earlier reconstructions of the virus at pH 7.4 and 5.9. By mapping high density centers in the structure, assigned as density nodes, the relocations of stably folded subdomains in the glycoproteins could be followed at the different pH. The results obtained point to unexpected features of the alphavirus fusion mechanism, where the E2 protein may play an essential role after the fusion loop is exposed.

EXPERIMENTAL PROCEDURES
Virus Purification and Low pH Treatment-The SFV was propagated and purified essentially as described previously (8,11,20). Briefly, monolayers of BHK-21 cells were grown in 225-cm 2 T-flasks using Glasgow's modified Eagle's minimal essential medium (Invitrogen), supplemented with 5% fetal calf serum, 10% tryptose phosphate broth, 2 mM glutamine, 20 mM HEPES (Sigma), and 20 g/ml cholesterol. At 90% confluency, the cells were infected with SFV (clone pSFV4) and further incubated at 37°C. Virus was harvested at 18-h postinfection. The virus-containing supernatant was cleared from cell debris, and the virus pelleted at 17,000 ϫ g, 4°C for 18 h. The virus pellet was soaked in TNM buffer (50 mM Tris-HCl, 50 mM NaCl, 10 mM MgCl 2 pH 7.4) and applied to isopyknic centrifugation in a 10 -30% (w/w) potassium tartrate gradient in 0.1 M MOPS, pH 7.4, run at 100,000 ϫ g for 5 h at 4°C. The virus was eluted from the gradient, diluted with TNM buffer, and pelleted by centrifugation at 100,000 ϫ g for 5 h at 4°C. The finally obtained virus pellet was soaked in TNM buffer and kept at 4°C to avoid structural damage. The quality of the virus was controlled by SDS-PAGE and negative stain EM.
For acid treatment, an adequate concentration of the virus was mixed with a predetermined volume of 50 mM MES, 50 mM NaCl, 10 mM MgCl 2 , pH 5.6, to reach the final pH. This was determined from measurements in a large volume model system to obtain an accurate measure of the final pH. After 60 s of treatment, the sample was applied on a cryo grid, plunge-frozen into liquid ethane, transferred to liquid nitrogen, and subsequently to liquid helium for microscopy imaging.
Cryo-EM and Three-dimensional Reconstruction-Images of the frozen-hydrated SFV particles were recorded on a Kodak SO163 electron film using a JEM3200F field emission gun (FEG) transmission electron microscope equipped with the top-entry type liquid helium cold stage at a magnification of ϫ40,000 and an accelerating voltage of 300 kV. Micrographs at 1-m underfocus were digitized on a Zeiss SCAI scanner using a 7-m step size that was further bin-averaged to give a step size of 14 m corresponding to 3.5 Å per pixel at the specimen. Micrographs from every pH condition displaying well separated virus particles were analyzed. Individual intact particles were manually boxed out to generate a stack of more than 3,000 particles using the RobEM software package. The determination of orientation parameters and their iterative refinement were done using the model-based polar Fourier transform routines (21,22). The final reconstructions were computed with low-pass filtering set to suppress frequencies higher than 1 ⁄15 Å Ϫ1 , which is within the first zero of the contrast transfer function (CTF). The resolution was determined by the Fourier shell correlation at 0.5 coefficient cut-off and ranged from 11 to 15 Å in the different samples. The structures to be compared were recalculated to the same resolution of 15 Å. The number of particles included in the computation of the different reconstructions were 834 (pH 7.4), 730 (pH 5.9), and 1,205 (pH 5.8) with orientations well covering the asymmetric unit. The threedimensional visualization was mainly carried out with the Iris explorer software (NAG, Inc., Downers Grove, IL) with custom-made modules. 3 Fitting the X-ray Atomic Structure of E1* into Cryo-EM Density Maps of SFV at Different pH-The atomic structures of the external domain of SFV glycoprotein E1 (E1*) and its homotrimer (E1*HT), were obtained from the Protein Data Bank (PDB accession codes: 2ALA and 1RER). Our Cryo-EM density maps were visualized using the program O run on Linux workstations. The E1* crystal structure (PDB: 2ALA) (16) was fit into our CTF-corrected 11-Å resolution density map of SFV at pH 7.4. To achieve a good fit of E1* also in the low pH density maps we initially applied the entire E1* as one rigid body. However, the fit was not appropriate unless the E1* structure, which represents a neutral form, was treated as three independent rigid bodies (DI, DII, and DIII). These three rigid domains were fit into low pH density maps based on the match of the gravity centers between the density and coordinates. The evaluation of the fit was done considering that the acid configuration would originate from the E1 domain fitting in the pH 7.4 density map and the constraints implied by the icosahedral symmetry of the whole virion. The rigid domains were manually rotated about the hinge axes to fit into the Cryo-EM-derived densities. The E1* fitting procedures were based on the maximal fit of the independent domains into four quasi-equivalent E1* monomers densities. Consequently, the independent fitting generated our E1* model of the acid conformation. The final optimized fitting was judged both by visual inspection and program O calculations.

