Prefusion Rearrangements Resulting in Fusion Peptide Exposure in Semliki Forest Virus*

Semliki Forest virus (SFV), like many enveloped viruses, takes advantage of the low pH in the endosome to convert into a fusion-competent configuration and complete infection by fusion with the endosomal membrane. Unlike influenza virus, carrying an N-terminal fusion peptide, SFV represents a less-well understood fusion principle involving an endosequence fusion peptide. To explore the series of events leading to a fusogenic configuration of the SFV, we exposed the virus to successive acidification, mimicking endosomal conditions, and followed structural rearrangements at probed sensor surfaces. Thus revealed, the initial phase involves a transient appearance of a non-linear neutralizing antibody epitope in the fusion protein, E1. Concurrent with the disappearance of this epitope, a set of masked sequences in proteins E1 and E2 became exposed. When pH reached 6.0–5.9 the virion transformed into a configuration of enlarged diameter with the fusion peptide optimally exposed. Simultaneously, a partly hidden sequence close to the receptor binding site in E2 became fully uncovered. At this presumably fusogenic stage, maximally 80 fusion peptide-identifying antibody Fab fragments could be bound per virion, i.e. one ligand per three copies of the fusion protein. The phenomena observed are discussed in terms of alphavirus structure and reported functional domains.

The external accessibility of structures in a virus controls its interaction with the target cell surface as well as with the membranes the virus has to penetrate to complete infection. With enveloped viruses this involves a fusion step, which takes place after the initial attachment to the cell surface receptor, and, in some cases, endocytosis. Fusion may be directly induced by interaction with the receptor or occur by later triggering by other means. For viruses taken up into endosomes the acid environment commonly initiates the activation leading to fusion with the endosomal membrane.
The fusion reaction as revealed in influenza virus and human immunodeficiency virus seems to be relevant for a series of viruses with N-terminal fusion peptides (1). The trimeric protrusions ("spikes") hold three molecules each of an external glycoprotein (in influenza the HA2) and a membrane-spanning fusion protein (HA1). These glycoproteins are derived from a trimeric precursor molecule, (HA0) 3 , cleaved such that a hydrophobic peptide (fusion peptide) forms the N terminus of the fusion protein. Thus primed, the trimeric structure shields its three fusion peptides in charged pockets. Activation requires their extrusion toward the fusion target membrane. Here, the refolding of a loop between two helical domains plays a key role in the formation of a long helix extending the central triplestranded coiled-coil of the protrusion, as proposed from modeling studies on synthetic peptides (2,3) and peptide inhibitor interaction (4). This relocates the fusion peptide over 100 Å from its previously buried position (5) allowing insertion into the target membrane. To provide the juxtaposition of the viral and the target membranes, the long helix is unfolded at the middle, forming a reverse turn. Thus, the second half of the long ␣ helix is jackknifed back to lie antiparallel against the first half, allowing the C-terminal membrane anchor to be repositioned at the same end of the rod-shaped molecule as the fusion peptide (6). This would bring the membranes in close proximity and promote fusion (7). The formation of a six-helical coiled-coil bundle that provides juxtaposition of the virus membrane anchor and an N-terminal fusion peptide (8,9) is a recurring feature characterizing this fusion mechanism I, extensively reviewed by Skehel and Wiley (10 -12) and others (13). A variation on this theme is proposed for the Newcastle disease virus, where the target membrane contacts may be established by the fusion peptide prior to formation of the extended coiled coil. In this case the helix extension and super coiling would, directly, force the virus and target membranes together to complete the fusion reaction (14). Less explored are mechanisms prevailing in viruses, which, like the flavi and alphaviruses, have a low content of alpha helical-structure in their envelope proteins and carry an endosequence fusion peptide. Furthermore, in these viruses, although derived from one precursor peptide chain, the two main spike proteins are both anchored in the virus membrane.
The sequence of folding and tertiary organization of the envelope proteins in the alphaviruses are essential, not only for the assembly and budding of the virion (39), but also for the capacity of the virus to properly fulfill infection. The transmembrane glycoproteins, pE2 (p62) and E1, originating from the same translation product, form heterodimers already in the endoplasmic reticulum compartment (40). The direct association with pE2 seems to be required to prevent aggregation of the E1 protein (41). The pE2-E1 dimers associate into trimers before cleavage, in the trans-Golgi, of the pE2 component into the membrane-anchored E2 subunit and the small external glycoprotein E3 (42)(43)(44). This processing establishes a proper interaction of the E2/E1 subunits in a mature, trimeric spike structure (41). It also provides an infectious structure, because, although they may bud normally, mutants with uncleaved pE2 analogues can neither bind well to the target cell nor respond to acidification as the wild type virus (32,33,41,43,45). Therefore, the E3 domain of the pE2 would not only scaffold the arrangement of the E1/pE2 glycoproteins for proper folding but also control premature unfolding. Such control would be released by split of pE2 into E2 and E3 and possibly by a following discharge of E3. The cleavage would allow receptor binding to occur and permit fusion to be triggered. In the Sindbis virus the E3 component is early separated from the mature virus. Structural studies on non-cleavable pE2 mutants imply that the E3 is located at the outer rim of the lobes in the spike-like protrusions (16,20).
