Alternative intermolecular contacts underlie the rotavirus VP5^* two‐ to three‐fold rearrangement

We expressed the RRV VP5^* antigen domain with an N‐terminal histidine tag in bacteria and insect cells. When expressed in Escherichia coli, the domain is insoluble (data not shown). When expressed in Sf9 insect cells from a recombinant baculovirus vector, the domain is soluble and readily purified by nickel affinity chromatography and size exclusion chromatography, with anion exchange chromatography interposed when higher purity is needed (Supplementary Figure 1A and B). This procedure yields up to 4 mg of purified protein from each liter of insect cell culture. The purified domain is soluble to 4 mg/ml. SDS–PAGE, mass spectrometry, and N‐terminal sequencing reveal some heterogeneity due to inconsistent cleavage of the histidine tag by adventitious proteases (Supplementary Figure 1B). The solubility of the directly expressed VP5^* antigen domain suggests that further structure‐based engineering may yield an optimized version of the domain that can be produced efficiently enough to be a practical immunogen for inclusion in a subunit rotavirus vaccine.

Equilibrium analytical ultracentrifugation demonstrates self‐association of the domain. The domain's hydrodynamic behavior can be modeled as a dynamic equilibrium between monomers and trimers with an association constant of between 18.7 and 62.5 (Supplementary Figure 1C). The domain's equilibrium distribution is similar at 4 and 22°C and at pH 5.6 and 8.0 (Supplementary Table). The hydrodynamic data cannot rule out the possibility that a small proportion of oligomers other than trimers are also present in solution, but the simple monomer–trimer equilibrium model is sufficient to fit the observed distributions closely.

At 25°C and pH 5.6, the VP5^* antigen domain crystallizes by the hanging drop method, using 2‐methyl‐2,4‐pentanediol (MPD) as a precipitant. When frozen, these crystals diffract X‐rays coherently to 1.5 Å interplanar spacing (although anisotropy limits the resolution of useable data to 1.6 Å). We determined the structure of these crystals by molecular replacement, using VP5CT as an initial phasing model (See Materials and methods and Table I).

This VP5^* antigen domain crystal structure reveals a dimer. Figure 3A shows the dimer in the orientation of the spike on a trypsin‐primed virion. In this orientation, the heads would be located above the diagram, and the foot would be buried under the VP7 shell below the diagram. The dimer has maximal dimensions of 78 Å (height), 45 Å (width), and 30 Å (depth), as measured on a Cα trace. The N‐ and C‐termini of each subunit are at the bottom, and three loops (B′C′, D′E′, and F″G′) with hydrophobic tips are at the top. In the dimer, the F″G′ loop, which resembles the Semliki Forest virus fusion loop in primary sequence (Mackow et al, 1988), is distal to the approximate two‐fold axis, and the B′C′ and D′E′ loops are adjacent to this axis. Like the protruding VP4 dyad on virions (Shaw et al, 1993; Yeager et al, 1994), the crystallized antigen domain dimer is slightly asymmetrical. A highly ordered molecule of MPD (not shown) is bound between the sheets of the β‐sandwich of the green subunit in Figure 3A, but is absent from the blue subunit. In solution at pH 5.6 and 22°C, the antigen domain is in a dynamic equilibrium, with monomers and trimers as the main species (Supplementary Table). The change in solvent characteristics caused by MPD and glycerol in the crystallization mother liquor or the asymmetry introduced by the insertion of an ordered MPD molecule into some molecules may alter the equilibrium in favor of dimer formation.

The contacts between subunits at the top of the dimer are insubstantial—salt bridges between E293 of one subunit and R341 of the other subunit and a hydrogen bond between the hydroxyl of S293 and its counterpart across the approximate two‐fold axis (not shown). Instead, the dimer is primarily held together by extensive two‐fold interactions near the bottom of the structure. These contacts include an intersubunit β‐sheet (red box in Figure 3A and C) and a hydrophobic core (blue and black boxes in Figure 3A, E, and F). The central strands of the intersubunit β‐sheet (strand G from each subunit) share eight backbone amide‐to‐backbone carbonyl hydrogen bonds and an additional hydrogen bond between the S412 side chain and its symmetry mate (Figure 3C). This interaction creates a continuous 10‐stranded β‐sheet (CDEHGGHEDC), which forms a saddle across the dimer interface (Figure 3A).