RESULTS
Morphological Features, a pH-dependent Swelling-Threedimensional SFV reconstructions derived from particles exposed to pH values of 7.4, 5.9, and 5.8 were compared at the similar resolution to reveal conformational changes related to the transformation into a fusogenic mode. Whereas an appar-ent feature of the acid-treated virus is a swelling, the overall particle morphology with icosahedral symmetry and T ϭ 4 arrangement of the 80 tri-lobed spikes is essentially retained (Fig. 1). The increased diameters are evident by comparing the SFV structures in cross-section views and in one-dimensional radial average density plots (Fig. 1B). The comparison shows that the diameters of the virus particles gradually increase as the pH is lowered. This not only occurs down to pH 5.9, as was earlier observed (8,11), but further down to pH 5.8. Thus, particles expand from 707 to 721 Å and 727 Å in diameter in response to 60-s treatments at pH 5.9 and 5.8, respectively. As seen in the one-dimensional plot (Fig. 1B, bottom panel), the lipid bilayer essentially retains its thickness and its radial loca-tion, whereas the external protein layer expands. This expansion comes with (i) an elongated limb 4 region, the shell moves about 5 Å (pH 5.9) and 7 Å (pH 5.8) out from the lipid bilayer, (ii) an expanded shell surface area, and (iii) an elongated stalk region where the narrow base becomes more elongated. The height of the stalk region measures 19 Å in the neutral virion and elongates to 21 and 22 Å in the pH 5.9-and pH 5.8treated virions, respectively (Fig.  2). As found earlier (8), the openings at the 2-and 5-fold axes in the glycoprotein shell are enlarged in the acid structures (Fig. 1A).
From a more detailed examination of the 3-fold spikes of the control and acid structures, a series of differences are observed, as demonstrated in Fig. 2, comparing the spikes of the virus at pH 7.4 (control) and pH 5.8 (acid). As the whole shell-spike layer is lifted outwards, away from the lipid bilayer, the elongated limbs become thinner. A separation of elongated structures in the stalk region can be seen in the acid structure ( Fig. 2A). The width of the spike is narrowed from 100 to 95 Å, as seen in the side and top views of the 3-fold spikes. In addition, the central hole of the spike is widened in the acid structure (Fig. 2B).