The recently proposed x-ray structure of the ectodomain of SFV fusion protein, the glycoprotein E1, suggests it to be the major constituent of the protein shell, or skirt, domain, seen in the cryoEM structure (46). The elongated molecule would cling to the stem of the protruding spikes, hiding its fusion peptide underneath the lobes. This leaves the bulk of the protrusions to be composed by glycoprotein E2. Similar conclusions on relative organization of the E1 and E2 ectodomains were derived from structural studies on sugar site deletion mutants (22). With these structural advancements the location of functional domains and their interplay can be envisioned.
The location of the receptor-binding domain has been determined, with the aid of a monoclonal antibody and cryoEM, to the outer tips of the three lobes of the spikes (18). The epitope is located in the middle of the E2 sequence, and thus provides the first sequence location in the folded external structure. It is close in sequence to a reported protective epitope (47)(48)(49). The involvement of the cell surface receptor in the immediate prefusion events is unclear (24). Its binding seems not to be a prerequisite for fusion to occur. Whether or not it facilitates fusion at a different pH than should otherwise be the case is not known.
The fusion peptide is an endosequence located about 80 amino acid residues away from the N-terminal in the E1 peptide (50,51). There are few alpha helices (46), or even alpha helix predictability sequences, in the E1 protein, thus explaining why a somewhat different strategy for infection and fusion would prevail than in the influenza and human immunodeficiency virus-related viruses (2,9,11,13,52). Ordered structures as such, i.e. beta-sheet-forming sequences, may be essential, as suggested by Delos et al. (53). In several viruses with internal fusion peptides, there is a central proline within this sequence. Such a reverse-turn function seems to be related to the efficiency in processing during maturation, as shown by a series of mutations in avian sarcoma/leucosis virus (53). Receptor binding was not much affected by point mutations at this proline site. However, the infectivity showed a bell-shaped dependence viz. the hydrophobicity of the residue, where proline is of a moderate level (53). In SFV the cholesterol dependence for fusion is linked to another proline, Pro-226, also in the E1 glycoprotein. A substitution, P226S, resulted in a more promiscuous pattern in the fusion behavior, abolishing the strict requirement for the sterol in the target membrane (54). Therefore, non-bulkiness and an intermediate level of hydrophobicity are essential in the metastable clusters related to the fusion mechanism.
Structural rearrangements into a fusion-competent configuration take place upon acidification and are essential for fulfillment of the infection. Attempts to approach the fusogenic form of the particle by cryoEM have given indications for a reshaping of the protrusion structure (19,20,55) and an expansion of the extra-membrane domain of the envelope (56). Recent findings on the structure of the E1 protein (46) prompt a re-evaluation of the glycoprotein organization and refolding mechanism (46,56,57). An acid-induced dissociation of the E1-E2 interaction, along with E1 homotrimerization, seems to occur at a stage prior to the actual fusion (58,59). The acidinduced E1 homotrimer is highly resistant to detergent and trypsin treatment (33, 59 -62) and may provide a tool for insertion into target membranes. The split of the E1-E2 interaction appears to be a separate and preceding step toward E1 homotrimerization (60). During the process, masked domains become available (35, 63,64) and should present the fusion peptide in a ready-to-act configuration for close encounter with the target lipid layer. Details on such rearrangements in relation to other structures within the virion would provide further clues for the mechanism involved (64).
Studies to monitor and identify pH-related rearrangements in the SFV and the related Sindbis and Ross River viruses have, in addition to cryoEM, used cell binding assays, binding assays with liposomes (38,62,65), and enzyme-linked immunosorbent assay, in conjunction with virus mutants, antibodies, and peptides. In the present report we introduce an additional approach for following environmentally triggered, successive changes in the virus envelope configuration. Thus, variations in exposed domains in the virus are revealed by real-time measurements based on surface plasmon resonance. As sensor surface probes we establish monoclonal antibodies against the SFV envelope glycoproteins and select a set for which linear sequence epitopes can be defined by PepScan analyses. This includes an antibody directed against a portion of the fusion peptide in glycoprotein E1 and one against a sequence close to proposed receptor-binding domain in E2. Although with a nonlinear epitope, a neutralizing antibody toward E1 is also included in the study. By introducing the virus to environments of different acidity, we demonstrate that rearrangements occur already at close to neutral pH, whereas a more drastic reshuffling of the structure takes place in the pH range 6.2-5.8. At the same time the virion is transformed into a particle of larger diameter, as observed by electron microscopy. These findings provide a piece of a functional map to be gradually fed into the detailed structure.