The hydrophobic core on the two‐fold axis is below the new β‐sheet. One dimer contact is made by the aromatic ring of Y367 stacking against its symmetry mate (Figure 3E). Below this, two W262 aromatic rings pack tangentially against each other, separated at their distal ends by a sandwiched pair of L473 side chains (Figure 3F). At the bottom of the structure, the L261 side chain contacts its symmetry mate (Figure 3F). An additional intersubunit contact, formed by the N‐terminus of one subunit (blue in Figure 3A) crossing over to pack as strand A against the other subunit, is discussed below under ‘Comparison of VP5^* subunit structures.’ The dimer contacts bury 1846 Ų (15.2%) of the surface of each subunit. Excluding the crossed‐over N‐terminus, 1371 Ų (11.9%) are buried.

A 12 Å resolution electron cryomicroscopy image reconstruction of a trypsin‐primed rotavirus virion provides a molecular envelope of the dimeric, protruding region of cleaved VP4 (Figure 4). Previously, we described the ‘in silico’ trimming and reorientation of two subunits of the VP5CT trimer to model a potential fit of a hypothesized dimeric arrangement of the subunits to this molecular envelope (Dormitzer et al, 2004). In that fit, the trimmed subunits were arranged as a parallel dyad (although they did not contact each other), with the hydrophobic apices occupying the ‘shoulders’ of the spike body and each F″G′ loop located distal to the approximate two‐fold axis (Supplementary Figure 2). The dimeric crystal structure demonstrates that the VP5^* antigen domain actually forms a dimer with parallel subunits in which the F″G′ loop of each subunit is distal to the two‐fold axis (Figure 3A). The VP5^* antigen domain dimer fits the electron cryomicroscopy envelope in the orientation reported for the models carved from the VP5CT structure (Figure 4). The gap between the subunits of the dimer in the crystal structure corresponds to the hole in the spike body, the hydrophobic apices correspond to the ‘shoulders,’ and the C‐termini extend into the stalk to connect with the buried foot domain.

Despite the good overall match of the dimeric crystal structure to the electron cryomicroscopy envelope, the fit is imperfect, with 25.6% of the atoms protruding beyond an envelope contoured at 0.5σ (Figure 4). In the previously reported fit of trimmed VP5CT subunits (Supplementary Figure 2, Dormitzer et al, 2004), which were not constrained in their orientation relative to each other by dimer contacts, 18.9% of the atoms protruded. Relative to the equivalent parts of the trimmed VP5CT subunits, the hydrophobic apices of the VP5^* antigen domain dimer extend further beyond the spike shoulders. The dimer subunits are closer to the approximate dyad axis of the molecular envelope, leaving the lateral parts unfilled. Although not apparent from the perspectives in Figure 4, the flexible tip (with high thermal parameters) of each GH hairpin of the dimer protrudes from the molecular envelope. A shift in the GH hairpin (described under ‘Comparison of VP5^* subunit structures’) from its position in VP5CT to form the central strands of the intersubunit β‐sheet in the dimer fills in some of the envelope at the base of the body on the approximate dyad axis. The fit of the dimeric crystal structure to the molecular envelope could be improved if the dimer were extended about the ‘joint’ formed by its proximal two‐fold contact so that the apices separated from each other. This distortion would produce a better match to the previously reported independent fit of each electronically trimmed subunit from VP5CT (Supplementary Figure 2).

The imperfect match between the dimeric crystal structure and the molecular envelope of the primed spike probably reflects the conformational effects of interactions of the VP5^* antigen domain with VP8^* and with the stalk and foot regions of VP5^* on the virion. In the primed spike, approximately 60 N‐terminal residues of each VP8^* fragment tether each head to the body (Figure 1A; Dormitzer et al, 2004). Many of these residues probably insert between the two hydrophobic VP5^* apices to form the distal dyad contact of the spike body. The alternative possibility, that the VP8^* N‐termini fill the lateral part of the electron cryomicroscopy envelope, is less likely, as antibody neutralization escape mutations are broadly distributed over the surface of the VP5^* antigen domain, indicating surface exposure on the virion (Dormitzer et al, 2004). Insertion of the N‐terminal residues of VP8^* into the insubstantial distal two‐fold contact of the dimeric crystal structure would force the VP5^* apices apart, producing the extension about the proximal two‐fold contact that improves the fit to the envelope. Because the VP8^* N‐terminus forms the distal two‐fold contact and tethers the heads to the body, VP8^* probably either dissociates from the folded back, trimeric form of VP5^* or has a substantially altered association. Early dissociation of the heads could allow the VP5^* antigen domains to flex about their proximal two‐fold contact and achieve the conformation in the dimeric crystal structure prior to trimerizing and folding back.