Fitting of E1* Crystal Structure into Cryo-EM Density Map to Assign E1-and E2-related Density Nodes-In a protein, secondary structures like ␤-sheets represent relatively nonflexible atomic arrays (23)(24)(25). These would be the plausible source of high density centers (nodes), seen in the Cryo-EM-derived structure at high stringency. We assign the individual nodes by letters from top of the spike to the shell region, as shown in Figs. 3 and 4. When co-localization of stable secondary structures with density nodes are taken into account in molecular fit-FIGURE 1. Cryo-EM reconstructions of SFV structure at different pH. Three-dimensional reconstructions derived from Cryo-EM image data of SFV particles that were prepared at pH 7.4 (Control) and further treated at pH 5.8 (Acid) are shown. The surface renderings at 100% mass shown in A represent the structures of the control and the acid-treated virions viewed along the 2-fold axes. The quasi 3-fold (q3) and 3-fold (3) spikes are indicated. The two structures are corrected to the same resolution (ϳ15 Å) and are presented with the radial color-code indicated. The particles have Tϭ4 icosahedral symmetry with 80 tri-lobed spikes (blue) protruding above a protein shell (green). The outer layer of the membrane (yellow) is seen through holes in the shell. After treatment at pH 5.8, the surface morphology is essentially retained, although the particles have expanded from 707 to 727 Å in diameter. The openings in the shell at 2-and 5-fold axes are widened in the acid structure, confirming earlier observations (8). From the shell, the protein structure connects to the outer density of the virus membrane via a limb region, by some authors referred to as the stem region (Footnote 4), and penetrates the bilayer membrane to contact the interior nucleocapsid, as evident from the cross-section view (B, top). The gray scale image shows the control data, with the ϭ 1 contour lines indicated (green). The corresponding contour of the acid structure is overlayered (red). A one-dimensional radial average-density plot is included (B, bottom); the curves represent virus treated at pH 7.4 (hatched), pH 5.9 (dotted), and pH 5.8 (solid). The plot is matched to the named layers of the virus structure, seen in the cross-section. Note that the radii of the nucleocapsid (NC) and membrane bilayer (inner, IL; outer, OL) remain at close to control levels in all three structures, whereas the external domains are moved outwards in response to a decrease in pH.
ting or modeling into a Cryo-EM density map, this provides accuracy in the fitting.
A 3-Å refined atomic resolution structure of soluble external domain of E1, E1*, is available in the Protein Data Bank (PDB: 2ALA, Roussel et al. (16)). This is a fine-tuning of the earlier published Lescar et al. structure (15). In the SFV, the domains DI and DIII of the E1* fit well in the shell, around the 2-and 5-fold axes, whereas the DII packs with a dimeric interaction between the spikes. Thus, the DII of the three E1* molecules that fit in the 3-fold spike are in contact with their quasi-2-fold (q2-fold)-related counterparts at the interfaces between projections (15).
The fitting of the E1* crystal structure into the shell-spike layer of our Cryo-EM-derived SFV structures points to a similar general organization of nodes and ␤-sheet domains within the contours of the molecule (Fig. 3). The fitting locates the DI and DIII to the shell region of the external SFV structure in an orientation such that the C terminus in DIII (Leu 384 ) is connected to the limb. The DII of the model fit atomic E1* crystal structure follows a path on the outside of the spike toward the spike head (Fig. 3). Density analyses imply that DI of E1 contributes to nodes g and h, and the DIII to node i. Furthermore, based on relative distances, the first part of the long DII, with its twisted ␤-sheet structure, would fit with node e, while, again based on distance calculation and the assumed rigidity of the E1* molecule, node c in the bottom of the head would represent the distal double ␤-hairpin part of the DII. Looking back at the fitting of E1* in the spike structure, it is evident that the fusion loop and the cholesterol determinant ␤-turn at the tip of the molecule, would be located distal to the outermost region of node c.
Like the E1 protein, E2 would be folded with a series of stable domains, giving rise to density nodes in the spike structure. If the fitting of E1* holds, the remaining density of the spike, with nodes a, b, d, and f would be formed by the E2 protein (Fig. 3). Nodes a and b are in the bulky head of the spike, and node d is the wing connector. Below node d, node f is an elongated structure that follows nodes e and g down to the shell. Thereby E2 would pass the shell region inside node i of E1. Therefore, E2 would be an elongated protein that follows an essentially vertical path inside E1, and covers E1 in the spike head. The E1-E2 dimeric heads are connected into a trimeric structure by node d of E2. Based on these node assignments, we will discuss the interplay of the two proteins during pre-fusion rearrangements induced by a lowered pH.
Demonstration of pH-dependent Node Relocation in the SFV Spike Structure-To demonstrate the relocation of the glycoproteins enforced by a low pH, the 3-fold spike is virtually cut in a vertical plane, following the orientation of one E1 subunit according to our fitting of E1*, as shown in Fig. 4A. The density nodes assigned are represented by a color-coded intensity scale, as described in the legend.
