Quaternary Structure of Coronavirus Spikes in Complex with Carcinoembryonic Antigen-related Cell Adhesion Molecule Cellular Receptors*

Oligomeric spike (S) glycoproteins extend from coronavirus membranes. These integral membrane proteins assemble within the endoplasmic reticulum of infected cells and are subsequently endoproteolyzed in the Golgi, generating noncovalently associated S1 and S2 fragments. Once on the surface of infected cells and virions, peripheral S1 fragments bind carcinoembryonic antigen-related cell adhesion molecule (CEACAM) receptors, and this triggers membrane fusion reactions mediated by integral membrane S2 fragments. We focused on the quaternary structure of S and its interaction with CEACAMs. We discovered that soluble S1 fragments were dimers and that CEACAM binding was entirely dependent on this quaternary structure. However, two differentially tagged CEACAMs could not co-precipitate with the S dimers, suggesting that binding sites were closely juxtaposed in the dimer (steric hindrance) or that a single CEACAM generated global conformational changes that precluded additional interactions (negative cooperativity). CEACAM binding did indeed alter S1 conformations, generating alternative disulfide linkages that were revealed on SDS gels. CEACAM binding also induced separation of S1 and S2. Differentially tagged S2 fragments that were free of S1 dimers were not co-precipitated, suggesting that S1 harbored the primary oligomerization determinants. We discuss the distinctions between the S·CEACAM interaction and other virus-receptor complexes involved in receptor-triggered entry.

Oligomeric spike (S) glycoproteins extend from coronavirus membranes. These integral membrane proteins assemble within the endoplasmic reticulum of infected cells and are subsequently endoproteolyzed in the Golgi, generating noncovalently associated S1 and S2 fragments. Once on the surface of infected cells and virions, peripheral S1 fragments bind carcinoembryonic antigen-related cell adhesion molecule (CEACAM) receptors, and this triggers membrane fusion reactions mediated by integral membrane S2 fragments. We focused on the quaternary structure of S and its interaction with CEACAMs. We discovered that soluble S1 fragments were dimers and that CEACAM binding was entirely dependent on this quaternary structure. However, two differentially tagged CEACAMs could not co-precipitate with the S dimers, suggesting that binding sites were closely juxtaposed in the dimer (steric hindrance) or that a single CEACAM generated global conformational changes that precluded additional interactions (negative cooperativity). CEACAM binding did indeed alter S1 conformations, generating alternative disulfide linkages that were revealed on SDS gels. CEACAM binding also induced separation of S1 and S2. Differentially tagged S2 fragments that were free of S1 dimers were not co-precipitated, suggesting that S1 harbored the primary oligomerization determinants. We discuss the distinctions between the S⅐CEACAM interaction and other virusreceptor complexes involved in receptor-triggered entry.
For enveloped viruses, efficient infection requires a regulated coalescence of virion and cellular membranes. Temporal and spatial regulation of this membrane fusion event must occur for viral genomes to enter into a milieu suitable for subsequent replicative processes. Protruding virion glycoproteins, each poised to induce membrane coalescence, have therefore evolved sensitivities to the environmental conditions found at entry sites. These conditions trigger coordinated and irreversible changes in virion glycoprotein conformations that can culminate in membrane fusion. Well known triggers for conformational change include cellular receptor binding (1)(2)(3)(4) and/or the low pH exposures that occur following engulfment of virus particles into endosomes (5)(6)(7).
Our studies have focused on murine hepatitis coronavirus (MHV) 1 as a model for understanding receptor-triggered entry processes. This virus is a well studied prototype member of the Coronaviridae, plus-strand RNA viruses that cause a wide range of diseases in humans and animals (8). Because the distinct species specificity and tissue tropism of coronavirus strains largely correlate with changes in the spike (S) protein (9 -11), details about S interactions with receptors can enhance our understanding of pathogenesis.
S proteins are classical type I membrane proteins, with ϳ1300 residue ectodomains, 18-residue transmembrane spans, and a 38-residue cytoplasmic tail (12). Oligomerization occurs rapidly after synthesis (13) and is followed by transport through the exocytic pathway. Within the trans-Golgi network, a furin-like protease cleaves the full-length spike into two similar-sized fragments, a peripheral S1, and a membraneanchored S2 (14,15). S1, which associates with S2 through noncovalent interactions, is responsible for binding to cellular receptors. S2 contains the core machinery necessary for membrane fusion (16).