Cell and Virus Preparations-Baby hamster kidney cells (BHK-21)
were grown in BHK-21 medium (Invitrogen) supplemented to contain 10% tryptose phosphate broth, 5% fetal calf serum, 2 mM glutamine, and 20 mM HEPES (pH 7.3). (In later experiments this was further supplemented with 17 g/ml cholesterol.) When close to confluent, the cells were infected with SFV wt at a multiplicity of infection of 10, in minimal essential medium (or BHK-21 medium), supplemented to contain 0.2% bovine serum albumin, 2 mM glutamine, and 20 mM HEPES, and incubated at 37°C in a 5% CO 2 atmosphere. After 1 h of infection the medium was replaced by fresh minimal essential medium (or BHK-21 medium), containing 2 mM glutamine and 20 mM HEPES, and incubated for another 18 -22 h before harvest of virus. The viruscontaining supernatant was cleared from cell debris by centrifugation in a Sorvall JA10 rotor at 11,000 ϫ g for 30 min. This was repeated until no visible pellet appeared. The virus was collected from the cell-free supernatant by pelleting it through an 8% (w/v) potassium-tartrate cushion, in an overnight centrifugation at 11,000 ϫ g (JA10 rotor). The potassium-tartrate stock solution was 30% (w/v) and contained 100 mM Tris and 10 mM MgCl 2 (final concentrations) with pH adjusted with KOH to 7.4. It was diluted to the final concentration with 100 mM Tris, 10 mM MgCl 2 , pH 7.4, and filtrated at 0.22 m before use. In later experiments the cushion was omitted and the centrifugation was performed at 17,000 ϫ g, which assured a high recovery. The virus pellet was overlayered with 100 mM Tris, 10 mM MgCl 2 , pH 7.4, and left for 4 h, or overnight, to form a suspension. The virus was applied to a 10-ml continuous gradient of 10 -30% (w/v) tartrate, established in a Gradient Master, using the program for 20 -50% sucrose. Ultra clear SW40 tubes (Beckman) were used, and the centrifugation was run in a Beckman Rotor SW40 Ti at 20,000 ϫ g (12,500 rpm) overnight. This careful handling was applied to avoid distortion and damage of the virus during purification. We found that with the alternative media, containing more glucose and supplemented with cholesterol, the virus retained its properties in a modified centrifugation protocol. Thus, in later experiments we used the tartrate gradient buffered with MOPS, pH 7.4, and ran it at about 100,000 ϫ g (25,000 rpm) for 5 h (52). The virus band was collected, and the concentration, purity, and quality of the virus were checked by light absorbance at 260, 280, and 310 nm, SDS-PAGE, and electron microscopy.
CryoEM Data Processing-Images, at 28,000-fold nominal magnification, were taken as focal pairs at Ϫ1.0 and Ϫ3.0 m defocus, using a Philips CM-120 microscope. In the digitized micrographs, individual intact particles were manually boxed out to generate a stack of above 700 particles for every pH condition. Data-processing steps included particle orientation and size determination with the model-based EMPFT and EMIMGCMP program packages (66,67). All calculation steps were carried out on DEC alphastations, running the system OSF with an XP1000 processor (Compaq, Houston, TX). In an iterative process, intermediate three-dimensional density maps were successively used as new models in refining the parameters of particle orientations (39). The particle size was assessed by comparing the raw image with a series of scaled back-projected images of the control threedimensional model (66,67). The size was assigned as the best fit among the back-projections, using systematic grid-interpolations in real space. With the pH 6.2 and 5.9 virus analyses, the reconstruction of the pH 7.4 virus was used as a control model. The particle diameter was determined with an accuracy of 0.1% and grouped in classes by intervals of 0.5% of control average diameter.
SDS-PAGE and Western Blot Analysis-Electrophoretic analyses were made in the PhastSystem (Amersham Biosciences, Uppsala, Sweden) or in the Bio-Rad Protean II system (Bio-Rad Laboratories, Hercules, CA), and run according to the recommendations of the manufacturers. Samples were regularly prepared by dilution with nonreducing sample buffer to obtain a good separation of the SFV E1 and E2 peptide bands, which under reducing conditions run close to each other in the gels. Electrotransfers to PVDF membranes were made by overnight runs (30 V, 90 mA) with 25 mM Tris, 192 mM glycine, and 0.01% (w/v) SDS, in 20% (v/v) methanol, as transfer buffer. After transfer the membrane was kept on a wet filter paper and covered with Parafilm, to prevent drying, and cut into appropriate sections for blocking (2 h in 10 mM Tris-HCl, 150 mM NaCl, pH 7.4 (TBS), supplemented with 0.5% Tween 20) and subsequent probing with antibodies or biotinylated lectins. After incubation with the probe for about 2 h, the membrane strips were washed four times in 10 mM Tris-HCl, 0.05% Tween 20 (TBS-T). The detection system included incubation with alkaline phosphatase-conjugated antispecies antibody (Blotting grade, Bio-Rad, Richmond, CA) or a complex formed by biotinylated alkaline phosphatase and avidin (ABC-AP Vectastain TM kit, Vector Laboratories, Burlingame, CA). After washing, the probed bands were visualized by zymogram staining using the substrate ␤-naphthyl phosphate in combination with Fast Blue-B salt.