All the nodes identified in the control structure can also be traced in the acid structure. Compared with their position in the control structure, the nodes c, e, g, and i, corresponding to the E1 protein (Fig. 3), have all moved upwards into a more vertical position relative to the membrane in the acid structure (Fig. 4A). In the spike head, nodes a and b remain at a higher radius than node c of the E1, but with a slight relocation relative Meanwhile, the width of the spike is narrowed, and the limb region becomes slimmer and more elongated than in the control. There is a widening inbetween pillar structures in the stalk region (asterisk). Top views of the spikes demonstrate that the hole, located in the center, has expanded in the acid structure, and there is a partial reshaping of the spike head morphology.
to each other. The density node b has moved partly out of the vertically cut plane, as can also be seen in a horizontal cut through the spike head (Fig. 4B).
Details of the acid-induced structural changes are further analyzed in a series of virtual sections from the head, via stalk, to shell following the 3-fold axis (Fig. 4, B-H). The cutting radii are chosen to reveal the same nodes in the control and acid structures. The density node relocations in the spike head follow an overall anticlockwise rotation in the acid structure (Fig. 4B). At lower radii, a twist of structures occurs on acidification, as demonstrated by triangles in Fig.  4, C-F. Overall, the nodes c, e, g, and h turn slightly anticlockwise in the acid structure; whereas node f in contrast displays a clockwise movement. Node d in E2, which holds the density of the wings of the spike together, appears less dense and vacates the center of the spike in the acid structure ( Figs. 4C and 5C, middle). Consequently, these reciprocal node movements would dissociate the two glycoproteins in the stalk region, an observation also evident from the renderings shown in Fig. 2A and Fig. 5, A and C, right. While this happens, the morphology of the spike is essentially retained in the acid structure. The described relocations in the spike structure would play a role in the swelling of the virus. The lift of the shell (Figs. 1B and 2A) and the widening of the openings at the 2-and 5-fold axes (Fig. 1A) are part of the process. Similarly, to demonstrate how that relates to the relocations in the spike, the domains around the 3-and 2-fold axes are included together with the radial sections through the shell in Fig. 4, G and H. As seen, the released contacts between nodes i from the neighbor spikes contribute to the enlarged opening around the 2-fold axis in the shell of the acid virus, a feature also seen at the 5-fold openings.
Modeled Bending of the Hinge Region of the E1-In the low pH structure, the morphological shell-limb area is lifted outwards away from the membrane, and the central hole in the spike, as well as the openings around the 2-and 5-fold axes, are widened (Figs. 1A, 2B, and 4H). Along with this, node i is tilted so that the connection to the limb region is thinner and more elevated (Figs. 2A and 4A and detail in Fig. 7). Following the assumed E1 orientation in the stalk, nodes e and g within the molecule are merged and tilted to rise at a slightly higher angle from the shell (Figs. 4A and 6). This is followed by a lift of the head with node c. The DI contacts to DII of E1 are assumed to be between nodes g and e (Fig. 3A). Node e has changed in relative position to nodes g and h, at the lowered pH. In the E1* crystal structure, the DI and DII are connected by a hinge, located between ␤-sheet strands k-l and g-f (15,16). Recalling that this hinge region between the DI and DII is bent by 15°in the E1* HT (12), as compared with E1*, we ask if the observed relocations may reflect a similar phenomenon. The best fit of E1* into our control density map was obtained if it was treated as one rigid body; however, in the acid density map the fit was improved when the E1* was treated as three independent rigid bodies (DI, DII, and DIII). Therefore, in the acid structure, the best fit is obtained at a bend in the hinge region of 6°, and in the same direction as the larger bend seen in the E1*HT (Fig. 6). A capacity for bending in the hinge is evident from a comparison of the crystal structures of E1* and monomers in the E1*HT. Therefore, the modeled bending derived from the relative tilt of nodes e to g and h would reflect an intermediate configuration. In addition, the distance of His 125 involved in E1-E1 dimer interaction at the q2-fold axis is widened (Fig. 6, highlighted box).
Release of the Lateral Shell Contacts-Together with an expansion of the external structure the widening of the openings in the shell region are the most prominent morphological changes observed in the acidified virus (Fig. 1A). The shell region is dominated by nodes g, h, and i, which can be followed in the acid structure (Figs. 3 and 4).