Receptors for the MHV S proteins include numerous members of the carcinoembryogenic antigen-related cell adhesion molecule (CEACAM) family, immunoglobulin-like glycoproteins that serve as entry portals for a relatively wide variety of pathogens (17)(18)(19)(20)(21). The prototype receptor for MHV, murine CEACAM isoform 1 a , is a type I transmembrane glycoprotein with four Ig-like ectodomains designated as N (amino-terminal)-A1 a -B a -A2 a (22). The N-domain binds to S proteins (17). After binding to a soluble N-CEACAM fragment, spikes undergo a conformational change that can, in some cases, be revealed as S1 shedding from S2 (23). This structural change may be relevant to MHV entry, as S1 separation from S2 correlates with increased membrane fusion activity (23). A conservative view is that the CEACAM binding to S releases free energy that drives the conformational changes required to promote coalescence of the virus and cell membranes. Indeed, soluble forms of CEACAM can, through binding S proteins, increase the propensity of S to fuse membranes (24).
Understanding the connections between CEACAM binding and membrane fusion depends in part on a view of the actual CEACAM-binding site(s) on the S protein. Kubo et al. (25) mapped the CEACAM-binding sites to the amino-terminal 330 residues of S1, but high resolution protein structures are cur-rently unknown. Questions also remain concerning quaternary structures, both S dimers and trimers have been proposed (13,26). Thus, we embarked on studies assessing the oligomeric organization of S and S⅐CEACAM complexes.
Here we report that S1, when shed from S2 or when produced independently from cDNAs, exists stably as a dimer. We discovered that S1 dimers, but not monomers, will bind to CEACAM receptors. In fact, we found that the region conferring dimerization resides at or near the CEACAM1 a -binding site, an unexpected finding because oligomerization determinants in functionally analogous spike proteins of other viruses reside within the integral membrane fragments (27)(28)(29)(30)(31)(32)(33)(34)(35). Remarkably, only one CEACAM bound each S1 dimer, and we identified a novel disulfide-linked S1 conformation in S1⅐CEACAM complexes. Our findings refine the current understanding of CEACAM receptors as mediators of conformational change, and they form the basis for a preliminary model of receptor-triggered entry with both parallels and deviations from established paradigms for enveloped virus entry (36 -41).
Mutagenesis of CEACAM and Spike cDNAs-We used murine CEACAM1 a cDNA (22,42) as template for PCR amplification of N-CEACAM 6ϫHis , using the primer 5Ј GTCGAGTCAGTGGTGGTGGT-GGTGGTGTACATGAAATCG 3Ј, which encodes a hexahistidine tag. We used cDNA of S (strain JHM) (43) as a template for PCR amplification of S gene and truncated S fragments. To create S T212S/Y214S/Y216S , mutagenic oligonucleotides 5Ј GGTGGTTCTTTTTCTGCGTCCTATGC-GGAT 3Ј and its complement were used in PCRs. To generate enhanced green fluorescent protein (EGFP)-tagged spikes, we engineered pTM1-S (42) with a unique NotI restriction site using the following oligonucleotide: 5Ј GGGCTCGAGTCAGCGGCCGCTCACAGGGATCCAGTGCAT-CCTCATGGGC 3Ј. EGFP DNA was PCR-amplified from pEGFP (Clontech), and 741-nucleotide Not-I/BamHI restriction fragment was cloned into the aforementioned pTM1-S (NotI) .
Mutations in the CEACAM and S genes were confirmed by DNA sequencing. Restriction fragment exchanges with the vaccinia virus insertion-expression vector pTM1-S and pTM3-S1 were all performed as described previously (42). All recombinant plasmids were cloned and amplified in E. coli DH5␣ (for pTM1 vector) or HB101 (for pTM3 vector). Plasmids pTM1-S ⌬DPR2 (23) and pTM1-S EGFP were used directly, without recombination into vaccinia vectors. The S EGFP protein includes the entire S ⌬DPR2 followed by an eight-residue linker (ALDPPVAT) and a C-terminal 238-residue EGFP.
Preparation of Soluble CEACAM and Spike Proteins-To obtain N-CEACAM Fc , 293 EBNA:N-CEACAM Fc cells (42) were incubated overnight in serum-free DMEM. Culture supernatant was collected, filtered through a 0.22-m membrane, dialyzed against PBS-P (PBS (pH 7.4) containing 0.01% protease inhibitor mixture (Sigma)), and concentrated 100-fold by ultrafiltration. In some cases, N-CEACAM Fc was further purified by affinity chromatography on Sepharose-protein G (Amersham Biosciences). Supernatants typically yielded ϳ2 g of N-CEACAM Fc per ml.
To obtain 35 S-labeled recombinant S proteins and CEACAM ECTO , monolayers of HeLa-tTA cells were inoculated at 2 plaque-forming units/cell for 1 h at 37°C with vTF7.3 and the respective recombinant vaccinia viruses. At 6 h post-infection, the medium was replaced with labeling media (DMEM, 1% dialyzed FCS lacking cysteine and methionine). After 1 h, the labeling media was replaced with serum-free labeling media supplemented with 25 Ci/ml Tran 35 S-label (ICN). After a 5-h incubation, the harvested media were clarified by centrifugation, dialyzed, and concentrated ϳ100-fold by ultrafiltration as described above.