During storage at 8°C at pH 8.0, the VP5^* antigen domain crystallizes in batch without added precipitant. These crystals, when frozen, diffract X‐rays coherently to 2 Å interplanar spacing. We determined the structure of these crystals by molecular replacement, using VP5CT as an initial phasing model (see Materials and methods and Table I).

The structure reveals a VP5^* antigen domain trimer (Figure 3B). Like the VP5CT trimer (Figure 2A), the VP5^* antigen domain trimer is shaped like an umbrella. Unlike the VP5CT trimer, the VP5^* antigen domain trimer lacks the coiled‐coil or β‐annulus that form the umbrella's ‘post’ due to the C‐terminal truncation of the recombinant construct. In the VP5^* antigen domain trimer, 1286 Ų (11.0%) of the surface of each subunit are buried. The additional three‐fold contacts of VP5CT raise its buried surface area to 3956 Ų (25.8%). The tips of the globular ‘shades’ of the VP5^* antigen domain umbrella approach each other more closely than those of VP5CT because they are not separated by a coiled‐coil ‘post.’ Except for the position of the GH loop (discussed under ‘Comparison of VP5^* subunit structures’), the globular regions of the VP5^* antigen domain and VP5CT trimers are essentially identical (RMSD of 0.76 Å for Cα of residues Y267‐T410 and S423‐L470).

The fold‐back rearrangement of VP5^* (Figure 1) inverts the antigen domain of the trimer relative to the dimer. Correspondingly, in Figure 3, the dimer and the trimer are both depicted with the probable position of the foot at the bottom, but the subunits of the trimer are inverted relative to those of the dimer. In Figure 3B, the trimer is presented with each subunit's termini at the top and hydrophobic apex at the bottom. The maximal dimensions of the trimer are 74 Å (height) and 33 Å (radius), as measured on a Cα trace.

The view of the blue subunit in Figure 3B displays features that are apparent at the 1.6 and 2.0 Å resolutions of the VP5^* antigen domain structures, but not at the 3.2 Å resolution of the VP5CT structure (Dormitzer et al, 2004). Specifically, adjacent pairs of glycines interrupt the β‐structure of the F′G loop, so that strand F′ from the VP5CT structure is divided into strands F′ and F″ (by G382 and G383) in the antigen domain structures, and strand G is divided into strands G′ and G (by G399 and G400). The more radial limb of the shortest loop at the bottom of the trimer structure (the loop that includes strand H′) had weak electron density in the VP5CT maps and was modeled as a coil (Figure 2A). In the new maps, these residues have strong density and form β‐strand I (Figure 3B). Accordingly, VP5CT strand I becomes strand J, and VP5CT strand J becomes strand K in the antigen domain structures.

The antigen domain trimer is held together by a hydrophobic core centered on the three‐fold axis near the top of the structure. The packing of this hydrophobic core is the same in the VP5^* antigen domain trimer and in VP5CT. Three main levels of hydrophobic interactions are apparent (red, blue, and black boxes in Figure 3B). At the bottom level (red box in Figure 3A and D), the F415 aromatic ring of each subunit packs about the three‐fold axis and forms the base of a large solvent filled cavity (volume 390 ų; Supplementary Figure 3). This cavity is also present in VP5CT and communicates with the molecule's exterior through narrow channels. During entry, it could allow room for movements associated with molecular rearrangements. The ceiling of the cavity is formed by Y367 aromatic rings (blue box in Figure 3B and G), which pack about the three‐fold axis, reinforced by interdigitating V366 side chains. Above this (black box in Figure 3B and H), a tight ‘propeller’ of three W262 aromatic side chains is held together by interactions between the hydrogen bound to each pyrrole nitrogen and the π electrons of the adjacent indole ring. The propeller is reinforced by interdigitating L473 side chains. The packing of each L261 side chain against the L473 side chain of the adjacent subunit caps the hydrophobic core (Figure 3H).

