Crystallographic and Mutational Data Show That the Streptococcal Pyrogenic Exotoxin J Can Use a Common Binding Surface for T-cell Receptor Binding and Dimerization*

The protein toxins known as superantigens (SAgs), which are expressed primarily by the pathogenic bacte-ria Staphylococcus aureus and Streptococcus pyogenes , are highly potent immunotoxins with the ability to cause serious human disease. These SAgs share a conserved fold but quite varied activities. In addition to their common role of cross-linking T-cell receptors (TCRs) and major histocompatibility complex class II (MHC-II) molecules, some SAgs can cross-link MHC-II, using diverse mechanisms. The crystal structure of the streptococcal superantigen streptococcal pyrogenic exotoxin J (SPE-J) has been solved at 1.75 Å resolution ( R (cid:1) 0.209, R free (cid:1) 0.240), both with and without bound Zn 2 (cid:2) . The structure displays the canonical two-domain SAg fold and a zinc-binding site that is shared by a subset of other SAgs. Most importantly, in concentrated solution and in the crystal, SPE-J forms dimers. These dimers, which are present in two different crystal envi-ronments, form via the same face that is used for TCR binding in other SAgs. Site-directed mutagenesis shows that this face is also used for TCR binding SPE-J. PCR

The common human pathogens Staphylococcus aureus and Streptococcus pyogenes secrete a number of potent protein toxins known as superantigens. These toxins derive their name from their primary functional attribute, which is to bind simultaneously to T-cell receptors (TCRs) 1 and MHC class II (MHC-II) molecules, outside the MHC peptide-binding groove and as intact molecules rather than processed peptides. This can cause massive overstimulation of the cellular immune response, with the overproduction of cytokines such as tumor necrosis factor ␣ and interleukin-2, as a result of uncontrolled T-cell activation (1)(2)(3). This activity is central to their involvement in many human diseases, such as toxic shock, scarlet fever, food poisoning, and possibly others such as rheumatoid arthritis (2,4,5).
The SAg family comprises staphylococcal enterotoxins (SEs) such as SEA, SEB, SEC1-3, and SED, streptococcal pyrogenic exotoxins (SPEs) such as SPE-A and SPE-C, and toxic shock syndrome toxin-1 (TSST-1). The sequencing of the complete genomes of several strains of S. aureus (6) and S. pyogenes (7) has led to the discovery of many more sag genes, including spe-j, and the realization that in these two organisms this is a widespread protein family that must play a major role in their pathogenicity. The SAgs share widely different levels of sequence identity. Some are so similar (for example SEA and SEE, with ϳ 90% sequence identity) as to make allelic variants between different strains difficult to distinguish, but many share much lower sequence identity, around 20%. Structurally, however, the SAgs share a highly conserved fold (8,9), comprising an N-terminal ␤-barrel domain with the well known OB-fold (10,11), and a C-terminal ␤-grasp domain comprising a ␤-sheet that wraps around a long central helix.
A striking feature of the SAg family, however, is that this conserved fold supports a wide variety of different binding modes. Most SAgs (for example SEB and TSST-1) have a single MHC-II-binding site, located on their N-terminal domains (12)(13)(14), often referred to as the generic MHC-II-binding site, whereas others (such as SMEZ and SPE-H) have instead a site on their C-terminal domains, mediated by a bound Zn 2ϩ ion (9). Still others, such as SEA, have both sites (15)(16)(17), giving them the ability to cross-link MHC-II on antigen-presenting cells (APCs) and thus elicit intracellular signaling in the APCs. A variation on this theme is given by several other SAgs, including SED and SPE-C, which can cross-link MHC-II by formation of homodimers. Thus, SED forms zinc-dependent homodimers through its C-terminal domain and can cross-link MHC-II through the N-terminal domain sites at each end of the homodimer (18). On the other hand, SPE-C dimerizes via its N-terminal domain and can cross-link MHC-II by the two Cterminal domain Zn 2ϩ sites of the dimer (19).
In contrast to the varied MHC-II-binding modes, the evidence so far suggests that most, if not all, SAgs bind to TCR via a common site, at the interface between the N-and C-terminal domains (20,21). The ability to select particular TCR V␤ subtypes appears to derive from sequence and structural diversity on the TCR, coupled with local SAg sequence variation at a common TCR-binding site.