Immunoprecipitations-S proteins were collected from media or from cytoplasmic extracts. Extracts were obtained by lysing infected cell monolayers with PBS-P containing 0.5% Nonidet P-40, followed by removal of nuclei by centrifugation at 3000 ϫ g for 15 min. S proteins were immunoprecipitated with N-CEACAM Fc , with polyclonal anti-JHM serum (R33 serum, a gift from Dr. Stanley Perlman, University of Iowa), or with monoclonal anti-S antibody J.2.6 (J.2.6 hybridoma (46), a gift from Dr. John Fleming, University of Wisconsin, Madison) or with monoclonal anti-S antibody number 2 (a gift from Dr. Fumihiro Taguchi, National Institute of Neuroscience, Tokyo, Japan). Briefly, these Igs were bound for 4 h at 4°C to Gamma Bind G-Sepharose beads (Amersham Biosciences). The beads were then rinsed three times with PBS-P by centrifugation and resuspension. After overnight incubation at 4°C with media or cytoplasmic extracts, beads were rinsed by five cycles of centrifugation and resuspension with PBS-P containing 0.5% Nonidet P-40. The final bead pellets were mixed with SDS solubilizer (2% SDS, 5% ␤-mercaptoethanol (␤-ME), 2.5% Ficoll, 0.005% bromphenol blue) for 5 min at 100°C. Dissolved proteins were then visualized after SDS-PAGE by fluorography or immunoblotting, as described previously (23).
Velocity Gradient Ultracentrifugation-Samples containing 35 S-labeled S1 or S1⅐CEACAM complexes were overlaid onto linear 5 ml of 5-20% w/w sucrose gradients in PBS-P containing 0.01% BSA. A parallel gradient was overlaid with an extract containing the sedimentation markers horseradish peroxidase (HRPO 4 S), human immunoglobulin G1 (IgG 7 S), and E. coli ␤-galactosidase (16 S). After sedimentation at 55,000 rpm at 5°C for 5.95 h in a Beckman Spinco SW55 rotor, fractions (20 per gradient) were collected. The S proteins in gradient fractions were then immunoprecipitated and visualized by fluorography after SDS-PAGE. Sedimentation standards were identified in the fractions by enzymatic assays (HRPO by turnover of 2,2Јazino-bis(3-ethylbenzthiazoline-6-sulfonic acid substrate; IgG by immunoblotting with goat anti-human IgG:alkaline phosphatase; ␤-galactosidase by turnover of chlorophenol red-␤-D-galactopyranoside substrate).
Co-production and Immunoprecipitation of S2 and S2 EGFP -vTF7.3infected HeLa-tTA cells were lipofected with pTM1-S and with pTM1-S EFGP , alone or together, using LipofectAMINE PLUS according to manufacturer's instructions (Invitrogen). At 4 h post-lipofection, media were removed and replaced for 1 h with labeling media and then 5 h with labeling media containing 25 Ci/ml Tran 35 S-label. Cell monolayers were lysed with PBS-P (pH 8.5) containing 0.5% Nonidet P-40, and nuclei were pelleted by centrifugation (3000 ϫ g for 10 min at 4°C).
To separate 35 S-labeled S1 from S2, leaving S1 associated with Sepharose beads, 35 S-labeled S proteins in clarified cytoplasmic extracts were captured by incubation for 4 h at 4°C with Sepharose G:N-CEACAM Fc beads, and suspensions were incubated for an additional 4 h at 37°C before pelleting beads (3000 ϫ g for 10 min at 4°C). Supernatants enriched in S2 were further depleted of residual S1 by two additional cycles of incubation with fresh Sepharose G:N-CEACAM Fc beads. The final supernatants were then incubated overnight at 4°C with Sepharose G:anti-GFP antiserum to capture S2 EGFP fragments. Beads were rinsed extensively with PBS-P containing 0.5% Nonidet P-40, suspended in SDS solubilizer, and heated to 100°C for 5 min. Dissolved proteins were visualized by immunoblot with anti-S2 mAb 10G (a gift from Drs. Stuart Siddell and Fumihiro Taguchi) (47) after SDS-PAGE. Fluorograms of immunoblots were obtained using a Molecular Dynamics Typhoon 8600 PhosphorImager.