The structure of the VP5^* antigen domain trimer has several implications for VP4 rearrangements. First, near the bottom of the structure, the subunits do not interact. In VP5CT, the interactions between the lower portions of each domain and the coiled‐coil are polar. Thus, the lack of a central coiled‐coil in the VP5^* antigen domain trimer does not expose hydrophobic patches. The overall hydrophilicity of the sides (but not the apex) of the globular domain allows for its free rotation through solvent during the fold‐back translocation. Second, alternative packing of the same residues in the dimer and the trimer allows the formation of two well‐ordered oligomers. Residues L261, W262, Y367, and L473 make key hydrophobic contacts about both the two‐fold axis of the dimer (Figure 3E and F) and the three‐fold axis of the trimer (Figure 3G and H). Other residues, such as V366 (Figure 3E and G) and F415 (Figure 3C and D), only make intersubunit contacts in the trimer.

Superposition of a subunit from the VP5^* antigen domain dimer on a subunit from the VP5CT trimer (Figure 5) demonstrates that most of the globular domain remains rigid during the two‐ to three‐fold reorganization (RMSD 0.85 Å for Cα of residues D270‐Q360, A372‐T410, and S423‐L470). However, the GH loop rotates by approximately 61° relative to the rest of the domain, displacing its tip (residue D417) by 18.4 Å (two‐headed arrow in Figure 5). In the dimer, the GH loop forms the central strands of the intersubunit β‐sheet (Figure 3C); in the VP5CT trimer, this loop forms six of nine strands of the intersubunit β‐annulus (Figure 2A). The shift of the loop disrupts the intersubunit β‐sheet of the dimer and allows the F415 aromatic rings to pack around the three‐fold axis of the trimer (Figure 3D). In VP5CT, hydrogen bonds in the β‐annulus between strand G and an additional β‐strand formed by residues just N‐terminal to the coiled‐coil substitute for the disrupted bonds in the intersubunit β‐sheet and clamp the subunits in the folded‐back conformation (Dormitzer et al, 2004). The shift in the GH loop is accompanied by the partial uncoiling of an adjacent, short α‐helix (labeled ‘α’ in the green subunits of Figure 3A and B) between β‐strands E and F. The GH loop of the VP5^* antigen domain trimer does not form part of an intersubunit β‐sheet or β‐annulus, and its position may not model a conformation attained during the rearrangements that lead to cell entry.

Comparison of the VP5^* antigen domain dimer and VP5CT trimer subunits also reveals significant rearrangements of the domains’ termini (Figure 5). The C‐termini of the dimer and trimer point in opposite directions, reflecting the fold‐back relative to the foot domain. The N‐terminus of one subunit of the dimer (the blue subunit in Figure 3A) crosses over to form intersubunit β‐strand A, which is hydrogen bonded to strand B of the other subunit. The equivalent residues of the other dimer subunit are either disordered or missing due to proteolysis. In the trimer, the N‐terminus of each subunit folds back, so that β‐strand A is hydrogen bonded to strand B of its own subunit (Figures 3B and 5). Whether strand A is exchanged or retained, the same residues on strands A and B hydrogen bond to each other in both dimer and trimer (E252–N268, I254–Q266, and V256–E264). The crossed‐over strand of the dimer does not fit the molecular envelope of the primed VP4 spike (and is omitted from the Cα trace in Figure 4). Therefore, it is doubtful that the crossover occurs in the primed conformation of the spike on the virion. Trypsin cleavage of VP4 to produce an N‐terminus at A248, but not an alternative N‐terminus at residue N242, primes particles to infect cells and to mediate cell–cell ‘fusion from without’ (Arias et al, 1996; Gilbert and Greenberg, 1998). The specificity of the priming cleavage suggests that the interactions of N‐terminal residues of VP5^* in the unprimed, primed, or folded‐back states of the spike may have an essential role in controlling VP5^* rearrangements.