Streptococcal pyrogenic exotoxin J (SPE-J) was first identified from the S. pyogenes genome sequence (22). The recombinant protein has been shown to be a highly potent mitogen, giving half-maximum responses at 0.1 pg/ml. In terms of sequence, SPE-J is most closely related to SPE-C (49% identity), and it appears to be functionally indistinguishable; its T-cell specificity is the same (for V␤2.1), and like SPE-C (23) it forms homodimers and induces rapid homotypic aggregation of LG-2 cells, implying an ability to cross-link MHC-II (24). This raises the question as to why a bacterial isolate should maintain two different genes that code for apparently functionally identical proteins.
Here we show, from its high resolution crystal structure, that SPE-J forms a completely different dimer from that of SPE-C. Intriguingly, the interface used for dimerization proves to be the same as that used for TCR binding. This leads to the conclusion that SPE-J must bind to TCR as a monomer but that concentration-dependent dimerization allows it also to stimulate intracellular signaling in APCs by cross-linking MHC-II as a dimer.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-SPE-J was cloned and expressed in Escherichia coli as described by Proft et al. (24). The protein was overexpressed as a glutathione S-transferase fusion protein and was initially purified using glutathione/agarose. After cleavage of glutathione S-transferase from the toxin with protease 3c, the protein was further purified by cation exchange chromatography (MonoS HR 5/5 column, Amersham Biosciences) followed by gel filtration (Superdex 75 HR 10/30 column, Amersham Biosciences). Small fractions were taken across the protein peak, and dynamic light scattering (see below) was used to decide which fractions were to be taken for crystallization trials. Only those with a Cp/R H ratio of less than 14% were used.
Light Scattering Analysis-Dynamic light scattering was performed by using a Protein Solutions (Charlottesville, VA) DynaPro molecular sizing instrument to determine not only the monodispersity of protein samples, prior to crystallization, but also to determine the relative molecular mass of the protein at various concentrations. Samples ranged in concentration from 0.8 to 12.0 mg/ml, and 30 measurements were made at each concentration. Results are summarized in Table I.
Crystallization-Crystals were grown at 18°C by the hanging drop method by mixing 1 l of protein solution (10 mg/ml protein in 50 mM HEPES/KOH, 100 mM NaCl, pH 7.0) with 1 l of reservoir solution (210 mM lithium acetate, 17% PEG 3350, pH 5.5). Small shield-shaped crystals of maximum dimension of 0.03 mm grew over a period of 2-3 weeks. These crystals were monoclinic, space group C2, with unit cell dimensions a ϭ 165.6, b ϭ 46.4, c ϭ 72.2 Å, ␤ ϭ 90.6°. This gave V m values of 3.0 Å 3 /Da (59% solvent) assuming two molecules per asymmetric unit, or 2.0 Å 3 /Da (39% solvent) assuming three molecules per asymmetric unit; the structure determination showed the latter to be correct.
Crystals of zinc-bound SPE-J (Zn-SPE-J) were obtained by soaking crystals in 100 mM zinc acetate, 20% PEG 3350, 230 mM lithium acetate, pH 5.8, for 1 h. This short, sharp soak gave much better diffraction than from crystals soaked in lower zinc acetate concentration (1 mM) for a longer period (4 -24 h). Crystals were mounted in a cryoloop and flashfrozen by plunging into liquid N 2 after a rapid pass through a cryoprotectant solution. The latter comprised 0.23 M lithium acetate, 18% PEG 3350, and 20% ethylene glycol, pH 5.8, for SPE-J and 100 mM zinc acetate, 0.23 M lithium acetate, 20% PEG 3350, and 20% ethylene glycol, pH 5.8, for Zn-SPE-J.
Data Collection-X-ray diffraction data to 1.7 Å resolution were collected for SPE-J at 110 K at the Stanford Synchrotron Radiation Laboratory. Zn-SPE-J data to 2.0-Å resolution were collected at 110 K using CuK ␣ radiation from a Rigaku RU-H3R x-ray generator equipped with Osmic mirrors, an Oxford cryostream, and a Mar345 imaging plate system. Raw data were processed using MOSFLM (25) and scaled and merged with SCALA (26). Data collection statistics are summarized in Table II.
Structure Determination and Refinement-The structure of SPE-J was solved by molecular replacement using AMoRe (27) with the closely related SAg structure SPE-C (19) (Protein Data Bank code 1AN8), as search model; SPE-J and SPE-C share 49% sequence identity. Two molecules were found and used for phasing to 2.0 Å resolution, after which an initial model was built with ARP/wARP (28). This gave an almost complete model for both molecules (391 of 422 residues) and also revealed the position of a third molecule, which was added to the model. Further refinement was with CNS (29), with cycles of refinement being interspersed with manual model building into the electron density using the graphics program TURBO FRODO (30). Solvent molecules, all treated as water, were added using the WATERPICK facility in CNS and were retained if they had spherical density and appropriate hydrogen bond geometry. The quality of the model was checked periodically with PROCHECK (31), and hydrogen bonds were identified following the distance and angle criteria of Baker and Hubbard (32). The Zn-SPE-J structure was solved using the final SPE-J structure as a starting model and was refined in the same way.