PepScan Analysis of Antibody Epitopes-Polyclonal rabbit antisera against the E1 and E2 proteins were a kind gift from Dr. Henrik Garoff. They had been raised in rabbits against the E1 and E2 glycopeptides and cut out from the gel after separation in SDS-PAGE. The source of the monoclonal antibodies, from where hybridomas were prepared, was kidney B cells of SFV wt-infected mice (A. Salmi). The monoclonal antibodies were purified on a protein A-cartridge (Bio-Rad) and desalted. The Fab fragments were prepared by papain cleavage, and the Fc fragment was removed by re-passage over the protein-A cartridge. Antibodies or Fab fragments were desalted and concentrated by vacuum dialysis. Linear epitopes for the antibodies were searched by a peptide-scanning system, essentially as described by Pulli et al. (68). Sequential peptides along the viral protein sequence (sequential peptides of target, SPOT) were synthesized in spots on a filter. The peptides were 20 residues long, and the sequential overlap was 3 residues. To reveal antibody epitopes, the SPOT filter was blocked for nonspecific interaction in TBS with 0.5% Tween 20 and incubated with the antibody, usually at a dilution of 1/1000 in TBS-T, for 2 h at room temperature. After washing the SPOT filter in TBS-T, the bound antibody was electrotransferred to a PVDF membrane at 100 V, 350 mA (max settings) during an overnight run in a transfer buffer with SDS and methanol, as described for the Western blots above. Because the mouse monoclonal antibodies may vary in isoelectric points, PVDF membranes were placed on both sides of the SPOT filter during blotting. The PVDF filters were blocked and probed for antibodies with alkaline phosphatase-conjugated anti-mouse or anti-rabbit antibodies and developed as described for the Western blots. The reactive spots were traced to the sequence. The SPOT filter was re-blocked and readied for reaction with another antibody.
Binding Analyses in Biacore Sensor Experiments-The affinity studies were done in the BIAcore2000 system (Biacore AB, Uppsala, Sweden) at 25°C, as standard. The sensor chips utilized were with carboxymethyl coating (Sensor Chip CM5) or attached streptavidin (Sensor Chip SA). The conjugation of antibodies or lectins to the carboxylated dextran matrix of the CM5 chips was done using aminecoupling chemistry according to standard wizard methods in the Biacore program. The SA surface was conjugated with biotinylated Galanthus nivalis lectin. The incorporation levels were usually established within the range 25-105 nmol of ligand, antibody, or lectin per mm 2 (about 3,000 -15,000 response units, total binding). The regeneration solution used during all the experiments was 10 mM Tris-HCl, pH 8.0, with 1 M NaCl, 500 mM methylmannoside, and 0.25% (v/v) Empigen BB TM (Calbiochem-Novabiochem Corp., La Jolla, CA). The composition of the running buffer and other experimental details are given in relation to the separate experiments.

SFV Particles at Different pH
To follow the morphology of SFV at different pH, aliquots of the virus were mixed with a series of buffers to obtain a final pH in the range of 7.4 -5.0. The treated virus was examined by electron microscopy, with negative contrasting, as well as by cryoEM. The latter was done without contrast or fixatives and should therefore closely reflect the native structure. The virions were found as free particles at physiological pH 7.4 and down to pH 5.9 (Fig. 1A). They essentially remained as free particles down to and somewhat below pH 5.8, where they could be seen attached to each other in small aggregates. At lower pH these gradually developed into large clusters inside which several core particles seemed to be contained.
The size of the free virus particles was found to vary with the pH. To estimate the variation, samples were treated at pH 7.4, 6.2, or 5.9 and examined by cryoEM. Images were taken at the same magnification and defocus level. Upon image analysis the sets of reconstruction basic data showed a significant difference in particle size. Thus, the diameter of the control at pH 7.4 was 679 Ϯ 7.4 Å, n ϭ 773 (mean Ϯ S.D., number of observations), that of the pH 6.2-treated ones were 683 Ϯ 7.9 Å, n ϭ 1328, and that of the pH 5.9-treated ones were 692 Ϯ 7.1 Å, n ϭ 1072. The smaller variation at pH 5.9 than in the other cases suggests that the particles had transformed into a more uniform population of larger diameter (Fig. 1B).

Antibody Characterization
To explore conformational consequences of the acid-induced transformation into a larger particle, as observed above, the SFV virion was examined for variation in epitope accessibility. For that purpose a set of mouse monoclonal antibodies against the virus glycoproteins were characterized by Western blot and PepScan analyzes.
Western Blot Analyses-The antibodies used in this study could all be assigned to their target glycoprotein by Western blot analysis (Fig. 2A). The monoclonal antibodies mabE1f, mabE1g, and mabE1n gave signals at the same, and single, position in the blot, corresponding to the SFV E1 peptide. mabE1n was less reactive with the reporter antiserum used. It was found to be of IgG3 type and reacted well when specific reporter was introduced. All the E1 mAbs were sensitive to reduction of the SFV sample, whereas only one of those against the E2 protein, mabE2v, shared this sensitivity. The anti-E2 mAbs all gave signals with the E2 peptide, as well as with its precursor peptide, pE2 ( Fig. 2A, p62).