The core of the VP5^* antigen domain is an eight‐stranded β‐sandwich, composed of β‐sheets BKF and CDEHG (Figure 3A and B). Functional appendages mounted on this core include three adjacent hydrophobic loops (B′C′, E′D′, and F″G′), which may interact with membranes, and the CD loop, which projects from the opposite end of the β‐sandwich and may bind an integrin (Graham et al, 2003). The ability to form dimers or trimers alternatively is added to the core β‐sandwich by residues in N‐ and C‐terminal extensions and by the GH loop.

The fold of the core framework that bears these functional appendages is unique—the DALI structural similarity search algorithm (Holm and Sander, 1993) reveals no molecule in the Protein Data Bank with the same fold. The closest match is the reovirus σ1 knob (z score 4.1), which is very similar and probably homologous to the adenovirus fiber protein (Chappell et al, 2002). Rotavirus and reovirus are both members of the family Reoviridae. VP4 and σ1 both form viral spikes that function in attachment by binding cell surface proteins and sialic acid, albeit using different structural elements. The σ1 eight‐stranded antiparallel β‐barrel superimposes strikingly well on the VP5^* antigen domain β‐sandwich (Figure 6A), with an RMSD of 2.58 Å for 76 corresponding Cα atoms, 9.2% of which are identical. However, the two proteins have somewhat different folds. The σ1 β‐barrel has a double Greek key fold (Figure 6B). Although the five C‐terminal strands (E, F, G, H, and K) of the VP5^* antigen domain β‐sandwich share this connectivity, the three N‐terminal strands (B, C, and D) are arranged in the opposite order (Figure 6B).

There are several possible explanations for the similarities and differences between the folds of the rotavirus VP5^* antigen domain and the reovirus σ1 head. The two proteins could lack common ancestry, yet have converged on closely related folds. However, there is no known functional significance to the details of β‐barrel folding that would drive the convergence to such similar structures in the spike proteins of two viruses in the same family. Another possibility is that small initial mutations in a common ancestral domain could have lead to a change in the pathway of protein folding, with a three‐strand section of the VP5^* β‐sandwich or σ1 β‐barrel flipping ‘inside out.’ However, such refolding would necessitate solvent exposure of a number of hydrophobic side chains that pack in the interior of the β‐structures. In a third scenario, genetic rearrangements could have altered the connectivity of the three N‐terminal strands. Altered connectivity by this mechanism has been inferred by comparing the sequences of the palm domains of the RNA‐dependant RNA polymerases (RdRps) of the tetraviruses and birnaviruses to the sequences of other RdRps of known structure (Gorbalenya et al, 2002). In the case of VP5^* and σ1, such a rearrangement would require exchange of sequences encoding strands B and D of VP5^* (or strands A and C of σ1).

A fourth possibility is suggested by the observation that the galectin‐like sialic acid‐binding domain of VP8^*, which is N‐terminal to the VP5^* antigen domain, bears no structural relationship to the β‐spirals in the sialic acid‐binding region N‐terminal to the σ1 head (Chappell et al, 2002). The overall organization of the rotavirus spike is consistent with the heads having arisen by the insertion of a host galectin‐like carbohydrate‐binding domain into an ancestral headless spike protein (Dormitzer et al, 2002). Thus, the three N‐terminal strands of the β‐structures may have separate origins in insertions that included a galectin‐like domain or a β‐spiral region added to the N‐terminus of a common Greek‐key ancestral core. Indeed, a systematic review of Greek‐key β‐barrels and β‐sandwiches shows that the arrangement of the five staves with shared connectivity in VP5^* and σ1 is conserved in many Greek‐key β‐structures (Zhang and Kim, 2000). Additional strands added to either terminus of this five‐stranded core are more variably connected, suggesting that they are appended to a common folding intermediate.

Several observations indicate that the VP5^* antigen domain dimer and trimer model two functionally important states of the rotavirus spike protein. The extensive two‐fold contacts in the dimeric crystal structure, including nine hydrogen bonds in the intersubunit β‐sheet and a well‐ordered hydrophobic core (Figure 3), provide evidence that the dimer is not simply an artefact of crystal packing. The fit of the dimer to the spikes on primed virions is confirmed by electron cryomicroscopy image reconstructions showing decoration of the ‘shoulders’ of the spike by the antigen‐binding fragment of a neutralizing monoclonal antibody that selects an escape mutation in the loop of the domain that fills the shoulders (Tihova et al, 2001). There are, however, subtle differences between the orientation of the domain in the crystal and on the virus, probably due to interactions with other parts of VP4 on the virion.