Site-directed Mutagenesis-Single-site mutants of SPE-J were generated by overlap PCR. Oligonucleotide primers pGEX.fw/SpeJmut.rev and pGEX.rev/SpeJmut.fw were used for 12 cycles of PCR with pGEX-3c:speJ (24) as template. See supplemental Material for primer sequences. The PCR products were then purified from agarose gels and used as templates for 18 cycles of PCR with pGEX.fw/pGEX.rev primer pairs. The PCR products were cloned into pGEX-3c vectors, and the recombinant SPE-J mutant proteins were produced as described previously for wild type SPE-J (24). The DNA sequences of the cloned SPE-J mutants were confirmed using a Licor automated DNA sequencer (model 4200).
Toxin Proliferation Assay-The mitogenic activity of the SPE-J mutants was determined in a peripheral blood lymphocyte (PBL) stimulation assay as described previously (22). In brief, PBLs were purified from blood of healthy donors and incubated with varying dilutions of SPE-J mutants (100 ng/ml to 1 fg/ml). After 3 days of incubation at 37°C, 0.1 Ci of [ 3 H]thymidine was added. After another 24 h, the PBLs were harvested and counted on a Cobra scintillation counter. The decrease in T-cell mitogenicity was calculated as the amount of mutant toxin needed to achieve half-maximum stimulation (P 50 value) of wild type SPE-J.

RESULTS
Crystal Structure of SPE-J-The three-dimensional structure of SPE-J was determined by molecular replacement and refined at 1.75 Å resolution to an R factor of 0.209 and free R factor of 0.240. The model has excellent geometry with 87.2% of non-glycine residues falling in the most favored regions of the Ramachandran plot, as defined in PROCHECK (31), with no outliers. The three molecules in the asymmetric unit of the crystal are organized in such a way that A and B form a putative dimer (see below) and C also forms a dimer, with another molecule C, related by 2-fold crystallographic symmetry. The final model for molecule A comprises the complete polypeptide for mature SPE-J, residues 1-209, but with two additional residues (Gly-2 and Ser-1) also modeled at the N terminus, left after cleavage of the glutathione S-transferase fusion domain. Sequence numbering here follows that of the mature protein. Molecule B lacks residues 97-101, and molecule C lacks residues Ϫ2, Ϫ1 and 1; these have no interpretable electron density and are assumed to be disordered. Further details are given in Table II.
Molecular Structure-SPE-J has the characteristic two-domain SAg fold (8,9), shown in Fig. 1. Following an N-terminal helix ␣2 (residues 2-17), which ends in the inter-domain cleft, the  (Fig. 2), which is considerably more extensive than any of their other packing interactions in the crystal, and has many of the properties expected of a protein dimer. This interface buries a total of 1360 Å 2 of accessible surface area (680 Å 2 per monomer, or 6.5% of the monomer surface), calculated using the Protein-Protein Interaction Server (www.biochem.ucl.ac.uk/bsm/PP/server); this uses the algorithm of Lee and Richards (33) with a probe radius of 1.4 Å. The interface is formed by the C-terminal half of helix ␣2 and its connection to strand ␤1, the ␤2-␤3 loop, the ␤4-␤5 loop, all from the N-terminal domain, and the end of helix ␣4 and start of helix ␣5, both from the C-terminal domain. The residues that make the greatest contribution to the interface are Tyr-14, Glu-17, and Ile-19 from ␣2, Phe-77, Arg-79, and Tyr-83 from ␤4-␤5, Gln-142 from ␣4, and Arg-181 from the start of ␣5, which hydrogen bonds across the interface to the carbonyl oxygen of Gly-17.
The third molecule in the asymmetric unit, molecule C, forms a very similar interaction with another molecule C, related by crystallographic symmetry. The dimerization of these two molecules buries a somewhat larger surface area of 2130 Å 2 (1065 Å 2 per monomer, 10.4% of the monomer surface). The structural elements that comprise it are the same as for the A-B dimer, however, involving residues in and around the interdomain cleft (Fig. 2). The principal contributors to the interface are Tyr-14, Glu-17, and Ile-19 from ␣2, Tyr-43, Lys-44, and Lys-45 from ␤2-␤3, Phe-77, Tyr-80, and Tyr-83 from ␤4-␤5, Gln-142 from ␣4, and Arg-181 from ␣5. In both the A-B and C-C dimers there are 6 -10 direct protein-protein hydrogen bonds across the interface, and a number of water molecules make bridging interactions.