RESULTS
Peripheral S1 Fragments Are Dimers-It is still unclear whether coronavirus peplomers are dimers or trimers. Vennema et al. (13) reported that MHV spikes are dimers, and Delmas and Laude (26) provided evidence for cross-linking of transmissible gastroenteritis coronavirus spikes into trimers. Noting the importance of quaternary structure in viral glycoprotein function, we decided to return to the question of S protein oligomerization in the MHV system. We initially used velocity gradient ultracentrifugation and chemical cross-linking to determine the quaternary structure of peripheral S1 fragments. We found that S1 produced independently from cDNA sedimented to an ϳ9 S position on sucrose gradients (Fig. 1A). Identical results were obtained for S1 fragments that had separated from S2 (data not shown). Formulas based on isokinetic sedimentation of globular proteins indicated that the ϳ9 S material would have a molecular mass of ϳ200 kDa, consistent with S1 homodimers (48). However, elongated molecules like the coronavirus spike peplomer (49) might exhibit unusual sedimentation behavior in sucrose gradients; therefore, we further addressed quaternary structure by cross-linking 35 S-labeled S1 with DSP, a thiol-cleavable chemical crosslinker. Cross-linked spikes were then immunoprecipitated and electrophoresed under non-reducing and reducing conditions (Fig. 1B). S1 dimers appeared with increasing concentrations of DSP, and ␤-ME reduced these dimers into monomers. Extraordinarily high DSP concentrations (25 mM) did not complex 35 S-labeled S1 into higher order oligomers (data not shown).
CEACAM Receptor-binding Sites Are Only Present in S Oligomers-In previous experiments, we found that polyclonal anti-spike antibodies captured newly synthesized 35 S-labeled S proteins, whereas N-CEACAM Fc , an immunoadhesion consisting of the N-domain of murine CEACAM1 a linked to a carboxylterminal IgG1 Fc, did not. The 35 S-labeled spikes bound N-CEACAM Fc only after ϳ30 min of maturation (23). One possible explanation for this finding was that S proteins oligomerized during the 30-min maturation process and that soluble receptors only recognized oligomers. This contention was consistent with numerous reports that glycoprotein oligomerization is required to maintain native tertiary structures (50,51). Therefore, we separated newly synthesized spikes by rate-zonal sedimentation through sucrose gradients prior to immunoprecipitation and detection on SDS-polyacrylamide gels. Polyclonal anti-spike serum captured a range of spike forms from ϳ6 S to ϳ14 S (Fig. 2, 0 hour anti-S). In contrast, N-CEACAM Fc specifically immunoprecipitated ϳ14 S macromolecules (Fig. 2, 0 hour N-CEACAM Fc ). When a 2-h chase period occurred prior to cell lysis and sedimentation, N-CEACAM Fc and anti-S antiserum captured only the ϳ14 S forms (Fig. 2, 2 hour panels). The two endoproteolytic cleavage products S1 (lower band) and S2 (upper band) indicated that most of the spikes had encountered a trans-Golgi-localized furin-like protease (14). These findings indicate that newly synthesized S proteins form CEACAM-binding sites concomitant with their oligomerization.
We next considered whether CEACAM-binding sites disappeared when S proteins dissociated into monomers. We could address this question because our S1 preparations moderately break down when incubated for 2 h at 37°C. On sucrose gradient sedimentation, the 37°C-treated S1 occupied two positions, ϳ9 S and ϳ6 S, with ϳ6 S being consistent with 110kilodalton monomers (48) (Fig. 3, top panel). N-CEACAM Fc only precipitated the ϳ9 S material (Fig. 3, bottom panel), suggesting that monomers do not contain a CEACAM-binding site. Interestingly, soluble ectodomain fragments of S2 co-sedimented with the ϳ6 S S1 monomers in these gradients.
Each S Dimer Binds One CEACAM Molecule-Because receptors can perform an essential role in reorganizing viral spikes into structures that can mediate membrane fusion (2)(3)(4)53), we sought further details on S⅐CEACAM interactions. Our data indicated that relatively small S fragments bound CEACAM receptors only when combined into dimers. However, it remained uncertain whether multiple receptors could coordinately bind a single S1 dimer. We addressed this question initially by sedimenting S1⅐N-CEACAM Fc complexes on sucrose gradients. If each S1 monomer contained a separate CEACAM-binding site, then two dimers might link to the bivalent N-CEACAM Fc to form an estimated 16 S complex (48). Higher order complexes also may form at equivalent S1:N-CEACAM Fc ratios. However, we observed only ϳ16 S complexes and no evidence of higher order aggregates (Fig. 5).
We obtained additional insight into the stoichiometry of S1⅐N-CEACAM Fc complexes by co-producing both ligands in 293 cells in the presence of Tran 35 S-label, ensuring that the synthesis of S1 exceeded that of N-CEACAM Fc . From the radioactive media, we then precipitated 35 S-labeled N-CEACAM Fc in complex with 35 S-labeled S1 using Sepharose-protein G beads. After electrophoretic separation of the precipitated proteins, we determined 35 S-labeled S1: 35 S-labeled N-CEACAM Fc ratios of 3.40 Ϯ 0.05 (n ϭ 3). Complexes in which one S1 dimer is tethered to each arm of the bivalent N-CEACAM Fc (Fig. 5A) would be expected to have an 35 Slabeled S1: 35 S-labeled N-CEACAM Fc ratio of 3.428 (42,43).