The reinforcement of the three‐fold contacts observed in the crystal of VP5^* antigen domain trimers by a triple coiled‐coil and β‐annulus in the more complete VP5CT fragment confirms the significance of the trimeric state of the antigen domain. Although analytical ultracentrifugation (Supplementary Figure 1C) indicates that the VP5^* antigen domain is in dynamic equilibrium between monomers and trimers in solution, the additional contacts in VP5CT yield a very stable oligomer. VP5CT is more resistant to denaturation by guanidine hydrochloride than uncleaved VP4 (unpublished data) and its subunits do not dissociate in the presence of SDS (Dormitzer et al, 2001). This stability suggests that, during cell entry, the two‐ to three‐fold rearrangement of VP5^* is essentially irreversible.

The two‐ to three‐fold reorganization and fold‐back of VP5^* is probably linked to membrane penetration. A two‐ to three‐fold reorganization, associated with a fold‐back rearrangement, mediates membrane fusion by class II‐enveloped virus fusion proteins, such as the Semliki Forest virus and dengue fever virus envelope glycoproteins (Gibbons et al, 2004; Modis et al, 2004). The structural bases for the oligomeric reorganizations of the class II fusion proteins and VP5^* are distinct. In the class II fusion proteins, different surfaces interact in the dimer and the trimer. In VP5^*, many of the same residues at the N‐ and C‐terminus and in the GH loop of the antigen domain have alternative packings with their symmetry mates in the dimer and the trimer. The formation of both a spike‐like dimer and an umbrella‐shaped trimer by the VP5^* antigen domain through alternative packings demonstrates a molecular property that underlies the reorganization of a nonenveloped virus membrane penetration protein from a metastable primed spike to a trimeric final state.

The coding sequence for RRV VP4 residues R247‐D479 was amplified by PCR from a plasmid containing a cDNA copy of the full‐length VP4 gene (GenBank accession code AY033150). This fragment was ligated into pET‐28b (Novagen), which had been cleaved with NdeI and BamHI. A fragment excised with XbaI and XhoI encodes an N‐terminal 6‐histidine tag separated by a thrombin cleavage site from VP4 and was ligated into pFastBac1 (Invitrogen). A recombinant baculovirus expressing this fusion was generated using the Bac‐to‐Bac system (Invitrogen).

Insect cell culture and baculovirus propagation were carried out as previously described (Willis et al, 1998). For protein production, suspension cultures of Sf9 cells, maintained in spinner flasks at 28°C, were infected with the recombinant baculovirus. Infected cells were harvested by centrifugation 72–86 h after infection and lysed by freezing and thawing in 20 mM Tris (pH 8.0), 100 mM NaCl, 10 mM imidazole, and 1 mM phenylmethylsulfonyl fluoride (PMSF). The lysate was clarified by centrifugation at 235 400 g for 1 h. Proteins in the clarified lysate were bound to nickel‐charged HiTrap Chelating HP resin (Amersham Biosciences) and eluted with imidazole. Some samples were further processed by anion exchange chromatography on Mono Q HR 5/5 (Amersham Biosciences) after concentration and buffer exchange to less than 10 mM NaCl using an Amicon Ultra‐4 centrifugal filter device (Millipore). Size exclusion chromatography on a Hi‐Load 16/60 Superdex 75 or 200 column (Amersham Biosciences), equilibrated in 20 mM Tris (pH 8.0), 100 mM NaCl, and 1 mM EDTA (TNE), was used as a final purification step. Chromatography was performed using an AKTA system (Amersham Biosciences).