PepScan Analyses-Sequential epitopes could be identified by PepScan analyses for four of the antibodies. As outlined under "Experimental Procedures" the antibody was allowed to react with 20-amino acid-long peptides, synthesized in spots on a filter. The consecutive spots hold peptides with 3 amino acids in sequential overlap. Antibody-binding spots were identified with enzyme-conjugated antispecies antibodies and zymogram staining (Fig. 2B). The results were evaluated according to the principle outlined for mabE1f in Fig. 2B (bottom panel). Thus, an epitope is assigned as the common sequence in the peptides of the reactive spots. The identified epitopes are listed in Table  I together with notations on the sequence region, found in the literature. The neutralizing antibody, mabE1n, and mabE1g and mabE2t, although binding to the E1 glycoprotein in Western blots and enzyme-linked immunosorbent assay ( Fig. 2A), were not reactive in the PepScan analysis. Their epitopes were therefore assigned as being conformational. Therefore, only the neutralizing antibody is further discussed in this report. Two rabbit antisera, prepared against denatured E1 and E2 glycopeptides, were also applied to PepScan analyses. In this case several sequence regions were labeled in the SPOT filter (Table  I and

Virus Antibody Interactions Studied at Sensor Surfaces
Reorganization in the virus exterior was recorded with the set of characterized monoclonal antibodies in Biacore binding experiments. This was done in two ways: the antibody was immobilized on the sensor surface and the virus introduced in the flow at different pH, or the virus was attached to the sensor and the antibodies introduced in the flow.
Variation in Epitope Exposure-To reveal a possible pH-dependent variation in epitope exposure, the virus was equilibrated to different pH and introduced in the flow over antibodycoated sensor surfaces (Fig. 3). As observed by electron microscopy the virus was retained as free particles down to, and somewhat below, pH 5.9 (Fig. 1A). Reproducible Biacore responses were obtained during more than 1 h after equilibration of the virus within this range. The level of nonspecific binding was established in control runs over non-related antibody surfaces. Conditions were accepted when background compared with virus-related antibody binding was less than 1%. To study the pH-dependent variation in binding to the antibody-coated surface, the virus was introduced at a subsaturating concentration (chosen as one giving about 70% surface saturation at the optimal pH). Six repeats were run over antibody-coated sensor surfaces, and the binding, at various pH, was recorded (Fig. 3). Tris-HCl buffers were used throughout this experiment, but essentially the same results were obtained with MES buffers in the lower pH range. At pH 7.4, the pH of the cell culture medium and of the buffers used throughout the virus purification, a relatively low interaction with the immobilized antibodies was noted. As pH was lowered to 7.0 the binding to the mabE1n sensor surface was significantly increased, and it increased further until pH 6.5 (Fig. 3A, mabE1n). However, at pH 6.0 the binding to this surface dropped to the pH 7.0 level. The virus binding to the antibody mabE2r proceeded in a two-step manner. A plateau was first established at pH 7.0 -6.5. Further decrease in pH raised the binding to its optimal level at around pH 6.0 (Fig.  3A, mabE2r). The virus binding to the mabE1f-coated sensor surface showed a continuous increase as the pH dropped from pH 7.4 and reached an optimum at around pH 6.0 (Fig. 3A,  mabE1f). Finally, for this series, the virus binding to the mabE2m surface sharply raised from about 20% at pH 7.0 to the optimal at pH 6.5, with an essentially retained level down to pH 6.0 (Fig. 3A, mabE2m). Virus binding to an additional antibody, mabE2v, increased from about 20% at pH 7.0 to optimal at pH 6.0 (Fig. 3B). The binding profiles of the virus to the different antibody surfaces versus pH are summarized in the plot in Fig. 3B. This plot also shows virus binding to a surface coated with the G. nivalis lectin, GNA. Contrary to the case with the antibodies, the binding to the GNA surface varied less with pH (Fig. 3B), probably reflecting that its target, the high mannose glycoconjugate present in the E2 glycoprotein, is fairly well exposed for external interaction all through the pH range tested.