The DI and DIII of the E1* crystal structure fit well within the shell region of the ϭ 3 surface rendered control SFV density map (Fig. 7A). Thereby the DIII allocates to node i and the DI to node g and h. In the acid structure, the whole shell is lifted, and a relocation of the nodes has occurred (Fig.  7). To reveal the initial action of DIII on acid trigger, the DI Ref. 16). The folding and the secondary structures match several of the nodes. The S-S bond-stabilized Ig-like ␤-sheet structure of DIII fits into node i, the ␤-sheet structure defining DI fits into nodes h and g. Rising from the shell into the stalk, the DII central domain with its twisted ␤-sheets fits node e. The two long ␤-hairpins in the distal part of DII, of which one holds the fusion loop at its turn, passes a less defined structure region of the stalk under the spike head, and fills into node c at the bottom of the lobe. Thereby, the fusion loop is under node a. Node b, more centrally oriented in the head lobe, seems not to be in contact with E1 in this fitting. Provided the fit of the E1* structure is correct, nodes a, b, d, and f represent the E2 protein. Furthermore, the convergence point of E1 (distal part of node i) and E2 (node f) proteins is in the shell region, shown by a black arrow in B. E1 and E2 together form the limb, which connects the shell to the membrane. and DIII atomic structures were handled similarly as described above for the DI and DII. Thus, the DI and DIII rigid bodies were independently fit into the density map of the acid-treated virus structure (Fig. 7).
The fitting analysis shows that interaction between E1 molecules around the 2-and 5-fold axes are created by the DI/III linker peptide and the DIII in the neighbor molecule, which is according to the literature (15,16). This contact is weakened in the acid structure (Fig. 7A). The shortest distance between adjacent E1-E1 molecules is formed by Val 291 in the linker and Thr 305 in the ␤-sheet strand "A" of DIII. This is increased from 5 to 12 Å. At the same time there is a relocation affecting the cleft between the DI and DIII in the molecule (Fig. 7B).
The crystal structure of E1* shows that interactions within the cleft between DI and DIII are created by the His 18 in DI and His 331 and His 333 in DIII, by the Glu 20 in DI and Arg 373 in DIII, and by the Lys 16 in DI and main chain carbonyl oxygen of Leu 339 in DIII. Their distances are in the range 2.75 to 5.7 Å. What we can observe in the acid structure is a tilt of node i relative to node h, as demonstrated in Fig. 7B. Furthermore, node i has merged with node f, assigned to E2 (Fig. 7B). This appears concomitant with an extension of the limbs and a raise of the overall shell-spike region. The independent modeling of the DI and DIII demonstrates the fit of the coordinates around density nodes h and i in the acid structure, where bending is found in the linker peptide, and caused by the lifted C-terminal end of the DIII (Fig. 7B).
In summary, the acidification of the virus results in the release of E1-E1 interactions around the DIII. In addition, the contacts between E1 and E2, while released in the stalk region, are more prominent both in the shell region where the two molecules enter into the narrowed limb region (Figs. 4G and 7B, shown with the radial location indicated. The gradient color-coding, from blue to red, represents ϭ 1 to ϭ 5 in the reconstructions. Major nodes are labeled as assigned in the legend to Fig. 3. The sections of the acid structure are cut at a 7-10 Å higher radius than in the control to account for the expansion. The side view section (A) is cut to pass through nodes a, b, c, e, g, and i. The head of the spike holds nodes a, b, and c, the stalk region encompasses nodes d, e and f, and the shell region holds nodes i, g and h (node h is not seen in this side view section). In the acid structure, node a is lifted relative to b; node b moves out of this vertical section; nodes c and cЈ, to which the distal DII of the E1* is fit in the control structure, are moved toward the center of the spike in the acid structure. The horizontal section of the spike head (B), which passes through nodes a and b shows an overall anticlockwise turning of node b, relative to node a in the acid structure (r ϭ 332 Å in Control; r ϭ 342 Å in Acid, respectively). A series of sections through the stalk (C-E) show the torsion of triangles connecting the major nodes of the trimeric spike of the control and acid structures. Nodes c, e, and g follow anticlockwise turning (dashed triangles); whereas, node f turns clockwise (solid triangles). Node d, which connects the three lobes of the spike (r ϭ 312 Å, Control) appears with less density in the acid structure (r ϭ 321 Å, Acid), along with widening of the hole of the spike center in the acid structure. The pH-dependent separation in the stalk region is here seen as a separation between nodes e and f (black asterisk in D, as is also observed in the ϭ 1 surface rendering in Fig. 2); whereas, nodes e and f Љ keep connected in the acid structure (r ϭ 303 Å, Acid). A series of sections through the shell (F-H) are shown. Node h follows the anticlockwise turning (dashed triangles). Node i viewed through both 3-fold and the 2-fold axes are shown in G and H, showing that the contacts (indicated by black small triangles) of node i with its neighbors are widened in the acid structure. The black arrow (see H, Control) shows the convergence point of E1 and E2 proteins toward the limb region. MARCH 2, 2007 • VOLUME 282 • NUMBER 9

The Dynamic Alphavirus Envelope
i-f merging) and in the spike head region (Fig. 5, nodes a to b and a to c  distances).