We next used a series of co-immunoprecipitation tests to further address whether more than one CEACAM could simultaneously bind an S oligomer. A soluble CEACAM bound to S1 may or may not prevent the subsequent binding of another alternatively tagged (and thus distinguishable) soluble receptor. We bound recombinant S1 dimers to 35 S-labeled CEAC-AM ECTO , which contains the four ectodomain fragments (N-A1-B-A2) of murine CEACAM1 a (17,22). We then immunoprecipitated these complexes with N-CEACAM Fc , which possesses the unique, easily captured, Fc tag. Immunoprecipi-2 T. M. Gallagher, unpublished observations. FIG. 3. Influence of S1 quaternary structure on its ability to bind CEACAM. 35 S-Labeled S ECTO in sucrose gradient fractions was precipitated with trichloroacetic acid (top panel) or with N-CEACAM Fc (bottom panel) and then detected by fluorography following SDS-PAGE. The sedimentation marker immunoglobulin G (IgG 7S) was identified in fractions from a parallel gradient by immunodetection assays, and its position is indicated above the electropherogram. The positions of S1 and S2 ECTO are also indicated.
FIG. 4. Co-immunoprecipitation of amino-terminal fragments S1 330 and S1 769 . Recombinant S1 fragments of 330 or 769 residues were synthesized alone or together in HeLa cells in the presence of Tran 35 S-label. S fragments in the media (top panel) and cell lysates (bottom panel) were immunoprecipitated with N-CEACAM Fc or with anti-spike mAb J.2.6. The J.2.6 epitope is between S1 residues 510 and 540. 35 S-Labeled proteins were visualized by fluorography following SDS-PAGE.
FIG. 5. Sucrose gradient sedimentation analysis of unbound S1 and S1⅐N-CEACAM Fc complexes. A, two possibilities for the CEACAM-binding site architecture on a S1 dimer are illustrated. If only one CEACAM binds each S1 dimer, then divalent N-CEACAM Fc would bind one or two S1 dimers. Alternatively, two CEACAM-binding sites could complex S1 and N-CEACAM Fc into higher order oligomers. B, unbound 35 S-labeled S1 and 35 S-labeled S1⅐N-CEACAM Fc complexes were sedimented on linear sucrose gradients and detected in gradient fractions after immunoprecipitation, SDS-PAGE, and fluorography. The sedimentation markers immunoglobulin G (IgG 7S) and ␤-galactosidase (B-gal 16S) were identified in fractions from a parallel gradient by enzyme or immunodetection assays, and their positions are indicated above the electropherograms. An ϳ16 S S1⅐N-CEACAM Fc complex was identified. There was no evidence of larger complexes in pellet fractions (not shown). tation of the 35 S-labeled CEACAM ECTO would indicate that both receptors bound simultaneously to S1 dimers. This receptor-binding state would be consistent with an S oligomer whose structure parallels those of influenza hemagglutinin and HIV gp120, in which each monomer (of the trimer) can bind a single cell-surface ligand (27,54).
We considered the possibility that the large carboxyl-terminal Fc "tags" might cause an "artificial" steric hindrance in these tests. Thus, in parallel co-precipitation tests, we replaced N-CEACAM Fc with N-CEACAM 6ϫHis , whose 6-residue appendage is roughly 100 times smaller than IgG Fc. N-CEACAM 6ϫHis also could not precipitate S1⅐ 35 S-labeled CEACAM ECTO complexes (Fig. 6, Lower, B), but it could readily bind and precipitate free S1 dimers (Fig. 6, Lower, A). Thus, for steric hindrance to account for this interference by CEACAM ECTO , separate binding sites on each S1 monomer would have to be very closely juxtaposed in the dimer.
We also considered the unconventional possibility that an S1 dimer contains only a single asymmetric binding site for CEACAM that is formed by different residues contributed by each S1 monomer. We synthesized S1 proteins with the changes T212S, Y214S and Y216S, which Suzuki and Taguchi (55) had shown to decrease CEACAM binding. We found that N-CEACAM Fc inefficiently precipitated this mutant, but its capture increased 2-fold when hetero-oligomerized with wildtype S 330 (Fig. 7 boxed bands). These results argue for a traditional view, in which each S1 monomer (in the context of a dimer) contains a complete CEACAM-binding site.
A CEACAM-induced Conformational Change of S1-Our data suggested that each monomer of the S1 dimer contained an independent CEACAM-binding site. Although sterically hindered sites might explain how only one CEACAM binds S1 oligomers, another possibility was negative cooperativity. For negative cooperativity to be a viable possibility, CEACAM would have to bind one S1 monomer (of the dimer) and induce structural rearrangements such that the adjacent monomer would have substantially reduced CEACAM affinity. Thus, we evaluated whether CEACAM binding induces structural rearrangements in S1.