Number of Epitopes and Antibody Binding Kinetics-To establish the number of available binding sites per virion and the kinetics for antibody Fab-fragment interaction, the virus was reversibly immobilized by attachment to an antibody-or lectincoated sensor surface. The amount of immobilized virus was related to the surface density of coating antibody and to the virus load (Fig. 4). Varying concentrations of free antibody, or Fab fragment, were then included in the flow over the immobilized virus, as demonstrated at pH 6.2 for fabE1f, the Fab fragment of mabE1f in the Fig. 4. Dissociation was slow, and a stable level of bound ligand made it possible to accurately determine saturation level and the average number of occupied binding sites at a particular Fab concentration. This was calculated based on the assumption that a binding response of 1000 response units represents 1 ng of protein, antibody, or virus bound per mm 2 of the sensor surface (69). Kinetic exper- FIG. 2. A, Western blot analyses, after SDS electrophoresis in a 10% polyacrylamide gel, of non-reduced and reduced SFV sample. The probing was done with the antibodies indicated, at about 1 g/ml. The mabE1f and mabE1n were also applied at a 10-fold higher concentration (lanes to the right in each set). Blanks were similarly treated but without monoclonal antibody. Alkaline phosphatase-labeled goat anti-mouse IgG and zymogram staining were used to reveal bound antibody. The positions of the glycoproteins E1, E2, and p62 are indicated, as are molecular masses of high and low range prestained standards (kDa). B, PepScan analysis of the mabE1f epitope. The spots reactive to the monoclonal antibody were identified by enzyme-conjugated anti-mouse IgG in a PVDF membrane replica (top panel, see "Experimental Procedures" for details). The algorithm used for identifying the epitope of the antibody is demonstrated in the bottom panel scheme. This shows the amino acid sequences of the reactive spots (highlighted) in a PepScan analysis, here exemplified by that of mabE1f. The common sequence is assigned as the epitope of mabE1f. This covers part of the fusion peptide in the SFV glycoprotein E1 sequence, shown in part in the bottom row. Note the disulfide bridge and the locations of the closest histidines in the sequence, relatively distal to the fusion sequence.  iments were run with the virions attached to mabE1f-or mabE1n-coated sensor surfaces (CM5 chips, amine-coupling chemistry for binding of the antibodies), at pH 6.0 and pH 7.0, respectively, revealing a high affinity of these antibodies to the virus at both pH values (Table II). Likewise, the number of virions bound to a GNA-coated sensor surface (Fig. 5, A-C, SA chip, coating with biotinylated lectin), and the successive number of GNA or antibody molecules saturating the virus particle, were determined. Some area of the virion should be shielded around the immobilization region. Nevertheless, the saturation level of GNA was close to the theoretical number of 240, as expected if each E2 molecule in the virus carried one high mannose structure, and that was available for binding. At high concentrations of the lectin the virus particles were released from the surface (Fig. 5A). Although saturating at a much lower concentration, the mabE1n bound maximally 80 mAbs per virion (Fig. 5B). The numbers of fabE1f and fabE1n molecules bound per virion were determined at different pH. It was observed that, with both Fab fragments, as with the corresponding monoclonal antibodies, the number of incorporated molecules never exceeded the number of spikes in the virion, i.e. 80 per particle. However, the availability of sites varied with pH such that at pH 7.0 the virus bound 77 fabE1n molecules, but only 11 fabE1f ones (experiment shown in Table II). At pH 6.0 the situation was reversed with 65 fabE1f, and 12 fabE1n molecules were incorporated. From similar experiments, the kinetic constants were approached. The dissociation constants, for both ligands, were in the nanomolar range (Table  II). Therefore, in the range studied, the difference in incorporation would rather reflect a pH-dependent exposure of the epitopes, than variation in affinity.
Irreversible Changes in the Virus Surface Structures-Because the antibodies reacted with the denatured proteins in the Western blots at neutral pH, they should bind to the particles at that pH also in the Biacore experiments, if the epitopes were accessible. Thus, to reveal "essentially irreversible" pHdependent changes in the virus structure, immobilized virus was treated at an acidic pH and returned to neutral before interaction with the antibody. The GNA-virus interaction was essentially stable throughout the pH range studied (Fig. 3B) and was therefore chosen as linker for immobilization of the  Table II). virus (Fig. 5). The interference of the lectin with the monoclonal antibodies was low. It slightly increased with a lowered pH, but a parallel control channel without virus allowed the appropriate adjustment of the background level. Flushes of successively lower pH were given, interchanged by probing with a monoclonal antibody, and introduced in a flow buffer of pH 7.4. There was essentially no binding of the antibodies to the immobilized virus at pH 7.4 unless the virus was first flushed at a lower pH. In analogy with the earlier experiments the antibodies gave individual patterns in response to variations in pH (Fig. 5C). After cycles with successively lower pH, mabE1n was the first to accumulate on the virus. The mabE2m started to accumulate after treatment at, and below pH 6.5, and successively build up a saturation level on flushes with decreasing pH between the probing steps. The mabE2v reacted after flushing at a slightly lower pH, where it accumulated on the virus within a short pH range. It was saturating every theoretical binding site in the E2 after treatment at around pH 6.0. This occurred at the same pH as the optimally bound mabE1n was released. Finally, the mabE1r epitope was not available for probing by the short pH flushes until pH 6.1 was introduced. Thus, the virus surface possesses a certain flexibility, or reversibility in pH adoption, but is definitely changed at around pH 6.0. Furthermore, the transformation applies differently to the different regions as identified by the antibodies.