DISCUSSION
It was found earlier that alphaviruses infect cells via receptormediated endocytosis and fusion with the membrane of acidic endosomes. To reveal mechanistic rearrangements on virus entry into a fusogenic stage, we have studied the SFV structures at different pH by Cryo-EM. Reaching a consistently higher accuracy in the structural details and managing a slightly lower pH than earlier reported (8), we confirm and provide better details on the increase in diameter at acidic pH and on the accompanying relocations at the limb level. However, by applying the node display approach to compare the structures, the present study reveals details, not earlier anticipated, on the behavior of the two glycoproteins in the acid-triggered virus.
The nodes would represent domains with stable secondary structures (23)(24)(25). Thereby their interplay in the virus structure under different pH conditions can be followed. Because the crystal structure of the external E1* is solved (15,16), the E1 can be allocated to a subset of these nodes, leaving the other ones as putative E2-related structures. With these assumptions the external part of the E1 protein is shown to move outwards with the expansion of the external domains of the virion and to acquire a more vertical position in the stalk region at acidic conditions. Meanwhile, the interactions of E1 and E2 in the stalk and shell resolve. However, the E2 protein keeps to one side of the E1 in the stalk and to the top of E1 in the head of the spike also at the more acidic conditions.
The Head of the Spikes Remains as a Morphological Unit While the Fusion Loop is Exposed and the Shell Region Opens Up-By gradual acidification it was previously shown that the fusion peptide loop is rela-  (nodes a, b, c, d, e, and f). The location of the centers of the density nodes are indicated by small balls, and distances between them given in Å. Although nodes a, b, and c rearrange within the head region, their individual shapes are essentially retained (C, left). The distances between centers of nodes a and b, as well as between a and c, are narrowed, and node c approaches a in the acid structure (C, left). Node b moves inward toward the center of the spike and is out of the section cut in Fig. 4B. Node d, which would be part of the E2 protein, vacates the center of the spike resulting in a widening of the central hole of the spike (Figs. 2B and 5, B and C, middle); whereas node c, representing a portion of the E1 protein, is moving centripetally (see details in the legend to Fig. 6). In addition, distance between nodes e and f around the central hole in the stalk region is widened (C, right) indicating a separation of the two glycoproteins in this domain. (See also Figs. 2A and 4D).