Samples of the VP5^* antigen domain were microdialyzed against TNE or ANE (20 mM Na acetate, pH 5.6, 100 mM NaCl, 1 mM EDTA). The diasylate was used as a blank and as diluent. Samples with protein concentrations of 75–300 μg/ml were centrifuged in an AN‐60 Ti rotor at speeds of 10 000 and 14 000 r.p.m. in an Optima XL‐A analytical ultracentrifuge (Beckman Coulter). Absorbance data were measured at 280 nm with 10 replicates. Attainment of equilibrium was confirmed by comparing absorbance curves at 6‐h intervals. Origin software (Microcal Software, Inc.) was used to fit the data to the following model:

Image (embj7601034-math-0001.gif) is missing or otherwise invalid.

where A_r is the absorbance at radius x; A_0,1 is the absorbance at the reference radius x₀; H is a constant to account for protein specific volume, solvent density, angular velocity, and temperature; M is the monomer MW; N is the stoichiometry of the association; K is the association constant; and E is the baseline offset. A_0,1 was constrained to be greater than 0. The monomer molecular weight was fixed at 27 644, based on MALDI mass spectrometry and SDS–PAGE (Supplementary Figure 1B). Protein‐specific volumes and solvent densities were calculated based on published data (McRorie and Voelker, 1993). Each reported association constant (Supplementary Table) was determined by simultaneously fitting curves to protein distributions from six combinations of starting concentration and rotor speed.

At a concentration of 4 mg/ml in TNE, the VP5^* antigen domain crystallized in batch at 8°C. The crystals belong to space group P2₁2₁2₁ with unit cell parameters of a=93.12 Å, b=94.03 Å, and c=97.50 Å and contain one trimer in each asymmetric unit. These crystals were flash frozen by immersion in liquid nitrogen, using TNE with 25% glycerol as a freezing buffer. A second crystal form was obtained by the hanging drop vapor diffusion method after 72 h at 25°C with a well solution containing 100 mM sodium acetate, pH 5.6, 7% MPD, and 2% glycerol and with 2 μl drops containing a 1:1 mixture of 4 mg/ml protein in TNE and well solution. The crystals of this form belong to space group P2₁2₁2 with unit cell parameters of a=163.36 Å, b=54.81 Å, and c=65.73 Å and contain one dimer in each asymmetric unit. These crystals were frozen after serial soaks in freezing buffers consisting of well solution with 12, 16, and 20% MPD.

X‐ray diffraction data were collected at 100 K at the Advanced Light Source, beamline 8.2.2. Diffraction data were integrated and scaled using HKL2000 (HKL Research, Inc.). Starting phases for both crystal structures were obtained by molecular replacement using AMoRe (Navaza, 1994) in CCP4 (Collaborative Computational Project, 1994). Residues Y267‐L470 from the VP5CT crystal structure (PDB identification code 1SLQ) were used as a search model. The structure was refined by multiple cycles of simulated annealing, energy minimization, and individual B‐factor refinement using CNS (Brunger et al, 1998) alternating with rebuilding in O (Jones et al, 1991). Simulated annealing omit maps used in rebuilding were calculated using CNS (Brunger et al, 1998). Data collection and refinement statistics are reported in Table I.

Structure geometry was analyzed using PROCHECK (Laskowski et al, 1993). Molecular alignments were obtained using LSQMAN (Kleywegt, 1996). Cavities were modeled using CAVENV (Collaborative Computational Project, 1994) and O (Jones et al, 1991), and their volumes were calculated using MAMA (Kleywegt and Jones, 1999). Buried surface area was calculated using AREAIMOL (Lee and Richards, 1971). The dimeric form of the VP5^* antigen domain was fit to the molecular envelope of a VP4 spike from an approximately 12 Å resolution electron cryomicroscopy‐based image reconstruction of trypsin‐primed SA11‐4F rotavirus particles, as previously described (Dormitzer et al, 2004). A structural similarity search was conducted using the DALI algorithm (Holm and Sander, 1993). Figures were made using MOLSCRIPT (Kraulis, 1991), O (Jones et al, 1991), Illustrator and Photoshop (Adobe Systems, Inc.), and Origin (Microcal Software, Inc.).

Atomic coordinates and structure factor amplitudes for the VP5^* antigen domain dimer and trimer have been deposited in the Protein Data Bank (http://www.rcsb.org). The identification code for the dimer is 2B4H; the code for the trimer is 2B4I.

Supplementary data are available at The EMBO Journal Online.