Structure of CD84 provides insight into SLAM family function

CD84 Is Dimeric in Solution.

Refolded CD84 ectodomain (22,385 Da) and IgV domain (12,646 Da) both eluted as symmetric monodisperse peaks on calibrated gel-filtration columns, with apparent molecular masses of 60 kDa and 32 kDa, respectively [supporting information (SI) Fig. 5]. These data suggested that CD84 is not monomeric in solution. A more quantitative assessment of the oligomeric state was provided by sedimentation velocity analysis. The observation that the normalized sedimentation factor g(s*) plots superimpose is strong evidence that, over the concentration range examined, reversible dissociation is not occurring (14, 15) (Fig. 1). Analysis of the complete data sets with Sedphat version 4.04 (16), using the model of a single ideal species, yielded molecular masses of 23.9 kDa and 44.1 kDa for the CD84 IgV domain and full-length ectodomain, respectively. These results indicated that over the concentration ranges studied both the IgV and the complete CD84 ectodomain exist as dimers (expected dimer molecular masses of 25.3 and 44.8 kDa, respectively), with no indication of dissociation to monomers or further aggregation beyond the dimers. The lack of detectable dissociation, over the concentration range examined, sets the upper limit for the K_d at several hundred nanomolars.

SLAM Family Members Possess a Noncanonical IgV Domain.

The asymmetric unit in the CD84 crystals contains six independent copies of the CD84 IgV domain (Fig. 2A), which exhibits the classic two-layer β-sandwich topology present in other IgV structures. The front and back sheets of the domain are composed of the A′GFCC′C′′ and BED strands, respectively (Fig. 2B). The six independent molecules are substantially similar with each other: Pairwise comparisons resulted in rms deviations ranging from 0.44 Å to 0.79 Å (based on 103 C_α atoms), with the BC, C′C′′, and FG loops exhibiting the greatest variability (Fig. 2B). Two notable features are present in the CD84 IgV domain. Canonical IgV domains start with the A strand, followed by the A′ strand, which form hydrogen bonds with the B and G strands, respectively; however, the N terminus of the CD84 IgV domain lacks the equivalent of the A strand and begins with the A′ strand. Furthermore, the hallmark disulfide bond that links the B and F strands in canonical IgV domains is absent in CD84. Based on sequence analysis, both of these unique features are predicted to be present in all SLAM/CD2 family members and, in fact, have been directly observed in CD2, CD58 (17), 2B4 (18), and NTB-A (12).

CD84 Forms Two Types of Dimers in the Crystalline State.

The six molecules in the asymmetric unit form two distinct interfaces. Head-to-tail dimers, involving the approximately orthogonal association of front-sheet strands (GFCC′C′′), are formed by the interaction of molecules A with B, molecules C with E, and molecules D with F (Fig. 2A). These three independent dimers are very similar with rms deviations that range from 0.6 to 0.8 Å. Parallel, side-to-side dimers, formed by the association of molecules A with C, molecules B with D, and molecules F with the symmetry mate E′, are stabilized by interfaces formed by strands on the back sheet (A′B) (SI Fig. 6). The rms deviations among the side-to-side dimers range from 0.82 to 1.14 Å. Given the similarities, only the AB head-to-tail dimer and the AC side-to-side dimer are discussed unless noted.