On considering possible assays for structural rearrangements, we noted that reduction strongly affects the electrophoretic mobility of some variant S1 fragments, ϳ70 and 90 kDa before and after ␤-ME treatment, respectively. Thus we entertained the possibility that CEACAM binding might induce conformational changes that rearrange complex disulfide architectures, thereby creating electrophoretic mobility shifts. It is important to note that we felt any positive results might even have some biological relevance, as we had shown earlier (56) that chemicals preventing disulfide rearrangements block MHV infection by arresting S-induced membrane fusion.
We exposed 35 S-labeled S1 dimers to the sulfhydryl alkylating agent N-ethylmaleimide (NEM) either before or after being complexed with N-CEACAM Fc at 37°C. In parallel, a monoclonal anti-S1 330 antibody (number 2) (25) was used in place of N-CEACAM Fc , with the expectation that its binding would not induce conformational changes. Electropherograms of the immunoprecipitated 35 S-labeled proteins revealed that a small proportion of CEACAM-bound S1 had complexed into an ϳ220-kDa disulfide-linked protein (Fig. 8A, lane 4). Thiols apparently had to be available to generate the ϳ220-kDa protein, as S1 that had been pretreated with NEM bound N-CEACAM Fc but did not couple into disulfide-linked oligomers (Fig. 8A, lane  3). Formation of the ϳ220-kDa protein required N-CEACAM Fc , because mAb number 2 bound S1 but did not generate any electrophoretic mobility changes (Fig. 8B, lanes 3 and 4). Collectively, these data indicate that CEACAM binding can induce structural rearrangements in the S1 dimer, revealed in this experiment by alternative disulfide linkages.
Quaternary Structure of S2 Fragments after Separation from S1-High resolution structural data are available for portions  7. Increased capture of mutant S proteins with low affinity CEACAM-binding sites by heteromerization with S1 330 fragments. Recombinant mutant S proteins with three point mutations within the CEACAM-binding site were synthesized alone or with increasing amounts of "wild-type" S1 330 . S1 330 synthesis was adjusted by the input multiplicity of infection (MOI) of the vTM3-S1 330 vector. 35 S-labeled spikes were immunoprecipitated with N-CEACAM Fc and visualized by fluorography following SDS-PAGE. The amount of 35 S associated with the boxed bands was quantitated with a Molecular Dynamics Typhoon 8600 PhosphorImager. The left box contained 2141 cpm and the right box contained 4329 cpm. of many different viral spike proteins, and the images reveal strong intersubunit interactions within integral membrane (fusion-inducing) post-translational fragments (6, 30 -34, 36, 57). Less structural data are available for the peripheral (receptor binding) fragments, but collected information often leads to models in which these peripheral subunits separate from each other during membrane fusion reactions (36,38). This "opening" of spike oligomers may expose hydrophobicity within the integral membrane fragments, a prerequisite for membrane fusion. Our discovery of stable intersubunit connections in coronavirus S1 led us to doubt whether these peripheral subunits separate during fusion, and also led us to speculate about additional oligomerization determinants in integral membrane S2 fragments.
We engineered S proteins with a relatively large 238-residue EGFP appended to their cytoplasmic tails. We found that the carboxyl-terminal additions had no effect on S protein transport through the exocytic pathway, indicating that oligomerization motifs in the S ectodomains remained. Moreover, S EGFP proteins induced membrane fusion (data not shown). Therefore, these tagged S proteins allowed us to identify intersubunit associations through co-immunoprecipitation experiments. In one set of tests, we synthesized S and S EGFP , either separately or together, in the presence of Tran 35 S-label. We then immunoprecipitated all EGFP-tagged spikes from total cytoplasmic extracts using GFP-specific rabbit antiserum (a gift from Dr. Katherine L. Knight, Loyola University Medical Center), and we visualized S2 proteins by immunoblotting (Fig. 9A). Uncleaved S EGFP and S2 EGFP were detected (lane 2), and when co-produced with S, the untagged S2 was also detected (lane 3). This association of S2 EGFP with untagged S2 was not due to a generalized aggregation of S2 chains after cell lysis, because there was no capture of S2 from mixtures of S and S EGFP lysates (lane 4). The abundance of 35 S-labeled S1 in the immunoprecipitated material suggested that S1 dimers might be responsible for holding S2 and S2 EGFP together (Fig. 9B).