DISCUSSION
Relevance of Biacore Binding Studies-One advantage of real-time sensor binding analysis is that the interaction between the sensor surface and the analyte in the flow is recorded directly as it happens. The technique does not require other reporters to interfere, like secondary antibodies or enzyme reactions. This makes the technique useful for revealing native structures that are accessible for external probing (70). It also applies to the study of fluctuation in their exposure under variable environmental conditions. The variation in antibody avidity, within the studied pH range, was assumed to be of less importance than the variations in the epitopes and their exposure (Table II). Covalent immobilization, which has been commonly used with small naked viruses in Biacore studies, was not useful with the enveloped SFV. Therefore, we have studied the exposure of epitopes in relation to pH variation by the interference of the free virus with antibody-coated sensor surfaces (Fig. 3). Assuming that the surface interaction reflects a probability function of the number and relative available area of the epitopes, a dose-dependent concentration of the virus had to be established for each coating of the sensor surface. In parallel, the virus was reversibly immobilized via antibody or lectin linkers to the sensor surface. Thereby the number of ligand binding sites and affinity could be determined by introducing the free lectin, antibody, or Fab fragments at varying concentrations in the flow over the virus particles, replenished between the experiments (Figs. 4 and 5).
Locations of Epitopes and Domains-Aided by the PepScan analysis, several antibodies were mapped to sequential epitopes ( Table I). Some of these fall in or are close to regions earlier reported to be of functional interest. Thus, the antibody mabE2v 142-146 epitope subsists in a variable, in the native virus-hidden, sequence region of the E2 glycoprotein reported as a T helper cell epitope (71). A potential receptor-binding domain is targeted by mabE2r 211-218 . Within this sequence region mutations in residues Thr-216 (49), Asn-218 (72), and Thr-219 (73) caused variations in receptor binding and virus tropism. In Sindbis and Ross River viruses this sequence represents an essential portion of a protective epitope against lethal encephalitis in mice (47,49). In the structure, this sequence was localized to the tip of the spike lobes (18). Together these reports suggest that the region is readily available at the exterior of the native virus. Also in the present study, this is the more accessible region for binding of the virus at the antibody-coated sensor surfaces (mabE2r, Figs. 3 and 6). However, we observed that the mabE2r does not saturate all theoretical sites at close to neutral pH. Only after treatment at below pH 6, was there a full access for antibody binding (Figs. 3A and 5C). This two-stage exposure could be caused by the site being in a cleft, which is sometimes the case for receptor binding sites; therefore, such exposure could be favorable for antibody interaction when the E1 and E2 subunits are less tightly associated than in the native, neutral pH structure. Not far from the mabE2r site in the sequence is the epitope of mabE2m 238 -248 (Table I and Fig. 6). This covers a protective B-cell epitope, residues 240 -246, in SFV protein E2 (71). It also covers part of a protective epitope, residues 227-243, found by Grosfeld et al. (48). The mabE2r 211-218 and mabE2m 238 -248 epitopes appear adjacent to two of the immunodominant regions identified by the polyclonal E2 antiserum. Actually, the region between the two sugar sites in the E2 sequence seems to be the ones most easily accessed for receptor binding and protective immunity, as well as being accessible for sensor surface FIG. 6. Schematic representation of the Semliki Forest glycopeptide sequences with secondary structure predictions, above a hydrophilicityhydrophobicity plot. The positions of the monoclonal antibody epitopes are indicated. The blue bars above the sequence represent the found linear epitopes of the polyclonal sera. interaction studied here (Fig. 6). In concordance with antibody site determination by cryoEM (18), this would indicate that it is located at the top domain of the spike-like protrusions in the SFV structure. The mabE2m epitope is optimally exposed at the pH where the conformational epitope of mabE1n begins to disappear, and they both may be part of the same event, most likely E1-E2 dissociation. Among the antibody epitopes studied, mabE2m is the one closest to the high mannose glycoconjugate site, Asn-262, the glycoconjugate of which is readily available for interaction with the GNA lectin over the pH range tested (Figs. 3B and 5A). From the study of Pletnev et al. (22) on Ross River virus, the location of this sugar domain would be at quasi-2-fold axis, between the spikes, in the shell domain. The positions were located by differential mapping of sugar site deletion mutants and wild-type virus. The glycosylation sites in Ross River virus are the same as in SFV; therefore, the results would have bearing also on SFV. The observed domain may not necessarily represent the true location of the sugars in the wild-type virion, because the carbohydrate would enforce a particular location of peptide domains, overruling a folding occurring in the absence of the glyco-moiety. Furthermore, the lectin only sees the terminal portion of the mannose structure, which may be distal from the attachment point in the peptide backbone. However, the lift of the shell domain at the base of the spikes was among the more prominent changes we observed in the structure at pH 5.9 (56) and thus is well related to the enhanced exposure of the epitope. Another antibody epitope that may reside in the shell region, at the base of the spikes, is the reported acid structure-specific antibody E1a-1 (74). The epitope of this antibody is lost during G157R mutation in the E1 glycoprotein (74). The Gly-157 resides in the central domain of the E1 structure, close to the sugar site Asn-141 and the N-terminal. This residue is located in the shell domain, when following the fitting of the E1 structure in the SFV map (46), and indicates that the exterior spike domains carrying the fusion peptide are highly dependent on structures in the lower regions.