The head-to-tail dimers are roughly twofold symmetric, with the interacting sheets crossing at a nearly orthogonal angle and burying ≈1,460 to 1,480 Ų of solvent-accessible surface area. This interface is composed predominantly of neutral polar residues, which participate in 14 hydrogen bonds and a number of van der Waals contacts (Fig. 2C and SI Table 1). No salt bridges contribute to this interface. Two predominant types of interactions, loop-to-strand and strand-to-strand, can be defined across this dimer interface. The strand-to-loop class includes side-chain-to-side-chain contacts formed by T38 and S39 on the CC′ loop from molecule A with A33 on the C strand and D90 on the F strand from molecule B; and T57, H58, R59, and Y62 on the C′′D loop from molecule A with D90, N92 on the F strand, and T99 on the G strand from molecule B. The strand-to-strand interactions include contacts between the side chains of A33 and T35 on the C strand of molecule A and side-chain and main-chain atoms on the C and C′ strands from molecule B, as well as side-chain-to-main-chain contacts involving Y42 and A41 from opposing molecules. In addition, a loop-to-loop interaction is provided by the side and main chains of R59 and P97, respectively. Because of the twofold symmetry of the homophilic dimer, there are two independent, symmetry-related sets of the above interactions. Notably, the symmetry-related equivalents of Y42 from the C′ strand of each molecule form a single stacking interaction that demarks the approximate middle of the interface. The interfaces in all three head-to-tail dimers are generally similar, and a number of the contributing residues are highly conserved (Fig. 2D).

Molecules A and C contribute eight residues to the side-to-side dimer interface, which buries ≈1,200 Ų of the surface area (SI Fig. 6 and SI Table 2). The side-to-side dimer interface has fewer direct hydrogen bonds (eight in the AC complex, three in the BD complex, and seven in the FE′ complex) than the head-to-tail dimer interface (14 in the AB complex), although more water-mediated polar interactions are present in the side-to-side dimer interface. The head-to-tail dimer has a significantly higher shape complementarity (Sc) than the side-to-side dimer, with average Sc values of 0.68 and 0.5, respectively (Sc values of 1.0 and 0 correspond to perfect complementarity and complete mismatch, respectively). The Sc value of 0.68 is slightly lower than protein–oligomer and protease–inhibitor interfaces (0.7–0.76) and is similar to that observed for antibody–antigen interfaces (0.65–0.68) (19).

CD84 Head-to-Tail Dimer Is the Predominant Dimer in Solution.

Based on their conservation and predicted contribution to the head-to-tail binding interface, 11 residues were mutated to alanine (T35, S39, Y42, T57, H58, R59, Y62, K88, D90, N92, and T99) and one to aspartic acid (Y42D) in the full-length ectodomain construct (CD84 IgVC domain). All mutants were expressed in Escherichia coli and successfully refolded in vitro, except for S39A. As judged by gel-filtration chromatography, seven mutant proteins (T35A, T57A, H58A, D90A, N92A, T99A, and Y42D) exhibited behavior consistent with a hydrodynamic radius significantly smaller than that of the wild-type protein (SI Fig. 7). One of these mutants, T99A, was examined by sedimentation velocity analysis. The superimposition of the normalized g(s*) plots is strong evidence that, over the concentration range examined, there is no reversible association occurring (Fig. 1C). Data for CD84-T99A were analyzed with Sedphat version 4.04 by using the hybrid model of a sedimentation coefficient distribution, c(s), plus a single ideal species (16) and yielded a molecular mass of 21.8 kDa. These results indicated that over the concentration range examined CD84-T99A is monomeric (expected monomer molecular mass of 22.4 kDa). It is likely that the T35A, T57A, H58A, D90A, and N92A mutations also disrupted the head-to-tail dimer interface because the corresponding proteins exhibited chromatographic behavior similar to that of T99A. The Y42A, R59A, Y62A, and K88A mutant proteins exhibited chromatography properties similar to the wild-type protein (SI Fig. 7), suggesting a lesser effect on the homophilic dimerization. Notably, the Y42D mutation, which likely introduces an unfavorable electrostatic interaction between symmetry-related side chains across the interface, resulted in the disruption of the dimer (SI Fig. 7). The N10A and Q24A mutants located in the crystallographically observed side-to-side dimer interface had no effect on the chromatographic behavior (data not shown), arguing against the existence of the side-to-side dimer in solution. Taken together, these results support a model in which the head-to-tail dimer is the predominant oligomeric state of CD84 in solution.

Cloning, Expression, and Purification of CD84 IgV and IgC Domain Proteins.