Zinc Binding-SPE-J has been shown to bind to MHC-II in a zinc-dependent manner. The native SPE-J structure contained no bound zinc, however, and of the three residues proposed to form the Zn 2ϩ -binding site (24), the side chains of His-201 and Asp-203 were close together but that of the third putative ligand, His-167, was turned away. In the Zn-SPE-J structure, however, after soaking the crystals very briefly in 100 mM Zn 2ϩ , the side chain of His-167 had moved, and these three residues are bound to a fully occupied Zn 2ϩ ion, with bond lengths of 2.1-2.2 Å. A water molecule is bound as a fourth ligand, completing a tetrahedral coordination site. The zinc site is located on the concave surface of the C-terminal domain and is equally accessible for MHC-II binding in both the monomeric and dimeric forms of SPE-J dimer. In the latter the Zn atoms are ϳ60 Å apart, at the two ends of the dimer.
Functional Analysis of the TCR-binding Site in SPE-J-SPE-J is most closely related to SPE-C by amino acid sequence (49% identity) and, like SPE-C, primarily stimulates T-cells carrying the V␤2 TCR (24). We therefore selected for mutagenesis those residues in SPE-J that were equivalent to the SPE-C residues shown to contact V␤ in the TCR V␤-SPE-C co-crystal structure (21) (Table III). The recombinant SPE-J mutant proteins were then analyzed for their mitogenicity in standard PBL proliferation assays. To ensure that any loss in proliferation activity was because of impaired TCR binding, MHC-II binding of the mutant proteins was confirmed in a standard binding assay (data not shown).
The strongest decrease in potency for T-cell stimulation (10,000-fold) was observed with the Y14A and R181Q mutants, which correspond to SPE-C residues Tyr-15 and Arg-181, respectively. 10-Fold and 100-fold decreases in potency were detected with mutants K44A (Arg-45 in SPE-C) and F77A (Leu-78 in SPE-C), respectively. A minor difference was observed with T78A (Asn-79 in SPE-C), which had a 2-fold reduction in mitogenicity compared with wild type SPE-J. In contrast, the mutants E17A, I19A, K46A, F48A, and S178A showed no differences to wild type SPE-J in proliferation (Table III).

DISCUSSION
Solution studies, using dynamic light scattering, show clearly that SPE-J forms dimers at higher protein concentrations (Ͼ3 mg/ml) and in fact has a somewhat greater propensity for dimerization than SPE-C. In contrast, another SAg, SMEZ-2, showed only monomers under the same conditions. SPE-J was also shown to stimulate the rapid aggregation of LG-2 cells, presumably by cross-linking of MHC-II molecules (24). Further functional studies indicate that SPE-J binds to MHC-II only through the C-terminal zinc site that it shares with other zinc-dependent SAgs (24), and the conclusion must be that, as in the case of SPE-C (23), it is its ability to form dimers that enables SPE-J to cross-link MHC-II.
The mode of dimerization found in our SPE-J crystals was a surprise, however. It does not involve the N-terminal OB-fold binding face, as is the case for dimers of SPE-C (19), nor does it involve the face of the C-terminal ␤-sheet as for SED (18). Instead, the dimer interface in SPE-J, as seen in the crystal structure, involves residues in and around the interdomain region, residues 10 -19 from helix ␣2 and the ␣1-␤1 loop, 42-45 from the ␤2-␤3 loop, 77-83 from the ␤4-␤5 loop, Gln-138 and Gln-142 from helix ␣4 in the C-terminal domain, and Arg-181 from helix ␣5, also in the C-terminal domain. Although it cannot necessarily be inferred that a mode of association seen in crystals also occurs in solution, especially if the surface area buried by this association is not great (34), in the present case SPE-J has been shown to form dimers in solution, at similar concentrations as were used to grow the SPE-J crystals. Most significantly, the same dimer is found in two completely different crystal environments, namely the A-B dimer between two independent molecules in the asymmetric unit and the C-C dimer between two molecules related by crystallographic symmetry. This argues strongly that the same dimer would be seen in solution. The buried surface area (1360 Å 2 for A-B and 2130 Å 2 for the more closely packed C-C dimer) is at the low end of the range for functional dimers (34,35) but is consistent with the solution data that shows dimerization only at higher concentrations.