tively early, accessible for interaction with antibodies, coated on a sensor surface or as Fab fragment binding to immobilized virions (11). The optimal exposure is reached at pH 6 and is maintained down to below pH 5.8 (11). Therefore, it would also be available for interaction with an external membrane at that pH. From the present study it is obvious that the contact between the two glycoproteins in the head of the spike is retained (Fig. 5, A and B). Thus, under the same conditions as the fusion loop is exposed to external probing (1), the top of the spike head remains in control of the upper region of the E1. As judged from the fitting of the E1*, the fusion loop would be exposed on the side of the spike head, below the protruding node a (Figs. 3 and 4A). Meanwhile, the reciprocal movements of E1-and E2-related nodes imply a separation of the glycoproteins in the stalk region (Fig. 4). Node d linking the three lobes of the spike head together in the stalk is dissolved in the pH 5.8 environment (Figs. 4C and 5, B and C, middle). Furthermore, relocation and narrowing of the limbs that connect the shell to the transmembrane domains show a reciprocal rotational movement of E1 and E2 domains, i.e. behaves as earlier described (8). The DIII of E1 in the shell region is to a large extent released from neighbor contacts as the radius of the shell FIGURE 6. The flexible hinge between DI and DII is utilized to model the E1* crystal structure into the acid SFV Cryo-EM density map. The source atomic structure of the E1* is fit into four unique monomers: three from the q3-fold spike and one from the 3-fold spike of the Cryo-EM density map. One-third of the spike with surface rendering is shown as an inset. The green and red ribbon structures show the fitting to the control and acid nodes c, respectively. The E1* crystal structure is bent in the hinge region between DI and DII to obtain the best fit of the DII into the Cryo-EM density map of the acid structure (see details under "Experimental Procedures"). The highlighted box in the right panel shows the released E1-E1 dimer interaction at q2-fold axis; the distance of His 125 with the other one is widened from 6 to 11 Å. In the bottom panel, we summarize the modeling and relate it to the atomic structure of one monomer in the reported postfusion E1 homotrimer, E1*HT. The E1 foldings are aligned at the B 0 I 0 H 0 G 0 ␤-sheet (16) of DI. The rotation angle about the hinge, located between the strands k, l, and g, f of the ␤-sheet, is 6°in our acid model (acid model, red), as compared with 15°in the E1*HT (E1*HT crystal, black). Here the acid model and the source structure (E1* crystal, green) are shown in overlay, to demonstrate the bend. The domains are indicated below, and the fusion peptide is colored yellow.
increases. The characterized, acid-induced rearrangements of glycoprotein interactions are likely to be part of the membrane fusion reaction in SFV. Phenomena observed that hints to a pre-organization for E1 trimerization are the flex of the DI/II hinge and the release of the DI to DIII interlinker along with a release of DIII to DI interphase interactions.
The E1 Domain III-On refolding from E1* to E1*HT (15,16) the relocation of DIII is remarkable. The DI and DIII of E1* are connected by an 11-residue long linker sequence (Pro 283 -Pro 294 ). In the refolding, the interphase between the DIII and DI in the E1* is totally lost, and the DIII ends up in contact with the neighbor monomer in the trimer. The C-terminal extension follows the junction between the monomers in the direction toward the fusion peptide (26). According to the crystal structure of the E1*HT, the DIII contact would be at the end of the D 0 and E 0 ␤-strands in DI of the neighbor monomer. It seems that exactly the corner of DIII, close to the C-terminal extension, is the contact point with the neighbor E1 in the trimer (16). The FIGURE 7. Alignment of the E1* crystal structure into the Cryo-EM density map of the shell region. A, E1* crystal structure is fit into the Cryo-EM density maps, here demonstrated in top views along the 2-fold axes of the pH 7.4 (Control, above the black dashed line) and pH 5.8 structures (Acid, below the black dashed line). The shell openings at the 2-fold axes are widened in the acid structure. The modeled control and acid structures are colored in green and red, respectively, and the contour indicates 100% mass. The modeling is based on independent rigid body fitting of the atomic structure. The interactions of DIII with its neighbors are released in the acid structure; the boxed area with the corresponding one in the acid case, as overlay, demonstrates the variation seen (A, right). The C terminus of E1* is indicated. The modeled E1-E1 contact areas, comprising the linker between DI and DIII and the half ␣-helix contact with the strand "A" of the neighboring molecule, are shown in black triangles (16). The distance between Val 291 and Thr 305 in the two domains, as modeled, is increased from 5 Å in the control to 12 Å in the acid case. B, nodes h of the control and acid Cryo-EM maps are aligned, which means a virtual radial lift of the control shell structure (control at ϭ 1, gray contour; at ϭ 3, green rendering; acid, red rendering). Thus matched, the structure overlay demonstrates that node i has lifted 18°in the acid case as compared with the control, together with the elongated limb in the acid structure; in addition, node i is merged with node f in the acid structure. The shown side view perpendicular to the 2-fold axes of the acid and control structures are aligned at node h (black font, left of the axis line in B). The location at a higher radius in the acid structure results in a larger distance, so that the other nodes h do not fully overlap, as seen in the structure to the right of the axis line. The atomic models (B, right) represent the same domains as shown in B, left. The acid (red) and control (green) ribbon structures are superimposed, aligned in the DI (the B 0 I 0 H 0 G 0 ␤-sheet of DI (16) of the two structures are aligned). This demonstrates the upwards tilt of DIII (node i) relative to DI (node h), implied from the differences seen in B, left.
importance of this contact is implied by the finding that exogenous DIII can function as a dominant-negative inhibitor of both fusion and infection (27). However, as reported by Liao and Kielian (27), the site is not prone to binding unless under conditions that trigger trimerization, i.e. acidification and the presence of a target membrane.