The extracellular IgV domain (110 amino acid residues excluding the initiator Met) and the entire ectodomain, composed of a membrane-distal IgV and a membrane-proximal IgC (IgVC; 199 amino acid residues excluding the initiator Met) of CD84, were expressed and refolded as described for B7–2 (22). Briefly, CD84 IgV and IgC domains were expressed in E. coli and refolded from inclusion bodies in vitro. Refolded IgV and IgC proteins were chromatographed on Superdex G-75 and Superdex 200 gel-filtration columns, respectively, followed by MonoQ chromatography. Expression and purification of selenomethionine-substituted IgV was similar to that described by Zhang et al. (22). The QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA) was used to generate a series of point mutations of CD84. Mutant sequences were confirmed by DNA sequencing, and the proteins were purified as described for the native material.

Crystallization and Data Collection.

Diffraction quality crystals of native and selenomethionine-substituted CD84 IgV domain were obtained by hanging drop vapor diffusion at 21°C by mixing 2 μl of 12 mg/ml protein in 10 mM (pH 8.5) Tris·HCl buffer with 1 μl of reservoir solution composed of 20% PEG3350, 0.3 M MgCl₂, and 0.1 M Hepes (pH 7.5) and equilibrating samples against reservoir solution for 2 days. Diffraction was consistent with the space group C222₁ (a, 61.02 Å; b, 170.77 Å; c, 148.68 Å; and six molecules of the CD84 IgV domain in the asymmetric unit). Crystals were briefly transferred to reservoir solution containing 20% PEG400 and flash-cooled in a nitrogen stream (Oxford Cryosystems, Oxford, U.K.) at 100°K. Native data were collected to a resolution of 2.0 Å from a single crystal at the National Synchrotron Light Source X-29 beamline. MAD data (wavelengths 0.97940, 0.97970, and 0.96100 Å) were collected to a resolution of 2.4 Å from a single selenomethionine-substituted crystal at the X-9A beamline. Intensities were integrated with HKL2000 (23) and reduced to amplitudes by using TRUNCATE (24) (SI Table 3). Heavy atom positions were determined with SOLVE (25) and SHELX (26). Initial phases were calculated with SOLVE and improved by using density modification methods in RESOLVE (27). The initial atomic model of the CD84 IgV domain was built with ARP/wARP (28), except for flexible loops, which required manual intervention. Crystallographic refinement used CNS1.1 (29) and Refmac (30). Model building was initially performed with O (31) and later with Coot (32). All other calculations used the CCP4 program suite (24). The quality of the final structure was verified with composite omit maps, and the stereochemistry was checked with PROCHEK (33). Figures were produced with PyMOL (DeLano Scientific LLC, Palo Alto, CA).

Multiple-sequence alignments were performed with MultAlin (34) and rendered with ESPript (35). Hydrogen bond distances, buried surface area, and shape complementarity were calculated by the programs CONTACT (with a cutoff of 3.5 Å), AREAIMOL, and SC, respectively, from the CCP4 package (24).

Analytical Ultracentrifugation.

The wild-type ectodomain, IgV domain, and T99A ectodomain mutant of CD84 proteins were exchanged into a buffer containing 20 mM Tris·HCl (pH 8.5) and 50 mM NaCl. For each sample, three or four concentrations (0.84–0.081 mg/ml for IgV, 0.89–0.076 mg/ml for ectodomain, and 0.64–0.15 mg/ml for the T99A mutant) were used in sedimentation velocity experiments. The data were collected at 20°C and centrifuged at 224,000 × g in a Beckman Coulter (Fullerton, CA) XL-I, and interference scans were acquired at 1-min intervals for 5 or 7 h depending on the sample. The AUC data were analyzed with the DcDt+ program of John Philo (version 1.99.1990) (14, 15). The data were also analyzed globally by using the direct boundary modeling program Sedphat (16), which allows the user to fit the data to various association schemes by using multiple as well as individual data sets.