The SPE-J mutagenesis results indicate that the TCR binding surface is very similar to that for SPE-C, consistent with their high structural similarity. The strongest decrease in Tcell mitogenicity was observed in the Y14A and R181Q mutants   (10,000-fold). The equivalent residues in SPE-C (Tyr-15 and Arg-181) hydrogen bond to the TCR ␤-chain (21), and in SEC3 mutation of the equivalent residues (Asn-23 and Gln-210) abrogates TCR binding completely (36). Thus these residues appear to be key residues for TCR binding, and the strong conservation between SPE-C and SPE-J may explain their shared T-cell specificity, as both toxins primarily target the TCR V␤2 chain. The mutational data also show that Phe-77 makes an important contribution to TCR binding by SPE-J, although like its equivalent in SPE-C (Leu-78) it can only make van der Waals contacts. The neighboring residue in SPE-C (Asn-79) is hydrogen-bonded to TCR V␤, but mutation of Thr-78 in SPE-J has little effect. We conclude that the ␤4-␤5 loop, to which these residues belong (as well as Tyr-90, shown to be important in SEC3 (36)), is important for TCR binding but through different residues in different SAgs; indeed, the loop conformation is slightly changed in SPE-J relative to SPE-C. Most of the other residues mutated make only van der Waals interactions, and it is likely that a single residue makes too small a contribution to the interface to significantly weaken binding when truncated to alanine.
What is most striking is that the surface that is used for binding to the T-cell receptor V␤ chain during T-cell stimulation is essentially the same surface that is used for dimerization of SPE-J (Fig. 3). This surface, which is centered around the cleft between the N-and C-terminal domains, is the site of TCR V␤ binding for all the SAgs for which SAg-V␤ complexes have been structurally characterized.
From a structural viewpoint, it is not unreasonable that the same (or very similar) surface can be used to bind different molecules, as is shown in a recent analysis of protein-protein interfaces (37). Such surfaces could be described as "moonlighting" surfaces, able to support diverse protein-protein interactions, especially those that are transient or of relatively low affinity. The interaction of SAgs with their TCR V␤ ligands is of this nature (K D ϳ 10 Ϫ4 to 10 Ϫ6 M) (38), and the surface area buried is usually correspondingly small (1268 Å 2 for SEB-V␤ and 1324 Å 2 for SPE-A-V␤) (21). This is very similar to the surface area buried in the SPE-J dimer, and this too is of low affinity as shown by its concentration dependence.
One of the fascinating features of proteins of the SAg family is their diversity of functional behavior. Among the true SAgs, with immunostimulatory activity, some have only a single MHC-II-binding site, whereas others have two, and even when the same site is employed the mode of MHC-II binding can vary significantly. Some SAgs can cross-link MHC-II but by different mechanisms. Some seem always to be monomeric, but others can dimerize and in different ways. Moreover, in the wider SAg family there is a large group of toxins, the SETs, that are clearly homologous, sharing the same fold and moderate sequence identity, but which do not have superantigen activity (39). Instead, these proteins target other components of the immune system. 2 The SET family has recently been renamed SSL (staphylococcal superantigen-like) to avoid confusion with the prototypical SAgs (40). This diversity of activity in proteins with a common fold can arise because of its dependence on relatively low affinity surface interactions, often employing the same or similar binding sites, albeit with different details.
What are the functional and physiological implications of the dimerization seen here for SPE-J? The clear conclusion must be that SPE-J cross-links MHC-II and TCR only as a monomer, because the TCR-contacting residues are buried in the dimer. We also conclude, however, that SPE-J can cross-link MHC-II molecules as a dimer, and so stimulate intracellular signaling and cytokine expression by antigen-presenting cells, but independently of its TCR activation ability. Which of these activities is expressed may depend on local concentration effects.
Seen in this light, the ability of SPE-J to express different activities under different conditions, stimulating T-cells as a monomer and cross-linking MHC-II as a dimer, is yet another expression of the diverse behavior of this family. It is reasonable to suppose that other activities will be uncovered, even for apparently well characterized family members. In this connection, we note that in at least one crystal form of TSST-1 (41), a crystal dimer is found that uses essentially the same binding surface as in the SPE-J dimer. The relative orientations of the two TSST-1 molecules are different from those of the two SPE-J molecules, but the interaction could imply that TSST-1, too, could express other activities through this association, if replicated in solution. Again by analogy with SPE-J, the interaction in the TSST-1 crystal dimer could provide a model for its TCR binding surface, which has not yet been defined crystallographically.