We next separated S1 dimers from S2 by tethering the spikes onto Sepharose:N-CEACAM Fc and then incubating at pH 8.5 and 37°C. This condition separates S1 from S2 (58), leaving "free" S2 chains in supernatants. Sequential incubations with Sepharose:N-CEACAM Fc generated S1-depleted supernatants from which anti-GFP serum immunoprecipitated free S2 chains. Here there was capture of S2 EGFP (Fig. 9C, lanes 2 and  4), but there was no evidence of co-immunoprecipitating untagged S2 (lane 3). We correlated this co-immunoprecipitation failure with the absence of S1 in the samples (Fig. 9D). Collec-tively, these findings suggested that S2, when free of S1, does not exist as an oligomer. DISCUSSION The oligomeric spike glycoproteins of many enveloped viruses are endoproteolytically cleaved into two fragments that act in concert to mediate virion binding to receptors and subsequent uncoating through virion:cell membrane fusion. Crystal structures of the influenza hemagglutinin reveal an exterior composed of peripheral residues and their receptor-binding sites, and a core which harbors much of the integral membrane fragments, thereby sequestering the hydrophobic residues involved in membrane fusion. The dramatic conformational changes of integral membrane fragments that link opposing membranes can only be accomplished in concert with some displacement of the peripheral fragments. How this displacement process takes place remains unclear, i.e. whether the peripheral fragments symmetrically dissociate into monomers to reveal the membrane fusion apparatus, as depicted in some models (36,38), or whether they displace asymmetrically as oligomers. We require additional insights into this process to understand the mechanisms by which antibodies neutralize virus infections and to develop chemicals that interfere with virus entry.
This issue of spike protein quaternary structures both before and after fusion activation has been studied in some detail with primate lentiviruses, and some interesting findings have emerged. The peripheral (gp120) fragments that shed from spike complexes during membrane fusion can be monomers, indeed they were crystallized in this form (54), but can also FIG. 8. CEACAM-induced conformational changes in S1. 35 S-Labeled S1 ⌬DPR1 was either incubated with (ϩ) or without (Ϫ) 10 mM NEM prior to incubation for 4 h at 4°C and then for 1 h at 37°C with 10 g of N-CEACAM Fc (A) or anti-S1 330 mAb number 2 (B) (25). Samples not previously treated with NEM were then incubated with 10 mM NEM (ϩ). 35 S-Labeled proteins were detected by fluorography using a Molecular Dynamics Typhoon 8600 PhosphorImager following SDS-PAGE under reducing (ϩ B-ME) and non-reducing (Ϫ B-ME) conditions. N-CEACAM Fc -induced disulfide-linked high molecular weight spikes are indicated by the *.
FIG. 9. Analysis of the oligomeric organization of S2 after separation from S1. A, cDNAs encoding S or EGFP-tagged S (S EGFP ) were transfected alone or together into vTF7.3-infected HeLa cells. Following metabolic labeling with Tran 35 S-label, cytoplasmic extracts were prepared, and EGFP-associated proteins were immunoprecipitated with polyclonal anti-GFP serum. Immunoprecipitates were then electrophoresed and immunoblotted for S2 fragments. Co indicates co-synthesis of S and S EGFP . Mix indicates that equal volumes of independently produced S and S EGFP lysates were mixed before immunoprecipitation. B, autoradiographic image of the immunoblot in A. C, cell lysates were depleted of S1 by sequential immunoprecipitations with N-CEACAM Fc at pH 8.5 and 37°C. Supernatants from the final depletions were collected, and EGFP-associated proteins were immunoprecipitated with anti-GFP serum. Immunoprecipitates were electrophoresed and immunoblotted for S2 fragments. D, autoradiographic image of the immunoblot in C. Arrows indicate the positions of uncleaved S EGFP (S unc-EGFP ), S2 EGFP , untagged S2, and S1. exist as stable dimers (59 -61) or trimers (62)(63)(64). By contrast, it is generally accepted that the integral membrane (gp41) fragments exist as trimers, at least in postfusion low energy conformations (29,30,32,34,65). Recent findings also indicate that different gp160 subunits, one with a lethal defect in receptor binding and the other unable to induce membrane fusion, can assemble together into functional heteromeric trimers (66). These findings naturally lead to the hypothesis that asymmetries can exist within gp120 -41 complexes, at least in some of the conformations existent during fusion activation.
Similarly, it is conceivable that asymmetries exist in murine coronavirus spikes. We found that the peripheral (S1) fragments of MHV exist as stable, homogenous dimers (Fig. 1). Although we view our data as convincing, we understand its apparent inconsistency with previous reports of trimeric associations within MHV integral membrane (S2) fragments (67). Collective findings therefore raise the possibility that asymmetric dimer-to-trimer transitions occur as part of the pathway to membrane fusion activation. In this view, closely spaced S1 fragments would dissociate as stable dimers, leaving S2 fragments to rearrange during membrane fusion into trimeric structures. Such dimer-trimer transitions are not unprecedented in virus entry, although they take place in the context of icosahedrally ordered glycoprotein lattice (68,69). No evidence exists for ordered arrangements of spikes on the pleiomorphic coronavirus envelopes.