A relocation of the DIII requires not only that the interphase between the DI and DIII in the monomer structure is opened but also that the linker sequence between DIII and DI is free to allow the extended relocation. This would relate to the plausible protonation of the histidyl cluster in the DIII to DI interphase, seen in our structure to open at a pH around pH 6. This would explain why the early relocations are seen in the shell and the stalk regions of the SFV envelope.
Resent observations from the Kielian group point to that the sequence between DIII and the membrane, i.e. the E1 part of the limb region, is partly exposed after membrane insertion. The study implies that this represents an intermediate stage, because the sequence is hidden in the postfusion structure (28 -30). From a mechanical point of view one can note that E1 homotrimers are very stable molecular arrangements, the formation of which would drive the fusion process if the components are made available to interact. However, for that to be possible, the E1-E2 contacts in the head, as well as in the limb regions, have to be released along with the internal E1 interactions. In the structure of the virion reported here, the areas of DII that establish the monomer contacts in the E1*HT are shielded by E2 structures. The DIII-DI linker and the DIII, set free from neighbor contacts in the shell region of the acid structure, could hardly by themselves act as a bender of the E1 molecule to create the required "hairpin" and force the virus and target membranes in proximity. However, the DIII may fix the molecule in the crucial trimeric configuration if remaining portions of the molecule are released. The elongated DI-DII of the E1 molecule need to be tilted and translocated such that contacts with the DIII of a neighboring unit can form and finally create the assumed structure of the stable homotrimer observed after fusion.
The pH Effect-The modest but significant difference in the structure of the virus treated at low pH and the control would represent early prefusion events. It is anticipated that they partly involve variation in ionic contacts imposed by the lowered pH. It is tempting to speculate that the widened opening in the shell with the lost interactions around the DIII to a great extent result from proton reorganization related to the cluster of histidyl residues in the DIII to DI interphase (16). Likewise, the histidyl protonation would play a part in the gradually reduced contact between the glycoproteins in the stalk region, and in the node relocation within the spike head, where the conserved His 230 in the cholesterol determining i-j ␤-hairpin would be located. According to the fitting of the E1* structure into our Cryo-EM density map, the His 230 would reside in node c. The observed movement of node c within the spike head region in response to a lowered pH may reflect protonation of this residue. Assumedly protonation of His 230 is essential, either for fusion loop exposure or for a proper response during membrane insertion and fusion. The fact that a mutation to alanine at this position renders the virus noninfectious points in that direction (31). As recently described by Chanel-Vos and Kielian (28,31), the H230A mutant would concert its infection deficiency at a late stage of the membrane-interacting process. This is in line with our observation that the spike head stays as a morphological unit with its nodes remaining in close contact, despite the fact that the fusion loop is exposed.
Mechanistic Considerations-With the described contact between the E1 and E2 glycoproteins in the spike head region, it is feasible that a coordinated separation of these would not happen until at or after insertion of the fusion loop in the target membrane. In this way, the E2 glycoprotein may actively aid the rotation, bending, and movement of the E1 subunit in proximity to each other and to force the virus and target membranes into close encounter for fusion to occur. Trimerization of E1 would then help to drive the membrane-merging reaction, but cannot occur until the head region contacts are released. The E2 glycoprotein would thereby be used not only as a scaffold during assembly and to shield the fusion loop during the maturation and transport, but also as a controlling lever to bring membranes together after insertion of the fusion loop in the target membrane. In addition, pH-dependent interactions just above and under the virus membrane bilayer, including the interaction with the nucleocapsid, may be of utmost importance to put the fusion mechanism into effect (20). To attest the applicability of the full mechanism, further details of virion structure and dynamics remain to be characterized.