It was recently noted that a series of SFV-neutralizing antibodies of the IgG1 type showed almost identical amino acid sequences in their binding domains (75). These were all composed only of the amino acids glycine, alanine, and threonine in addition to cysteine in the flanking regions. Neutralization would therefore be based an accurate three-dimensional fit in a neutral, moderately hydrophobic cluster. The mabE1n may well conform to this pattern. If so, the pH dependence observed would not originate from the area itself but might reflect changes in remote domains affecting the epitope structure. The mabE1n epitope is the most easily assessed among the epitopes here studied. Along with the other monoclonal antibody epitopes in protein E1, it was essentially lost upon reduction. The fact that the maximal binding represents one ligand per spike indicates that the epitope domains in three E1 peptides are in close proximity to each other or may even together form the epitope. It is tempting to assume that these domains may be located in the center of the protrusions rather in the shell domain.
The only monoclonal antibody epitope in protein E1 that we could identify by the PepScan technique was that of mabE1f 85-95 . It covers the amino acid 85-95 in the E1 sequence (Table I). This partly overlaps the fusion peptide, proposed to comprise the residues 79 -97 in the E1 glycoprotein sequence (50), and is well conserved among alphaviruses (25). This function is supported by the fact that the mabE1f epitope becomes embedded in lipid rafts upon experimental acidification in the presence of cholesterol-containing liposomes (62).
The immunodominant regions in the SDS-denatured E1, as judged from the PepScan analysis of the polyclonal E1 serum (Table I), are within the loop between the central domain I and C-terminal domain III (residues 280 -285), in the mid part of the E1-dimerization domain II (residues 186 -195), and at the top of the ␤-turn (residues 220 -230), parallel to the one carrying the fusion peptide in the atomic E1 structure (46). The top of the ␤-turn comprises the Pro-226 residue, where a mutation to serine has been reported to abandon the dependence for cholesterol in target membranes. Thus, both the fusion peptidecarrying ␤-sheet at the top of the E1 structure and the parallel one would be involved with the target membrane upon fusion.
Although the Pro-226-carrying domain obviously was immunogenic in the denatured E1 peptide, the fusion peptide was not, as far as could be seen from the PepScan analysis. This indicates that the fusion peptide is either not a strong immunogen or hidden in the detergent-treated protein used for immunization.

Rearranging Domains and Intermediate Structures-The
Biacore results indicate that all our monoclonal antibody epitopes are shielded in the virus particle at pH 7.4. Charged components, like sialic acid in the complex glycoconjugates, may ward off the interaction in a pH-dependent manner or the structures could be shielded by remaining E3 glycopeptides. Virus exposed to lower pH gradually made the epitope regions available for interaction with sensor surfaces. The mabE1n epitope showed a transient behavior, peaking at pH values between 7.0 and 6.5, whereas the interaction with other antibody surfaces increased in a continuous manner, as pH was lowered. The mabE2r, representing part of the putative receptor-binding domain in the E2 sequence, established a first level of interaction, about 50% of maximal, at pH 7-6.5, but then increased to its optimal value at a pH of about 6.0. The mabE2m-and then the mabE2v-coated surfaces became increasingly more attractive for the virus, with differences in onset and peaking as the pH decreased (Fig. 3). As discussed above, the full appearance of mabE2m epitope relates to the dissociation of the E1-E2 interaction, which should be needed to allow rearrangements in the well-developed protein network between the spikes in the shell domain (56).
The fusion peptide, identified by mabE1f, or its Fab fragment, was maximally exposed at around pH 6 (Figs. 3 and 6 and Table II). As judged from our results, the maximal exposure of this epitope, which approaches one per three theoretical sites in the virion, follows an essentially irreversible stage in the adoption to a lowered pH. This happens concomitantly to the observed transformation of the virion to a larger, and notably, still free particle (Fig. 1). The pH threshold for these transformations conforms to reported data on E1 oligomerization, possibly trimerization (61,62). The loss of the neutralization domain, probed by mabE1n, occurs in parallel to proposed E1-E2 dissociation, but before the formation of a "stable" E1 oligomer. Therefore, E1 oligomerization is a late step on the way to exposing the fusion peptide domain (62). Together, the maximal exposure of the mabE2r and mabE1f epitopes reflects the hoisting of the external protein domain (56) with a slight screwing move in the protrusions, as noted in the acid structure (21,56).
Then, when the high resolution structure of the alphaviruses are emerging, due to cryoEM advancement (15)(16)(17)(18)(19)(20)(21)(22)56), the location of functional domains awaits to be identified. Unveiling the nature of the fusion mechanism would be of particular value. Because analogous motifs for a type I, influenza type, mechanism are lacking in the structure, the mechanism has been assigned as type II (46,76). Prefusion rearrangements in the alphaviruses involve dissociation of the glycopeptides E1-E2 in the external domain of the virion. This would facili-tate formation of an observed, stable E1 oligomer, assumed to be an essential element for fusion peptide exposure (33,64,77). The present study provides appropriate probes for following the triggering steps involved and gives their relative order, as the environmental pH is lowered.