Another interesting and potentially relevant asymmetry likely exists in S⅐CEACAM complexes. Only a single CEACAM binds to an S1 dimer. This was most convincingly demonstrated in our co-immunoprecipitation tests, where differentially tagged N-CEACAM fragments never captured S1⅐CEACAM complexes (Fig 6). Further support for a one S1 dimer:one CEACAM ratio came from our stoichiometric analyses of S1⅐N-CEACAM Fc complexes (Fig. 5). As we consider it unlikely that only a single asymmetric binding site exists on each S1 dimer (Fig. 7), we propose two alternative possibilities. Either the two CEACAM binding sites on an S1 dimer juxtapose very closely (steric hindrance) or the binding of a single CEACAM rapidly induces global structural changes in S1 that destroy the adjacent binding site (negative cooperativity). This latter possibility has received consideration in the SIV gp140⅐CD4 interaction, whose stoichiometry has recently been identified as an asymmetric complex of one gp140 trimer bound to a single monomeric CD4 (63). Our data, although not yet able to distinguish between the two possibilities, nonetheless provides evidence of structural flexibility in the S proteins and thus points toward negative cooperativity as a likely scenario. We and others know that CEACAM binding induces separation of S1 from S2, a readily observed "global" conformational change (23,58). The CEACAM-binding site is itself dependent on a more global conformation, being formed with assembly of S into oligomers (Fig. 2), and eliminated on S1 dissociation into monomers (Fig. 3). This relationship between S oligomerization and CEACAM binding may provide the structural contexts for conformational changes across S1 monomers once a CEACAM molecule binds. In support of this view, we demonstrated that N-CEACAM Fc binding induced the formation of high molecular weight disulfide-linked S1 structures whose formation was blocked by pre-incubation with the sulfhydryl-alkylating agent N-ethylmaleimide (Fig. 8A). The inability of S1 330 -specific mAb number 2 to induce comparable disulfide rearrangements (Fig.  8B) further supports the contention that entry-specific conformational changes in S1 are unique to CEACAM binding. Although these findings are interesting, perhaps even suggesting a role for disulfide exchanges during coronavirus entry (56), we acknowledge that additional CEACAM-induced changes in S1 structure must be investigated to fully address how S conformations bring about virus:cell membrane fusion.
We were surprised to find oligomerization control in peripheral S1 fragments, because analogous viral glycoproteins oligomerize into trimers through interactions in their integral membrane fragments. Our attempts to determine whether the S1-interacting sites represent the only oligomerization motif prompted us to construct epitope-appended S proteins, and we discovered that tags as large as the 238-residue EGFP could be added to the carboxyl (cytoplasmic) termini of S2 without interrupting oligomeric assembly, intracellular transport, and function. With these epitope-tagged S proteins, it became relatively straightforward to determine whether S2 fragments retained an oligomeric structure even after separation from S1. Thus we bound soluble CEACAMs and shifted to 37°C, a condition known to dissociate S1 from S2 (23). We expected that this would leave free S2 chains in an oligomeric state, as primary S2 sequences predict oligomeric coiled-coils (70) similar to those found in many other viral spike proteins that carry out membrane fusion reactions (29 -34, 65). However, we did not co-immunoprecipitate S2⅐S2 EGFP complexes when free of S1, suggesting S2 monomers (Fig. 9). This interesting finding raises questions about the ways that MHV S2 might fuse membranes. The generally accepted view is that the integral membrane fragments of enveloped viruses collapse into ␣-helical coiled-coils on bringing opposing membranes together (36 -41). It remains to be determined whether MHV S2 monomers independently adopt an anti-parallel ␣-helical coiled-coil arrangement that is similar to that observed for other viruses.
A preliminary model of S proteins during virus entry is shown in Fig. 10. In this figure, we depict a single CEACAM that is bound to peripheral S1 (25,55). This bound CEACAM may preclude additional binding by steric hindrance (Fig. 10A) or alternatively may induce conformational changes in S1 oligomers that precludes additional CEACAM binding (Fig. 10B). To explain a single CEACAM-binding site on S1 dimers, two models are illustrated. Model A appeals to steric hindrance, and model B suggests that a single CEACAM induces global structural rearrangements (illustrated as the change of S1 from an oval to a rectangle). These conformational changes preclude additional CEACAM binding. In both cases, CEACAM binding is hypothesized to displace S1 from S2 (23) and to permit the insertion of an internal fusion peptide (green triangle) into the target cell membrane (71).
Additional conformational changes are thought to occur at S1-S2 connections, resulting in the displacement of the two fragments from each other (23). These events are considered prerequisites for the insertion of a hydrophobic portion of S2 into cellular membranes (71). As little is yet known regarding the later stages of the membrane fusion process, subsequent depictions of S2 structures are not illustrated. Coronavirus S2 fragments do contain putative M-helix and C-helix regions (Fig. 10), which are predicted to form coiled-coils (70). A collapse into coiled-coil structures may bring about membrane fusion in a fashion similar to that hypothesized for the enveloped myxoviruses (39), retroviruses (29,30,72), and filoviruses (33,73). We intend to use this model to generate and refine hypotheses on coronavirus entry into cells.